B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H CRIME AND JUSTICE

NSW Bureau of Crime Bulletin Statistics and Research

Contemporary Issues in Crime and Justice Number 45

June 1999 The effect of arrest on Indigenous employment prospects B. Hunter1 and J. Borland2

The effects of criminal conviction on an individual’s employment prospects are a matter of policy significance, especially for Australian Aboriginals who are overrepresented both among the unemployed and among those who come before the criminal justice system. In this study the effect of arrest on the employment status of indigenous is examined using data from the 1994 National Aboriginal and Torres Strait Islander Survey. Having been arrested is found to significantly reduce the probability of indigenous employment by 18.3 and 13.1 percentage points for males and females respectively. The effect also varies according to the reason for a person’s most recent arrest. Differences in arrest rates between indigenous and non- may explain about 15 per cent of the difference in the level of employment between those groups. Promoting diversion options should be a priority for governments keen to break the nexus between indigenous unemployment and crime.

INTRODUCTION and results the reader is referred to the The potential effect of arrest on full report (Borland & Hunter 1999). employment outcomes of indigenous Criminologists have long been interested There are a number of reasons why Australians is of interest for a number of in the relationship between unemployment linkages might exist between arrest or reasons. First, it may be important for and crime but have historically spent crime and employment. Arrest may understanding differences in employment much more time examining the effect affect a person’s employment outcome. outcomes between indigenous and other of unemployment on criminal behaviour A person who has been arrested and/or Australians. Arrest rates for indigenous than the effect of a criminal conviction on convicted of an offence may be Australians are significantly greater than an individual’s employment prospects. stigmatised by employers and hence be for other Australians – for example, in This second issue is particularly salient less likely to obtain employment. 1994 indigenous Australians comprised for Australian Aboriginals who are Alternatively, employers may be deterred 2.6 per cent of the population in Western overrepresented among both the from locating in regions with high levels yet accounted for 20.2 per cent unemployed and those who have been of criminal activity and hence there may of total arrests (Broadhurst 1997, p. 426). arrested. be limited employment opportunities for Hence, if having been arrested has a Preliminary analysis of the 1994 National persons living in those regions. There negative effect on a person’s probability Aboriginal and Torres Strait Islander are other possibilities as well. Contact of employment, the disparity in arrest Survey (NATSIS) indicates that arrest is with the criminal justice system may rates may explain part of the difference in affect a person’s motivation to work, or one of the major factors underlying the employment outcomes between perceptions of the expected benefits from poor employment prospects of the indigenous and other Australians. seeking employment, and hence lower indigenous population (Australian Bureau the probability of employment. of Statistics 1996a; Hunter 1997). Second, understanding the relationship between an individual’s arrest record and Unfortunately, these early studies could It is important to bear in mind, though, not determine the direction of causality that, not only can a person's arrest history employment outcome provides an insight between arrest and employment. This affect the likelihood of being employed, a into the social costs of contact with the bulletin reports on the results of a person’s employment outcome may also criminal justice system for indigenous statistical analysis which examines how affect the likelihood of being arrested. For Australians. This seems particularly having an arrest history affects an example, a response to being unable to important where there is a possibility that Australian Aboriginal’s employment obtain employment may be to engage in much of this contact for indigenous status. The bulletin provides an overview drinking which increases the probability of Australians arises from differential of a detailed study conducted by the being arrested for offences relating to treatment of indigenous and non- authors. For further details of the analysis drunkenness. indigenous Australians rather than

