CRIME SCARS: RECESSIONS AND CAREER CRIMINALS

Transcription

1 CRIME SCARS: RECESSIONS AND CAREER CRIMINALS Brian Bell*, Anna Bindler** and Stephen Machin*** July 2012 * Department of Economics, University of Oxford and Centre for Economic Performance, London School of Economics ** Department of Economics, University College London *** Department of Economics, University College London and Centre for Economic Performance, London School of Economics Abstract Recessions lead to short-term job loss, lower levels of happiness and falls in income. There is also growing evidence that workers who join the labor market in downturns suffer from poorer job matches that have a sustained detrimental effect on their wages and career progression. This paper uses UK data to document a more disturbing long-run effect of recessions. Youths who leave school in the midst of recessions are significantly more likely to lead a life of crime than those graduating into a buoyant labor market. These effects are long-lasting and substantial. Recessions characterized by high-levels of long-term unemployment drive the result short-term unemployment levels do not generate scarring effects. JEL Codes: K42, J64 Keywords: Crime, Unemployment, Scarring Corresponding Author: Brian Bell (b.bell1@lse.ac.uk) Acknowledgements This research was funded by the Economic and Social Research Council at the Centre for Economic Performance. We are grateful to Kory Kantenga for research assistance. 0

2 I. INTRODUCTION What makes a career criminal? 72% of males aged over 25 in the U.K. who were convicted of a crime in 2002 had a criminal record that went back to their teenage years. Adult onset offenders are very much the exception. 1 Therefore, the experience of crime during the formative years of youth seems to have a persistent and long-run effect on the likelihood of being a career criminal. In this paper we explore whether the labour market conditions that the young encounter when they first leave school has an important role to play in this process. The standard Becker model of crime can help us understand how this mechanism can play out. As youths leave school, they face a trade-off between legal and illegal activities. As unemployment rises, the expected returns to legal activity i.e. work falls. All else equal, this encourages some youths to commit crime who would have successfully avoided such a result in a more buoyant labour market. As these youths age, they continue to face the trade-off. Two obvious mechanisms now link their earlier decisions with later ones. First, earlier experiences of crime increase their stock of criminal knowledge and potentially reduce the costs of criminal activity. Second, a previous criminal record reduces the expected wage in the legal labour market. Both of these effects work to increase the likelihood that the individual will end up as a career criminal. Our analysis contributes to two distinct literatures. First there has been an extensive, though unresolved, debate over the link between recessions and crime (see Freeman, 1999). This literature has focused on the issue of whether crime rates, and in particular property crime rates, are countercyclical. Much of the early work based on 1 Evidence suggests the crime-age profile peaks for men around age

3 macroeconomic data or on cross-sections of US cities or states finds it hard to uncover a relationship between crime and unemployment, leading Freeman (1999) in his review of the literature to state that the evidence is 'fragile'. More recent work also studies the impact of low wages. Gould, Weinberg and Mustard (2002) examine the impact of contemporaneous unemployment and wages on the criminal behaviour of less educated males. Exploiting a panel of U.S. counties, they find significant effects for both wages and unemployment, though the impact of wages is more significant given the substantial decline in the real earnings of less educated males in the U.S. over the 1980s and 1990s. Similarly, Raphael and Winter-Ebmer (2001) find a positive correlation between current unemployment and property crime for a panel of U.S. states, though the evidence for any link with violent crime is considerably weaker. Machin and Meghir (2004) also find evidence that low wages matter more than unemployment. This evidence tends to suggest, as predicted by the Becker model, that crime becomes a more attractive option in the face of bleaker labor market conditions. Second, there is a growing literature on the effects of first entering the labor market during recessions (starting with Beaudry and DiNardo, 1991). The aim of this work has been to assess whether such workers experience sustained long-run negative effects. Thus, for example, Oreopoulos, von Wachter and Heisz (2012) exploit a large Canadian longitudinal dataset to show that the cost of recession for new graduates is substantial and long-lasting. A typical recession a 5 percentage point rise in the unemployment rate is associated with an initial loss of earnings of about 9 percent that halves within 5 years, and finally fades to zero by 10 years. The process operates via initial placement with lower paying employers and recovery through gradual job mobility to better firms. 2

