The Pennsylvania State University
The Graduate School
Crime, Law, and Justice Program, Department of Sociology
DETERRING SEX OFFENDERS:
IDENTIFYING THE FACTORS THAT AFFECT POLICY
AND THE POLICIES THAT AFFECT RECIDIVISM
A Dissertation in
Crime, Law, and Justice
by
Sarah Koon-Magnin
2011 Sarah Koon-Magnin
Submitted in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
August 2011
The dissertation of Sarah Koon-Magnin was reviewed and approved* by the following:
R. Barry Ruback. Professor of Sociology and Crime, Law, and Justice Dissertation Advisor Chair of Committee
Derek A. Kreager Assistant Professor of Sociology and Crime, Law, and Justice
Jeremy Staff Associate Professor of Sociology and Crime, Law, and Justice
Aaron L. Pincus Professor of Psychology
John McCarthy Professor of Sociology Head of the Department of Sociology
*Signatures are on file in the Graduate School
iii ABSTRACT
Sexual offending is considered the most deplorable type of crime, especially when the victims are children. A few particularly sensational cases of sexual assaults against children have led to public outrage and, ultimately, the introduction of new strategies for controlling the sex offender population. Two of the most controversial techniques for controlling sex offender behavior, registration/notification policies and civil commitment statutes, were widely implemented in the 1990‟s. Registration/notification policies
(adopted in all 50 states) require that law enforcement officials both maintain a centralized database of all sexual offenders in each jurisdiction and make that list available to the public. Civil commitment statutes (adopted in 20 states) allow the state to keep high-risk offenders under state control (i.e., in state institutions) even after the offenders‟ criminal sentence is complete. This dissertation focused on the adoption of these policies (Study 1) and the impact of these policies (Study 2).
Study 1 tested two explanations that have traditionally been offered to explain the risk and timing of policy adoption: internal determinants and diffusion. Internal determinants are factors that are unique to the state, such as the political party of the governor or the unemployment rate. Diffusion is a temporal process in which states respond to the policy choices that other states are making. Studies of state-level adoption of multiple types of policies have found that both internal determinants and diffusion theories can be used to explain the timing of policy adoption (Berry & Berry, 1990).
These two theoretical approaches were tested using annual, state-level data from
1986 through 2007. The data used for analysis was drawn from multiple sources,
iv including the Statistical Abstracts of the United States, the Bureau of Labor Statistics, and the Uniform Crime Reports. Original variables (e.g., the percentage of neighboring states that have adopted the policy) were also created to capture the diffusion of the two policies and diffusion mechanisms.
The results of Study 1 suggest that the factors that led to the adoption of sexual offending legislation differed from those traditionally used to explain policy adoption.
Only one of the internal determinant variables was significantly associated with the risk and timing of adopting registration/notification policies. Region of the country was an important predictor of adoption risk, such that states in the Southern region were significantly less likely to adopt the policies, or did so later, than states in other regions of the country. Two additional internal determinants were associated with the timing of adopting a civil commitment statute. First, states with higher populations were more likely to adopt such policies, or to adopt them sooner, than states with smaller populations. Second, states with higher levels of unemployment were less likely to adopt civil commitment statutes, or adopted them later, than states with lower levels of unemployment. There was a significant diffusion effect on both policy outcomes. That is, as more time passed and more states adopted the policy, the remaining states were increasingly likely to adopt the policy, as well. Diffusion mechanisms (specifically, the passage of Federal legislation in the Jacob Wetterling Act and the Supreme Court ruling in the case of Kansas v. Hendricks) were negatively associated with the timing of adopting a civil commitment policy.
v Study 2 tested the impact of the adoption of these two policies on rates of sexual offending within states over time. The rates of six types of sexual offending were tested: rape (as measured by both the Uniform Crime Reports and the National Incident-Based
Reporting System), forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest (all five measured by the National Incident-Based Reporting
System). The findings of Study 2 provided no support for deterrence theory. Neither the adoption of a registration/notification policy nor a civil commitment statute led to a significant decrease in the rates of sexual offending. In fact, for four of the outcomes studied here, there were positive relationships between policy adoption and the rate of sexual offending. Failure to find a deterrent effect may mean that no such effect exists.
However, it is also possible that the limited scope of NIBRS data (only 15 states were available for analysis because states‟ reporting is voluntary) did not provide adequate power to detect an effect.
Four covariates were significantly associated with the rates of sexual offending within states. First, there was a positive association between the murder rate in the state and the rape rate (as measured by the UCR). Second, there was a positive association between the percentage of the population living in poverty and the rape rate (both measured by NIBRS). Third, the divorce rate within a state was positively associated with the rape rate (as measured by the UCR). Fourth, the incarceration rate within a state was positively associated with the rates of rape (as measured by both the UCR), forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest.
vi These two studies provide a comprehensive view of registration/notification policies and civil commitment statute. There are three important implications of this research. First, the success or failure of a policy, once implemented, should be tested as a likely predictor of policy diffusion. Second, future policy research should continue to address both adoption and evaluation, as these processes are intertwined. Third, legislators should consider empirical research when implementing sexual offending policy. Although there is symbolic value in implementing policies regardless of their effectiveness, sexual offending is a serious enough problem to demand policies that will also serve to reduce the rate of sexual offending.
vii TABLE OF CONTENTS
LIST OF FIGURES ...... xi
LIST OF TABLES ...... xiii
ACKNOWLEDGEMENTS ...... xx
Chapter 1 Introduction to the Research...... 1
Sexual Offenses in the United States ...... 3 Memorial Laws ...... 7 Jacob Wetterling Crimes Against Children and Sexually Violent Predators Act .... 7 Megan‟s Law ...... 8 Adam Walsh Act ...... 9 Punishments Uniquely Applied to Sexual Offenders ...... 10 Registration/Notification ...... 11 The Effect of Megan‟s Law on Recidivism ...... 14 The Effect of Megan‟s Law on Re-entry ...... 16 Civil Commitment ...... 18 A Comparison of Registration/Notification and Civil Commitment Policies ...... 23 Empirical Knowledge of Sexual Offender Behavior ...... 26 Sex Offender Recidivism ...... 27 Sex Offenders as a Homogenous Group ...... 29 Research Questions ...... 31
Chapter 2 Theoretical Framework ...... 33
An Introduction to the Theories of Policy Adoption...... 35 Internal Determinants Theory ...... 36 Demographic Internal Determinants ...... 37 Political Internal Determinants ...... 41 Racial Threat/Conflict Internal Determinants ...... 43 Social Disorganization Internal Determinants ...... 44 Structural-Functional Internal Determinants ...... 46 The Strengths and Weaknesses of Internal Determinants Theory ...... 47 Diffusion Theory ...... 48 Social Networks ...... 50 Neo-Institutionalism ...... 51 Neighboring States ...... 52 States within a Region ...... 54 States within the Nation ...... 55 Physical Diffusion ...... 56 The Strengths and Weaknesses of Diffusion Theory ...... 57 Diffusion Processes ...... 57 Hub States...... 58 National Pressure ...... 59
viii Media Pressure ...... 61 Social Movement Organizations Pressure ...... 62 A Hybrid Theory ...... 65 Hypotheses ...... 65
Chapter 3 Study 1: Identifying the Factors that Affect Policy ...... 72
Planned Analyses ...... 72 Event History Analysis ...... 73 Logic of Analysis ...... 74 The Hazard Rate ...... 75 Modeling Time in an Event History Analysis ...... 77 Data ...... 77 Dependent Variables ...... 78 Registration and Notification ...... 78 Civil Commitment Statutes ...... 82 Independent Variables: Internal Determinants ...... 85 Demographic Internal Determinants...... 85 Political Internal Determinants ...... 93 Racial Threat/Conflict ...... 96 Social Disorganization ...... 96 Structural-Functional Internal Determinants ...... 97 Independent Variables: Diffusion ...... 98 Independent Variables: Diffusion Processes ...... 102 Hub States ...... 102 Federal Government ...... 103 Media ...... 103 Social Movement Organizations...... 104 Full Model ...... 104 Results of an Event History Analysis Predicting the Adoption of a Registration/Notification Policy ...... 105 Specifying Time ...... 105 Demographic Internal Determinants...... 108 Political Internal Determinants ...... 110 Racial Threat/Conflict Internal Determinants ...... 111 Social Disorganization Internal Determinants ...... 112 Structural-Functional Internal Determinants ...... 113 Full Internal Determinants Model ...... 113 Diffusion ...... 114 Diffusionat the Neighboring Level ...... 115 Regional Diffusion ...... 116 National Diffusion ...... 121 Physical Diffusion Characteristics ...... 123 Diffusion Processes – Hub States ...... 124 Diffusion Processes – Federal Government ...... 126 Diffusion Processes – Media ...... 126 Diffusion Processes – Social Movement Organizations ...... 127 Full Diffusion Model ...... 128
ix Full Model ...... 131 Summary of Registration/Notification Findings ...... 132 Results of an Event History Analysis Predicting the Adoption of a Civil Commitment Statute ...... 134 Specifying Time ...... 134 Demographic Internal Determinants...... 136 Political Internal Determinants ...... 137 Racial Threat/Conflict Internal Determinants ...... 138 Social Disorganization Internal Determinants ...... 139 Structural-Functional Internal Determinants ...... 140 Full Internal Determinants Model ...... 141 Diffusion ...... 142 Diffusionat the Neighboring Level ...... 142 Regional Diffusion ...... 143 National Diffusion ...... 148 Physical Diffusion Characteristics ...... 149 Diffusion Processes – Hub States ...... 150 Diffusion Processes – Federal Government ...... 152 Diffusion Processes – Media ...... 153 Diffusion Processes – Social Movement Organizations ...... 153 Full Diffusion Model ...... 154 Full Model ...... 155 Summary of Civil Commitment Findings ...... 157 Discussion of Study 1 ...... 158 Hypotheses ...... 159 Implications ...... 165
Chapter 4 Study 2: Examining Whether Laws Against Sexual Offending Deter Sexual Offending ...... 169
Deterrence ...... 170 Deterrence and Sex Offending Legislation ...... 172 The Present Research ...... 173 Hypotheses ...... 175 Data ...... 176 Uniform Crime Reports ...... 177 National Incident-Based Reporting System ...... 178 Methods ...... 181 Results ...... 185 Between States ...... 185 Within States ...... 200 Fixed Effects Model Predicting the Effect of Registration/Notification on the Rape Rate ...... 201 Rape Rate as Measured by the UCR ...... 201 Rape Rate as Measured by NIBRS ...... 204 Fixed Effects Model Predicting the Effect of Registration/Notification on the Forcible Sodomy Rate ...... 207
x Fixed Effects Model Predicting the Effect of Registration/Notification on the Fondling Rate...... 209 Fixed Effects Model Predicting the Effect of Registration/Notification on the Rate of Sexual Assault with an Object ...... 212 Fixed Effects Model Predicting the Effect of Registration/Notification on the Statutory Rape Rate ...... 215 Fixed Effects Model Predicting the Effect of Registration/Notification on the Incest Rate ...... 216 Fixed Effects Model Predicting the Effect of Civil Commitment on the Rape Rate ...... 219 Rape Rate as Measured by the UCR ...... 219 Rape Rate as Measured by NIBRS ...... 222 Fixed Effects Model Predicting the Effect of Civil Commitment on the Forcible Sodomy Rate ...... 225 Fixed Effects Model Predicting the Effect of Civil Commitment on the Fondling Rate ...... 228 Fixed Effects Model Predicting the Effect of Civil Commitment on the Rate of Sexual Assault with an Object ...... 230 Fixed Effects Model Predicting the Effect of Civil Commitment on the Statutory Rape Rate ...... 231 Fixed Effects Model Predicting the Effect of Civil Commitment on Incest the Rate ...... 234 Discussion of Study 2 ...... 235 Hypotheses ...... 236 Summary of Findings ...... 237
Chapter 5 Discussion and Conclusions ...... 246
Summary of Findings ...... 246 Strengths, Limitations, and Considerations for Future Research ...... 249 Implications ...... 252 Uniqueness of Sexual Offenses ...... 253 Diffusion ...... 255 Deterrence ...... 256 Need for Evaluation ...... 256 Interdependence of Adoption and Evaluation ...... 258 Conclusion ...... 259
References ...... 261
Appendix A Addressing Missing Data ...... 275
Appendix B Newspapers and Wires searched to quantify the amount of Media Coverage .... 278
xi LIST OF FIGURES
Figure 3-1: State Diagram of Adopting the Event...... 75
Figure 3-2: Frequency of Sex Offender Registry Adoption, by State...... 81
Figure 3-3: Cumulative Frequency of Sex Offender Registry Adoption, by State...... 81
Figure 3-4: Frequency of Civil Commitment Adoption, by State...... 83
Figure 3-5: Cumulative Frequency of Civil Commitment Adoption, by State...... 84
Figure 3-6: Constructing the Regional Diffusion Variable (Part 1)...... 99
Figure 3-7: Constructing the Regional Diffusion Variable (Part 2)...... 100
Figure 3-8: Functional Forms of Time...... 106
Figure 3-9: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Northeast Region...... 117
Figure 3-10: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Midwestern Region...... 118
Figure 3-11: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Southern Region...... 120
Figure 3-12: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Western Region...... 121
Figure 3-13: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, National Diffusion...... 123
Figure 3-14: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Northeast Region...... 144
Figure 3-15: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Midwestern Region...... 145
Figure 3-16: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Southern Region...... 146
Figure 3-17: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Western Region...... 147
Figure 3-18: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, National Diffusion...... 149
xii Figure 4-1: UCR Rape Rate for States with Both Policies Compared to States with only Registration Policies...... 187
Figure 4-2: NIBRS Rape Rate for States with Both Policies Compared to States with only Registration Policies...... 190
Figure 4-3: NIBRS Forcible Sodomy Rate for States with Both Policies Compared to States with only Registration Policies...... 192
Figure 4-4: NIBRS Forcible Fondling Rate for States with Both Policies Compared to States with only Registration Policies...... 193
Figure 4-5: NIBRS Sexual Assault with an Object Rate for States with Both Policies Compared to States with only Registration Policies...... 194
Figure 4-6: NIBRS Statutory Rape Rate for States with Both Policies Compared to States with only Registration Policies...... 196
Figure 4-7: NIBRS Incest Rate for States with Both Policies Compared to States with only Registration Policies...... 198
Figure 4-8: Rates of Sexual Offending Iowa and the Timing of Policy Adoption...... 241
Figure 4-9: Rates of Sexual Offending New Hampshire and the Timing of Policy Adoption...... 242
Figure 4-10: Rates of Sexual Offending North Dakota and the Timing of Policy Adoption...... 243
Figure 4-11: Rates of Sexual Offending South Carolina and the Timing of Policy Adoption...... 244
Figure 4-12: Rates of Sexual Offending Virginia and the Timing of Policy Adoption...... 245
Figure 5-1: The Interconnectedness of Policy Adoption and Evaluation...... 259
xiii LIST OF TABLES
Table 1-1: Rape in the United States, Total and Rate per 100,000 Females, 1989-2008...... 4
Table 2-1: Expected Relationships Between Internal Determinants and Policy Adoption...... 40
Table 2-2: Expected Relationships Between Diffusion and Policy Adoption...... 53
Table 3-1: Year of Adoption of a Sex Offender Registry with Notification, by State...... 80
Table 3-2: Year of Adoption of a Civil Commitment Statute, by State...... 83
Table 3-3: Descriptive Statistics...... 86
Table 3-4:. Correlations Among Independent Variables ...... 88
Table 3-5: Specifying Time for Registration/Notification Models...... 107
Table 3-6: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Demographic Internal Determinants...... 110
Table 3-7: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Political Internal Determinants...... 111
Table 3-8: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Racial Threat/Conflict Internal Determinants...... 112
Table 3-9: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Social Disorganization Internal Determinants...... 112
Table 3-10: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Structural-Functional Internal Determinants...... 113
Table 3-11: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Internal Determinants Model...... 114
Table 3-12: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Neighboring)...... 115
Table 3-13: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Northeast Region)...... 116
Table 3-14: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Midwest Region)...... 118
xiv Table 3-15: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (South Region)...... 119
Table 3-16: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (West Region)...... 120
Table 3-17: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (National)...... 122
Table 3-18: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Physical Diffusion Characteristics...... 124
Table 3-19: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Networks (Walker). .... 125
Table 3-20: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Networks (Canon & Baum)...... 125
Table 3-21: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Federal Government...... 126
Table 3-22: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Media...... 127
Table 3-23: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Movement Organizations...... 127
Table 3-24: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Diffusion Model (Neighboring)...... 129
Table 3-25: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Diffusion Model (National)...... 130
Table 3-26: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Model...... 132
Table 3-27: Specifying Time for Civil Commitment Models...... 135
Table 3-28: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Demographic Internal Determinants...... 137
Table 3-29: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Political Internal Determinants...... 138
Table 3-30: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Racial Threat/Conflict Internal Determinants...... 139
xv Table 3-31: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Social Disorganization Internal Determinants...... 140
Table 3-32: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Structural-Functional Internal Determinants...... 140
Table 3-33: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Internal Determinants Model...... 142
Table 3-34: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Neighboring)...... 143
Table 3-35: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Northeast Region)...... 144
Table 3-36: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Midwest Region)...... 145
Table 3-37: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (South Region)...... 146
Table 3-38: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (West Region)...... 147
Table 3-39: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (National)...... 149
Table 3-40: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Physical Diffusion Characteristics ...... 150
Table 3-41: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Hub States (Walker)...... 151
Table 3-42: Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Hub States (Canon & Baum)...... 151
Table 3-43: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Federal Government...... 152
Table 3-44: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Media...... 153
Table 3-45: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Social Movement Organizations...... 154
Table 3-46: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Diffusion Model...... 155
xvi Table 3-47: Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Model...... 157
Table 4-1: All National Incident-Based Reporting System Certified States...... 179
Table 4-2: States with Greater than 80% of Crime Reported to NIBRS...... 180
Table 4-3: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor)...... 201
Table 4-4: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 202
Table 4-5: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 203
Table 4-6: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Linear Registration/Notification Policy Predictor)...... 204
Table 4-7: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor)...... 204
Table 4-8: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 205
Table 4-9: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 206
Table 4-10: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Linear Registration/Notification Policy Predictor)...... 206
Table 4-11: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor)...... 207
Table 4-12: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 207
Table 4-13: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 208
Table 4-14: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Linear Registration/Notification Policy Predictor)...... 209
Table 4-15: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor)...... 210
xvii Table 4-16: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 210
Table 4-17: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 211
Table 4-18: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Linear Registration/Notification Policy Predictor)...... 212
Table 4-19: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor)...... 213
Table 4-20: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 213
Table 4-21: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 214
Table 4-22: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Linear Registration/Notification Policy Predictor)...... 214
Table 4-23: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor)...... 215
Table 4-24: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 216
Table 4-25: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 216
Table 4-26: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Linear Registration/Notification Policy Predictor)...... 217
Table 4-27: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor)...... 217
Table 4-28: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year)...... 218
Table 4-29: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged Two Years)...... 218
xviii Table 4-30: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Linear Registration/Notification Policy Predictor)...... 219
Table 4-31: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor)...... 220
Table 4-32: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 221
Table 4-33: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 222
Table 4-34: Fixed Effects Model Predicting the Rate of Rape According to the UCR (Linear Civil Commitment Policy Predictor)...... 222
Table 4-35: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor)...... 223
Table 4-36: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 223
Table 4-37: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 224
Table 4-38: Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Linear Civil Commitment Policy Predictor)...... 225
Table 4-39: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor)...... 225
Table 4-40: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 226
Table 4-41: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 227
Table 4-42: Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Linear Civil Commitment Policy Predictor)...... 227
Table 4-43: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor)...... 228
Table 4-44: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 228
Table 4-45: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 229
xix Table 4-46: Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Linear Civil Commitment Policy Predictor)...... 229
Table 4-47: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor)...... 230
Table 4-48: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 230
Table 4-49: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 231
Table 4-50: Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Linear Civil Commitment Policy Predictor)...... 231
Table 4-51: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor)...... 232
Table 4-52: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 232
Table 4-53: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 233
Table 4-54: Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Linear Civil Commitment Policy Predictor)...... 233
Table 4-55: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor)...... 234
Table 4-56: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged One Year)...... 234
Table 4-57: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor and Covariates Lagged Two Years)...... 235
Table 4-58: Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Linear Civil Commitment Policy Predictor)...... 235
Table 4-59: Summary of Significant Effects Across Crimes...... 236
xx ACKNOWLEDGEMENTS
First and foremost, I thank my chair and mentor, Barry Ruback, for the countless hours that he has spent helping me develop into a researcher over the past five years.
Barry has helped me to cultivate multiple skills that I will need to be a successful assistant professor next year. I thank him for providing thoughtful and thorough feedback on written work, for introducing me to the grant process, and for including me as both a research assistant and a teaching assistant in his work.
This dissertation would not have been possible without the help of my committee members: Derek A. Kreager, Jeremy Staff, and Aaron L. Pincus. Their careful reading of my dissertation proposal and thoughtful comments at the proposal defense helped to shape this dissertation into a stronger final product. I appreciate their time and effort as committee members throughout the past year.
To all of the faculty members at Penn State who have helped me throughout my graduate career, especially when I was on the job market, I thank you.
Finally, I thank my family for their constant support throughout my graduate education. My biggest supporter has been my husband, Jacob. I am truly grateful for all of the kind words and encouragement that he has given me for the past five years.
1
Chapter 1
Introduction to the Research
Sexual crimes against children are so repugnant that they provoke media outrage and trigger political repercussions. These offenses are considered more heinous than virtually any other type of crime because children are perceived as the most vulnerable part of the population (Sample, 2011). Furthermore, sexuality is perceived as something so deeply personal and important that it requires protection (Kohm & Lawrence, 1997).
Because of this perception, the transition from virgin to non-virgin status is sometimes referred to as “losing one‟s innocence.” For the many Americans who endorse this belief, early sexual activity is believed to rob children of their childhoods (Kohm &
Lawrence, 1997). Sexual activity perpetrated by an adult against a child is especially invasive because of the violation of trust that is often involved. For this reason, perpetrators of sexual crimes are considered despicable, even by incarcerated offenders
(Struckman-Johnson, Struckman-Johnson, Rucker, Bumby & Donaldson, 1996).
The most extreme and disturbing cases of sexual offending against children receive inordinate media attention and are typically treated as though they are the norm
(Tonry, 2009; Terry, 2011). In addition, because sex offenders are believed to commit crimes at extremely high rates, there is a general fear that these crimes could happen to large numbers of children at any time. In response to several particularly heinous crimes that took place in the late 1980‟s and the early 1990‟s, and the fear and outrage that
2 followed, new measures were introduced in many jurisdictions to help control the sex offender population (Terry & Ackerman, 2009).
With the implementation of this new sex offender legislation, convicted sex offenders are now subject to some of the most restrictive punishments available in the
U.S. criminal justice system (e.g., residency restrictions, castration). The amount of resources being poured into initiatives intended to reduce sexual offender recidivism is enormous: as much as $100,000 per year to incarcerate a single sex offender, $7,000 per year, per offender for drugs that reduce libido (Scott & Busto, 2009), $500,000 per state to maintain an online public registry (Sample & Evans, 2009), and millions of dollars for
GPS monitoring (Meloy & Coleman, 2009). Although these punitive sanctions may serve to make communities feel safer, empirical evidence suggests that these policies may not, in fact, be serving a protective function. Moreover, a balance has yet to be struck between protection of the community and just and effective treatment of sex offenders in the criminal justice system.
This research had two goals. First, annual measures of state characteristics and legislative changes throughout the nation were included in an event history analysis to identify the factors that affect policy adoption across states. The results of this analysis indicated that states make policy decisions based on one of three possible sources of influence: (1) the characteristics of the individual state, (2) the policy decisions that other states are making, or (3) a combination of within-state characteristics and between-state interaction. Second, fixed effects analysis of panel data were used to determine whether and to what extent sex offender policies impact rates of sexual offending within states.
The results of this analysis indicated whether new legislative initiatives meet their
3 intended goal of reducing sexual offending. Taken together, these two studies provided new and important findings about the sources and usefulness of sex offender legislation in the United States.
Sexual Offenses in the United States
Offenses that are labeled “sexual offenses” vary across the 50 states, with some states including non-violent offenses such as possession of child pornography or exhibitionism. However, every state has statutes preventing rape, sexual assault, and child molestation, and these offenses are treated seriously by the criminal justice system.
Because these laws cover the most serious sexual offenses, they have also been the focus of much research. This section outlines the major findings of this line of research and explains the incidence and prevalence of sexual offending in the United States.
According to the FBI‟s Uniform Crime Reports, the rate of rape reported to the police in the United States peaked in 1992 when the national rate of forcible rape was
42.8 per 100,000 women, for a total of nearly 110,000 rapes. Since then, however, the rate of rape has decreased (see Table 1.1). Some research suggests that there has been a similar decline in “substantiated”1 child sexual abuse cases (Williams, 2009). Indeed, between 1990 and 2004, reported sexual abuse of children decreased 49% (Finkelhor &
Jones, 2006). The decline has been even larger for teenage sexual assault, which decreased 67% over the same time period. Despite this decrease, between 1993 and
1 More claims of child abuse are filed than those that are “substantiated.” For a case to be substantiated, there must be evidence that a crime did in fact take place.
4 2002, the number of incarcerated sex offenders increased by 74% compared to a general prison population increase of 49%.
Table 1.1: Rape in the United States, Total and Rate per 100,000 Females, 1989-2008. Year Forcible Rape Forcible Rape Rate* 1989 94,504 38.3 1990 102,555 41.1 1991 106,593 42.3 1992 109,062 42.8 1993 106,014 41.1 1994 102,216 39.3 1995 97,470 37.1 1996 96,252 36.3 1997 96,153 35.9 1998 93,144 34.5 1999 89,411 32.8 2000 90,178 32.0 2001 90,863 31.8 2002 95,235 33.1 2003 93,883 32.3 2004 95,089 32.4 2005 94,347 31.8 2006 92,757 31.0 2007 90,427 30.0 2008 89,000 29.3 Source: FBI Uniform Crime Reports * Rate per 100,000 women
Although there are many types of perpetrators and victims of sexual assault, there are a few characteristics that can be generalized to most sexual offense scenarios. Most sexual offenders are male, and most sexual assault victims are female (Williams, 2009).2
Furthermore, it is well-established that most sexual assaults involve a known perpetrator such as a family member, an intimate partner, or an acquaintance (Williams, 2009).
2 There is a major gap in the literature on sexual offenders, in that there are few comprehensive studies of female sexual offenders (Williams, 2009). Thus, little is known about this population and female sexual offenders are not included in this literature review.
5 Although crimes against children tend to attract the most media attention and public outrage, the demographic at greatest risk of rape or sexual assault victimization is young
(i.e., in their early twenties or younger) females (Williams, 2009).
Sexual offenses remain among the most underreported offenses (Zilney & Zilney,
2009). Certain serious crimes (e.g., homicide, auto theft) are nearly always reported.
However, because of the intimate nature of a sexual offense, many victims choose not to report their attacks. In cases in which the victims knew their offender (e.g., father- daughter incest), the victims may be unwilling to report their crime because they do not want to get their attacker into trouble with the criminal justice system. Even victims who know that they have been victimized and do not have a relationship with their rapist may still be hesitant to report their attack because of the additional stress that comes with reporting a crime to the police. To process a case through the criminal justice system, the victim will likely be required to make a statement and face difficult questions from police, prosecutors, and defense attorneys (Bryden & Lengnick, 1997).
The level of media-attention that a case is likely to attract can also affect the victim‟s willingness to report. High profile cases typically develop when a crime was particularly atrocious. The abductions of Dylan and Shasta Groene following the murder of their mother, brother, and mother‟s boyfriend in Coeur d‟Alene, Idaho in 2005, are an example. Dylan was later found murdered, and, though Shasta survived, she was molested repeatedly during her abduction. High profile cases also tend to develop when a celebrity was involved. The 2010 incidents involving NFL Hall of Fame football player Lawrence Taylor and Pittsburgh Steelers quarterback Ben Roethlisberger are examples, as was the Laker‟s star Kobe Bryant‟s case in Colorado in 2003. If the case is
6 high profile, the victim may be featured in the media, despite rape shield laws, which are intended to protect the identity of the victim. Publication of such a private incident can be extremely traumatic for the victim (Benedict, 1992; Zilney & Zilney, 2009).
Cases involving child victims pose additional, unique reasons for underreporting.
Child victims may be unsure of what has happened to them, and whether or not it was wrong. If the offender is in a position of power over the child (which is likely, if the offender is an adult), the child may be afraid to tell even his or her parents about the abuse, or the child may believe that he or she has done something to deserve this treatment. This confusion is not limited to child victims. Even adult victims may not be sure that they have been victimized, or may feel that they were somehow responsible for their victimization.
Once reported, many sexual assault cases are never fully processed through the criminal justice system. Some victims decline to press charges, or drop the charges to avoid the emotional trauma that can result from a trial (Zilney & Zilney, 2009). Some victims do want to press charges, but there is not sufficient evidence to result in a conviction, in which case the defense may argue that the sex was consensual, turning the trial into a “he-said she-said” exchange (Zilney & Zilney, 2009). Those cases that do ultimately lead to convictions can result in serious punishments for the offenders. In general, sex offenders are subject to restrictions and punishments far more severe than any other criminal group (with the exception of murderers, who can be subject to capital punishment. However, sex offenders were also subject to capital punishment prior to the
2008 Supreme Court decision, Kennedy v. Louisiana, banning this practice). A major
7 explanation for this phenomenon is that particularly heinous sexual offenses often provoke outrage in the public, which can lead to quick changes in legislation.
Memorial Laws
Laws serve a variety of purposes: to ensure public safety, to provide justice for victims (thereby reducing the risk of vigilante justice), and to express public sentiment about inappropriate behavior. Some laws are further motivated by public outrage following a particularly repulsive crime, which can serve as a galvanizing event to adopt new legislation. Oftentimes the laws that result from specific events are named for their victims, and are known as memorial laws (Zilney & Zilney, 2009). The two best known sexual victim memorial laws are the Jacob Wetterling Crimes Against Children and
Sexually Violent Predators Act (1994) and Megan‟s Law (1996), both of which pertain to sex offenders. Additional memorial laws have also been implemented to control sex offender behavior, most recently the Adam Walsh Act (2006). Brief descriptions of the cases for which each of these laws was named, and the primary purpose of the laws, follow.
Jacob Wetterling Crimes Against Children and Sexually Violent Predators Act
In 1988 in Minnesota, 11-year old Jacob Wetterling, his brother, and a friend, were biking home from a store. A stranger approached the three boys and abducted
Jacob. The boy has not been seen nor heard from since. Because his body has never
8 been found, there is no proof that he was sexually assaulted. However, the theory embraced by law enforcement is that Jacob was abducted by a sexual predator. After the attack, and as a result of the public fear and outrage that followed, Congress passed the
Jacob Wetterling Crimes Against Children and Sexually Violent Predators Act of 1994.
The Act required states, as a tool for law enforcement, to develop and maintain a centralized database of sexual offenders living in the state. Two years later, an amendment to the Act, known as Megan‟s Law, expanded the role of these databases.
Megan’s Law
In 1994, a 7-year old New Jersey girl named Megan Kanka was invited inside a neighbor‟s house to meet his puppy. Megan followed the neighbor inside the house, and later he raped and killed her. Unbeknownst to Megan and her family, the neighbor, Jesse
Timmendequas, was a repeat sexual offender, as were his two roommates. Megan
Kanka‟s mother, Maureen, has since fervently argued that if she had known that three sex offenders lived in the house across the street she would have warned her daughter to stay away from it.
The New Jersey community was shaken and “Maureen Kanka went on a crusade to change the law, arguing that registration as established by the Wetterling Act was not a sufficient form of community protection from sex offenders. Her goal was to mandate community notification” (Terry & Ackerman, 2009; page 80). Within three months of
Megan‟s death, Maureen‟s campaign was successful and New Jersey passed the first
Megan‟s Law.
9 Added as an amendment to the federal Jacob Wetterling Crimes Against Children and Sexually Violent Predators Act in 1996, Megan‟s Law requires that states notify community members when a sexual offender moves into the area. This notification, now federally required, may involve an active approach, such as the police knocking on all doors in the neighborhood to inform the residents that a sex offender has moved in to the community, or a passive approach, requiring only that, upon inquiry, police confirm that a sex offender lives in the area. The form of notification varies both between and within states based on law enforcement availability, the size of the community, and the risk posed by the offender.
Adam Walsh Act
The most recent and “one of the most comprehensive acts ever created to supervise and manage sex offenders” (Terry & Ackerman, 2009; page 91) is the federal
Adam Walsh Act, passed in 1996 and named for the son of “America‟s Most Wanted” host, John Walsh. Adam Walsh was abducted from a Florida mall in 1981, when he was
6-years old. He was found dead 16-days later, but no perpetrator was ever arrested for the crime. There are seven major components of the Adam Walsh Act: the creation of a federal DNA database, mandatory minimum sentences for certain sexual offenses, funding for GPS monitoring devices, elimination of statutes of limitations for certain offenses, authorization for child victims of sexual assault to sue their offenders in civil court, consistency in the content of online registries, and a three-tier system that characterizes offenders based on level of risk (Zilney & Zilney, 2009). The Adam Walsh
10 Act is federal and must be implemented by all states; failure to adopt the necessary provisions results in the loss of 10% of federal funding for the Omnibus Crime Control and Safe Streets Act within the state (Terry & Ackerman, 2009).
Punishments Uniquely Applied to Sex Offenders
This dissertation seeks to determine the state-level factors associated with the adoption of the two most important laws applied uniquely to sexual offenders: registration/public notification and civil commitment. These two laws were chosen because they share the same goal, but use different means to achieve it. Both public notification and civil commitment statutes are intended to protect the community.
However, public notification attempts to protect the public by increasing awareness of potentially dangerous citizens, whereas civil commitment relies solely on incapacitation of the offender.
Because this is a state-level analysis, laws that are applied by individual jurisdictions (e.g., cities, counties) and that can vary within states are not included. For example, residency restrictions and GPS-monitoring are both laws that have been applied only in some jurisdictions, most notably in Florida following the sexual assault and murder of 9-year old Jessica Lunsford in 2005. Furthermore, these laws are too new to study using panel data. Although they are becoming more common, these statutes were not widely implemented until after Lunsford‟s death in 2005 and there are still many states that have not passed legislation regarding residency restrictions or GPS-
11 monitoring. The Adam Walsh Act of 2006 provides states with funding to pursue the implementation of GPS-monitoring technology.
Castration is not included in this study because the implementation varies so greatly across the few states (n = 9) that have a statute allowing castration. Four states allow only chemical castration, and although the remaining five states allow both surgical and chemical castration, they surgically castrate only on a voluntary basis (Scott & Busto,
2009). With so few cases, and such variation in the nature and intent of these laws across states, a study of the adoption of castration laws would not be relevant to this dissertation.
Registration/Notification
Under the 1994 Jacob Wetterling Crimes Against Children and Sexually Violent
Offenders Registration Act, states were threatened with a reduction in federal support if they failed to establish and maintain a centralized database of sex offenders in the state
(Tewksbury & Lees, 2007). Two years later, in 1996, an amendment to the Wetterling
Act required states to have “procedures in place to inform the public about sex offenders who live in (the area)” (Levenson and Cotter, 2005; page 49).
The amount of information that is placed on a sex offender registry has traditionally been left to the states‟ discretion, although one goal of the Adam Walsh Act is to increase the level of consistency. In some jurisdictions, only basic information, such as name and prior offense, has been released. In other jurisdictions, the offender‟s home address, place of employment, and photograph have been included on the registry. In fact, the released information “usually includes name, age, Social Security Number, race,
12 height, weight, sex, hair, and eye color, permanent address, and working address”
(Elbogen, Patry, & Scalora, 2003; page 207).
Proponents of notification laws argue that the public cannot protect itself unless this information is made available to them (Petrosino & Petrosino, 1999). The safety of the general public is believed to outweigh the rights of the individual sex offender being forced to register and to face the embarrassment of registration/notification.
Critics of registration and public notification argue that the laws violate the due process rights of sex offenders by continuing to punish sex offenders after their prison sentence has been served. The United States Supreme Court disagreed in the 2003 case of Connecticut Department of Public Safety v. Doe, in which these laws were upheld as
Constitutional (Welchans, 2005; Levenson, D‟Amora, & Hern, 2007a). A second legal challenge to Megan‟s Law was that it violates the ex post facto clause of the Constitution because offenders are forced to register even after they have completed their prison sentence. Once again, the Supreme Court disagreed and upheld the registration and notification policies as Constitutional in the 2003 case of Smith v. Doe (Welchans, 2005;
Levenson et al., 2007a). According to the Court‟s decision, sex offenders‟ liberties are not being violated simply because their name appears on the registry (Terry & Ackerman,
2009).
A third criticism of sex offender registration is that it distracts the public from the real threats to their children – acquaintances, family members, and other known assailants
(Quinn, Forsyth, & Mullen-Quinn, 2004; Levenson et al., 2007a). By creating a false sense of security, these laws may allow parents to become complacent and miss the sexual abuse taking place in their own homes, and their children‟s schools (Avrahamian,
13 1998). Fourth, it has been argued that public reactions, such as an increased anxiety, may counteract the potential positive outcomes of making offenders‟ information public
(Elbogen et al., 2003). Finally, there is a fear that using the information available under
Megan‟s Law might lead citizens to take the law into their own hands. And, in rare cases, the public availability of personal information about sexual offenders has led to citizen vigilantism (Avrahamian, 1998).
There are also practical reasons why some individuals oppose Megan‟s Law, particularly because of the additional burden that is now placed on the criminal justice system. New laws must be enforced (e.g., maintaining the accuracy of registry information; Petrosino & Petrosino, 1999), additional hours of work are required to notify the public of sex offenders in the area, and the courts must handle all challenges to the new law. Furthermore, there is some question over how often and to what extent sex offender registries are even accessed. According to a study of Nebraska citizens (n =
1,821), although most people do know that such a registry exists, they rarely if ever access this information (Anderson & Sample, 2008). And even if they do, some research suggests that the information may not be accurate (see Levenson & Cotter, 2005, in which greater than 50% of the sample who had viewed their registry listing indicated that their information contained errors, although the nature of these errors was not reported).
Sex offenders now have a strong motivation to register – the Adam Walsh Act made “failure to register” a federal crime which entails up to 10 years in prison and a fine of up to $25,000 (Zilney & Zilney, 2009). However, research is mixed on whether or not sex offenders attempt to keep their registry entries up to date and accurate. The average respondent in a study of 121 registered sex offenders reported that he would make an
14 effort to maintain accurate personal information with law enforcement (Tewksbury,
2006). However, less than 39% of the respondents in the sample reported that they had ever viewed their personal registry entry. Importantly, two groups of sex offenders were significantly less likely to maintain accurate information than others: those who fell under lifetime registration status, and those who had been registered for at least five years. This finding suggests that perhaps the most dangerous potential offenders are those least likely to maintain accurate information. In order for Megan‟s Law to have the deterrent effect that lawmakers hope for, the registry and notification procedures must be properly implemented and maintained.
The Effect of Megan’s Law on Recidivism
The additional work of developing, implementing, and maintaining a public registry will be worth the work, according to law makers, if it reduces the risk of sexual victimization. However, research has found mixed results in whether or not Megan‟s
Law has helped to decrease sexual recidivism.
One study focused on released sex offenders in Massachusetts, a state that adopted Megan‟s Law in 1999, the final year of adoption by any state (Petrosino &
Petrosino, 1999). The authors examined recidivism in a sample of offenders (n = 136) who were convicted, served their time, and were released prior to the implementation of
Megan‟s Law. By studying the nature of the recidivism of a group who would have been subject to Megan‟s Law had they committed their crime in the present day, the authors sought to determine whether having registration/notification policies in place would have
15 led to a decrease in recidivism. In total, 36 of the 136 offenders being studied committed new crimes. However, 24 of these were attacks on known persons. Megan‟s Law is intended to prevent stranger attacks, which were perpetrated by only 12 of the 136 sex offenders under study (9%). Furthermore, 6 of the 12 stranger attacks were perpetrated outside of the area in which the offenders lived, which makes it unlikely that their victims would have been made aware of the sex offenders‟ status, even if Megan‟s Law provisions had been in place. The authors concluded that the social control impact of
Megan‟s Law is “limited” (Petrosino & Petrosino, 1999; page 154), in that according to the author‟s reasoning, 30 of the victims would still have been victimized even if
Megan‟s Law had been in place.
Using a research design involving two control groups, another study assessed the effectiveness of Megan‟s Law and found that it reduced recidivism (Duwe & Donnay,
2008). The recidivism rate of a group of registered sex offenders was compared with the recidivism rates of two different control groups: a group that was not subject to notification because they committed their offenses prior to the enactment of Megan‟s
Law3 and a group that was not required to register but had an identical propensity to sexually recidivate (Duwe & Donnay, 2008). Compared with both control groups, the registered sex offenders were less likely to recidivate, regardless of whether recidivism was measured as rearrest, reconviction, or reincarceration. The authors thus concluded
3 Megan‟s Law is applied retroactively. That is, an offender who committed a sex crime in Massachusetts in 1998 would still be forced to register, even though the state did not adopt Megan‟s Law until 1999. In their study, Duwe & Donnay‟s focused on the years that these sex offenders were released and thus able to recidivate prior to the enactment of Megan‟s Law (i.e., when they were not registered on a public registry).
16 that public registration policies decrease the risk of recidivism among sex offenders
(Duwe & Donnay, 2008).
The Effect of Megan’s Law on Re-entry
Returning to the community after a period of incarceration (“re-entry”) is a critical point in any former offender‟s life. Individuals who are going to reoffend usually do so relatively quickly (Gray, Fields, & Maxwell, 2001). Therefore, it is imperative that released offenders return to stable environments. Factors that are theoretically associated with maintaining a crime-free lifestyle include stable residence, employment, and social support (Sampson & Laub, 2005). Many of the restrictions uniquely applied to sex offenders indirectly work to undermine these anchors to a successful re-entry process.
Evidence of difficulties with re-entry is supported by a growing line of research that focuses on the effects of registration and notification on sex offenders (Welchans,
2005). A study of 183 participants at several outpatient sex offender counseling centers throughout the State of Florida attempted to identify the “collateral consequences” faced by sex offenders as a result of Megan‟s Law (Levenson & Cotter, 2005). Respondents reported that they were forced to move, were threatened, were harassed, and were victims of property damage. They also reported forms of psychological suffering, such as fear and stress. These findings were replicated in a study of the effect of Megan‟s Law on re- entry for 239 sex offenders in two states (Levenson et al., 2007a). As in the previous sample, the sex offenders reported experiencing a variety of negative consequences after they had re-entered their communities, including physical and verbal abuse of both the
17 offender and his/her family members, vandalism or other damage to the offender‟s property, and an inability to maintain employment. For some sex offenders (specifically, child molesters), the publication of their status led to eviction.
In order to assess the collateral consequences of Megan‟s Law on sex offenders, researchers have primarily relied upon qualitative research. In-depth interviews of 22 registered sex offenders in Kentucky revealed that they struggled to gain and maintain employment (Tewksbury & Lees, 2007). The respondents also reported numerous difficulties in their personal relationships, including physical and verbal harassment.
Additionally, a study of 40 sexual offenders being housed in a psychiatric hospital (some civilly committed, some voluntarily committed) revealed that the subjects believed that the release of their personal information was unfair (Elbogen et al., 2003). However, the majority of the sample admitted that they considered the registry a motivation not to recidivate.
