ArticlePDF Available

Why Are Criminals Less Educated than Non-Criminals? Evidence from a Cohort of Young Australian Twins

Authors:

Abstract and Figures

Many studies find a strong negative association between crime and education. This raises the question whether crime reduces investment in human capital or whether education reduces criminal activity. This article investigates posed question by using fixed-effect estimation on data of Australian twins. We find early arrests (before the age of 18) both to have a strong effect on human capital accumulation, as well as strong detrimental effects on adult crime. Schooling does not have an effect on adult crime if there is variation in early arrests. However, schooling reduces crime if there is little variation in early crime. (JEL code: I2, K42).
Content may be subject to copyright.
CPB Discussion Paper
No 114
November 2008
Why are criminals less educated than non-
criminals?
Evidence from a cohort of young Australian twins
Dinand Webbink, Pierre Koning, Sunčica Vujić, Nick Martin
(QIMR Brisbane)
The responsibility for the contents of this CPB Discussion Paper remains with the author(s)
2
CPB Netherlands Bureau for Economic Policy Analysis
Van Stolkweg 14
P.O. Box 80510
2508 GM The Hague, the Netherlands
Telephone +31 70 338 33 80
Telefax +31 70 338 33 50
Internet www.cpb.nl
ISBN 978-90-5833-382-7
3
Abstract in English
Many studies find a strong negative association between crime and education. This raises the
question whether crime reduces investment in human capital or whether education reduces
criminal activity. This paper investigates this question by using fixed effect estimation on data
of Australian twins. We find that early arrests (before the age of 18) have a strong effect on
human capital accumulation. In addition, we find that education decreases crime. However,
controlling for early arrests and early behaviour problems reduces the estimated effect of human
capital on crime to less than on third of the previously estimated association. From this, we
conclude that the strong association between human capital and crime is mainly driven by the
effect of early criminal behaviour on educational attainment. The strong detrimental effects of
early criminal behaviour become also transparent if we consider the estimated effects of early
arrests on three measures of crime. We find large effects of early criminal behaviour on
participation in crime later on. This suggests that programs that succeed in preventing early
criminal behaviour might yield high social and private returns.
Key words: Education, crime, causal effects
JEL code: I2, K42
Abstract in Dutch
In veel studies is een negatieve samenhang gevonden tussen onderwijs en criminaliteit. Dit
roept de vraag op of criminaliteit leidt tot het volgen van minder onderwijs of dat onderwijs
leidt tot minder criminaliteit. Deze studie onderzoekt deze vraag door gebruik te maken van
gegevens van Australische tweelingen en door rekening te houden met genetische en sociaal-
economische factoren die gedeeld worden door tweelingen. Arrestaties op jonge leeftijd (vóór
het 18e jaar) hebben een sterk effect op het bereikte onderwijsniveau. Bovendien vinden we dat
onderwijs leidt tot minder criminaliteit. Echter, het effect van onderwijs op criminaliteit daalt
met meer dan tweederde als rekening gehouden wordt met arrestaties op jonge leeftijd en
antisociale gedragsstoornissen. Dit betekent dat de sterke samenhang tussen onderwijs en
criminaliteit grotendeels bepaald wordt door het effect van arrestaties op jonge leeftijd op
onderwijs. Ook vinden we dat vroeg crimineel gedrag een sterk effect heeft op crimineel gedrag
op latere leeftijd. Dit suggereert dat programma’s die erin slagen crimineel gedrag op jonge
leeftijd terug te dringen, hoge private en sociale opbrengsten kunnen genereren.
Steekwoorden: Onderwijs, criminaliteit, causale effecten
Een uitgebreide Nederlandse samenvatting is beschikbaar via www.cpb.nl.
