State of West Virginia Department of Military Affairs & Public Safety Criminal Justice Statistical Analysis Center Division of Criminal Justice Services

July 2006 Report Highlights • The methodology introduced in this report offers a valid approach for Establishing the “Statistical Accuracy” assessing the statistical accuracy of UCR crime statistics in WV and the of Uniform Crime Reports (UCR) in nation. West Virginia • Of the 31,084 reported crimes in the population, 1,297 were estimated to contain a classification error. James Nolan, Ph.D. - West Virginia University Stephen M. Haas, Ph.D. - Criminal Justice Statistical Analysis Center • An estimated 4.17% of all records Theresa K. Lester, M.A. - Criminal Justice Statistical Analysis Center reported for the population of 12 law enforcement agencies in this study Jeri Kirby, M.A. - West Virginia University were misclassified. Carly Jira, B.A. - West Virginia University • The UCR Part I crimes of aggravated assault, burglary, larceny, The Uniform Crime Reporting studies have sought to ascertain the and robbery contained a statistically significant amount of classification (UCR) Program is a national initiative magnitude of error resulting from the error. involving more than 17,000 city, county, erroneous classification of crimes. and state law enforcement agencies That is, few studies have assessed the • A significant level of error was found for nonindex crimes such as which voluntarily report crime data to amount of error found in crime totals simple assault/intimidation, unfounded the FBI (FBI, 2004).1 The main objective due to the misclassification of crime offenses, and general incidents. of the UCR Program is to generate a types on the part of law enforcement • A great deal of overlap in valid set of crime statistics for use in officers and agencies. Classification classification error was found between law enforcement administration, error (or the misclassification of crime some individual crimes such as operation, and management. types) and the impact of this error on aggravated-simple assaults and Over the years, however, the official UCR statistics provides the basis larceny-burglary offenses. information gathered and reported for this report. • Both the Violent Crime and Index through the UCR Program has become It is anticipated that an totals for the state were significantly a social indicator for the nation. The understanding of classification error and undercounted in reported UCR statistics. public looks to the statistics generated its consequences for crime reporting will from the UCR Program for information have notable implications for both state • The Violent Crime Total for WV on fluctuations in the level of crime. and federal UCR Program was undercounted by 22.49% compared to the Index Total at 2.35%. Meanwhile, criminologists, sociologists, administrators. The identification of legislators, municipal planners, the classification error and its sources may • The Property Crime Total for the media, and other students of criminal provide a basis for the modification of state did not contain a significant justice use the statistics for varied law enforcement training curricula. In amount of classification error at 0.18%. research and planning purposes (FBI, addition, such information may lead to 2004). more accurate crime reporting as well • The differentiation between Such widespread use of UCR as provide a means for adjusting future aggravated and simple assault crimes accounted for a disproportionate information has underscored the crime data based on known error. amount of classification error in importance of ensuring its accuracy. This report seeks to assess the reported UCR statistics in WV. While considerable attention has been statistical accuracy of crime reporting • The inclusion of simple assault into focused on errors associated with victim in West Virginia. In this pursuit, Violent Crime and Index Totals reduced reporting and missing data (Hart and however, a central purpose of this report classification error for these categories Rennison, 2003; Maltz, 1999), few is to introduce a methodology for to nonsignificant levels. assessing the “statistical accuracy” of strict accounting of all crimes, it remains occur when state UCR programs and crime estimates produced by the UCR a tremendously useful resource for the FBI make decisions about how to Program. Currently, no uniform method gaining knowledge about crime. deal with problems of missing data at exists for identifying and assessing the As a result, the value of UCR is not the time of publication and reporting impact of classification error in WV or contingent on FBI and state UCR (Maltz, 1999). Multiple methods exist the nation. Thus, this report applies an program officials eliminating all errors for dealing with problems associated original methodology designed to in reported statistics. Instead, it is more with missing data. For instance, some examine classification error which may important that the managers of the jurisdictions may choose to not report in turn influence how such assessments program understand these errors and missing data while others may produce are conducted in other states and the make every attempt to measure them. estimates for missing data and adjust nation. By doing so, program managers will crime statistics based on those In an effort to illustrate the merit become more cognizant of the limitations estimates. Missing information and the of this methodology, this report utilizes in UCR and can begin to engage in handling of such data is a common a random sample of crimes reported to efforts to improve the accuracy of crime source for error found in the reporting the UCR Program in WV. An analysis reporting. of statistical data and is not unique to was conducted to identify the under- and Toward this end, considerable UCR statistics. overreporting of incidents across various attention in recent years has focused on However, there have been no true crime types. Both the sources of errors that result from victims deciding systematic studies of classification classification error and the most not to report crimes and by the police errors in the UCR statistics. As a result, common reasons for misclassifications electing not to record them (Bachman, little is known about the degree to which are discussed. To begin, this report 1993; Black, 1974; delFrate and classification error influences the provides a brief overview of Goryainov, 1994; Gove, Hughes, and statistical accuracy of crime statistics. classification error and its impact on the Geerken, 1985; Greenberg and Ruback, Moreover, research has not progressed accuracy of official crime statistics. 1987; Greenberg and Ruback, 1992; to the point of establishing a single Schwind and Zenger, 1992; Shah and methodology for assessing classification Assessing the Accuracy of Pease, 1992; Skogan, 1976; Warner and error in UCR data. For these reasons, Crime Statistics Pierce, 1993; Wexler and Marx, 1986; the current study may not only assist Since the inception of the UCR White and Mosher, 1986). For example, UCR Program administrators in WV, but Program more than 70 years ago, in a study based on the 2000 National may have implications for how statistical police records have been the primary Crime Victimization Survey (NCVS), accuracy is examined on a national source of data in these national crime Hart and Rennison (2003) found that basis. To provide a foundation for this statistics. Although the quality of police nearly 4 in 10 property crimes and 6 in study, the following discussion provides records has often come into question, 10 violent crimes were never reported a brief description of classification error UCR has been found to be a valid to police. Hence, it is clear that not all and why it is often present in crime indicator of the “index” crimes (Gove, crimes are reported to law enforcement statistics. Hughes, & Geerken, 1985). and not all crimes are accounted for in It is well known, however, in the official reporting. This type of error Classification Error: fields of criminology and criminal justice occurs at the input stage of the process Definition and Sources that UCR is a “statistical program,” and is due to nonreporting of crimes by A classification error occurs when meaning that it is not an actual victims and law enforcement. the police officers record the facts of accounting of all crimes. Nevertheless, On the other hand, error can also an incident correctly, but misclassify the UCR has been invaluable to police and occur at the latter stages of the process. crime type. For example, an criminologists alike in their efforts to Recently, there has been an investment “aggravated assault” that involves a understand the nature and extent of on the part of the U.S. Department of weapon is sometimes recorded by the crime locally and nationally. Therefore, Justice to learn more about errors at the police as a “simple assault” when the in spite of the fact that UCR is not a output stage (Maltz 1999). These errors victim is not seriously injured. Such a