1 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H differences in behaviour (see, for Table 1 shows the aggregate Table 3 shows the reasons for the most example, Gallagher & Poletti 1998). employment/population rate,3 and the recent arrest. The most common In this bulletin we use data from the employment/population rate reasons for arrest relate to drinking – NATSIS to examine the determinants of disaggregated by arrest record and by 23.1 per cent of males and 6.6 per cent employment outcomes for indigenous type of employment – in Community of females had charges for drink driving Australians. The NATSIS was Development Employment Projects or drinking in public in their most recent undertaken by the Australian Bureau of (CDEP) employment and in non-CDEP arrest in the previous five years. Theft is Statistics (ABS) in April to July 1994 in employment. The CDEP scheme is a a relatively minor component of response to a recommendation by the Commonwealth government program indigenous arrests. Given that theft is Royal Commission into Aboriginal under which unemployed indigenous the only category of indigenous arrest for Deaths in Custody (Commonwealth of persons of working age forego individual which there may be a pecuniary Australia 1991). The NATSIS covered a entitlements to unemployment benefit incentive, the preponderance of total of 4,205 households, which yielded payments in return for a grant to their indigenous arrests are not for offences 15,726 indigenous respondents, 3,076 local community council which is used to committed for financial gain. non-indigenous persons living in the fund job creation in community same household as an indigenous development activities.4 person and 158 prisoners (Australian DESCRIBING EMPLOYMENT Bureau of Statistics 1996b). For our Table 1 shows that the employment/ IN THE CONTEXT OF HIGH analysis a restricted sample was selected population rate is lower for persons who ARREST RATES from the NATSIS. First, only working-age had been arrested than for persons who members of the population (15-64 years) had not been arrested in the five years The aim of our analysis is to determine who were not in full-time schooling were prior to the survey. It is also evident that the effect of an arrest history on an included. Second, persons who were in this pattern in aggregate employment is Australian Aboriginal’s chances of being at the time of the survey were solely due to the fact that those with an employed. One issue which arises in excluded; this group represented 1.8 per arrest record are much less likely to be analysing arrest data from a self- cent of the total sample population. found in non-CDEP employment. There were 8,402 respondents in this response survey is the possibility of restricted sample. The sample was Table 2 presents information on the under-reporting of arrest. For example, further reduced to 7,568 persons after proportions of the indigenous population Freeman (1994, p. 16) notes that it is deleting observations that did not have a arrested in the last five years by selected common to find under-reporting of crime complete record of all the information characteristics. The table shows that in the by black youths. To required for the analysis. 33.3 per cent of males, and 11.2 per cent examine potential under-reporting of of females, had been arrested in the last arrest by indigenous Australians we are five years. For those arrested the restricted to a comparison between LABOUR FORCE average number of arrests is 3.0 for NATSIS data and official police data for CHARACTERISTICS males and 2.2 for females. Generally, as this is the only AND HIGH INDIGENOUS males in younger age groups and with State which reports official police arrest ARREST RATES lower levels of educational attainment data disaggregated by indigenous and have the highest incidence of arrest. non-indigenous persons. Estimates We begin with some descriptive There does not appear to be any pattern based on the official police data indicate information on employment and arrest in the incidence of arrest for persons that the proportion of indigenous persons experience for the indigenous population. living in different regions. arrested in Western Australia between

Table 1: Labour force status by whether arrested in past five years, Indigenous Australians aged 15-64, 1994

Males Females

Not arrested Arrested Not arrested Arrested Employment status (%) (%) (%) (%)

Employed-total 55.5 40.9 31.3 22.5 CDEP 19.2 21.5 9.6 10.9 non-CDEP 36.3 19.4 21.7 11.6 Unemployed 23.0 39.4 15.7 32.3 Not in labour force 21.5 19.7 53.0 45.2 Total 100.0 100.0 100.0 100.0

Number of observations 2,628 1,217 4,071 486

Source: NATSIS, unit record file.

2 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H

Table 2: Arrests in last five years by selected characteristics, Indigenous Australians aged 15-64, 1994

Males Females

Proportion Average Proportion Average arrested number of arrested number of Characteristics (%) arrests (%) arrests

Overall 33.3 3.0 11.2 2.2

Age: 15 to 24 40.6 3.2 13.6 2.5 25 to 44 37.0 2.8 12.3 2.1 45 to 64 15.0 2.9 4.7 1.7

Education: Degree / diploma 17.9 4.1 6.5 2.0 Vocational qualifications 34.3 2.5 11.6 1.8 Completed Year 12 20.3 2.1 7.9 2.5 Left school Years 10 to 11 34.9 2.8 10.7 1.9 Left school Years 6 to 9 36.1 3.2 13.3 2.3 Left school in Year 6 or below 23.4 2.9 6.5 2.8

Region of residence: Capital city 36.1 2.9 16.1 2.3 Other urban 35.4 3.1 11.9 2.1 Rural 27.5 2.8 7.5 2.3 Remote 32.4 2.9 9.8 2.3

Note: Average number of arrests is for the subset of persons arrested in the previous five years. Persons with 10 or more arrests were assumed to have 10 arrests. Source: NATSIS, unit record file.