4 Graduates in the lower quintile of the ability distribution suffer permanently lower wages, while the more able graduates quickly bounce back. Similar results are reported by Kahn (2010) using longitudinal data on U.S. college graduates, though some of her results suggest that the wage penalty is longer-lasting. By contrast, Benedetto, Gathright and Stinson (2010) find no evidence of a persistent impact of graduation-year unemployment on earnings using U.S. social security earnings data. Taking a somewhat different approach, Oyer (2006, 2008) has examined the career paths of particular occupations, namely economists and investment bankers, to assess the importance of initial conditions. He shows for example, that stock market conditions at time of graduation have a strong effect on whether MBA students go directly to Wall Street, or instead pursue alternatives such as consulting. Further, starting directly in investment banking after graduation causes a person to be more likely to work there later on and earn significantly more. The size of these effects are substantial, amounting to several million dollars in present value. Outside of the labour market, MacLean (2011) finds that males who graduate from high school in a recession have worse health outcomes at age forty than those graduating in a more auspicious labour market. This is true for both self-reported health measures and objective measures of physical and mental health. The rest of the paper is structured as follows. In the next section we formalize the intuition of the link between initial labour market conditions at entry and the future path of criminal behaviour using a simple dynamic economic model of crime. In Section III we discuss the empirical strategy and the various datasets that we use in the empirical work. Our main results are presented in Section IV, where we present both panel 3

5 evidence across geographic areas for successive cohorts of males and micro evidence using self-reported arrest data. Our conclusions are presented in Section V. II. A SIMPLE MODEL We consider a simple dynamic Becker crime model. In the standard Becker crime model, individuals are rational decision makers and choose between legal and illegal activity. Their choice is based on the expected returns to both options. In this simple framework, returns to legal activity are solely determined by the market earnings from employment whereas returns to illegal activity take into account the potential crime payoff, the probability of getting caught and the expected sanction if caught. If the expected return to illegal activity outweighs the expected return to legal activity, the individual chooses to commit crime. More formally, let U(.) denote an individual s utility function, W t market earnings from employment in time period t, W ct the crime payoff, p t the probability of getting caught and S t the sanction if caught. In this model, an individual commits crime if: U(W ) < (1 p)u(w ) + pu(s) Therefore, an individual chooses between legal and illegal activity at each point in time t (and we assume, for simplicity, that the probability of being caught and the sanction are time invariant. At labour market entry, the individual initially decides between both options and commits crime if: U(W ) < (1 p)u(w ) + pu(s) where t=0 denotes the labour market entry period. Note that W 0 decreases with an increasing unemployment rate as expected returns to work fall with an increasing 4

6 unemployment rate. Thus, all else equal, a higher unemployment rate in that initial labour market period triggers an individual at the margin to commit crime. As individuals age, they re-decide between legal and illegal activity following the same decision rule: U(W ) < (1 p)u(wt ) + pu(s) where t now denotes period t>0. There are two mechanisms linking the current decision with earlier decisions. Firstly, let the legal wage be a decreasing function of past criminal behaviour up to period t, summarised by C t : U(W ) U(W (C )) with W (C ) C Past criminal behavior reduces the expected return to legal activity. Assuming that the criminal history of an individual is observable for an employer, signaling effects may lead to a wage reduction. Alternatively, past criminal activity reduces the amount of time spent obtaining human capital relevant to the labour market. Secondly, let the cost of crime be a decreasing function of past criminal behavior up to period t. Assuming that the underlying mechanism is a criminal learning process which reduces the probability of < 0 getting caught and thus decreases the expected cost to crime, we can write: p p(c ) with p(c ) C Both mechanisms increase the probability of an individual choosing to commit crime in later careers. Taking everything into account, we thus expect higher unemployment rates at labour market entry to trigger higher crime rates in later periods by generating career criminals. < 0 5

