Professional Documents
Culture Documents
Abstract—Recessions lead to short-term job loss, lower happiness, and Quetelet (1831) showed that crime in early nineteenth-
decreasing income levels. There is growing evidence that workers who
first join the labor market during economic downturns suffer from poor century France peaked when individuals were in their late
job matches that can have sustained detrimental effects on wages and teens. Subsequent research has confirmed the strong age-
career progressions. This paper uses U.S. and U.K. data to document a crime pattern, with crime peaking in the late teens and
more disturbing long-run effect of recessions: young people who leave
school during recessions are significantly more likely to lead a life of declining with age quite rapidly.1 Unsurprisingly in our
crime than those entering a buoyant labor market. Thus, crime scars data, the same patterns emerge.2 The peak is at age 17 or 18
resulting from higher entry-level unemployment rates prove to be long and then declines reasonably smoothly. However, it is still
lasting and substantial.
the case that the offender rate at age 29 is a lot higher than
at age 39—showing criminality to be not just a feature of
to uncover a strong and precisely estimated link between labor market. This is true for both self-reported health mea-
youth unemployment and crime for both property and vio- sures and objective measures of physical and mental health.
lent crimes. Giuliano and Spilimbergo (2014) show that those who enter
Second, there is a growing literature on the effects of the labor market in recession are more likely to believe that
entering the labor market during recessions on outcomes success in life depends more on luck than effort and support
later in life. That literature so far has focused on whether more government redistribution. Again, these effects are
such workers experience sustained long-run negative con- seen to be long lasting. The protective effect of education
sequences. Early contributions by Ellwood (1982) and for cohorts who graduate in recessions is studied by Cutler,
Gardecki and Neumark (1998) found somewhat contrasting Huang, and Lleras-Muney (2014) in their analysis of Euro-
evidence on whether initial labor market experience barometer data. They report evidence of lower wages and
affected subsequent outcomes, with Ellwood finding signifi- life satisfaction, together with higher obesity and a greater
cally are less attached to the legal labor market than older involved in criminal activity but is not exposed to these
individuals further on in their careers. conditions for a long enough period to develop sufficient
The model has proved valuable in highlighting the eco- criminal capital in order to yield higher returns in the crime
nomic incentives associated with criminal activity and its market than in the legal market once the recession ends.
basic predictions on incentive and deterrence effects on If an individual is exposed to an unexpectedly long reces-
crime have received substantial empirical support (see the sion, the decision between legal and illegal activity changes
reviews of Draca & Machin, 2015; Freeman, 1999; and in the same way as in a short recession. However, the indi-
Chalfin & McCrary, 2017). Its weakness and limitation for vidual’s criminal human capital stock grows over a longer
our purposes is that it is explicitly static. Individuals make a time period, whereas the legal human capital stock is
one-off decision to commit crime or work in the legal sec- expected to decline even more than in a shorter recession.
tor. There is no process through which decisions made in These two effects may result in higher returns to criminal
their first crime aged 21 or over). They find a significant vior. In effect, the model allows us to estimate the average
effect of high unemployment at ages 16 to 18 as a predictor effect of entering the labor market in a recession on crime
of subsequent offending (relative to a nonoffending control over the life cycle, given the usual pattern of regional
group). To address this in our empirical analysis, we con- unemployment that cohorts experience after entry. For the
sider an approach that is flexible enough to permit differen- focus of this paper, we are arguably more interested in the
tial timing of the effects of labor market entry unemploy- effect of entry unemployment net of subsequent labor mar-
ment effects on crime. ket conditions. To isolate this effect, we can include region-
al unemployment rates that the cohort experiences in the
III. Empirical Strategy years after labor market entry. We measures these as Ucr,i,
where i > 0 is the number of years since entry. This gives
Our empirical analysis exploits both panel data on year- us a second, more general model to estimate:
Our final and most general estimating equation allows for and assault. We produce arrest rates by dividing the number
the unemployment experienced after labor market entry to of arrests by the annual population in the observational unit,
have permanent or transitory effects: and scale by 1,000. Population data are retrieved from the
U.S. Census population estimates.