4
5
Contents
Summary 7
1 Introduction 9
2 Previous studies 11
3 Empirical strategy 13
4 Description of data 15
5 The effect of early arrests on educational attainment 21
6 The effect of human capital on crime 25
7 Robustness 29
7.1 Missing values due to the routing of the questionnaire 29
7.2 Measurement error 31
8 Decomposing the association between crime and education 33
9 Conclusions and discussion 35
References 37
Appendix 41
7
Summary
This paper aims at disentangling the strong association between human capital and crime by
investigating whether crime reduces investment in human capital or whether education reduces
criminal activity. Heretofore, we exploit two aspects of the Australian survey data on education
and crime we use. First, as the data are obtained from twins, we are able to control for many
unobserved characteristics affecting both criminal behaviour and the schooling decisions.
Second, as criminal behaviour is measured over different periods of time – prior to and after
senior high school completion – we can address the causality between crime and education as
well. As early criminal behaviour may affect human capital formation, and human capital may
influence criminal behaviour in later stages of life, we follow a two step analysis.
First, we address the effects of early criminal behaviour on educational attainment. The
estimates suggest that early criminal behaviour is detrimental to investment in human capital.
Within pairs of twins we find that early arrests (before the age of 18) reduce educational
attainment with .7 to .9 years and lower the probability of completing senior high school with
20 to 23 percentage points. In addition, the timing of the early arrest matters, arrests at age 13,
14 or 15 are most detrimental for educational attainment. These estimates are found after
controlling for conduct disorder and early school performance.
Second, we focus on the effect of human capital on crime. As early criminal activity might
be an important confounder, we control for early arrests. The estimates suggest that human
capital has a negative effect on crime. Completing senior high school reduces the probability of
incarceration with 2 to 3 percentage points. We find similar but statistically insignificant effects
on the probability of being arrested since the age of 18 and on the number of arrests. The size of
these estimates might be downward biased because of measurement error in schooling. IV-
estimates using a second independent measure of schooling suggest that the effect of human
capital might be larger. Lochner and Moretti (2004) report IV-estimates of the effect of high
school completion on imprisonment of 8 percentage points for blacks and 0.9 for whites.
When combining these findings, it seems that the causality between human capital and
crime runs in both directions. Still, the impact of early criminal behaviour on human capital
formation dominates the impact of human capital formation on future crime behaviour.
Controlling for early arrests and early behaviour problems reduces the estimated effect of
human capital on crime to less than a quarter of the previously estimated association. From this,
we conclude that early criminal behaviour explains most of the association between human
capital and crime.
The strong detrimental effects of early criminal behaviour become also transparent if we
consider the estimated effects of early arrests on all three measures of crime. Early arrests
increase the probability of incarceration with 20 percentage points and the probability of being
arrested since the age of 18 with 10 percentage points. These effects are much larger than the
estimated effects of human capital. For instance, the estimated effect of being arrested before
8
the age of 18 on incarceration is almost ten times higher than the estimated effect of completing
high school.
In line with previous studies (Lochner and Moretti, 2004, Machin and Vujic, 2006) our
findings suggest that policies that succeed in raising investment in human capital might reduce
crime. However, the (direct) returns to polices that succeed in preventing early criminal
behaviour might be much larger. The estimated effects of early criminal behaviour and conduct
disorder stress the importance of the early stages of life for preventing crime. Programs that
keep children on ‘the right track’ not only may yield high private returns but also may yield
high social returns through their impact on crime reduction. Studies on the effects of effective
early schooling programs in the US show that these program have large social returns mainly
through their impact on preventing crime (Carneiro, et. al, 2003).
Our main conclusion is that the strong association between human capital and crime is
mainly driven by the effect of early criminal behaviour on educational attainment. This finding
based on within-twin estimation confirms one of the main conclusions from a synthesis of the
literature on the causes of crime: ‘We must rivet our attention on the earliest stages of the life
cycle, for after all is said and done, the most serious offenders are boys who begin their criminal
careers at a very early age.” (Wilson and Hernstein, 1985, cited in Dilulio, 1996).
9
1 Introduction
Many studies document a strong negative association between education and crime. For
instance, in the US two-thirds of all incarcerated men in 1993 had not graduated from high
school (Freeman, 1996). Studies that use self-reported and (administrative) arrest data find large
differences in property and violent crime across education groups (Tauchen et al. 1994,
Lochner, 2004). However, the relationship between crime and education is not straightforward.