7 III. EMPIRICAL APPROACH AND DATA Our empirical analysis focuses on the UK and we use a variety of different data sources. First, we exploit panel data on year-of-birth cohorts over regions and time. This comes from administrative sources and provides data on conviction rates. Thus we can observe cohorts as they enter the labour market and then follow them throughout their working life.second, we explore repeated cross-section micro data on self-reported criminal behavior of a sample of individuals. This has the advantage of providing accurate data on year of labour market entry and a rich set of observable characteristics, but suffers from the usual problem of self-reporting of criminal behavior. The strength of our analysis is that we use two different types of data and find similar effects. For the panel analysis, our unit of analysis is at the year-of-birth cohort (c), region (r), and calendar year (t) level. We estimate the long-run effects of initial labour market conditions by exploiting the regional variation in unemployment rates across time: (C i ) U t + + where we have cohort, region and time fixed-effects.the vector X contains controls for time-varying characteristics.the unemployment rate is measured at the time of labour market entry, denoted t=0. The main coefficient of interest is on the initial unemployment rate and this is allowed to vary with potential labour market experience. Thus we measure the extent to which any effect of initial unemployment on criminal behaviour persists as length of time since labour market entry increases. Our principal data source for the panel is the Offenders Index Database (OID). The OID records every criminal conviction in England and Wales for four selected weeks in each year over the period 1963 to For every convicted individual we have data on 6

8 date of birth, region, sex and nature of each conviction 2. As our empirical strategy is based on individual year-of-birth cohorts in each region, we aggregate the OID data each calendar year by year-of-birth cohort and region. We then merge in annual population data by sex, year-of-birth and region to produce conviction rates. We restrict the sample to those males aged since only y% of offences are committed by women and only x% of offences are committed by those aged 40 and over. We have nine regions within England and Wales. Unemployment rates (both youth and total) at the regional level are computed from the Labour Force Survey from 1975 and merged with administrative unemployment data before then. In addition in some specifications we split the unemployment rate into short-term and long-term unemployment, defined as unemployment durations greater than 52 weeks. This allows for the possibility that recessions in which there is a substantial rise in expected unemployment duration may have more detrimental effects as the job-finding probability declines more significantly and the erosion of human capital increases (see Machin and Manning, 1999). There are two key issues with these data. First, we assume that cohorts do not substantially move across regions over time since we link the current crime data for a particular cohort in a region to the initial unemployment rate of the cohort in the same region. In other words, the criminal behaviour of 30 year-olds in the West Midlands in 2000 is assumed to be affected by the unemployment rate in the West Midlands in This is of course only strictly valid if there is no inter-regional mobility since schoolleaving age. If mobility is random since school exit, the estimates will merely be noisy. However if mobility is driven by self-selection, the coefficient of interest may be biased 2 Individuals convicted of multiple offences have each offence recorded in the data. 7

9 (Dahl, 2002). In Section IV, we present some robustness tests conducted using mobility data from the Census to examine this issue (to be done). Second, we use the compulsory school-leaving age to identify the entry year into the labour market. The key justification for this is that we know that the majority of arrested and/or convicted criminals have low educational attainment and generally do leave school at this age. Thus for example in 1991, 89% of convicted offenders in the UK had left school at the compulsory school leaving age, compared to 77% of the overall population (Lynch et al, 1994). Our second set of results use cross-sectional data on self-reported arrests. Such data provide information on the age at which the respondent left full-time education and so allow us to precisely date the year of labour market entry. In addition, we can also identify the region of residence at age 16, so we can be sure of allocating the correct entry unemployment rate to each individual. Finally, the data provide an extensive set of personal characteristics, which we would expect to be correlated with criminal activity. There are two key disadvantages in using such micro data. First, the usual concern associated with the self-reporting of arrests. In the context of this study however, this would only bias our estimates if the self-reporting probability varied within a cohort depending on the initial regional unemployment rate. It seems to us hard to make such a case. Second, we have no information of when the arrest occurred the question is simply whether the individual has ever been arrested. If poor labour market conditions at entry generated a rise in the probability of arrest for the relevant cohort but no subsequent scarring effects, we would still find a positive effect of entry unemployment on the our dependent variable. However, since such an effect is a necessary condition for our story, 8