Although the findings of these studies are consistent, the studies all used relatively small samples. The selection of the samples is also problematic. Two of the samples were nonrandom, selected from sex offenders participating in outpatient treatment programs (Levenson & Cotter, 2005; Levenson et al., 2007a). One study used a systematic sample, but obtained only a 12% response rate (Tewksbury & Lees, 2007).
Given that the respondents were selected from treatment programs, or had the motivation to complete and return a survey, it is likely that only the most motivated offenders replied. This likely bias in the survey probably means that even the individuals who are trying hardest to return to a life without crime face difficulties as a result of Megan‟s
Law.
18 In sum, there are only a few studies on the experiences of sex offenders subject to registration and notification. Although these findings should be replicated using larger, randomly selected samples, they are the best that sex offender research currently has to offer.
Civil Commitment
Civil commitment statutes, which allow for the long-term commitment of non- criminally incarcerated individuals, are not a new concept. In the 1930‟s and 1940‟s many states had “sexual psychopath laws,” which were used to indefinitely commit mentally ill offenders until they were “cured” and no longer posed a threat to society
(Sutherland, 1950). However, these laws were rarely used in the 1950‟s and 1960‟s and were phased out by the 1970‟s, as the rehabilitative model lost popularity (Janus, 2000).
Rather than perceiving sex offenders as sick and in need of treatment, “The current trend in dealing with sex offenders is toward incarceration (including longer sentences and civil commitment), as offenders cannot reoffend if they are removed from society” (Cohen &
Jeglic, 2007; page 369).
In 1988, the State of Washington saw two violent homicides committed by released sex offenders. In 1989, a 7-year old boy was sexually assaulted and mutilated by a third released sex offender. The State‟s citizens were outraged and afraid, and pressure for government action was high. As a result, Washington State passed a civil commitment statute even though the laws had fallen out of favor decades earlier. To
19 date, a total of 20 states have adopted similar policies regarding involuntary civil commitment.
The concept of civil commitment is a simple one: individuals who have completed their prison sentence but who are still believed to pose a threat to society are kept in state or federal custody until they are deemed fit to reenter society. The criminal justice system, or any other government agency cannot simply commit an individual because the agency wants to do so. The ability to involuntarily civilly commit an individual requires extenuating circumstances.
The state has two primary sources of power for removing individual rights: police power and parens patriae power (Janus, 2000). Police power is practiced through the criminal justice system and is a response to a criminal act that has already been committed. In contrast, parens patriae power is civil, and can be employed to protect an individual or society from individuals who are unable to control their own behavior and who are likely to harm themselves or others. In general, an order of civil commitment is issued when an individual meets four requirements: (1) a history of a sexual offense, (2) evidence of a mental “abnormality,” (3) evidence that the individual poses future risk, and (4) evidence that the potential for future risk is directly related to the mental abnormality (Janus, 2000).
In May 2010, the U. S. Supreme Court ruled 7-2, that it is constitutional for the federal government to civilly commit individuals (Liptak, 2010). The plaintiff in the relevant case, United States v. Comstock, was six days away from the end of his 37- month sentence for receiving child pornography when the U. S. Attorney General declared that Comstock was a danger to society and should be civilly committed. The
20 most important ruling on the involuntary civil commitment of sex offenders came in the prior Supreme Court case of Kansas v. Hendricks (1997). The plaintiff argued that the
Kansas commitment statute violated both the ex post facto and double jeopardy clauses of the Constitution. In a 5-4 ruling, the U. S. Supreme Court upheld the Kansas civil commitment statute as Constitutional because it is not punitive in nature or intent.
Based on the Hendricks decision, there are four primary differences between the early sexual psychopath laws and civil commitment as it exists today (2000). First, the state has to show only a “mental abnormality,” not evidence of mental illness. Because rates of mental illness are quite high among the incarcerated population, “this definition fails to limit civil commitment in any meaningful way” (Janus, 2000).
Second, a recent history of violent or predatory behavior is not required.
Evidence that was admitted at the criminal trial years earlier, or even acts that the offender admitted to during his or her psychological treatment while incarcerated, can be used to establish the dangerousness of the individual. Offenders are supposed to reveal all of their prior offenses, including those that have not been brought to the attention of the criminal justice system. However, these admissions are not confidential and may be used against them in determining the length of commitment. In some cases, the previously unreported crimes can result in new criminal charges (LaFond, 1998).
Third, whereas the sexual psychopath statutes provided treatment rather than incarceration, current civil commitment statutes require completion of a prison sentence prior to civil commitment. Fourth, there is no requirement that the state provide treatment to the committed individual, a fact that one researcher has called “disturbing”
(LaFond, 2000; page 167) and others have called an “unfortunate historical tendency,
21 whereby the convenience of untempered responses to alarming events overrides the consciences of both law and science” (Lucken & Latina, 2002; page 38). Overall, the goal of civil commitment statutes is to selectively incapacitate sexual offenders who are believed to present the greatest risk of recidivism if returned to the community.
The Court‟s decision in Kansas v. Hendricks paid special attention to the intent of the civil commitment statute (Carlsmith, Monahan, & Evans, 2007). If the laws are in place in order to protect society through incapacitation of a dangerous person, civil commitment is constitutional. However, if the laws are simply adding further punishment to an individual who has completed his or her punishment in the criminal justice system, then the laws are unconstitutional. In order to determine the motivation that drives citizens to recommend civil commitment, Carlsmith et al. (2007) used two samples: one consisted of 175 college students, the second consisted of 200 jury-eligible citizens. The researchers provided a vignette describing an individual who had been convicted of two sexual offenses against children. The authors manipulated the length of the offender‟s original sentence (as a measure of retribution) and his risk of recidivism
(as a measure of dangerousness). Respondents were significantly more likely to recommend civil commitment for offenders in the vignettes in which a short sentence had originally been handed down, suggesting that their motivation for applying civil commitment was based on retribution rather than public safety. The risk of recidivism significantly interacted with length of original sentences. In cases in which the original sentence was longer, the risk of recidivism had a significant positive effect on recommendations for civil commitment.
22 Proponents of civil commitment argue that removing an individual‟s liberty is worthwhile if it prevents the sexual violation of an innocent citizen. However, critics of civil commitment argue that the costs of such policies are far too high. In most states, new facilities must be built, new positions opened, and new agencies formed (LaFond,
1998). Furthermore, the judicial costs of implementing and then responding to challenges of such a law are substantial. A study by the Washington State Institute for
Public Policy found that, on average across the U.S., the annual cost for civil commitment is $97,000 per person (Gookin, 2007).
Finally, some researchers are concerned about the discretion involved in determining who should be committed, given that no completely predictive tool exists as to who will recidivate. However, two studies on the use of civil commitment in the State of Florida found that the law is being applied to the offenders who pose the greatest risk to society (Levenson, 2004; Lucken & Bales, 2008).
The first study analyzed a sample of 450 male sex offenders who were under consideration for civil commitment (Levenson, 2004). Ultimately, 229 of these offenders were recommended for civil commitment and the remaining 221 were not. Offenders who appeared to pose greater risk based on assessment instruments were significantly more likely to be recommended for civil commitment. Additionally, those who were diagnosed as antisocial or as having a paraphilia (a psychiatric disorder involving unusual sexual fantasies or behaviors) were more likely to be recommended for commitment.
Based on these results, the author concluded that the Florida officials were doing a good job of committing the individuals at greatest risk of reoffending (Levenson, 2004).
23 A follow-up study in Florida replicated and extended the previous findings
(Lucken & Bales, 2008). Using a sample of 773 male sex offenders, the authors found that the individuals at greatest risk of reoffending were significantly more likely to be referred for civil commitment. They based their assessment of recidivism risk largely on the results of various tests which are frequently used to assess recidivism risk of sex offenders.
Taken together, these studies provide strong evidence that civil commitment laws, as currently implemented in Florida, are successfully meeting their intended goal of keeping the offenders with the highest risk of recidivism under state control.
Furthermore, it does not appear that the laws are being used to keep low risk offenders under state control beyond their period of incarceration.
A Comparison of Registration/Notification and Civil Commitment Policies
Although registration/notification statutes and civil commitment statutes are quite different policies in application, their goal is the same: to protect society. Another similarity between registration/notification policies and civil commitment statutes is that both have been subject to legal challenges, as discussed above. Furthermore, both registration/notification and civil commitment statutes are implemented after the prison term to which the individual was initially sentenced is complete. In fact, according to some researchers, “sex offender legislation is unprecedented in its ability to penalize a specific type of offender after his/her judicially prescribed punishment has been served”
(Sample & Bray, 2003; page 66, original emphasis). Most sexual offending legislation
24 assumes that sex offenders cannot be treated or rehabilitated, will pose a constant threat to society, and are a relatively homogenous group.
In contrast, civil commitment statutes appear to assume that some sex offenders pose a more serious threat to society than others, a difference that warrants differential treatment of certain sex offenders. In this way civil commitment statutes are unique because they selectively incapacitate those who are believed to pose the greatest risk to society. This distinction implicitly acknowledges that a select group of sexual offenders pose a very high risk of recidivism, but that most sex offenders can be released into the community and supervised in less restrictive ways (e.g., informing law enforcement of address changes, maintaining an accurate registry entry). The fact that civil commitment statutes are less common than registration/notification statutes may be because this distinction between higher and lower risk sex offenders runs contrary to the conventional wisdom about sex offending, that all sex offenders are a threat at all times.
In sum, there are five important differences between registration/notification policies and civil commitment statutes. First, registration/notification laws are now federally required, with threat of financial penalty for states that do not comply. No federal requirement exists for civil commitment statutes. Second, as stated above, the primary goal of both of these statutes is to protect society. However, the routes that they take are quite different. Registration and public notification policies attempt to protect the public through increased awareness, whereas civil commitment statutes protect society through the incapacitation of potential offenders. Third, while sex offenders subject to registration and notification policies complain of the inconveniences that they face in the community, civilly committed sex offenders have extremely limited freedom
25 and are subject to far more restrictions than their counterparts in the community. Fourth, registered sex offenders are generally not required to participate in any treatment, but those who are civilly committed generally have access to some level of psychological treatment. Finally, registration/notification statutes appear to be based entirely on the inaccurate assumptions that sex offenders can be treated in a uniform manner, as though all pose an equal threat to society, whereas civil commitment statutes acknowledge that some sex offenders pose a greater risk than others if returned to society. The most common ideas about sex offenders include that they are not rehabilitatable and will constantly threaten society. The accuracy of these assumptions has been empirically evaluated in many ways, using a variety of samples, and a variety of analytic methods.
The findings of this line of research indicate that these assumptions are generally inaccurate (Wright, 2009).
These differences between the two statutes should be evidenced in terms of the states that have adopted the two policies. Internal determinants should have a smaller impact on the adoption of registration/notification policies, because these policies are now federally required. The role of internal determinants should be more pronounced in the adoption of civil commitment statutes, because they are not federally required and are in place in only a selection of states (n = 20). Specifically, states that have adopted civil commitment statutes should have higher incarceration rates, Republican governors, and higher violent crime rates.
26 Empirical Knowledge of Sexual Offender Behavior
Sex offenders are a major source of concern for many citizens, legislators, and law enforcement officials. The increasingly punitive treatment of sex offenders (such as sex offender registries, increasingly long sentences, and mandatory psychological interventions) are evidence of this concern (Quinn et al., 2004). The nature of this treatment, which is more severe than the treatment of many other types of offenders, is based on misconceptions about sex offenders (Fortney, Levenson, Brannon, & Baker
2007),
Reporters and politicians who have caught on to the “hot button” issue of sexual offending suggest that sex offenders are a bigger risk now than ever before. Again, empirical evidence contradicts this claim. The number of incarcerated sex offenders is not increasing as quickly as the general prison population (Turner & Rubin 2002).
Furthermore, the rate of recidivism for sexual offenders has not “changed dramatically since 1962” (Turner & Rubin 2002, page 39).
Reserving unique punishments for sex offenders implies that there is a fundamental difference between sex offenders and other criminals. Specifically, requiring them to register with the state throughout their lifetime implies that sex offenders are stable in their offending over time and cannot be rehabilitated. However, empirical evidence suggests that these assumptions are unfounded (Fortney et al., 2007).
Rather, as a group, sex offenders recidivate relatively less than other criminals (Bureau of
Justice Statistics, 2003).
27 Sex Offender Recidivism
The term “recidivism” refers to committing a new criminal act after a prior conviction. Since not every crime is reported, social scientists cannot accurately measure recidivism as commission of a new crime. Several operationalizations have been proposed, but the two most commonly used measurements are rearrest and reconviction.
However, there is a nontrivial chance of overestimating the amount of recidivism when using rearrest as a measure, because sex offenders may be arrested as early suspects but never face charges for the crime. Reconviction is a more stringent measure because it requires greater evidence than arrest that the offender did commit a crime. However, this measure is also problematic because it likely underestimates the true number of criminals who recidivate. The most responsible research choice is to measure recidivism using multiple operationalizations. In sex offender research, a distinction is typically made between sexual recidivism and nonsexual recidivism. As their names suggest, sexual recidivism is defined as committing a new sex crime and nonsexual recidivism is defined as committing a new crime that is not sexual in nature.
Both the public and sex offenders vastly overestimate the rate of recidivism by sexual offenders (Fortney et al., 2007; Levenson et al., 2007b). Sex offenders are generally less likely to recidivate than many other types of offenders (Turner & Rubin,
2002). According to a report by the Bureau of Justice Statistics (2003), 9,700 sex offenders were released from prison in 1994. Forty-three percent of released sex offenders recidivated during the three year follow-up period; whereas 68% of non-sex offenders recidivated. In terms of sexual recidivism, sex offenders were more likely than
28 non-sex offenders to recidivate, but the rates for both groups were low. Among the sex offenders, 5.3% recidivated sexually, whereas 1.3% of non-sex offenders committed a sex crime during the three year follow-up period. Although this recidivism rate may sound like a substantial portion of sex offenders, it is actually lower than the recidivism rate of many other types of criminals.
Survey respondents generally do not believe that treatment is effective in reducing recidivism (Levenson, Brannon, Fortney, & Baker, 2007b). As of now, research suggests that respondents are probably wrong; that is, that treatment does have some positive effects. A meta-analysis of 61 studies, with a cumulative total of 23,393 sexual offenders, found an average sexual recidivism rate of 13.4% (Hanson & Bussiere, 1998).
The average follow-up period was just under 5 years, which is a reasonable length of time, given that most offenders who reoffend do so quickly upon release from prison.
Studies with follow-up periods of less than 6-months received low “quality” ratings in the meta-analysis, whereas the highest “quality” ratings went to studies with follow-up periods of 10-years or longer. Rates of recidivism were higher among sex offenders who did not complete any treatment program and lower among sex offenders who did complete a treatment program (Hanson & Bussiere, 1998). The meta-analysis that found this relationship included 61 studies that were not all described in detail, so the possibility of selection bias cannot be ruled out. That is, the authors did not report whether the treatment was mandatory or voluntary. It is reasonable to expect that offenders who are motivated enough to pursue treatment are also less likely to recidivate. However, treatment reduces recidivism even when treatment is court mandated (Kruttschnitt,
Uggen, & Shelton, 2000). In 2002, Hanson did another meta-analysis, this time focused
29 on the effects of age on recidivism rates. In his examination of 10 studies (with a total of
4,673 sex offenders), Hanson found an overall recidivism rate of 17.5%.
A study of 5,331 sexual offenders on probation revealed that over three years,
30.7% of the sample recidivated (Freeman, 2007). Of the entire group, 5.5% (n = 294) sexually recidivated. Another study of sex offenders on probation (n = 917) in 17 states showed that within 3 years only 4.5% of the offenders sexually recidivated (Meloy,
2005). Sixteen percent of the sample were rearrested for any crime. The participants were less likely to recidivate when they had stable housing or “an intimate partner”
(Meloy, 2005, page 228). Given these relatively low rates of recidivism, the author suggested that placing sex offenders on probation is generally successful. Thus, allowing sex offenders to return to the community may be a much cheaper alternative to civil commitment. However, politicians and the public are hesitant to take this course of action because the potential harm even if it is experienced by only a small number of individuals, is so great.
Sex Offenders as a Homogenous Group
Sex offenders are often perceived as a homogenous group, a fact that is reflected in the laws applied to them (Sample & Bray, 2006). This assumption is incorrect, as there are important distinctions among different types of sex offenders. One of the most important of these distinctions is between child molesters and rapists (Robertiello &
Terry, 2007). According to Prentky and Lee (2007), “Rape is fundamentally predatory antisocial behavior that is subject to the same type of age-related decline observed with
30 non-sexual antisocial behavior…Child molestation, on-the-other-hand, is characterized by anomalous sexual preference…a form of sexual deviance, with persistence patterns that reflect greater longevity” (page 57). Clearly, sex offense researchers believe that there is an important distinction to be made between rapists and child molesters. Perhaps the causes of recidivism are different, and the two groups need to be dealt with separately. Treating all sex offenders as though they are pedophiles would be a mistake, if the causes of pedophilia differ from the causes of other types of sexual offending. In fact, pedophiles make up less than one half of the incarcerated sex offender population in
Pennsylvania, according to recent statistics provided by the Pennsylvania Department of
Corrections (Emery & Lategan, 2009). It does not seem appropriate to design policies based on what is probably the most feared group of sexual offenders, given that they are only one group of a larger population.
Within criminology, there is a debate over whether criminals tend to “specialize” or “generalize” in their criminal behavior. A specialist is a criminal who is active in primarily only one type of crime (e.g., a white collar criminal or a regular marijuana user). A generalist, on the other hand, engages in a variety of criminal behaviors (e.g., uses drugs, steals to support the drug habit, and fights when under the influence).
Overall, child molesters are more likely than rapists to sexually recidivate and are more likely to specialize in offending (Hanson, Scott, & Steffy, 1995). Rapists are more likely to be generalists and have higher rates of nonsexual crime. They show recidivism rates that are similar to the general population of offenders and do not seem to specialize.
Rapists resemble the general population of criminal offenders far more than child molesters. In Hanson and Bussiere‟s meta-analysis (1998), the average rate of non-
31 sexual violent recidivism for the sample was 12.2%, but there was a significant difference between rapists and child molesters. Rapists showed a nonsexual violent recidivism rate of 22.1% whereas child molesters showed a nonsexual violent recidivism rate of only
9.9%. This finding supports the conclusion that child molesters are more specific and tend to specialize in one area of offending. Moreover, child molesters are more likely than rapists to reoffend sexually than nonsexually (Doren 2006; Hanson 2002).
Compared to non-sexual offenders, child molesters are significantly more likely to be married, older, have lower educational attainment, and have a shorter criminal history (Hanson et al., 1995). Among child molesters, certain commonalities exist, including poorly developed relationships and poorly developed social skills (Robertiello
& Terry, 2007). However, poor self-image is also a major issue for most child molesters who tend to suffer from very low self-esteem (Robertiello & Terry, 2007). Moreover, there are different types of sexual offenders and it is erroneous to treat rapists, child molesters, and other sexual offenders as one homogenous group.
Research Questions
The empirical findings outlined above suggest that the assumptions upon which lawmakers shape policy (namely, that sex offenders are sure to reoffend and are a homogenous group) are inaccurate. But, if they are not based in fact, where do these laws come from? This dissertation attempts to identify the factors that affect the adoption of sex offender legislation and the legislation that affects rates of sexual offending.
32 In Study 1, discrete-time event history analysis is used to predict the adoption of specific legislation: sex offender registry/notification policies and involuntary civil commitment statutes. For both policies, two types of factors are empirically tested: state- level factors believed to affect adoption (internal determinants), and the effect of other states‟ legislative activity (diffusion).
In Study 2, data from the Uniform Crime Reports and the National Incident-Based
Reporting System is used to identify changes in the prevalence and incidence of various sexual offenses over time. Specifically, this study attempts to determine whether the policies examined in Study 1 (registration/notification and civil commitment) have deterrent effects on sexual offending (in Study 2).
33
Chapter 2
Theoretical Framework
Although the Adam Walsh Act (2006) has made sexual offense laws more consistent across states, there are still substantial discrepancies in terms of the number and nature of statutes, the activities that they prohibit, and the punishments that they prescribe. But what can account for these differences? It is very unlikely that some acts are more dangerous or more reprehensible in one state than they are in another, yet states reach different conclusions about how to best control sexual offending. In order to address cross-state differences in sex offender legislation, this study looks at the adoption of two specific sex offender policies: registries requiring public notification and civil commitment statutes.
As noted in the previous chapter, the passing of Megan‟s Law in 1996 required the implementation of public sex offender registries. However, prior to this federal mandate, 31 states introduced the legislation (19 others waited until after the passing of the federal statute). Civil commitment statutes are much more controversial and, to date, are in effect in only 20 states. The first state to adopt a modern civil commitment statute was Illinois, in 1988. Most recently, in 2007, New York and New Hampshire adopted the legislation. The differences in the timing of policy adoption are directly related to the first goal of this research. Using these two sexual offense statutes as the outcomes of
34 interest, this research examines the factors that affect whether and when states adopt sexual offender policies.
There are both practical and political reasons for passing laws. For instance, if a neighboring state were to pass a law that required sex offenders to publicly register and the state of interest did not, it is possible that sex offenders would move to the state where they were not required to register. Thus, it would serve a practical purpose for the state where the sex offenders were beginning to take up residence to pass similar legislation, so that the state did not become a haven for sex offenders. From a political standpoint, allowing sex offenders from other states to move in to the state of interest would be unpopular, and citizens would likely be unhappy.
In 1990, the State of Washington, passed what was then the nation‟s most severe sexual offense legislation as a response to moral outrage among the citizenry. In the two years prior to the adoption of this legislation, three heinous sexual offenses had been committed by released sex offenders, and the citizens, political leaders, and lawmakers, used these new laws to express their outrage at the situation. This legislation reflected both the practical goal of incarcerating sex offenders and the political goal of responding to the public‟s concerns.
However, to argue that laws are passed for both practical and political purposes oversimplifies the process of lawmaking. In political science, a literature on policy adoption has existed for many years, but has advanced substantially in the 20 years following the work of Berry and Berry (1990). Their research, and much of the research that it spawned, is described in terms of the two major theoretical approaches to understanding how public policies are adopted.
35 An Introduction to the Theories of Policy Adoption
In the literature on policy adoption, there are two points-of-view as to how changes in legislation come about. The first explanation, that changes in policy are based on factors within the state, has been called an “intrastate approach” (Soule & Zylan,
1997), “hierarchical diffusion” (Cohen & Tita, 1999), and most commonly, “internal determinants” (Berry & Berry, 1990). The second explanation, that the actions being taken by other states affect the policies adopted within a state, has been called an
“interstate approach” (Soule & Zylan, 1997), “contagious diffusion” (Cohen & Tita,
1999), and most commonly, “diffusion” (Berry & Berry, 1990). Berry and Berry (1990) made the case that both of these major explanations should be tested simultaneously, and there is empirical support for this argument (Newmark 2002). A hybrid model, incorporating both internal determinants and diffusion is essentially the state of the art in policy adoption (Buckley & Westerland, 2004).
The types of policies being analyzed affect the relative importance of factors leading to the policy‟s adoption. For instance, a controversial statute (e.g., one decriminalizing sodomy) will likely be affected by the religious composition of a state, because conservative Christians are likely to oppose sodomy on religious grounds (Kane,
2003). A statute that is less controversial may not be affected by demographic or political characteristics. The adoption of a sex offender registry is one such law, because few citizens or lawmakers oppose statutes that are intended to protect the public
(especially children) from sexual assault, which is the primary stated purpose of publicly available registries.
36 Internal Determinants Theory
Internal Determinants models of policy adoption assume that states operate autonomously and are not affected by the legislative actions taken by other states or jurisdictions. States make different policy choices based upon their own unique and specific needs (e.g., demographic characteristics, political makeup). In this research, five types of internal determinants are tested: demographic characteristics, political composition, the presence of a racial threat, social disorganization, and the structural- functional needs of the state.
In 2010, Arizona passed a law requiring that all aliens carry documentation with them at all times and allowing police to question anyone whom they suspect is not in the
United States legally. This law illustrates how multiple types of internal determinants can affect the types of policy decisions made in a state. First, the demographics of the
State of Arizona made it a likely place for this type of legislation to be enacted. Because the state has a large Hispanic population (nearly 30%, according to the U.S. Census
Bureau) and a comparatively large population of Mexican immigrants, it is an appealing destination for Mexicans entering the country (legally or illegally).
Second, the political characteristics of the State made it possible for this law to be passed. Both Arizona‟s House and Senate are controlled by the Republican party, and the
Governor of Arizona is a Republican. These characteristics of the electorate indicate the generally conservative nature of the state (a fact which was also evidenced 1992, when
Arizona began observance of the Martin Luther King, Jr., Day holiday only after being
37 faced with a boycott by professional conventions, academic conferences, and athletic teams that refused to hold their events in the State).
Third, the structural-functional characteristics of the State made politicians and many citizens believe that this legislation was necessary. If illegal immigration was not a problem in Arizona, this law would serve only a symbolic purpose. However, because of its proximity to Mexico, and the fact that the border between Mexico and Arizona currently has the most illegal crossings of any area in the country, the State of Arizona faces a unique challenge in responding to illegal immigration in ways that other states do not.
The controversial immigration legislation in Arizona is a current example of policy adoption that can show, in general terms, the effects of multiple state-level characteristics. Beyond their intuitive appeal, however, there is empirical support for each of these five types of internal determinants as explanations for policy adoption.
Demographic Internal Determinants
The population of a state generally has an effect on the types of laws enacted there. It has long been established that factors such as race, class, and education affect voting behavior. Specifically, whites, people of higher social classes, and people with more education are likely to be Republican (Knoke & Hout, 1974). As noted previously, different variables are expected to affect different types of policies. I did not expect population, population density, and per capita income to be significantly related to the likelihood of adopting a registration/notification policy. Few citizens are upset by the
38 idea of a public registry, and therefore these indicators would be unlikely to affect
(positively or negatively) its implementation (see Table 2.1 for a list of the expected relationships between the independent variables and the adoption of sex offender legislation). The same should not be true of civil commitment statutes because these laws are more controversial.
For two reasons, I expected that larger states and densely populated states would be more likely to adopt civil commitment statutes sooner than states with smaller populations or that were less densely populated. First, it is possible that states with larger populations and higher population densities are more reactive to crime within their states.
As an example, Florida has been a source of much innovation in sex offender legislation
(GPS monitoring and residency restrictions following the murder of Jessica Lunsford).
Although it is not an extremely large state, New Jersey, where the original Megan‟s Law was passed, is densely populated and contains crime-ridden urban areas such as Newark.
Second, the size of the population increases the pool of potential victims, and the population density increases the risk of potential perpetrators and potential victims coming into contact. By increasing the number of potential victimizations in a state, I expected that population and population density would be significantly and positively related to the risk and timing of adopting a civil commitment statute.
Per capita income was included as a control variable for policy adoption because the policies being studied here were costly to implement. Thus, it was possible that a state‟s financial situation may have impacted their likelihood of adopting these policies, such that states in better economic shape would be more likely to adopt them.
39 The final set of indicators related to demographic internal determinants was a set of dummy variables representing region of the country. The Southern region was treated as the reference category. I expected that the Western region would be at a significantly higher risk of adopting both registration/notification and civil commitment statutes because there were a number of important sexual offending cases that took place in the
West and led to legislative changes there (i.e., three violent assaults committed by released sex offenders between 1988 and 1989). I expected that being in the Northeast region would be significantly and positively related to the adoption of a registration/notification policy because these laws originated in New Jersey with the case of Megan Kanka. To be clear, I expected that being in the Northeast would be positively associated with the adoption of a registration/notification policy but not significantly associated with the adoption of a civil commitment statute. This expectation of differing effects for registration/notification policies and civil commitment statutes was the only expected difference based on region.
40
Table 2.1: Expected Relationships Between Internal Determinants and Policy Adoption Expected Impact on the Adoption of Variable Registration/Notification Civil Commitment Internal Determinants Demographic Internal Determinants Population 0 + Population Density 0 + Per Capita Income 0 0 Region Northeast + 0 Midwest 0 0 West + + Political Internal Determinants Republican Governor + + Percentage Republican + + Percentage Females in the Legislature 0 0 Professionalism 0 0 Competition 0 0 Innovation 0 0 Racial Threat/Conflict Hispanic + + Black + + Unemployment Rate + + Social Disorganization Divorce Rate + + Percentage Living in Poverty + + Rate of Females in the Labor Force + + Structural-Functional Internal Determinants Murder Rate + + Incarceration Rate + + Death Penalty + + + a positive effect was expected – a negative effect was expected 0 no effect was expected
Other demographic variables that have been shown to affect policy in empirical research were included in this research as controls. A study of the adoption of state lotteries showed support for the importance of demographic internal determinants in that it found that states with lower levels of per capita income were less likely to adopt state
41 lotteries (Berry & Berry, 1990). The authors attributed this finding to the fact that individuals with less disposable income were probably perceived by legislators as less likely to purchase lottery tickets. In a study of hate crime legislation, states with higher per capita incomes were found to adopt hate crime policies more quickly than states with lower per capita incomes (Soule & Earle, 2001). This finding can likely be explained by the fact that having a higher income is generally reflective of having a higher education, and more educated people are generally less prejudiced.
Support for the importance of demographic internal determinants has also been found in a study of worker‟s compensation laws (Fishback & Kantor, 1998). In general, states with more workers in the manufacturing sector were significantly more likely to adopt the laws sooner. In a study of the adoption of policies decriminalizing sodomy laws (Kane, 2003), states with higher percentages of conservative Protestants in the population were significantly less likely to decriminalize sodomy. This finding is consistent with the conservative position opposing gay-rights (Kane, 2003). Overall, characteristics of the population of a state have an effect on the types of policies adopted there and must be included in any model attempting to explain changes in policy.
Political Internal Determinants
In general, the political structure of American society is a continuum with two major parties serving as the poles: Democrats and Republicans. Democrats tend to support government involvement in the economy, protecting the environment, the separation of church and state, and social welfare programs. Republicans tend to oppose
42 government involvement in business, and they support lower taxes, fewer social welfare programs and increases in military funding (Hershey, 2005). Because Republicans tend to support more punitive policies, I expected that states with a larger percentage of the population identifying themselves as Republican and states with Republican governors, would be more likely to adopt the policies sooner than states with a smaller percentage of the population identifying themselves as Republican and states with Democratic governors.
It is possible, if traditional gender norms hold, that females would be more receptive to policies intended to protect (primarily) women and children. Thus, the percentage of females in the legislature was included as a control variable. Three additional political internal determinants were included as controls: the levels of legislative professionalism, legislative innovation, and political competition within a state. These variables may impact the adoption of new policies because they characterize the legislative atmosphere. Thus, it is important to account for these potential differences.
There is empirical evidence to support the need to include political characteristics in a study of policy adoption. For instance, a study of the adoption of hate crime legislation showed that states with higher percentages of liberal voters were likely to adopt hate crime legislation sooner than states with lower percentages of liberal voters
(Grattet, Jenness, & Curry, 1998). This finding is consistent with the liberal ideology, which focuses on civil rights (Soule & Earl, 2001). However, this effect was reduced to non-significance once the state‟s level of innovativeness (in terms of original legislation
43 put forth) was accounted for. This finding likely means that the same underlying variable is associated with both the percentage of liberal voters and a state‟s level of innovation.
In another study of the adoption of hate crime legislation, Soule and Earl (2001) replicated Grattet et al.‟s (1998) findings that once other variables were controlled, the percentage of liberal voters did not significantly affect the adoption of hate crime legislation. More importantly, they replicated the finding that more innovative states were also quicker to adopt hate crime legislation. Furthermore, they found that states with a higher percentage of Democrats in the legislature were significantly more likely to pass the laws quickly, compared to states with lower percentages of Democrats in the legislature. In sum, the political makeup of a state‟s citizens has an impact on the type of legislation pursued and ultimately implemented.
Racial Threat/Conflict Internal Determinants
Racial Threat theory is based on a conflict theory of sociology. Conflict theorists argue that when resources are scarce (e.g., employment), there will be competition for them. However, because of power differentials in society, the terms of the competition for resources may be manipulated. If the majority population (typically white) feels threatened by an increasingly large minority population, they may implement new or additional formal social controls to attempt to maintain a position of power.
Extreme situations characterized by racial threat theory include factors that were present in the Jim Crow laws of the South. Certain jobs, residences, and schools were available only to white citizens. Separate, and typically inferior, jobs, residences, and
44 schools were available to black citizens, the racial minority. Although such blatant discrimination is no longer present in the U.S., there are still some arenas in which racial conflict may be present.
The criminal justice system is one area in which the presence of a racial threat is associated with the level of social control. An advocate of the racial threat position would expect states with higher minority populations and higher rates of unemployment to have more formal social controls in place, including sex offender legislation. Prior research indicates that higher minority populations are associated with more police officers (Kent & Jacobs, 2005) and an increased likelihood of allowing the death penalty
(Jacobs & Carmichael, 2002).
The vast majority of sex offenders are Caucasian. However, if previous findings regarding criminal justice reactions to the presence of a racial threat hold, I would expect states with higher percentages of racial minorities and higher unemployment rates to be more likely to adopt sex offender policies sooner.
Social Disorganization Internal Determinants
A classic theory in criminology, social disorganization theory, focuses on macro- level variables rather than micro-level variables. Certain characteristics are associated with higher levels of social control. Traditionally, tests of social disorganization include measures of three social phenomenon: racial/ethnic heterogeneity, residential instability, and poverty. Informal social control is established when residents of a community are invested in their community. This investment is most likely to occur when residents
45 share common backgrounds (low racial/ethnic heterogeneity), they own homes or reside in the area for an extended period (low residential instability), and have the financial security to focus on things outside of their own homes (low poverty).
Residents are able to maintain the social norms expected in their community through a process called collective efficacy. This process is characterized by working together to speak up when inappropriate behavior is taking place (e.g., children skipping school, loitering in the neighborhood) and monitor both public and private property within the neighborhood.
An important measure of social disorganization is the role of women in the home.
The rate of female headed households is commonly used as an indicator of social disorganization. The logic behind this measure is that in areas in which traditional nuclear families are present and females stay at home, informal relationships develop and informal social control is present. Female residents are home during the day, can monitor the area, and can ensure that all behavior taking place is consistent with the expected behavior in the area. Based on a social disorganization perspective, I expected that states with higher levels of poverty, divorce, and females in the labor force would have higher levels of formal social control (i.e., registration/notification and civil commitment policies) because the levels of informal social control would not be sufficient to maintain the expected standards of behavior.
46 Structural-Functional Internal Determinants
Structural-functional internal determinants refer to those factors that are closely related to the outcome of interest. The logic of a structural functional approach is that society is organized around a consensus of values (Liska, 1987). Thus, laws are implemented as a response to problems within society. For instance, a state with an unemployment rate of zero would have little need to implement policies addressing unemployment. Therefore, a structural-functional theorist would not expect that state to adopt policies relating to unemployment. As a more specific example, the state legislature in Pennsylvania is currently considering legislation to outlaw certain chemical compounds that are taken from bath salts and used to create a substance with similar properties to methamphetamine (Adler, 2011). Only recently has abuse of this substance
(called “blizzard”) become a problem in the State, and the State is responding by trying to adopt policies to regulate the substance (Bock, 2011). In other states, in which this substance abuse is not taking place, or is taking place on only a very small scale, such legislation has not been drafted.
According to the logic of structural-functionalism, a state with a high rate of sexual offending is more likely than a state with a low rate of sexual offending to adopt policies intended to decrease sexual offending. Consistent with this logic, states with higher rates of violent crime and generally punitive criminal justice policies were expected to respond more quickly by adopting these laws compared to states with lower rates of violent crime and less punitive criminal justice policies.
47 Berry and Berry (1990) found support for structural-functional theory in their study of state lottery adoption. States with poorer “fiscal health” were substantially (but not significantly) more likely to adopt lotteries, an initiative that provides financial incentives for the state. Fishback and Kantor‟s (1998) study of the adoption of worker‟s compensation laws also found support for a structural-functional argument. They found that in states where employers faced high liability for accidents on the job, worker‟s compensation laws were more likely to be enacted sooner. That is, in states that were already dealing with issues involving employer liability, the addition of worker‟s compensation benefits were more readily adopted.
Policy changes in the Aid to Families with Dependent Children (AFDC) were also related to structural-functional variables, according to an event history analysis (Zylan &
Soule, 2000). States with the highest rates of AFDC use and states that provided the highest amount of benefits were the states that first attempted to change the structure of their welfare programs. Because these states were facing financial difficulties related to
AFDC, legislators were most likely to change their policies, and quickly.
The Strength and Weakness of Internal Determinants Theory
The strength of internal determinants theory is that it acknowledges the importance of the characteristics of each individual state as predictors of policy innovation. Differences in the demographic characteristics, political characteristics, presence of a racial threat, level of social disorganization, and structural-functional characteristics of a state lead to different policy choices. California and Nevada, for
48 instance, are neighboring states that have important differences in population density, urbanicity, economic policy, and various other characteristics that would likely lead them to make distinct policy decisions. An internal determinants theorist would expect
California‟s and Nevada‟s state legislatures to reach different policy decisions based on these differences.
However, the strength of the theory is directly related to its weakness. Internal
Determinants theory fails to consider the public and political pressure placed on a state when surrounding states have adopted some policy that it has not. Diffusion theory, conversely, focuses on the effects of other states‟ legislative processes on the state being analyzed.
Diffusion Theory
Simply put, diffusion theory assumes that states‟ legislative activity is based on the legislative activity taking place in other states. The foundation of diffusion as a theory of policy adoption is a factor analysis conducted forty-two years ago (Walker,
1969). Eighty-eight policies that were in place in at least 20 states by the year 1965 composed the sample, and included various types of legislation (from civil rights to taxes to conservation). By studying whether and when each state passed each of the 88 policies, Walker assigned each state an “innovation” score. A factor analysis of these innovation scores suggested that policy behavior was regionally clustered. That is, states that were located within the same region were more similar to each other than to states in other regions in terms of whether and when they adopted a variety of policies. These
49 findings indicate that location and interaction among states, not just the states‟ individual characteristics impact policy adoption.
If surrounding states have adopted some policy, the state being analyzed is likely to adopt that policy, as well. Of course, states do not blindly accept policies just because other states have adopted them. Rather, lawmakers watch the actions being taken by their counterparts in other states when deciding which states to emulate (Mooney, 2001).
When a policy is implemented in one state and it is highly successful, other states are more likely to pass similar legislation (Volden, 2006). Lotteries, which provide revenue within a state, are an example of a policy that would likely be emulated (Berry & Berry,
1990). The 2010 change to immigration policy in Arizona led to numerous protests, criticism from the President, and economic boycotts of the state (Montopoli, 2010).
Because of these negative consequences, this policy is an example of one that is unlikely to diffuse to other states.
The likelihood of diffusion to different states is believed to be related to the proximity of states. However, in the current age of technology, physical proximity has become less salient, presenting the possibility that state legislators may interact more regularly with states elsewhere throughout the nation than they do with legislators in their neighboring states. This concept has been explored by sociologists who study the role of interaction within social networks.
50 Social Networks
Social networks are composed of all of the people (or in this research, states) in a sample universe. Although each member of the network operates as a single unit, the impact that some members have on the network is stronger than others. In a study of the spread of a prescription drug, researchers found that when doctors who were friends with a higher proportion of the other doctors in the sample (i.e., links in the network) began prescribing the drug, so did many of the other members of the social network (Barabási,
2002). The doctors with greater influence on the social network, are known as “hubs.”
They are connected with many other doctors, and serve as conduits for the flow of information (Barabási, 2002). The network of states in the nation operates in much the same way, with certain states serving as hubs in the network. When California or New
York makes a legislative change, it is likely to impact more states than when Wyoming or
Idaho makes a legislative change (for example, see Volden‟s 2006 study of the children‟s health insurance program).
The study of social networks has also revealed a trend in the adoption of innovations, including policies: a histogram representing the timing of innovation adoption takes the approximate shape of a normal curve (Barabási, 2002). First, only a few members of the network adopt the innovation. As time passes, additional members adopt the innovation, as well, forming the highest part of the curve. Finally, there are only a few members left who have yet to adopt the policy, and when they do, they make up the final tail of the histogram.
51 Research on social networks has been a valuable addition to the study of innovation and diffusion. Some researchers have expanded on the concepts of networks by introducing “neo-institutionalism” into sociology.
Neo-Institutionalism
Neo-institutionalism focuses on interactions among micro-level actors in macro- level social institutions (Nee, 1998). For example, and relevant to this research, legislatures are composed of individual legislators who bring their own backgrounds, points of view, and pressures from constituents. However, legislatures are also nested within states, impacted by powerful interest groups, pressured by political leaders, bound to legal precedent, and subject ultimately to being overruled by the federal government.
Thus, multiple levels of study and multiple sources of information should be considered in a study of policy diffusion.
The underlying logic of neo-institutionalism is consistent with economic theory, and focuses on the “concept of choice within constraints” (Nee, 1998; page 8). Among the possible choices, actors are assumed to be rational in their decision making: actors will work to minimize pain or cost, and to maximize pleasure or benefit (Nee & Ingram,
1998). Although actors may possibly pursue an infinite number of actions, there are only certain actions that are actually normative within the social network (i.e., consistent with the expectations of the group). Furthermore, the fact that the social network is embedded within other social organizations (e.g., accountable to the federal government) provides additional constraints on possible outcomes. Only the courses of action that are both
52 normative within the social network and consistent with the expectations of the institutions in which the network is embedded are likely to be pursued.
This study tested the notion of diffusion using several different approaches. First, the traditional concept of diffusion was operationalized in three ways: neighboring states
(i.e., share a border), states within a region, and states within the nation. Second, additional diffusion measures, or diffusion processes, inspired by the social network and neo-institutionalism literature, were also included. These additional measures included physical characteristics of the state, the role of hub states, the role of the federal government, the role of media coverage, and the presence of social movement organizations to help explain the physical and temporal process of diffusion.
Neighboring States
The first way to operationalize diffusion is to measure the percentage of neighboring states that have adopted some policy. Under this measurement, it was expected that as the percentage of bordering states that had adopted the policy increased, the likelihood that the state being analyzed would adopt the policy increased as well (see
Table 2.2). If the state that had not yet adopted the policy continues to disregard the policy, there may be negative effects from the other jurisdictions that cause lawmakers to reconsider and ultimately pass the policy.
53
Table 2.2: Expected Relationships Between Diffusion and Policy Adoption Expected Impact on the Adoption of Variable Registration/Notification Civil Commitment Traditional Diffusion Neighboring X X Regional X X National X X Physical Diffusion Length of Border X X Shared Border X X Number of Neighbors X X Interstates + + Airport Hub + + Federal Pressure Hub State (Walker) + + Hub State (Canon & Baum) + + Federal Pressure Jacob Wetterling Act + 0 Megan‟s Law + 0 Kansas v. Hendricks 0 + Adam Walsh Act 0 + Media Media Coverage + + Social Movement Organizations “Association for the Treatment of – – Sexual Abusers” “Stop it Now!” + – + a positive effect is expected – a negative effect is expected 0 no effect is expected X effect is expected to be consistent with neighbor(s) (i.e., adoption if other have adopted and no adoption if others have not adopted)
Although this analysis focuses on diffusion at the state-level, the theory of diffusion can be applied to any jurisdiction, such as counties. For instance, multiple counties in Florida adopted residency restrictions for sex offenders after the rape and murder of 9-year old Jessica Lunsford in 2005. In response, many of the surrounding counties adopted residency restrictions as well. Had they not, lawmakers may have left
54 their counties open to an influx of sex offenders, who could live more freely in their counties compared to counties that had adopted residency restrictions. No mayor, legislator, politician, or other elected official wants to appear soft on crime. Thus, it is common for jurisdictions to adopt more restrictive policies if other jurisdictions are also doing so.