1990 and 1994 was 24.6 per cent.5 Data from NATSIS show that 25.4 per Table 3: Reasons for most recent arrest in last five years, Indigenous Australians aged 15-64, 1994 cent of the indigenous population in Western Australia in 1994 had been Percentage of population arrested in the previous five years (Australian Bureau of Statistics 1995, Reason for most recent arrest Males Females Table 51). The closeness of the estimates of the proportion of the Drink driving 10.7 1.9 indigenous population arrested for Drinking in public 12.4 4.7 Western Australia from the NATSIS and Outstanding warrant 6.2 1.5 official police data gives us some confidence that, at least at an aggregate Assault 6.3 2.1 level, under-reporting of arrest is not a Theft 4.9 1.0 serious problem.

Source: NATSIS unit record file. METHOD OF ANALYSIS

To examine the effect of arrest on employment. The Appendix describes the issue in measuring employment status is employment we use a multiple regression method of analysis for dealing with these whether to include persons working model to predict employment, using possibilities. It is sufficient to note here under the CDEP scheme as employed or arrest and other relevant variables as that the analysis shows no evidence of not employed. Our objective is to predictors. However, in analysing the any bias in our estimates of employment characterise the determinants of data we first need to take account of induced by an effect of employment on employment status in an environment (1) the fact that a person’s employment arrest or of any effects from unobserved where direct government intervention status may have an effect on the variables. The effect of arrest on through job creation schemes is absent. likelihood of being arrested (as well as employment can therefore be assessed Since employment under the CDEP an arrest record affecting employment using a single-equation multiple scheme is generally regarded as a regression model to predict employment. i.e. the causal direction may flow both substitute for receipt of unemployment ways) and (2) the fact that unobserved Employment is measured as the benefit payments, we classify persons variables may be responsible for the employment status of the respondent at employed under the CDEP scheme as relationship between arrest and the time of the survey. One important not employed.6