logðCÞcrt ¼ ac þ ar þ at þ aa The sample covers men aged between 16 and 39, the
X
E group of individuals with the highest crime propensity. The
þ be IðExp ¼ eÞ Ucr;0 original data are grouped by age. Up to the age of 24, the
e¼1 data are reported by single age year, while for ages 25 and
I X
X E above, the data are grouped in five-year age bands (25–29,
þ die IðExp ¼ eÞ Ucr;i 30–34, 35–39). As our empirical strategy exploits year-of-
i¼1 e¼1
birth cohorts, we assume that year-of-birth cohorts within
Second, in our empirical work for the United States, we show that our conclusions are robust to this alternative mea-
use the average unemployment rate that the cohort experi- sure of labor market entry.
enced at ages 16 to 18 as our measure of entry unemploy-
ment. This is motivated by the observation that the majority C. U.S. Results
of arrested criminals have low educational attainment and
generally leave school at or around the compulsory school- We begin our analysis of the panel data by presenting evi-
leaving age. In the U.S. Census data that we use in our dence on the average effect of initial labor market conditions
microdata analysis, 86% of those incarcerated over the on criminal activity. In terms of the equations, this corre-
1980–2010 sample had high school or less (twelve or fewer sponds to equation (1), which restricts the coefficient b to be
years of education) as their highest level of education. Since the same across experience groups. Panel a of table 1 shows
school-leaving ages differ slightly across time and states the results for the United States. The dependent variable is the
and unemployment within a cohort or state observation is log of the crime rate. Columns 1, 3, and 5 consider the national
autocorrelated, we use the age 16 to 18 average unemploy- unemployment rate at labor market entry, while columns 2, 4,
ment rate to characterize the state of the labor market that and 6 use the state unemployment rate at labor market entry
the cohort first experiences. An alternative would be to use (our preferred specification). All regressions include year,
the age 16 (or indeed age 17 or 18) unemployment rate state and age fixed effects, and cohort composition variables.
only. We will show that our results are robust to these alter- The national results control for a quadratic cohort trend, while
native approaches to defining entry unemployment. the state results include a full set of cohort fixed effects. The
regressions are weighted by cell population, and robust stan-
dard errors are clustered at the state-cohort level.
B. Details of U.K. Panel Data Columns 1 and 2 of table 1 show a strong, positive esti-
mated coefficient on the entry unemployment rate, whether
Crime data for the U.K. panel come from the Offenders we use the national or state-level variation in entry
Index Database (OID) and the Police National Computer unemployment. For the state-level entry unemployment rate
(PNC). The measure of crime is convictions. This has the specification in column 2, the average arrest rate for a cohort
advantage of capturing actual offenders (subject, of course, entering the labor market in a recession is estimated to be
to wrongful conviction) rather than the proportion of a par- around 10.2% higher than for a similar cohort entering into a
ticular cohort that come into contact with the police. The normal labor market (using a 5 percentage point increase
OID/PNC provides data on gender, date of birth, region of in unemployment as a measure of recession relative to nor-
conviction, and offense category. This data sample again mal).14 The estimate is statistically significant at the 1% level.
runs from 1980 to 2010. Our variable construction matches
that of the U.S. panel. However, in contrast to the United 14
We have used a 5 percentage point increase to measure a recession.
States, there is a standard national school-leaving age in the This is in line with the scale of the unemployment rate increase in the
Great Recession in the United States and also with the recessions seen in
United Kingdom. For those leaving school by 1972, the the United Kingdom over the whole period we analyze. However, this
compulsory school-leaving age was 15, and from 1973 may be a little high for the United States for pre-2000 recessions, where
onward, it was 16. We use this compulsory age to date labor perhaps a 3 percentage point increase is a more appropriate magnitude of
increase. Of course, the impact on the arrest rate would need to be scaled
market entry for each cohort rather than taking the age 16 down by three-fifths for this recession size, so in the example here, it
to 18 average. However, results are reported that again would drop to 6.1%.