In their study of lottery adoptions, Berry and Berry (1990) found support for diffusion to neighboring states. As the percentage of surrounding states that had adopted a lottery increased, so did the state‟s likelihood of adopting a lottery.
The problem with this operationalization is that certain states (e.g., Maine) are quite isolated from their “neighbors,” so such a measure would not hold much meaning.
However, an alternate operationalization of diffusion, regional diffusion, addresses this issue.
States within a Region
A second way to operationalize diffusion is to treat each state as a part of a region, and to track the percentage of states in the region that have adopted the policy. It was expected that as the percentage of states within the region that had adopted the policy increased, the likelihood that the state being analyzed would adopt the policy also increased. In this research, the term region refers to a large geographic area (i.e.,
Northeast, Midwest, South, and West) that shares similar characteristics, including geographic similarities, historical background (e.g., slavery in the South, the rise of industry in the Northeast), agriculture, and climate.
55 Support for regional diffusion was found in Fishback and Kantor‟s (1998) study of worker‟s compensation laws. In what the researchers call a “contagion effect,” the more states within the region that had adopted worker‟s compensation laws, the more likely that the state being analyzed was to adopt such a law. Grattet et al. (1998) also found support for regional diffusion when they measured the contagion effect using the nine census regions. In a contrary finding, Soule and Earl (1998), found that the number of states in the region that had adopted a hate crime policy was negatively related to the state‟s likelihood of adopting such a policy. This finding is inconsistent with the majority of the policy literature, which generally supports regional diffusion. This finding is likely explained by the fact that lawmakers in a nearby state who witness the adoption of a policy that they perceive as “threatening, misguided, or irrational” (page 298), may be especially unlikely to adopt a similar policy (Soule & Earl, 1998). Thus, the effect of regional diffusion appears to be moderated by the perceived success of the policy.
States within the Nation
Additionally, two studies of the diffusion of state hate crime legislation found that the number of states throughout the nation at large (not just within the region) had an effect on policy adoption (Grattet et al., 1998; Fishback & Kantor, 1998). The authors explained this finding by arguing that as more and more states adopt some policy, the likelihood that the remaining states will adopt the policy increases, regardless of their regional location. Based on these findings, it was expected that as the percentage of
56 states in the nation that had adopted the policy increased, the likelihood that the state being analyzed would adopt the policy increased, as well.
Therefore, a comprehensive test of diffusion theory should allow for diffusion at three levels: neighboring states (they share a border), states within a region, and all states in the nation.
Physical Diffusion
Diffusion is both a temporal and physical process. The actual sharing of information is a physical process that requires contact between individuals or organizations. Despite the increased use of telephones and the internet for business purposes, it is still expected that people are more interested in the policy choices of areas nearer to them than the areas farther away from them. The amount of interaction between states is not entirely dependent on proximity, but Tobler‟s first rule of geography states that, “Everything is related to everything else, but near things are more related than distant things” (1970). Thus, the physical connectedness of states may be an important indicator of diffusion. Measures of the physical connectedness of states that are included in this research were the number of neighboring states, the length of a state‟s shared border, the percentage of the state‟s border that is shared with another state, the number of interstates running through a state, and whether a major airport hub is present within the state.
I expected that states that were more “connected” to other states (e..g, greater percentage of border shared) would be more likely to emulate the policy choices of their
57 neighbors. States with greater levels of connectedness within the nation (e.g., interstates, airport hubs) were expected to adopt the policies sooner than states with lower levels of connectedness within the nation.
The Strengths and Weaknesses of Diffusion Theory
The primary strength of diffusion theory is that it focuses on what effect the actions that some states have on the actions subsequently taken by other states. However, a weakness of diffusion theory is that it largely ignores the unique characteristics of individual states and relies only on the actions of other states to explain changes in policy over time. Moreover, the processes of diffusion have been underexplored. As described above, most research measures diffusion using the percentage of states (neighboring, in the region, or in the nation) that have adopted the policy. Though the percentage of states in the defined area that have adopted the policy is an important variable, it is unclear precisely how the process of diffusion operates.
Diffusion Processes
This research sought to explore specific diffusion processes by testing various mechanisms that may explain how information is spread or shared across states. First, the role of the state social network was assessed by measuring the impact of hub states on policy adoption. Second, four specific sources of pressure that may contribute to
58 diffusion across states were tested: status as a hub state in the social network, the Federal government, media coverage, and social movement organizations.
Hub States
States do not have equal influence on each other. In his factor analysis of 88 policies, Walker identified certain states as “hub” states. These hub states were typically among the first to adopt a new policy and were “innovative” in terms of creating new policies. Based on his 1969 factor analysis, the hub states included New York (.66),
Massachusetts (.63), California (.60), New Jersey (.59), Michigan (.58), and Connecticut
(.57).
More recent research continues to indicate the importance of a state‟s level of innovation in determining his likelihood of adopting a policy. A study of hate crime laws found that states with higher ratings of “innovativeness” in legislation were likely to adopt legislation earlier than states with lower ratings of “innovativeness” (Grattet, et al.,
1998). The state‟s level of innovativeness was assessed using two different scales (one measuring civil rights innovation, the other measuring general policy innovation).
The present research uses a scale measuring judicial policy innovation, created by
Canon and Baum (1981). Consistent with Walker‟s methodology, the authors used a factor analysis of 23 reforms to tort law between 1876 and 1975 to produce factor scores for each of the 50 states. The states with the highest innovation scores, considered hub states, included Minnesota (.51), Texas (.46), Kentucky (.46), Washington (.43),
California (.42), and Missouri (.40).
59 I expected that the more innovative states identified by Walker (1969) and Canon and Baum (1981) would be likely to adopt sex offender legislation sooner than states that were not identified as hubs of innovation.
National Pressure
Given that all 50 states are united under one federal government, it makes sense that pressure from the federal government also affects the likelihood and timing of policy adoption by mandating a policy and threatening states with a financial loss if they fail to implement it. Megan‟s Law, the memorial law discussed earlier, is an example of the use of such coercive methods. Overall, federal pressure may affect the likelihood or timing of policy adoption. Three Federal laws (the Jacob Wetterling Act, Megan‟s Law, and the
Adam Walsh Act) and one Supreme Court case (Kansas v. Hendricks) were passed during the time period of interest in this research and were sources of Federal pressure to adopt certain policies. The role of these four federal actions are included in the present research.
I expected that the passage of the Jacob Wetterling Act and Megan‟s Law would be positively associated with the timing of adopting registration/notification policies because these two laws were directly related to such policies. Recall that the Jacob
Wetterling Act (1994) required states to create registries of information of all sex offenders in the state. Megan‟s Law (1996) expanded the role of these registries by requiring states to make them publicly available. Thus, after the adoption of the Jacob
60 Wetterling Act and Megan‟s Law, I expected that the risk of adopting a registration/notification policy would increase in the remaining states.
The Kansas v. Hendricks ruling dealt specifically with civil commitment, so I did not expect that it would impact states‟ likelihood of adopting registration/notification policies. The event history analysis of the timing of adoption of registration/notification policies ended in 1999, as the final four states adopted this policy. Thus, no relationship between the Adam Walsh Act (passed in 2006) and adopting a registration/notification policy was hypothesized.
Because they dealt specifically with sex offender registries, I did not expect a relationship between either the Jacob Wetterling Act or Megan‟s Law and the timing of adopting a civil commitment statute. However, I did expect that the Kansas v. Hendricks ruling would be positively associated with the timing of adopting civil commitment statutes. That is, after the Supreme Court ruling that civil commitment at the state level was Constitutional, I expected that the remaining states would be at greater risk of adopting civil commitment statutes. The Adam Walsh Act, though it did not apply directly to civil commitment, represented a Federally-sanctioned punitive shift in the legal treatment of sex offenders. Thus, I expected that remaining states would be at an increased risk of adopting civil commitment statutes after the passage of the Adam Walsh
Act in 2006.
61 Media Pressure
Another potential mechanism of diffusion included in this research is the effect of media coverage. In their study of the adoption of hate crime legislation, Soule and Earl
(2001) found that states in regions that had greater levels of media coverage of hate crimes, were more likely to adopt the policy earlier. In this study the extent of regional media coverage and the extent of national media coverage is included. The national media can indirectly affect state legislatures by shaping public opinion or by encouraging citizens to mobilize on certain issues. Policies that are believed to affect the public‟s safety or morality are especially vulnerable to such effects (see the discussion of
Memorial Laws in Chapter 1).
A single galvanizing event may gain media attention because it is heinous and sensational. As a result, the public may encourage their congressional representatives to take action in response to the event. This public outcry can lead to proposals for new legislation, which are also covered extensively by the media. The cycle of interaction between the media, the public, and lawmakers can continue until action is taken that satisfies the public. A review of the effect that public opinion has on policy adoption indicated that public opinion had a strong effect on public policy (Burstein, 2003).
Furthermore, the strength of the relationship between public opinion and public policy increased when the issue being considered was a prominent one. The effect of public opinion on policy was consistent, even after controlling for political variables (Burstein,
2003). Given that media both shapes and is shaped by public opinion, it is a potentially
62 important diffusion process. It was expected that policy adoption would be positively associated with the amount of media coverage devoted to the issue of sexual offending.
Social Movement Organizations Pressure
Perhaps the most important factor in determining whether a policy is adopted is the level of public support for it among the state‟s constituents. In a study of sodomy laws, the lower the percentage of people who believed that same-sex relationships were wrong, the higher the probability that sodomy would be decriminalized (Kane, 2003).
That is, when more members of the public reported that they believed that same-sex relationships were wrong, the probability of decriminalization decreased. Social movement organizations work to educate the voting public about certain issues and typically promote a certain point-of-view. Not surprisingly, an issue as controversial as sex offender legislation has social movement organizations promoting multiple stances.
The “Association for the Treatment of Sexual Abusers” (“ATSA”) is an organization with chapters in 29 states that aims to “foster research, develop guidelines for practice, facilitate information exchange, further professional education and provide for the advancement of professional standards and practices in the field of sexual offender evaluation and treatment” (ATSA, 2001). The organization works towards these goals through participating in research, releasing reports, educating the community, and attempting to affect policy. Brief discussions of both registration/notification and civil commitment statutes, along with ATSA‟s stance on both policies, are available on the website (http://ATSA.com) and are summarized below.
63 Although the “ATSA” Board of Directors fail to state whether or not they support registration/notification laws, they provide recommendations for the most effective use of these policies to protect the public (2010a). Specifically, they encourage these policies to take into account empirical findings of research on the usefulness of registration/notification. Moreover, the written statement seems to support registration/notification policies, but only when these laws are used selectively (e.g.,
“Public safety can be enhanced, and limited resources used more efficiently, when the most active notification practices are reserved for those offenders who are at highest risk to reoffend sexually”). Based on these statements, I expected that states with “ATSA” chapters would be among the first to adopt registration/notification statutes.
The “ATSA” does not officially endorse or denounce civil commitment, but they provide recommendations for states interested in implementing civil commitment policies that are evidence-based (“ATSA” Board of Directors, 2010b). These recommendations may be difficult for some jurisdictions to implement due to financial and practical constraints (e.g., “…treatment should be consistent with current research and professional standards and guidelines, and it should reflect each individual‟s qualifying mental disorder(s), relative risk, and criminogenic needs. Individualized treatment plans are critical and should provide for systematic measurements of the sex offender‟s progress in treatment.”). Given the high standards that “ATSA” sets for what they define as appropriate use of civil commitment, I expected that states with “ATSA” chapters would be less likely to adopt civil commitment policies early.
In 1992, a sexual abuse survivor founded a community-based organization, and called the organization “Stop it Now!” With offices in three states (Georgia,
64 Pennsylvania, and Massachusetts) and the United Kingdom, “Stop it Now!” has educated countless citizens about the dangers of child abuse and techniques for prevention. The group advocates a medical model of sexual offending and encourages public health professionals, not just the criminal justice system, to become involved in the effort. In line with this view, “Stop it Now!” spends a great deal of time and resources on educational programs and campaigns. The more that people know about sexual offending, the organization believes, the better equipped they will be to identify or prevent abuse. Relevant to this research, “Stop it Now!,” as an organization, attempts to affect sex offender policy through research and community involvement. Moreover, the organization takes a “nonpunitive approach” (Giustina, 2009) and would almost certainly oppose civil commitment statutes. The group‟s stance on registration/notification may be positive, if they view the goal of registries as educational rather than punitive.
There is reason to believe that the presence of these groups may impact policy adoption at the state-level. Kane (2003) made important contributions to the study of diffusion processes, finding, for instance, that the number of gay and lesbian organizations operating within a state affected the likelihood that sodomy would be decriminalized within that state. On the other hand, in a study of the diffusion of hate crime legislation, the presence of an Anti-Defamation League office within a state was not found to affect the likelihood of adopting a hate-crime statute (Soule & Earl, 2001).
Thus, the presence of relevant social movement organizations in a state should be considered in a study of policy adoption. It is important to test these potential mechanisms given that they may spread information through regular contact (e.g.,
65 meetings, emails, or organizational initiatives). In this research, the two organizations discussed above (“ATSA” and “Stop it Now!”) were both included.
A Hybrid Theory
Although they are typically pitted against each other, empirical evidence suggests that the most successful explanations for policy adoption incorporate elements of both
Internal Determinants theory and Diffusion theory (Berry and Berry, 1990; Berry, 1994).
In their seminal 1990 piece, Berry and Berry were the first to include elements of both
Diffusion and Internal Determinants in one analysis: they used an event history analysis to predict the adoption of state lotteries. Their findings were consistent with both Internal
Determinants and Diffusion theories. That is, states‟ adoption trends were similar to surrounding states, but also to states with similar internal determinants, such as political and economic characteristics.
Combining the two approaches, internal determinants and diffusion in one model is the current state of the art in policy analysis, but these factors do not tell the complete story. Diffusion processes are also important for consideration and can easily be incorporated into an event history model predicting policy adoption.
Hypotheses
Study 1 served as a test of the role of both internal determinants and diffusion characteristics on the adoption of two pieces of sex offender legislation:
66 registration/notification policies and civil commitment statutes. The outcome variable in these analyses is the hazard rate, a variable that encompasses both the state‟s likelihood of adopting the policy as well as the timing of that adoption. That is, the single outcome variable represents both risk and timing. The hazard rate is discussed at length on pages
92 and 93.
I had seven hypotheses regarding the effect of internal determinants on the adoption of registration/notification policies and civil commitment statutes.
1. I expected that the impact of demographic internal determinants variables would
be different for registration/notification policies and civil commitment statutes,
such that:
a. I expected that there would be no significant relationship between
population, population density, and per capita income, and the risk or
timing of adopting a registration/notification policy. This expectation was
based on the fact that registration/notification policies are not very
controversial, and find support among many different people.
b. I expected that states with larger populations and more densely populated
states to adopt a civil commitment statute sooner than states with smaller
populations and less densely populated states. This expectation was based
on the fact that civil commitment statutes are very controversial. Thus, the
demographic makeup of a state may impact the likelihood that civil
commitment laws are pursued or implemented.
2. I expected that Western states would adopt both registration/notification policies
and civil commitment statutes sooner than Southern states. This expectation was
67 based on the fact that Washington State was home to three brutal murders
committed by released sex offenders in a period of two years, resulting in public
outrage and demand for legislative action.
3. I expected that Northeastern states would adopt registration/notification policies
sooner than Southern states. However, I expected that there would be no impact
of being in a Northeastern state on the adoption of a civil commitment statute.
This expectation was based on the fact that Megan Kanka‟s murder took place in
New Jersey, and a state level version of Megan‟s Law was implemented in New
Jersey before it became a federal law.
4. I expected that more Republican states (in terms of the state‟s citizenry and
governor) to adopt both registration/notification policies and civil commitment
statutes than states sooner than states with less Republican influence. This
expectation was based on the fact that Republicans tend to support more punitive
policies, and the two policies considered here are somewhat punitive (although
that is not their primary intent).
5. I expected states with higher percentages of racial minorities and higher
unemployment rates to adopt sex offender policies sooner than states with lower
percentages of racial minorities and lower unemployment rates. This expectation
was based on conflict theory. That is, I expected that states with higher values on
these variables would be more likely to introduce social control mechanisms, such
as the policies studied here, to maintain the power structure favoring the racial
majority.
68 6. I expected states with higher levels of social disorganization (i.e., poverty,
divorce, and females in the labor force) to adopt both registration/notification
policies and civil commitment statutes sooner than states with lower levels of
social disorganization. This expectation was based on the fact that states with
high levels of social disorganization tend to have low levels of informal social
control, and thus may require additional formal social control to keep crime rates
low.
7. I expected states with higher rates of violent crime and generally punitive criminal
justice policies to adopt both registration/notification policies and civil
commitment statutes sooner than states with lower rates of violent crime and less
punitive criminal justice policies. This expectation was based on the idea that
both registration/notification policies and civil commitment statutes are somewhat
punitive, and would be more likely in states that support other punitive legislation.
I had an additional 11 hypotheses regarding the effect of diffusion variables on the adoption of registration/notification policies.
8. I expected that as the percentage of neighboring states that had adopted the policy
increased, the likelihood that the state being analyzed would adopt the policy
would also increase. This expectation was based on diffusion theory, which
suggests that states emulate the policy decisions that neighboring states are
making, particularly when those policies are successful.
9. I expected that as the percentage of states in the region that had adopted the policy
increased, the likelihood that the state being analyzed would adopt the policy
69 would also increase. Like Hypothesis 8, this expectation was based on diffusion
theory.
10. I expected that as the percentage of states in the nation that had adopted the policy
increased, the likelihood that the state being analyzed would adopt the policy
would also increase. Like Hypotheses 8 and 9, this expectation was based on
diffusion theory.
11. I expected that states that are more physically connected to their neighbors (e.g.,
greater percentage of border shared) would be more likely to emulate the policy
choices of their neighbors. This expectation was based on the fact that states that
are more involved with other states are more likely to care about and be
influenced by the legislative actions other states are taking.
12. I expected that states with greater levels of connectedness with the nation as a
whole (e.g., interstates, airport hubs) would be likely to adopt both
registration/notification policies and civil commitment statutes sooner than states
with lower levels of connectedness with the nation. As in Hypothesis 12, this
expectation was based on the fact that states that are more involved with other
states are more likely to care what legislative actions other states are taking and
consider taking the same actions themselves.
13. I expected states that Walker (1969) identified as more innovative would be likely
to adopt both registration/notification policies and civil commitment statutes
sooner than states that were not identified as hubs of innovation (innovative states
were those that created or were among the first to adopt new policies). This
70 expectation was based on the fact that these policies, particularly civil
commitment, were innovative ways to deal with the sex offender population.
14. I expected states that Canon and Baum (1981) identified as more innovative
would be likely to adopt both registration/notification policies and civil
commitment statutes sooner than states that were not identified as hubs of
innovation. This expectation was based on the fact that these policies, particularly
civil commitment, were innovative ways to deal with the sex offender population.
15. I expected that the role of pressure placed on states by the Federal government
would differ for registration/notification policies and civil commitment statutes,
such that:
a. The passage of the Jacob Wetterling Act and Megan‟s Law would increase
the risk and timing of adopting registration/notification policies. This
expectation is based on the fact that both acts were directly related to
registration/notification (Jacob Wetterling Act required states to establish
registries and Megan‟s Law required states to make those registries
public).
b. The U.S. Supreme Court ruling in Kansas v. Hendricks and the passage of
the Adam Walsh Act would increase the risk and timing of adopting civil
commitment statutes. This expectation is based on the fact that the
Kansas v. Hendricks ruling dealt specifically with civil commitment
statutes (upheld them as constitutional) and the Adam Walsh Act was a
federal law that took a markedly punitive stance toward sex offenders.
71 16. I expected that as the level of media coverage of sexual offending increased, the
risk and timing of adopting both registration/notification policies and civil
commitment statutes would increase. This expectation was based on the idea that
media coverage is associated with public opinion, and may pressure legislators to
take action on a public issue.
17. I expected that the presence of an “ATSA” (Association for the Treatment of
Sexual Abusers) chapter in a state would operate differently for
registration/notification policies and civil commitment statutes, such that:
a. The presence of an “ATSA” chapter would be positively associated with
the risk and timing of adopting a registration/notification policy. This
expectation was based on the organization‟s stance in favor the policy.
b. The presence of an “ATSA” chapter would be negatively associated with
the risk and timing of adopting a civil commitment statute. This
expectation was based on the organization‟s stance against the policy.
18. I expected that the presence of a “Stop it Now!” chapter would operate differently
for registration/notification policies and civil commitment statutes, such that:
a. The presence of a “Stop it Now!” chapter would be positively associated
with the risk and timing of adopting a registration/notification policy.
This expectation was based on the organization‟s stance in favor of the
policy.
b. The presence of a “Stop it Now!” chapter would be negatively associated
with the risk and timing of adopting a civil commitment statute. This
expectation was based on the organization‟s stance against the policy.
Chapter 3
Study 1: Identifying the Factors that Affect Policy
Study 1 of this dissertation focuses on the state-level adoption of sex offender registries requiring public notification and state-level adoption of civil commitment.
Following the approach set forth by Berry and Berry (1990), I used an event history analysis of the timing of policy adoption.
Planned Analyses
Traditionally, Internal Determinants models have been tested using regression on cross-sectional data, whereas diffusion models have been tested using factor analysis
(Berry, 1994). However, Berry demonstrated that analyzing the two as separate constructs using different methods is inadequate and can lead to erroneous conclusions
(1994). For instance, a state may adopt the same policies as its neighbors because they are close, and lawmakers are following the standard set by their neighbors. However, neighboring states also tend to be more similar to each other than they are to more distant states (Buckley & Westerland, 2004). Therefore, it is also possible that they are adopting similar policies because the states are more similar to each other, not because of their spatial relationship. Failure to consider the effects of internal determinants in a test of diffusion is also problematic. It is reasonable to assume that states operate autonomously, based upon their own needs and agendas. However, they do not operate
73 in a vacuum, and are likely to take the policy actions of neighboring states into account when they are making their own policy choices. Thus, the appropriate analytic method for evaluating the variables that affect policy adoption is event history analysis, which accounts for both theories simultaneously (Berry 1994; Buckley & Westerland, 2004;
Newmark 2002).
Event History Analysis
The advantage of event history analysis is that internal determinants, diffusion, and pressure variables can be tested simultaneously. Furthermore, data that span multiple years are far more informative than cross-sectional data for studies of policy adoption
(Newmark, 2002). Because diffusion is a temporal process, it can be captured only in longitudinal data.
An assumption of event history analysis, like most regression techniques, is that the observed data are independent. However, a test of regional diffusion explicitly violates this assumption, by instead hypothesizing that states within the same region are more similar to each other than they are to states in other regions. In order to address this violation, all analyses in this research cluster standard errors within region. This technique results in larger standard errors, thus making tests of statistical significance more conservative.
74 Logic of Analysis
Event history analysis was designed to help researchers statistically model the timing and occurrence of qualitative events in individuals‟ lives (e.g., the timing of death, periods of unemployment, length of marriage prior to dissolution). This method quantifies the risk of a qualitative event occurring by calculating the probability that it will occur at each time point being analyzed. Only certain types of data are appropriate for event history analysis. In order to use this method, three conditions must be met: (1) there must be a beginning of time at which each case became at risk of the event occurring, (2) there must be a qualitative event (that may or may not happen), and (3) there must be a clear end point at which the case is no longer at risk of experiencing the event.
With regard to the first condition, the “beginning of time” is the time at which all states become at risk of adopting a piece of sex offender legislation. There are a variety of options for defining the time at which all states became “at risk” of adoption (such as the year the state officially joined the union, the date of the first sexual offense, the year the policy was first discussed in the legislature). The only limitation when selecting a beginning of time is that it must occur before any state has experienced the event and it must not be correlated with any event times (that is, there must be no systematic relationship between the beginning of time and any of the event times). If there is no compelling argument for one specific beginning of time, an arbitrary starting point may be selected by the researcher (Singer & Willett, 2003).
75 An event must be clearly defined as a transition from one state4 to another. In the
state diagram shown in Figure 3.1, the arrow represents the event occurrence. The case
shifts from being at risk of experiencing the event to having experienced it.
At Risk of Experiencing the Event No Longer At Risk of Experiencing the Event State Has No Sex Offender Sex Offender Registry Registry Requiring Public Requiring Public Notification Notification has Been Adopted
Experiences the Event
Figure 3.1: State Diagram of Adopting the Event.
Once this transition has occurred, the case has reached the “end of time” and is no
longer at risk of experiencing the event. Thus, the case is dropped from all subsequent
analyses. In order to have a complete measure of time, each case must have a beginning
of time at which it comes at risk of adopting a policy and an end of time at which it
adopts the policy or the research period ends, with T representing the time between these
two points, the duration of time at risk.
The Hazard Rate
The discrete-time method of event history analysis yields results for each defined
time interval (in this research, the intervals are years). The cases at risk of experiencing
4 Here, the term “state” refers to a clearly definable pre-event condition and a clearly definable post-event condition.
76 the event in each interval compose the risk set. Once a case experiences the event, it is removed from subsequent analyses and no longer contributes to the risk set because it is no longer at risk of experiencing the event. Dividing the number of cases that actually experienced the event within a specific interval by the number of cases at risk of experiencing the event in that interval yields a ratio known as the hazard rate. The hazard rate is calculated using the following equation:
h(tij) = number of eventsj . number of cases at riskj where the numerator is a count of all of the events experienced in interval j and the denominator is the total number of cases in the risk set at the start of interval j. In intervals in which no case experienced the event, the hazard function is undefined (i.e., no states adopted sex offender registries in 1991, so the hazard for 1991 in this analysis is undefined).
The hazard rate is expressed as a probability, represented in the following equation:
h(tij) = Pr[Ti = j | T > j]
The outcome of interest (i.e., the dependent variable) in this research is the hazard rate, which is the probability that state i will experience the event (adopt the policy) at time j, given that it had not already experienced the event (adopted the policy) prior to time j. Because the hazard rate is expressed as a probability, it is bounded between 0 and
1.
77 Modeling Time in an Event History Analysis
There are several ways to model time in an event history analysis. The most general measure of time includes a dummy variable for each of the n-1 time intervals under study. Although this procedure is often the most complete way of measuring time, it can become complicated because it introduces so many intercepts to the model.
Other options such as linear, quadratic, or cubic functions, can provide accurate depictions of time, yet be simpler to model and interpret. However, these forms constrain the data by restricting the shape of the function. If the data are linear, and a quadratic function is employed, the estimates provided will be inaccurate. Multiple specifications of time was attempted, and model fit was assessed using the deviance statistics for each model. The deviance statistic is calculated by first taking the positive difference between the -2 Log Likelihood values for the two models being compared. The resulting value is then tested using a chi-square distribution based on degrees of freedom equal to the difference in the number of variables between the models.
Data
The research project proposed here features data taken from the State Politics and
Policy Quarterly Data Resource, which houses a database called the “State Politics and the Judiciary.” Compiled by political scientist Stefanie A. Lindquist, this is an annual, state-level data set spanning the period 1965-2006. Although this dataset is extremely valuable and is the best annual, state-level compilation available, there are two additional issues to be addressed. First, a fair amount of data is missing, including all variables
78 beyond the scope of the dataset (2007). However, because Lindquist‟s codebook contains the original research sources of her data, the majority of this missingness could be filled in using newer versions of those sources (e.g., the Statistical Abstracts of the
United States).
Second, I created multiple variables, including all of the diffusion and pressure variables. The unit of analysis in the present research is the state, but because the data are structured in a state-period format there are more than 50 cases for the event history analysis. Each state appears in the dataset multiple times, disappearing from the analysis only after it is no longer at risk of adopting the policy.
Dependent Variables
In both the event history analysis predicting the adoption of registration/notification policies and the analysis predicting the adoption of civil commitment statutes, the dependent variable was the hazard rate. The hazard rate is the state‟s probability of adopting the policy of interest in each year of study.
Registration and Notification
As shown in Table 3.1, all states adopted a sex offender registry requiring public notification between 1990 and 1999. The frequency and cumulative frequency of adoption are shown in Figures 3.2 and 3.3, respectively. Early on (1990 through 1994) there were few adopters (n = 10). The majority of states (n = 28) adopted policies
79 between 1995 and 1997, cumulatively accounting for more than half of the 50 states. The remaining states adopted registries in 1998 and 1999 (n = 12).
80
Table 3.1: Year of Adoption of a Sex Offender Registry with Notification, by State Alphabetically Chronologically Alabama 1998 Washington 1990 Alaska 1994 Louisiana 1992 Arizona 1996 Idaho 1993 Arkansas 1997 New Jersey 1993 California 1996 Oregon 1993 Colorado 1998 West Virginia 1993 Connecticut 1998 Alaska 1994 Delaware 1994 Delaware 1994 Florida 1997 Kansas 1994 Georgia 1996 Kentucky 1994 Hawaii 1998 Iowa 1995 Idaho 1993 Maine 1995 Illinois 1996 Maryland 1995 Indiana 1998 Michigan 1995 Iowa 1995 Mississippi 1995 Kansas 1994 Missouri 1995 Kentucky 1994 Montana 1995 Louisiana 1992 New Mexico 1995 Maine 1995 New York 1995 Maryland 1995 North Dakota 1995 Massachusetts 1999 South Dakota 1995 Michigan 1995 Arizona 1996 Minnesota 1998 California 1996 Mississippi 1995 Georgia 1996 Missouri 1995 Illinois 1996 Montana 1995 New Hampshire 1996 Nebraska 1997 North Carolina 1996 Nevada 1998 Pennsylvania 1996 New Hampshire 1996 Rhode Island 1996 New Jersey 1993 Utah 1996 New Mexico 1995 Vermont 1996 New York 1995 Arkansas 1997 North Carolina 1996 Florida 1997 North Dakota 1995 Nebraska 1997 Ohio 1997 Ohio 1997 Oklahoma 1998 Tennessee 1997 Oregon 1993 Virginia 1997 Pennsylvania 1996 Wisconsin 1997 Rhode Island 1996 Alabama 1998 South Carolina 1999 Colorado 1998 South Dakota 1995 Connecticut 1998 Tennessee 1997 Hawaii 1998 Texas 1999 Indiana 1998 Utah 1996 Minnesota 1998 Vermont 1996 Nevada 1998 Virginia 1997 Oklahoma 1998 Washington 1990 Massachusetts 1999 West Virginia 1993 South Carolina 1999 Wisconsin 1997 Texas 1999 Wyoming 1999 Wyoming 1999 Source: Walker, Maddan, Vasquez, Van Houten, & Ervin-McLarty, 2004
81
Number of States that Adopted Sex Offender Registries in Each Year
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Year
Figure 3.2: Frequency of Sex Offender Registry Adoption, by State.
Total Number of States that Adopted Sex Offender Registries by Year
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year
Figure 3.3: Cumulative Frequency of Sex Offender Registry Adoption, by State.
82
This S-shaped distribution is typical of policy adoption, in which a few states start the process off, many states follow suit a couple of years later, and the remaining states adopt the policy shortly thereafter (Cocca, 2004; Barabasi, 2002).
For this research, the years of analysis are limited to 1986 through 1999. The first year of study is defined as 1986 because it provides multiple years of observations prior to the first state‟s adoption of a registry requiring public notification. The final year is
1999 because in 1999 the final four states adopted registries requiring public notification, and thus, no more states were at risk of adopting such policies after this time.
Civil Commitment Statutes
As shown in Table 3.2, 20 states adopted civil commitment statutes between 1988 and 2007. The frequency and cumulative frequency of adoption are shown in Figures 3.4 and 3.5, respectively. The cumulative distribution in Figure 3.5 may reflect the upward trend at the start of the s-shaped distribution that policy adoption typically follows.
However, it is too soon to tell, since less than half of the 50 states have adopted a civil commitment statute for sex offenders.
83
Table 3.2: Year of Adoption of a Civil Commitment Statute, by State. Alphabetically Chronologically Arizona 1996 Illinois 1988 California 1996 Washington 1990 Florida 1999 Wisconsin 1994 Illinois 1988 Kansas 1994 Iowa 1998 Missouri 1994 Kansas 1994 New Jersey 1994 Massachusetts 1998 Arizona 1996 Minnesota 1999 California 1996 Missouri 1994 North Dakota 1997 Nebraska 2006 South Carolina 1998 New Hampshire 2007 Iowa 1998 New Jersey 1994 Massachusetts 1998 New York 2007 Florida 1999 North Dakota 1997 Minnesota 1999 Pennsylvania 2003 Texas 1999 South Carolina 1998 Pennsylvania 2003 Texas 1999 Virginia 2003 Virginia 2003 Nebraska 2006 Washington 1990 New York 2007 Wisconsin 1994 New Hampshire 2007 Source: Gookin, 2007
Number of States that Adopted Sex Offender Registries in Each Year
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Figure 3.4: Frequency of Civil Commitment Adoption, by State.
84
Total Number of States that Adopted Sex Offender Registries by Year
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Figure 3.5: Cumulative Frequency of Civil Commitment Adoption, by State.
For the event history analysis predicting the timing of adoption of civil
commitment statutes, the years of analysis are limited to 1986 through 2007. Again, the
first year of study was selected because it provides multiple years of observations prior to
the first state‟s adoption of the statute. The final year is 2007 because that is the most
recent year in which states had adopted civil commitment statutes, and is also the most
recent year in which data were available for most of the included variables.
85 Independent Variables: Internal Determinants
Demographic Internal Determinants
Previous research suggests that state-level demographic variables are associated with policy adoption (see Table 3.3 for descriptive statistics and Table 3.4 for correlations among independent variables). In the present study, a variety of demographic variables are examined. The population of each state is included as a time- varying predictor, as reported in the annual editions of the Statistical Abstracts of the
United States. The population was reported in thousands in the original source data, but is recalculated in this research to millions. Population density, measured per square mile, was also based on measures from the Statistical Abstracts of the United States, and was calculated by dividing the population by the state‟s land area (provided by the State
Politics and the Judiciary dataset). These measures are intended to capture how large the state‟s citizenry is, as well as its rural/urban representation.
Per capita income is included to examine the individual level of wealth in the state. To adjust for inflation, all years of per capita income are reported in dollars constant with the value of a dollar in 1996. Per capita income was reported by the
Statistical Abstracts of the United States.
86
Table 3.3: Descriptive Statistics Mean or Standard Variable Percentage Deviation Minimum Maximum Demographics Population (in 100,000) 4.81 5.03 .433 31.70 Population Density .17 .23 .002 1.05 Per Capita Income (in 1,000) 22.30 6.93 2.99 50.76 Region South 35% N/A 0 1 Northeast 22% N/A 0 1 Midwest 20% N/A 0 1 West 23% N/A 0 1 Political Republican Governor .47 .50 0 1 Percentage Republican 33.14 7.77 6.8 75 Percentage Females in the 19.15 7.74 2.1 37 Legislature Professionalism .19 .13 .03 .66 Competition 38.23 11.41 9.26 56.58 Innovation .26 .11 .02 .51 Racial Threat/Conflict Hispanic 5.74 7.87 .4 44.43 Black 10.23 9.79 .2 37.31 Unemployment Rate 5.38 1.58 2.2 13.1 Social Disorganization Divorce Rate 4.69 1.47 0 14.1 Percentage Living in Poverty 13.13 3.97 2.9 27.2 Rate of Females in the Labor 59.43 4.61 39.6 73.1 Force Structural-Functional Murder Rate 6.12 3.54 .2 20.3 Incarceration Rate .31 .15 .05 .85 Death Penalty .78 .41 0 1 Traditional Diffusion Registration Neighboring .10 .23 0 1 Registration Regional .10 .10 0 .92 Registration National .11 .20 0 .92 Civil Commitment .13 .18 0 .75 Neighboring Civil Commitment Regional .13 .15 0 .67 Civil Commitment National .15 .13 0 .36
87 Mean or Standard Variable Percentage Deviation Minimum Maximum Physical Diffusion Length of Border 1215.58 491.32 160 3029 Shared Border .75 .22 .18 1.17 Number of Neighbors 4.41 1.71 0 8 Interstates 4.03 2.25 0 12 Airport Hub .22 .42 0 1 Social Networks – Hub .11 .32 0 1 States Hub State (Walker) .11 .32 0 1 Hub State (Canon & Baum) .08 .28 0 1 Federal Pressure Jacob Wetterling Act .55 .50 0 1 Megan‟s Law .45 .50 0 1 Kansas v. Hendricks .31 .46 0 1 Adam Walsh Act .04 .19 0 1 Media Media Coverage 5328.20 5859.02 129 20929 Social Movement Org. “Association for the Treatment .50 .50 0 1 of Sexual Abusers” “Stop it Now!” ,07 .26 0 1
Table 3.4. Correlations Among Independent Variables
VARIABLES
ation
rcentage Living rcentage
Popul Region:South Region:Northeast Region:Midwest Region:West Republican Governor Percentage Republican Females Percentage Legislature the in Professionalism Competition Innovation Black Percentage Percentage Hispanic Unemployment Rate Rate Divorce Pe Poverty in Females Percentage Force Labor the in Rate Murder Rate Incarceration Population 1 Region: South .10 1 Region: Northeast .01 -.39 1 Region: Midwest .02 -.36 -.27 1 Region: West -.14 -.40 -.29 -.28 1 Republican Governor -.04 .15 -.05 -.13 -.003 1 Percentage -.02 -.09 -.39 .16 .34 -.08 1 Republican Percentage Females -.19 -.51 .25 -.01 .34 -.03 -.08 1 in the Legislature Professionalism .71 -.19 .19 .15 -.12 -.04 -.16 -.06 1 Competition .02 -.69 .21 .39 .19 -.06 .08 .37 .24 1 Innovation .47 .25 -.19 .23 -.30 -.0002 -.02 -.32 .29 -.07 1 Percentage Black .28 .69 -.15 -.20 -.43 .09 -.11 -.41 .12 -.52 .30 1 Percentage Hispanic .31 -.19 -.06 -.23 -.50 -.07 .10 .24 .17 .05 -.09 -.16 1 Unemployment Rate .19 .22 -.16 -.18 .08 .17 -.09 -.32 .19 -.14 .15 .16 .12 1 Divorce Rate -.17 .23 -.36 -.27 .36 .19 .14 -.11 -.24 -.32 -.17 -.05 .14 .24 1
89
spanic
VARIABLES
Population Region:South Region:Northeast Region:Midwest Region:West Republican Governor Percentage Republican Females Percentage Legislature the in Professionalism Competition Innovation Black Percentage Hi Percentage Unemployment Rate Rate Divorce Living Percentage Poverty in Females Percentage Force Labor the in Rate Murder Rate Incarceration Percentage Living in .06 .40 -.28 -.16 -.03 .03 -.01 -.36 -.03 -.34 .11 .27 .12 .45 .22 1 Poverty Rate of Females in -.27 -.45 .13 .25 .14 -.11 .08 .53 -.20 .25 -.19 -.31 -.07 -.64 -.21 -.51 1 the Labor Force Murder Rate .41 .56 -.32 -.21 -.11 .06 -.05 -.42 .28 -.43 .31 .69 .22 .42 .30 .40 -.48 1 Incarceration Rate .20 .45 -.27 -.20 -.05 -.06 .08 -.12 .04 -.40 .22 .62 .13 -.04 .12 .20 -.11 .45 1 Death Penalty -.02 .26 -.39 -.21 .29 .06 .41 -.15 -.33 -.39 .19 .26 .15 -.03 .31 .05 .02 .29 .40 Neighboring -.07 -.06 .02 .04 .01 -.06 -.02 .32 -.09 .08 .07 -.01 .12 -.32 -.21 -.08 .25 -.20 .32 Regional -.04 -.28 -.04 .37 .004 -.20 .12 .32 -.07 .25 -.04 -.13 .03 -.38 -.30 -.15 .37 -.34 .25 National -.06 .04 .02 -.10 .02 -.17 .02 .31 -.21 -.03 -.09 .07 .11 -.38 -.21 -.05 .24 -.24 .51 Length of Border .37 .13 -.66 .19 .33 .003 .35 -.15 .12 -.03 .29 -.10 .32 .20 .28 .26 -.14 .28 .17 Shared Border -.34 .16 -.34 -.11 .26 .03 .32 -.13 -.42 -.36 -.14 .10 .004 -.06 .25 .05 .04 .02 .09 Number of -.05 .03 -.35 .07 .24 .04 .26 -.09 -.04 -.20 .10 -.22 .08 -.06 .19 .03 .03 -.05 -.03 Neighbors Interstates .64 .27 -.09 -.03 -.19 .06 -.001 -.27 .39 -.13 .42 .28 .06 .23 -.18 .13 -.34 .37 .12 Airport Hub .59 -.06 .03 -.10 .13 -.05 .14 -.05 .35 .01 .27 .11 .27 .03 -.11 -.05 -.07 .14 .05 Jacob Wetterling Act -.04 .05 .02 -.09 .02 -.18 -.01 .30 -.19 -.03 -.08 .07 .09 -.39 -.17 -.06 .26 -.19 .47 Linear Time -.06 .04 .02 -.09 .02 -.16 .04 .33 -.20 -.03 -.09 .08 .12 -.34 -.22 -.02 .25 .22 .52 Squared Time -.06 .03 .02 -.09 .03 -.14 .05 .29 -.20 -.03 -.09 .07 .12 -.30 -.22 -.01 .20 .23 .48 Cubic Time -.06 .03 .02 -.08 .03 -.12 .06 .26 -.19 -.03 -.08 .06 .12 -.26 -.21 .002 .17 .22 .44 4th Order Time -.06 .02 .01 -.07 .03 -.10 .06 .24 -.18 -.03 -.07 .06 .11 -.22 -.20 .01 .14 .20 .40
90
Rate
VARIABLES
entage entage
Population Region:South Region:Northeast Region:Midwest Region:West Republican Governor Perc Republican Females Percentage Legislature the in Professionalism Competition Innovation Black Percentage Hispanic Percentage Unemployment Rate Rate Divorce Living Percentage Poverty in Females Percentage Force Labor the in Rate Murder Incarceration Per Capita Income .09 -.14 .30 -.09 -.05 -.14 -.10 .37 .04 .14 -.05 .05 .16 -.42 -.35 -.26 .37 -.24 .39 Megan‟s Law -.06 .04 .02 -.09 .01 -.18 -.04 .22 -.22 -.03 -.08 .07 .08 -.38 -.20 -.06 .22 -.24 .46 Kansas v. Hendricks -.07 .02 .02 -.08 .03 -.12 .05 .16 -.18 -.02 -.08 .06 .10 -.28 -.21 -.05 .16 -.23 .41 Media -.06 .03 .02 -.08 .03 -.14 .04 .18 -.20 -.03 -.08 .07 .11 -.31 -.21 -.01 .19 -.22 .46 “ATSA” .32 .01 -.08 .19 -.11 .05 -.06 .08 .30 -.09 .33 .22 -.07 -.01 -.23 -.09 .08 .17 .11 “Stop it Now!” .17 .21 .04 -.14 -.15 .13 -.04 .03 .03 -.24 .07 .33 -.09 -.06 -.18 -.10 .03 .20 .10 Hub State (Walker) .49 -.26 .38 .02 -.10 -.09 -.18 -.10 .74 .29 .20 .09 .17 .08 -.32 -.12 -.10 .09 .01 Hub State (Canon & .35 .09 -.16 .08 -.02 .004 -.04 -.03 .13 -.07 .53 -.09 .14 .13 .05 .10 -.03 .12 -.06 Baum)
91
ndricks
VARIABLES
Neighboring Regional National Border of Length Border Shared Neighbors of Number Interstates Hub Airport (Walker) State Hub & (Canon State Hub Baum) Act Wetterling Jacob Law Megan‟s He v. Kansas Media “ATSA” it “Stop Now!”) Penalty Death Income Capita Per Neighboring 1 Regional .65 1 National .71 .77 1 Length of Border -.03 .02 -.06 1 Shared Border .03 -.04 .06 -.13 1 Number of .11 .04 .03 .28 .54 1 Neighbors Interstates -.12 -.09 -.06 .32 -.06 .28 1 Airport Hub -.10 -.05 -.04 .13 .05 .17 .51 1 Hub State (Walker) .0002 .03 -.05 -.08 -.44 -.14 .14 .17 1 Hub State (Canon & -.05 -.12 -.13 .38 -.08 .12 .21 .08 .03 1 Baum) Jacob Wetterling Act .59 .64 .87 -.04 .04 .02 -.03 -.03 -.05 -.11 1 Megan‟s Law .64 .70 .91 -.05 .05 .03 -.04 -.04 -.05 -.12 .82 1 Kansas v. Hendricks .64 .70 .88 -.06 .06 .03 -.06 -.03 -.03 -.13 .61 .74 1 Media .67 .72 .92 -.06 .05 .02 -.06 -.04 -.04 -.12 .74 .83 .81 1 “ATSA” .02 .05 -.05 .17 -.09 .09 .49 .35 .12 .03 -.04 -.04 -.05 -.05 1 “Stop it Now!” .06 -.08 .02 -.17 .23 .09 .38 .28 -.10 -.08 .03 .02 .02 .01 .27 1 Death Penalty .01 -.05 .05 .17 .52 .26 .08 .14 -.31 .02 .03 .04 .05 .04 .10 .15 1 Per Capita Income .65 .66 .83 -.23 -.07 -.06 -.03 .03 .25 -.12 .71 .74 .72 .80 .04 .11 -.01 1
92
VARIABLES
Neighboring Regional National Border of Length Border Shared Neighbors of Number Interstates Hub Airport (Walker) State Hub & (Canon State Hub Baum) Act Wetterling Jacob Law Megan‟s Hendricks v. Kansas Media “ATSA” it “Stop Now!”) Penalty Death Income Capita Per Linear Time .69 .75 .97 -.05 .05 .02 -.06 -.04 -.04 -.13 .84 .86 .83 .93 -.05 .02 .04 .85 Squared Time .69 .75 .96 -.06 .05 .02 -.06 -.04 -.04 -.13 .76 .84 .88 .97 -.05 .01 .04 .83 Cubic Time .66 .71 .90 -.06 .05 .02 -.06 -.04 -.04 -.12 .67 .77 .87 .97 -.05 .01 .04 .79 4th Order Time .63 .67 .85 -.06 .04 .02 -.06 -.04 -.03 -.11 .59 70 .83 .96 -.05 .01 .04 .75
VARIABLES
Order Time Order
th
Linear Time Linear Time Squared CubicTime 4 Linear Time 1 Squared Time .97 1 Cubic Time .92 .99 1 4th Order Time .86 .96 .99 1
Political Internal Determinants
There are also several characteristics of the state‟s legislative system that are important variables in this analysis. First, the party of the governor was included as a measure of the political leadership in the state. The party of the governor was adapted from the measure provided by The State Politics and Policy Quarterly Data Resource. In years in which leadership changed, the original source provided decimals based on the proportion of time each leader was in office (Republicans were coded 0, Democratic governors were coded 1, Independents were coded 0.5). The adapted variables used in this research were dummy coded, with Republicans as the reference category.