Does crime reduce investment in human capital or does education reduce criminal activity?
This paper studies the relationship between human capital and crime using data of a sample
of young Australian twins. We exploit two aspects of the Australian survey data on education
and crime. First, as the data are obtained from fraternal and identical twins, we are able to
control for many unobserved characteristics affecting both criminal behaviour and schooling
decisions. Second, as criminal behaviour is measured over different periods of time – prior to
and after senior high school completion – we can address the causality between crime and
education as well. As early criminal behaviour may affect human capital formation, and human
capital may influence criminal behaviour in later stages of life, we follow a two step analysis.
First, we study the relationship between early crime and the accumulation of human capital.
In particular, we estimate the effect of arrests before the age of 18 on educational attainment by
using within-twin estimation. In addition, we investigate whether the timing of the arrest
matters for educational attainment. Second, we estimate the effect of educational attainment on
three measures of crime: incarceration, arrests since the age of 18 and number of arrests. As
early criminal behaviour might be an important confounder in the estimation, we control for
early arrests and measures of conduct disorder within pairs of twins.
Our paper contributes to the economic literature on the relationship between education and
crime in several aspects. First, the empirical economic literature on human capital and crime
that takes unobserved factors into account is limited. Two previous studies use arguably
exogenous variation in human capital to investigate the effect of education on crime (Lochner
and Moretti, 2004; Machin and Vujic, 2006). Both studies use changes in compulsory schooling
laws as an instrument for educational attainment, so as to find that education reduces crime. We
add to this literature and use an identification strategy that has not been applied before – that is,
we exploit the longitudinal nature of our data so as to estimate the relationship between human
capital and crime in both directions. Second, we investigate the effect of early criminal
behaviour on investment in human capital while controlling for fixed effects within pairs of
twins. We are not aware of studies in the economic literature that estimate the causal effect of
early criminal activity on educational attainment. Third, there is growing interest in the
economic literature for the effects of early conditions in life on adult outcomes (Currie and
Stabile, 2006, 2007; Borghans, et. al, 2008). Our paper addresses similar issues.
We find early arrests (arrests before the age of 18) to have a strong effect on human capital
accumulation. In particular, early arrests reduce educational attainment with .7 to .9 years of
10
education and lower the probability of completing senior high school with 20 to 23 percentage
points. These effects are largely driven by the timing of the early arrest; arrests at age 13, 14 or
15 are most detrimental for educational attainment. We also find human capital to reduce crime.
Completing senior high school reduces the probability of incarceration with 2 to 3 percentage
points. Similar but statistically insignificant effects are obtained for the probability of being
arrested since the age of 18 and for the number of arrests. When controlling for early arrests and
early behaviour problems, the estimated effect of human capital on crime reduces to less than a
quarter of the previously estimated association. The strong detrimental effects of early criminal
behaviour become also transparent if we consider the estimated effects of early arrests on all
three measures of crime. We then find large effects of early criminal behaviour on participation
in crime later on. These effects are much larger than the (isolated) impact of human capital on
crime.
We conclude that the strong association between human capital and crime is mainly driven
by the effect of early criminal behaviour on educational attainment. Programs that succeed in
preventing early criminal behaviour might yield high social and private returns.
11
2 Previous studies
The major difficulty in studying the relationship between human capital and crime is that both
variables are driven by a multitude of unobserved factors. For instance, a person’s level of
schooling is typically not randomly determined but the result of individual choices and ability.
These individuals might also have unobserved factors that prevent them from committing
crimes. Unobserved factors that are both correlated with the decision to invest in human capital
and the decision to participate in crime will confound the empirical relationship between
education and crime. As such, OLS estimates of the effects of human capital on crime or OLS
estimates of the effects of crime on human capital are likely to be biased.