2 “Statistical Accuracy” of UCR Crime Statistics crime classification may be correct for it actually had (McCoy, Matza, & agency revealed that 3.00% of all larceny criminal prosecution, but not for Fazlollah, 1998). records contained a record error. Yet, “statistical” purposes. Given that this Lastly, errors can occur at the only a fraction of this 3.00% actually incident involved a weapon, it should be point of automation or when resulted in a reclassification of UCR recorded as an aggravated assault. automated systems are upgraded or crime. This is because many of the record Classification errors can occur for revised. Automated systems are errors were discovered to be the result of multiple reasons. Some of the most often programmed to allow for the slight discrepancies in the dollar amounts common sources of error include the automatic translation of reported recorded for theft loss (i.e., incorrectly inaccurate interpretation of UCR crimes to UCR definitions. In these recorded as $400.00 as opposed to the definitions, purposive actions on the part instances, reported crimes are actual amount of $450.00). Since a of law enforcement agencies to automatically translated from state mistakenly reported dollar amount for theft downgrade particular crimes, and faulty code to the UCR. These loss does not change the fact that a theft automation of police records. computerized systems can contain was committed, it would not result in First, error can occur when criminal programming or algorithm problems changes to UCR estimates for larceny- definitions rather than UCR definitions that may result in the routine theft. Instead, this is record error that are used by law enforcement officers misclassification of reported offenses does not affect UCR statistics. to classify crimes. In some instances, into erroneous UCR definitions or As it pertains to the accuracy of UCR UCR definitions are slightly different crime categories. statistics, record accuracy is of limited use. from state and local criminal laws and Regardless of the source, It only reflects the rate of error in a ordinances. In the absence of regular classification error can have a particular crime type. In the review of training to help police recognize these substantial impact on the statistical simple assault records, for instance, the conflicting definitions and purposes (i.e., accuracy and interpretation of UCR only classification error that will be for “statistical gathering” versus crime estimates. To better understand discovered are offenses that should have “criminal investigations”), police may how classification error can impact the been something else, but were instead sometimes begin to apply criminal accuracy of reported crime, the reported as simple assaults. As such, it definitions when classifying crimes for following discussion offers a captures only the type of errors that statistical purposes. description of two basic measures for inflate the crime in a particular crime Second, classification error has accuracy in reported crimes. category. In other words, it is simply the been found to be a result of conscious rate at which a crime is overreported. decision-making on the part of law Record Versus Statistical Statistical accuracy, on the other enforcement agencies. As a matter of Accuracy hand, refers to the errors found in the policy or practice, for instance, law Two types of accuracy must be crime totals after all crime types have enforcement officers may be considered when assessing the been examined and offsetting encouraged to record some crimes as presence or absence of classification misclassifications have been considered. less serious offenses in order to keep error in UCR statistics: record Since some of the misclassifications result the crime rate down. This source of accuracy and statistical accuracy. in overcounting UCR crimes while others error was discovered recently in the city Record accuracy refers to the errors result in undercounting them, the correct of Philadelphia. For years, the found in a particular record or group UCR number can be obtained by Philadelphia Police Department had of records in a given crime type. This considering the canceling effect of the two been downgrading certain major crimes type of accuracy may be defined types of errors—overcounting and to exclude them from official crime differently by police agencies for undercounting. statistics. This long-standing practice different purposes. In addition, record To illustrate the concept of statistical of “going down with crime,” as referred errors may or may not result in the accuracy, consider the hypothetical case to by police officers, resulted in that city misclassification of crime. of a police department that wants to reporting a much lower crime rate than Consider the following: A review check the accuracy of all its reports for a of larceny records for a particular given year (N = 50,000 records). Due to

“Statistical Accuracy” of UCR Crime Statistics 3 0 15 366 125 142 137 673 8,428 1,056 6,651 6,408 4,480 2,522 31,003

Total 285,898 1 1 4 1 7 0 0 67 12 90 32 620 138 267 10,822

Agency 12 0 0 25 17 49 16 22 34 494 194 817 688 982 3,338 32,980

Agency 11 2 4 7 0 0 5 0 0 0 18 96 32 447 283 17,193 8 9 Agency 10 1 0 73 39 20 13 886 335 228 222 718 2,552 25,713

Agency 9 6 2 0 4 6 26 82 21 12 210 277 467 247 1,360 11,410

Agency 8 4 3 7 1 0 29 46 20 365 403 222 103 381 1,584 19,028

Agency 7 0 0 9 0 0 5 0 0 0 95 12 27 32 10 13,342

Agency 6 7 0 Table 1 Table 1 Table 1 Table 1 Table 1 27 39 47 461 163 306 840 1,786 1,352 2,650 1,306 8,984 53,230

Agency 5 5 6 5 0 0 18 10 33 64 181 162 420 386 1,290 15,633 Agency 4 7 4 2 3 0 0 13 17 82 278 331 280 262 1,279 20,338 Agency 3 4 0 30 91 43 14 311 106 913 714 790 1,942 1,581 6,539 51,291 Agency 2 6 2 0 0 The Twelve Largest Municipal Police Agencies in West Virginia The Twelve Largest Municipal Police Agencies in West Virginia The Twelve Largest Municipal Police Agencies in West Virginia The Twelve Largest Municipal Police Agencies in West Virginia The Twelve Largest Municipal Police Agencies in West Virginia 55 76 31 17 339 173 822 426 968 2,915 14,918 Agency 1 The total number of incidents (31,003) does not include unfounded or general incident reports.

Note: Note: Note: Resident Population Resident Population Total Incidents Total Incidents Resident Population Total Incidents Arson Aggravated Assault Simple Assault/Intimidation Burglary Murder Other Homicide Larceny Motor Vehicle Thefts Robbery Rape Other Sex Offenses Other Group A Group B Resident Population Resident Population Total Incidents Total Incidents

4 “Statistical Accuracy” of UCR Crime Statistics time and resource constraints the burglary. The officers also found 5 is multiplied by the total number of department can review no more than burglary reports (5.00%) misclassified, reports in each category to determine 1,500 records. The officers assigned which should have been recorded as the estimate. to this project decide that the best way larceny. No other errors were found in Once the calculation is complete, the to proceed is to conduct a study that the entire review of 1,500 records. estimated number of burglaries for this will allow them to make inferences about At first glance it appears that the 5 department is 725 and the estimated the error in the population of reports misclassifications from each crime number of larcenies is 4,775. The based on the review of a random sample category (burglary and larceny) would calculations were made as follows: of only 1,500 reports. cancel each other out and the reported The first step taken is to separate number of crimes would be correct. Burglary all police reports for that year into 15 However, determining statistical 725 = 500 (original number) - 25 categories. Some of these categories accuracy for the entire year’s records (overcounts: burglaries that should have are single UCR categories, such as requires a calculation that estimates the been larceny) + 250 (undercounts: robbery, rape, and burglary. Other impact of the error found in the sample larcenies that should have been categories include several types of to the population of records for the year. burglaries) crimes, such as all sex offenses other To perform the calculation, the total than rape. number of reports in each category Larceny Once all of the 50,000 records are where the errors were found must be 4,775 = 5,000 (original number) - 250 partitioned into the 15 categories, the determined. In this hypothetical case, (overcounts: larcenies that should have officers select a random sample of 100 there were 5,000 larcenies and 500 been burglaries) + 25 (undercounts: reports from each category. Upon burglaries reported for the year. burglaries that should have been reviewing the records officers found 5 Therefore, the 5.00% error found in larcenies). larceny reports (5.00%) misclassified, each category (5 out of 100 larceny which should have been recorded as reports and 5 out of 100 burglary reports)

Table 2 Sample Agencies and Reporting Categories

Crime Agency 5 Agency 7 Agency 9 Stratum Total N n N n N n N n Arson 47 47 4 4 8 8 137 59 Aggravated Assault 306 166 29 26 73 18 673 210 Simple Assault/Intimidation 1,306 118 222 60 228 127 4,480 305 Burglary 840 226 103 43 222 175 2,522 444 Murder 7 7 1 1 1 1 15 9 Other Homicide 0 0 0 0 0 0 0 0 Larceny 2,650 375 381 61 718 242 8,428 678 Motor Vehicle Thefts 461 94 46 38 39 32 1,056 164 Robbery 163 100 20 18 20 17 366 135 Rape 27 16 3 3 13 13 125 32 Other Sex Offenses 39 36 7 7 9 7 142 50 Other Group A 1,786 168 365 69 886 62 6,651 299 Unfounded 31 31 unkn 0 unkn 0 31 31 General Incidents 48 48 2 2 unkn 0 50 50 Group B 1,352 71 403 53 335 99 6,408 223 Total 9,063 1,503 1,586 385 2,552 801 31,084 2,689

Note: The total under “N” in the “Stratum Total” column (31,084) includes both unfounded and general incident reports. These reports were not included in Table 1.