3 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H

A number of data items from the NATSIS significant relationship between a probability of employment. For males were used as explanatory variables to person’s arrest record and employment having been arrested reduces the predict employment. Individuals’ status. This suggests that the effect of probability of employment from about observed employment outcomes depend arrest on employment is not simply a 0.410 to 0.227 (a decrease of 0.183); on the interaction of labour demand and proxy for a wider set of social and for females the probability of labour supply factors. Hence we seek to influences. employment falls from about 0.291 to include, as explanatory variables for 0.160 (a decrease of 0.131). Both A wide range of factors is found to affect employment, a range of factors that are marginal effects are statistically a person’s employment outcome. As likely to capture the effects of both labour significant. demand and labour supply on well as establishing the existence of a employment. These factors can be significant negative relationship The size of the effect of indigenous arrest classified as four main types of variables: between arrest and employment status, is large relative to other factors that skill, location, family, and socioeconomic. the probability of employment is found determine employment. For males, To capture skill factors, the following to be lower for younger and older age arrest is the third largest influence on variables were included: groups, for persons with low levels of employment behind education and living educational attainment and training, in a family which includes non- • age (15-24, 25-44, and 45-64 years); with difficulty in speaking English, and indigenous people. For example, • years completed at high school; higher for persons who were married, completing a qualification for males • post-school qualification (degree/ had voted in a recent election, or were increases the probability of employment diploma, vocational qualification, living in a mixed family. by between 17.1 and 25.1 percentage other, or none); Estimates of the marginal effect on the points. That is, arrest is almost as • whether or not a training course was probability of employment are presented important as education, at least for completed in the previous year; in Table 4 for several regressors. The indigenous males, in explaining table shows the estimated marginal employment. • whether or not respondent has effects, on the probability of being difficulty in speaking English. employed, measured relative to a DIFFERENCES BETWEEN Locational determinants of employment hypothetical reference person. The INDIGENOUS AND NON- status are proxied for by region of hypothetical reference person, or so- INDIGENOUS EMPLOYMENT residence (capital city, other urban, rural, called ‘base case’, is taken to be an or remote). Family-type variables included Australian Aboriginal, who: What are the implications of the effect of as predictors of employment are variables • is aged 25-44; arrest on employment status for for whether the respondent is married, understanding differences in employment lives in a mixed family (that is, with • lives in an urban region outside a outcomes between indigenous and indigenous and non-indigenous members), capital city; non-indigenous Australians? Since and the number of children in the family. • did not complete a training course in arrest rates are substantially lower for Possible socioeconomic or social the previous 12 months; non-indigenous than for indigenous influences on employment status are Australians it might be expected that the represented by variables for whether the • has