Second, the percentage of the population who identified themselves as Republican was included to measure the political climate in the state. The percentage of the state population who reported being Republican was taken from the State Politics and the
Judiciary dataset (its original source: CBS News/New York Times Poll of the 48 continental United States).
Finally, three characteristics of the state‟s legislative system were included: legislative professionalism, a measure of political competition in the state, and a measure of the level of innovation within the state. The measure of professionalism is an index, which was created by a political scientist (Squire, 2007). Three traditional indicators of legislative professionalism were equally weighted to form the scale: the amount of pay that legislators received, the length of time spent in session each year, and the amount of resources available to the staff (Squire, 2007). The amount of pay that legislators receive
94 is relevant because it helps to determine whether their service as a state legislator is a full time job or a role that the individual takes on in addition to a career. Legislators who are paid sufficient salaries to pay their bills without obtaining an additional job are likely to spend more time working on new legislation and working to meet their constituents‟ needs. In contrast, legislators whose salaries are not sufficient to serve as a sole source of income are required to pursue additional employment and, thus, may not consider their role as a legislator their first priority. Second, the length of time spent in session each year is traditionally considered in measures of legislative professionalism because legislators who spend more time in session are generally more familiar with the rules and protocols of the state. This increased familiarity may increase the effectiveness and efficiency of the legislature as a whole. Third, the resources available to the staff are taken into account in the scale. Legislators with larger staffs, and more equipment available to those staff members, are likely able to accomplish more than legislators with little or no support from a staff. Moreover, states with higher ratings of legislative professionalism are likely more able to consider changes in legislation compared to states with lower levels of legislative professionalism.
The measure of political competition used in this research was created by two political scientists (Holbrook & Van Dunk, 1993) who included four indicators: the percentage of the popular vote that went to the winner of the election, the margin of victory, whether the seat was contested, and whether the seat was considered “safe” (a safe seat is one in which the winner received at least 55% more of the vote than did their opponent). Measures of these indicators were taken between 1982 and 1986. Typical measures of political competition are measured using the results of state-level elections
95 (e.g., the Ranney index). Holbrook and Van Dunk‟s (1993) measure improves upon these measures by also considering elections that take place at the district-level, rather than only the state-level elections. Each state‟s score on the scale was calculated by subtracting the average percentage of the popular vote that went to the winner across jurisdictions, the average margin of victory for the winner across jurisdictions, the average percentage of uncontested seats across jurisdictions, and the average percentage of safe seats across jurisdictions from 100. Thus, a low score would indicate that there is little political competition in the state, whereas a high score (closer to 100) would indicate that there is a higher level of political competition in the state. No scores of political competition were estimated for the state of Louisiana because no election data were available (see Appendix A for a table displaying the presence of missing data and how it was handled).
Following the method described on page 58, Canon and Baum (1981) assigned each state a score representing its level of innovation in judicial policy. This score was included as a control variable in the political internal determinants model.
The political model took the following form:
h(tij) = βDemograticGovernorXDemograticGovernor + βIndependentGovernorXIndependentGovernor +
β%DemX%Dem + β%Rep X%Rep + βProfessionalismXProfessionalism + βCompetitionXCompetition +
βInnovationXInnovation + ε
96 Racial Threat/Conflict
In this research, the percentage of the population that is black and the percentage of the population that is Hispanic were both included, as reported in the Statistical
Abstracts of the United States. No estimates of the percent black and percent Hispanic were included in the Statistical Abstracts of the United States for 1986 or 1987. Thus, the values that were reported for 1988 were also used for the 1986 and 1987 values in this analysis.
The unemployment rate, measured per 1,000 people, was taken from the
Geographic Profile of Employment and Unemployment, compiled by the Bureau of Labor
Statistics.
The racial threat/conflict model took the following form:
h(tij) = β%HispanicX%Hispanic + β%BlackX%Black + βUnemploymentRate XUnemploymentRate + ε
Social Disorganization
This research included three measures of social disorganization. First, the divorce rate, taken from the Statistical Abstracts of the United States, was measured per 1,000 people in the population. Second, the rate of female participation in the labor force, taken from the Statistical Abstracts of the United States, was included. This value was calculated as the percentage of females in the non-institutionalized civilian population who were employed. Because 2007 values were not available, the values for 2006 were also used for 2007.
97 Third, the percentage of the state population living in poverty was included. The
2001 and 2002 values for this variable were based on the average of the percent living in poverty in 2000, 2001, and 2002, as these were the only values reported by the Statistical
Abstracts of the United States for 2001 and 2002
Relevant to sexual offending legislation, an advocate of the social disorganization perspective would expect areas with higher levels of social disorganization (and thus, lower levels of informal social control) to require additional formal controls (such as sex offender laws) to maintain order in the jurisdiction.
The social disorganization model took the following form:
h(tij) = βDivorceXDivorce + βFemaleParticipationXFemaleParticipation βPovertyXPoverty + ε
Structural-Functional Internal Determinants
According to a structural-functional view of policy adoption, jurisdictions adopt policies based on their specific needs. The primary structural-functional predictor variable was the rate of murder within the state, a measure of violent crime. Taken from the Uniform Crime Reports, compiled annually by the FBI, the murder rate was measured per 100,000 people in a state. The rate of a sex crime may be appropriate here; however, in the following study (see Chapter 4) the rate of sexual crimes was the dependent variable. Thus, it would be inappropriate to use them as independent variables in this analysis.
In order to assess the state‟s punitiveness, the incarceration rate and whether or not the state allows the death penalty were included in the analysis. The incarceration
98 rate was originally measured per 100,000 people living in a state, as reported by the
Sourcebook of Criminal Justice Statistics, compiled annually by the Bureau of Justice
Statistics (Lindquist, 2007). In this research, the incarceration rate was recalculated and the measure used was the number of incarcerated persons per 1,000 people in the population. Data were available only through 2005. Thus, the values from 2005 were also used as the values for 2006 and 2007.
Whether or not a state allows the death penalty was a dummy variable, in which states that allow the death penalty were coded high.
The structural-functional model took the following form:
h(tij) = βMurderRateXMurderRate + βIncarcerationXIncarceration + βDeathPenaltyXDeathPenalty + ε
Independent Variables: Diffusion
Diffusion was operationalized in three ways: the percentage of neighboring states
(sharing a border) that have adopted the policy, the percentage of states in the region that have adopted the policy, and the percentage of states in the nation that have adopted the policy.5 Regions were defined using the same categories used by the Bureau of Justice
Statistics.
Each of these variables was constructed using the same procedure. As an example, consider how I coded certain states in the Western region. Washington, in
1990, was the first state (in the West and in the nation) to adopt a public sex offender
5 Because of anticipated problems of multicollinearity, the three operationalizations (neighboring, region, nation) of diffusion appear individually in three separate models.
99 registry. In the two years of data leading up to Washington‟s adoption of a public sex offender registry, the regional diffusion value would have been 0, because no other state in this region had experienced the event (see Figure 3.6). However, in 1990, every remaining state had a value of 8% because in that year, one state in the region (1/13 =
8%) had experienced the event. By the time Wyoming adopted the policy in 1999, every other state in the region had also adopted the policy, so the 1998 value of this variable is
12/13 = 92% and the 1999 value is 13/13 = 100% (see Figure 3.7). There were no missing values for any of the created diffusion variables.
Figure 3.6: Constructing the Regional Diffusion Variable (Part 1)
100
Figure 3.7: Constructing the Regional Diffusion Variable (Part 2)
The three diffusion models took the following form:
h(tij) = β%NeighboringX%Neighboring + ε
h(tij) = β%RegionX%Region + ε
h(tij) = β%NationX%Nation + ε
101 Two measures of the physical connection between states were also included in these analyses.
First, data from the Worldmark Encyclopedia of the States (Gall, 2007) and the
State Border Dataset (Holmes, 1998) were used to calculate the percentage of a state‟s border that is shared with other states. Variables were provided for the length of a state‟s perimeter by the Worldmark Encyclopedia of the States. The length of each state border was taken from the State Border Dataset. In cases in which a state bordered multiple other states (e.g., Alabama and Florida, Alabama and Georgia, Alabama and Tennessee, etc.), the multiple borders were summed. The length of shared border was then divided by the length of the entire border to provide the percentage of the border that was shared.6
The number of interstate highways that go through a state is a physical indicator of the amount of traveling that takes place into and out of the state. Thus, a variable was included in these analyses, which represents the number of interstates present in each state. This variable was treated as non-time-varying and was measured using a Rand
McNally Road Atlas (2006).
These physical measures of diffusion were then be added to each of the three diffusion models, as shown:
h(tij) = βDiffusionMeasureXDiffusionMeasure + β#Neighbors X#Neighbors + β%SharedBorder
X%SharedBorder + βInterstatesXInterstates + ε
6 Because the length of the states‟ shared borders was calculated by hand using GIS software, the possibility of measurement error was introduced. Thus, three states (Pennsylvania, Kansas, and Mississippi) had longer shared borders calculated by Holmes (1998) than the reported length of their border reported by the encyclopedia (2007). The resulting values of percentage shared border were left at their calculated amounts of more than 100 percent.
102 Independent Variables: Diffusion Processes
Although the process of physical diffusion is important, information may also be disseminated in other ways. In order to account for the communication that may take place between states, dummy variables representing “hub” states that have been identified by researchers as leaders in innovation were included in the analyses. Additionally, three institutions that are relevant to the creation of policy are the Federal government, the media, and social movement organizations. In order to gauge the role of these three institutions in the adoption of sex offender policy, variables representing the Federal government, media coverage, and social movement organizations were included in these analyses.
Hub States
In both Walker‟s (1969) analysis and Canon and Baum‟s (1981) analysis, states were assigned innovation scores. Those states that were most innovative in terms of policy creation may have undue influence on the social network of states. Other states may observe and emulate the more innovative states before making their own policy choices or changes. In order to account for this phenomenon, the states that Walker identified as the most innovative were indicated by dummy variables (coded high if they were among the most innovative). A second dummy variable was created using Canon and Baum‟s innovation scores (again, the states were coded high if they were among the most innovative).
103 Federal Government
The Federal government can hand down legislative guidelines to the states. In this analysis, three dummy variables representing federal pressure were included. One of the dummy variables was coded high after the adoption of the Jacob Wetterling Act
(1994), which required all states to create databases of sex offenders living within the state. The second dummy variable was coded high after the adoption of Megan‟s Law
(1996), which required that a public notification system be put into place. A final variable representing federal pressure was included in the analysis predicting civil commitment statutes, and was coded high after the Kansas v. Hendricks ruling. This
Supreme Court case, from 1997, upheld civil commitment statutes as Constitutional at the state level.
Media
American public opinion both shapes and is shaped by the media. Thus, a measure of the media‟s role in information sharing about sex offenders during this time was included in the analyses. The search engine LexisNexis Academic, a major media database, was searched using the terms “sex offender,” “sexual offender,” and “sexual predator.” Any article that was published in the LexisNexis Academic database of U.S. newspapers (see a list of included papers, Appendix B), that included any of the three search terms counted towards the total number of articles. Because this variable was measured at the national level, the values are consistent across states, but vary over time.
104 Social Movement Organizations
As a final measure of social pressure, dummy variables were included to represent the presence of national social movement organizations in a state. If the organization is present, the variable was coded high. The social movement organizations that are included in this study are the “Association for the Treatment of Sexual Abusers” and
“Stop it Now!”.
Taking all four major institutions of information dissemination and investment in the policy process in to account, the pressure model will take the following form:
h(tij) = βHubStatesXHubStates + βJacobWetterlingActXJacobWetterlingAct + βMegan‟sLawXMegan‟sLaw +
βKansas v. HendricksXKansas v. Hendricks + βNationalMediaXNationalMedia + β”ATSA”X”ATSA” + βStopXStop + ε
Full Model
Finally, I ran models containing predictors representing each of the theoretical perspectives discussed above, trimmed according to whether the relationship between those variables and the outcome were statistically significant, theoretically significant, or both. The full and final models were used to test whether internal determinants, diffusion, or pressure best explain policy adoption, or if some combination of the three was most appropriate.
105 Results of an Event History Analysis Predicting the Adoption of a Registration/Notification Policy
Policies requiring registration and public notification are in place in all 50 states, but the timing of adoption varied across a full decade (1990 to 1999). The following analyses were aimed at determining what state-level factors or diffusion processes were associated with an increased risk of policy adoption. Event history analysis was used when time plays a role in the outcome variable, as is the case in this research. Because of the temporal nature of diffusion, and the changing pressures on states to adopt policies based upon the legislative actions that other states are taking, it was necessary to include a measure of time in all of the included analyses.
Specifying Time
There are multiple ways to measure time in an event history analysis. A general model includes a dummy variable for all but one of the time periods being studied, and is the most flexible model in terms of fit (Singer & Willet, 2003). The general model, thus, will always produce the best goodness of fit statistics, such as the deviance test.
However, because of its complex nature, and the fact that including the series of dummy variables consumes so many degrees of freedom, many event history researchers prefer other models of time. One alternate method is to take the log of the linear measure of time. This transformation uses only one degree of freedom, but does not allow the model to vary with each year of data, as does the general model.
106 A linear measure constrains the effect of time to a straight line (see Figure 3-8).
The line may trend upwards, downwards, or have no slope, but is limited because the effect of X on Y does not change over time. A squared measure adds an additional term, using one additional degree of freedom and allowing the effect of X on Y to change over time, but only once (see Figure 3.8). Similarly, a cubed measure adds one additional term, using one additional degree of freedom, and allowing the effect of X on Y to change once more (see Figure 3.8). A fourth-order function follows the same logic, adding an additional term (and an additional degree of freedom) and permitting the effect of X on Y to change a total of three times (see Figure 3.8).
Cubic Fourth-order Linear Squared
Figure 3.8: Functional Forms of Time
In order to determine which model provides the best fit to the data, I conducted a chi-square test. Models using each measure of time and containing no other predictor variables resulted in the -2 Log Likelihood scores shown in Table 3.5.
107
Table 3.5. Specifying Time for Registration/Notification Models 2 x [(-2 Log Likelihood of General Model) Measure -2 Log Likelihood Df – (-2 Log Likelihood)] General -94.79 0 13 Logged -222.43 255.28 1 Linear -243.78 297.98 1 Squared -124.55 59.52 2 Cubic -106.15 22.72 3 Fourth- -99.18 4 Order 8.78 Fifth- -99.15 5 Order 8.72
As expected, the model with the lowest -2 Log Likelihood was the general model.
However, the general model requires 13 degrees of freedom, which makes it more cumbersome to use once multiple independent variables are introduced to the model. The logged model requires only one degree of freedom, but it clearly provides a very poor fit based on the absolute value of the difference between the -2 Log Likelihood of the general model and the logged model, multiplied by two (255.28). The linear model provides an even worse fit than the logged model (297.98 with no additional degrees of freedom saved).
The squared model is a significant improvement over the linear model (χ2 =
124.32, 1 df, p < .001), but its -2 Log Likelihood is still substantially higher than the general model. Although the difference between the absolute value of two times the -2
Log Likelihood of the cubic model subtracted from the -2 Log Likelihood general model is significantly smaller than the squared model (χ2 = 29.62, 1 df, p < .001), the fourth- order model clearly provides a superior fit (χ2 = 8.33, 1 df, p < .01). A fifth-order model
108 was also run, and decreased the -2 Log Likelihood difference slightly, but not enough to warrant an increase in the degrees of freedom (χ2 = 0.06, 1 df, p = .80).
Thus, the most appropriate specification of time for this model is the fourth-order model. Although it has a somewhat higher 2 Log Likelihood than does the general model
(-94.79 and -99.18, respectively), the 9 degrees of freedom that are saved by using the fourth-order model rather than the general model make the fourth-order model a more parsimonious and desirable choice. In the subsequent analyses, the event history models used to assess the role of state-level factors on the adoption of registration/notification policies included four terms to account for the fourth-order measure of time: a linear term, a squared term, a cubic term, and a fourth-order term.
Demographic Internal Determinants
Characteristics of the state and the state‟s citizenry were used to predict the risk and timing of registration/notification adoption across states (see Table 3.6). Two factors that are frequently related to policy adoption, population and population density, were not significantly related to policy adoption in this model. These findings were consistent with Hypothesis 1A, that there would be no significant effect of these demographic variables on the timing of adopting a registration/notification policy.
However, the region of the country did have a significant impact on a state‟s risk and timing of adopting a registration/notification policy. States located in the Midwest and states located in the West were significantly more likely to adopt a registration/notification policy sooner than were states located in the South (there was no
109 significant difference between the risk of adoption in the Northeast and the South). The strength of this association (odds of adoption were increased 94% and 84%, respectively) indicated that region had an important effect on the adoption of registration/notification statutes.
This finding regarding region was consistent with Hypothesis 2, that states in the
Western region would be at an increased risk of adopting a registration/notification policy, but inconsistent with Hypothesis 3, that states in the Northeast would also be at an increased risk of adopting a registration/notification policy. The fact that states in the
Midwest were at an increased risk of policy adoption was not hypothesized. However, this finding could also be expressed as the South being at a decreased risk of policy adoption. Although it is not significant, the coefficient for the Northeast region was strong and positive (1.02). This finding, combined with the significant and positive findings for both the Western and Midwestern regions, suggested that the Southern region is less likely than any other region of the country to adopt a registration/notification policy. Although Southern states do tend to have punitive criminal justice policies, in general, they also tend to value personal privacy and limit government control of individuals. This hesitance to apply governmental control to individuals may explain the South‟s reluctance to adopt registration/notification policies.
110
Table 3.6. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Demographic Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -4.56** 1.61 .01 Squared Measure of Time 1.32** .43 3.75 Cubic Measure of Time -.13** .04 .88 Fourth-order Measure of Time .004** .001 1.004 Population .001 .04 1.001 Population Density .42 .95 1.53 Per Capita Income -.15 .08 .86 Region (South is the reference category) Northeast 1.02 .65 2.76 Midwest .66** .20 1.94 West .61*** .17 1.84 ** p-value <.01 *** p-value <.001
Political Internal Determinants
Although characteristics of the state and its citizenry are important for understanding major changes that take place at the state level, the political climate is particularly important when considering changes in legislation. In this analysis, states with a Republican governor were at a significantly higher risk of adopting a registration/notification policy sooner than were states with a Democratic governor (see
Table 3.7). This finding, partially supportive of Hypothesis 4, was expected because
Republicans tend to be more punitive in criminal justice policy, and registration/notification statutes limit the freedom of sex offenders in the community.
The second expectation expressed in Hypothesis 4, that the percentage of Republicans in the state would impact the likelihood of policy adoption, was not supported. The coefficient was non-significant and near zero. None of the other political internal
111 determinants were significantly associated with the risk or timing of adopting a registration/notification statute.
Table 3.7. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Political Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -7.88*** 1.40 .0004 Squared Measure of Time 2.15*** .38 8.55 Cubic Measure of Time -.21*** .03 .81 Fourth-order Measure of Time .007*** .001 1.01 Republican Governor (Democrat is the reference .54* .26 1.71 category; 4 Independent observations dropped) Percentage Republican -.01 .01 .99 Percentage Females in the Legislature -.01 .01 .99 Professionalism -.46 1.58 .63 Competition .04 .02 1.04 Innovation .81 1.94 2.26 * p-value < .05 *** p-value <.001
Racial Threat/Conflict Internal Determinants
None of the three measures of racial threat/conflict theory were significantly associated with the risk or timing of adopting a registration/notification policy (see Table
3.8). Although states with higher percentages of the population that were racial minorities were expected to adopt the policy sooner, according to Hypothesis 5, the effects found here were near zero and non-significant. Although the coefficient for unemployment rate was positive, as hypothesized, it was not significantly associated with the risk or timing of policy adoption.
112
Table 3.8. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Racial Threat/Conflict Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.66*** 1.66 .001 Squared Measure of Time 1.83*** .45 6.26 Cubic Measure of Time -.18*** .04 .84 Fourth-order Measure of Time .006*** .001 1.006 Percentage Hispanic -.03 .04 .97 Percentage Black -.02 .01 .98 Unemployment Rate .14 .12 1.15 *** p-value <.001
Social Disorganization
None of the social disorganization indicators had significant impacts on the risk or timing of adopting a registration/notification policy (see Table 3.9). The effects presented here were near zero and non-significant, providing no support for Hypothesis 6.
In fact, the social disorganization variables were such a poor fit to the data, that even the four measures of time (linear, squared, cubic, and fourth-order) failed to reach significance, though they are significant in every other model presented in this study.
Table 3.9. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Social Disorganization Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -2.07 2.67 .13 Squared Measure of Time .77 .64 2.15 Cubic Measure of Time -.08 .06 .93 Fourth-order Measure of Time .003 .002 1.003 Divorce Rate -.08 .17 .92 Rate of Females in the Labor Force -.08 .05 .92 Percentage Living in Poverty -.05 .07 .95
113 Structural-Functional Internal Determinants
Surprisingly, and contrary to Hypothesis 7, none of the three structural-functional internal determinants examined here were significantly associated with the risk or timing of adopting a registration/notification policy (see Table 3.10). States with more punitive criminal justice policies (i.e., higher incarceration rates and the use of the death penalty) were expected to adopt the policy sooner, but results of the analyses did not support this expectation.
Table 3.10. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute Based on Structural-Functional Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.95*** 1.15 .003 Squared Measure of Time 1.69*** .32 5.40 Cubic Measure of Time -.16*** .03 .85 Fourth-order Measure of Time .01*** .001 1.01 Murder Rate .09 .08 1.10 Incarceration Rate -5.02 3.36 .01 Death Penalty .25 .22 1.28 *** p-value <.001
Full Internal Determinants Model
The significant predictors from all of the preceding internal determinants models were included in a full model (see Table 3.11). Two significant internal determinant predictors of the risk and timing of registration/notification policy adoption, region and governor‟s party, remained significant in the full model. The South was significantly less likely than all other regions of the country to adopt the registration/notification policy, or did so later than states in other regions of the country. In the demographic internal
114 determinants model, there was no significant difference between the Northeastern and
Southern regions in the risk and timing of adopting a registration/notification policy.
However, in this full model the difference became significant, and the effect was consistent with Hypothesis 3. The difference between the Western and Southern regions remained significant, and consistent with Hypothesis 2. The significant difference between the Midwestern and Southern regions that was reported in Table 3.4 was also found here.
As expected, states with Republican governors were more likely to adopt registration/notification policies, or to adopt them sooner, than were states with
Democratic governors. This finding is consistent with Hypothesis 4.
Table 3.11. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Internal Determinants Model Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.49*** .002 .002 Squared Measure of Time 1.79*** 1.85 6.01 Cubic Measure of Time -.17*** .02 .84 Fourth-order Measure of Time .01*** .001 1.01 Region (South is the reference category) Northeast .47*** .16 1.60 Midwest .41*** .15 1.51 West .33*** .10 1.39 Republican Governor (Democrat is the reference .37* .26 1.44 category; 4 Independent observations dropped) * p-value < .05 *** p-value <.001
Diffusion
Three operationalizations of diffusion are included in this research. First, the percentage of neighboring states that had adopted, defined as states showing a border,
115 was tested. Second, the percentage of states within the region of the country that had adopted the policy was tested. Third, the percentage of states within the continental
United States that adopted the policy was tested. The results of these three analyses are presented in the section that follows.
Diffusion at the Neighboring Level
The first measure of traditional diffusion that was included in this analysis was the percentage of states sharing a border that had adopted the registration/notification policy. As shown in Table 3.12, the effect of neighboring states adopting a registration/notification policy was both negative and significant. As the percentage of states that had adopted the policy increased, the risk of adopting the policy for remaining states decreased or the timing of adoption was later. This pattern is contrary to
Hypothesis 8, that there would be a significant positive effect of diffusion at the neighboring level, and may suggest that states hesitated to implement the policy change based on what they were seeing in other states.
Table 3.12. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Neighboring) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.78*** .95 .003 Squared Measure of Time 1.59*** .26 4.88 Cubic Measure of Time -.15*** .02 .86 Fourth-order Measure of Time .01*** .001 1.01 Neighboring -1.39* .63 .25 * p-value < .05 *** p-value <.001
116 Regional Diffusion
The regional measure of diffusion was highly correlated with time. The correlation of the regional measure of diffusion with the linear measure of time was .77, with the squared measure of time was .89, with the cubed measure of time was .94, and with the fourth-order measure of time was .96. Thus, the regional diffusion variable that was calculated as the percentage of states within the region that had adopted the policy could not be included in addition to the time variables without introducing multicollinearity problems. Given that the correlations were so strong and positive, the measures of time themselves are probably indicators of the role of regional diffusion.
In order to assess the role of regional diffusion, the four regions of the country were examined separately with the measures of time included as the only predictor variables. Table 3.13 shows the impact of time on the risk and timing of adopting a registration/notification policy in states in the Northeast region.7 Linear, squared, and cubic measures of time were all significantly associated with the risk and timing of policy adoption.
Table 3.13. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Northeast Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.54*** .71 .03 Squared Measure of Time .63*** .14 1.88 Cubic Measure of Time -.03*** .01 .97 *** p-value <.001
7 The same procedure for specifying time that was described on pages 105-107 was used to determine the appropriate measure of time within each region.
117 In order to illustrate the magnitude of these effects, Figure 3.9 demonstrates the change in the hazard rate over the 14 years of analysis, by graphing the impact of these time variables. It is evident from Figure 3.9 that the risk of adopting a registration/notification policy differed significantly and substantially over time. In the early years, when few states within the region had adopted a registration/notification policy, the risk of remaining states adopting the policy was very low. Approximately half way through the period of study, the risk of adopting the policy had increased substantially. However, in the final years of analysis, the risk of adopting a registration/notification policy spiked substantially. This meant that remaining states were extremely likely to adopt the policy, as well.
200 180 160 140
120 100 80 60 40 20 0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Figure 3.9: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Northeast Region
118 The findings for the Midwestern region, presented in Table 3.14 and Figure 3.10,
were consistent with the findings for the Northeastern region. That is, states were at a
very low risk of adopting a registration/notification policy in early years of analysis, but
gradually their risk increased. By the final years of analysis it was almost certain that the
remaining states would adopt registration/notification policies themselves.
Table 3.14. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (Midwest Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -4.41*** .66 .01 Squared Measure of Time .79*** .15 2.21 Cubic Measure of Time -.03*** .01 .97 *** p-value <.001
350
300
250
200
150
100
50
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Figure 3.10: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Midwestern Region
119 In the Southern region, a fourth-order measure of time, in addition to the linear, squared, and cubed terms significantly improved the fit of the model. The significant effects of the four measures of time are presented in Table 3.15.
Table 3.15. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (South Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.18*** 1.17 .002 Squared Measure of Time 1.78*** .40 5.93 Cubic Measure of Time -.18*** .05 .83 Fourth-order Measure of Time .01*** .002 1.006 *** p-value <.001
Figure 3.15 depicts the effect of time on the risk and timing of policy adoption in
Southern states. Much like the previous two figures, during the early years of analysis, the risk of adopting a registration/notification policy was extremely low. However, by the end of the period of analysis the risk of adoption was substantially higher. That is, all states that had not yet adopted a registration/notification policy were at an extremely high risk of doing so.
120
800 700
600
500
400
300
200
100
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Figure 3.11: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Southern Region
The effect of time in the Western region was consistent with the results in the
Northeast and Midwest. Results are presented in Table 3.16 and Figure 3.12. Linear, squared and cubic measures of time were all significantly associated with the risk and timing of policy adoption.
Table 3.16. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (West Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -2.40*** .57 .09 Squared Measure of Time .40*** .12 1.49 Cubic Measure of Time -.02** .01 .98 *** p-value < .001
121
60
50
40
30
20
10
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Figure 3.12: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Western Region
Overall, there was a significant and positive impact of diffusion at the regional level. In the Northeast, Midwest, and West, a cubic time trend was present. In the South a fourth-order measure of time was also significantly associated with the risk and timing of policy adoption. In each region, the time trend indicated that over time, as more states adopted registration/notification policies, the risk of the remaining states adopting the policy increased to near certainty. These findings support Hypothesis 9, that there would be a positive and significant effect of regional diffusion on the risk and timing of policy adoption.
National
The correlation between the original measure of national diffusion and the measures of time were so highly correlated that it can be argued that they are essentially measuring the same thing. The correlation of the linear measure of time with the national
122 measure of diffusion was .80 (with the squared measure of time r = .92, with the cubed measure of time r = .97, and with the fourth-order measure of time r = .99). Thus, the measure of national diffusion was excluded from this analysis, and only time variables were included as predictor variables (see Table 3.17).
Table 3.17. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Traditional Diffusion (National) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.01*** 1.07 .002 Squared Measure of Time 1.67*** .30 5.34 Cubic Measure of Time -.16** .03 .85 Fourth-order Measure of Time .01** .001 1.01 *** p-value <.001
To illustrate the significant effect of time, Figure 3.13 shows the change in the hazard rate over the 14 years of analysis. In the early years, the risk of adopting a registration/notification policy was near zero. In the early 1990‟s, when a handful of states had adopted the policy, the risk of the remaining states adopting it as well increased substantially. However, by the final years of analysis, the risk of adopting a registration/notification policy was extremely high, meaning that all remaining states were very likely to adopt the policy. This finding supports Hypothesis 10, that there would be a positive effect of national diffusion on the risk and timing of adopting a registration/notification policy.
123
700
600
500
400
300
200
100
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Figure 3.13: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, National
Physical Diffusion Characteristics
As shown in Table 3.18, none of the physical characteristics of the state were
significantly associated with the risk or timing of adopting a sex offender
registry/notification policy. This finding is contrary to both Hypotheses 11 and 12, that
states‟ levels of physical connectedness to other states and within the nation, respectively,
would impact the risk and timing of their policy adoption.
124
Table 3.18. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Physical Diffusion Characteristics Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.40** 1.56 .005 Squared Measure of Time 1.53*** .42 4.61 Cubic Measure of Time -.15*** .04 .86 Fourth-order Measure of Time .005*** .001 1.005 Length of Border -.0003 .0004 .9997 Shared Border -.36 .61 .70 Neighbors .02 .14 1.02 Interstates -.04 .09 .97 Airport Hub .20 .63 1.22 * p-value < .05 ** p-value <.01 *** p-value <.001
Diffusion Processes – Hub States
As shown in Table 3.19, states that were identified as leaders in policy innovation by Walker (1969), but were no more or less likely than other states to adopt the policy (or to do so either sooner or later). Although inconsistent with Hypothesis 13, this finding may be explained by the fact that Walker‟s analysis was conducted 42 years ago.
125
Table 3.19. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Networks (Walker) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.99*** 1.07 .002 Squared Measure of Time 1.67*** .30 5.33 Cubic Measure of Time -.16*** .03 .85 Fourth-order Measure of Time .01*** .001 1.01 Walker‟s Hub States -.21 .42 .81 *** p-value <.001
Canon and Baum‟s identification of innovative states was 12 years more recent than Walker‟s. However, contrary to Hypothesis 14, the risk and timing of policy adoption was also not significantly associated with being identified as an innovative state by Canon and Baum (1981). These findings, shown in Table 3.20, may suggest that registration/notification statutes are not, in fact, innovative statutes. They were adopted relatively quickly across the 50 states (within a decade) and were federally required.
Thus, being a state that is willing to implement new legislative actions may not be as relevant when the policy being considered was not, in itself, innovative.
Table 3.20. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Networks (Canon & Baum) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.998*** 1.05 .002 Squared Measure of Time 1.67*** .29 5.33 Cubic Measure of Time -.16*** .03 .85 Fourth-order Measure of Time .01*** .001 1.01 Canon & Baum‟s Hub States -.11 .55 .89 *** p-value <.001
126 Diffusion Processes – Federal Government
Neither of the federal initiatives governing the implementation of registries and notification policies had a significant impact on the risk or timing of policy adoption (see
Table 3.21). This result is surprising, and contrary to Hypothesis 15a, because these
Federal laws required states to take action on this policy and gave them a timeline for doing so.
Table 3.21. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Federal Government Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.24*** 1.17 .002 Squared Measure of Time 1.87*** .36 6.49 Cubic Measure of Time -.20*** .04 .82 Fourth-order Measure of Time .01*** .001 1.01 Jacob Wetterling Act 1.36 .72 3.90 Megan‟s Law -.22 1.53 .80 *** p-value <.001
Diffusion Processes – Media
Media coverage throughout the nation was not significantly associated with the risk or timing of adopting a registration/notification policy (see Table 3.22). According to Hypothesis 16, the amount of media coverage of sex offenders was expected to be positively related to the risk and timing of policy adoption. The fact that there was no significant relationship suggested that registration/notification legislation was passed based on factors other than pressure placed on legislators by the media.
127
Table 3.22. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Media Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.10*** 1.65 .002 Squared Measure of Time 1.70** .54 5.49 Cubic Measure of Time -.17** .06 .85 Fourth-order Measure of Time .01** .002 1.01 Media Coverage .0001 .001 1.0001 ** p-value <.01 *** p-value <.001
Diffusion Processes – Social Movement Organizations
Contrary to both Hypotheses 17a and 18a, neither the presence of a chapter of the
“Association for the Treatment of Sexual Abusers” nor of “Stop it Now!” was associated with the risk or timing of policy adoption across states (see Table 3.23). This finding is somewhat surprising, given that both groups advocate for the treatment of sexual offenders. However, both groups also advocate for public education, so perhaps these two stances neutralize each other, resulting in a non-significant effect.
Table 3.23. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Social Movement Organizations Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -5.93*** 1.34 .003 Squared Measure of Time 1.65*** .45 5.23 Cubic Measure of Time -.16** .05 .85 Fourth-order Measure of Time .01** .002 1.01 “Association for the Treatment of Sexual Abusers” -.20 .38 .82 “Stop it Now!” .61 .74 1.85 ** p-value <.01 *** p-value <.001
128 Full Diffusion Model
Because the only significant diffusion predictors of adopting a registration/notification policy were traditional measures, and the effect of diffusion at the neighboring-level was opposite of the effect of both regional and national diffusion, full models were run testing multiple diffusion measures. The full model shown in Table
3.24 includes the traditional measure of diffusion at the neighboring level as well as all of the diffusion processes that were tested. This model tested the robustness of the traditional diffusion effect at the neighboring level. Once all of these diffusion processes were taken into account, the effect of diffusion at the neighboring state level was attenuated to non-significance.
The only variable that was statistically related to the risk and timing of policy adoption was the presence of a chapter of “Stop it Now!” in the state. Recall that the presence of a “Stop it Now!” chapter was not associated with the timing of policy adoption in Table 3.23. Only after introducing variables relevant to the amount of interaction between neighboring states did “Stop it Now!” become significant, indicating a suppression effect.
The four variables that, when included with “Stop it Now!,” produced a suppression effect were the traditional measure of diffusion (neighboring), the number of interstates running through the state, the number of neighbors that a state had, and the percentage of the border that was shared with other states. “Stop it Now!” was present in three states during the period of analysis covered here: Georgia, Pennsylvania, and
Massachusetts.
129 This model accounted for some of the physical characteristics of the states, which likely led to the grouping of certain states. For example, there are many large states out
West (e.g., Wyoming, Idaho) that have few interstates and share most or all of their border with another state. The three “Stop it Now!” chapters that were in place during the period of analysis were located exclusively along the East coast in states with many interstates and only a portion of their border shared (due to their coastal locations). Once the physical characteristics of the state were taken into account, by controlling for these variables, the presence of a “Stop it Now!” chapter was able to reach significance.
Table 3.24. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Diffusion Model (Neighboring) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.38*** 1.43 .002 Squared Measure of Time 1.93*** .40 6.87 Cubic Measure of Time -.20*** .04 .82 Fourth-order Measure of Time .01*** .001 1.01 Neighboring -.96 1.01 .38 Length of Border -.0004 .001 .9996 Shared Border -.29 .84 .75 Neighbors .04 .13 1.04 Interstates -.03 1.14 .97 Jacob Wetterling Act .31 2.71 1.37 Megan‟s Law -2.13 2.71 .12 Media Coverage .002 .001 1.002 “Association for the Treatment of Sexual Abusers” -.13 .45 .88 “Stop it Now!” .65* .27 1.91 * p-value < .05 *** p-value <.001
Table 3.25 shows the results of a model including time variables that represent the national measure of diffusion with all diffusion process variables included. All time variables remained significant indicating that as more states in the nation adopted a
130 registration/notification policy, the risk of other states adopting the policy increased.
This finding is consistent with what was reported in Table 3.24 and supports Hypothesis
10, that the national diffusion effect would be positive.
The effect of “Stop it Now!” remained positive and significant when controlling for diffusion at the national level, rather than diffusion at the neighboring level. This finding suggests that the presence of a “Stop it Now!” chapter is a significant indicator of adopting a registration/notification policy, regardless of which significant traditional diffusion measure is included in the model.
Table 3.25. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Model (National) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.23*** 1.52 .002 Squared Measure of Time 1.91*** .42 6.74 Cubic Measure of Time -.20*** .04 .82 Fourth-order Measure of Time .01*** .001 1.01 Length of Border -.0003 .001 .9997 Shared Border -.47 .73 .63 Neighbors .03 .13 1.03 Interstates -.04 .08 .96 Jacob Wetterling Act .63 1.71 1.88 Megan‟s Law -1.79 3.52 .17 Media Coverage .001 .002 1.001 “Association for the Treatment of Sexual Abusers” -.11 .47 .90 “Stop it Now!” .66* .27 1.94 * p-value < .05 *** p-value <.001
131 Full Model
Table 3.26 shows the results of a full model including all of the variables that were significantly associated with the timing of adoption of a registration/notification policy in the preceding models. The effects of each dummy variable representing region of the country remained positive and significant. This finding was consistent with
Hypothesis 2 and Hypothesis 3 and indicated that states in the South were least likely to adopt registration/notification policies, or did so later than states in the Northeast,
Midwest, or West.
Consistent with Hypothesis 4, states with Republican governors were at an increased risk of adopting a registration/notification policy, or adopting such a policy sooner, compared to states with Democratic governors.
The final significant relationship was the effect of national diffusion on the timing of policy adoption. The coefficients for each measure of time were significant, indicating that over time, as other states in the nation adopted a registration/notification policy, the likelihood of the remaining states adopting the policy increased significantly and substantially. This finding supports Hypothesis 10.
“Stop it Now!” was no longer significant, however, when none of the variables related to interaction with neighboring states were included in this analysis. The model shown in Table 3.26 was also run using the traditional measure of diffusion at the neighboring level, rather than the national level, and including the additional variables that appeared to suppress the effect of having a “Stop it Now!” chapter in the state according to Table 3.24. The coefficient for “Stop it Now!” remained non-significant
132 even when these additional variables were included. Overall, this analysis does not support Hypothesis 18b.
Table 3.26. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Statute: Full Model Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -6.50*** 1.13 .002 Squared Measure of Time 1.79*** .30 5.99 Cubic Measure of Time -.17*** .03 .84 Fourth-order Measure of Time .006*** .001 1.01 Region (South is the reference category) Northeast .40*** .10 1.48 Midwest .47*** .12 1.61 West .39*** .10 1.48 Republican Governor (Democrat is the reference .35 .20 1.42 category; 4 Independent observations dropped) “Stop it Now!” .52 .35 1.68 *** p-value <.001
Summary of Registration/Notification Findings
The results of the preceding analyses partially supported and partially refuted the hypotheses set forth in this study. The results supported Hypothesis 1, that none of the demographic characteristics of the state (population, population density, and per capita income) would affect the risk or timing of adopting a registration/notification policy.
Support was also found for Hypotheses 2 and 3: that states in the West and Northeast, respectively, would be at an increased risk of policy adoption compared to states in the
South. In addition, the analyses indicated that states in the Midwest were significantly more likely than states in the South to adopt registration/notification policies.