“Statistical Accuracy” of UCR Crime Statistics 5 To illustrate how statistical Table 3 accuracy was assessed using a sample Sample Sizes Based on Error Estimates of records in WV, the following section provides a detailed description of the Total Estimated methodology applied in the present Reports Error Sample study. Crime N % n Arson 137 N/A 59 Methodology Aggravated Assault 673 8.00% 210 Simple Assault/Intimidation 4,480 8.00% 305 The methods used to assess Burglary 2,522 15.00% 444 classification error and statistical Murder 15 N/A 9 accuracy in the WV UCR statistics Other Homicide 0 N/A 0 involved three distinct stages: 1) pre- Larceny 8,428 22.00% 678 sampling, 2) selecting and reviewing Motor Vehicle Thefts 1,056 5.00% 164 sampled records, and 3) analyzing the Robbery 366 5.00% 135 results. Rape 125 2.00% 32 Other Sex Offenses 142 2.00% 50 Other Group A 6,651 8.00% 299 Pre-Sampling Unfounded 31 N/A 31 The pre–sampling stage involved General Incidents 50 N/A 50 two steps: 1) the partitioning of records Group B 6,408 5.00% 223 and 2) the calculation of the appropriate Total 31,084 2,689 sample size. Note: There are 12 municipal police agencies with populations over 10,000 that Partitioning of Records. The 12 comprise the stratum represented in this table. largest municipal police departments in WV comprised the population for this participating agencies. The “n” column confidence and error. First, given the study. Table 1 describes the resident provides the sample sizes by crime size of the population, a desired level of populations and number of crimes category for each of the 3 agencies. confidence for the interval estimate had reported within each of these 12 The “N” under the “Stratum Total” to be established. A 95.00% confidence agencies in 2002 (i.e., the largest column represents the number of reports level was chosen and is reflected in the municipal police departments in WV). in each crime category, for all 12 z score 1.96. Second, the acceptable The last column labeled “Total” refers municipal police agencies, representing level of error was established as .03, to the total number of residents and the the total population for this study (31,084 meaning plus or minus three percent. total number of reported incidents, incidents). The “n” under the “Stratum Lastly, a proportion of error was excluding the unfounded and general Total” column is the total number from estimated based on what was expected incidents among all 12 agencies (31,003 the 3 participating agencies (2,689 to exist in each of the 15 crime incidents). incidents). This was the total desired categories. These estimates were From the group of 12 police sample size for this study based on the based on the results of previous record departments, 3 were randomly selected estimated error in each offense audits. to participate. The records in each of category. Once the above information was these 3 agencies were partitioned into established, the sample size was 15 categories. Calculating Sample Size calculated according to equation 1: Table 2 describes these 3 police In order to establish point estimates k 2 NPQ departments and the 15 categories, of statistical accuracy, a sample of (1.) n = k 2 PQ + NE 2 , where including unfounded and general records from each of the 15 reporting incident reports. The column “N” under categories was drawn. Prior to drawing k = confidence level (1.96 each agency describes the total number the sample, however, it was necessary represents 95.00% confidence) of crime reports within each of these 3 to decide on an appropriate level of

6 “Statistical Accuracy” of UCR Crime Statistics P = estimated proportion of the established to ensure a high level of categories included in this report are population reliability between reviewers. For each described below. record, a system of verifying the inter- Q = (1-P) rater agreement among reviewers was Arson. To unlawfully and intentionally constructed by having multiple damage, or attempt to damage, any real E = acceptable error (set at .03 reviewers examine those records with or personal property by fire or incendiary or 3.00%) a suspected classification error. device. When a classification error was The estimated sample size for each suspected in a given record, at least 3 Aggravated Assault. An unlawful reporting category and participating members of the research team attack by one person upon another police agency is provided in Table 2. In (including the first author) reviewed the wherein the offender uses a weapon Table 2, “n” represents the sample size report to either confirm the classification or displays it in a threatening manner, and “N” corresponds to the total error or reaffirm the accuracy of the or the victim suffers obvious severe or number of reports. As previously noted, original classification.3 If it was difficult aggravated bodily injury involving the desired sample size for the entire to determine whether the report was apparent broken bones, loss of teeth, study was set at 2,689 records. classified correctly, the record was possible internal injury, severe Table 3 identifies the desired judged to not contain a classification laceration, or loss of consciousness. sample size for each crime category error. Essentially, this procedure gave based on the estimate of error expected the reporting officer the benefit of any Simple Assault/Intimidation. Simple to be found. The sample “n” was doubt in the absence of clear assault is the unlawful physical attack calculated according to equation 1. information. Thus, the research team by one person upon another where relied on the judgment of the officer on neither the offender displays a Selection and Review of the scene when crime classification was weapon, nor the victim suffers Sampled Records difficult to determine because the obvious severe or aggravated bodily An automated random sample narrative was vague.4 injury. Severe or aggravated bodily generator was used to select the records injury includes apparent broken bones, for the present study. All incident Key Terms and Definitions loss of teeth, possible internal injury, numbers from each report category The UCR definitions for all fifteen severe laceration, or loss of were entered into the program. A list crime reporting categories included in consciousness. of selected cases based on incident this report are provided below. Group Intimidation is to unlawfully place numbers was generated. With the A offenses include crimes of arson, another person in reasonable fear of assistance of agency representatives, aggravated assault, simple assault/ bodily harm through the use of hardcopies of all reports identified for intimidation, burglary, murder, other threatening words and/or other the sample were manually pulled and homicide, larceny, motor vehicle theft, conduct, but without displaying a assessed for accuracy. robbery, and rape. These same weapon or subjecting the victim to an In reviewing each record, offenses, excluding simple assault/ actual physical attack. definitions provided by the UCR intimidation, are also referred to as Part program officials at the FBI were I crimes under the UCR Program. Part Burglary. The unlawful entry into a applied. Prior to reviewing the records, I crimes are also referred to as Index building or other structure with the intent however, a one day training was offenses in this report. to commit a felony or a theft. provided by an official FBI trainer which The UCR Program also collects focused on the proper classification of arrest data on 19 other offenses. In Murder. The willful, nonnegligent killing crimes in the UCR. The FBI provided National Incident-Based Reporting of one human being by another. materials to assist in this process. System (NIBRS) these offenses are A systematic procedure for the referred to as Group B crimes.5 The Other Homicide. Includes negligent assessment of each record was UCR definitions for each of the offense manslaughter, which is the killing of