The first part of this paper focuses on the effect of early criminal behaviour on human
capital formation. To our knowledge there are no previous economic studies that empirically
estimate the effect of early crime on investment in human capital while taking unobserved
factors into account. Related studies can be found in health economics. Some recent studies
investigate the effect of childhood mental health problems such as ADHD, aggression, anti-
social behaviour and depression on human capital accumulation later in life (Le et al., 2005;
Slade & Wissow, 2006; Currie & Stabile, 2006, 2007; Fletcher & Wolfe, 2007). These studies
typically find large negative effects of childhood mental health problems on educational
attainment. Another related literature focuses on the importance of cognitive and non-cognitive
skills for labour market outcomes and social behaviour (Carneiro & Heckman (2003), Heckman
et al. (2006), Heckman & Masterov (2007), Borghans et al. (2008)). These studies stress the
importance of skills development early in life for human capital accumulation and success later
in life. Early schooling programmes, like the Perry Preschool Programme (PPP), the Syracuse
Programme (SP) or the Head Start Programme (HSP) have proven to be highly effective in
reducing criminal activity, promoting socioeconomic skills, and integrating disadvantaged
children into mainstream society (see for instance Schweinhart et al. 1993; Donohhue &
Siegelman,1998; Lally et al. 1988; and Garces et al. 2002). These social, motivational, and
emotional skills affect performance in school and in the workplace. Programmes that aim at
intervening in the lives of children in their teenage years only attempt to redress the damage of
bad childhoods (Carneiro & Heckman (2003)).
The second part of this paper studies the causal effect of human capital on crime. So far,
only two papers in the economic literature try to establish a causal relationship between
education and crime (Lochner & Moretti (2004), Machin & Vujić (2006)). Both studies use
changes in compulsory school leaving age laws in order to account for the endogeneity of
schooling decisions. Using US Census data Lochner & Moretti (2004) show that one more year
of schooling reduces the probability of incarceration by 0.37 percentage points for blacks, and
0.10 for whites. They corroborate these results using FBI Uniform Crime Reports (UCR) data
for different types of offences, and conclude that the greatest impacts of graduation are
associated with murder, assault, and motor vehicle theft. The authors also calibrate the social
12
savings from crime reduction associated with completing secondary education. They show that
a 1% increase in male high school graduation rates would yield $1.4 billion dollars in social
benefits in 2004 dollars. Machin & Vujić (2006) study the relationship between crime and
education using two British data sources and making use of the raisings of the school leaving
age that occurred in Britain in 1947 and 1973. These data sources are twofold: individual-level
data on imprisonment from the 2001 Census, as well as cohort-level panel data on offending
rates from the Home Office Offenders Index Data (OID) in the period from 1984 to 2002. The
main finding is that schooling significantly reduces imprisonment rates and property crime
offending. As mentioned before, these two studies use an instrumental variable approach and
typically estimate a local treatment effect for the particular subgroup of the population that is
affected by the instrument (a change in compulsory schooling). We expect that this subgroup
consists of those at the lower end of the education distribution. Our approach (see next section)
uses variation over the whole distribution of education which may bring the advantage that our
estimates are applicable to a broader population. Theoretical work on the relationship between
human capital and crime has been done by Lochner (2004). He developed a model of crime in
which human capital increases the opportunity costs of crime. The model predicts that older,
more intelligent and more educated adults should commit fewer street (unskilled) crimes. It also
expected that white collar crime should decline less with age and education than unskilled
crime. These predictions receive broad empirical support in self-report data from the US.
13
3 Empirical strategy
In this paper, we use variation within pairs of twins for studying the relationship between
education and crime. Obviously, the advantage of twin data is that many (unobserved) variables
that twins share – like socioeconomic background and family factors – can be controlled for.
Within twin estimation has been used in several studies on the returns to schooling (see for
instance, Ashenfelter & Krueger, 1994, Miller, et al. 1995) and recently on the effect of parents’
education on the education of their children (Behrman and Rosenzweig, 2005).