10 it is instructive to examine the results from such an exercise. More specifically, the micro data allows us to estimate the average impact of initial unemployment on all future criminal behaviour and does not allow us to investigate the persistence of such effects. Our micro-data comes from the British Crime Survey (BCS). The BCS is a large (45,000 individuals) annual cross-section survey used to construct measures of crime victimization. A sub-sample each year are asked whether they have ever been arrested by the police. There is no information on the type of crime for which they were arrested nor on the eventual outcome. However as we use conviction data in the panel analysis, it is useful to have an alternative measure of criminal behaviour (as in Lochner and Moretti, 2004) to check the robustness of our findings. We have a broad array of personal characteristics including educational attainment, ethnicity, marital status, housing status and employment and income measures. In addition, the survey asks both about current area of residence and length on residence in the same area. Thus we can identify a sub-set of individuals whose region of residence at age 16 is known to be the same as their current region of residence and so overcome any self-selected mobility bias. 3 IV. RESULTS We begin by presenting evidence on the average effect of initial labour market conditions on criminal activity. In terms of equation (1), this specification restricts the to be the same across experience groups. The first three columns of Table 1 use the total crime rate as the dependent variable, whereas the final two columns examine property and violent crime respectively. The key explanatory variable, the entry-year unemployment rate, is 3 The data is not quite as clean as this suggests. Individuals are asked how long they have lived in the same area rather than region. However given the size of our regions (9 for England and Wales), we assume that area is smaller than region and fully contained within it. 9

11 measured for the whole labour force in the first column and using only the youth unemployment rate in subsequent columns. All regressions include year, cohort and region fixed effects, are weighted by cell-population and robust standard errors are clustered at the cohort level. Column (1) shows a strong positive effect of initial unemployment on average conviction rates. In terms of magnitude, a fairly typical recession increases the aggregate male unemployment rate by around 5 percentage points. Thus, the average conviction rate for a cohort entering the labour market in recession is around 10% higher than for a similar cohort entering into a growing labour market. This is a substantial effect but in some respects understates the detrimental effect of recessions. Within a cohort, there will be a substantial share for which the marginal effect is zero, since their Becker trade-off thresholds will be such that the recession has no effect on their optimal decision i.e. they are at an interior solution that results in no illegal behaviour and the recession does not move them across the threshold. Thus the average effect we estimate is a combination of a zero effect for a large share of the cohort and a substantial effect for those close to the legal/illegal threshold in the absence of a recession. In column (2) we replace the aggregate male unemployment rate at entry with the youth unemployment rate. Again we find a strong positive effect - the coefficient is smaller of course because of the higher average youth unemployment rate. In column (3) we allow for both the aggregate and youth unemployment rate. Perhaps surprisingly, the aggregate unemployment rate is driving the results. It may be that the aggregate rate better reflects the overall state of the economy at the time as the youth rate can be more affected by schooling decision etc or it may be that the aggregate rate is measured with 10

12 less error (both rates come from the same Labour Force Survey so cell-sizes at the region/year level are naturally higher for the aggregate as opposed to the youth rate). In any case, columns (1) and (2) show the same detrimental effects regardless of the precise unemployment rate used. Finally in columns (4) and (5) we split the conviction rate into property and violent crime. The results suggest that the unemployment effect is slightly higher for property than violent crime, but is substantial and significant for both. Are all recessions as bad for future crime outcomes? Recessions differ in their severity and in their impact on unemployment durations. During the 1980s, recessions were increasingly associated with a rise in the expected duration of unemployment spells. By the mid 1980s, around one-half of all the unemployed had durations in excess of 52 weeks. There is substantial evidence of the scarring effects of such long durations (see Machin and Manning, 1999). In terms of our theoretical model, the rise in duration reduces the probability of employment and also potentially reduces the returns to legal work as skills atrophy. To examine these effects, we split the unemployment rate at labour market entry into the short-term component and the long-term component (more than 52 weeks unemployed). We have data on regional long-term unemployment by year from 1970, so our panel sample is reduced since we cannot identify entry conditions for any cohort entering the labour market prior to In Table 2 we begin by replicating the basic specification on this reduced sample. As column (1) shows, the average effect is almost identical to that estimated in Table 1. In the second column, we split the initial unemployment by duration. As expected, we find that the effect of initial unemployment on future crime comes through long-term unemployment rather than short-term unemployment. In other words, recessions characterized by rising short-term 11