“Statistical Accuracy” of UCR Crime Statistics 7 another person through negligence. Unfounded. Crimes that were Undercounts. When reports that should Justifiable homicide, defined as the reported to the police but were have been in category X are found in killing of a perpetrator of a serious subsequently determined by the police another category such as Y. The reports criminal offense by a peace officer in to be false or baseless. result in an undercount of category X. the line of duty; or the killing, during the commission of a serious criminal General Incidents. These are reports Statistical Definition. The UCR offense, of the perpetrator by a private filed by the police for noncriminal definition for each crime. individual, is also included. matters, such as suspicious person investigations, false burglary alarms, Criminal Definition. The criminal Larceny. The unlawful taking, carrying, community problems/disputes, and so definition of crime found in state code. leading, or riding away of property from forth. the possession or constructive These offense categories or crimes possession of another person. Group B. These offenses include bad and statistical terms are referred to checks, curfew/loitering/vagrancy throughout the remainder of this report. Motor Vehicle Theft. The theft of a violations, disorderly conduct, driving The following section presents the results motor vehicle. under the influence, drunkenness, of the current study. nonviolent family offenses, liquor law Robbery. The taking, or attempting to violations, peeping tom, runaway, Results take, anything of value under trespassing, and all other offenses not The results of this study focus on the confrontational circumstances from the considered Group A offenses. These statistical accuracy of crimes reported control, custody, or care of another are crimes that are only reported upon in selected municipal police departments person by force or threat of force or an official arrest. in WV. The analysis begins with a violence and/or by putting the victim presentation of results based on an in fear of immediate harm. There are also several general as assessment of agency records. The well as statistical terms used frequently findings center on the degree to which Rape. The carnal knowledge of a in this report. A list of these terms and overcounts and undercounts were found person, forcibly and/or against that their definitions are presented below. in the classification of crimes. Emphasis person’s will; or not forcibly or against is placed on the nature of classification the person’s will where the victim is Confidence Intervals. The interval of across crime types. incapable of giving consent because of values surrounding the point estimate These results are followed by an his/her temporary or permanent mental in which researchers can be confident assessment of classification error found or physical incapacity or because of his/ that the true population parameter (e.g., between individual crimes and the her youth. the number of crimes) falls. associated impact on aggregate crime totals. Based on a review of police Other Sex Offenses. Includes forcible Point Estimate. A statistic provided records, this discussion is followed by a sodomy, sexual assault with an object, without indicating a range of error. The qualitative description of why many of forcible fondling, incest, and statutory best guess of the true number of crimes the errors in classification occurred. This rape. in each crime category in the population discussion highlights the primary reasons under study. for the presence of classification errors Other Group A. Includes the offenses in UCR statistics. This report concludes of bribery, counterfeiting, vandalism, Overcounts. When reports in crime with a review of the major findings and drug crimes, gambling, extortion, fraud, category X are examined, overcounts potential implications for UCR program kidnapping, prostitution, and weapons represent reports that should have administrators and officers as well as offenses. actually been in another category such for future research. The analysis begins as category Y. These reports are with a review of the results presented in deemed overcounts of category X. Table 4.

8 “Statistical Accuracy” of UCR Crime Statistics

Overcounts Overcounts Overcounts Overcounts Overcounts 4 9 8 0 0 0 0 7 2 6 3 2 2 5 13 42 26 47 26 44 12 12 35 10 15 15 395 324 220 146 N/A N/A 133 1,297 9 0 0 59 15 32 50 31 31 50 50 Total Total Total Total Total 137 210 673 294 431 677 165 136 366 125 142 300 219 N/A N/A Reports Reports Reports Reports Reports 4,480 2,522 8,428 1,056 6,651 6,408 2,663

31,084 Sample Size Sample Size Sample Size Sample Size Sample Size Group B Group 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 2 2 5 30 22 54 214

219

6,262 6,316 Incidents

3 7 1 3 2 0 0 0 0 0 0 2 0 0 0 0 0 0 1 3 2 1 1 0 0

General 30 25 44 47 35 35 12

148 113 Unfounded 0 0 1 3 0 0 2 0 0 0 0 2 3 1 3 0 0 3 9 0 0 0 0 0 0

12 25 19 41 29 29 12 71 represent the targeted sample sizes, while

100 Group A Group

1 2 3 0 0 2 0 0 0 0 5 0 0 0 0 0 0 0 0 1 1 6 6 0 0 Other 10 12 62 18 93 308 290

6,524 6,431

Offenses Other Sex Other 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 2 0 0 0 0 0 0 5

12 44 43 56 38

163 107 Rape 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 8

23 37 23 29

136 113 Robbery 0 0 3 4 0 0 0 0 0 0 2 1 6 0 0 0 0 0 0 0 0 0 0 0 0

10 61 25 10

144 462 102 134 360 MV Theft MV 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 29 29 159 158

1,041 1,012 Table 4 Table 4 Larceny Table 4 Table 4 Table 4 0 0 0 0 0 0 4 0 0 0 0 3 1 3 0 0 0 0 2 0 0 1 1 1 23 19 44 29 12 663 119 651

8,223 8,104

Homicide

Other 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Murder 0 0 0 0 0 0 0 0 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9 0 0

15 15 Matrix of Overcounts and Undercounts Matrix of Overcounts and Undercounts Matrix of Overcounts and Undercounts Matrix of Overcounts and Undercounts Matrix of Overcounts and Undercounts Burglary 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 1 1 0 0 15 44 18 187 441 232 423

2,707 2,475 Intimidation Simple Assault Simple 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 3 16 88 10 278 106 268

4,191 4,085 Assault

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 3 0 0 Aggravated 18 21

274 218 908 277 197 631 Arson 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 22 56 22 55 150 128 The actual sample sizes in this table differ slightly from the Table 3. This is because numbers 3

Note: Note: Note: Table 4 includes the number of reports in each crime category that were actually obtained and reviewed. Undercounts Undercounts Undercounts Arson Aggravated Assault Simple Assault/Intim. Burglary Murder Other Homicide Larceny MV Thefts Robbery Rape Other Sex Offenses Other Group A Unfounded General Incidents Group B Total Actual Undercounts Undercounts

“Statistical Accuracy” of UCR Crime Statistics 9 Assessing Overcounts and (considering overcounts and Similar to arson, the crime of rape Undercounts in Crime undercounts). Based on this sample contained few classification errors. It Classification study, it was estimated that 1,297 is estimated that a total of 35 records in The results shown in Table 4 are classification errors were contained in the population contained an error. Of displayed in a matrix of overcounts and the population of 31,084 records. Thus, these 35 instances, all of the undercounts. The rows reflect the approximately 4.17% of all records misclassifications occurred between the initial classification of the reports (that reported in the population of 12 agencies offense of rape and other sex offenses. is, by the law enforcement officer), and were estimated to be misclassified. A total of 12 crimes were originally the columns show the classifications Beyond the total number of classified as rape and were based on the reviewers assessment classification errors found in agency subsequently deemed to be other sex using UCR crime reporting definitions. records, an examination of individual offenses. Meanwhile, 23 crimes initially Two numbers are presented in the crime types suggested a great deal of recorded as other sex offenses were matrix cell. The top number is the variation in the number of misclassified assessed by the reviewers as rape cases. sample and the bottom number is offenses. As a result, the This resulted in a net reduction of 11 population estimate. For example, misclassification of crimes was more offenses that were actually rape, but under arson there are two numbers, 55 pronounced for some offenses, classified otherwise. and 128. The 55 represents the number compared to others. The crime In addition, a close examination of of sample reports out of 59 that were categories that had the highest error Table 4 reveals a general pattern for initially reported as arson and were estimates include: simple assault/ much of the classification error that confirmed to be arson. The number intimidation (501), larceny (443), occurred between individual crime 128 is the estimate of the number of aggravated assault (319), other Group types. As noted previously, a greater arsons in the population of reports that A (313), burglary (279), Group B (200), amount of error was found among were recorded accurately by the police. and general incident (128). (The certain crimes. The crimes with the Following along the top row of numbers in parentheses represent the most error included aggravated assault, Table 4, notice that there is a 1 under combined total of over- and simple assault/intimidation, larceny, other “other Group A” offenses. This undercounts.) Group A, and burglary. However, an indicates that 1 incident was originally On the other hand, far less or no assessment of the classification error reported as arson, but was assessed by classification error was found in other across these crime types confirmed that the reviewers to be an other Group A crime types. For instance, the crime of much of the error found in larceny offense rather than an arson. Below murder contained no misclassified tended to overlap with error associated the 1 in this cell is a 2. This indicates records. Of the records sampled, all with burglary and vice versa. Similarly, that in the population (i.e., all 12 were assessed as murders by the a great deal of the error associated with municipal agencies combined) it is reviewers. aggravated assaults tends to overlap estimated that 2 arsons should have The crimes of motor vehicle theft, with the error related to simple assault/ actually been classified as other Group arson, and rape contained fewer intimidation. Though the impact was A offenses. classification errors. In the case of arson much less in terms of total number of Finally, the last column provides the for instance, only 31 records were records, this type of pattern was also overcounts in the sample and the estimated to contain an error. A total of present in rape and other sex offense population estimates of overcounts by 9 overcounts and 22 undercounts were cases discussed above. offense category. While the bottom estimated for arson. As a result, Consider the case of larceny. A row denotes the undercounts in the misclassifications associated with the majority of undercounts and overcounts sample and the population estimates of offense of arson did not contribute a occurred in relation to the offense of undercounts for each offense. great deal to the overall level of burglary (see Table 4). While the crime As shown in Table 4, there were classification error in UCR statistics in of larceny contained an estimated 443 133 out of 2,663 records in the sample WV. classification errors, (324 overcounts that contained classification errors and 119 undercounts) 210 of the errors