In order to get a full picture of the relationship between human capital formation and
criminal behaviour, our estimation strategy consists of two steps. First, we focus on the
relationship between early criminal behaviour and educational attainment. Early criminal
behaviour is measured as the event of being arrested before the age of 18. It is likely that these
early criminal activities occur during the time that the accumulation of human capital is still in
progress because compulsory schooling laws force individuals in Australia to attend schooling
until the age of 15 to 17, depending on the State of residence. For estimating the effect of early
arrests on educational attainment we use the usual linear (probability) model for within-family
estimation:
ijjijijij fXAS
εγβα
++++= 17 (3.1)
where ij
S is the educational attainment of individual i in family j, 17
ij
A is a dummy for being
arrested before the age of 18, ij
X a vector of covariates, j
f is an unobserved family effect
common to all twins in family j and ij
ε
is a random error term. In this model the family fixed
effect, which consists of all shared socioeconomic and genetic factors, is removed by
differencing between twins. In equation (1), we expect that the causality primarily runs from
early arrests towards educational attainment, as early arrests occur before the completion of
schooling. We argue that we can largely control for reverse effects – i.e. bad school
performance driving kids to start criminal activities – by including several measures of early
school performance as additional controls. Moreover, we control for differences in early
behaviour within pairs of twins by including an indicator of conduct disorder (see next section).
The second part of our analysis addresses the effect of human capital on crime since the age of
18, which is usually the perspective that is taken in the literature. The model we estimate is very
similar to equation (1):
ijjijijijij fAXBSC
εδγα
+++++= 1718 (3.2)
with 18
ij
C is criminal activity since the age of 18. As early criminal activity is likely to be an
important confounder for the estimated effect of human capital on crime, we include early
14
arrests as an additional control. We argue that these lagged arrests can be treated as exogenous
variables.
Obviously, the twin setup – together with the use of lagged information – helps us to cancel out
many possible sources of endogeneity. Still, there are two important concerns in the use of
within-twin estimation (Bound & Solon, 1999) that need to be addressed to check the
robustness of our results. First, measurement error in (self-) reported schooling (or crime) may
bias the estimates towards zero (‘attenuation bias’). A solution for this problem has been
introduced by Ashenfelter and Krueger (1994). They obtained two measures of schooling of a
twin by asking the twins to report both on their own schooling as on the schooling of their
sibling. The second measure of schooling can then be used as an instrument to correct for
measurement error. This approach has been used in several studies (for instance Miller et al.
1995, Behrman and Rosenzweig, 2005). In these studies, the size of the estimated effects
increases after instrumenting for measurement error. This paper follows the same approach to
address any attenuation biases.
The second concern in within-twin models is with respect to endogeneity bias within twin
pairs. Although (identical) twins share many genes and were raised in the same social
environment, they are not exactly identical. Bound and Solon (1999) show that the bias in the
within-family estimator may not always be smaller than the bias in the cross-sectional
estimator. This depends on the importance of the fixed family component in the unobservables.
We address this possible bias by using additional controls in the within-twin models, such as
conduct disorder and early arrests.
15
4 Description of data
We use data from the so-called younger cohort of twins of the Australian Twin Register (ATR).
The ATR data were gathered in two surveys, in 1989-1990 and in 1996-2000. In 1980-1982 a
sample of 4,262 twin pairs, born between 1964 and 1971, were registered with the ATR as
children by their parents in response to media appeals and systematic appeals through the
school system. The data were collected in two surveys among this sample of twins. In 1989-
1992, when the twins were 18-25 years old, the first survey by mailed questionnaire was
conducted, called Alcohol Cohort 2. The response rate of this questionnaire survey was 63%. In
1996-2000, the second survey was launched, called TWIN89. For this survey, telephone
interviews were completed with 6,267 individuals, 2,805 men (889 complete and 1,027
incomplete pairs) and 3,462 women (1,215 complete and 1,032 incomplete pairs), who were 30
years old on average (range from 24 to 39) at the time of the interview. The individual response
rate for this telephone interview was 86%.
The surveys gathered information on the respondent’s family background (parents, siblings,
marital status, and children), socioeconomic status (education, employment status, and income),
health behaviour (body size, smoking and drinking habits), conduct disorder, personality,
feelings and attitudes. Zygosity was determined by a combination of diagnostic questions plus
blood grouping and genotyping.
