biology & criminal behavior

THE IMPACT OF YOUTH CRIMINAL BEHAVIOR ON ADULT EARNINGS Sam Allgood University of Nebraska [email protected] B. Mustard University of Georgia [email protected] S. Warren, Jr.

University of Georgia [email protected] 1999 Abstract Individuals charged with or convicted of a criminal offense when young complete fewer years of schooling and accumulate less work experience as young adults than those with no contact as a youth with the criminal-justice system. Because both schooling and experience are positively correlated with earnings, having a criminal background when young indirectly lowers earnings as an adult. We show, however, that – holding these human-capital variables constant – youth criminal behavior directly reduces subsequent earnings as an adult.

We combine data from the 1980 wave of the National Longitudinal Survey of Youth, which provides detailed, self-reported information on criminal background, with socioeconomic and demographic variables to specify and estimate a model of the determinants of earnings in 1983 and 1989. The results imply that having been convicted prior to 1980 of a crime when young reduces 1983 earnings by at least 12%. However, having been charged - but not convicted - of an offense as a youth has no statistically significant effect on such earnings. A criminal case adjudicated in juvenile court reduces 1983 earnings by at least 9%, while having a charge decided in adult court lowers those earnings by about 14%. The magnitudes of these earnings effects persist over the subsequent six years. 2I. Introduction It is well known that young people are more likely to engage in illegal activity than are older individuals. However, the extent to which illegal behavior engaged in as a youth influences adult socioeconomic outcomes is less clearly understood. For example, does such activity as a youth persistently affect subsequent labor-market opportunities, or are its effects relatively short-lived? Our paper analyzes this relationship by estimating the impact of youth criminal activity on adult labor-market earnings.

Few studies have examined how youth criminal activity affects adult labor-market outcomes. Instead, the literature has focused on how adult criminal activity affects adult outcomes. Previous studies have reached conflicting conclusions about the effect of an adult conviction on subsequent income. Lott (1989, 1992a, 1992b) examined the earnings of adult federal offenders, and concluded that their post-conviction reduction in income is statistically significant and is largest for high-income offenders. He argued that the most important aspect of society’s sanction against criminals is the reduced legitimate earnings of offenders upon their return to the labor force. Waldfogel (1994b) also studied adult federal offenders, and found that a first-time conviction reduced employment probabilities and significantly depressed legitimate income. These effects were largest for offenders whose pre-conviction jobs required trust.

Conversely, several studies have found that the labor-market effects of a criminal background are modest in magnitude and duration. Grogger (1995), using a sample of male arrestees from California, concluded that earnings and employment effects are relatively short-lived, that convictions have little effect on earnings, and that probation has no effect on arrestees' subsequent earnings. Waldfogel (1994a) also addressed the 3persistence of labor-market penalties for criminal participation and found that prior to their current conviction ex-offenders earned less and were less likely to work than first- time offenders. These earnings and employment gaps grew with the number of prior convictions.

Nagin and Waldfogel (1998) maintained that criminal participation increases observed wages shortly after conviction. They argued that conviction reduces access to career jobs offering stable, long-term employment, and relegates offenders to spot-market jobs that have higher initial pay, but do not offer stable employment or steadily rising wages. Consequently, a first conviction has a positive effect on income for those under age 25 and an increasingly negative earnings impact for offenders over age 30. Nagin and Waldfogel (1995) studied about 300 London offenders, and concluded that prior criminality has no effect on job performance, whereas a criminal conviction increases both job instability and pay. This result is consistent with their other findings that conviction increases both the income and employment instability of young offenders.

This study is distinguished from the previous literature in two ways. First, our observations are drawn randomly from the young-adult population. In contrast, other studies have confined attention to labor-market outcomes for offenders.1 If, however, offenders are systematically different from non-offenders, previous results may be affected by this sample-selection bias. Second, the longitudinal nature of our data allows us to examine the extent to which labor-market penalties for previous criminal activities persist over workers' early careers. Most studies have examined the effect on income for only a few (usually no more than three) years after conviction. However, our study 1 Grogger (1992) examined the effect of conviction on employment, and reported results from one regression that used data from non-offenders. 4follows labor-market performance for at least 10 years after data were collected on prior contact with the criminal-justice system.