no difficulty in speaking English; negative effect of arrest on employment respondent voted in any recent election, • left school in Years 6 to 9; has a long-term health condition, and is a would provide some explanation for why • has no post-school qualification; Torres Strait Islander. employment/population rates of • is not married; indigenous Australians are lower than Arrest is also included as an explanatory those of non-indigenous Australians. variable for employment. For this • voted in a recent Federal, State or purpose we use two alternative ways of ATSIC election; One approach to estimating the size of measuring arrest – one as a variable for • is not living in a racially mixed family; this effect is to multiply the difference in whether a person had been arrested in and arrest rates between indigenous and the five years and the other as a series of non-indigenous Australians by the effect • does not have long-term health variables for whether a person’s most of arrest on the probability of condition. recent arrest involved charges for employment. To undertake this exercise drinking offences, theft, assault, or an For females the base case was also a we use data on the difference in arrest outstanding warrant. woman with no children. The base case rates between indigenous and non- is intended to represent a ‘representative’ indigenous persons between 1990 and member of the indigenous population – 1994 from Western Australia (Ferrante & RESULTS and hence was chosen using sample Loh 1996, p. 39), and estimates of the information on the most likely outcome effect of arrest on employment from the for each explanatory variable.7 The regression analysis in this paper. As OVERALL EFFECT OF ARREST probability of being in non-CDEP data on arrest rates of indigenous and ON INDIGENOUS PEOPLE’S employment for the base case is 0.410 non-indigenous Australians are not EMPLOYMENT PROSPECTS and 0.291 for males and females, available disaggregated by gender the respectively. analysis is restricted to total persons. The results show that even after controlling for skill, location, family and The results show that having been Based on estimates reported earlier in socioeconomic factors there is a arrested has a significant effect on the the paper, the proportion of indigenous

4 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H

Table 4. Marginal effect of a change in the circumstance of the base case on probability of non-CDEP employment, Indigenous Australians aged 15-64, 1994

Males Females

Change in Change in probability of Standard probability of Standard Change from the base case: employment error employment error

Arrested -0.183* 0.019 -0.131* 0.025 Age 15 to 24 -0.109* 0.023 -0.110* 0.019 Age 45 to 64 -0.034 0.027 -0.060* 0.025 Torres Strait Islander -0.062 0.035 -0.020 0.031 Living in capital city 0.078* 0.031 -0.015 0.024 Living in rural area -0.049 0.025 -0.054* 0.022 Living in remote area -0.140* 0.024 -0.060* 0.022 Difficulty in English -0.095* 0.028 -0.087* 0.024 Completed training course 0.113* 0.038 0.176* 0.040 Left school before Year 6 -0.094* 0.034 -0.045 0.033 Left school in Year 10 or 11 0.073* 0.022 0.106 0.021 Completed Year 12 0.150* 0.039 0.229* 0.034 Degree / diploma and completed Year 12 0.171* 0.056 0.255* 0.035 Vocational qualification 0.251* 0.030 0.170* 0.033 Other qualification 0.192* 0.040 0.169* 0.041 Married 0.102* 0.020 0.004 0.018 Did not vote in recent election -0.090* 0.022 -0.125* 0.020 Living in a racially mixed family 0.210* 0.024 0.138* 0.023 Long-term health condition -0.026 0.021 -0.033 0.017 One child under 12 -0.068* 0.020 Two / three children under 12 -0.140* 0.020 Four plus children under 12 -0.175* 0.026