10 “Statistical Accuracy” of UCR Crime Statistics were estimated to be related to the as simple assaults. After the net gains Crime Estimates and offense of burglary (187 overcounts and and loses were considered, a total of Classification Error Across 23 undercounts). 258 misclassifications occurred due the Crime Types A similar pattern emerges when the difficulty in differentiating between Tables 5 and 6 provide a summary magnitude of classification error these two crimes. The end result was of point estimates, confidence intervals, between the crimes of simple and a considerable downgrading of and statistical error rates for each aggravated assault was examined. aggravated assaults to the less serious reported crime category. Both tables Again, a majority of misclassifications offense of simple assault. also offer an estimate of the total in both crimes occurred when trying to These results underscore the nature number of Index, violent, and property distinguish between reporting an and variability in classification error crimes. incident as a simple versus an across crime classifications. While The “Reported” columns in Tables aggravated assault. An estimated 274 there was considerable variability in the 5 and 6 show the number of reports filed aggravated assault offenses were magnitude of error across crime types, in each category. The “Estimate” misclassified by law enforcement a majority of misclassifications tended columns provide the point estimates of officers as simple assaults/intimidation. to occur in a predictable fashion. The crimes based on the review of the Far fewer aggravated assaults were following section illustrates the impact sampled records. The “Ratio” columns misclassified as simple assaults. Only of the classification error found among compare the point estimates to the 16 offenses were initially recorded as individual crimes on the total population number of offenses originally reported. aggravated assaults and later assessed of reported offenses. For example, in the arson row in Table

Table 5 Crime Estimates, Error Rates, and Confidence Intervals

Crime Estimate Reported Ratio Error Low High Arson 150 137 0.91333 -8.67% 107 194 Aggravated Assault 908 673 0.74199 _-25.88% 788 1,028 Simple Assault/Intimidation 4,191 4,480 1.06896 6.90% 4,019 4,363 Burglary 2,707 2,522 0.93166 -6.83% 2,595 2,819 Murder 15 15 1.00000 0.00% N/A N/A Other Homicide 0 0 N/A N/A N/A N/A Larceny 8,223 8,428 1.02493 2.49% 8,077 8,369 Motor Vehicle Thefts 1,041 1,056 1.01441 1.44% 977 1,105 Robbery 462 366 0.79221 -20.78% 394 530 Rape 136 125 0.91911 -8.09% 120 152 Other Sex Offenses 163 142 0.87117 -12.88 101 227 Other Group A 6,524 6,651 1.01947 1.95% 6,379 6,665 Unfounded 100 31 0.31000 -69.00% 57 143 General Incidents 148 50 0.33784 -66.22% 68 228 Group B 6,316 6,408 1.01457 1.46% 6,178 6,454 Total 31,084 31,084 1.00000 0.00% N/A N/A

Violent Crime Total 1,521 1,179 0.77515 -22.49% 1,382 1,660 Property Crime Total 12,121 12,143 1.00182 0.18% 11,923 12,319 Index Total 13,642 13,322 0.97654 -2.35% 13,399 13,885

Note: Percentages highlighted in lighter shaded boxes denote statistically significant levels of error. The Violent Crime Total includes: aggravated assault, murder, other homicide, robbery, and rape. The Property Crime Total includes: arson, burglary, larceny, and motor vehicle theft.

“Statistical Accuracy” of UCR Crime Statistics 11 5, the 0.91333 in the “Ratio” column statistical significance. Large The Impact of Classification results from comparing the reported percentages as well as statistical Error on Aggregate Crime number to the estimate (137 divided by significance are, to some extent, Totals 150). influenced by sample size.7 The totals reported at the bottom of This statistical error percentage is To assess where the observed Table 5 and 6 provide the amount of reported in the “Error” column in both errors in each crime category were error once all offenses were collapsed Tables 5 and 6. Negative percent error statistically significant, a 95.00% into Index categories. The Violent indicates an undercount of crimes in a confidence interval was calculated for Crime Total was comprised of murder, given category. Meanwhile, positive the point estimate. If the reported aggravated assault, robbery, and rape. percent error is indicative of overcounts number did not fall within this confidence The Property Crime Total consisted of in a given crime type. The “low” column interval it was identified as statistically arson, burglary, larceny, and motor in each table reports the lower bounds significant at p < .05. vehicle theft. All of the offenses that of the confidence interval, and the The Part 1 offenses with reported make up the Violent Crime and Property “high” column represents the upper numbers outside the calculated Crime indices constituted the Index bounds. confidence interval , included aggravated Total. All of the results in Table 5 and 6 assault, burglary, larceny, and robbery. As shown in Table 5, the offenses are identical, with the exception of the As expected based on the review of that comprised the Violent Crime Total information used to calculate the Violent under- and overcounts in Table 4, the were undercounted by 22.49%. There Crime and Index totals. Table 6 includes crimes of aggravated assault and simple were 1,179 violent crimes reported by simple assault/intimidation as an Index assault contained a significant amount the 12 law enforcement agencies that crime. The results and implications for of classification error. Similarly, a comprised the population for this study. the inclusion of simple assault/intimidation significant amount of error was present Based on the study’s findings, however, as an Index crime are discussed later in for larceny and burglary. 1,521 offenses were estimated to have this report. But first, a discussion of the In the case of robbery, a total of 366 occurred based on the review of sampled findings shown in Table 5 is provided. records were reported by law records. This undercount in violent The results displayed in Table 5 enforcement during the year under crimes was statistically significant at p illustrate the amount of classification study. Based on a review of a sample < .05. error estimated for each crime type in of records, it was estimated that 462 Similar to violent crimes, the the population. As shown in the error robberies should have been reported. undercount in reported cases that column, the greatest amount of percent The estimated difference of 96 records comprised the Index Total was also error was observed for unfounded (- resulted in a statistically significant level statistically significant. A total of 13,322 69.00%), general incident (-66.22%), of classification error for robbery. Index crimes were reported by law aggravated assault (-25.88%), and In addition to Part 1 Index offenses, enforcement; however, the number was robbery (-20.78%) offenses.6 All of the the error in the number of unfounded estimated to actually be 13,642. error is negative, indicating that these and general incident cases reported by Of the Index offenses, only the crimes were undercounted in original law enforcement was significant. aggregate measure for property crime crime reports. Though fewer records involved these failed to reach statistical significance. To a lesser extent, error offenses, the greatest amount of In the case of property crimes, the percentages were reported for burglary classification error occurred in these number of records reported and (-6.83%), other sex offenses (-12.88%), crimes. A total of 31 unfounded and 50 estimated on the part of the reviewers arson (-8.67%), and rape (-8.09%). general incident records were reported was nearly equal, indicating the presence Similar to the other crimes, these reports for the population of 12 agencies in this of little classification error. A total of were undercounted in original UCR study. Following the review the actual 12,143 property crimes were reported statistics. It is important to note, number of records was estimated to be by law enforcement, compared to 12,121 however, that large percentages of error 100 unfounded and 148 general incidents records deemed to be property crimes do not automatically translate into cases.