We find that individuals who were convicted of a crime as youths experience a 12% reduction in earnings when they are young adults, holding constant various human- capital characteristics like education and work experience. However, those who were charged, but not convicted, of a criminal offense when young suffer no reduction in early-career earnings, ceteris paribus. Young adults who had one or more criminal cases adjudicated in juvenile court earned 9% less than their non-offender counterparts, but adjudication in adult court reduces earnings by an additional 5%. These estimated effects are found to persist over the subsequent six years. However, individuals who had contact with the criminal justice system as youths also complete fewer years of schooling and accumulate less work experience as young adults. Because schooling and experience increase future earnings, these estimated partial effects of a criminal background underestimate its total effect on such earnings.

The paper is organized as follows. Section II describes the data. Section III presents the model, and discusses how we control for person-specific heterogeneity.

Section IV reports the empirical results, and Section V concludes.

II. Data We use data on males from the 1980, 1984, and 1990 waves of the National Longitudinal Survey of Youth (NLSY), a stratified random sample of individuals who were between 14 and 22 years old in 1979. The 1980 wave included a special section 5about the respondents' self-reported participation in delinquent and criminal activities.

This section of the survey provides detailed information about each respondent's history of criminal charges and convictions, the nature of any offenses committed, and whether adjudication of a criminal case was in juvenile or adult court. We combine this information with standard demographic and labor-market data to estimate earnings equations augmented by a variety of criminal participation variables. The 1984 and 1990 surveys record labor-market earnings for 1983 and 1989, respectively.

Our empirical work uses two distinct samples: one includes individuals through the 1984 wave of the NLSY, and the second includes individuals through the 1990 wave.

For the first data set we omitted all individuals younger than 21 at the time of the 1984 interview, because many were still in school or just beginning their labor-market experiences.2 Furthermore, observations were deleted for those reporting zero weeks of work or zero income and those responding inappropriately.3 Finally, we deleted people who were students during the week of the interview.4 There are 2897 respondents with complete records for all variables of interest in 1984.

The 1990 data set was constructed by imposing the same restrictions used to create the 1984 data, with the exception of the age restriction. We did not impose an age 2 We also ran, but do not report, regressions that do not impose this restriction. The estimated effects of the criminal-participation variables were slightly larger in these regressions.

3 Missing observations are those defined as REFUSAL, DON’T KNOW, INVALID SKIP, or NONINTERVIEWS. Variables also include the code VALID SKIPS, but this is not necessarily a missing observation. For example, VALID SKIPS for the variables ADLTCRT, NUMCHAR, and NUMCNVC reflect those not charged or convicted of crimes. These valid skips are recoded as zeros. This reduces the sample from 12,686 to 5,400. Of those remaining, 16.7% report having been charged with a crime and 9.9% report having been convicted.

4 This is done using a variable in the NLSY called Employment Status Recode (R15199), which reflects employment status during the week of the interview. Individuals coded “Going to School” were deleted. 6restriction for the 1990 sample because respondents to the survey were not of typical school-going age. There are 3280 respondents with complete records for all variables of interest in 1990. The 1990 sample is larger than the 1984 sample because the age restriction was relaxed. We adjusted 1989 income data to constant 1983 dollars. Table 1 contains the summary statistics for the two samples.

III. Model We estimate the model( )it it i i itV F C Ye b b b a+ + + + = 3 2 0 1 ln(1) where it Y is annual earnings in 1983 or 1989, 0 i C is a set of criminal participation variables for each person i, as of the interview year 1980, i F is a vector of fixed individual characteristics, such as race, ethnicity, age and AFQT5 score, it V is a vector of characteristics that vary over time, such as educational attainment, marriage, work experience, union membership and whether one lives in a Metropolitan Statistical Area, and it e is the individual-specific error term.