Hypothesis 4 was partially supported. States with Republican governors were significantly more likely to adopt registration/notification statutes or to adopt them
133 sooner, compared to states with Democratic governors. However, the percentage of
Republican citizens in the state did not significantly affect the risk or timing of policy adoption. Of the remaining internal determinants hypotheses, no support was found for
Hypothesis 5 (regarding racial threat internal determinants), Hypothesis 6 (regarding social disorganization internal determinants), or Hypothesis 7 (regarding structural- functional internal determinants).
Among the hypotheses regarding traditional measures of diffusion, two were supported. Both the regional and national measures of diffusion were significantly related to the timing of policy adoption, as anticipated. In multiple models, national diffusion had a negative effect on the risk and timing of adopting a registration/notification policy. Only Hypothesis 8, regarding the measure of diffusion at the neighboring level, was not supported.
Most of the hypotheses regarding the diffusion processes tested here were not supported. Specifically, no support was found for Hypothesis 11(regarding connectedness with other states), Hypothesis 12 (regarding connectedness with the nation), Hypothesis 13 (regarding Walker‟s innovation hubs), Hypothesis 14 (regarding
Canon and Baum‟s innovation hubs), Hypothesis 15a (regarding Federal legislation),
Hypothesis 16 (regarding media), Hypothesis 17a (regarding the presence of an American
Society for the Treatment of Sexual Abusers chapter), and Hypothesis 18a (regarding the presence of an “Stop it Now!” chapter).
Though civil commitment and registration/notification are both policies intended to protect society from sexual offenders, they differ in important ways and these differences may be reflected in the factors that impacted their adoption. The following
134 section uses event history analysis to test for the state-level factors and diffusion processes that are associated with the timing of adoption of civil commitment statutes within each state.
Results of an Event History Analysis Predicting the Adoption of a Civil Commitment Statute
Specifying Time
As with registration/notification, an appropriate measure of time was determined for the civil commitment models. Table 3.27, below, shows the values produced by models with each functional form of time that may be appropriate for these analyses. The method for determining the best model fit was the same as that described on pages 115 and 115. However, because the length of time included in analyses to predict the adoption civil commitment statutes was longer (22 years) than the length of time included in analyses to predict the adoption of a registration/notification statute (14 years), the general model required 21 degrees of freedom.
135
Table 3.27. Specifying Time for Civil Commitment Models 2 x [(-2 Log Likelihood of General Model) Measure -2 Log Likelihood Df – (-2 Log Likelihood)] General -76.22 0 21 Logged -151.13 149.82 1 Linear -177.06 201.68 1 Squared -130.90 109.36 2 Cubic -110.01 67.58 3 Fourth- -91.88 4 order 31.32 Fifth- -90.91 5 Order 29.38
Once again, the model with the lowest -2 Log Likelihood was the general model.
However, the 21 degrees of freedom required by the general model introduces substantial complexity, which could be reduced by using an alternate specification of time. As with the registration/notification analysis, both the logged model and the linear model, though they require only one degree of freedom, were poorly suited to the data.
The squared model was a significant improvement over the linear model (χ2 =
86.59, 1 df, p < .001), but was still inferior to the general model, as was the cubic model, though it was a significant improvement over the squared model (χ2 = 30.43, 1 df, p <
.001). Consistent with the registration/notification model, the fourth-order model for civil commitment provided a good fit to the data (χ2 = 26.37, 1 df, p < .001) compared to the cubic model. A fifth-order model decreased the -2 Log Likelihood difference slightly, but not enough to warrant an increase in the degrees of freedom (χ2 = 1.93, 1 df, p = .16).
Thus, the most appropriate specification of time for this model is the fourth-order model. Although it has a somewhat higher -2 Log Likelihood than does the general
136 model (-91.88 and -76.22, respectively), the 17 degrees of freedom that are saved by using the fourth-order model rather than the general model make the fourth-order model a more parsimonious and desirable choice. In the subsequent analyses, all of the event history models used to assess the role of state-level factors on the adoption of civil commitment policies included four terms (a linear term, a squared term, a cubic term, and a fourth-order term) to account for this specification of time.
Demographic Internal Determinants
Two demographic internal determinants were significantly related to the timing of adopting a civil commitment statute: population and region (see Table 3.28). Consistent with Hypothesis 1b, states with larger populations were slightly but significantly more likely to adopt a civil commitment statute early, compared to states with smaller populations. In fact, an increase in the population (measured in 100,000‟s) increased the odds of policy adoption by 9%. Also, compared to the South, states located in the
Midwest of the country were significantly and substantially more likely to adopt civil commitment statutes sooner. This finding is contrary to both Hypothesis 2 and
Hypothesis 3, but is consistent with the finding that Midwestern states are more likely than Southern states to adopt registration/notification policies.
137
Table 3.28. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Demographic Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -4.03*** .82 .02 Squared Measure of Time .78*** .12 2.19 Cubic Measure of Time -.05*** .01 .95 Fourth-order Measure of Time .001*** .0001 1.001 Population .09** .03 1.09 Population Density .19 1.58 1.21 Per Capita Income -.02 .02 .98 Region (South is the reference category) Northeast .39 .56 1.47 Midwest 1.10** .42 3.01 West -.13 .43 .88 ** p-value <.01 *** p-value <.001
Political Internal Determinants
States with Republican governors were significantly less likely to adopt civil commitment statutes, or adopted them later, than states with Democratic governors. This finding is contrary to Hypothesis 4, that states with Republican governors would be at an increased risk of adopting a civil commitment statute. This hypothesis was based on the fact that states run by Republican governors tend to be more punitive. However, the fact that Republican politicians also tend to oppose “Big Government” applies to this finding in two ways. First, the cost of civil commitment is high (approximately $100,000 per offender, per year), a factor that may deter Republicans from adopting the policy.
Second, Republicans generally do not like to see the government involved in people‟s personal lives. A civil commitment statute would run contrary to this ideology,
138 especially because it is implemented after the term imposed by the criminal justice system is complete. None of the other political internal determinants was significantly associated with the timing of adopting registration/notification policy (see Table 3.29).
Table 3.29. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Political Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.60*** .62 .03 Squared Measure of Time .72*** .11 2.05 Cubic Measure of Time -.05*** .01 .95 Fourth-order Measure of Time .001*** .0002 1.001 Republican Governor (Democrat is the reference -.63** .19 .53 category; 12 Independent observations dropped) Percentage Republican -.02 .02 .98 Percentage Females in the Legislature .02 .05 1.02 Professionalism .89 1.92 2.42 Competition -.01 .02 .99 Innovation 3.03 2.16 20.69 ** p-value <.01 *** p-value <.001
Racial Threat/Conflict Internal Determinants
Although the racial composition of a state was not associated with the risk or timing of adopting a civil commitment statute, the unemployment rate had a significant, negative effect (see Table 3.30). States with higher unemployment rates were less likely to adopt civil commitment statutes, or adopted them later, compared to states with lower unemployment rates. This finding is contrary to Hypothesis 5, that factors related to the presence of a racial threat in a state would be associated with the timing of adopting a civil commitment statute. The fact that the unemployment rate was negatively related to the risk and timing of adopting a civil commitment statute may suggest that the economy
139 is an important consideration (note that this effect was opposite of the expected direction in Hypothesis 5). States with weaker economies, here indicated by a higher unemployment rate, may be hesitant to adopt civil commitment statutes because these policies are so costly to implement and maintain.
Table 3.30. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Racial Threat/Conflict Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -2.24*** .46 .11 Squared Measure of Time .46*** .09 1.59 Cubic Measure of Time -.03*** .01 .97 Fourth-order Measure of Time .001*** .0002 1.001 Percentage Hispanic .04 .03 1.04 Percentage Black -.002 .02 .998 Unemployment Rate -.33* .13 .72 * p-value < .05 *** p-value <.001
Social Disorganization Internal Determinants
Only one of the three indicators of social disorganization was significantly associated with the risk and timing of adopting a civil commitment statute (see Table
3.31). States with higher divorce rates were less likely to adopt the policy, or adopted it later, compared to states with lower divorce rates. This finding provides partial support for Hypothesis 6, that socially disorganized states would require the existence of a formal control mechanism (civil commitment) due to a lack of informal social control within the community.
140
Table 3.31. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Social Disorganization Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -1.73** .65 .18 Squared Measure of Time .39*** .11 1.48 Cubic Measure of Time -.03*** .01 .97 Fourth-order Measure of Time .001*** .0002 1.001 Divorce Rate -.34** .12 .71 Females in the Labor Force -.01 .02 .99 Poverty -.06 .06 .94 ** p-value <.01 *** p-value <.001
Structural-Functional Internal Determinants
Contrary to Hypothesis 7, no structural-functional characteristics of the state were associated with the risk and timing of adopting a civil commitment statute. States that appear to be more punitive in other respects (for example, by allowing the death penalty or having a high incarceration rate) were at no greater or lesser risk of adopting a civil commitment statute compared to states that are generally non-punitive (see Table 3.32).
Table 3.32. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute Based on Structural-Functional Internal Determinants Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.01*** .37 .05 Squared Measure of Time .60*** .07 1.83 Cubic Measure of Time -.04*** .01 .96 Fourth-order Measure of Time .001*** .0001 1.001 Murder Rate -.03 .19 .97 Incarceration Rate -1.0001 5.49 .37 Death Penalty -.40 .55 .67 *** p-value <.001
141 Full Internal Determinants Model
As shown in Table 3.33, only significant variables from the preceding models were included in the full model. States with higher populations remained at an increased risk of adopting a civil commitment statute, or of adopting the policy sooner, than states with lower populations. This finding provided support for Hypothesis 1.
Contrary to Hypothesis 4, the effect of having a Republican governor became non-significant in the full model. This finding, combined with the fact that the percentage of Republicans in the state was non-significant, provided no support for
Hypothesis 4.
Compared to the South, states in the Midwest remained at a significantly higher risk of adopting the policy or doing so sooner. Neither states in the Northeast nor the
West differed significantly from Southern states, consistent with Hypothesis 2, but contrary to Hypothesis 3.
States with higher unemployment rates remained less likely to adopt registration/notification policies than were states with lower unemployment rates. This finding, which provided partial support for Hypothesis 5, suggested that the state‟s economic characteristics may have an effect on the risk and timing of adopting a civil commitment statute.
Although the coefficient remained negative, the effect of divorce rate on policy adoption became non-significant and provided no support for Hypothesis 6.
142
Table 3.33. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Internal Determinants Model Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -2.04*** .13 .13 Squared Measure of Time .43*** .03 1.54 Cubic Measure of Time -.03*** .003 .97 Fourth-order Measure of Time .001*** .0001 1.001 Population .11* .05 1.12 Republican Governor (Democrat is the reference .09 .20 1.09 category; 12 Independent observations dropped) Region (South is the reference category) Northeast .02 .20 1.02 Midwest .88*** .24 2.42 West -.10 .28 .90 Unemployment Rate -.35* .17 .71 Divorce Rate -.24 .13 .78 * p-value < .05 *** p-value <.001
Diffusion
As in the analysis of the adoption of registration/notification policies, all three operationalizations of diffusion were tested here. The modifications made to the regional and national diffusion models in the previous results section were also necessary here.
The findings of each set of analyses are reported in the following section.
Diffusion at the Neighboring Level
Diffusion defined as neighboring states (i.e., sharing a border) was not significantly associated with the risk of adopting a civil commitment statute (see Table
3.34). This result, which is contrary to Hypothesis 8, may be due to the fact that less than
143 half of the states have adopted civil commitment statutes, so the amount of variation in this variable was somewhat limited.
Table 3.34. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Neighboring) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.49*** .39 .03 Squared Measure of Time .68*** .05 1.98 Cubic Measure of Time -.05*** .002 .95 Fourth-order Measure of Time .001*** .0001 1.001 Neighboring -.90 2.93 .41 *** p-value <.001
Regional Diffusion
As with the analysis of registration/notification policies, the effect of regional diffusion could not be measured directly due to high correlations between that variable and the measures of time (The correlation of the regional measure of diffusion with the linear measure of time was .75, with the squared measure of time was .75, with the cubed measure of time was .71, and with the fourth-order measure of time was .67). Thus, the same procedure used in the analysis of the registration/notification policy, page 115, was also used to analyze the adoption of civil commitment statutes. Findings for the
Northeast region are presented in Table 3.35. Four measures of time contributed significantly to the fit of the model: linear, squared, cubic, and a fourth-order measure.
144
Table 3.35. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Northeast Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -4.003*** 1.01 .02 Squared Measure of Time .78*** .22 2.18 Cubic Measure of Time -.05** .02 .95 Fourth-order Measure of Time .001** .0004 1.001 ** p-value <.01 *** p-value <.001
To help illustrate the effects of time, Figure 3.14 shows the changing hazard rate
over the 22-year period of analysis. For the first several years of analysis, the risk of
adopting a civil commitment statute was very low. However, over time this effect
increased significantly. That is, by the end of the period of analysis, states in the
Northeast that had not yet adopted a civil commitment statute were at an increased risk of
doing so.
40
35
30
25
20
15
10
5
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Figure 3.14: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Northeast Region
145
The findings of the Midwestern Region, shown in Table 3.36 and Figure 3.15., are consistent with the findings of the Northeastern region. That is, in early years of analysis the risk of adopting a civil commitment statute was very low, but by the end of the period of analysis, this risk had increased significantly and substantially.
Table 3.36. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (Midwest Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -2.96*** .65 .05 Squared Measure of Time .62*** .16 1.85 Cubic Measure of Time -.04*** .01 .96 Fourth-order Measure of Time .001*** .0003 1.001 *** p-value < .001
20 18
16 14 12 10 8 6 4 2
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Figure 3.15: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Midwestern Region
146 In the Southern region, the appropriate specification of time included a linear, squared, and cubic measure, but not a fourth-order measure of time. These three coefficients are presented in Table 3.37 and graphed in Figure 3.16. Overall, the effect of regional diffusion in the South, represented by these time variables, is consistent with the findings for the Northeast and Midwest. That is, in early years, states are at a low risk of adopting the policy but in later years, this risk increased significantly.
Table 3.37. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (South Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.96*** .75 .02 Squared Measure of Time .52*** .13 1.68 Cubic Measure of Time -.02*** .01 .98 ** p-value <.01 *** p-value <.001
500
450 400 350 300
250 200 150 100 50 0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Figure 3.16: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Southern Region
147 The effect of time in the Western region, shown in Table 3.38 and Figure 3.17, was consistent with the findings of the South, in that only the first three measures of time contributed significantly to the fit of the model.
Table 3.38. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (West Region) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.28** 1.28 .04 Squared Measure of Time .60* .25 1.81 Cubic Measure of Time -.03* .01 .97 * p-value <.05 ** p-value <.01
700
600
500
400
300
200
100
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Figure 3.17: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, Western Region
148 Consistent with the findings in the other regions, the overall impact of diffusion in the West was positive. In the final years of analysis, the risk of policy adoption were significantly higher than it was in the early years of analysis.
As with the analysis of registration/notification statutes, the effect of regional diffusion was significantly associated with the risk of adopting a civil commitment statute. This effect was ultimately positive, consistent with Hypothesis 9. Over time, as more states within each region adopted a civil commitment statute, the risk of the remaining states adopting the policy also increased.
National Diffusion
The national measure of diffusion was also highly correlated with the measures of time, which precluded a model including the national diffusion measure along with the measures of time (The correlation of the regional measure of diffusion with the linear measure of time was .97, with the squared measure of time was .96, with the cubed measure of time was .90, and with the fourth-order measure of time was .85). Ultimately, the measures of time represent the process of national diffusion. The effects are presented in Table 3.39 and Figure 3.18. The findings of this analysis to predict the risk and timing of adopting a civil commitment statute were consistent with the findings of analysis of the registration/notification policy. In the early years of analysis, the risk of adopting a civil commitment statute was extremely low. However, by the end of the 22- year period of analysis, the risk of adopting a civil commitment policy had increased significantly and substantially. This finding was consistent with Hypothesis 10, which stated that as more states within the nation adopted the poloicy, the likelihood of the remaining states adopting the policy increased.
149
Table 3.39. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Traditional Diffusion (National) Robust Variable Coefficient Standard Odds Ratio Error Linear Measure of Time -3.48*** .39 .03 Squared Measure of Time .68*** .05 1.98 Cubic Measure of Time -.05*** .003 .95 Fourth-order Measure of .001*** .0001 1.001 Time ** p-value <.01 *** p-value <.001
30
25
20
15
10
5
0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Figure 3.18: Changes in the Risk and Timing of Adopting a Registration/Notification Policy as a Result of Time, National Diffusion
Physical Diffusion Characteristics
Contrary to Hypothesis 11, none of the physical characteristics of the state had a significant impact on the risk or timing of policy adoption (see Table 3.40). This finding is surprising because it indicates that a state‟s physical connectedness with other states does not impact its legislative activity. Furthermore, and contrary to Hypothesis 12, a
150 state‟s level of connectedness within the nation, measured by the number of interstates and the presence of a major airport hub, did not significantly impact its risk of adopting a civil commitment statute.
Table 3.40. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Physical Diffusion Robust Variable Coefficient Standard Odds Ratio Error Linear Measure of Time -2.81*** .71 .06 Squared Measure of Time .57*** .13 1.76 Cubic Measure of Time -.04*** .01 .96 Fourth-order Measure of .001*** .0002 1.001 Time Length of Border -.0002 .0002 .9998 Shared Border -1.50 .99 .22 Neighbors -.09 .20 .92 Interstates .12 .14 1.13 Airports .42 .35 1.52 *** p-value <.001
Diffusion Processes – Hub States
Contrary to Hypothesis 13, being an innovative state, according to Walker‟s
(1969) classification, was not significantly related to the risk or timing of adopting a civil commitment statute. However, as noted earlier, Walker‟s study is more than 40 years old. Thus, a more recent study of state innovativeness may provide support for
Hypothesis 14.
151
Table 3.41. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Hub States (Walker) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.57*** .51 .03 Squared Measure of Time .70*** .08 2.01 Cubic Measure of Time -.05*** .004 .95 Fourth-order Measure of Time .001*** .0001 1.001 Walker‟s Hub States .67 .81 1.96 *** p-value <.001
In fact, being identified as an innovative state in Canon and Baum‟s (1981) study was significantly and positively related to the risk or timing of adopting a civil commitment statute (see Table 3.42). These states were at an increased risk of adopting a civil commitment statute, or of doing so earlier, consistent with Hypothesis 14. Unlike registration/notification, civil commitment statutes are in place in fewer than half the states and are not Federally required. Thus, the fact that these laws are more likely to be adopted or adopted sooner in more innovative states is not surprising.
Table 3.42. Event History Analysis Predicting the Timing of Adopting a Registration/Notification Based on Diffusion Processes: Hub States (Canon & Baum) Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.71*** .37 .02 Squared Measure of Time .72*** .05 2.06 Cubic Measure of Time -.05*** .003 .95 Fourth-order Measure of Time .001*** .0001 1.001 Canon & Baum‟s Hub States 1.39** .53 4.02 ** p-value <.01 *** p-value <.001
152 Diffusion Processes – Federal Government
Consistent with Hypothesis 15b, two federal initiatives were significantly associated with the risk and timing of adopting a civil commitment statute (see Table
3.43). After the adoption of the Jacob Wetterling Act (1994) and after the Supreme Court
Case of Kansas v. Hendricks (1997), which upheld the constitutionality of civil commitment, the risk of adoption decreased significantly. This negative relationship could indicate that states waited prior to adopting the statute to see how the new legislation was received by the public in their state. The Adam Walsh Act was not significantly related to the timing of adopting a civil commitment statute, as hypothesized, but this non-significant effect is probably due to the fact that the law was passed in 2006 and the analysis presented here ended in 2007.
Table 3.43. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Federal Government Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -4.04*** .49 .02 Squared Measure of Time .78*** .11 2.18 Cubic Measure of Time -.05*** .01 .95 Fourth-order Measure of Time .001*** .0003 1.001 Jacob Wetterling Act -3.34* 1.60 .04 Megan‟s Law -1.73 1.69 .18 Kansas v. Hendricks -3.57*** .80 .03 Adam Walsh Act .18 2.97 1.19 * p-value < .05 *** p-value <.001
153 Diffusion Processes – Media
Media coverage of sexual offending was not significantly related to the risk of adopting a civil commitment statute (see Table 3.44). This finding provided no support for Hypothesis 16 and suggested that the amount of media coverage of sexual offending had no effect on the risk or timing of adopting a civil commitment statute.
Table 3.44. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Media Robust Variable Coefficient Standard Odds Ratio Error Linear Measure of Time -3.51*** .44 .03 Squared Measure of Time .69*** .08 1.997 Cubic Measure of Time -.05*** .005 .95 Fourth-order Measure of .001*** .0001 1.001 Time Media Coverage -.00003 .0002 .99997 * p-value < .05 ** p-value <.01 *** p-value <.001
Diffusion Processes – Social Movement Organizations
The presence of a social movement organization in the state did not significantly impact whether or when a state adopted a civil commitment statute (see Table 3.45).
Contrary to Hypothesis 17b, the presence of an “ATSA” chapter did not impact the timing of adopting a civil commitment statute. And, contrary to Hypothesis 18b, the presence of a “Stop it Now!” chapter did not impact the timing of adopting a civil commitment statute.
154
Table 3.45. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Based on Diffusion Processes: Social Movement Organizations Robust Odds Variable Coefficient Standard Ratio Error Linear Measure of Time -3.78*** .72 .02 Squared Measure of Time .74*** .12 2.09 Cubic Measure of Time -.05*** .01 .95 Fourth-order Measure of Time .001*** .0001 1.001 “Association for the Treatment of Sexual Abusers” .69 .74 1.9996 “Stop it Now!” -.83 1.04 .44 *** p-value <.001
Full Diffusion Model
In the full diffusion model, three variables were significantly related to the risk and timing of policy adoption. First, as more states within the nation adopted a civil commitment statute, the risk of adopting the policy in other states increased significantly.
This finding provided strong evidence that the national measure of diffusion was an important indicator of the risk and timing of adopting a civil commitment statute as stated in Hypothesis 10.
There were also significant, negative effects of both the adoption of the Jacob
Wetterling Act and the Supreme Court decision in Kansas v. Hendricks. These findings were the opposite of what were expected in Hypothesis 15a and Hypothesis 15b, respectively. Given that both of these measures are dummy variables coded high after a specific event, they are essentially capturing a time trend. Thus, it is possible that what appears to be a significant effect may in fact be the result of multicollinearity with the time variables.
155
Table 3.46. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Diffusion Model Robust Variable Coefficient Standard Odds Ratio Error Linear Measure of Time -4.03** 1.40 .02 Squared Measure of Time .77* .31 2.16 Cubic Measure of Time -.05* .02 .96 Fourth-order Measure of .001 .01 1.001 Time Length of Border -.0004 .01 .9996 Shared Border -.89 1.12 .41 Neighbors -.10 .18 .91 Interstates .11 .19 1.12 Airport .27 .58 1.31 Jacob Wetterling Act -3.56* 1.37 .03 Megan‟s Law -2.20 1.98 .11 Kansas v. Hendricks -3.19*** .80 .04 Adam Walsh Act .46 2.33 1.59 Canon & Baum‟s Hub States 2.61 2.63 13.54 Media Coverage .0001 .0002 1.0001 “Association for the Treatment of Sexual .41 .59 1.51 Abusers” “Stop it Now!” -1.01 1.16 .37 * p-value < .05 ** p-value < .01 *** p-value < .001
Full Model
The full model, including both the significant internal determinants and the significant diffusion variables, indicated that three of the effects shown above remained significant. Two additional variables were also significantly associated with the risk and timing of adopting a civil commitment statute. First, there was a positive and significant effect of population on the hazard rate, consistent with Hypothesis 1B.
156 Second, states with higher unemployment rates were less likely to adopt civil commitment statutes (or to adopt them later) than were states with lower unemployment rates. A positive relationship was hypothesized, but this finding can be explained by the fact that a state‟s economic situation is sometimes related to the way that the criminal justice system functions (Pratt & Maahs, 1999). Because these policies are expensive to implement, it makes sense that states with stronger economies were able to adopt them first.
The negative effects of the adoption of the Jacob Wetterling Act and the Supreme
Court decision in the Kansas v. Hendricks were significant, and, as above, are probably the result of multicollinearity
Also, the national diffusion effect was positive, which indicated that as more states adopted civil commitment statutes, the likelihood of adoption by the remaining states increased.
157
Table 3.47. Event History Analysis Predicting the Timing of Adopting a Civil Commitment Statute: Full Model Robust Variable Coefficient Standard Odds Ratio Error Linear Measure of Time -2.46** .77 .09 Squared Measure of Time .51* .20 1.66 Cubic Measure of Time -.03* .02 .97 Fourth-order Measure of .001 .0004 1.001 Time Population .12*** .08 1.13 Region (South is the reference
category) .17 .09 1.19 Northeast .9993 .18 2.72 Midwest .10 .19 1.11 West Republican Governor (Democrat is the reference .08 .22 1.08 category; 12 Independent observations dropped) Unemployment Rate -.37* .16 .69 Divorce Rate -.24 .09 .79 Jacob Wetterling Act -4.03* 1.58 .02 Kansas v. Hendricks -3.55** 1.20 .03 * p-value < .05 ** p-value < .01 *** p-value < .001
Summary of Civil Commitment Findings
Overall, eleven hypotheses received no support in these analyses: Hypothesis 2
(regarding the Western region), Hypothesis 3 (Northeastern region) Hypothesis 4
(regarding political internal determinants), Hypothesis 6 (regarding social disorganization, Hypothesis 7 (regarding structural-functional internal determinants),
Hypothesis 8 (regarding diffusion at the neighboring level), Hypothesis 9 (regarding diffusion at the regional level), Hypothesis 11 (regarding connectedness within the state),
158 Hypotheses 12 and 13 (regarding state hubs of innovation), Hypothesis 16 (regarding media coverage), Hypothesis 17b (regarding the presence of an “ATSA” chapter in the state), and Hypothesis 18b (regarding the presence of a “Stop it Now!” chapter in the state).
There were four primary findings regarding the adoption of civil commitment statutes across states. First, there was a negative and significant relationship between the unemployment rate in a state and the likelihood of adopting a civil commitment statute, which indicated that the state‟s economic situation may impact their legislative decisions.
This finding was opposite of what was expected in Hypothesis 5.
Two pieces of Federal legislation, the passage of Jacob Wetterling Act and the
Kansas v. Hendricks Supreme Court decision, were significantly, negatively associated with the timing of policy adoption. These negative coefficients were opposite of expectation, which is likely the result of multicollinearity with time.
Fourth, there was a strong significant effect of diffusion at both the regional and the national level. As more and more states adopted civil commitment statutes, the risk of adopting a civil commitment statute in the remaining states increased. These findings were consistent with Hypothesis 9 (regarding regional diffusion) and Hypothesis 10
(regarding national diffusion).
Discussion of Study 1
Study 1 tested the impact of state-level characteristics concerning internal determinants and diffusion mechanisms on the adoption of two types of sexual offending
159 legislation. Specifically, Study 1 used event history analysis to examine the risk and timing of adopting a registration/notification policy and the risk and timing of adopting a civil commitment statute. The results of this Study as they relate to the hypothesized relationships are discussed below.
Hypotheses
Hypothesis 1a stated that no significant relationship would be found between population, population density, or per capita income and the adoption of a registration/notification policy. However, Hypothesis 1b stated that there would be a significant effect of these variables on the adoption of a civil commitment statute. The findings of Study 1 were consistent with Hypothesis 1a, and partially consistent with
Hypothesis 1b. Because it is a more controversial statute, it was expected that the demographic characteristics of a state (particularly states with larger populations) would impact the likelihood of adopting a civil commitment statute. In fact, states with larger populations were more likely to adopt civil commitment policies, or to adopt them sooner, compared to states with smaller populations.
Partial support was found for Hypothesis 2, that Western states would be more likely to adopt both policies, or to adopt them sooner, than states in the South. In fact,
Western states were at an increased risk of adopting registration/notification but not civil commitment statutes, compared to Southern states. Support was also found for
Hypothesis 3, that Northeastern states would be more likely to adopt registration/notification policies, or to adopt them sooner, than states in the South. These
160 relationships were hypothesized because of the specific cases that took place in these two regions (i.e., the murder of Megan Kanka in New Jersey and the three assaults in two years by released sex offenders in Washington). Although it was not hypothesized, states in the Midwestern region were also more likely to adopt registration/notification policies, or to adopt them sooner, than states in the South.
Moreover, it is clear that Southern states were less likely than states in any other region to adopt registration/notification policies, or they did so later. This finding may indicate a general preference in the South for personal autonomy and freedom from government involvement. In fact, the state motto of Alabama is “Audemus Jura Nostra
Defendere,” which translates to “We Dare Maintain Our Rights” (www.alabama.gov).
This is not meant to imply that Southern states are less concerned than other states about sexual offending. In fact, until 2008, when the Supreme Court ruled it unconstitutional,
Louisiana allowed the death penalty for rape of a child. However, it is possible that
Southern states would take more traditional approaches to controlling the sex offenders
(e.g., very long periods of incarceration without the possibility of parole) rather than the more recent policies studied here.
No support was found for Hypothesis 4, that states in which a higher percentage of the population identified themselves as Republican or states that have a Republican governor would be at an increased risk of adopting both policies, compared to states in which a lower percentage of the population identified themselves as Republican or states that have a Democratic governor. Compared to states with Democratic governors, states with Republican governors were no more likely to adopt either registration/notification or civil commitment policies. Political variables are traditional predictors of policy
161 adoption. Particularly when considering criminal justice system policies, Republican leaders tend to be more conservative than do Democratic leaders. The fact that political party does not significantly impact the risk or timing of adopting the sexual offense legislation studied here suggests that these laws are not the same as other types of legislation.
Partial support was also found for Hypothesis 5, that states with the presence of indicators of political or racial conflict (e.g., percentage black, percentage Hispanic, and percentage unemployed) would be more likely to adopt both registration/notification policies and civil commitment policies. None of the three variables used to measure conflict in this study (percentage of the population that was black, percentage of the population that was Hispanic, and the unemployment rate) were related to the risk or timing of adopting a registration/notification policy. However, civil commitment statutes were more likely to be adopted, or to be adopted sooner, in states with lower levels of unemployment compared to states with higher levels of unemployment. This finding provides partial support for Hypothesis 5, and indicates that there is likely an economic element to legislator‟s decision making. A poor economy may cause states to reduce criminal justice funding, which would make this policy a burden to implement because of its high cost.
No support was found for Hypothesis 6, that states with higher levels of social disorganization would be more likely to adopt both policies than would states with lower levels of social disorganization. Although social disorganization variables are traditionally related to crime, for two reasons it is understandable that these variables were not related to sexual offending. First, sexual offending takes place across all
162 socioeconomic statuses (e.g., college aged women are a common target of sexual assault, though many of these women come from middle-class or better homes). Second, most sexual assaults are committed by individuals who know each other. Thus, a poverty- ridden neighborhood with low levels of community supervision is not a necessary environment in which to commit this crime.
No support was found for Hypothesis 7, that states with structural-functional reasons to adopt both policies would be more likely to adopt them than would states without such structural-functional reasons. This finding was surprising, because it indicated that the states that are typically most punitive (i.e., states that allow the death penalty and states with higher incarceration rates) were at no greater risk of adopting these policies, or of adopting them sooner, than states that are not typically punitive. This finding is consistent with the idea that sex offender legislation is passed for multiple reasons (e.g., to quell moral outrage), not simply as a way to control crime.
No support was found for Hypothesis 8, that as the percentage of neighboring states that adopted the policy increased, the likelihood that the states being analyzed would adopt the policy would also increase. Because there are many states with only one or two neighbors, there is limited variation in the values of this variable. This lack of variation may explain why it did not have a significant impact on either outcome, once internal determinant variables were taken into account.
The two additional operationalizations of diffusion, regional and national, did significantly and negatively impact the adoption of both registration/notification and civil commitment. As more states in the region adopted registration/notification policies, the likelihood that the remaining states would adopt a registration/notification policy
163 increased, providing support for Hypothesis 9. Similarly, as more states in the nation adopted civil commitment statutes, the likelihood that the remaining states would adopt civil commitment statutes decreased, providing support for Hypothesis 10.
The importance of time may suggest that states copied the actions of other states blindly. However, because the effect was not linear, it may also suggest that states hesitated to adopt these changes in sex offender legislation, to see the impact of these policies in states that had already adopted them. Even if the states did ultimately adopt the policy (as was the case for all states regarding registration/notification policies), they may have waited until the policies were adopted elsewhere simply to see the way that the laws were implemented and received in other states. Both registration/notification policies and civil commitment statutes are very expensive to implement and maintain. It is estimated that a state spends approximately $500,000 per year to provide a publicly available sex offender registry (Sample & Evans, 2009) and approximately $100,000 per year to civilly commit a sex offender (Scott & Busto, 2009). These high costs may serve to deter states from implementing the policies too quickly.
No support was found for Hypothesis 11 or 12, that states that were more physically connected to their neighbors or to the nation, respectively, would be more likely to emulate the legislative actions of their neighbors. In fact, none of the variables included to measure a state‟s level of connectedness were significantly associated with their likelihood of adopting changes in sex offender legislation. This finding may suggest that internal determinants, the factors present within the state, are more important the legislative trends present in other states.
164 Hypothesis 13, that states which Walker (1969) identified as hubs of innovation would be more likely to adopt registration/notification policies and civil commitment statutes, was not supported. No support was found for Hypothesis 14, that states that were identified as hubs of innovation by Canon and Baum (1981) would be at an increased risk of adopting these policies, or of doing so sooner than states that were not identified as hubs of innovation.
Partial support was found for both Hypothesis 15a (that the passage of the Jacob
Wetterling Act and Megan‟s Law would impact the adoption of registration/notification policies) and Hypothesis 15b (that the Supreme Court ruling in Kansas v. Hendricks and the passage of the Adam Walsh Act would impact the adoption of civil commitment statutes). Both the adoption of the Jacob Wetterling Act and the Supreme Court ruling
Kansas v. Hendricks were negatively related to the timing of adopting a civil commitment statute. These findings are contrary to expectation, and may indicate that
Federal opinions on sexual offense legislation does not have the intended impact. These findings suggest that states prefer some degree of autonomy in deciding their own policies to control the sex offender population in their state.
Hypothesis 16, that the amount of media coverage of sexual offending would be positively associated with the timing of policy adoption, was not supported. Although the role of media was expected to have a significant impact on the adoption of both policies, it is possible that the measures of time included in this study accounted for this effect. The amount of media coverage of sexual offenders increased consistently each year, which was consistent with the basic linear measure of time included in each event history model.
165 No support was found for Hypotheses 17a (that the presence of an “ATSA” chapter in a state would increase the risk of adopting a registration/notification policy) or
17b (that the presence of an “ATSA” chapter in a state would decrease the risk of adopting a registration/notification policy). Similarly, no support was found for
Hypothesis 18a (that the presence of “Stop it Now!” chapter in a state would increase the risk of adopting a civil commitment statute) or 18b (that the presence of “Stop it Now!” chapter in a state would decrease the risk of adopting a civil commitment statute). In fact, once both internal determinants and diffusion characteristics were taken into account, there was no effect of having either social movement organization present in the state. At the time included in these analyses, “Stop it Now!” was present in only three states, which suggests that it had a limited influence. Although “ATSA” was present in a larger sample of states, the membership of this organization (n = 2,700 members at present) was still limited.
Implications
These findings have both theoretical and practical implications. Studies of the adoption of multiple different types of policies indicate that both characteristics of the state and the legislative actions that other states took impact the risk and timing of policy adoption (see, for example, Berry & Berry, 1990). However, many of the internal determinants that were tested in this research were not significantly associated with policy adoption. The only internal determinant that impacted both registration/notification policy and civil commitment statutes was region of the country.
166 None of the characteristics of the state‟s political composition or structural-functional needs were significantly associated with the risk and timing of policy adoption.
This finding may indicate that sex offender laws are unique, and cannot be predicted using the same variables that are traditionally used to predict the adoption of other types of policies. Essentially all policies serve practical purposes, but some policies also serve moral interests. Sexual offending legislation is particularly sensitive to the effects of single, galvanizing events, rather than general trends in offending behavior.
These responses to a single case (e.g., Megan‟s Law), clearly have symbolic value, even if the usefulness of the law is not clear. This type of legislative response is a practice that has been called “irrational” (Mears, 2010; page 11), and sets sexual offending policies apart from other types of policies.
Although internal determinants had less of an impact on sex offender legislation than expected, diffusion mechanisms did affect the risk and timing of policy adoption.
As more states within the region (in the case of registration/notification) or within the nation (in the case of civil commitment) adopted the policy, the risk of the remaining states adopting that policy increased. Typically, as shown here, policy adoption follows an s-shaped pattern in which a few states initially adopt the policy (the bottom tail of the
“s”) then many states imitate the first group by adopting the policy, as well. After most states have adopted the policy, the few states remaining adopt the policy, as well, and form the top tail of the “s” (Cocca, 2004). However, the perceived success or failure of sexual offense legislation may impact the rate of policy adoption. Some researchers have argued that states are more likely to adopt policies that other states have adopted only when those policies were successful (Volden, 2006). In other words, policies that appear
167 to have a positive impact on a state (such as the positive economic impact of adopting a state lottery; Berry & Berry, 1990) are more likely to be adopted in other states compared to policies that are poorly implemented or do not result in any positive outcomes. In future research, the success or failure of a statute should be considered as a likely moderator of the diffusion process.
Currently, although all 50 states have registration/notification policies and 20 states have civil commitment statutes, the effectiveness of these laws has been questioned. Empirical research indicates both that the online registries are rarely accessed by the citizens whom they are intended to protect (Anderson & Sample, 2008) and that inaccuracies on these registries limit their usefulness (Levenson & Cotter, 2005).
A recent study (Bandy, 2011), comparing the individual behaviors of residents who were notified by police that a sexual offender was moving in or lived within a three-block radius to the individual behaviors of residents who lived outside of the three-block radius, produced mixed results. The types of behaviors that the researchers focused on were those that should increase an individual‟s level of safety, or perceived safety.
Respondents were asked, for example, whether they had purchased a weapon in order to protect themselves. Other types of “protective behaviors” included limiting the amount of time that the respondent spent outside alone or walking at night. There was no difference in the amount of “protective behaviors” (in total, 17 protective behaviors were assessed) that adults took as a result of this notification, compared to adults who were not notified of an offender moving into the neighborhood. However, there was a slight but significant increase in the amount of protective behavior that parents notified of a sex offender in the neighborhood took on behalf of their children, compared to parents who
168 did not receive such notification. This finding suggests that there may be some effect of having a registration/notification policy on the amount of protective behavior employed by the public.
Ultimately, the goal of sex offender legislation is to reduce the occurrence of sexual offending. Study 2 of this dissertation focused on the impact of these policies on rates of sexual offending within the states. Two types of comparisons were conducted.
First, Study 2 included a comparison of the rates of multiple types of sexual offending in states that adopted only registration/notification policies compared to states that adopted both registration/notification policies and civil commitment statutes. Second, Study 2 included an analysis of the rates of sexual offending within states before and after the implementation of these two policies.
169
Chapter 4
Study 2: Examining Whether Laws Against Sexual Offending Deter Sexual Offending
Having established the factors that affect the timing of policy adoption across states, I next turned to the effect of those policies. Specifically, I used two data sources, the Uniform Crime Reports and the National Incident-Based Reporting System, to examine whether laws aimed specifically at sexual offenders affected the commission of sexual crimes. Using these two databases, I looked at six sexual crimes: rape, forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest.
The primary purpose of these laws is to reduce the rate of sexual offending, which could occur through different mechanisms because there are different ways that the criminal justice system is believed to alter criminal behavior. The four traditionally accepted philosophies of sentencing (Cotton, 2000) are retribution, rehabilitation, incapacitation, and deterrence. Retribution refers to punishing offenders for the crime that they have committed. Severe sentences and sentences that are very punitive (e.g., the death penalty) are the most extreme examples of retributive sentencing.8 A rehabilitation approach assumes that the offender is suffering from criminal tendencies, but can be trained to stop or reduce his or her criminal behavior. Rehabilitation was a central goal of sentencing in the first half of the 1900‟s, when the medical model of criminal
8By requiring long sentences for sex offenders, it is clear that at least one purpose of sex offender legislation is to make the offenders pay for their crimes.
170 offending was popular.9 Incapacitation refers to keeping offenders in prison or jail so that they will not commit crimes against citizens in the community. If offenders are off the streets (i.e., incapacitated), they are unable to commit crime.10
Deterrence has led to the most research, and comes in two forms: general deterrence and specific deterrence. General deterrence is the effect that a sanction has on all potential offenders. Awareness of the sanctions being applied to actual offenders should make the potential offenders less likely to commit a crime. Specific deterrence, on the other hand, is aimed at the specific individual being punished. By facing an unpleasant sanction, the theory suggests that the particular offender will be unlikely to commit another crime in the future.
Deterrence
Deterrence theory is based on the ideas set forth by Cesare Beccaria in his 18th century work, On Crimes and Punishments. According to his perspective, crime can be reduced by making the anticipated pain worse than the potential pleasure that could be gained from engaging in crime. In order for deterrence theory to be truly effective,
Beccaria stated that three conditions must be met. First, the offenders must believe with
9 Sexual psychopath laws, early forms of civil commitment in the United States, were intended to help rehabilitate offenders and provide them with treatment that would ultimately prevent them from future offending. The rehabilitation approach is generally inconsistent with sex offender legislation as it exists today. 10 Civil commitment statutes are consistent with an incapacitation approach. The major goal of sex offender legislation, according to the legislators who have passed it, is to ensure the safety of American citizens, especially children. The primary way that lawmakers have approached this goal is by introducing more severe sanctions for sex offenders, including longer incapacitation.
171 certainty that the law will be applied. If only a small percentage of offenders actually face punishment for their crimes, potential offenders are unlikely to be deterred from committing crime. Second, the punishment must be swiftly implemented. If an individual commits a crime, but does not face punishment for months or years, the imposed punishment will have only a weak deterrent effect. Finally, the punishment must be appropriately severe. If the punishment is too trivial, potential offenders will be more likely to decide that the potential reward for engaging in the crime is worth the possible sanction. None of these three requirements is weighted more highly than the others: all three requirements serve an important role and must be present for deterrence to take place.
There have also been studies to test measures of certainty, severity, and swiftness of punishment, the three necessary elements of deterrence, according to Beccaria.
Consistent with classical deterrence theory, these three elements have been found to affect human behavior such that when an individual believes that a punishment is certain to be applied, and that the punishment will be applied swiftly and severely, individuals are likely to avoid the behavior (Singer, 1970).11 However, deterrence may not operate the same way for all people.
Deterrence theory is popular among “law and order” politicians who promote legislation to “get tough” on crime. Three-strikes laws, mandatory sentences, and other
11 Although studied in different ways and under different names, psychologists have studied the concepts of deterrence theory. Traditionally, their studies focused on the effects of punishment and reward on the behavior of animals. Early studies involved training animals to complete a task (e.g., pressing a lever) by rewarding them, and then training them not to complete the task by punishing them (Singer, 1970). The consensus among researchers was that animals could be conditioned to engage in or avoid certain behaviors through a series of rewards and punishment (Skinner, 1951). Empirical support suggests that these principles can be applied to humans, as well (Rogers & Skinner, 1956; Singer, 1970).