12 “Statistical Accuracy” of UCR Crime Statistics by the researchers. As a result, there crimes may be contributing to a in this manner resulted in undercounts was only a slight overcount in the substantial amount of the error in the that were no longer considered to be number of reported property crimes. Violent Crime Totals and Index Totals. statistically significant. Based on all of the findings To explore this prospect, simple assault/ Such an examination, which presented in Table 5, it appeared that a intimidation was added to the Violent aggregates all assaults together, revealed substantial amount of the error attributed Crime and Index calculations in Table that these offenses were likely being to the Violent Crime Total and Index 6. recorded correctly as assaults. Total may have resided within the two As shown in Table 6, once all However, the high level of error was categories of assault — aggravated and reported assaults (aggravated and more likely to be associated with making simple assault. This was evidenced by simple) were included in the calculation the finer distinctions between the the fact that aggravated assault was of the Violent Crime and Index Totals, different types of assault. As a result, undercounted by 25.88%, while simple the amount of classification error was this analysis underscored the notion that assault/intimidation was overcounted by reduced considerably. The Index Total the impact of classification error on 6.90%. was now undercounted by only 0.17% aggregate crime totals may be Given the prevalence and nature of and the Violent Crime Total was significantly reduced by limiting the the error associated with the assault undercounted by less than one percent number of misclassifications between offenses, it was hypothesized that these (0.93%). The aggregation of all assaults these two crimes. To understand why

Table 6 Crime Estimates, Error Rates, and Confidence Intervals with Simple Assault Considered an Index Crime

Crime Estimate Reported Ratio Error Low High Arson 150 137 0.9133 -8.67% 107 193 Aggravated Assault 908 673 0.7412 -25.88% 788 1,028 Simple Assault/Intimidation 4,191 4,480 1.0690 6.90% 4,019 4,363 Burglary 2,707 2,522 0.9317 -6.83% 2,595 2,819 Murder 15 15 1.0000 0.00% N/A N/A Other Homicide 0 0 N/A N/A N/A N/A Larceny 8,223 8,428 1.0249 2.49% 8,077 8,369 Motor Vehicle Thefts 1,041 1,056 1.0144 1.44% 977 1,105 Robbery 462 366 0.7922 -20.78% 394 530 Rape 136 125 0.9191 -8.09% 120 152 Other Sex Offenses 163 142 0.8712 -12.88% 101 227 Other Group A 6,524 6,651 1.0195 1.95% 6,379 6,665 Unfounded 100 31 0.3100 -69.00% 57 143 General Incidents 148 50 0.3378 -66.22% 68 228 Group B 6,316 6,408 1.0146 1.46% 6,178 6,454 Total 31,084 31,084 1.0000 0.00% N/A N/A

Violent Crime Total 5,712 5,659 0.9907 -0.93% 5,491 5,933 Property Crime Total 12,121 12,143 1.0018 0.18% 11,924 12,319 Index Total 17,833 17,802 0.9983 -0.17% 17,536 18,130

Note: Percentages highlighted in lighter shaded boxes denote statistically significant levels of error. Small differences of plus or minus one may exist in some total figures due to rounding. The crimes of aggravated assault, simple assault/intimidation, murder, other homicide, robbery, and rape comprise the Violent Crime Total. The Property Crime Total is comprised of arson, burglary, larceny, and motor vehicle theft.

“Statistical Accuracy” of UCR Crime Statistics 13 such errors in classification occur, the examined. Using the crimes found to for overcounts that ensued in the following section provides a qualitative have a significant amount of reporting of 5 crimes. Each of these analysis of police records. classification error in this report, this crimes was found to contain a significant section offers a description for why amount of classification error in the Explanations for the many of the most common errors previous analysis. The 5 offenses include Misclassification of Crime occurred. Qualitative information aggravated assault, simple assault, Types gathered from the review of agency burglary, robbery, and larceny.8 The previous sections of this report records was used to illustrate the The original crime classification by identified the sources of under- and shortcomings in interpretation that law enforcement officers is noted in the overcounts in the classification of UCR occurred in the reporting crimes by law row labeled “Reported Crime.” The crimes. The magnitude of classification enforcement officers. section labeled “Amended Crime error by crime type as well as its impact To facilitate the discussion, Table 7 Classification and Explanation for on aggregate crime estimates was also provides a set of selected explanations Overcount” lists the crimes that should

Table 7 Common Classification Errors by Crime Type

Aggravated Assault Simple Assault Robbery Burglary Larceny Reported Crime Reported Crime Reported Crime Reported Crime Reported Crime

Simple Assault Aggravated Assault Larceny Larceny Burglary • Minor injuries and • Weapon involved • Grabbed money • Theft from vehicle • Breaking and no weapon involved with minor injuries from person; no force or • Theft from associate entering to commit a • Severe injuries with or threat of force reported theft Robbery without a weapon Other A • Threats with weapon Unfounded • Destruction of Robbery and items stolen Robbery • Lost property, found property • Weapon involved • Weapon or threat of after initial report • Entered building and or threat of force legally, deleted Other A force to commit theft to commit a theft • Destruction of computer files property Group B Other A • Public intoxication Unfounded • Warrant Unfounded • Disorderly conduct • Misplaced property • Forgery • Self defense • Destruction of • Destruction of property (no one property General Incident assaulted) • Not enough Unfounded information to General Incident • Credit card complete report or • Civil disputes malfunction, no confirm aggravated • Suspicious activity crime assault • Items missing, not stolen

General Incident Amended Crime Classification and Explanation for Overcount Amended Crime Classification and Explanation for Overcount Amended Crime Classification and Explanation for Overcount Amended Crime Classification and Explanation for Overcount Amended Crime Classification and Explanation for Overcount • Civil dispute