We use four alternative measures of youthful contact with the criminal-justice system: (i) a dummy variable indicating whether the individual had been charged with a crime; (ii) a dummy variable indicating whether the individual had been convicted of a crime; (iii) a pair of dummy variables indicating, respectively, whether an individual had been charged but not convicted, and whether he had been convicted; and (iv) a pair of 5 AFQT denotes the normalized score on the Armed Forces Qualification Test, administered in 1980 to over 90% of the NLSY panel, and measures pre-market skills. 7dummy variables denoting whether an individual’s criminal case was adjudicated in juvenile or adult court. We estimate these four specifications for both the 1984 and 1990 samples, and therefore report eight sets of estimates on subsequent adult earnings.

Because characteristics that lead to high wages and employment also reduce participation in criminal activity, estimates that do not control for this heterogeneity will be biased toward finding the expected negative relationship – that youth criminal participation leads to lower earnings. Several papers have attempted to control for heterogeneity in a variety of ways. Grogger (1995) chose a comparison group for the California arrestees comprising his sample to control statistically for any time-invariant, individual-specific, unobservable characteristics. Waldfogel (1994b) and Lott (1992a, 1992b) estimated differences between pre- and post-conviction income as a function of changes in criminal participation.

Unfortunately, because the NLSY records criminal participation only in the initial year (1980), we do not observe changes in criminal participation, and cannot control for unobserved heterogeneity with a fixed-effects, panel-data model. Instead we control for heterogeneity in two ways. First, the NLSY contains an extensive set of demographic variables that allow us to control for many observed individual characteristics. One of these variables, AFQT, is frequently omitted from earnings regressions, and as a proxy for ability captures much of the heterogeneity. Grogger (1995) pursued a similar strategy by incorporating various demographic variables, but he excluded AFQT.6 Second, the full model specification in (1) includes many characteristics over which individuals have 6 Grogger also notes a problem with the NLSY arrest data – blacks and whites have the same number of self-reported arrests on average. In most other samples, however, the arrest rate for blacks is about 3 times that of whites. 8some degree of choice–these are captured in it V above. Because educational attainment, marital status, and work experience are functions of criminal activity, the indirect effect of youth criminal activity on adult earnings is absorbed by the coefficients on these variables. Consequently, the estimate of 1 b in the full specification understates the total effect of youth criminal background on adult earnings.

Our analysis is limited to young adults who reported positive labor-market earnings. However, both Freeman (1991) and Grogger (1992) found that having a criminal record when young reduces the probability of legal employment as an adult.

Consequently, by restricting our sample to employed individuals, we further underestimate the total effect of youth criminal background on adult earnings, inclusive of its effect on employment status.

IV. Empirical Results We begin our empirical analysis by estimating the raw, unadjusted difference in adult earnings between individuals who, when young, had formal contact with the criminal justice system (criminal charges and/or convictions) and those who did not. This estimated difference does not control for either fixed, pre-market traits that affect adult earnings (such as race or ability) or for other human-capital variables (like schooling and work experience) that help determine adult earnings, but also could be affected by youth criminal activity. We obtain this raw difference by estimating a bivariate regression in which the dependent variable is either 1983 or 1989 log annual income.