172 forms of serious sanctions are intended to deter offenders from future crime. However, the bulk of the empirical literature suggests that these laws do not have deterrent effects
(Tonry, 1996). Deterrence theory has not been supported in a line of research on the most severe penalty available in the criminal justice system, the death penalty. In general, researchers argue that use of the death penalty neither lowers nor raises the homicide rate (Cochran, Chamlin, & Seth 1994). However, some studies suggest that in fact the death penalty increases homicide rates, through modeling of state-sanctioned violence, a phenomenon known as the brutalization hypothesis (Cochran & Chamlin,
2000).
A major criticism of deterrence theory is that criminals tend to act quickly, and without premeditation. Thus, thoughts about the potential punishment that they may face if caught are unlikely to cross the offenders‟ minds. A longitudinal study of more than
7,000 respondents addressed this issue and found that deterrence was most effective among the individuals who were most prone to criminal behavior (Wright, Caspi, Moffitt,
& Paternoster, 2004), probably because non-criminal offenders do not need to be deterred from offending. However, the finding that deterrence does impact the criminal behavior of individuals with higher criminal propensities provides evidence to refute the argument that deterrence will not work because criminals act impulsively (Wright et al., 2004.
Deterrence and Sex Offending Legislation
Sex offender legislation is generally consistent with the specific deterrence approach because individual offenders face unpleasant criminal sentences. In addition,
173 these individuals are subject to registration/notification and civil commitment as a preventive measure, to protect society from the perceived certainty of sexual recidivism.
However, the threat of these severe sanctions, as well as the other potential sanctions faced by sex offenders, may also work to deter potential sex offenders, a form of general deterrence. That is, individuals who are tempted to commit a sexual offense may refrain from doing so in order to avoid the criminal sentence that would result if they were caught.
The Present Research
This study served as a simple test of deterrence theory. Public notification laws meet all three requirements of deterrence theory. Registration/notification policies are imposed on all sex offenders (certainty),12 immediately upon the determination that the individuals must register as sex offenders (swiftly),13 and are perceived as punitive by those who are subject to them (severity; Levenson & Cotter, 2005). Thus, consistent with deterrence theory, a decrease in the rate of sexual offending in the state in the year(s) following the implementation of registration/notification policies was expected.
The effect of adopting a civil commitment statute on rates of sexual offending was also analyzed here. However, civil commitment statutes do not fulfill the three requirements of deterrence theory, as do registration/notification policies. First, only a
12 Of course, we cannot be sure that no sex offender slips through the cracks and fails to be registered. However, as written, the laws require all sex offenders to be registered.
13 Admittedly, the criminal justice system takes time between commission of the sexual offense and conviction for the act, which lessens the “swiftness” of the applied punishment.
174 small percentage of sexual offenders are ultimately sentenced to civil commitment, which is inconsistent with the theory‟s standard of certainty. Second, individuals are not sentenced to involuntary civil commitments until they have completed their criminal sentence, which does not meet the theory‟s standard of punishment being applied swiftly.
The third standard, and the only standard that is adequately met, is severity (being involuntarily civilly committed results in a substantial loss of liberty). Thus, it was not expected that civil commitment would have a deterrent effect on sexual offending.
Despite the fact that civil commitment statutes meet only one criterion of deterrence, it is important to note that there are two ways that these statutes may still have a negative effect on rates of sexual offending. First, offenders who are currently committed would be unable to offend due to incapacitation. Second, it is possible that released offenders who were at one time civilly committed would be less likely to recidivate due to specific deterrence. That is, the severe sanction of being civilly committed could deter released offenders from future offending, in order to avoid the risk of experiencing involuntary civil commitment again. In general, however, because the registration/notification laws more clearly meet the standards of deterrence, I expected that registration and notification laws would be more effective than involuntary civil commitment laws at reducing sex related offenses.
175
Hypotheses
There were two hypotheses regarding deterrence theory in this study.
1. I expected that after the adoption of registration/notification policies, rates of
sexual offending in the state would decrease. This expectation was based on
the fact that registration/notification policies were consistent with the three
requirements of deterrence theory: certainty, swiftness, and severity.
2. I expected that after the adoption of civil commitment statutes, rates of sexual
offending in the states would decrease, but not as substantially as they did
after the adoption of registration/notification policies. This expectation was
based on the fact that civil commitment statutes do not sufficiently meet the
three requirements of deterrence theory. However, because civil commitment
is a severe punishment, it was expected that some deterrent effect may be
present.
With regard to the types of sexual offenses included as outcomes, I had three additional hypotheses.
3. I expected that the adoption of registration/notification and civil commitment
policies would be associated with a decrease in the rate of rape in the state (as
measured by both the Uniform Crime Reports and the National Incident-
Based Reporting System). This expectation was based on the fact that rape is
the most commonly reported sexual offense being studied here. Thus, these
176 policies would be applied with the greatest frequency to rapists, providing the
most opportunity for deterrence to take place.
4. I expected that there would be a deterrent effect, though smaller than the
effect observed for rape, for forcible sodomy, forcible fondling, and sexual
assault with an object. This expectation was based on the fact that these
crimes are generally less common than rape. Thus, the specific deterrent
effect that results from punishments for these crimes will be limited.
5. I expected that the adoption of registration/notification and civil commitment
statutes would be significantly and negatively associated with the rates of
statutory rape and incest. This expectation is based on the fact that these
crimes are most commonly committed against children or adolescents, the
population that is targeted for protection by most sex offender legislation.
Because these policies are intended to protect the youth of society, I expected
to see a larger deterrent effect of these policies on statutory rape and incest
than the other outcomes being considered.
Data
In order to examine the deterrent effects of these laws on the rates of sexual offending, two different data sources were used, each with its own set of strengths and limitations. The first dataset was the Uniform Crime Reports (UCR) which reports the rates of forcible rape in each of the 50 states, by year. The second dataset was the
National-Incident Based Reporting System (NIBRS) which includes a variety of sexual
177 offenses (i.e., not just forcible rape), but is only available for certain states. The differences, strengths, and weaknesses of these two datasets are discussed, in turn.
Uniform Crime Reports
The Uniform Crime Reports are compiled annually by the FBI. All states participate, and have for decades (the UCR has been in existence since the 1930‟s), so the available data spanned the entire length of analysis relevant to this research (1986-2007).
I looked at the rate of forcible rape starting in 1986, two years prior to the adoption of the first policy (Illinois‟s adoption of a civil commitment statute in 1988). This lead time of two-years was consistent with the lead time used in the event history analyses in Study 1.
I looked at the rate of forcible rape by state, by year, through the final year, 2007, in which New York and New Hampshire adopted civil commitment statutes. I used data through this final adoption year to ensure that the most accurate picture of trends in sex offending were observed.
Annual, state-level data spanning 20 years helped to identify the deterrent effect of registration/notification and civil commitment statutes. However, these statutes were implemented largely in response to crimes against children. Therefore, an analysis of only forcible rape was insufficient. Thus, an additional dataset, which included a greater variety of sexual offenses, was also used.
178 National Incident-Based Reporting System
Introduced in the early 1990‟s as an extension of the UCR, NIBRS contains information on victims, offenders, and criminal incidents for 46 different offenses
(Chilton & Jarvis, 1999), including sexual offenses. Thus, it is an appropriate complement to the UCR data, which includes only one sexual offense (forcible rape) in the annual crime statistics. NIBRS, on the other hand, includes statistics for four forcible sex crimes: rape, sodomy, fondling, and sexual assault with an object. The database also included three non-forcible sex crimes: statutory rape, prostitution, and assisting prostitution. Finally, the rate of incest was included in the NIBRS database. This variety in offenses makes NIBRS a very useful and complementary data source.
Prostitution and assisting prostitution were not analyzed in this research for three reasons. First, both crimes are vastly underestimated in official statistics. Second, the legal status of these crimes is inconsistent across states. Third, both crimes often involve consenting adults. Moreover, the legislative initiatives intended to reduce rates of sexual offending were not aimed at controlling prostitution. Rather, they were aimed at reducing rates of rape and sexual assault, particularly when those crimes were committed against children.
States are not required to report to NIBRS, a fact that poses a major limitation of the data. Although states are reporting at an increasing rate, there are few states that have achieved 100% reporting. Table 4.1 provides a listing of all states that currently report to
NIBRS. However, as the table shows, not all precincts within participating states may be reporting. In fact, as Table 4.2 shows, only seven states had 100% of precincts reporting
179 by the time this study was conducted. In an additional two states, although 100% of precincts did not report to NIBRS, 100% of crime in the state was reported.
Table 4.1. All National Incident-Based Reporting System Certified States. Date of Adoption % of % of Certification State State Civil Date State Public Registry Commitment Pop. Crime Alabama 1 1998 N/A Sept. 2006 2% 2% Arizona 1996 1996 July 2004 3% 1% Arkansas 1997 N/A April 2000 78% 66% Colorado 1998 N/A Nov. 1997 85% 81% Connecticut 1998 N/A July 1999 67% 40% Delaware 1994 N/A July 2001 100% 100% Georgia1 1996 N/A Sept. 2000 .01% .02% Idaho 1993 N/A July 1992 91% 100% Illinois1 1996 1998 Dec. 2006 4% 6% Iowa 1995 1998 Aug. 1992 92% 97% Kansas 1994 1994 Feb. 2001 90% 72% Kentucky1 1994 N/A Jan. 2005 6% 8% Louisiana 1992 N/A Jan. 2002 9% 8% Maine 1995 N/A July 2003 22% 24% Massachusetts 1999 1998 Aug. 1995 76% 69% Michigan 1995 N/A Feb. 1996 100% 100% Missouri 1995 1994 July 2005 2.06% .03% Montana 1995 N/A July 2000 98% 98% Nebraska 1997 2006 Feb. 1997 36% 21% New Hampshire 1996 2007 May 2003 87% 85% North Dakota 1995 1997 Feb. 1991 93% 94% Ohio 1997 N/A Jan. 1999 70% 70% Oregon 1993 N/A June 1997 29% 27% Rhode Island 1996 N/A May 2002 100% 100% South Carolina 1999 1998 Jan. 1991 100% 100% South Dakota 1995 N/A Feb. 2001 90% 98% Tennessee 1997 N/A July 1998 100% 100% Texas 1999 1999 July 1998 13% 15% Utah 1996 N/A April 1994 79% 80% Vermont 1996 N/A April 1994 98% 100% Virginia 1997 2003 Nov. 1994 100% 100% Washington 1990 1990 2007 1% .04% West Virginia 1993 N/A Sept. 1998 100% 100% Wisconsin 1997 1994 Feb. 1997 17% 30% 1 State is not NIBRS-certified but agency data are individually accepted by the FBI. Source: The Justice Research and Statistics Association‟s Incident-Based Reporting Resource Center (http://www.jrsa.org/ibrrc/background-status/nibrs_states.shtml)
180 Table 4.2. States with Greater than 80% of Crime Reported to NIBRS Date of Adoption % of % of Certification State State State Public Civil Date Registry Commitment Pop Crime States with More than 80% of Crime Reported Colorado 1998 N/A Nov. 1997 85% 81% New Hampshire 1996 2007 May 2003 87% 85% Utah 1996 N/A April 1994 79% 80% States with More than 90% of Crime Reported Iowa 1995 1998 Aug. 1992 92% 97% Montana 1995 N/A July 2000 98% 98% North Dakota 1995 1997 Feb. 1991 93% 94% South Dakota 1995 N/A Feb. 2001 90% 98% States with 100% of Crime Reported Delaware 1994 N/A July 2001 100% 100% Idaho 1993 N/A July 1992 91% 100% Michigan 1995 N/A Feb. 1996 100% 100% Rhode Island 1996 N/A May 2002 100% 100% South Carolina 1999 1998 Jan. 1991 100% 100% Tennessee 1997 N/A July 1998 100% 100% Vermont 1996 N/A April 1994 98% 100% Virginia 1997 2003 Nov. 1994 100% 100% West Virginia 1993 N/A Sept. 1998 100% 100% Source: The Justice Research and Statistics Association‟s Incident-Based Reporting Resource Center (http://www.jrsa.org/ibrrc/background-status/nibrs_states.shtml) States that are shaded in grey have adopted both registration/notification and civil commitment statutes.
The percentage of crime being reported is more relevant than the percentage of precincts that report. Table 4.2 listed those states that have achieved at least 80% of crime reported. As shown, three states had greater than 80% of crime reported, four states had greater than 90% of crime reported, and nine states had 100% of crime reported.
As discussed previously, all 50 states had some form of public registry in place by
2007, whereas only 20 states had adopted civil commitment statutes. In the first analysis
181 presented here, I compared those states that had adopted both policies to states that had adopted only registration/notification policies.
Of the 20 states that had adopted both public registration and civil commitment statutes, five (California, Minnesota, New Jersey, New York, and Pennsylvania) are currently developing or testing the NIBRS system. Only one state (Florida) is not currently participating in, developing, or testing the NIBRS system. Of the remaining 14 states available for analysis, only five (New Hampshire, Iowa, North Dakota, South
Carolina, and Virginia) had a sufficient percentage (greater than 80%) of crime reported to be considered for analysis. These five states were analyzed and compared to the ten14 states that had not adopted civil commitment statutes in terms of the six sex crimes
(forcible rape, forcible sodomy, forcible fondling, forcible sexual act involving an object, statutory rape, and incest) by year.
A second limitation of the NIBRS data is that it is so new. Although the fifteen states analyzed did have sufficient reporting, they did not have 100% reporting for nearly as long as the more established UCR. Thus, I was unable to analyze the entire period that was analyzed using the UCR data. Data were used as far back as it was available in each of the fifteen states, respectively.
Methods
The panel data used in this research covered 1986 to 2007 (UCR) or 1991 to 2007
(NIBRS), and allowed for two different types of analyses. First, descriptive analyses
14 Montana was excluded from analyses because no data were available prior to 2005.
182 were conducted to identify differences between states that adopted both policies compared to states that have adopted only civil commitment statutes. Second, fixed effects analyses were used to look at changes within states before and after adopting a policy.
A fixed effects regression model provided a test of the impact of adopting a registration/notification policy or a civil commitment statute on the rate of sexual offending within each state. The results of a fixed effects analysis allow inferences to be made about the changes in sexual offending before and after policy adoption, within each state. Fixed effects models are more conservative, stringent tests than are random effects models. It is conventionally accepted that the best predictor of criminal behavior is prior criminal behavior. If an effect on criminal behavior is found, controlling for criminal behavior at a prior time point, researchers can be somewhat confident in that effect. A fixed effects model takes a minimum of three time points into account. Thus, the presence of a significant relationship in a fixed effects model is a strong indicator that there is truly such a relationship present in the data.
A regression model for panel data commonly takes the following form:
Y = α + β1X1 + … βnXn + u + ε where Y represents the outcome variable and α represents the intercept. All explanatory predictors are represented by an X, and the corresponding β indicates the effect of a one- unit change in that predictor on the outcome variable. Because panel data involve repeated measures of the same units over time, there are two error terms. The first, represented by u, captures the error present at the second level (in this research, state-
183 level) and the second error term, represented by ε, captures error present within state observations.
The level-2 error term is critical to understanding the difference between fixed effects and random effects models. Random effects models require that there is a zero correlation between the predictor variables and the level-2 error term. If level-2 predictors are correlated with the level-2 error term, then a fixed effects model must be used. Using random effects models when fixed effects are the appropriate choice would produce both inaccurate standard errors and coefficient estimates.
Rather than including a shared constant, α, fixed effects models treat each level-1 unit (annual data observations) as nested within its level-2 group (state). In this way, each state is used as a control for itself, by using earlier years of data to control for later years of observations (Halaby, 2004). In order for a fixed effects analysis to run, the data must be in panel form, with the dependent variable measured at a minimum of two time points, and there must be variation in the independent variables over time.
There are three important considerations to keep in mind when using fixed effects. First, factors that do not vary within-units over time are dropped from the analyses. For example, Alabama was located in the Southern region of the United States at every time point, and thus region could not be used as a predictor variable because it did not change within states. Because the focus of this analysis was to determine which factors led to within-unit change in an outcome, there was no need to include variables that did not change over time. This is a powerful property of fixed effects analysis, because it accounts for omitted or unobserved variables, so long as those variables do not change over time.
184 Second, and related to the first consideration, is that fixed effects models provide no information on potential relationships between level-2 units (i.e., states). Because of this characteristic fixed effects models are not an appropriate method to address research questions focused on between group differences (e.g., regional differences). Finally, fixed effects models are sometimes characterized by large standard errors, making it harder to find statistically significant relationships. The problem of larger standard errors is most pronounced when there is little variation over time in the explanatory variables.
Higher levels of variation over time produce more reasonable standard errors (Allison,
2006; Halaby, 2004).
Random effects models have two properties that fixed effects models do not.
First, they allow for between-unit (cross-state) comparisons. Second, random effects models will measure the effects of variables that are not time-varying (such as region of the country). A strength of random effects models is that they provide more degrees of freedom, which results in higher statistical power and improves the chance of finding statistically significant relationships. However, recall that random effects regression models are based on the assumption that there is no systematic relationship between the level-2 explanatory variables and the level-2 error term (Hardin, 2011).
In this research, I tested the effect of adopting certain policies on the rates of sexual offending within states, a question that was most appropriately addressed using fixed effects models.
185 Results
The following results section includes two primary parts: between-state comparisons and within-state comparisons. The between-state comparisons focus on the changes in crime rates over time in states that adopted only registration/notification policies compared to states that adopted both registration/notification policies and civil commitment statutes. The rape rate as measured by the UCR is presented first, followed by the rape rate as reported by NIBRS. The rates of five additional sexual offenses are then reported, each measured by NIBRS: forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest.
The within-state comparisons focus on the change in sexual offending rates over time within each state. The same sexual offense outcomes are examined, first including a predictor of having registration/notification policy, then including a predictor of having a civil commitment statute.
Between States
Although all 50 states have adopted registration/notification policies, only 20 have adopted civil commitment statutes. The observations presented in Figures 4.1 through 4.7 are split into two groups: states that adopted only registration/notification policies prior to 2007 and states that adopted both registration/notification and civil commitment policies to 2007. A comparison of these two groups over time may suggest that they differ on their rates of sexual offenses.
186 The rape rate as reported in the UCR is shown in Figure 4.1. At every time point, the rape rate was lower in states that had adopted both registration/notification and civil commitment policies than in states that had adopted only registration/notification policies. Both sets of states experienced very similar trends over time (e.g., a rise in reported rapes in the early 1990‟s followed by a gradual decline). The fact that states with only registration/notification policies have slightly but consistently higher rates of rape from 1986 through 2007 does not necessarily mean that these policies are having an impact on the rape rate, as a third variable might explain both effects. Moreover, the fact the trend lines follow the same pattern suggests that adopting a civil commitment statute may have little or no impact on the rape rate. If this policy had a large deterrent effect on rape, the trend line for states with both policies should diverge, showing a decrease, compared to the trend line representing states with only registration/notification policies.
Figure 4.1. UCR Rape Rate for States with Both Policies Compared to States with only Registration Policies
50
45
40
35
Rape 30 Rate Registration/Notification (UCR) 25 Only
20 Registration/Notification and Civil Commitment 15
10
5 0
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Figure 4.2 shows the rates of rape from 1991 to 2007 as reported in the NIBRS.
These rates are markedly higher than the rape rates reported in the UCR, likely due to two factors. First, the UCR has a hierarchical rule in crime reporting. For each criminal incident, only the most serious crime is included in the UCR statistics. Thus, for example, in any case in which someone was raped and murdered only the murder would appear in the UCR statistics because it was the most serious offense that took place during the criminal incident. In contrast, the NIBRS includes multiple crimes, even if they occur as part of one criminal incident. Thus, because more crimes are counted,
NIBRS has higher rates of most crimes. Second, there is a slight bias in the states that participate in NIBRS, such that the states available for this analysis had higher average rape rates (M = 36.69) than the remaining states (M = 36.07). Because states participate on a voluntary basis, it is possible that those states with very little crime do not consider this participation a priority and therefore do not report to NIBRS. The data shown here was based on observations from 15 states and the rates expressed in this study may be higher than the national average.
The pattern shown in Figure 4.2 began with a substantial difference between the two groups of states (those with registration/notification and civil commitment versus those with only registration/notification), but that gap lessened considerably by the mid-
90‟s and the two lines converged by the turn of the 21st century. This convergence suggests that there is no difference between the rape rates in the two sets of states.
However, it may also indicate that civil commitment statutes were effective. Perhaps, if civil commitment statutes had not been implemented, the upward trend in rape in the
189 states that had only registration/notification policies would have also been seen in states that adopted both policies.
Figure 4.2. NIBRS Rape Rate for States with Both Policies Compared to States with only Registration Policies
400
350
300
250 Rape Rate 200 Registration/Notification (NIBRS) Only
150 Registration/Notification and Civil Commitment
100
50
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
The rate of sodomy from 1991 to 2007 as reported by the NIBRS is shown in
Figure 4.3. The rate of fondling from 1991 to 2007 as reported by the NIBRS is shown in
Figure 4.4. The rate of sexual assault with an object from 1991 to 2007 as reported by the NIBRS is shown in Figure 4.5. All three of these figures follow a pattern very similar to that presented in Figure 4.2. Although there is some divergence in the rates at the first points of measurement, the two groups converge by the middle of the time period and stay similar from that point forward.
At first glance, this pattern suggests that there is no difference in crime rates between the states that adopted both policies and the states that adopted only registration/notification policies. However, these findings may suggest that that civil commitment statutes had a deterrent effect on rates of sodomy, fondling, and sexual assault with an object. If the 20 states had not adopted civil commitment statutes, it may be that the line representing the 30 states with both policies would have followed the same upwards trend that the line representing states with only registration/notification policies did in the early 2000‟s.
Figure 4.3. NIBRS Forcible Sodomy Rate for States with Both Policies Compared to States with only Registration Policies
80
70
60
Sodomy Rate 50 (NIBRS) Registration/Notification 40 Only
Registration/Notification 30 and Civil Commitment
20
10
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
193
Figure 4.4. NIBRS Forcible Fondling Rate for States with Both Policies Compared to States with only Registration Policies
450
400
350
300 Forcible Fondling 250 Rate Registration/Notification (NIBRS) Only 200 Registration/Notification and Civil Commitment 150
100
50
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
194
Figure 4.5. NIBRS Sexual Assault with an Object Rate for States with Both Policies Compared to States with only Registration Policies
50
45
40
35 Sexual Assault 30 with an Object Rate 25 Registration/Notification (NIBRS) Only
20 Registration/Notification and Civil Commitment 15
10
5
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Figure 4.6 shows the rate of statutory rape from 1991 to 2007, as reported by
NIBRS. Although the rate in states with only registration/notification policies and the rate in states with both policies are consistent through 2002, the two lines then diverge.
The rates of statutory rape increased in both sets of states, but the increase was more pronounced among the states with both civil commitment statutes and registration/notification policies. This trend may suggest that civil commitment policies exacerbated statutory rape problems in those states. It may also suggest that states that adopted civil commitment statutes are those states with greater need for such laws. If there were greater rates of sexual crimes involving teenage victims in those states (such as statutory rape), then perhaps those states were in need of a policy more severe than just registration/notification.
Figure 4.6. NIBRS Statutory Rape Rate for States with Both Policies Compared to States with only Registration Policies
140 00
120
100 Statutory Rape Rate (NIBRS) 80 Registration/Notification Only
60 Registration/Notification and Civil Commitment 40
20
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Regarding incest, at first glance, there appears to be substantial difference between the two lines presented in Figure 4.7. However, the rates of incest, as reported by the NIBRS, which these lines represent, actually vary very little (note the value labels on the Y-axis). Incest is an uncommon crime, particularly in official reports. However, in 2002 and later, the rate of incest was consistently higher in states with both policies than in states with only registration/notification policies. Similar to the pattern shown in
Figure 4.6, this may suggest one of two things. First, it is possible, but unlikely, that the introduction of civil commitment statutes led to an increase in the rate of incest. Second, it is possible that states with higher rates of crimes against children (such as incest) required a more severe response, and they turned to civil commitment statutes as one way to control the sex offender population.
Figure 4.7. NIBRS Incest Rate for States with Both Policies Compared to States with only Registration Policies
30
25
20 Incest Rate (NIBRS) Registration/Notification 15 Only
Registration/Notification and Civil Commitment 10
5
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Although these descriptive trend lines provide some insight into the potential impact of policy adoption on the rates of sexual offending across states, they are by no means definitive. A more useful way of addressing this question would involve a comparison of rates of sexual offending within each state before and after each policy was adopted. That was the goal of the analyses presented in the next section.
Previous research has attempted to predict differences in rape rates across states using various theoretical perspectives. For example, Baron and Straus (1987) isolated the effects of certain theoretical variables on the rape rate within each of the 50 states over a three year period. Social disorganization had significant effects on the rape rate, which indicated that social disorganization is an important theory to include in any aggregate- level study of rape. Although the goal of the present study was not explicitly to identify the factors that impact the rape rate across states, two measures of social disorganization were still included as covariates: the percentage of the population living in poverty and the divorce rate.15
Two covariates were also included to account for state-level trends in the criminal justice system. The murder rate served as a gauge of the rate of violent crime and the incarceration rate served as a measure of the state‟s punitiveness. States with higher crime rates tend to have higher incarceration rates, which may indicate two possible underlying trends. First, it is possible that states with higher crime rates have higher incarceration rates because they have more crime and need somewhere to house the
15 All covariates included in these models were operationalized in the same way as those used in Study 1. See page 105 for a discussion of how the percentage of the population living in poverty and the divorce rate were measured. See page 106 for a discussion of how the murder rate and the incarceration rate were measured. See pages 101-102 for a discussion of how the party of the governor was measured.
200 offenders. Second, it is possible that this relationship is spurious and that both higher crime rate and higher incarceration rate are the result of greater levels of law enforcement. All outcomes studied here are based on official measures of crime, so only crime that is reported is included. Thus, an upwards bias as a result of law enforcement is possible in these two variables.
The final covariate included here is a dummy variable indicating the political party of the governor. More punitive criminal justice policies are generally embraced by conservative politicians so it is possible, if deterrence theory holds, that having a
Republican governor would result in lower rates of sexual offending.
The primary explanatory variable in these analyses is a time-varying dummy variable, coded high once a state has adopted the policy being tested (either registration/notification or civil commitment). The goal of this study was to isolate the effects, if any, of the adoption of specific sex offender legislation on the rates of sexual offending across states.
Within States
Fixed effects models provided estimates of the effect of adopting a registration/notification policy or a civil commitment statute on rates of sexual offending at the state-level. These models, which gauged changes in rates of sexual offending within each state, are presented below.
201 Fixed Effects Model Predicting the Effect of Registration/Notification on Rape Rate
Rape Rate as Measured by the UCR
The model predicting the rape rate according to the UCR is shown in Table 4.3.
Although the effect of registration/notification policies on the rape rate was negative, consistent with Hypothesis 1, it was not significant. Three covariates were positive and significant predictors of the rape rate: murder rate, divorce rate, and incarceration rate.
These findings indicate that as a state‟s rate of murder, divorce, or incarceration increased, so did the rape rate. The adoption of a registration/notification policy did not have a significant impact on the outcome.
Table 4.3. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 14.83** 4.87 Registration/Notification Policy -.53 1.09 Murder Rate 1.83*** .28 Percentage Living in Poverty -.02 .05 Divorce Rate 1.52* .60 Republican Governor -.79 .95 Incarceration Rate 16.82** 6.38 * p-value < .05 ** p-value < .01 *** p-value < .001
The problem with this analysis is that the predictors were contemporaneous with the dependent measure. A better way to gauge the effect of these predictors on the outcome would be to lag all predictor variables, so that they precede the outcome variable
(rape rate as measured by the UCR) by one year. This way, all coefficients represent the effect of the predictors on the rate of rape one year later. The lagged model, shown in
202 Table 4.4., is consistent with the findings of Table 4.3. The adoption of a registration/notification policy did not have a significant impact on the rape rate, measured by the UCR when the independent variables were lagged by one year.
Table 4.4. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 14.40** 4.79 Registration/Notification Policy .13 1.07 Murder Rate 1.84*** .30 Percentage Living in Poverty -.07 .05 Divorce Rate 1.56* .59 Republican Governor -.43 .93 Incarceration Rate 17.52* 4.79 * p-value < .05 ** p-value < .01 *** p-value < .001
Although there was no significant one-year lag effect, it is possible that the effects of the predictor variables take longer than one year to appear. Table 4.5 shows the results of a fixed effects model where the outcome variable (rape rate as measured by the UCR) was measured two years after all of the predictor variables. The findings of this model indicate that the same three predictors, murder rate, divorce rate, and incarceration rate, were significantly and positively associated with the rape rate. However, the magnitude of these effects was slightly larger than in the preceding models. The adoption of a registration/notification policy did not have a significant impact on the rape rate, measured by the UCR, when lagged by two years.
203
Table 4.5. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 11.76** 4.55 Registration/Notification Policy .81 1.08 Murder Rate 1.93*** .27 Percentage Living in Poverty -.11 .05 Divorce Rate 1.87* .56 Republican Governor -.60 .76 Incarceration Rate 20.75* 6.40 * p-value < .05 ** p-value < .01 *** p-value < .001
One additional way to assess the potentially deterrent effect of the adoption of a registration/notification policy on the rape rate is to examine its linear effect, rather than simply whether the policy was adopted. That is, the effect of the predictor variable may increase cumulatively over time. In Table 4.6, the policy adoption variable was coded in a linear way, starting with 1 the first year in which the policy was adopted, and running through the fifth year following the adoption of the policy (assigned a numeric value of
5). Although the coefficient shown in this model is negative (suggesting that adopting the policy was associated with a decrease in the rape rate), this relationship was not significant.
204
Table 4.6. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 14.55** 4.73 Linear After Adoption -.13 .18 Murder Rate 1.86*** .27 Percentage Living in Poverty -.03 .05 Divorce Rate 1.59** .58 Republican Governor -.84 .95 Incarceration Rate 15.69* 4.73 * p-value < .05 ** p-value < .01 *** p-value < .001
Rape Rate as Measured by NIBRS
The significant effect of incarceration rate on rape rate which was found in the
UCR data was also present in the sample of 15 states in which NIBRS data were reported
(see Table 4.7). However, in this data, the relationship was negative, indicating that when the rate of incarceration within a state was higher the rates of rape were lower. As in Table 4.3, there was no significant effect of having a registration/notification law on the rape rate. This finding is inconsistent with Hypothesis 1, that registration/notification would have a significant negative impact on the rate of sexual offending in the state.
Table 4.7. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 35.78*** 6.28 Registration/Notification Policy -.40 1.77 Murder Rate 1.76* .65 Percentage Living in Poverty -.01 .02 Divorce Rate -1.34 1.08 Republican Governor .20 1.39 Incarceration Rate -.01* .01 † p-value <.10 * p-value < .05
205
Once the outcome variable, rape rate as measured by NIBRS, was lagged one year, the effect of incarceration rate became non-significant (see Table 4.8). The negative effect of murder rate also lost significance. However, an additional variable, percentage of the population living in poverty, became significant, suggesting that within states as the rate of poverty increased, the rate of rape decreased. There was no deterrent effect of having a registration/notification policy on the rape rate, when it was lagged by one year.
Table 4.8. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 37.87*** 4.90 Registration/Notification Policy -.61 2.12 Murder Rate 1.22 .81 Percentage Living in Poverty -.07** .02 Divorce Rate -.97 .88 Republican Governor -.46 1.46 Incarceration Rate -.01 .01 † p-value <.10 * p-value < .05
The findings presented in Table 4.9 (a lag of two years) are consistent with those presented in Table 4.8. There was no deterrent effect of having a registration/notification policy on the rape rate, measured two years later.
206
Table 4.9. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant `35.05*** 5.89 Registration/Notification Policy -.07 2.48 Murder Rate .64 .88 Percentage Living in Poverty -.08* .03 Divorce Rate .27 1.02 Republican Governor -1.36 1.77 Incarceration Rate -.01 .01 † p-value <.10 * p-value < .05
Once a linear measure of the policy predictor was taken into account, in which the policy adoption variable was coded in a linear way (1 in the first year in which the policy was adopted through 5 in the fifth year), the effects of murder rate and incarceration rate returned to significance (see Table 4.10). However, the adoption of a registration/notification policy did not significantly impact the rape rate as measured by the UCR, even measured over time.
Table 4.10. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 31.02*** 3.90 Linear After Adoption .26 .28 Murder Rate 1.81* .61 Percentage Living in Poverty -.02 .02 Divorce Rate -.53 .45 Republican Governor .09 1.32 Incarceration Rate -.02** .01 † p-value <.10 * p-value < .05
207 Fixed Effects Model Predicting the Effect of Registration/Notification on the Forcible Sodomy Rate
In the model shown in Table 4.11, predicting the rate of sodomy within states as reported in NIBRS, the incarceration rate was both positive and significant. Consistent with the preceding models, there was no significant effect of having a registration/notification policy on the outcome.
Table 4.11. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 185.42 108.87 Registration/Notification Policy 20.46 32.89 Murder Rate 14.16 19.89 Percentage Living in Poverty .88 .50 Divorce Rate -23.55 13.32 Republican Governor 10.64 20.68 Incarceration Rate .53*** .10 † p-value <.10 * p-value < .05
The findings of Table 4.12 (a lag of one year) were consistent with those presented in Table 4.11. That is, there was a significant, positive effect of the incarceration rate on the rate of forcible sodomy.
Table 4.12. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 228.06 107.69 Registration/Notification Policy -7.09 44.72 Murder Rate 8.36 14.36 Percentage Living in Poverty .57 .86 Divorce Rate -27.01 15.12 Republican Governor 26.60 31.86 Incarceration Rate .55*** .12 † p-value <.10 * p-value < .05
208
The sodomy rate as measured by the NIBRS, lagged by two years, was significantly impacted by both the divorce rate and the incarceration rate. As shown in
Table 4.13, when the divorce rate in a state increased, the sodomy rate decreased two years later. As shown in the preceding models, there was a significant and positive effect of incarceration on the sodomy rate. However, the policy predictor remained non- significant, even when lagged by one year.
Table 4.13. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 337.10*** 82.89 Registration/Notification Policy -47.81 54.93 Murder Rate -7.05 13.19 Percentage Living in Poverty .84 .64 Divorce Rate -34.93* 12.39 Republican Governor 54.76 33.428 Incarceration Rate .51** .14 † p-value <.10 * p-value < .05
As shown in Table 4.14, there was a significant effect of adopting a registration/notification policy on the sodomy rate over time, but not in the anticipated direction. The linear measure of policy adoption (1 in the first year in which the policy was adopted through 5 in the fifth year) indicated that the longer the policy remained in effect, the higher the forcible sodomy rate became. This is the opposite of a deterrent effect. Rather, it appears to suggest that the policy led to an increase in sexual offending of this type. However, there is a more likely explanation for this increase. An increase in the state‟s level of vigilance towards sexual crimes, manifested in the adoption of a registration/notification policy, may have also been associated with an increase in the
209 reporting and enforcement of these crimes. That is, acts of forcible sodomy that may have gone unreported or uninvestigated in a previous era, were brought to the attention of law enforcement officials after the implementation of a more serious, punitive approach to sexual offending was put into place. A study of changes in rape legislation showed that analogous types of policy changes can impact rates of reporting (Horney & Spohn,
1991).
Table 4.14. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Linear Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 7.80 122.30 Linear After Adoption 13.37*** 2.95 Murder Rate 11.71 18.10 Percentage Living in Poverty .51 .33 Divorce Rate 14.31 16.17 Republican Governor 2.94 17.25 Incarceration Rate .35** .11 † p-value <.10 * p-value < .05
Fixed Effects Model Predicting the Effect of Registration/Notification on the Forcible Fondling Rate
In a model predicting the rate of forcible fondling, as reported by NIBRS, the only significant predictor was incarceration rate. When states reported higher incarceration rates, they also reported higher rates of forcible fondling (see Table 4.15). The adoption of a registration/notification statute was not a significant predictor of the outcome.
210
Table 4.15. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 44.25 25.89 Registration/Notification Policy 2.15 12.48 Murder Rate -1.05 3.84 Percentage Living in Poverty .20 .10 Divorce Rate -2.70 3.34 Republican Governor 7.03 6.27 Incarceration Rate .11*** .03 ** p-value < .01
The results shown in Table 4.16 (a lag of one year) were consistent with those presented in 4.15. That is, the incarceration rate was a positive, significant predictor of the rate of forcible fondling one year later.
Table 4.16. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 53.76* 19.52 Registration/Notification Policy -.23 13.20 Murder Rate -1.90 3.51 Percentage Living in Poverty .05 .17 Divorce Rate -4.19 2.68 Republican Governor 13.13 7.70 Incarceration Rate .11** .03 ** p-value < .01
When all predictors were lagged by two years, the divorce rate and having a
Republican governor became significant predictors of the rate of forcible fondling. As with forcible sodomy, an increase in the divorce rate was associated with a decrease in the rate of forcible fondling two years later. According to the logic of social disorganization theory, for which divorce rate is an indicator, higher rates of social disorganization lead to decreases in informal social control. To compensate for this
211 breakdown, formal social controls are then adopted, leading to the increased enforcement of sexual assaults, such as forcible fondling. Ultimately, the effect of divorce rate may be an indicator of increased social disorganization, resulting in an increase in formal social control mechanism, which would result in the increase in the official rate of forcible fondling.
When states had Republican, rather than Democratic, governors the rate of forcible fondling was significantly higher. This effect may be the result of increased enforcement, as Republican governors tend to be more punitive in terms of criminal justice policy than are Democratic governors. The positive effect of incarceration rate remained significant, and there was no significant deterrent effect of adopting a registration/notification policy.
Table 4.17. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 56.67* 19.50 Registration/Notification Policy -4.05 12.83 Murder Rate -2.04 2.99 Percentage Living in Poverty .18 .16 Divorce Rate -5.14* 1.92 Republican Governor 24.03* 9.49 Incarceration Rate .09* .04 ** p-value < .01
When a linear measure of time from policy adoption was included, ranging from one in the year of adoption to five in the fifth year following adoption, there were positive effects of both the percentage of the population living in poverty and the incarceration rate (see Table 4.18). Thus, as the rate of poverty or incarceration within a state increased, so too did the state‟s rate of forcible fondling. More importantly, there
212 was a positive effect of adopting a registration/notification policy on the rate of forcible fondling. That is, in the years following the adoption of a registration/notification statute, the rate of forcible fondling increased. This increase may be the result of increased enforcement of forcible fondling, rather than an actual increase in the amount of forcible fondling taking place within the state.
Table 4.18. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 11.35 34.06 Linear After Adoption 2.30* 1.00 Murder Rate -1.30 4.00 Percentage Living in Poverty .14* .06 Divorce Rate 3.96 6.51 Republican Governor 5.76 5.14 Incarceration Rate .08** .03 ** p-value < .01
Fixed Effects Model Predicting the Effect of Registration/Notification on the Rate of Sexual Assault with an Object
In a model predicting the rate of sexual assault with an object, as reported by
NIBRS, the only significant predictor was incarceration rate. When states reported higher incarceration rates, they also reported higher rates of sexual assault with an object
(see Tables 4.19). The adoption of a registration/notification did not have a significant effect on the rate of sexual assault with an object.
213
Table 4.19. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 98.80 161.69 Registration/Notification Policy 58.66 47.78 Murder Rate 19.60 22.72 Percentage Living in Poverty .19 .52 Divorce Rate -6.84 13.50 Republican Governor 7.43 22.97 Incarceration Rate .71** .18 * p-value < .05
The findings shown in Table 4.20 (lagged by one year) were consistent with those of Table 4.19. That is, there was a positive and significant effect of the incarceration rate on the rate of sexual assault with an object.
Table 4.20. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 157.04 144.63 Registration/Notification Policy 27.74 43.25 Murder Rate 7.58 218.17 Percentage Living in Poverty .40 .81 Divorce Rate -8.58 16.33 Republican Governor 31.52 29.97 Incarceration Rate .68*** .17 * p-value < .05
The findings shown in Table 4.21 (a two-year lag) were consistent with those of both Table 4.19 and Table 4.20. In none of the three was there a significant effect of adopting a registration/notification statute.
214
Table 4.21. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 339.60* 121.51 Registration/Notification Policy 2.31 51.43 Murder Rate -18.32 17.39 Percentage Living in Poverty .73 .75 Divorce Rate -25.40 15.21 Republican Governor 53.47 41.04 Incarceration Rate .57** .18 * p-value < .05
In Table 4.22 a linear measure of time from adoption of a registration/notification to five years after adoption, was included. The effect of the incarceration rate remained significant and positive. As with forcible sodomy and forcible fondling, there was a positive effect of adopting a registration/notification policy on the rate of sexual assault with an object. This finding suggests that in the years following the adoption of a registration/notification statute, the rate of sexual assault with an object increased. This increase in the rate of the crime is likely the result of increased reporting.
Table 4.22. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant -61.96 220.05 Linear After Adoption 16.35* 6.64 Murder Rate 12.41 17.12 Percentage Living in Poverty -.23 .22 Divorce Rate 36.06 37.44 Republican Governor -3.34 19.81 Incarceration Rate .51*** .13 * p-value < .05
215 Fixed Effects Model Predicting the Effect of Registration/Notification on the Statutory Rape Rate
As shown in Table 4.23, no variables were significantly associated with the rate of statutory rape, suggesting that there was no significant deterrent effect of having a registration/notification policy on the outcome. A caveat, however, is needed here and with the other analyses in this section, in that a larger sample of states over a longer period of time would provide more statistical power to detect an effect if it were present.
Table 4.23. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant -1.50 6.13 Registration/Notification Policy -3.41 2.65 Murder Rate 2.08 1.19 Percentage Living in Poverty -.01 .05 Divorce Rate .73 .69 Republican Governor -2.23 1.51 Incarceration Rate .04 .02 ** p-value < .01 *** p-value < .001
Although the effect of having a registration/notification policy remained non- significant when the predictor variables were lagged one year, there was a significant positive effect of incarceration rate (see Table 4.24).
216
Table 4.24. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 10.04 6.05 Registration/Notification Policy -5.10 3.75 Murder Rate .04 .92 Percentage Living in Poverty -.01 .04 Divorce Rate .51 .89 Republican Governor .32 2.90 Incarceration Rate .02** .01 ** p-value < .01 *** p-value < .001
When the predictor variables were lagged by two years, there were significant, positive effects of the percentage of the population living in poverty, on the outcome variable, statutory rape rate.
Table 4.25. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 15.47 16.43 Registration/Notification Policy -5.17 3.87 Murder Rate -.63 .55 Percentage Living in Poverty .12* .05 Divorce Rate -.48 1.21 Republican Governor 2.27 4.09 Incarceration Rate .01 .01 ** p-value < .01 *** p-value < .001
There was no significant effect of adopting a registration/notification policy over time, as shown in Table 4.26. None of the substantive predictors included in this model were significantly associated with the rate of statutory rape.
217
Table 4.26. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant -5.77 6.63 Linear After Adoption -.09 .32 Murder Rate 2.51 1.37 Percentage Living in Poverty -.01 .05 Divorce Rate .80 1.01 Republican Governor -2.04 1.55 Incarceration Rate .04 .02 ** p-value < .01 *** p-value < .001
Fixed Effects Model Predicting the Effect of Registration/Notification on the Incest Rate
With regard to the incest rate as reported by NIBRS, the incarceration rate was positively and significantly predictive (see Table 4.27). However, there was no significant effect of having a registration/notification policy on the rate of incest.