Note: Asterisk denotes significance at the 5 per cent level.

persons arrested between 1990 and the incidence of arrest by the marginal TYPE OF ARREST AND INDIGENOUS 1994 is taken as 24.6 per cent. For effect of arrest. For example, if the EMPLOYMENT OUTCOMES the proportion of non-indigenous indigenous arrest rate was reduced to persons arrested in the same period, that of other Australians, then the How does each type of arrest affect employment outcomes? Table 5 we take an upper bound estimate of indigenous employment/population rate presents the marginal effect of type of 8.5 per cent and a lower bound estimate would increase by between 3.3 and 2.4 8 arrest on the probability of employment of 2.6 per cent. The marginal effect of percentage points (depending on for the base case scenario.10 The arrest is assumed to be to reduce the whether non-indigenous arrest rate is probability of employment by 15 probabilities of being in non-CDEP assumed to equal the lower or upper employment for the base case for the percentage points. This is approximately bounds given above). As the difference equal to the average of the marginal Table 5 model are 0.384 and 0.287, for in the employment/population rate males and females, respectively. For effects of arrest on employment for between these groups in 1994 was 19.5 both males and females there are males and females. percentage points,9 the effect of arrest significant reductions in the probability of It can be shown that an estimate of the accounts for between 12 and 17 per cent employment where a person’s most effect of a reduction in indigenous arrest of the difference in employment/ recent arrest involved charges for on employment can be obtained by population rates between indigenous and drinking-related offences or was for an multiplying the hypothesised reduction in non-indigenous Australians in 1994. outstanding warrant. Arrest for theft has

5 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H

Table 5: Effect on probability of employment of arrest record by reason for last arrest, Indigenous Australians aged 15-64, 1994

Males Females

Change in Change in probability of Standard probability of Standard Reason for last arrest employment error employment error

Drinking related offence -0.120* 0.023 -0.137* 0.033 Theft -0.084 0.050 -0.170* 0.073 Assault -0.105* 0.040 -0.070 0.059 Outstanding warrant -0.140* 0.042 -0.190* 0.064

Note: Asterisk denotes significant at the 5 per cent level. no significant effect on the probability of CONCLUDING REMARKS indigenous and non-indigenous employment for males, although it does Australians could account for about 15 for females, and arrest for assault The major conclusion of this study is that per cent of the difference in employment significantly reduces the probability of the experience of arrest rather than any outcomes between these groups. Third, employment only for males. unobservable characteristics of those we find that the effect of arrest on arrested is driving the relationship employment differs by reason for most There are a number of potential between arrest and employment. That is, recent arrest. explanations for why the effect of our analysis indicates that it is not arrest on employment might differ The findings of the study resonate with possible to claim that low marketable depending on the reason for most the recommendations of the Royal ‘ability’ among certain NATSIS recent arrest. One possibility is that Commission into Aboriginal Deaths in respondents both circumscribes their the variables for the reason for arrest Custody. For example, recommendation employment prospects and explains their are a proxy for the number of arrests 62 indicates that: higher arrest rates. The analysis in the employment equation. For provides a strong empirical and [T]here is an urgent need for example, it might be thought that being governments and Aboriginal theoretical justification for the policy arrested on an outstanding warrant organisations to negotiate together to recommendations of the Royal makes it more likely that a persons will devise strategies designed to reduce Commission into Aboriginal Deaths in have been arrested on multiple the rate at which Aboriginal juveniles Custody (Commonwealth of Australia occasions and that this explains the are involved in the welfare and 1991). In particular, ensuring that large size of the effect on employment criminal justice systems and, in indigenous citizens are dealt with in ways of having been arrested on an particular, to reduce the rate at which which minimise contact with the formal Aboriginal juveniles are separated outstanding warrant. However, analysis criminal justice system should be a from their families and communities, of the number of arrests by reason for priority policy issue for governments who whether being declared in need of last arrest revealed that there was no are concerned about indigenous care, detained, imprisoned or difference in the number of arrests in employment outcomes. otherwise. (Commonwealth of each arrest category. Australia 1991, p. 83.) This study has also provided a number of Another possibility is that each type new insights into the determinants of The preponderance of alcohol-related of arrest is treated differently by employment for indigenous Australians. offences in the indigenous population employers when choosing which First, we find that a wide variety of factors also emphasises the direct benefits of potential worker to hire or has a different are related to employment outcomes. In decriminalising drunkenness effect on an individual’s motivation to addition to explanatory variables that (Commonwealth of Australia 1991, seek employment. It is difficult to see, seek to capture skill, location or family pp. 87-88). While the majority of States however, why an employer would not differences between individuals, it is also had decriminalised public drunkenness take into account an arrest for theft but found that a set of socioeconomic before 1990 (Commonwealth of Australia would take into account arrest for variables are significant determinants of 1992, pp. 279-280), the results indicate drinking-related offences. A more employment. Second, we find that that substantial economic gains can still likely scenario is that arrest for persons who have been arrested have a be made by addressing problems relating drinking-related offences indicates that significantly lower probability of to the policing of statutes relating to an individual is in an environment employment – 13.1 percentage points disorderly conduct or drinking in public. where lack of employment opportunities lower for females and 18.3 percentage An alternative strategy, nominally or social conditions reduces the points lower for males. On the basis of supported by all State and Federal perceived returns to seeking these estimates it is calculated that governments in their responses to the employment. differences in arrest rates between Royal Commission in Aboriginal Deaths