Table 2 contains ordinary least-squares estimates of four bivariate regressions 9using 1983 log annual earnings as the dependent variable and each of the alternative measures of youth criminal background. Column 1 indicates that individuals who were charged with a crime when young (whether convicted) earned approximately 27% less in 1983, on average, than individuals who were not criminally charged. Of course, because this regression does not control for observed (and unobserved) differences in characteristics that affect earnings, this point estimate is equivalent to a simple difference-in-means. The bivariate regression results reported in column 2 imply that young adults convicted of a crime as youths earned about 29% less in 1983, on average, than those who were not. As expected, the coefficient on having been convicted is larger than the one on having been charged reported in column 1. Column 3 shows that those youths who were charged but not convicted of a criminal offense earned approximately 21% less as young adults than individuals with no criminal charges against them, while persons convicted of crimes when young earned about 31% less as young adults than did those who had no criminal convictions. In column 4, finally, youths whose criminal charges were adjudicated in juvenile and adult court experienced a 27% and 26% decrease, respectively, in 1983 earnings compared with uncharged individuals.

Table 3 replicates the same four specifications for 1989 earnings, and shows the same general results—the coefficients on the criminal sanction variables are uniformly negative and significantly different from zero. The coefficient estimates on being charged and convicted are slightly higher than for 1983 earnings.

An analysis of the effect of youth criminal background on adult earnings must assign to (observable) pre-market characteristics some of the explanatory power for differences in subsequent earnings between youthful offenders and non-offenders. 10Inherent skill (or ability or aptitude), along with ethnicity and age, are important determinants of labor-market earnings that are unaffected by subsequent human-capital investment but may be correlated with criminal behavior when young.

Tables 4 and 5 report least-squares estimates of the effect of our four alternative measures of youth criminal activity, controlling for the pre-market variables, on 1983 and 1989 earnings, respectively. The point estimate in column 1 implies that, holding ethnicity, skill, and age constant, individuals who were charged with a crime when young earned almost 29% less in 1983 than those who were not. The magnitude of the CHARGED coefficient is smaller in this specification than in the simple bivariate mode, because in the latter, the estimated coefficient captures effects on subsequent earnings more properly attributed to the pre-market variables included here. As expected, the estimated coefficient on BLACK is negative and significantly different from zero, and implies that blacks earn about 32% less than whites, holding pre-market skills and age constant. However, this specification is extremely parsimonious, and does not control for variables such as education and work experience that are typically included in earnings regressions and are correlated with race. In contrast, the estimate of the HISPANIC coefficient is small and not significantly different from zero. The estimated coefficients on AFQT and AGE are positive and significantly different from zero, as expected.

Column 2 reports the results of estimating the same specification discussed above, with criminal background now represented by a dummy variable indicating whether one was convicted of a crime as a youth. The coefficient estimate on CONVICTED is positive, significantly different from zero, and somewhat larger than the estimated coefficient on the CHARGED variable reported in column 1. The estimated coefficients 11on the included pre-market variables are virtually identical to those in column 1.

Of course, individuals convicted of a crime when young were also charged with that crime, so it is of interest to separate out the marginal effect on earnings of having been convicted of a youthful crime, given that one has been charged with the crime. The estimates in column 3 indicate that someone who was charged but not convicted earns about 22% less than his uncharged counterpart. However, an individual who was charged and subsequently convicted experienced a 34% reduction in 1983 labor-market earnings.Therefore, the marginal impact of a prior conviction on 1983 earnings is about -11.5% [- 33.9 - (-22.4)], ceteris paribus.

Finally, the data permit us to distinguish between the subsequent earnings effects of a criminal charge adjudicated in juvenile court rather than in adult court. Column 4 reports the empirical results for this specification, and shows that individuals whose criminal cases were handled in juvenile court earned approximately 20% less than those having had no contact when young with the criminal-justice system. However, those youths whose cases were adjudicated in adult court experienced a 36% reduction in 1983 earnings. This large difference in coefficient estimates may reflect one or both of the following phenomena: (i) because of the confidentiality of juvenile-court proceedings, the “scarring” or “signaling” aspects of criminal charges handled in that setting are less than in cases dealt with in open adult court; (ii) youths who commit crimes of such severity that they are tried in adult court are different from their juvenile-court counterparts in ways that adversely affect subsequent labor-market earnings. As before, the 1989 results for the criminal sanction variables are very similar to the 1983 findings.