The measures of crime used in the analysis are self reported data on arrests and
incarceration. The survey contains questions on the age of first and last arrest, the number of
arrests and incarceration. The questions explicitly exclude arrests for traffic violations, drunken
behaviour or drunk driving. The question on incarceration excludes time spent in jail for using
drugs or alcohol.
The reliability of these self-report data is an important issue. In criminology, the use of self
report data is well established. Self-report has been the dominant technique used for measuring
criminal behaviour since its introduction in the 1950s by Short and Nye (1957). A large
literature shows that self-report data have consistently acceptable reliability and validity. Many
studies find high correlations of self-report data with other criterion related measures of
criminal frequency and arrest histories (Farrington, 1973; Hardt & Hardt, 1977; Horney &
Marshall, 1992; Huizinga & Elliott, 1986; Maddux & Desmond, 1975; Mieczkowski, 1990;
Weiss, 1998). Thornberry and Krohn (2000) conclude that ‘self-reported measures of
delinquency are as reliable as, if not more reliable than, most social science measures’. A recent
study among street-drug users recruited in 11 cities throughout the United States revealed that
lifetime arrest and incarceration items demonstrated good to excellent reliability (Fisher et al.
2004). In addition, it has been shown that substance abuse factors and mental illness factors did
not affect the quality and accuracy of self-reported arrest history (Nieves et al. 2000).
Educational attainment was measured in the first survey using a seven point scale and
translated into years of education (Miller et al., 1995). The second survey of the younger cohort
16
uses an eight point scale which we also translate into years of education (Miller, et al., 2006).
We prefer to use this more recent measure, as it contains less missing values for our main
estimation sample.
As covariates we use mothers and fathers education and age. In addition, we control for
conduct disorder and early school performance. Our data contains self-reported information on
21 statements that reflect behavioural problems before the age of 18 (see Table A.1). In the
second survey the twins were asked to reflect on their experiences before the age of 18. We
constructed a measure of conduct disorder by summing occurrences of these 21 statements (see
Vujic et al. 2008). This approach is similar to Currie and Stabile (2007) who use 6 questions to
form a conduct disorder scale. The survey contains four questions on early school performance.
Marks in primary and secondary education were measured using a three point scale: better than
average, average and below average. Respondents were also asked about the teacher’s view on
their school achievements: did as well as could, could have done much better, don’t know.
Finally, grade repetition was measured.
In our total sample of 6267 individuals, 70 twins reported having spent time in jail and 340
twins reported having been arrested, which is 1.1% or 5.4% of our sample. Approximately 10%
of male twins and 2% of female twins reported having been arrested. A direct comparison with
population statistics is complicated because of differences in reporting measures. Statistics on
alleged offenders in Australia for 1995 to 2005 show that among males aged 15-19
approximated 9 to 13% gets arrested and among females 2 to 3% (Australian Institute of
Criminology, 2007). For individuals aged 20-24, the rates drop to 6 to 9% for men and 2% for
women, for individuals older than 24 the rates drop further to approximately 1%. It should be
noted that the number of alleged offenders does not equal the number of distinct offenders
during a year because police may take action against the same individual for several offences,
or the individual may be processed on more than one occasion for the same offence type. In
addition, we might expect that many of those arrested since the age of 20 will be recidivists. As
such, a direct comparison of the arrest rates found in our data with population statistics is
difficult. However, the difference between males and females seem in line with the population
statistics. In addition, the total arrest rates in our sample do not seem implausibly high or low.
The sample we use in the main estimations consists of pairs of twins with information on
educational attainment and criminal participation. If this information is missing for one or both
of the twins, we dropped the complete pair. In these samples, 47 twins reported having spent
time in jail and 224 twins reported having been arrested. This includes 6 twin pairs (12 twins)
who both report having spent time in jail and 28 twin pairs (56 individuals) who both report
having been arrested. Our data contain information on the zygosity of the twins, enabling us to
distinguish fraternal and identical twins. However, we only focus on the sample of all twins,
including fraternal and identical twins. A separate analysis on the sample of identical twins
strongly reduces the sample size and especially the variation within pairs of twins on the main
variables of criminal behaviour. The intra-class correlation for being arrested (incarcerated) is
17
0.31 (0.41) for identical twins and 0.07 (0.13) for fraternal twins. Unfortunately, due to the
routing of the questionnaire twins with a conduct disorder score of zero did not answer
questions on criminal behaviour. As this may bias the estimates we did some sensitivity
analysis with imputations for missing values on these outcomes for twins with no childhood
conduct disorder (see section 7).