Table 4.27. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant 22.79 10.63 Registration/Notification Policy 2.98 6.20 Murder Rate 1.05 2.28 Percentage Living in Poverty .03 .06 Divorce Rate -2.73 2.56 Republican Governor 2.68 3.02 Incarceration Rate .07** .02
The findings presented in Table 4.28 (one year lag) were consistent with those in
Table 4.27. That is, there was a positive and significant impact of incarceration rate on the rate of incest.
218
Table 4.28. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor and Covariates Lagged One Year) Variable Coefficient Robust Standard Error Constant 27.14 9.24 Registration/Notification Policy 1.02 6.39 Murder Rate -.27 2.20 Percentage Living in Poverty .09 .08 Divorce Rate -2.97 2.19 Republican Governor 5.49 3.35 Incarceration Rate .07** .02
No substantive predictors included in Table 4.29 (lagged two years) were significantly associated with the rate of incest, measured two years later.
Table 4.29. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Registration/Notification Policy Predictor and all Covariates Lagged Two Years) Variable Coefficient Robust Standard Error Constant 28.87* 11.22 Registration/Notification Policy -3.07 5.34 Murder Rate -1.23 1.93 Percentage Living in Poverty .04 .07 Divorce Rate -2.45 1.60 Republican Governor 6.66 4.33 Incarceration Rate .08 .03
There was also a significant effect of the linear measure of policy adoption on the rate of incest. Contrary to expectation, but consistent with the findings for several other outcomes presented here, the effect was positive. That is, the longer that the registration/notification policy was in place, the higher the rate of incest became. This effect is likely the result of increased reporting and enforcement of incest, rather than an increase in the true rate of incest. The effects of the substantive predictors presented in
Table 4.30 are consistent with the findings of Tables 4.27 and 4.28, in which there was a significant, positive effect of incarceration rate on incest rate.
219
Table 4.30. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Linear Registration/Notification Policy Predictor) Variable Coefficient Robust Standard Error Constant .71 15.04 Linear After Adoption 1.71* .79 Murder Rate .70 2.39 Percentage Living in Poverty -.02 .05 Divorce Rate 2.08 2.16 Republican Governor 1.68 3.28 Incarceration Rate .04** .01
Overall, the models presented here provide no evidence for Hypothesis 1, that having a registration/notification policy would deter sexual offending in the state. In fact, for four measures (forcible sodomy, forcible fondling, sexual assault with an object, incest) the linear effect over time showed an increase. This finding is also inconsistent with Hypothesis 3, that the deterrent impact of having a registration/notification policy would be larger on the rape rate than on the other types of sexual offending analyzed here. However, it is possible that civil commitment statutes do provide some deterrent effect. The models presented next test for such an effect.
Fixed Effects Model Predicting the Effect of Civil Commitment on the Rape Rate
Rape Rate as Measured by the UCR
In Table 4.31, the dummy variable representing the presence of a registration/notification policy was replaced with a dummy variable representing the presence of a civil commitment statute, and the outcome was the rape rate as measured by
220 the UCR. This predictor was not significant. However, the murder rate, divorce rate, and incarceration rate, were all positively and significantly associated with the rape rate, as measured by the UCR. That is, as the rates of murder, divorce, and incarceration within a state increased, so did the rate of rape.
Table 4.31. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 14.64*** 4.83 Civil Commitment Policy -1.20 1.62 Murder Rate 1.86*** .28 Percentage Living in Poverty -.02 .05 Divorce Rate 1.51* .61 Republican Governor -.70 .99 Incarceration Rate 16.58** 4.83 * p-value < .05 ** p-value < .01 *** p-value < .001
The findings presented in Table 4.32, in which all predictor variables were lagged by one year, were consistent with the findings presented in Table 4.31. That is, there were positive, significant effects of the murder rate, divorce rate, and incarceration rate on the rape rate.
221
Table 4.32. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 14.56** 4.74 Civil Commitment Policy (one year later) -.65 1.55 Murder Rate 1.82*** .30 Percentage Living in Poverty -.06 .05 Divorce Rate 1.51* .59 Republican Governor -.39 .96 Incarceration Rate 18.63** 6.22 * p-value < .05 ** p-value < .01 *** p-value < .001
All findings that were significant in Table 4.31 and 4.32 are also significant in
Table 4.33, in which the outcome was measured two years later. However, the percentage of the population living in poverty was also significantly, negatively associated with the rate of rape as measured by the UCR. This means that as the state‟s level of poverty increased, the rate of rape two years later decreased. One possible explanation for this surprising effect relates to how law enforcement might react to increased poverty. Recall that, according to social disorganization theory, an increase in the percentage of the population living in poverty should lead to a decrease in informal social control. Thus, new or additional formal social control mechanisms should be put in place, which should lead to the increased enforcement of rape. The policy predictor remained non-significant.
222
Table 4.33. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 12.32* 4.59 Civil Commitment Policy (two years later) -.41 1.46 Murder Rate 1.86*** .27 Percentage Living in Poverty -.11* .05 Divorce Rate 1.75** .57 Republican Governor -.60 .79 Incarceration Rate 23.60*** 6.14 * p-value < .05 ** p-value < .01 *** p-value < .001
The findings presented in Table 4.34 (the linear effect of the policy) were consistent with the findings of the preceding models. The murder rate, divorce rate, and incarceration rate were all positively and significantly associated with the rape rate.
However, the linear measure of adopting a civil commitment policy, which ranged from a value of one in the year of adoption to five in the fifth year following adoption, was not significant, indicating no deterrent effect of the policy, even over time.
Table 4.34. Fixed Effects Model Predicting the Rate of Rape According to the UCR (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 14.52*** 4.77 Linear After Adoption -.06 .26 Murder Rate 1.87*** .27 Percentage Living in Poverty -.02 .05 Divorce Rate 1.57** .59 Republican Governor -.79 .95 Incarceration Rate 15.33* 6.25 * p-value < .05 ** p-value < .01 *** p-value < .001
223 Rape Rate as Measured by NIBRS
With regard to the NIBRS rape rate, the incarceration rate was negatively, and significantly associated with the rape rate, as shown in Table 4.35. This finding suggests that as the incarceration rate within a state increased, the rate of rape decreased.
Table 4.35. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 33.64*** 6.57 Civil Commitment Policy 1.77 2.30 Murder Rate 1.91 .69 Percentage Living in Poverty -.01 .01 Divorce Rate -1.08 .99 Republican Governor -.14 1.17 Incarceration Rate -.01** .004 † p-value <.10 ** p-value < .01
After all predictors of the rate of rape (measured by NIBRS) were lagged by one year, the effect of having a civil commitment policy became non-significant. The incarceration rate remained significant, and the percentage of the population was negative and significant. That is, as the state‟s rate of poverty and incarceration increased, the rate of rape decreased.
Table 4.36. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 34.76*** 5.17 Civil Commitment Policy 2.51 2.18 Murder Rate 1.43 .73 Percentage Living in Poverty -.07** .02 Divorce Rate -.60 .79 Republican Governor -.94 1.12 Incarceration Rate -.12* .004 † p-value <.10 ** p-value < .01
224
When the rate of rape (measured by NIBRS) was lagged by two years, as shown in Table 4.36, the civil commitment policy predictor remained non-significant. The effects of poverty and incarceration rate were consistent with their effects in Table 4.37.
Table 4.37. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 33.14*** 6.14 Civil Commitment Policy 2.17 2.31 Murder Rate .77 .80 Percentage Living in Poverty -.08** .02 Divorce Rate .55 .98 Republican Governor -1.79 1.60 Incarceration Rate -.01* .005 † p-value <.10 ** p-value < .01
The final method of assessing the potentially deterrent effect of adopting a civil commitment statute was to include a linear measure of time following the adoption of the policy. As shown in Table 4.38, the adoption of a civil commitment statute did not significantly impact the rape rate over time. The effect of incarceration rate remained significant, and consistent with the preceding models. A positive, significant effect of murder rate was also present in this model, which indicated that when a state‟s murder rate increased, so did its rape rate (as measured by NIBRS).
225
Table 4.38. Fixed Effects Model Predicting the Rate of Rape According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 32.48*** 6.07 Linear After Adoption .43 .40 Murder Rate 1.80* .69 Percentage Living in Poverty -.005 .01 Divorce Rate -.72 .85 Republican Governor -.08 1.07 Incarceration Rate -.01** .004 † p-value <.10 ** p-value < .01
Fixed Effects Model Predicting the Effect of Civil Commitment on the Forcible Sodomy Rate
With regard to the NIBRS sodomy rate, the findings presented in Table 4.12 show that both the percentage of the population living in poverty and the incarceration rate were significantly and positively associated with the rate of sodomy. These effects suggest that as the percentage of the population living in poverty and the incarceration within a state increased, so did the rate of forcible sodomy. There was no significant impact of having a civil commitment statute on the rate of forcible sodomy.
Table 4.39. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 236.30* 109.23 Civil Commitment Policy -19.96 45.66 Murder Rate 10.48 20.30 Percentage Living in Poverty .89* .43 Divorce Rate -28.11 16.25 Republican Governor 13.84 33.47 Incarceration Rate .54*** .12 * p-value < .05
226 Once the predictors were lagged by one year, the percentage of the population living in poverty no longer had a significant effect on the rate of forcible sodomy (see
Table 4.40). The effect of incarceration rate was consistent with that presented in Table
4.39.
Table 4.40. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 243.78* 100.56 Civil Commitment Policy -32.70 44.00 Murder Rate 6.84 14.77 Percentage Living in Poverty .58 .78 Divorce Rate -30.11 16.55 Republican Governor 34.18 36.87 Incarceration Rate .56*** .13 * p-value < .05
Consistent with the preceding models, the outcome variable of forcible sodomy measured two years after all included predictors was not significantly impacted by the adoption of a civil commitment statute (see Table 4.41). The positive and significant impact of incarceration rate was consistent with the preceding models. There was also a negative and significant impact of the divorce rate on the rate of forcible sodomy. This finding suggests that as the rate of divorce increased within a state, the rate of sodomy two years later decreased. This finding may be the result of an increase in formal social control, following the breakdown of informal social controls, consistent with the tenets of social disorganization theory.
227
Table 4.41. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 301.02*** 66.34 Civil Commitment Policy -53.16 43.95 Murder Rate -5.02 15.51 Percentage Living in Poverty .84 .59 Divorce Rate -36.89** 11.94 Republican Governor 67.82 36.33 Incarceration Rate .52*** .12 * p-value < .05
In the final model predicting forcible sodomy, using a linear measure of policy adoption, the only significant substantive predictor was the positive incarceration rate, shown in Table 4.42. There was no significant deterrent effect of adopting a civil commitment statute, over time.
Table 4.42. Fixed Effects Model Predicting the Rate of Forcible Sodomy According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 234.02* 106.47 Linear After Adoption -2.36 6.86 Murder Rate 11.64 21.38 Percentage Living in Poverty .87 .41 Divorce Rate -28.76 17.86 Republican Governor 11.43 29.96 Incarceration Rate .54*** .12 * p-value < .05
228 Fixed Effects Model Predicting the Effect of Civil Commitment on the Rate of Forcible Fondling
The findings presented in Table 4.43 show that the incarceration rate was a positive, significant predictor of the rate of forcible fondling. However, the adoption of a civil commitment statute did not significantly impact the outcome.
Table 4.43. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 55.02* 24.96 Civil Commitment Policy -8.64 12.67 Murder Rate -1.81 4.12 Percentage Living in Poverty .20 .09 Divorce Rate -3.99 3.13 Republican Governor 8.68 7.54 Incarceration Rate .12** .03 ** p-value < .01
The findings shown in Table 4.44, in which the predictors of the rate of forcible fondling were lagged by one year, are consistent with the findings of Table 4.43. That is, there was a positive, significant effect of incarceration rate on the rate of forcible fondling.
Table 4.44. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 57.28* 22.70 Civil Commitment Policy -4.86 11.32 Murder Rate -2.22 3.91 Percentage Living in Poverty .05 .16 Divorce Rate -4.71 2.89 Republican Governor 14.27 8.70 Incarceration Rate .12** .03 ** p-value < .01
229
When the rate of forcible fondling was measured two years later than the predictor variables, there were three significant predictors. There was a negative effect of divorce, suggesting that as the rate of divorce within a state increased, the rate of forcible fondling decreased. Having a Republican governor or an increase in the incarceration rate positively impacted the forcible fondling rate.
Table 4.45. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 53.59*** 17.08 Civil Commitment Policy -4.47 10.36 Murder Rate -1.86 3.31 Percentage Living in Poverty .18 .16 Divorce Rate -5.30* 2.11 Republican Governor 25.61* 10.07 Incarceration Rate .09* .04 ** p-value < .01
Even a linear measure of the policy predictor did not result in a significant effect.
In this model, shown in Table 4.46, the only significant predictor was the positive impact of incarceration rate.
Table 4.46. Fixed Effects Model Predicting the Rate of Forcible Fondling According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 57.97* 22.84 Linear After Adoption -1.67 2.04 Murder Rate -1.29 4.46 Percentage Living in Poverty .19 .09 Divorce Rate -5.14 2.81 Republican Governor 8.08 6.87 Incarceration Rate .12** .03 ** p-value < .01
230 Fixed Effects Model Predicting the Effect of Civil Commitment on the Rate of Sexual Assault with an Object
In a model predicting the rate of sexual assault with an object, the only significant predictor was the incarceration rate. As the incarceration rate within a state increased, so did the rate of sexual assault with an object.
Table 4.47. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 202.87 151.06 Civil Commitment Policy -6.85 41.02 Murder Rate 11.89 23.13 Percentage Living in Poverty .25 .47 Divorce Rate -13.62 16.22 Republican Governor 6.46 27.90 Incarceration Rate .74** .22 * p-value < .01
The findings shown in Tables 4.48 and 4.49, in which the policy predictor and covariates were lagged one year and two years, respectively, are consistent with those of
Table 4.47. That is, only the incarceration rate was a significant predictor of the rate of sexual assault with an object.
Table 4.48. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 207.84 151.17 Civil Commitment Policy -13.40 40.24 Murder Rate 3.57 19.39 Percentage Living in Poverty .46 .80 Divorce Rate -11.85 19.96 Republican Governor 35.16 32.55 Incarceration Rate .70** .20 * p-value < .01
231
Table 4.49. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 362.49 107.09 Civil Commitment Policy -33.47 41.36 Murder Rate -20.15 19.08 Percentage Living in Poverty .76 .71 Divorce Rate -28.64 16.10 Republican Governor 63.30 40.86 Incarceration Rate .59** .19 * p-value < .01
A linear measure of the policy predictor did not significantly impact the rate of sexual assault with an object. Consistent with the preceding models, the only significant predictor of the rate of sexual assault with an object was the incarceration rate.
Table 4.50. Fixed Effects Model Predicting the Rate of Sexual Assault with an Object According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 197.35 147.35 Linear After Adoption -.03 5.48 Murder Rate 12.28 23.16 Percentage Living in Poverty .25 .45 Divorce Rate -12.80 16.75 Republican Governor 5.10 25.50 Incarceration Rate .73** .22 * p-value < .01
Fixed Effects Model Predicting the Effect of Civil Commitment on the NIBRS Statutory Rape Rate
The model predicting the rate of statutory rape within states, shown in Table 4.51, shows that none of the substantive predictors analyzed here was significantly associated with the outcome.
232 Table 4.51. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant -3.12 8.64 Civil Commitment Policy -4.93 2.95 Murder Rate 2.23 1.33 Percentage Living in Poverty -.01 .04 Divorce Rate .45 .74 Republican Governor -1.10 1.94 Incarceration Rate .04 .02 † p-value < .10 ** p-value < .01 *** p-value < .001
Once predictors of the rate of statutory rape were lagged one year, the effect of incarceration rate became positive and significant. This finding, shown in Table 4.52, suggests that when the incarceration rate in a state increases, so does the rate of statutory rape one year later. One possible explanation for this pattern is that as citizens recognize that there is more enforcement of the laws, they are more likely to report violations.
Table 4.52. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 6.42 3.70 Civil Commitment Policy -4.74 2.23 Murder Rate .27 .88 Percentage Living in Poverty -.02 .04 Divorce Rate .34 .83 Republican Governor 1.35 2.95 Incarceration Rate .02** .01 † p-value < .10 ** p-value < .01 *** p-value < .001
When the outcome was measured two years later than all predictors (see Table
4.53), there was a significant, positive effect of the percentage of the population living in
233 poverty on the rate of statutory rape. That is, when the poverty rate in a state increased, the rate of statutory rape two years later also increased.
Table 4.53. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 11.33 7.11 Civil Commitment Policy -5.34 3.42 Murder Rate -.39 .54 Percentage Living in Poverty .12* .05 Divorce Rate -.65 1.27 Republican Governor 3.56 3.59 Incarceration Rate .01 .01 † p-value < .10 ** p-value < .01 *** p-value < .001
In the model including a linear policy predictor, shown in Table 4.54, there were no significant predictors. This finding, combined with the findings of the preceding three models, indicate that there was no deterrent effect of a civil commitment policy on statutory rape.
Table 4.54. Fixed Effects Model Predicting the Rate of Statutory Rape According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant -2.39 6.62 Linear After Adoption -.80 .37 Murder Rate 2.52 1.35 Percentage Living in Poverty .02 .04 Divorce Rate .01 .72 Republican Governor -1.55 1.54 Incarceration Rate .04 .02 † p-value < .10 ** p-value < .01 *** p-value < .001
234 Fixed Effects Model Predicting the Effect of Civil Commitment on the Incest Rate
As shown in Table 4.55, the incarceration rate was positively and significantly associated with the rate of incest. No other predictors were significant.
Table 4.55. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 34.62 15.52 Civil Commitment Policy -8.23 5.31 Murder Rate .22 2.43 Percentage Living in Poverty .03 .05 Divorce Rate -4.05 3.38 Republican Governor 4.21 3.50 Incarceration Rate .07* .02
The findings reported in Table 4.56 represent the effect of the included predictors on the rate of incest, measured one year later. The positive effect of incarceration rate that was present in Table 4.55 remained significant here. There was also a significant positive effect of having a Republican governor.
Table 4.56. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor Lagged One Year) Variable Coefficient Robust Standard Error Constant 35.20* 12.03 Civil Commitment Policy -8.29 5.17 Murder Rate -.97 2.41 Percentage Living in Poverty .10 .07 Divorce Rate -3.94 2.58 Republican Governor 7.46* 3.33 Incarceration Rate .07* .03
As in the two preceding models, there was a positive and significant effect of the incarceration rate on the rate of statutory rape, measured two years later (see Table 4.57).
As above, a possible explanation for this pattern is that as citizens recognize that there is
235 more enforcement of the laws, they are more likely to report violations.No other predictors were significant.
Table 4.57. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Civil Commitment Policy Predictor Lagged Two Years) Variable Coefficient Robust Standard Error Constant 28.67 8.35 Civil Commitment Policy -7.03 4.98 Murder Rate -1.27 1.99 Percentage Living in Poverty .04 .07 Divorce Rate -2.91 1.67 Republican Governor 8.55 4.00 Incarceration Rate .08** .04
The linear measure of policy was not significantly associated with the rate of incest within the state. As in the preceding models, there was a significant, positive effect of incarceration rate on the outcome.
Table 4.58. Fixed Effects Model Predicting the Rate of Incest According to NIBRS (Linear Civil Commitment Policy Predictor) Variable Coefficient Robust Standard Error Constant 34.00 16.99 Linear After Adoption -1.02 .92 Murder Rate .70 2.57 Percentage Living in Poverty .02 .05 Divorce Rate -4.39 3.92 Republican Governor 3.26 3.19 Incarceration Rate .07** .02
Discussion of Study 2
Study 2 was a test of the deterrent effect of registration/notification policies and civil commitment statutes on sexual offending. Fixed effects models, controlling for five theoretically important variables, were used to assess the impact of the two primary explanatory variables, on the rate of rape (as reported by the UCR and the NIBRS,
236 respectively), forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest. As shown in Table 4.59, the impact of these policies were largely positive, contrary to the hypothesized results.
Table 4.59. Summary of Significant Effects Across Crimes Analysis Deterrence Effect Opposite Effect (Decrease in Rate) (Increase in Rate) Registration/Notification Policy Predictor Lagged by Lagged by 0 years 1 year 2 years Linear 0 years 1 year 2 years Linear Rape (UCR) mdi mdi mdi mdi Rape (NIBRS) p p i mi m Forcible Sodomy d i i i *i Forcible Fondling d i i gi *pi Sexual Assault with an i i i *i Object Statutory Rape i pi i Incest i i *i Civil Commitment Policy Predictor Lagged by Lagged by 0 years 1 year 2 years Linear 0 years 1 year 2 years Linear Rape (UCR) p mdi mdi mdi mdi Rape (NIBRS) i pi pi i m Forcible Sodomy d pi i i i Forcible Fondling d i i gi i Sexual Assault with an i i i i Object Statutory Rape i p Incest i gi i i * = policy predictor i = incarceration rate d = divorce rate m = murder rate p = percentage living in poverty
Hypotheses
With regard to Hypothesis 1, that adoption of a registration/notification policy would lead to a decrease in sexual offending, no support was found. In some models, the
237 effect of adopting a registration/notification policy was negative, but it never reached significance. In fact, in four models, the impact of the registration/notification policy predictor was in the opposite direction.
With regard to Hypothesis 2, that adoption of a civil commitment statute would lead to a decrease in sexual offending, no support was found. There were also no models in which the civil commitment policy predictor had a positive impact on the rates of sexual offending.
Because there were no deterrent effects of either policy, no support was found for
Hypotheses 3, 4, or 5, regarding the magnitude of these effects across crimes.
Summary of Findings
Two measures relevant to the state‟s criminal justice system were included in these models: murder rate and incarceration rate. In this study, an increase in the state‟s murder rate was significantly associated with an increase in the state‟s rape rate as measured by the UCR. This finding suggests that as the murder rate within a state increased, so did the rate of rape. The murder rate was also positively and significantly associated with the rate of rape, as measured by NIBRS, in half of the models predicting rape rate.
The most consistent predictor of the rates of sexual offenses studied here was incarceration rate. In models predicting all outcomes except the rape rates as measured by NIBRS, an increase in the state‟s incarceration rate was associated with an increase in the rate of sexual crimes. This finding suggests that when states experienced increases in
238 incarceration rates, they also experienced increases in the rates of sexual crimes. As suggested in the results, this pattern may result from citizens believing that there is more enforcement of the laws and that therefore their report of violations are more likely to lead to the punishment of offenders.
As discussed in Chapter 2 (see pages 44-45), and demonstrated by Baron and
Straus (1987), social disorganization characteristics are related to rates of criminal activity. A large literature in criminology suggests that areas that have higher rates of social disorganization also have higher crime rates. Consistent with that general finding, one of the most robust indicators of social disorganization, percentage living in poverty, was a significant predictor of the rate of rape as measured by NIBRS in most models included in this study. Another social disorganization indicator, divorce rate, was a significant predictor of the rape rate as measured by the UCR.
In contrast to this study‟s support for the importance of social disorganization, this study provided no support for the deterrence theory prediction regarding the adoption of new laws, which would have predicted a decrease in sexual offending after the implementation of additional sex offender legislation, such as registration/notification policies and civil commitment statutes. Yet, it is possible that some of the positive policy impacts on rates of sexual offending found here, are the result of an effective policy. By adopting registration/notification policies and civil commitment statutes, states are making clear that sexual offending is a serious crime which requires serious consequences, and will be treated seriously within the criminal justice system. Such a clear stance on sexual offending may, in fact, encourage victims to report crimes that they otherwise would not have reported. If they believe that they will be treated with sincerity
239 and respect, victims may be more likely to come forward, thus, increasing the official rates of sexual offending.
In a study of the impact of changes in rape legislation, Horney and Spohn (1991), found that changes in the rape legislation resulted in an increase in the number of reported rapes in two of the six jurisdictions that they examined. The authors‟ explanation for this finding was that the publicity associated with the new laws brought increased attention to the crime of rape among both the public and criminal justice officials. These changes and the attention they drew made victims more comfortable and more likely to report their victimization. It was also possible that, as a result of these changes, law enforcement officials handled rape victims better, thus making it easier for others to come forward.
Even though there is no clear deterrent effect of these policies according to this study, these null results do not definitively indicate that these policies are ineffective. It is possible that the small sample sizes currently available in the NIBRS, and the fact that only five of the 15 states in this analysis had both policies in place, meant there was insufficient power to detect effects.
However, despite the limited sample size, it is possible to look for evidence of a deterrent effect, using changes in rates of sexual offending within states over time.
Graphs representing the rate of sexual offending over a 17-year period within the five states that had adopted both policies are shown in Figures 4.8 through 4.12. If the two policies of interest had had deterrent effects on offending, I would have expected to see decreases in sexual offending following the adoption of these policies.
Registration/No tification Only
Registration/No tification and Civil Commitment 240 With the exception of South Carolina, this pattern is clearly absent. South
Carolina adopted both policies one year apart (civil commitment in 1998 and
registration/notification in 1999). The years closely following these two adoptions did
show decreases in the rates of sexual offenses, even though the other four states showed
increases during those years. However, the decrease was short, followed by a substantial
increase in the rates of sexual offending. If this brief decrease was in fact due to the new
legislation, it appears that those effects were very short-lived. If this brief decrease was
due to other factors, then it appears that the policies being studied here were ineffective,
particularly combined with the fact that none of the other four states showed decreases in
sexual offending following policy adoption.
Figure 4.8. Rates of Sexual Offending in Iowa and Timing of Policy Adoption
400 Registration/Notification Civil Commitment Adopted Adopted 350 00 300 UCR Rape Rate
Rates of NIBRS Rape Rate Sexual 250 Offending Forcible Sodomy Rate 200 Forcible Fondling Rate 150 Rate of Assault with an Object
100 Incest Rate
50 Statutory Rape Rate
0
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
242
Figure 4.9. Rates of Sexual Offending in New Hampshire and Timing of Policy Adoption
450 Registration/Notification Civil Commitment Adopted Adopted 400
350 UCR Rape Rate 00
Rates of 300 Sexual NIBRS Rape Rate Offending 250 Forcible Sodomy Rate
200 Forcible Fondling Rate
150 Rate of Assault with an Object
100 Incest Rate
50 Statutory Rape Rate
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
243
Figure 4.10. Rates of Sexual Offending in North Dakota and Timing of Policy Adoption
450 Registration/Notification Adopted Civil Commitment 400 Adopted
350 UCR Rape Rate 00 Rates of 300 Sexual NIBRS Rape Rate
Offending 250 Forcible Sodomy Rate
200 Forcible Fondling Rate
150 Rate of Assault with an Object
100 Incest Rate
50 Statutory Rape Rate
0
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
244
Figure 4.11. Rates of Sexual Offending in South Carolina and Timing of Policy Adoption
700 Registration/Notification Adopted 600 Civil Commitment Adopted Rates of 500 UCRUCR Rape Rape Rate Rate Sexual Offending NIBRSNIBRS Rape Rape Rate Rate 400 ForcibleForcible Sodomy Sodomy Rate Rate
Forcible Fondling Rate 300 Forcible Fondling Rate
RateRate of of Assault Assaul twith with an an Object Object 200 IncestIncest Rate Rate
100 StatutoryStatutory Rape Rape Rate Rate
0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
245
Figure 4.12. Rates of Sexual Offending in Virginia and Timing of Policy Adoption
400 Registration/Notification Civil Commitment Adopted Adopted 350 00
300 UCR Rape Rate Rates of Sexual 250 NIBRS Rape Rate Offending Forcible Sodomy Rate 200 Forcible Fondling Rate 150 Rate of Assault with an Object
100 Incest Rate
50 Statutory Rape Rate
0
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Chapter 5
General Discussion and Conclusions
This dissertation sought to answer two questions. First, what state-level factors lead to the adoption of sex offender legislation? Second, what impact does sex offender legislation have on rates of sexual offending within the state? Two studies were used to address these questions. The findings of these two studies are summarized below. In addition, the strengths and weaknesses of the research are presented, and the theoretical and policy implications of the findings are discussed.
Summary of Findings
Study 1 used an event history analysis to examine the risk and timing of adopting a registration/notification policy and the risk and timing of adopting a civil commitment statute. Study 1 had four primary findings.
First, states in the Southern region were less likely than states in all other regions to adopt registration/notification policies, or they did so later. This finding is consistent with the Southern culture of respect for state‟s rights and freedom from government involvement. Moreover, the Northeast and Western regions both had historical incidents of sexual violence, which led to changes in legislation (i.e., the murder of Megan Kanka in New Jersey and the three assaults in two years by released sex offenders in
Washington). Although the Jacob Wetterling Act originated at the federal level, the boy
247 for whom the law is named was abducted in the Midwestern state of Minnesota. Thus, these important cases may also explain why states in other regions of the country, compared to states in the South, were at an increased risk of adoption. This regional effect was the only finding that was consistent for both the adoption of registration/notification policies and the adoption of civil commitment statutes.
The second primary finding of this research was that states with higher levels of unemployment were at a decreased risk of adopting civil commitment statutes, or did so later, than states with lower levels of unemployment. Recent research on the use of incarceration has found that a state‟s economic situation has impacted criminal justice decisions. For example, a study of the effect of sentencing guidelines found that states with full prisons, or prisons that are more expensive to operate, were unlikely to see increases in prison populations based on sentencing guidelines (Nicholson-Crotty, 2004).
In these resource burdened states, either the incarcerated population stayed the same, or it decreased, after the guidelines were implemented.
The economic downturn that has taken place in the United States over the past several years has brought to light some critical shortfalls in criminal justice funding. The state of California, in particular, needs to decrease the size of its prison population by 25 percent (Barker, 2011). Incarceration is a staple of the criminal justice system process, and has been used as a form of punishment since the founding of the United States.
When even this most basic element of the criminal justice system is at risk of reduction and is affected by the economy, the likelihood of adopting new policies is extremely limited. Consistent with this logic, Study 1 found that the unemployment rate was negatively related to the risk and timing of policy adoption. Thus, it appears that states
248 struggling with economic hardship are generally less likely to adopt policies that are expensive to implement, particularly when the effectiveness of those policies is unproven.
Third, traditional diffusion measures had significant, positive impacts on both policies investigated here. Regarding the adoption registration/notification policies, a measure of diffusion at the regional level had a significant impact such that as more states in the region adopted a registration/notification policy, the remaining states were at an increased risk of adopting a registration/notification policy or doing so sooner. Civil commitment statutes were also positively impacted by diffusion at the national level. As more states within the nation adopted a civil commitment statute, the remaining states were at an increased risk of adopting a civil commitment statute or doing so sooner.
Finally, one diffusion mechanism was associated with the risk and timing of adopting a civil commitment statute. Federal pressure, as measured by the adoption of the Jacob Wetterling Act and the Supreme Court ruling in the case of Kansas v.
Hendricks, was negatively related to the risk and timing of policy adoption. Because of their high correlation with time, these effects are likely the result of multicollinearity.
Study 2 used two sets of panel data to examine the deterrent effect, if any, of adopting a registration/notification policy or a civil commitment statute, on the rates of six sexual offenses (forcible rape, forcible sodomy, forcible fondling, sexual assault with an object, statutory rape, and incest). Results did not indicate a deterrent effect of these policies on rates of sexual offending either between states or within states. In fact, for four of the outcomes (forcible sodomy, forcible fondling, sexual assault with an object, and incest), the adoption of a registration/notification policy had a positive impact on the
249 rates of offending. Although this finding might suggest that the policy led to more sexual offending, there is another possible explanation for this finding. It may be that the political and social climate in the state that led to the adoption of this policy also made victims feel more comfortable reporting sexual offending. It may also be that the increase in attention to the issue of sexual offending led to a crackdown by the criminal justice system, in which police and prosecutors responded seriously to accusations of sexual offending. The lack of significant effects in the remaining models may be due to a limited sample size and resulting low statistical power. It is possible that with data from additional states, or data spanning additional years, a deterrent effect of adopting these policies would be present.
Strengths, Limitations, and Considerations for Future Research
This dissertation research has several strengths. Two strengths characterized both
Study 1 and Study 2. First, the two studies included multiple years of data and used multiple datasets. Second, there was sufficient variety in the two policies being analyzed
(registration/notification and civil commitment) to make comparisons. Another strength of Study 1 was that it used the entire population (the states), not a sample. In general, a state-level analysis was the appropriate level of specificity for addressing this research question, because the policies being studied were adopted by each state. However, some policies are adopted at the county or city-level (e.g., residency restrictions or GPS monitoring), which would mean that in future studies another level of analysis would be appropriate when examining those policies. Richer data is sometimes available at the
250 national, county, or city-level. Nevertheless, for the policies considered here, a state- level analysis was the appropriate approach. An additional strength of Study 2 was the use of a fixed effects design, which is a conservative test for effects.
However, both Study 1 and Study 2 also had limitations. Additional explanatory variables may have been useful in Study 1 for predicting the timing of policy adoption.
In particular, a measure of the religious composition of the state may have been informative. However, no such variable is available at the state level that is measured annually. Although the U.S. Census Bureau provided estimates of the percentage of the population in each state who identify themselves as Christians, this data was reported only per decade, and was reported in different formats in 1990 and 2000 (with more denominations listed in the later year).
Two types of future studies could help to clarify the role of various state-level characteristics on policy adoption. First, the study of an additional criminal justice policy
(though not a sexual offending policy) should be analyzed using the same data and methods as used here. This analysis would provide a baseline level for each of the effects. I believe that sexual offending statutes are unique and differ from other policies that have been evaluated using the same methodology employed here. However, it may be that criminal justice policies in general differ from other types of policy. By using an event history analysis to predict the risk and timing of adopting a different criminal justice policy, future researchers could determine whether criminal justice policy in general is more similar to other types of policies (e.g., welfare) or is more similar to sexual offense legislation.
251 Second, research gauging the motivations behind those who actually write and pass the legislation would be useful. One such study has been conducted, which reported interviews of a sample of legislators in Illinois (Sample & Kadleck, 2008). These interviews revealed that legislators‟ ideas about sex offenders, even though generally not based in fact, impacted their support for sex offender legislation. The content of these interviews was consistent with the ideas presented here, that laws governing sexual offenders are qualitatively different from other types of legislation. The authors described a nonrecursive media-public-legislative environment: “a cycle emerges in which high-profile cases lead to extensive media coverage, which permits public outcry and concern and influences public officials‟ perceptions. Public officials then feel the need to respond to their and the public‟s concerns, which prompts more media attention, and the cycle continues” (Sample & Kadleck, 2008, page 60). This cycle exists outside of the scope of traditional internal determinants and diffusion variables that are typically considered in theories of policy adoption.
A content analysis of the transcripts of legislative sessions could contribute to an understanding of the cycle of sexual offense legislation and illuminate additional motivations or intentions for passing the laws. Residence restrictions are relatively new additions to sexual offender legislation, and have been implemented in some, but not all, states. Transcripts of legislative discussions debating this policy may be useful because they are timely and the policy is somewhat controversial.
Study 2‟s primary limitation was the fact that NIBRS data was not available for all states. The number of states that report to NIBRS has increased dramatically over the past 10 years, and will likely continue to increase. However, for the period of analysis
252 covered here, there were only 15 states available, only 5 of which had adopted both registration/notification policies and civil commitment statutes. If a larger sample of states were available, a significant deterrent effect of policy adoption may be detected.
However, it is also possible that there is no deterrent effect of these laws on the rates of sexual offending. For example, a study of statutory rape legislation indicated that adolescent sexual behavior did not vary as a result of state-differences in the definition of statutory rape, suggesting that the law did not impact behavior (Koon-Magnin, Kreager,
& Ruback, 2010).
To determine whether or not a deterrent effect of these policies truly exists, a follow-up of Study 2 should be conducted in the future, once additional years of data or more states are available for analysis. It would also be informative to do a case study of one state from the start of discussing policy adoption, to policy implementation, through evaluation of the policy. Because it is the only state in which there appears to be a deterrent effect of these two policies, South Carolina may be an appropriate place to conduct this qualitative analysis. Such a study would provide information about additional considerations that should be taken into account in studies of policy adoption and evaluation.
Implications
Important theoretical and policy implications can be drawn from the findings presented here. Study 1 provided partial support for both Internal Determinants and
Diffusion theory, but it also found some inconsistencies with both theories. Some of the
253 internal determinants that were included in Study 1 had a significant impact on the risk and timing of policy adoption. However, many of the variables that are traditionally associated with policy adoption (e.g., the percentage of the population who identify themselves as Republican, and structural-functional characteristics such as the murder rate and whether the state allowed the death penalty) were not predictive of the risk or timing of adopting these policies. This finding suggests that the laws regulating sexual offending are qualitatively different than other types of policies.
The most recent and most far-reaching piece of sexual offense legislation, the
2006 Adam Walsh Act, contains seven sections named for specific victims of sexual crimes (Bandy, 2011). These types of laws are passed out of concern for the public, especially children, and often as a direct response to a particularly gruesome crime.
Thus, the internal determinants that are traditionally used to explain policy adoption may not be as useful for predicting the adoption of sexual offending legislation.
Uniqueness of Sexual Offenses
There are two primary factors that make sexual offense legislation unique. First, because of the controversial nature of sexual offending, there are few, if any, legislators who are willing to oppose more restrictive measures. No legislators want to appear “soft on crime,” but this effect is particularly pronounced when considering sexual crimes or crimes against children. Even if a legislator believes that a law may be too strict or invasive, it would be a politically unpopular move to oppose it, and would put his or her chance of reelection at risk.
254 Second, laws governing sexual offenders are not based on scientific evidence.
These laws treat sexual offenders as a homogenous group with a consistently high rate of offending. In fact, sex offenders differ in many important ways, and have lower recidivism rates than most other types of offenders. Policies, particularly those that are expensive to implement and maintain, are generally adopted with the hope of impacting some desired outcome. However, if there is no true relationship between the policy being adopted and the desired impact, the policy will not have the desired effect (Mears, 2011).
Laws directed at sexual offenders may be adopted for symbolic reasons (i.e., to express public sentiment and to show citizens that legislators are responsive to their needs), rather than to reduce the rates of sexual offending. If lawmakers truly want to reduce the rates of sexual offending, they should pursue initiatives that are consistent with empirical knowledge about sexual offender behavior, rather than initiatives that are consistent with public perceptions.
The assumptions upon which most sex offender policies are based are not empirically supported (Wright, 2009). One example is Megan‟s Law, which aims to protect the public from sexual predators by informing the public about who the predators are and where they live. All types of sexual offenders appear on the public registry, including statutory rape offenders who engaged in consensual sex, parents who abused their own children, and sexually violent predators who sexually assaulted strangers prior to murdering them. The level of risk posed by these offenders differs substantially, but the registry presents all three of these hypothetical offenders as equally dangerous.
Another misleading element of Megan‟s Law is that the majority of sexual assaults are committed by acquaintances (Williams, 2009). Encouraging citizens to focus on the
255 strangers listed in a sex offender registry may distract them from the possibility of victimization at the hands of someone they know.
Diffusion
Study 1 provided support for diffusion theory. However, additional information could be gained by including measures of not just whether a policy was implemented, but how successful it was, when enacted in other states. As Volden (2006) has argued, the success of the policy in other states impacts the likelihood of its adoption elsewhere.
Legislators generally do not want to spend resources implementing a policy that has minimal effect on the desired outcome. Thus, they may wait and see how successfully the policy is implemented elsewhere. If it proves effective, legislators in the remaining state will likely move towards implementing the policy in their own states, as well.
Future research on this issue should take into account the success of the laws in surrounding states (i.e., the kinds of information examined in Study 2 here), not just the proximity of the states.
In a future study predicting the risk and timing of adopting sex offender legislation, diffusion measures based on proximity should be paired with measures based on shared political affiliation of state leadership. Imagine that a state run by a Republican governor has three neighbors (sharing a border), two of which are led by Democratic governors and one of which is led by a Republican governor. It is reasonable to assume, even holding constant other state-level similarities, that the state in question would be more likely to follow the same policy adoption trend as its neighbor with leadership by
256 the same party. However, traditional diffusion analysis does not include this level of nuance, focusing only on physical proximity. An analysis including measurements composed of both proximity and state-level characteristics would be consistent with the concept of a hybrid model, incorporating elements of both internal determinants and diffusion theories.
Deterrence
Study 2 provided no direct support for deterrence theory and no deterrent effect was detected for either registration/notification policies or civil commitment statutes on rates of sexual offending. The null finding for civil commitment statutes is less surprising than the null finding for registration/notification because civil commitment statutes do not meet the three conditions required by deterrence theory: swiftness, certainty, and severity. Registration/notification meets all three of these requirements, so the lack of a significant deterrent effect of registration/notification laws is a more appropriate evaluation of deterrence theory. However, failure to find support for deterrence theory in Study 2 does not necessarily mean that there is no deterrent effect, because the sample employed here may have been too small to detect it.
Need for Evaluation
Because of the great cost associated with adopting these policies, it makes sense to conduct some sort of evaluation prior to implementing them fully. A pilot program at
257 the city or county-level may serve this purpose. Or, the policies could be implemented for a probationary period, which would require a vote to renew or eliminate the policy after a specified amount of time. Advances in the field of criminal justice policy evaluation (see, for example, Mears, 2010) could be used to appropriately and completely assess the effectiveness and efficiency of the policy.
Moreover, the legislators who adopt policy, the practitioners who implement it, and the researchers who evaluate it, operate autonomously. More communication between these three groups could benefit the entire process. As noted by Mears, the process of implementing a policy is a critical element to the policy‟s success. A policy may appear to be ineffective, but if those implementing the policy have been doing so incorrectly, the lack of an effect on the desired outcome may not be because of the policy, but because of the fact that it was improperly implemented. Although individuals who evaluate policy should be objective and unbiased, they also need to know the outcome that the policy is intended to produce. If the evaluators focus only on the recidivism rate, when the goal of the policy was to reduce costs, then the evaluation does not provide all of the desired information. Evaluators should focus on a variety of outcomes, including both cost and other instrumental goals.
But, regardless of their cost or effectiveness, some policies may still be worthwhile (Sample, 2011). As discussed throughout this paper, many sexual offense laws are passed in response to specific cases that generate outrage or fear in the community. The presence of publicly available sex offender registries, though they appear to be largely ineffective at reducing the rates of sexual offending, may make the public feel safer and the citizenry feel as though their legislators are responding to their
258 concerns (Sample, 2011). This symbolic impact may be just as important as the instrumental goal of reducing sexual offending, and should be considered by researchers of sex offender legislation.
Interdependence of Adoption and Evaluation
Scholars who focus on criminal justice policy may be well served to think more broadly about the issues involved. That is, most researchers of policy adoption focus solely on adoption, and not evaluation. Most researchers who conduct policy evaluation do not study policy adoption. More comprehensive research would be possible if researchers were familiar with both literatures and had a broader view of the policy process.
It is a mistake to treat these two literatures as separate because the process of adoption and the process of evaluation are conceptually and practically intertwined. As shown in Figure 5.1, the legislative actions that states take can impact the actions in other states. Every policy adoption originates somewhere, and is then implemented. If a policy is relevant to more states than just the original adopter (as most policies are), then lawmakers in other states may observe and evaluate the policy as it is implemented in the state of original adoption (effect depicted as “a” in Figure 5.1).
Assuming that the policy is successful, or that it meets some need (instrumental or symbolic), states that observed the original adoption and implementation may adopt the policy as well (effect depicted as “b” in Figure 5.1). If there are elements that they find
259 undesirable or that they can improve upon, they may adopt a modified version of the policy.