6 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H in Custody (Commonwealth of Australia years prior to the NATSIS) using a However, the nature of the data do not 1992, pp. 281-284), is to ensure ongoing number of explanatory variables. The allow us to predict arrest from the funding and maintenance of adequate explanatory variables included as employment variable because non-custodial facilities for the care and predictors of arrest are the same set of employment is measured at the time of treatment of intoxicated persons. variables used to predict employment plus the survey whereas arrest is based on two extra variables, namely, whether the what has occurred in the five years prior The importance of general respondent had ever drunk alcohol and to the survey. Therefore to predict arrest socioeconomic and family factors in whether the respondent was taken from we can only use proxy measures of prior determining employment outcomes has his or her natural family. Members of the employment status, and, as already general implications for employment ‘stolen generation’ who were taken from discussed, we have to rely on current studies in the population at large. For their natural families have experienced social and demographic measures to act example, the significance of the as proxy measures of these variables socioeconomic indicators, such as social dislocation and alienation which over the previous five years. whether a person voted in a recent anecdotal evidence suggests has election, means that labour economists significantly increased contact with the The model for arrest therefore includes should consider controlling for such criminal justice system (Commonwealth of the same explanatory variables as are to factors, where possible, as a matter of Australia 1997, pp. 12-16). be used to predict employment (plus two course. In considering explanatory variables for additional explanatory variables, as discussed above). This model does not The economic and social costs of low the arrest outcome we note that the arrest only provide estimates of the effects of rates of employment for indigenous record variable represents a summary of the measured explanatory variables. Australians are significant and represent arrest outcomes in previous time periods The effects of other unobserved a major problem for policy-makers in and hence it is actually a function of the variables on arrest can be estimated Australia. Direct labour market lagged values of explanatory variables from the unexplained variation in the intervention is frequently advocated as a (because, for example, an arrest four fitted arrest model. The generalised solution for poor indigenous employment years prior to the survey date would be residual from this fitted model measures outcomes. The analysis in this bulletin predicted by explanatory variables at that the unexplained variation and can suggests that it will also be necessary to time). Unfortunately, information on therefore be regarded as a variable address the social environment in which explanatory variables in previous time capturing the effects on arrest of all individuals make decisions about labour periods is not available from the NATSIS explanatory variables not specifically supply and labour demand – and in so that it is necessary to include included in the model. particular, to address the problem of the explanatory variables from the current high arrest rates among indigenous time period as proxies for effects from The next step is to fit a model to predict Australians. previous time periods. employment (that is, whether or not a For some variables which are relatively person was employed at the time of the ‘permanent’, such as age, educational NATSIS). The explanatory variables in APPENDIX the employment model include the attainment, whether a person was taken economic and social characteristics from their natural family, and whether There are two issues to address in the described above, as well as arrest and they drink alcohol, use of current period data analysis. The first is that not only is the generalised residual from the arrest variables should not cause a significant it possible for a person’s arrest history to model. The model for employment loss of information. On the other hand, affect the person’s employment, it is also therefore includes as an explanatory high rates of geographic mobility in the possible that a person’s employment variable a measure of the effects of all indigenous population are likely to mean status might affect the person’s chances other unobserved variables which affect that variables related to current location of being arrested (in other words, the arrest. may be less accurate as proxies for causal relationship can go in both previous location. This two-stage process, first fitting a directions). The second issue to model for arrest, then using information consider is that the observed relationship It should be noted though that the from it to fit a model for employment, between arrest and employment may be primary objective in fitting a regression deals with our two issues of concern. driven by a third factor that is not model to predict arrest is to control for Using the predictors of employment as adequately captured by the data. For unobserved characteristics of the explanatory variables in the arrest model example, unmeasured regional variations NATSIS respondents, rather than to seek accounts for the effect of prior in outcomes, arising from the failure to to interpret the coefficient estimates in employment status on arrest, and using include a State indicator in the NATSIS that regression or to explain arrest the generalised residual from the arrest unit record data, may bias the estimated outcomes. To undertake the latter task of model as a predictor in the employment relationship between arrest and explaining arrest outcomes it would be model accounts for the effects of employment. Other factors that are necessary to take proper account of the unobserved variables. intrinsically difficult to measure, such as wide range of theoretical work on the The generalised residual term was not ‘ability’, may also induce the appearance determinants of criminal activity significant in any of the employment of a relationship. (Broadhurst 1997, pp. 413-415). regressions analysed for this bulletin. To resolve these issues we first fit a To allow for an effect of employment on Therefore (1) the potential bias (resulting model to predict arrest (that is, whether or arrest we would like to include from arrest depending on employment as not a person had been arrested in the five employment as a predictor for arrest. well as employment depending on arrest)