The results reported in Tables 4 and 5 control only for exogenous pre-market 12variables that, along with youth criminal background, affect the subsequent earnings of young adults. However, the model on which these estimates are based is an under- specified representation of the process determining such earnings. In particular, this model specification excludes variables such as schooling and work experience which proxy human-capital investment affecting earnings as a young adult. To redress this shortcoming, we specify a more complete model of earnings incorporating additional variables that are exogenous to earnings but whose values are determined by choices made after adolescence.

Tables 6 and 7 report the results of this more completely specified earnings model. Because we include both schooling and work experience in this regression and use a sample of males for whom post-schooling work experience is, on average, highly continuous, we excluded age from the estimated regressions. The estimated coefficients on the pre-market variables HISPANIC and AFQT are very similar to those from the more parsimonious specification reported in Table 4. Interestingly, the size of the coefficient on BLACK is reduced by almost three-fifths after controlling for the post- adolescence explanatory variables, suggesting considerable heterogeneity among the black population with respect to these additional observable determinants of earnings.

The signs, sizes, and significance levels of the coefficients on the additional explanatory variables in column 1 conform to standard results reported in the empirical earnings literature. In particular, the coefficients on schooling (grades completed), married, urban residence, and union membership are positive and significantly different from zero. Additional weeks of work experience increase earnings, but at a decreasing rate. Individuals who were charged with a crime when young earned approximately 1311.4% less in 1983 than their non-charged counterparts, and this adverse earnings effect is significantly different from zero. However, the size of the criminal-background discount on adult earnings is lowered by about three-fifths with the inclusion of additional controls for observable influences on adult earnings. We interpret this reduction in the estimated effect of youth criminal background to mean that a portion of the total effect of having been charged when young with a criminal offense is now beingattributed to variables – such as labor-market experience and years of completed schooling – that are affected by adolescent criminal activity. As a consequence, the estimated coefficient on CHARGED is a downward-biased estimate of the true effect of a youthful criminal charge on subsequent earnings. This downward bias offsets to an unknown degree the upward bias in the estimated effect associated with any individual heterogeneity arising from omitted (unobservable) variables that are correlated with both youth criminal background and adult earnings.

In column 2 the point estimate of the CONVICTED coefficient is slightly higher than that on CHARGED, reported in the previous column, and is significantly different from zero. As before, the model specification in column 3 permits us to separate the marginal effect of being convicted when young of a criminal offense from the effect of having been charged but not convicted. The point estimates of the coefficients on both criminal-participation variables are substantially lower than before, again suggesting that the total effects of these variables are being attributed partly to post-adolescent individual characteristics that are, in turn, affected by youth criminal behavior. The evidence from this specification implies that an individual charged with a crime when young experiences about a 9% reduction in earnings as a young adult, ceteris paribus, while the 14marginal effect on earnings of a conviction, having been charged, is -12.8 - (-8.8) = - 4.0%.

Column 4 reports the results of estimating the model with dummy variables indicating adjudication of any criminal case(s) in adult or juvenile court. Again, the point estimate of the coefficient on the adult-court variable is substantially lower than the estimated coefficient on the juvenile-court variable (-0.131 versus -0.095). Moreover, themagnitudes of both coefficients are lower in this estimated regression than in the more parsimonious model reported in Table 4, as expected.

Compared with the 1983 results, the estimated effects on 1989 earnings of being black, living in an urban area, being a union member, previous work experience, and being married are smaller, while the estimated return to schooling is substantially larger.

The coefficient estimates on the variable CHARGED in column 1 across the three tables, show essentially no difference in the magnitudes of the estimated effects on 1983 and 1989 earnings. The point estimate of the effect on 1989 earnings of having been convicted is slightly higher than on 1983 earnings for each of the model specifications.