13 unemployment but no significant rise in long-term unemployment (as occurred in the 1960s in the UK and commonly typify the US experience prior to the Great Recession) do not produce crime scars on newly entering cohorts. As unemployment rises, these entrants experience a fall in job-finding rates that is not substantial or long-lived enough to switch them to a criminal career. In contrast, rises in long-term unemployment are very detrimental. The same pattern emerges if examine property and violent crime separately, though the average effects on violent crime are smaller and less precisely estimated on this reduced sample. Thus far, we have demonstrated a statistically significant and economically substantial effect of initial unemployment conditions on the average conviction rate of cohorts over their entire lifetime. We now turn to an examination of the persistence of this effect. Is the effect primarily driven by a very large impact on crime in the early years of labour market entry that then subsides as the young age and establish a stable legal career? Or is the effect persistent, with those affected by harsh labour market conditions at entry pushed into a criminal career that become self-perpetuating for the reasons discussed in Section II. To examine this, we allow the coefficient on initial unemployment to vary by years since labour market entry. We group experience into four categories (0-5 years, 6-11 years, years and years) and allow for differential effects for each category. The specification is otherwise identical to that of Table 1. The key result is shown in Column 1. The effect of initial unemployment on criminal behaviour is persistent and shows no sign of declining as length of time since labour market entry increases. Indeed the effect in the very early years of labour market experience is small and insignificant. This may seem surprising since the wage impact 12

14 literature tends to find the largest negative effects at impact, with the effect declining over time. Two considerations help explain our result. First, the dynamics of the crime-work choice suggest that the impact of initial unemployment will not be immediate. New labour market entrants will take time, perhaps a quite substantial length of time, to determine that the labour market conditions are serious and long-lasting enough to affect the present value of the crime-work tradeoff. This will delay the move into criminal activity and weaken the immediate impact of unemployment. Second, it should be recalled that our dependent variable is conviction in a criminal court rather than criminal activity. Again, there will be substantial and variable lags between beginning a criminal career and being convicted of a crime. These are functions of both detection rates and the non-conviction sanctions often used for early offenders e.g. cautions. For all these reasons, we might expect a smaller estimated effect on the contemporaneous effect of unemployment on convictions. No such considerations apply to the persistent effect we identify and suggest that the initial unemployment conditions lead to a sub-set of cohort members pursuing a criminal career that would otherwise have been avoided. The remaining columns of the Table show the effects are the same if the youth unemployment rate is used and across different crime categories though less precisely estimated. Finally, we turn to the micro evidence on self-reported arrests to examine whether we find the same key link between crime and initial labour market conditions. We estimate probit models with the dependent variable taking the value one if the respondent reports having ever been arrested by the police. We include survey year dummies and an extensive set of personal controls. Table 4 reports the results. In column one, the estimated coefficient on entry unemployment is positive and significant at the 5% level. 13