14 “Statistical Accuracy” of UCR Crime Statistics have been reported based on UCR gun and [taking] a DVD player.” On such as “took cigarettes and beat up definitions. The short descriptions the other hand, other police reports person,” “stole items from store and beat identified by an bullet offer a brief described incidents that were more up store employee,” and “shoplifted and explanation for why each of the 5 closely related to destruction of property. then attempted to strike employee with offenses were misclassified; thereby, For instance, 2 police records noted that his car.” Since these incidents involved resulting in an overcount. a “dog was shot” or “windows were shot the taking of something by force and/or The first column in Table 7 shows out” with no mention of bodily injury to by putting the victim in fear of immediate the sources for overcounts in aggravated the victims nor mention that the victims harm, they were most appropriately assaults. A total of 5 different crimes were in danger of being injured. classified as robberies. were originally reported as aggravated As shown in the second column of In other instances, however, robbery assault, but later assessed to be another Table 7, many simple assaults were in was overcounted by the inclusion of acts crime. These crimes include simple turn reported as aggravated assaults. of larceny, aggravated assault, and assault, robbery, other A, unfounded, and Of the 26 overcounts for simple assault, unfounded incidents. In one case, a general incident. more than two-thirds of these errors crime originally classified as a robbery The classification error associated were more appropriately classified as only described that the perpetrator with the reporting of aggravated assault aggravated assault. “stabbed [a] guy with a razorblade, very occurred most often in relation to simple Most of these crimes, as reported serious injury, victim life-flighted to assault. For a crime to be classified as by law enforcement officers, either hospital.” This incident, as recorded by aggravated assault, an offender must involved a weapon or described severe the police officer, contained no indication have used a weapon or displayed the bodily injuries to the victim(s). Various that this act was in pursuit of taking, or weapon in a threatening manner, or the reports described the use of a weapon attempting to take, anything of value. victim must have suffered severe or such as a beer or glass bottle or club. Therefore, this incident should have been aggravated bodily injury. In the absence Other nonweapon-use examples classified as an aggravated assault. of a weapon or such injury, a crime is involved incidents in which a victim Another major source of typically classified as a simple assault. suffered “a broken nose and slight classification error in UCR statistics As shown in Table 7, crimes of concussion” or statements noting severe reported in this study involved the simple assault were in fact reported as injury such as “lacerated the victim’s relationship between burglary and aggravated assault in some instances. finger.” Aggravated assault was the larceny. In many instances, burglary When misclassifications occurred, they correct classification for these crimes was overcounted by the misclassification were most often due to either no due to the presence of a weapon(s) and/ of larceny offenses and vice versa. mention of a weapon being involved in or severe injury to the victim(s). Burglary simply involves the unlawful the incident or the absence of severe To a lesser extent, some crimes entry into a building or other structure bodily injury. Instead of serious injury, were classified as simple assaults when with the intent to commit a felony or these reports tended to describe such they were in fact robberies. Robbery is theft. Since burglary often involves injuries as a “knot on the head,” “redness defined in the UCR as the taking, or larceny, these crimes were found to be to eyes,” “bruises,” and other “minor attempting to take, anything of value...by more prone to misclassification by abrasions.” No weapons were reported force or threat of force or violence and/ officers. as being involved in these incidents. or putting the victim in fear of immediate Most reported cases that produced Other aggravated assault harm. In most instances where a an overcount of burglary offenses overcounts were more accurately robbery was mistakenly classified as a involved the taking of items from a car. classified as robbery and other A simple assault, police records described The UCR definition for burglary pertains offenses such as destruction of property. acts in which force was used to take or solely to entry into a building or other In the case of robbery, reports completed attempt to take something of value from structure with the intent to commit a by law enforcement officers often another person. felony. According to the FBI, a building described acts of “shoplifting with a For instance, police reports of simple or other structure does not include a car knife” or breaking “into a home with a assaults sometimes included statements or personal vehicle. Instead, these

“Statistical Accuracy” of UCR Crime Statistics 15 offenses are to be reported as larceny/ these qualitative accounts of cases may significant levels of classification error theft offenses or, more specifically, theft help UCR administrators pinpoint where in WV, such as aggravated assault, from a motor vehicle. Over half of all particular difficulties in interpretation burglary, larceny, and robbery. With the overcounts for burglary involved the may reside among officers. This exception of larceny, which was misclassification of larceny offenses. information may further lead to efforts overreported by law enforcement Larceny offenses, on the other on the part of UCR administrators and officers, all Part I offenses were hand, involve the unlawful taking, police agencies to work at better significantly underestimated in official carrying, leading, or riding away of preparing officers at making more UCR statistics. Likewise, 3 nonindex property from the possession, or precise distinctions between crimes for offenses contained a significant level of constructive possession, of another UCR reporting purposes. error. These included simple assault/ person. Similar to burglary offenses, a The following section provides a intimidation, unfounded offenses, and large proportion of overcounts in larceny brief review of the findings and a general incidents. crimes were due to the misclassification discussion of plausible implications of this The classification error in Part I of burglary offenses. study for UCR administrators, law crimes was determined to have a Of the burglary offenses enforcement training curricula, as well profound impact on aggregate crime erroneously reported as larceny, most as future research. totals. When individual crimes were involved the simple breaking into a house, aggregated into Violent Crime and Index apartment, or storage garage. Officer Summary and totals, this study found a significant reports included such statements as Conclusions undercounting of these offenses. The “broke into house through window and This report sought to assess the Violent Crime and Index totals for the stole items,” “entered building after it statistical accuracy of crime reporting population were estimated to have been was closed and stole items,” and in West Virginia. The population for undercounted by 22.49% and 2.35%, “storage lock broken, entered building the study consisted of the 12 largest respectively. Meanwhile, the Property and stole $2,000 worth of items.” Since municipal police agencies in the state of Crime Total for the state was slightly these incidents involved illegal entry into WV. From the group of 12 police overcounted, however this error was not a building to commit the larceny, they departments, 3 were randomly selected found to be statistically significant. were more accurately classified as to participate in the study. A total of This study also found a substantial burglary offenses. 2,663 reports were randomly selected amount of overlap in classification error In addition to the misclassification from the 3 participating agencies. The between the individual crimes. Thus, it of burglary offenses as larcenies, other records in each of the participant appears many misclassifications tend to crimes such as robberies, other A’s, agencies were partitioned into the 15 occur in a rather predictable fashion. For unfounded offenses, and general crime categories which served as a instance, much of the error associated incidents were also counted as larceny. focal point for this study. with the crime of larceny occurred in Instead of being classified as robbery, The results underscored the relation to burglary and vice versa. A offenses that involved threats to “cut a importance of assessing the impact of similar pattern emerged between the victim’s throat” and a shop employee classification error on crime statistics. crimes of aggravated and simple assault. being “shoved, pushed, and punched” Of the 31,084 offenses reported by the These findings suggest that, during a shoplifting incident were 12 agencies that comprised the perhaps, law enforcement officers have originally classified as larceny offenses. population for this study, a total of 1,297 some difficulty in making the fine Likewise, some other A offense records were estimated to have been distinctions that are necessary for including forgery, embezzlement, and misclassified. As a result, approximately accurately classifying crimes that are destruction of property were incorrectly 4.17% of all records reported by law conceptually close in nature. Thus, while reported as larceny crimes. enforcement were estimated to contain officers were often correct in Given the inherent difficulty in a classification error. determining that an assault had occurred making the fine distinctions necessary Several Part I and other crime in this study, the error occurred when for classifying crimes, it is hoped that categories were found to contain making the decision of whether the