V. Conclusion We have used data from a stratified random sample of young adults to estimate the effect of youth criminal arrests, charges, and convictions on labor-market earnings as an adult. Individuals charged with or convicted of a criminal offense when young have lower adult earnings because they complete fewer years of schooling and accumulate less work experience than those with no contact as a youth with the criminal-justice system. 15However, we show that youth criminal behavior when young also directly reduces adult earnings, even after controlling for these human-capital variables. Having been charged but not convicted decreases earnings by between 5-8% and having been convicted as a youth permanently lowers adult earnings by at least 12%. Adjudication in a juvenile court lowers adult earnings by at least 9%, while having one’s case adjudicated in an adult court lowers earnings an additional 5%. 16References Freeman, Richard (1991) “Crime and the Employment of Disadvantaged Youths.” NBER Working Paper no. 3875.

Grogger, Jeff (1992) “Arrests, Persistent Youth Joblessness, and Black/White Review of Economics and Statistics, Vol. 74 (February): 100-106.

Grogger, Jeff (1995) “Effect of Arrests on the Employment and Earnings of Young Quarterly Journal of Economics, Vol. 110 (February): 52-71.

Lott, John R. Jr. (1989) “The Effect of Conviction on the Legitimate Income of Economics Letters, Vol. 34, no. 4: 381-385.

Lott, John R. Jr. (1992a) “An Attempt at Measuring the Total Monetary Penalty from Drug Convictions: The Importance of an Individual’s Reputation.” Journal of Legal Studies, Vol. 21, (January): 159-187.

Lott, John R. Jr. (1992b) “Do We Punish High-Income Criminals Too Heavily?” Economic Inquiry, Vol. 30, (October): 583-608.

Nagin, Daniel and Joel Waldfogel (1995) “The Effects of Criminality and Conviction on the Labor Market Status of Young British Offenders.” International Review of Law and Economics, Vol. 15 (January): 109-126.

Nagin, Daniel, and Joel Waldfogel (1998) "The Effect of Conviction on Income Through the Life Cycle." International Review of Law and Economics, Vol. 18 (March): 25-40.

Waldfogel, Joel (1994a) “Does Conviction Have a Persistent Effect on Income and International Review of Law and Economics, Vol. 14 (March) 103-119.

Waldfogel, Joel (1994b) “The Effect of Criminal Conviction on Income and the Trust The Journal of Human Resources, Vol. 29, (Winter):

62-81. 17Table 1 Summary Statistics VariableNumberMeanSt. Dev.Min.Max.1984 DataAge289723.651.762127 Income83289711,2378,2102575001 Black28970.230.4201 Hispanic28970.140.3501 AFQT89289745.0729.90199 SMSA28970.760.4201 Grade289712.462.14220 Married28970.330.4701 Experience (weeks)2897206772312 Experience2 289748,45229,837497344 Union Member28970.210.4101 Charged28970.180.3901 Just Charged28970.090.2901 Convicted28970.110.3101 Adult Court28970.100.2901 Juvenile Court28970.090.2801 1990 data Age3280 Income89328017,96312,13340138204 Black32800.250.4301 Hispanic32800.160.3601 AFQT89328042.8730.39199 SMSA32800.790.4101 Grade328012.922.47320 Married32800.520.5001 Experience (weeks)3280435.31132.203624 Experience2 3280206,963107,0099389376 Union Member32800.200.4001 Charged32800.140.3501 Just Charged32800.070.2501 Convicted32800.080.2801 Adult Court32800.060.2401 Juvenile Court32800.080.2701 18Table 2 The Effect of Criminal Participation on 1983 Wages Bivariate Regression (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.268-5.40Convicted-0.288-4.62-0.308-4.94Just Charged-0.208-3.09Adult Court-0.263-4.03Juvenile Court-0.273-3.97Intercept9.009426.428.991444.109.012422.829.009426.34Num. of Obs.28972897289728972897289728972897F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1983 income.

Standard errors are in parentheses.