15 So again, there is a positive link between initial labour market conditions and the average arrest probability over subsequent periods. In the second column we restrict to the focus to those whose highest qualification is GCSE and therefore will have left education at age 16. This is the first sample for which we can tightly link exit from education and the initial unemployment rate and is a sample that contains a larger fraction of individuals at risk of criminal behaviour. As expected, we find a substantially larger and more strongly identified impact of entry unemployment for this group. Next, we further restrict this sample of high-school leavers to those who graduated in the same region in which they currently live. This avoids biases from self-selected migration. The sample size falls substantially and the coefficient is poorly determined. However, the magnitude of the coefficient is almost identical to that of the previous column suggesting that the mobility bias may be small. In the final column, we conduct a placebo experiment. The arrest record of individuals who report graduating from college should not be affected by the unemployment rate when they were 16 since this did not affect them. Sure enough we no longer find a positive effect for these individuals indeed the effect is estimated to be significantly negative. V. CONCLUSIONS Despite there being a literature examining whether poor labour market entry conditions have a scarring effect on individuals' labour market outcomes in subsequent years, the issue of whether a crime scar from recessions exists has not yet been explored. In this paper, we present the first evidence that recessions can lead to substantial and persistently higher rates of criminal behaviour among those likely to be most impacted by such 14

16 conditions those newly entering the labour market. In contrast to much of the evidence on the long-run effect of initial unemployment on wages and career trajectories, we find that the effect on criminal behaviour shows no sign of diminishing as time since labour market entry increases. As a consequence, recessions play a role in the production of career criminals. 15

17 REFERENCES Beaudry, P., and DiNardo, J. (1991) The effect of implicit contracts on the movement of wages over the business cycle: Evidence from micro data, Journal of Political Economy, 99(4), Benedetto, G., Gathright, G., and Stinson, M. (2010) The Earnings Impact of Graduating from College during a Recession, mimeo. Dahl, G. (2002) Mobility and the Return to Education: Testing a Roy Model with Multiple Markets, Econometrica, 70(6), Freeman, R. (1999) The Economics of Crime, in O. Ashenfelter and D. Card (eds.) Handbook of Labor Economics, North Holland Press. Gardecki, R., and Neumark, D. (1998) Order from Chaos? The effects of youth labor market experiences on adult labor market outcomes, Industrial and Labor Relations Review,51(2), Genda, Y., Kondo, A., and Ohta, S. (2010) Long-Term Effects of a Recession at Labor Market Entry in Japan and the United States, Journal of Human Resources, 45(1), Gould, E. D., Weinberg, B. A., and Mustard, D. B. (2002) Crime Rates and Local Labor Market Opportunities in the United States: , Review of Economics and Statistics, 84(1), Kahn, L. (2010) The Long-Term Labor Market Consequences of Graduating from College in a Bad Economy, Labour Economics, 17(2), Lochner, L., and Moretti, E. (2004) The Effect of Education on Crime: Evidence from Prison Inmates, Arrests and Self-Reports, American Economic Review, 94(1), Lynch, J. P., Smith, S. K., Graziadei, H. A., and Pittayathikun, T. (1994) Profile of Inmates in the United States and in England and Wales, 1991 Bureau of Justice Statistics. Machin, S. and Manning, A. (1999) The Causes and Consequences of Long-Term Unemployment in Europe, in O. Ashenfelter and D. Card (eds.) Handbook of Labor Economics, North Holland Press. Machin, S. and Meghir, C. (2004) Crime and Economic Incentives, Journal of Human Resources, 39,

18 MacLean, J. C. (2011) The health effects of leaving school in a bad economy, mimeo, Cornell University. Oreopoulos, P., von Wachter, T., and Heisz, A. (2012) The Short- and Long-Term Career Effects of Graduating in a Recession, American Economic Journal: Applied Economics, 4(1), Oyer, P. (2006) Initial labor market conditions and long-term outcomes for economists, Journal of Economic Perspectives, 20(3), Oyer, P. (2008) The Making of an Investment Banker: Macroeconomic Shocks, Career Choice and Lifetime Income, Journal of Finance, 63, Raphael, S., and Winter-Ebmer, R. (2001) Identifying the Effect of Unemployment on Crime, Journal of Law and Economics, 44,