Table 4.1 Summary statistics for the main estimation samples
Spent time in jail
Ever arrested
No
Yes
No
Yes
Education 11.9
10.3
12.0
11.0
(2.4)
(2.3)
(2.4)
(2.4)
Senior high school 75.0
36.2
76.7
52.2
(43.3)
(48.6)
(42.3)
(50.0)
Education (twin report) 11.7
10.2
11.7
10.9
(2.3)
(2.1)
(2.3)
(2.3)
Education father 10.4
9.5
10.4
9.9
(2.7)
(2.5)
(2.7)
(2.6)
Education mother 10.4
9.7
10.4
10.1
(3.1)
(2.9)
(3.1)
(2.8)
Male 53.1
85.1
51.1
78.1
(50)
(36)
(50.0)
(41.4)
Age in 1996 29.8
29.8
29.8
29.9
(2.5)
(2.7)
(2.5)
(2.5)
Conduct disorder 3.3
8.1
3.3
5.8
(2.5)
(3.4)
(2.4)
(3.4)
Marks primary school (1-3) 2.3
2.1
2.3
2.2
(0.6)
(0.7)
(0.6)
(0.6)
Marks secondary school (1-3) 2.2
2.0
2.2
2.0
(0.6)
(0.7)
(0.6)
(0.6)
Underachiever (%) 71.1
76.0
70.8
77.0
(45.3)
(43.1)
(45.5)
(43.1)
Grade repetition (%) 18.1
31.9
18.1
24.6
(38.5)
(47.1)
(38.2)
(0.43)
Age of first arrest 20.3
18.6
19.9
(4.6)
(4.7)
(4.7)
Identical twin 40.7
40.4
40.7
40.6
(49.1)
(49.6)
(49.1)
(49.2)
Estimation sample 2199
47
2028
224
Total sample 6197
70
5927
340
Table 4.1 shows sample means and proportions for educational attainment and background
characteristics by criminal participation. The first two columns compare twins that spent time in
jail with twins that have not been incarcerated. The last two columns compare twins that have
ever been arrested with twins that have not been arrested. The sample size slightly differs
between the first two columns and the last two columns because of missing values of
18
‘incarceration’ or ‘having been arrested’. Clearly, the sample statistics show a strong
association between educational attainment and participation in crime. Twins that have been
incarcerated attain on average 1.6 years less education than twins that have not been
incarcerated. The difference in educational attainment between those that have been arrested
and those that have not been arrested is on average 1 year. Very remarkable are the differences
in completion of senior high school, especially between those who spent time in jail and those
who did not. Two thirds of those who have been incarcerated did not graduate from senior high
school, compared to only one quarter of the remaining group of those who have not been
incarcerated. Twins that participated in crime have lower educated parents, the difference
between the columns is larger for those who spent time in jail. Male twins are more likely to be
involved in criminal activity.
The bottom panel shows the statistics on conduct disorder and early school performance.
The difference in conduct disorder is striking: twins that were incarcerated score approximately
5 points (2 standard deviations) higher on the indicator of conduct disorder. For twins that have
been arrested this difference is more than two points. We also observe that twins that have been
incarcerated or arrested have a higher grade repetition rate. The differences in self reported
marks in primary and secondary school seem quite modest. Moreover, the first arrest occurs
much earlier for twins that have been incarcerated than for other twins that have been arrested.
We further explore the association between human capital and crime by looking at the
relationship between education and arrests.