7 B U R E A U O F C R I M E S T A T I S T I C S A N D R E S E A R C H

is not a significant issue in the data and Ferrante, A. & Loh, N. 1996, Crime and the proportion of the population who were (2) the unobserved characteristics of justice statistics for Western Australia: arrested in that period.) The adjustment uses a measure of the average number of arrests per people who have been arrested cannot 1994, mimeo, Crime Research Centre, arrested person over the previous five years in explain the correlation between arrest University of Western Australia, Perth. Western Australia derived from the NATSIS data and employment. This result provides (3.26 arrests). clear evidence that it is the experience of Freeman, R. 1994, ‘Crime and the job 6 Classifying persons employed under the CDEP arrest that adversely affects employment market’, NBER Working Paper No. 4910, scheme as non-employed does not involve any prospects for indigenous people. National Bureau of Economic Research, judgement about the ‘genuineness’ of CDEP employment. It simply derives from an New York. The final stage of the analysis is to fit a assumption that those persons would not be in probit regression model to predict employment in the absence of government Gallagher, P. & Poletti, P. 1998, intervention through the CDEP scheme. employment using both arrest and the Sentencing Disparity and the Ethnicity of social and economic characteristics as 7 There was one exception. The base case Juvenile Offenders, Judicial Commission was chosen as a ‘not married’ person although predictors, but excluding the generalised of , Sydney. there were more married respondents than residual term. there were single respondents in the sample. Hunter, B. 1997, ‘The determinants of (The standard presentation of results in labour market studies is to focus on the effect of Indigenous employment outcomes: the marriage.) REFERENCES importance of education and training’, 8 The upper bound estimate is based on the Australian Bulletin of Labour, vol. 23, assumption that each person is arrested Australian Bureau of Statistics 1994, no. 3, pp. 177-192. only once; it equals the total number of Labour Force, Australia, Cat. no. 6204.0, non-indigenous arrests as a proportion of the ABS, . non-indigenous population. The lower bound estimate, which adjusts for the possibility NOTES that the same individual is arrested on Australian Bureau of Statistics 1995, multiple occasions, equals the total number of National Aboriginal and Torres Strait non-indigenous arrests as a proportion of the 1 Boyd Hunter is Ronald Henderson Research Islander Survey: Detailed Findings, non-indigenous population divided by the Fellow at the Centre for Aboriginal Economic Cat. no. 4190.0, ABS, Canberra. estimate of arrests per person in Western Policy Research, The Australian National Australia from the NATSIS data. University, Canberra. Australian Bureau of Statistics 1996a, 9 In 1994 the employment/population rate for Employment Outcomes for Indigenous 2 Jeff Borland is Associate Professor in the indigenous Australians was 35.9% (Australian Australians, Cat. no. 4199.0, ABS, Department of Economics, University of Bureau of Statistics 1996a) and the Melbourne and Visiting Fellow at the Centre for employment/population rate for non-indigenous Canberra. Economic Policy Research, Research School of Australians was 55.4% (Australian Bureau of Social Sciences, The Australian National Statistics 1994). Australian Bureau of Statistics 1996b, University, Canberra. National Aboriginal and Torres Strait 10 In order to fit the model for this analysis there was some minor respecification of the model, Islander Survey - Unit Record File, 3 The employment/population rate is calculated as the number of employed persons of working age with some variables omitted. The base case for Cat. no. 4188.0, ABS, Canberra. divided by the total population of working age, this model was an Australian Aboriginal who: where working age is taken to be 15 to 64. This • is aged 25-44; Borland, J. & Hunter, B. 1999, ‘Does rate differs from the usual ‘participation rate’ for crime affect employment status? - The which the labour force (those employed or • lives in an urban region outside a capital case of Indigenous Australians’, seeking work) is used as the denominator in the city; Economica, forthcoming. rate calculation. • has no difficulty in speaking English;

4 The CDEP scheme began in 1976/77 with the • left school before Year 9; Broadhurst, R. 1997, ‘Aborigines and participation of a single community of 100 crime in Australia’, Crime and Justice, persons. By the time of the NATSIS survey, • has no post-school qualification; vol. 21, pp. 407-468. however, the scheme had expanded to 262 • is not married; communities with over 27,000 participants and Commonwealth of Australia 1991, Royal accounted for more than 30 per cent of • voted in a recent Federal, State or Commission into Aboriginal Deaths in expenditure on indigenous persons. ATSIC election; Custody, National Report, Volume 5, 5 Official police data show that in each year from • does not have long-term health condition. AGPS, Canberra. 1990 to 1994 total arrests were 15.9 per cent, 16.9 per cent, 15.9 per cent, 15.6 per cent, and Commonwealth of Australia 1992, 15.9 per cent of the indigenous population in Aboriginal Deaths in Custody: Response Western Australia (Ferrante & Loh 1996, p.39). by Governments to the Royal To make the official police data comparable with the NATSIS data it is necessary to convert the Commission, Volume 1, AGPS, annual percentages to an estimate of the Canberra. proportion of the indigenous population arrested over the previous five years. This calculation is Commonwealth of Australia 1997, made by summing total arrests as a percentage Bringing Them Home: Report of the of the indigenous population across the five-year National Inquiry into the Separation of period from 1990 to 1994 and then adjusting to Aboriginal and Torres Strait Islander take account of persons who were arrested multiple times throughout the period. (The Children from Their Families, Human adjustment is necessary because, where some Rights and Equal Opportunity persons were arrested more than once over five Commission, Sydney. years, the official police data will over-estimate

NSW Bureau of Crime Statistics and Research Website: www.lawlink.nsw.gov.au/bocsar/

Level 8, St James Centre, 111 Elizabeth Street, Sydney 2000 Tel:8 9231-9190 ISSN 1030 - 1046 ISBN 0 7313 2610 5