19 TABLE 1: U.K. COHORT PANEL BASIC SPECIFICATION (1) (2) (3) (4) (5) Crime Type: All All All Property Violent Entry U rate 2.146** 2.888** (0.439) (0.845) Entry U rate ** ** 0.675** (0.165) (0.302) (0.173) (0.132) Year Fixed Effects x x x x x Cohort Fixed Effects x x x x x Region Fixed Effects x x x x x Sample Size Notes: Dependent variable is the convicted male offender rate from the OID. Individual year-of-birth cohorts run from We assume that cohorts enter the labour market at age 16. All unemployment rates are measured in year of labour market entry. Sample runs from All regressions include year, cohort, and region fixed effects. Standard errors are clustered at the cohort level and regressions are weighted by the male cell-population. ** indicates significance at the 1% level. * indicates significance at the 5% level. 18

20 TABLE 2: U.K. COHORT PANEL LONG TERM UNEMPLOYMENT RATE All Property Violent (1) (2) (3) (4) (5) (6) Entry U Rate 1.907** 1.926** (0.388) (0.380) (0.323) Entry Long Term U Rate 3.063** 3.635** (0.501) (0.540) (0.552) Entry Short Term U Rate (0.968) (0.759) (0.858) Year Fixed Effects x x x x x x Age Fixed Effects x x x x x x Region Fixed Effects x x x x x x Sample Size Notes: Dependent variable is the convicted male offender rate from the OID. Individual year-of-birth cohorts run from The Long term unemployment rate is defined as the unemployment rate of those unemployed for more than 12 months. Sample runs from All regressions include year, cohort, and region fixed effects. Standard errors are clustered at the cohort level and regressions are weighted by the male cell-population. ** indicates significance at the 1% level. * indicates significance at the 5% level. 19

21 TABLE 3: U.K. COHORT PANEL LABOR MARKET EXPERIENCE EFFECTS (1) (2) (3) (4) Crime Type: All All Property Violent Entry U rate*(0_5) (0.533) Entry U rate*(6_11) 2.595** (0.365) Entry U rate*(12_17) 2.476** (0.327) Entry U rate*(18_23) 2.150** (0.486) Entry U rate 16-19*(0_5) ** (0.220) (0.199) (0.193) Entry U rate 16-19*(6_11) 1.031** 1.047** 0.674** (0.133) (0.150) (0.166) Entry U rate 16-19*(12_17) 1.042** 0.826** 0.362* (0.122) (0.163) (0.143) Entry U rate 16-19*(18_23) 1.056** (0.194) (0.262) (0.211) Year-of-arrest Fixed Effects x x x x Year-of-birth Fixed Effects x x x x Region Fixed Effects x x x x Sample Size Notes: Dependent variable is the convicted male offender rate from the OID. Individual year-of-birth cohorts run from We assume that cohorts enter the labor market at age 16 and experience is defined as years since entering the labor market. The experience interacting variable is a dummy taking on value of 1 if the years of experience are as indicated and 0 otherwise. All unemployment rates are measured in year of labor market entry. Sample runs from All regressions include year, cohort, and region fixed effects. Standard errors are clustered at the cohort level and regressions are weighted by the male cell-population. ** and * indicate significance at the 1% and 5% level respectively. 20

22 TABLE 4: BRITISH CRIME SURVEY SELF-REPORTED OFFENDING Males Males/GCSE Males/GCSE Same Area Males/College (1) (2) (3) (4) Entry U rate 0.230* 0.614** ** (0.113) (0.198) (0.499) (0.101) Year Dummies Yes Yes Yes Yes Personal Controls Yes Yes Yes Yes Mean of Dependent Variable Sample Size 20,522 5,985 1,988 6,546 Notes: Table reports probit estimated marginal effects. Standard errors in parentheses are robust to arbitrary heteroskedasticity and autocorrelation (clustered at the government office region level). Personal controls include age (10 categories), gender (where appropriate), ethnic group (5 categories), education (9 categories where appropriate), student status, marital status (4 categories), income (18 categories), economic status (15 categories), number of children (10 categories), housing tenure (8 categories), years at address (9 categories), years in area (9 categories), and government office region (10 categories). The sample covers ages 16 to 65 of pooled British Crime Surveys, to * indicates significance at the 5% level. ** indicates significance at the 1% level. 21