16 “Statistical Accuracy” of UCR Crime Statistics particular incident should be classified Based on the pertinent information of crime estimates is likely to be as an aggravated versus simple assault. generated from this inquiry, the improved only by greater knowledge of The qualitative analysis based on police methodology introduced in this report where such errors occur and their reports also seems to support this notion. seems to offer a valid approach for related impact on aggregate crime The crime of simple assault, albeit assessing the statistical accuracy of totals. not a Part I crime, is important in this UCR crime statistics in WV and, Based on the results of this study, it analysis because it was overcounted by potentially, the nation. The importance is clear that classification error is present approximately 7.00%. Many aggravated of this study resides in the fact that it in UCR statistics and that it varies by assaults were misclassified as simple provides a concrete example of how to type of crime. Future efforts may seek assault/intimidation resulting in a assess classification error at the state ways to statistically adjust crime data substantial “downgrading” of violent level, where multiple police agencies for greater accuracy based on the crimes. The impact of this contribute to the overall state crime magnitude and variation of known error misclassification on the statistical statistics. in individual crime types as well as accuracy of the Index and Violent Crime To date, research has not aggregate totals. Given the widespread totals was particularly noteworthy. progressed to the point of establishing a reliance on UCR data, it is vitally When simple assault/intimidation single methodology for assessing important to continue efforts designed was not included in the Index and Violent classification error in UCR data. Future to improve the accuracy of crime Crime totals for the population, this research should build on the methods reporting. resulted in a significant undercounting used in this report as a means for either of these crime totals. Once simple building on extant approaches and/or Notes assault was added to the calculation for generating new methods for assessing 1 There are no federal statutes that these aggregate crime totals, the classification error in UCR statistics. It require state and local law enforcement undercounts for each declined is anticipated that the methodology agencies to submit UCR reports to the considerably. In the case of the Violent applied in this report will be readily FBI. However, many states mandate Crime Total, the percentage of transported to other states that report UCR reporting from law enforcement undercounts dropped from 22.49% to UCR data. agencies. In WV, law enforcement 0.93% with the addition of simple assault. This research also provides insight agencies are required to submit NIBRS In like manner, the error associated with into ways to improve the accuracy of data to the Bureau of Uniform Crime the Index Total declined from 2.35% to the crime statistics. First, the results of Reports maintained by the WV State 0.17%. Both of these aggregate this study may assist UCR Police under state code §15-2-24 measures of crime no longer contained administrators and agency personnel in subsections (i) and (j). a significant level of classification error. training police officers to classify crimes 2 Unfounded and general incident crime In addition to examining the extent according to UCR definitions. When the categories are not included in Table 1. of classification error in WV, this report quantitative and qualitative results of this Since such records are not routinely set out to determine the validity of an study are used in conjunction, ample captured by the WV’s Incident-Based original methodology for assessing information is provided for both the Reporting System (WVIBRS), these statistical accuracy. While a great deal identification of and explanation for records could only be identified and of attention has been given to the study classification error in crime statistics. accessed during on-site visits to each of factors that may influence victim and The results of this study could be drawn law enforcement agency. Table 2 law enforcement reporting of crimes, upon to improve law enforcement captures the total population, including few studies have systematically training through the use of scenarios that unfounded and general incidents. assessed the impact of classification illustrate common errors. 3 It is important to note that the first error on the statistical accuracy of UCR Finally, this study has implications for author has had more than 20 years estimates. Thus, this report introduced the proper reporting of UCR statistics experience in classifying crimes, both a method for estimating the statistical and the aggregation of crimes for as a police officer and an FBI official. accuracy of UCR statistics in WV. analysis and research. The accuracy He recently worked as unit chief in the

“Statistical Accuracy” of UCR Crime Statistics 17 UCR Program where he was actively increase. Therefore, smaller error Black, D. (1974), Production of Crime involved in setting policies regarding percentages (or differences in the actual Rates, American Sociological Review, crime classification. In all cases where number of reported and estimated 35 (4), 733-748. errors were detected, the first author reports) may yield statistically significant assisted the research team to determine results for large samples. On the other del Frate, A. A., & Goryainov, K. (1994). whether an error had actually occurred. hand, greater differences in the number Latent crime in Russia. Rome, Italy: Due to cost and time constraints, inter- of reported and estimated records are United Nations Interregional Crime and rater reliability tests were not conducted. required to yield statistically significant Justice Research Institute. 4 An example of this is a report where results for small samples. the victim reported a burglary because This relationship between sample Federal Bureau of Investigation (2004). she found items missing from her home. size and statistical significance can be Crime in the United States, 2003. It was unclear at the time of the initial illustrated by examining the crimes of Washington, DC: Government Printing report whether the incident was actually simple assault/intimidation and other sex Office. a burglary or a theft (i.e., a family offenses in Table 5. Notice that the member was a suspect). Therefore, the percent of error for other sex offenses Gove, W., Hughes, M., & Geerken,M. investigating officer’s opinion as to the and simple assault/intimidation is 12.88% (1985), “Are Uniform Crime Reports a proper crime classification was and 6.90% respectively. However, the Valid Indicator of Index Crimes? An accepted. number of simple assault/intimidation Affirmative Answer with Minor 5 The National Incident-Based Reporting cases in the population is roughly 4,000 Qualifications,” Criminology, 23, pp. System (NIBRS) is a comprehensive compared to fewer than 200 other sex 451-501. system by which crime statistics on crimes. offenses, victims, property, and arrests Despite a greater percentage of Greenberg, M. S. & Ruback, R. B. are submitted to the UCR Program. The error associated with other sex crimes (1985). A model of crime victim decision FBI began accepting NIBRS data into compared to simple assault/intimidation, making. Victimology, 10(1-4), 600-616. the UCR from states in 1989. it is not statistically significant. Yet, the 6 The actual number of unfounded and 6.90% of error associated with simple Hart, Timothy, and Rennison, Callie general incident reports are relatively assault/intimidation did achieve (March 2003). Reporting Crime tot he small compared to the other crime statistical significance. Police. Washington, DC: Bureau of categories. Given the small numbers, it 8 Unfounded and general incident Justice Statistics. is necessary to interpret these large offenses were also found to contain a error percentages with caution. significant amount of classification error. Maltz, M. (1999). Bridging the Gaps Moreover, a large percentage of For the purposes of this exercise, in Police Crime Data: A Discussion classification error in crimes with small however, the researchers chose to Paper from the BJS Fellows Program. numbers does not necessarily translate report the explanations for the most Washington, DC: Bureau of Justice into a large contribution to overall prevalent offenses. A total of only 248 Statistics. statistical accuracy. Even though the unfounded and general incident offenses percentage of error associated with were estimated to have been reported McCoy, C. R., Matza, M., and Fazlollah, unfounded and general incident records by the population of law enforcement M. (1998, November 1). Statistical was more than other crimes, the majority agencies during the study period. manipulation by police goes back of classification error in UCR statistics decades. Philadelphia Inquirer, in WV cannot be attributed to these References Philadelphia, PA. records. Bachman, R. (1993) Predicting the 7 Large percent differences can be reporting of rape victimization: Have Schwind, H., & Zwenger, G. (1992). The readily obtained with small numbers. At rape reforms made a difference? dark number analysis of motives for the same time, statistical significance is Criminal Justice and Behavior, 20(3), nonreporting theft in three studies. more easily obtained as sample sizes 247-251.

18 “Statistical Accuracy” of UCR Crime Statistics Studies on Crime and Crime provided hands on training in the Prevention, 1(1), 115-126. reporting of NIBRS data and the interpretation of offense definitions. Shah, R., & Pease, K. (1992). Crime, Their expertise was crucial to the race and reporting to the police. Howard success of this study. Journal of Criminal Justice, 31(3), 192-199. Funding Source This project was funded by a grant Skogan, W. (1976). Crime and crime from the U.S. Bureau of Justice Statistics rates. In Wesley Skogan (ed.) Sample (Grant # 2004-BJ-CX-K004) and in part surveys of victims of crime. Cambridge, by the ASA/BJS Statistical Methodology MA: Ballinger. Research Program, a joint program of the American Statistical Association’s Wexler, C. & Marx, G. T. (1986). When Committee on Law and Justice Statistics law and order works: Boston’s and the Bureau of Justice Statistics. innovative approach to the problem of racial violence. Crime & Delinquency, DCJS Administration 32(2), 205-223. J. Norbert Federspiel, Director Jeff Estep, Deputy Director White, B. B. & Mosher, D. L. (1986). Experimental validation of a model for predicting the reporting of rape. Sexual Coercion & Assault, 1(2), 43-56

U.S. Department of Justice. (1992). 1204 Kanawha Boulevard, East Uniform Crime Reporting Handbook, Charleston, WV 25301 NIBRS Edition. Washington, DC: Phone: (304) 558-8814 Federal Bureau of Investigation. Fax: (304) 558- 0391 www.wvdcjs.com Acknowledgments This study would not have been possible without the cooperation and support of the various police agency officials. Their guidance and knowledge in the collection and interpretation of data at the local level was invaluable. To the staff of the Division of Criminal Justice Services who assisted in the manual collection of data, our thanks is also extended. A special thanks to Dr. Yoshio Akiyama, the principal architect of the methodology used in this research, and Kevin MacFarland of the FBI. Dr. Akiyama provided support over the course of this study. Mr. MacFarland

“Statistical Accuracy” of UCR Crime Statistics 19