398 THE REVIEW OF ECONOMICS AND STATISTICS
This amounts to a substantial estimate of labor market where p is the proportion of cohort c in state s born in state
entry effects on crime, but in some respects, the average k.
effect of recessions may not be the most relevant parameter Panel B of table 1 reports estimates using this mobility-
of interest. Within a cohort, there will be a substantial share adjusted entry unemployment rate. The results are robust to
for which the marginal effect is 0 since their optimal deci- the new specification, in that a positive and substantial
sion will be unaffected—that is, they are at an interior solu- entry-level unemployment rate effect on crime remains.
tion that results in no illegal behavior and the recession The effect tends to be slightly larger in magnitude (at 2.470
does not move them across the threshold. Thus, the average compared to 2.039 in column 2), but very much validates
effect that we estimate is a combination of a zero effect for the earlier results. Hence, this robustness check offers a use-
potentially a large share of the cohort and a substantial ful corroboration of our main results; if anything, we appear
effect for those close to the legal/illegal threshold in the to marginally underestimate the effect of initial unemploy-
TABLE 3.—U.S. COHORT PANEL ESTIMATES, EFFECTS BY LABOR MARKET EXPERIENCE GROUPS
(1) (2) (3) (4)
Crime Type All All Property Violent
State entry unemployment rate at age 16–18Exp(0–5) 3.609*** 3.290*** 1.481** 5.151***
(0.626) (0.702) (0.717) (1.124)
State entry unemployment rate at age 16–18Exp(6–11) 1.926*** 1.705*** 0.965 2.615***
(0.535) (0.617) (0.737) (0.821)
State entry unemployment rate at age 16–18Exp(12–17) 1.475*** 1.558** 2.151** 0.883
(0.556) (0.643) (0.911) (0.752)
State entry unemployment rate at age 16–18Exp(18–21) 1.515*** 2.421*** 3.345*** 2.032**
(0.566) (0.707) (0.959) (0.859)
State, year, cohort, and age fixed effects X X X X
Compositional adjustment X X X X
TABLE 4.—U.K. COHORT PANEL ESTIMATES, EFFECTS BY LABOR MARKET EXPERIENCE GROUPS
(1) (2) (3) (4)
All All Property Violent
Region entry unemployment rate at age 16Exp(0–5) 0.861*** 0.967*** 0.972*** 1.034*
(0.305) (0.316) (0.362) (0.532)
Region entry unemployment rate at age 16Exp(6–11) 0.913*** 0.970*** 1.050*** 0.996**
(0.284) (0.298) (0.365) (0.444)
Region entry unemployment rate at age 16Exp(12–17) 0.832** 0.809** 0.733 1.369***
(0.343) (0.358) (0.448) (0.435)
Region entry unemployment rate at age 16Exp(18–23) 0.582 0.529 0.124 2.701***
(0.369) (0.405) (0.502) (0.504)
Region, year, cohort, and age fixed effects X X X X
Compositional adjustment X X X X
Subsequent unemployment interactions X X X
Sample size 7,440 7,440 7,440 7,440
Exp: experience. As for columns 2, 4, and 6 specifications of panel a, table 2. Significant at *10%, **5%, and ***1%.
key result of column 3 is that none of the subsequent three- We group experience into four categories (0–5, 6–11,
year average unemployment rates that the cohort experi- 12–17, and 18–21 years) and use an identical regression
ences has an individually significant effect on arrests. This specification as in table 1. Experience is set to 0 for ages 16
helps us to better understand a puzzle in the literature that to 18. The results are shown in tables 3 and 4 for the United
we referred to in section I: the overall link between crime States and United Kingdom, respectively, with columns 1
and unemployment appears quite weak in many studies. and 2 showing results for all crimes and columns 3 and 4
Our results show that the key effect from unemployment on for property and violent crime, respectively. Column 1 is
a cohort’s crime trajectory is the early experience of unem- estimated without controlling for subsequent unemploy-
ployment rather than the average unemployment experi- ment, whereas columns 2 to 4 allow for these effects inter-
enced over the life cycle. acted with experience dummies—equation (4). There are
Tables 1 and 2 demonstrated a statistically significant strong, positive effects of entry unemployment on arrests in
and economically substantial effect of initial unemployment the early years in the labor market that fall as experience
conditions on the arrest rates of cohorts over their entire increases. However, even a decade after leaving school,
lifetime. But we are also interested in the persistence of there remain significant positive effects from entry unem-
this. Is the entry unemployment effect primarily driven by a ployment on crime rates. Juveniles who leave school in a
very large impact on crime in the early years after labor recession have higher crime rates during their first few
market entry that subsides as the young age and go on to years in the labor market and higher crime rates over a
establish a stable legal career? Or is the effect persistent, decade later than juveniles who leave school in a buoyant
with some of those affected by harsh labor market condi- economy.