Table 3 The Effect of Criminal Participation on 1989 Wages Bivariate Regression (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.279-6.78Convicted-0.325-6.33-0.339-6.59Just Charged-0.184-3.25Adult Court-0.220-3.67Juvenile Court-0.324-6.11Intercept9.594626.439.583644.639.596622.059.594626.50Num. of Obs.32803280328032803280328032803280F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1989 income. 19Table 4 The Effect of Criminal Participation on 1983 Wages with Fixed Factors (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.286-5.99Convicted-0.314-5.26-0.339-5.64Just Charged-0.224-3.49Adult Court-0.362-5.76Juvenile Court-0.202-3.06Black-0.322-6.50-0.315-6.36-0.324-6.56-0.324-6.56Hispanic0.0300.540.0280.500.0260.470.0260.47AFQT0.1215.830.1245.990.1195.730.1195.73Age0.11911.270.11911.240.12011.390.12011.39Intercept6.26725.046.25124.936.24224.946.24224.94Num. of Obs.28972897289728972897289728972897F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1983 income.

Standard errors are in parentheses.

Table 5 The Effect of Criminal Participation on 1989 Wages with Fixed Factors (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.271-7.11Convicted-0.322-6.81-0.336-7.08Just Charged-0.160-3.08Adult Court-0.308-5.48Juvenile Court-0.245-5.02Black-0.162-4.73-0.158-4.61-0.163-4.76-0.162-4.74Hispanic0.0611.610.0591.540.0581.530.0611.59AFQT0.26517.890.26818.190.26517.880.26517.89Age0.0477.940.0457.690.0477.930.04857.98Intercept8.57164.658.59664.908.57564.738.55163.65Num. of Obs.32803280328032803280328032803280F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1989 income. 20Table 6 The Effect of Criminal Participation on 1983 Wages Full Specification (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.114-2.73Convicted-0.117-2.26-0.128-2.46Just Charged-0.088-1.58Adult Court-0.131-2.41Juvenile Court-0.095-1.66Black-0.125-2.81-0.123-2.77-0.126-2.82-0.125-2.81Hispanic-0.024-0.50-0.024-0.51-0.025-0.53-0.024-0.51AFQT0.1245.420.1235.360.1235.400.1245.43SMSA0.1493.980.1453.860.1483.940.1493.97Grade0.0171.730.0191.950.0171.740.0171.73Married0.3259.400.3249.380.3259.400.3259.40Experience0.01212.470.01212.440.01212.480.01212.47Experience2 0.000-6.400.000-6.350.000-6.410.000-6.40Union0.2827.180.2847.250.2827.180.2817.18Intercept6.78344.476.75144.656.78344.476.78244.46Num. of Obs.28972897289728972897289728972897F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1983 income.

Standard errors are in parentheses.

Table 7 The Effect of Criminal Participation on 1989 Wages Full Specification (1)(2)(3)(4)VariableCoeff.T-statCoeff.T-statCoeff.T-statCoeff.T-statCharged-0.117-3.39Convicted-0.140-3.28-0.145-3.36Just Charged-0.049-1.04Adult Court-0.148-2.99Juvenile Court-0.093-2.09Black-0.091-2.86-0.090-2.83-0.091-2.86-0.091-2.85Hispanic0.0371.090.0371.070.0371.060.0371.08AFQT0.1297.360.1297.350.1307.370.1307.38SMSA0.1194.100.1153.940.1163.970.1194.09Grade0.0619.350.0629.560.0619.390.0619.35Married0.2289.260.2299.320.2289.280.2289.26Experience0.00611.930.00611.830.00611.840.00611.96Experience2 0.000-7.580.000-7.490.000-7.500.000-7.60Union0.1615.490.1615.490.1615.490.1615.50Intercept7.01956.387.01156.437.02556.247.01456.31Num. of Obs.32803280328032803280328032803280F-StatisticAdj. R2Notes: Dependent variable is the natural log of 1989 income.