Once adopted in other states, the original adopter may reevaluate its own policy for effectiveness (effect depicted as “c” in Figure 5.1), or may modify its policy based on successful modifications in other states (effect depicted as “d” in Figure 5-1). This modification may then be witnessed, observed, and ultimately implemented in other states (effect depicted as “e” in Figure 5.1). This cycle can continue until all states have policies that their lawmakers are pleased with.
Policy Adoption by Other States Observe Other States Adopt Policy Original Adopter a Implementation or b (perhaps with modifications) Modification by Original c Adopter and Evaluate
Original Adopter May Reevaluate Policy based on Modifications in
Other States
d
Original Adopter may Modify Policy based on Modifications in e Other States
Figure 5.1. The Interconnectedness of Policy Adoption and Evaluation
Conclusion
The research presented here generated as many new questions as it did answers.
Overall, the factors that affect sexual offending legislation are complex, and the
260 effectiveness of these laws is unclear. Additional research is required in order to uncover the motivations for and impacts of these laws. However, because of the symbolic role that these laws serve as an expression of public sentiment against sexual offending, it appears that registration/notification policies and civil commitment statutes are here to stay.
References
Adler, L. (2011). “„Blizzard‟: Like Meth, Only Legal.” WTAJ News. Accessed online:
http://wearecentralpa.com/fulltext/?nxd_id=236806.
Allison, P. D. (2006). “Fixed Effects Regression Methods in SAS.” Paper 184-31.
Anderson, A. L. & L. L. Sample. (2008). “Public Awareness and Action Resulting from
Sex Offender Community Notification Laws.” Criminal Justice Policy Review,
19(4), 371-396.
Association for the Treatment of Sexual Abusers. (2001). “ATSA Membership
Benefits.” Accessed online: www.ATSA.com/memBene.html.
Association for the Treatment of Sexual Abusers Board of Directors. (2010a). “The
Registration and Community Notification of Adult Sex Offenders.” Accessed
online: www.ATSA.com/ppnotify.html.
Association for the Treatment of Sexual Abusers Board of Directors. (2010b). “Civil
Commitment of Sexually Violent Predators.” Accessed online:
www.ATSA.com/ppcivilcommit.html.
Avrahamian, K. A. (1998). “A Critical Perspective: Do Megan‟s Laws‟ Really Shield
Children from Sex-Predators?” Journal of Juvenile Law, 19, 301-317.
Bandy, R. (2011). “Measuring the Impact of Sex Offender Notification on Community
Adoption of Protective Behaviors.” Criminology & Public Policy, 10(2), 237-
263.
262 Barker, V. (2011). “Decarceration : Political Possibility, Social Sentiment, and
Structural Reality.” Criminology & Public Policy, 10(2), 283-286.
Baron, L. & M. A. Straus. (1987). “Four Theories of Rape: a Macrosociological
Analysis.” Social Problems, 34(5), 467-489.
Barabási, A-L. (2002). Linked: The New Science of Networks. Cambridge, MA:
Perseus Publishing.
Berry, F. S. & W. D. Berry. (1990). “State Lottery Adoptions as Policy Innovations: An
Event History Analysis.” American Political Science Review. 84(2), 395-415.
Berry, F. S. (1994). “Sizing Up State Policy Innovation Research.” Policy Studies
Journal, 22(3), 442-456.
Benedict, H. (1992). Virgin or Vamp: How the Press Covers Sex Crimes. New York:
Oxford University Press.
Bock, G. (2011). “Stern: Ban Blizzard.” Altoona Mirror. Accessed online:
http://www.altoonamirror.com/page/content.detail/id/547284.html.
Buckley, J. & C. Westerland. (2004). “Duration Dependence, Functional Form, and
Corrected Standard Errors: Improving EHA Models of State Policy Diffusion.”
State Politics and Policy Quarterly, 4(1), 94-113.
Bryden, D. P. & S. Lengnick. (1997). “Rape in the Criminal Justice System.” The
Journal of Criminal Law and Criminology, 87(4), 1194-1384.
Bureau of Justice Statistics. (2003). “Recidivism of Sex Offenders Released from Prison
in 1994.” Washington, DC: U.S. Department of Justice.
Burstein, P. (2003). “The Impact of Public Opinion on Public Policy: A Review and an
Agenda.” Political Research Quarterly, 56(1), 29-40.
263 Canon, B. C., & L. Baum. (1981). “Patterns of Adoption of Tort Law Innovations: An
Application of Diffusion Theory to Judicial Doctrines.” American Political
Sceicne Review, 75, 975-987.
Carlsmith, K. J., J. Monahan, & A. Evans. (2007). “The Function of Punishment in the
Civil Commitment of Sexual Violent Predators.” Behavioral Sciences and the
Law, 18, 111-130.
Chilton, R. & J. Jarvis. (1999). “Victims and Offenders in Two Crime Statistics
Programs: A Comparison of the National Incident-Based Reporting System
(NIBRS) and the National Crime Victimization Survey (NCVS).” Journal of
Quantitative Criminology, 15, 193-205.
Cocca, C. E. (2004). Jailbait: The Politics of Statutory Rape Laws in the United States.
Albany, NY: State University of New York Press.
Cochran, J. K., & M. B. Chamlin. (2000). “Deterrence and Brutalization: the Dual
Effects of Executions.” Justice Quarterly, 17(4), 685-704.
Cochran, J. K., M. B. Chamlin, & M. Seth. (1994). “Deterrence or Brutalization? An
Impact Assessment of Oklahoma‟s Return to Capital Punishment.” Criminology,
32, 107-134.
Cohen, M. & E. L. Jeglic. (2007). “Sex Offender Legislation in the United States: What
do we Know?” International Journal of Offender Therapy and Comparative
Criminology, 51(4), 369-383.
Cohen, J. & G. Tita. (1999). “Diffusion in Homicide: Exploring a General Method for
Detecting Spatial Diffusion Processes.” Journal of Quantitative Criminology,
15(4), 451-493.
264 Cotton, M. (2000). “Back with a Vengeance: The Resilience of Retribution as an
Articulated Purpose of Criminal Punishment.” American Criminal Law Review,
37, 1313-1363.
Doren, D. M. (2006). “What Do We Know About the Effect of Aging on Recidivism
Risk for Sexual Offenders?” Sexual Abuse: A Journal of Research and
Treatment, 18(2), 137-157.
Duwe, G. & W. Donnay. (2008). “The Impact of Megan‟s Law on Sex Offender
Recidivism: The Minnesota Experience.” Criminology, 46(2), 411-446.
Elbogen, E. B., M. Patry & M. Scalora. (2003). “The Impact of Community Notification
Laws on Sex Offender Treatment Attitudes.” International Journal of Law and
Psychology, 26, 207-219.
Emery, M. & D. Lategan. (2009). “Annual statistical report: Pennsylvania department of
corrections.” Accessed online at: http://www.cor.state.pa.us/portal/server.pt/
community/research___statistics/10669.
Finkelhor, D. & L. Jones. (2006). “Why Have Child Maltreatment and Child
Victimization Declined?” Journal of Social Issues, 62(4), 685-716.
Fishback, P. V. & S. E. Kantor. (1998). “The Adoption of Workers‟ Compensation in
the United States, 1900-1930.” The Journal of Law and Economics, XLI, 305-
341.
Fortney, T., J. Levenson, Y. Brannon, & J. N. Baker (2007). “Myths and Facts about
Sexual Offenders: Implications for Treatment and Public Policy.” Sexual
Offender Treatment, 2 (1), 1-17.
265 Freeman, N. J. (2007). “Predictors of Rearrest for Rapists and Child Molesters on
Probation.” Criminal Justice and Behavior, 34(6), 752-768.
Gall, T. L. (2007). “Worldmark Encyclopedia of the States, 7th Edition.” Detroit, MI:
Gale.
Giustina, J. D. (2009). “GPS Monitoring of Sex Offenders.” In R. G. Wright (Ed.), Sex
Offender Laws: Failed Policies, New Directions, (pp. 243-266). New York:
Springer Publishing Company.
Gookin, K. (2007). “Comparison of State Laws Authorizing Involuntary Commitment of
Sexually Violent Predators: 2006 Update, Revised.” Olympia, WA: Washington
State Institute for Public Policy, Document No. 07-08-1101.
Grattet, R., V. Jenness, & T. R. Curry. (1998). “The Homogenization and
Differentiation of Hate Crime Law in the United States, 1978 to 1995: Innovation
and Diffusion in the Criminalization of Bigotry.” American Sociological Review,
63, 286-307.
Gray, M. K., M. Fields, & S. R. Maxwell. (2001). “Examining Parole Violations: Who,
What, and When.” Crime and Delinquency, 47(4), 537-557.
Halaby, C. N. (2004). “Panel Models in Sociological Research: Theory into Practice.”
Annual Review of Sociology, 30, 507-544.
Hanson, R. K. (2002). “Recidivism and Age: Follow-Up Data From 4,673 Sexual
Offenders.” Journal of Interpersonal Violence, 17(10), 1046-1062.
Hanson, K. R. & M. T. Bussiere. (1998). “Predicting Relapse: A Meta-Analysis of
Sexual Offender Recidivism Studies.” Journal of Consulting and Clinical
Psychology, 66(2), 348-362.
266 Hanson, R. K., H. Scott, & R. A. Steffy. (1995). “A Comparison of Child Molesters and
Nonsexual Criminals: Risk Predictors and Long-Term Recidivism.” Journal of
Research in Crime and Delinquency, 32, 325-337.
Hardin, J. (2011). “Fixed-, Between-, and Random-Effects and xtreg.” Stata Data
Analysis and Statistical Software. Accessed online:
www.stata.com/support/faqs/state/xtreg.html.
Hershey, M. R. (2005) Party Politics in America. 11th Edition. Pearson Longman: New
York.
Holbrook, T. M. & E. Van Dunk. (1993). “Electoral Competition in the American
States.” American Political Science Review, 87, 955-962.
Holmes, T. J. (1998). The State Border Dataset. Accessed online:
www.econ.umn.edu/~holmes/data/BorderData.html.
Horney, J. & C. Spohn. (1991). “Rape law reform and instrumental change in six urban
jurisdictions.” Law and Society Review, 25(1), 117-153.
Jacobs, D. & J. T. Carmichael. (2002). “The Political Sociology of the Death Penalty: A
Pooled Time-Series Analysis.” American Sociological Review, 67, 109-131.
Janus, E. S. (2000). “Sexual Predator Commitment Laws: Lessons for the Law and the
Behavioral Sciences.” Behavioral Sciences and the Law, 18, 5-21.
Kane, M. D. (2003). “Social Movement Policy Success: Decriminalizing State Sodomy
Laws, 1969-1998.” Mobilization, 8(3), 313-334.
Kent, S. L. & D. Jacobs. (2005). “Minority Threat and Police Strength from 1980 to
2000: A Fixed-Effects Analysis of Nonlinear and Interactive Effects in Large U.S.
Cities.” Criminology, 43(3), 731-759.
267 Klarner, C. (2007). “Measure of Partisan Balance of State Government.” Accessed
online: http://academic.udayton.edu/sppq-TPR/klarner_datapage.html.
Kohm, L. M. & M. E. Lawrence. (1997). “Sex at Six: The Vitimization of Innocence
and Other Concerns Over Children‟s „Rights.‟” Brandeis Journal of Family Law,
36, 361-406.
Koon-Magnin, S., D. A. Kreager, & R. B. Ruback. 2010. “Partner Age Differences,
Educational Contexts and Adolescent Female Sexual Activity.” Perspectives on
Sexual and Reproductive Health, 42(3), 206-213.
Knoke, D. & M. Hout. (1974). “Social and Demographic Factors in American Political
Party Affiliations, 1952-72.” American Sociological Review, 39(5), 700-713.
Kruttschnitt, C., C. Uggen, & K. Shelton. (2000). “Predictors of Desistance Among Sex
Offenders: the Interaction of Formal and Informal Social Controls.” Justice
Quarterly, 17(1), 61-87.
La Fond, J. Q. (1998). “The Costs of Enacting a Sexual Predator Law.” Psychology,
Public Policy and Law, 4(1/2), 468-504.
La Fond, J. Q. (2000). “The Future of Involuntary Civil Commitment in the U.S. after
Kansas v. Hendricks.” Behavioral Sciences and the Law, 18, 153-167.
Levenson, J. (2004). “Sexual Predator Civil Commitment: A Comparison of Selected
and Released Offenders.” International Journal of Offender Therapy and
Comparative Criminology, 48(6), 638-648.
Levenson, J. S. & L. P. Cotter. (2005). “The Effect of Megan‟s Law on Sex Offender
Reintegration.” Journal of Contemporary Criminal Justice, 21(1), 49-66.
268 Levenson, J. D. A. D‟Amora, & A. L. Hern. (2007a). “Megan‟s Law and its Impact on
Community Re-entry for Sex Offenders.” Behavioral Sciences and the Law, 25,
587-602.
Levenson, J. S., Y. N. Brannon, T. Fortney, & J. Baker. (2007b). “Public Perceptions
about Sex Offenders and Community Protection Policies.” Analyses of Social
Issues and Public Policy. 7(1), 137-161.
Lindquist, S. A. (2007). “State Politics and the Judiciary Codebook.” Accessed online:
http://academic.udayton.edu/SPPQ-TPR/DATASETS/state_politics_and_the_
judiciary_codebook.pdf.
Liptak, A. (2010). “Extended Civil Commitment of Sex Offenders is Upheld.” New
York Times, May 17, 2010.
Liska A. E. (1987). “A Critical Examination of Macro Perspectives on Crime Control.”
American Sociological Review, 13, 67-88.
Lucken, K. & J. Latina. (2002). “Sex Offender Civil Commitment Laws: Medicalizing
Deviant Sexual Behavior.” Barry Law Review, 15, 1-19.
Lucken, K. & W. Bales. (2008). “Florida‟s Sexually Violent Predator Program: An
Examination of Risk and Civil Commitment Eligibility.” Crime and
Delinquency, 54(1), 95-127.
McNally, R. (2006). “The Road Atlas & Complete State-by-State Travel Guide.”
Chicago, IL: Rand McNally.
Mears, D. P. (2010). American Criminal Justice Policy: An Evaluation Approach to
Increasing Accountability and Effectiveness. New York: Cambridge University
Press.
269 Meloy, M. L. (2005). “The Sex Offender Next Door: An Analysis of Recidivism, Risk
Factors, and Deterrence of Sex Offenders on Probation.” Criminal Justice Policy
Review, 16, 211-236.
Meloy, M. & S. Coleman. (2009). “GPS Monitoring of Sex Offenders.” In R. G.
Wright (Ed.), Sex Offender Laws: Failed Policies, New Directions, (pp. 243-266).
New York: Springer Publishing Company.
Mooney, C. Z. (2001). “Modeling Regional Effects on State Policy Diffusion.” Political
Research Quarterly, 54(1), 103-124.
Montopoli, B. (2010). “Obama Criticizes „Misguided‟ Arizona Immigration Bill.”
Accessed online: http://www.cbsnews.com/8301-503544_162-20003274-
503544.html.
Nee, V. (1998). “Sources of the New Institutionalism.” The New Institutionalism in
Sociology (pp. 1-16). Ed. M. C. Brinton & V. Nee. New York: Russell Sage
Foundation.
Nee, V. and P. Ingram (1998). “Embeddedness and Beyond: Institutions, Exchange, and
Social Structure.” The New Institutionalism in Sociology (pp. 19-45). Ed. M. C.
Brinton & V. Nee. New York: Russell Sage Foundation.
Newmark, A. J. (2002). “An Integrated Approach to Policy Transfer and Diffusion.”
The Review of Policy Research, 19(2), 152-178.
Nicholson-Crotty, S. (2004). “The Impact of Sentencing Guidelines of State-Level
Sanctions: An Analysis Over Time.” Crime & Delinquency, 50(3), 395-411.
270 Petrosino, A. J. & C. Petrosino. (1999). “The Public Safety Potential of Megan‟s Law in
Massachusetts: An Assessment from a Sample of Criminal Sexual Psychopaths.”
Crime and Delinquency, 45(1), 140-158.
Pratt, T.C. & Maahs, J. (1999). Are Private Prisons more Cost-Effective than Public
Prisons? A Meta-Analysis of Evaluation Research Studies. Crime and
Delinquency, 45, 358-371.
Prentky, R. A., & A. F. S. Lee (2007). “Effects of Age-at-Release on Long Term Sexual
Re-Offense Rates in Civilly Committed Sexual Offenders.” Sex Abuse, 19, 43-
59.
Quinn, J. F., C. J. Forsyth, & C. Mullen-Quinn. (2004). “Societal Reaction to Sex
Offenders: A Review of the Origins and Results of the Myths Surrounding their
Crimes and Treatment Amenability.” Deviant Behaviors, 25, 215-232.
Robertiello, G. & K. J. Terry. (2007). “Can we Profile Sex Offenders? A Review of Sex
offender Typologies.” Aggression & Violent Behavior, 12, 508-518.
Rogers, C. R. & B. F. Skinner. (1956). “Some Issues Concerning the Control of Human
Behavior.” Science, 124(3231), 1057-1066.
Sample, L. L. (2011). “The Need to Debate the Fate of Sex Offender Community
Notification Laws.” Criminology & Public Policy, 10(2), 265-274.
Sample, L. L. & T. M. Bray. (2006). “Are Sex Offenders Different? An Examination of
Rearrest Patterns.” Criminal Justice Policy Review, 17(1), 83-102.
Sample, L. L. & M. K. Evans. (2009). “Sex offender Registration and Community
Notification.” In R. G. Wright (Ed.), Sex Offender Laws: Failed Policies, New
Directions, (pp. 211-242). New York: Springer Publishing Company.
271 Sample, L. L. & c. Kadleck. (2008). “Sex Offender Laws: Legislators‟ Accounts of the
Need for Policy.” Criminal Justice Policy Review, 19(1), 40-62.
Sampson, R. J. & J. H. Laub. (2005). “A Life-Course View of the Development of
Crime.” The ANNALS of the American Academy of Political and Social
Science, 602, 12-45.
Scott, C. & E. del Busto. (2009). “Chemical and Surgical Castration.” In R. G. Wright
(Ed.), Sex Offender Laws: Failed Policies, New Directions, (pp. 291-338). New
York: Springer Publishing Company.
Singer, B. (1970). “Psychological Studies of Punishment.” California Law Review, 58,
405-442.
Singer, J. D. & J. B. Willett. (2003). Applied Longitudinal Data Analysis: Modeling
Change and Event Occurrence. New York: Oxford University Press.
Skinner, B. F. (1951). "How to Teach Animals.” Scientific American, 185(6), 26-29.
Soule, S. A. & J. Earl. (2001). “The Enactment of State-Level Hate Crime law in the
United States: Intrastate and Interstate Factors.” Social Perspectives, 44(3), 281-
205.
Soule, S. A. & Y. Zylan. (1997). “Runaway Train? The Diffusion of State-Level
Reform in ADC/AFDC Eligibility Requirements, 1950-1967.” American Journal
of Sociology, 103(3), 733-762.
Squire, P. (2007). “Measuring State Legislative Professionalism: The Squire Index
Revisited.” State Politics and Policy Quarterly, 7, 211-227.
272 Struckman-Johnson, C., D. Struckman-Johnson, L. Rucker, K. Bumby, & S. Donaldson.
(1996). “Sexual Coercion Reported by Men and Women in Prison.” The Journal
of Sex Research, 33(1), 67-76.
Sutherland, E. H. (1950). “The Diffusion of Sexual Psychopath Laws.” American
Journal of Sociology, 56, 142-148.
Terry, K. (2011). “What is Smart Sex Offender Policy?” Criminology & Public Policy,
10(2), 275-282.
Terry, K. J. & A. R. Ackerman. (2009). “A Brief History of Major Sex Offender Laws.”
In R. G. Wright (Ed.), Sex Offender Laws: Failed Policies, New Directions, (pp.
65-98). New York: Springer Publishing Company.
Tewksbury, R. & M. Lees. (2007). “Perceptions of Punishment: How Registered Sex
Offenders View Registries.” Crime & Delinquency, 53(3), 380-407.
Tobler W. (1970). "A computer movie simulating urban growth in the Detroit region".
Economic Geography, 46(2): 234-240.
Tonry, M. (1996). Sentencing Matters: Studies in Crime and Public Policy. New York:
Oxford University Press.
Tonry, M. (2009). “Foreword.” In R. G. Wright (Ed.), Sex Offender Laws: Failed
Policies, New Directions, (pp. ix-xi). New York: Springer Publishing Company.
Turner, Elizabeth & Stephen Rubin (2002). “Once a Sex Offender…Always a Sex
Offender: Myth or Fact?” Journal of Police and Criminal Psychology, 17(2), 32-
43.
273 Volden, C. (2006). “States as Policy Laboratories: Emulating Success in the Children‟s
Health Insurance Program.” American Journal of Political Science, 50(2), 294-
312.
Walker, J. L. (1969). “The Diffusion of Innovations Among The American States.” The
American Political Science Review, 63, 880-899.
Walker, J. T., S. Maddan, B. E. Vasquez, A. C. Van Houten, & G. Ervin-McLarty.
(2004). “The Influence of Sex Offender Registration and Notification Laws in the
United States.” Arkansas Information Center. Accessed online: www.acic.org.
Welchans, S. (2005). “Megan‟s Law: Evaluations of Sexual offender Registries.”
Criminal Justice Policy Review, 16(2), 123-140.
Williams, F. M. (2009). “The Problem of Sexual Assault.” In R. G. Wright (Ed.), Sex
Offender Laws: Failed Policies, New Directions, (pp. 17-64). New York:
Springer Publishing Company.
Wright, B. R. E., A. Caspi, T. E. Moffitt, & R. Paternoster. (2004). “Does the Perceived
Risk of Punishment Deter Criminally Prone Individuals? Rational Choice, Self-
Control, and Crime.” Journal of Research in Crime and Delinquency, 41(2), 180-
213.
Wright, R. G. (2009). “Introduction: The Failure of Sex Offender Policies.” In R. G.
Wright (Ed.), Sex Offender Laws: Failed Policies, New Directions, (pp. 1-16).
New York: Springer Publishing Company.
Zilney, L. J. & L. A. Zilney. (2009). Perverts and Predators: The Making of Sexual
Offending Laws. New York: Rowman and Littlefield Publishers, Inc.
274 Zylan, Y. & S. Soule. (2000). “Ending Welfare as We Know It (Again): Welfare State
Retrenchment, 1989-1995.” Social Forces, 79(2), 623-652.
Appendix A:
Addressing Missing Data
VARIABLE 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Demographic Internal Determinants Population * * * * * * * * * * * * * * * * * * * * * * Area * ------Population Density C C C C C C C C C C C C C C C C C C C C C C Per Capita Income * * * * * * * * * * * * * * * * * * * * * * Poverty * * * * * * * * * * * * * * * I I * * * * I Region * ------Racial Threat/Conflict Internal Determinants Hispanic I I * I * I * I * I I * * * * I * * * * * * Black I I * I * I * I * I I * * * * I * * * * * * Unemployment Rate * * * * * * * * * * * * * * * * * * * * * * Social Disorganization Internal Determinants Divorce Rate P P P P * P P P * * P * * * * * I I * * * I Females in the Labor * * * * I * * * * * * * I * * * * * * * I * Force Political Internal Determinants Party of Governor * * * * * * * * * * * * * * * * * * * * * * 1 Percent Democrat * * * * * * * * * * * * * * * * * * I I I I Percent Republican1 * * * * * * * * * * * * * * * * * * I I I I Females in the * * * * * * * * * * * * * * * * * * * * * * Legislature
276
VARIABLE 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Professionalism * ------Competition2 * ------Innovation * ------Structural-Functional Internal Determinants Rape Rate * * * * * * * * * * * * * * * * * * * * * * Murder Rate * * * * * * * * * * * * * * * * * * * * * * Incarceration Rate * * * * * * * * * * * * * * * * * * * * I I Death Penalty * ------Traditional Diffusion Neighboring A A A A A A A A A A A A A A A A A A A A A A Regional A A A A A A A A A A A A A A A A A A A A A A National A A A A A A A A A A A A A A A A A A A A A A Physical Diffusion Length of Border * ------Shared Border * ------Neighbors A ------Interstates A ------Media Pressure Media Coverage * * * * * * * * * * * * * * * * * * * * * * Federal Government Pressure Jacob Wetterling Act A A A A A A A A A A A A A A A A A A A A A A Megan‟s Law A A A A A A A A A A A A A A A A A A A A A A Kansas v. Hendricks A A A A A A A A A A A A A A A A A A A A A A Adam Walsh Act A A A A A A A A A A A A A A A A A A A A A A Social Movement Organizations Pressure
277
VARIABLE 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 “Association for the Treatment of Sexual A A A A A A A A A A A A A A A A A A A A A A Abusers” “Stop it Now!” A A A A A A A A A A A A A A A A A A A A A A * indicates that the data was recorded directly from its original source --- Indicates that the variable is not time-varying C indicates that the variable was calculated by using other variables P indicates that the data was characterized as “preliminary” in the original source I indicates that the data were interpolated based on the average of the closest years of data available A indicates that the author created this variable 1 The percentage Democrat and percentage Republican in a state were not collected for Alaska and Hawaii 2 The measure of political competition was not calculated for Louisiana
Appendix B
Newspapers and Wires searched to quantify the amount of Media Coverage
The “US Newspapers and Wires” search within LexisNexis Academic comprises only domestic newspapers and wire services. These sources include the following:
Advertising Age Architecture
Advertising Age Creativity Arizona Capitol Times
AdvisorOne Arkansas Democrat-Gazette
ADWEEK AScribe Newswire
Agent Sales Journal Associated Press Financial Wire
Aircraft Value News Associated Press Online
Airline Business Report ATM and Debit News
Airport Security Report Automotive News
Air Safety Week Automotive News Europe
Alameda Times-Star (Alameda, CA) Automotive News German Auto Industry American Agent & Broker & AABBreakingNews Newsletter
American Artist AutoWeek
American Banker Aviation Maintenance
American Health Line Avionics
American Time BackStage
Amusement Business (VNU) Bangor Daily News (Maine)
Ann Arbor News (Michigan) Bank Systems & Technology
Architectural Lighting Bay City Times (Michigan)
279 Bedford Journal California Builder and Engineer
Belleville News-Democrat Campaign News and Associated Press
Benefits Selling Candidate Biographies
BestWire Capital Times (Madison, WI)
BI Industry Focus Cardline
Billboard Cashflow news
Billboard Radio Monitor CED
Bill Overviews & Outlooks Centre Daily Times (State College, PA)
Birmingham News Chapel Hill Herald
Bluff Country Reader (Spring Valley, Charleston Daily Mail Minnesota) Chattanooga Times Free Press Boomer Market Advisor Chemweek Daily Newswire Brandweek Chicago Daily Herald Brattleboro Reformer (Vermont) Chicago Sun-Times Broadcasting and Cable Chlor-Alkali Marketwire Broker Christian Newswire B to B CityBusiness North Shore Report (New Building Design & Construction Orleans, LA)
Business Insurance City News Service
Business Travel News City Pages (Minneapolis/St. Paul, MN)
Business Wire Claims Magazine
C4I News Clearing Quarterly Directory
CableFax Clovis Livestock Market News
CableFAX's CableWORLD Collections & Credit Risk
Cahners Publications Collegiate Presswire
280 Colorado Springs Business Journal Control Engineering Europe (Colorado Springs, CO) Convenience Store News Columbus Ledger-Enquirer Converting Magazine Commercial Property News Couture Jeweler Committee Markups – Historical Cox News Service Communications Technology CQ Congressional Press Releases Communications Today CQ Federal Department and Agency Commuter/Regional Airline News Documents
ComplyNet (Sourcemedia) Crain's Chicago Business
Computer Reseller News Crain's Cleveland Business
Computerworld Crain's Detroit Business
Conexion Crain's New York Business
CongressNow Creators Syndicate
Connecticut Post Online Credit Union Journal
Construction Credit Union Times & CUTBreakingNews Construction Bulletin Criticas Construction Digest CRNtech Constructioneer CT's Pipeline Construction Equipment CT's Voice Report Construction News CT Reports Consulting-Specifying Engineer Custom Builder Contemporary Women's Issues Unique Daily Journal of Commerce (Portland, Contra Costa Times OR)
Contract Daily News (New York)
Control Engineering Daily Variety
281 Dallas Observer (Texas) Electronic Chemicals News
Dayton Daily News Electronic Engineering Times
DCD Business Report Electronic Gaming Business
De Baca County News Electronic News
Defense Daily Embedded Systems Design
Defense Daily International Embroidery Monogram Business
Deming Headlight (New Mexico) Employee Benefit Advisor
Denver Westword (Colorado) Employee Benefit News Canada
Deseret Morning News (Salt Lake City) Employee Benefit News Canada InBrief
Digital Archives Enterprise Record (Chico, California)
Discover America's Story Environment and Energy Daily
Display & Design Ideas Ethnic News
Dixie Contractor Eureka Times-Standard (California)
Dodge County Independent-News European Rubber Journal
Dolan Publications EWorldWire
Drug Discovery and Development Facts on File World News Digest
E&E News PM Fairbanks Daily News-Miner (Alaska)
East Bay Express (California) Feather River Bulletin (Quincy, California) Eastern Express Times (Pennsylvania) Fiber Optics Forecast ECN Finance & Commerce (Minneapolis, Editor & Publisher Magazine MN)
Edmonds Beacon (Washington) Flint Journal (Michigan)
Education Week Florida Shipper
Electronic Business Florida Underwriters
282 FNS DAYBOOK Hollis Brookline Journal
Foodservice Equipment and Supplies Holmes County Herald
Fort Worth Star-Telegram Home Equity Wire
Futures & FuturesBreakingNews Hospitality Design
Game Developer Hotels
Gannett News service Houston Press (Texas)
General Accounting Office Reports Huntsville Times (Alabama)
Genomics and Proteomics Idaho Falls Post Register
GIANTS Illinois Legal Times
Global Round Up - ADRs and Impressions Depository Receipts Incentive GlobeNewswire Industrial Distribution Gloucester County Times (New Jersey) Industrial Maintenance and Plant Gourmet Retailer Operation
Government Procurement Report Information Bank Abstracts
Government Publications & Documents InformationWeek
Grand Rapids Press (Michigan) InfoWorld
Graphic Arts Monthly Inland Valley Daily Bulletin (Ontario, CA) Greenwire Inside Bay Area (California) Gulf Shipper InsideCounsel Haxtun-Fleming Herald (Colorado) Inside Digital TV HD/Studio Insurance & Technology Health Data Management Insurance Networking & Data Helicopter News Management
Herald News (Passaic County, NJ) Insurance Networking News
283 Intelligencer Journal /Lancaster New Era LA Weekly (Pennsylvania) Lawyers USA Intelligencer Journal/New Era (Lancaster, Pennsylvania) Learning Quarterly
Interior Design Legal News Line
InternetWeek Lewiston Morning Tribune
Investment News Lexington Herald Leader
Investor's Business Daily Library Journal
Investrend Library Journal's Reviews
IRA Bank Wire Life Insurance Services Selling
ISO&Agent Lincoln Journal Star (Nebraska)
Ivanhoe Times (Minnesota) Logistics Management
Jackson Citizen Patriot Long Beach Press-Telegram (Long Beach, CA) JCK-Jewelers Circular Keystone Long Island Business News (Long Jersey Journal (New Jersey) Island, NY)
Journal of Commerce Los Angeles Business Journal
Journal Record Legislative Report Lowell Sun (Lowell, MA) (Oklahoma City, OK) Mabel/Harmony News-Record Kalamazoo Gazette (Michigan) (Minnesota)
Kansas City Daily Record (Kansas City, MacReport/eTeligis MO) Madison County Record Kirkus Reviews Managed Services Insider Kitchen & Bath Business Manufacturing Business Technology Land Letter Marin Independent Journal (Marin, CA) Las Vegas Review-Journal Marketing Y Medios Latah Eagle Marketwire
284 Maryland Gazette Missouri Lawyers Media
Massachusetts Lawyers Weekly Mobile Register (Alabama)
Mattawa Area News (Washington) Modern Healthcare
McClatchy-Tribune Business News Modern Materials Handling
McClatchy-Tribune News Service Modern Physician
McKenzie River Reflections (McKenzie Modesto Bee Bridge, Oregon) Monterey County Herald (CA) Media Business Mortgage Servicing News Media Industry Newsletter Mortgage Technology MediaWeek Mukilteo Beacon (Washington) Mermigas on Media Multichannel News Merrimack Journal (New Hampshire) Multi-Housing News Metropolitan News Enterprise Muskegon Chronicle (Michigan) Miami Herald National Jeweler Miami New Times (Florida) National Mortgage News Michigan Contractor and Builder National Underwriter Life & Michigan Lawyers Weekly Health/Financial Services
Midnight Trader Network World
Midwest Contractor New England Construction
MiMegasite Newhouse News Service
Min's Advertising Report New Jersey Lawyer
MIN's B2B New Orleans CityBusiness (New Orleans, LA) Mississippi Business Journal (Jackson, MS) News Bites US Markets
Mississippi Press New Times, Inc. publications
285 New Times Broward-Palm Beach PR News (Florida) PR Newswire New York Observer Professional Builder New York Sun Professional Remodeler North Carolina Lawyers Weekly Property and Casualty 360 OC Weekly PUBID;SDPNWS Omaha World Herald Public Opinion (Chambersburg, Origination News Pennsylvania)
Oroville Mercury Register (California) Publisher's Weekly Reviews
Pacific Builder and Engineer Publishers Weekly
Pacific Shipper Purchasing Magazine
Pasadena Star-News (Pasadena, CA) R & D
Patriot News (Harrisburg, Pennsylvania) Raleigh Extra
Payments Source RCR Wireless News
Pensions and Investments RDS Business & Industry Selected Documents Phoenix New Times (Arizona) RDS Business and Management Photo District News Practices - Selected Documents
Pittsburgh Post-Gazette Religion News Service
Pittsburgh Tribune Review Republican-Leader (Preston, Minnesota)
Plant Engineering Research & ResearchBreakingNews
Plastics News (tm) ResourceInvestor
Portland Press Herald Resource News International
Potentials Restaurants and Institutions
Prepaid Trends Retirement Income Reporter
Presentations Reuters Health Medical News
286 Rhode Island Lawyers Weekly Securities Industries News
Richmond Times Dispatch Semiconductor International
Riverfront Times (St. Louis, Missouri) Send2Press
Rocky Mountain Construction Senior Market Advisor
Roll Call Sentinel & Enterprise (Fitchburg, Massachusetts) Rotor & Wing SF Weekly (California) Rubber & Plastics News Shipping Digest Ruidoso News (New Mexico) Sioux County Index Reporter (Hull, Sacramento Bee Iowa)
Saginaw News (Michigan) SmarTrend
Sales & Marketing Management SourceMedia Pubs
San Antonio Express-News South Bend Tribune
San Bernardino Sun (San Bernardino, South Carolina Lawyers Weekly CA) Southeast Texas Record San Gabriel Valley Tribune (San Gabriel Valley, CA) Space & Missile Defense Report
San Jose Mercury News (California) Sporting Goods Business
San Mateo County Times (San Mateo, Sporting Goods Dealer CA) Springfield Republican (Massachusetts) Sarasota Herald-Tribune Spring Grove Herald (Minnesota) Satellite News Spring Valley Tribune (Minnesota) Satellite Today SqueezeOC School Library Journal St. Charles County Business Record (St. School Library Journal's Reviews Charles, MO)
Seattle Post-Intelligencer St. Louis Daily Record/St. Louis Countian (St. Louis, MO) Seattle Weekly
287 St. Louis Post-Dispatch The Argus (Fremont, CA)
St. Paul Pioneer Press (Minnesota) The Associated Press
St. Petersburg Times The Associated Press Election Results (Historical) Star-News (Wilmington, NC) The Associated Press State & Local Star Tribune (Minneapolis MN) Wire
Stateline.org The Atlanta Journal and Constitution
Staten Island Advance (New York) The Augusta Chronicle
States News Service The Austin American-Statesman
Stockdiagnostics The Banner (Bernice, Louisiana)
Sunday News (Lancaster) The Bellingham Herald (Washington)
Supply Chain Management Review The Berkshire Eagle (Pittsfield, Massachusetts) Targeted News Service The Bismarck Tribune TechNews The Bond Buyer Technology Decisions The Bonney Lake & Lake Tapps TechWeb Courier-Herald Telegram & Gazette (Massachusetts) TheBookStandard.com Telegraph Herald (Dubuque, IA) The Boomerang! (Palouse, Washington) Television Week The Boston Herald Terror Response Technology Report The Bradenton Herald Texas Contractor The Buffalo News The Advocate (Baton Rouge, Louisiana) The Bulletin's Frontrunner The Alamogordo Daily News (New Mexico) The Business Press / California
The Albuquerque Journal The Cabinet (Milford, New Hampshire)
The Anchorage Daily News The Capital (Annapolis, MD)
288 The Charleston Gazette The Fresno Bee
The Charlotte Observer The Herald (Rock Hill, S.C.)
The Chatfield News (Minnesota) The Herald-Sun
The Christian Science Monitor The Hill
The Chronicle of Higher Education The Hollywood Reporter
The Chronicle of Philanthropy The Hotline
The Clatskanie Chief (Oregon) The Houston Chronicle
The Columbian (Vancouver, WA) The Idaho Business Review (Boise, ID)
The Columbia Star (South Carolina) The Idaho Statesman (Boise)
The Daily Deal The Indiana Lawyer
The Daily News of Los Angeles The Indianapolis Business Journal
The Daily Record (Baltimore, MD) The Island Packet
The Daily Record of Rochester The Journal of Jefferson Parish (Rochester, NY) (Louisiana)
The Daily Reporter (Milwaukee, WI) The Journal Record (Oklahoma City, OK) The Daily Review (Hayward, CA) The Kansas City Star The Dallas Morning News The Lebanon Daily News The Delhi Express (California) (Pennsylvania)
The Denver Post The Ledger (Lakeland)
The Enumclaw Courier Herald The Legal Ledger (St. Paul MN) (Washington) The Macon Telegraph The Evening Sun (Hanover, PA) The Main Wire The Florida Times Union The Mecklenburg Times (Charlotte, NC) The Forward The Merced Sun-Star (California) The Free Press (Buda, Texas) The Metropolitan Corporate Counsel
289 The Minnesota Lawyer (Minneapolis The Plain Dealer MN) The Post and Courier (Charleston, SC) The Myrtle Beach Sun-News The Post-Standard (Syracuse, NY) The News and Observer The Press Enterprise The News Tribune (Tacoma, Washington) The Progressive Grocer
The New York Post The Providence Journal-Bulletin
The New York Times The Raton Range (New Mexico)
The New York Times – Biographical The Record (Bergen County, NJ) Materials The Roanoke Times (Virginia) The New York Times - Government Biographical Materials The Salt Lake Tribune
The Oakland Tribune (Oakland, CA) The San Francisco Chronicle
The Oklahoman The San Luis Obispo Tribune
The Olathe News (Johnson County, The Santa Fe New Mexican Kansas) The Spokesman-Review The Olympian (Olympia, Washington) The Sporting News The Orange County Register The Sports Network The Oregonian The Star-Ledger (Newark, New Jersey) The Original Irregular (Kingfield, Maine) The State (Columbia, S.C.)
The Palm Beach Post The State Journal-Register (Springfield, IL) The Pantagraph The Sun Herald (Biloxi, MS) The Patriot Ledger The Tampa Tribune The Philadelphia Daily News (PA) The Taos News The Philadelphia Inquirer The Times of Trenton (New Jersey) The Pitch (Kansas City) The Times-Picayune
290 The Times Union (Albany, NY) TWICE
The Trinidad Times Independent University Wire (Colorado) UPI (United Press International) The Tulsa World UPI Energy The Union Leader USA Today The Village Voice US Fed News The Virginian-Pilot (Norfolk, VA) US States News The Washington Daybook Vallejo Times-Herald (California) The Washington Post VARBusiness The Washington Post Biographical Stories Variety
The Washington Times Very Light Jet Report
The White House Bulletin Via Satellite
The Wichita Eagle Video Business
The Wyoming Tribune-Eagle Virginia Lawyers Weekly
The York Dispatch (York, PA) VNU Entertainment News Wire
TickerTech/MarketNewsVideo Wall Street & Technology
Tire Business Wall Street Horizon
Today's Sunbeam (New Jersey) Wall Street Journal Abstracts
Topeka Capital-Journal Waste News
Training Wealth Manager &
Treasury & Risk & TRBreakingNews WealthManagerBreakingNews
Tribune-Review Westbrook Sentinel Tribune (Minnesota)
Tri-City Herald Western Builder
Tri-Valley Herald (Pleasanton, CA) West Virginia Record
291 Whittier Daily News (California) Workforce Management
Winston-Salem Journal WorldSources, Inc, Emerging Markets Datafile Wireless Week World Stock Wire Wisconsin Law Journal (Milwaukee, WI) Yearbook of Experts News
Wisconsin State Journal
Sarah Koon-Magnin
EDUCATION Ph.D. Crime, Law, and Justice, The Pennsylvania State University, August 2011 M.A. Crime, Law, and Justice, The Pennsylvania State University, August 2008 B.A. Sociology, B.A. Religious Studies, University of Missouri, May 2006 Summa Cum Laude, Phi Beta Kappa
PUBLICATIONS Koon-Magnin, S., D. A. Kreager, & R. B. Ruback. (2010). “Partner age differences, educational contexts and adolescent female sexual activity.” Perspectives on Sexual and Reproductive Health, 42(3), 206-213. Ruback, R. B., R. Collins, S. Koon-Magnin, W. Ge, L. Bonkiewicz, & M. Park. (2011). “People transitioning across places: A multimethod investigation of how people go to football games.” Environment and Behavior, 43(5), 1-28.
Manuscripts Under Review Koon-Magnin, Sarah and R. Barry Ruback. “Young Adults' Perceptions of Statutory Rape: The Effects of Victim Gender.” Revise & Resubmit. Sex Roles. Koon-Magnin, S. & R. B. Ruback. “The Perceived Legitimacy of Statutory Rape Laws: The Effects of Victim Age, Perpetrator Age, and Age-Span.” Revise & Resubmit. The Journal of Applied Social Psychology.
GRANTS AND AWARDS 1st Place, Crime, Law, and Justice Graduate Student Paper Competition, 2010. Harold F. Martin Graduate Assistant Outstanding Teaching Award, 2008. Graduate Scholars Award, College of the Liberal Arts, Penn State University, 2006.
PRESENTATIONS Koon-Magnin, S. & D. A. Kreager. “Older Romantic Partners and Female Adolescent Risk Behavior: A Counterfactual Approach.” Annual Meeting, American Society of Criminology, San Francisco, CA, November 2010. Koon-Magnin, S., R. B. Ruback, & O. Unger. “Differences in Sex Offenses and Sex Offender Legislation Across Time and Place.” Annual Meeting, American Society of Criminology, Philadelphia, PA, November 2009. Koon-Magnin, S. & D. A. Kreager. “Are Seniors who Date Freshmen Sexual Predators? Re-Assessing the Link between Partner Age-Span and Girls‟ Reproductive Health.” Annual Meeting, American Sociological Association, San Francisco, CA, August 2009. Koon-Magnin, S. & R. B. Ruback. “Adolescent Sexual Activity and Statutory Rape: A Multi-method Investigation.” Annual Meeting, American Society of Criminology, St. Louis, MO, November 2008.