tions at labor market entry pushed into a criminal career An alternative specification to examine the persistence of
that becomes self-perpetuating for the reasons discussed in entry unemployment allows the interaction term with
section II? To study this, we allow the coefficient on initial experience to vary with individual years of experience
unemployment to vary by years since labor market entry as (rather than group experience as in tables 3 and 4). Figures
outlined in equation (3) of section III. 1 and 2 plot the estimated coefficients (together with the
CRIME SCARS 401
FIGURE 1.—ENTRY UNEMPLOYMENT EFFECTS BY EXPERIENCE, UNITED STATES lized population using the Group Quarters variable con-
Crime and Entry Unemployment Effects by Experience, US tained in these data sources. However, only in the 1980
sample is that variable available at a detailed enough level
8
among the younger ages we focus on, the share of the total
2
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Years of Potential Experience covariates from the Census data include race, marital status,
Derived from specification comparable to that in column 1 of table 3, with separate estimates for each
veteran status, and education.
year of experience.
Derived from specification comparable to that in column 1 of table 4, with separate estimates for each
year of experience. C. U.S. Results
Panel A of table 5 reports marginal effects from a probit
95% confidence intervals) for every year of labor market regression on the Census incarceration data. We use state at
experience for the United States and United Kingdom, birth to identify the state in which the individual went to
respectively.20 Again, the drop in the effect after the first school (Dahl, 2002) and so restrict the data to those born in
few years of labor market entry is clear, but the individual- the United States. Column 1 reports the results for the full
year estimates suggest a consistent longer-lasting scarring sample of males aged 18 to 39, and the subsequent three col-
effect for both countries. umns focus on samples defined by educational attainment.
All regressions include a full set of year, state of residence,
state of birth, and cohort effects, a quartic in age, and con-
V. Individual-Level Evidence trols for race, education, marital status, and veteran status.
The estimated coefficient on entry unemployment in col-
A. Details of U.S. Microdata umn 1 is 0.038. The mean of the dependent variable is
0.030 (i.e., 3% of males aged 18 to 39 are incarcerated).
The microlevel data on U.S. incarceration of individuals
Thus, entering the labor market in a time of recession
come from the U.S. decennial Census and American Com-
(defined as the unemployment rate being 5 percentage
munity Survey (ACS) data. We study all males aged 18 to
points higher than normal) results in a 6.3% increase in the
39 from the 5% samples of the 1980, 1990, and 2000 Cen-
probability of being incarcerated at the time of subsequent
sus and the 2008–2012 ACS from IPUMS-USA (Integrated
Census survey dates. However, we can see from the subse-
Public Use Microdata Series). We identify the institutiona-
quent columns that this effect is almost entirely due to the
20 high school dropouts. A recession increases this group’s
Figures A4 and A5 in the online appendix show the respective results
when one allows unemployment rates later in life to enter the regression. probability of incarceration by 6.6% from an already high
The figures show patterns that are substantively the same. mean of 8.9%. These are sizable effects, taking into account
402 THE REVIEW OF ECONOMICS AND STATISTICS
that this is averaged over more than twenty years of the so allow us to precisely date the year of labor market entry.
individual’s postschool experience. The data also provide an extensive set of personal charac-
Finally, columns 3 and 4 show only weak effects for teristics, which we would expect to be correlated with crim-
those who successfully graduate from high school and no inal activity. There are two key disadvantages in using these
effect at all for those with four years of college—who microdata. First, there is the usual concern associated with
should, of course, not be affected by the unemployment rate the self-reporting of arrests. In the context of this study,
at the compulsory school-leaving age.21 The results in panel however, this would bias our estimates only if the self-
B show that redefining the 1980 measure of incarceration reporting probability varied within a cohort depending on
by explicitly excluding those not in correctional facilities the initial entry unemployment rate. It seems hard to us to
(see the online data appendix for discussion) does not alter make such a case. Second, we have no information of when
the conclusions. This suggests that policy focused explicitly the arrest occurred; the question is simply whether the indi-
on the least educated during periods of high unemployment vidual has ever been arrested. That means that these data
would likely reduce crime substantially more within that allow us to estimate the average impact of initial entry
group than the average estimate from the previously reported unemployment on the probability of being arrested in adult-
panel regressions would imply. hood but do not allow us to investigate the time pattern of
the persistence of such effects.
We estimate probit models with the dependent variable
D. U.K. Results taking the value 1 if the respondent reports having ever
been arrested by the police. We include survey year dum-
For the United Kingdom, we look at individual-level data
mies and an extensive set of personal controls. Table 6
on self-reported arrests. The data provide information on
reports the results. Column 1 shows an estimated significant
the age at which the respondent left full-time education and
positive coefficient on the entry unemployment rate: a
21 recession (again defined as a 5 percentage points higher
To further examine the sensitivity of our results to mobility, we
restricted the Census sample in table 5 to include only those observations than normal unemployment rate) is associated with a 5.7%
where the state of residence was the same as the state of birth. Reestimat- increase in the probability of ever being arrested.
ing table 5 with this restricted sample produces estimated coefficients In the second column, we restrict attention to those
(and associated standard errors) of 0.047 (0.024) and 0.161 (0.074) for the
first two columns. The coefficients in the final two columns remain small whose highest educational qualification was achieved at age
in magnitude and statistically insignificant. 16 and therefore definitely left education at age 16. Here,
CRIME SCARS 403
we can more closely link the exit from education and the inality that have occurred in both countries in the past few
initial unemployment rate, which results in a sample that is decades, the evidence of crime scars does demonstrate a
likely to contain a larger fraction of individuals at risk of rather more disturbing long-run effect of recessions. In
criminal behavior. As expected, we find a substantially lar- doing so, it also adds to the research picture that the state of
ger and more significant impact of entry unemployment for the business cycle when people leave school and enter the
this group: a recession raises the probability of ever being labor market can have profound and sizable impacts on eco-
arrested by 8%. nomic and social outcomes across their life.
In the final column, we conduct a placebo-type experi-
ment. We examine the arrest record of individuals who
report educational qualifications that require school atten- REFERENCES
dance at least to age 18. This group should not have been Becker, Gary S., ‘‘Crime and Punishment: An Economic Approach,’’
Hansen, Kirstine, ‘‘Education and the Crime-Age Profile,’’ British Journal Oreopoulos, Philip, Till von Wachter, and Andrew Heisz, ‘‘The Short-
of Criminology 43 (2003), 141–168. and Long-Term Career Effects of Graduating in a Recession,’’
Hershbein, Brad J., ‘‘Graduating High School in a Recession: Work, Edu- American Economic Journal: Applied Economics 4 (2012), 1–
cation and Home Production,’’ B. E. Journal of Economic Analysis 29.
and Policy 12 (2012), 1–32. Oyer, Paul, ‘‘Initial Labor Market Conditions and Long-Term Outcomes
Hirschi, Travis, and Michael Gottfredson, ‘‘Age and the Explanation of for Economists,’’ Journal of Economic Perspectives 20 (2006),
Crime,’’ American Journal of Sociology 89 (1983), 552–584. 143–160.
Kahn, Lisa, ‘‘The Long-Term Labor Market Consequences of Graduating ——— ‘‘The Making of an Investment Banker: Macroeconomic Shocks,
from College in a Bad Economy,’’ Labor Economics 17 (2010), Career Choice and Lifetime Income,’’ Journal of Finance 63
303–316. (2008), 2601–2628.
Lochner, Lance, ‘‘Education, Work and Crime: A Human Capital Pfaff, John, ‘‘The Myths and Realities of Correctional Severity: Evidence
Approach,’’ International Economic Review 45 (2004), 811–843. from the National Corrections Reporting Program,’’ American Law
Lochner, Lance, and Enrico Moretti, ‘‘The Effect of Education on Crime: and Economics Review 13 (2011), 491–531.
Evidence from Prison Inmates, Arrests and Self-Reports,’’ Ameri- Piquero, Alex, David Farrington, and Alfred Blumstein, ‘‘The Criminal
can Economic Review 94 (2004), 155–189. Career Paradigm,’’ Crime and Justice 30 (2003), 359–506.