Table 4.2 Arrests by schooling level (%)
Years of schooling
7
8-10
11-12
13
15
17
Ever arrested (%) 40
18.3
7.1
6.8
7.2
6.1
First arrest
15 years 20
4
0.6
0.5
0.7
1.1
16 years 0
1.1
0.4
0
0.3
0.6
17 years 0
2.9
0.8
1.0
0
0
18 years 0
2.3
1.1
1.9
0.7
0.5
19 years 0
0.9
0.8
1.0
0.7
0.6
Ever arrested since 18 years 40
14.3
6.0
5.3
6.2
4.5
Number of arrests
0 60
81.9
93.2
93.2
92.8
93.9
1 0
10.5
4.5
5.3
5.8
5.0
2 0
3.7
1.3
1.5
1.4
0.6
3 40
4.0
1.0
0
0
0.6
Spent time in jail (%) 20
5.1
1.1
0.5
1.4
0.6
N 5
574
995
207
292
179
19
Table 4.2 shows for each schooling level the proportions for several measures of criminal
participation. Criminal participation is concentrated at the two lowest schooling levels. Twins
that did not complete 11 years of education are more likely to be arrested and to be incarcerated.
In addition, the number of arrests is higher for those with less than 11 years of education. We
also observe that many arrests of those with less than 11 years of education already take place at
an early age. Moreover, their arrest rates since the age of 18 are much higher than those for
twins with at least 11 or 12 years of education. Table 2 also makes apparent that criminal
participation is fairly stable for those with at least 11 or 12 years of education. This suggests a
non linear relationship between human capital and crime. Completion of senior high school (11-
12 years of education) seems to be a critical boundary in this respect. Lochner and Moretti
(2004) report a similar nonlinear relation between education and crime for the US. In particular,
they find a steep drop in criminal participation at the level of high school graduation.
Figure 4.1 Age of first arrest by schooling level
0 .1 .2
10 20 30 40 10 20 30 40
0 1
Density
kdensity agefirstarrest
Density
agefirstarrest
For many countries and time periods, it has well been established that crime rates increase
during the teenage years, peak around the age of twenty and decrease afterwards (Lochner,
2004). This age-crime profile is well-documented in criminology. Figure 4.1 shows age-crime
profiles from our data based on the self reported age of first arrest. The left figure shows an age-
crime profile for individuals with less than 11 years of education. The right figure shows an
age-crime profile for individuals who completed at least 11 years of education (senior high
school).
20
The patterns in figure 4.1 confirm the typical features of age-crime profiles from the
criminology literature. That is, participation in crime increases until the age of twenty and drops
afterwards. A comparison of the left and right figures suggests that individuals with less than 11
years of education start earlier with criminal activities.
21
5 The effect of early arrests on educational attainment
The strong association between education and criminal activity might be the result of early
participation in crime. Early criminal involvement might be detrimental for human capital
investment because of various reasons such as ‘meeting the wrong friends (building criminal
capital)’, ‘getting stigmatized’, changes in motivation or aspirations. In this section we
investigate the effect of early arrest on human capital accumulation by estimating linear
(probability) models of early arrests on education. Table 5.1 shows estimates of the effect of
early criminal participation on human capital. We use the information on the age of first arrest
as an indicator for early criminal participation and constructed a dummy for early arrests, which
equals 1 (0) if someone had (not) been arrested before the age of eighteen. Column (1) shows
the OLS-estimates of the effect of early arrests on educational attainment controlling for gender,
age, age squared and education of parents. Column (2) includes conduct disorder as additional
control. In column (3) additional controls for early school performance have been included:
marks in primary school (1-3), marks in secondary school (1-3), grade repetition and teachers
view on underachievement. Column (4) shows the fixed effect estimates controlling for gender,
column (5) also controls for conduct disorder and column (6) also includes controls for early
school performance. The top panel shows the effect of early arrests on years of education, the
effects on completing senior high school are shown at the bottom of table 5.1.
Table 5.1 Estimates of the effect of early arrests on educational attainment
OLS
OLS
OLS
FE
FE
FE
Years of education (1)
(2)
(3)
(4)
(5)
(6)
Arrest before 18 1.534
1.103
0.954
0.856
0.740
0.775
(0.235)***
(0.236)***
(0.232)***
(0.329)***
(0.329)**
(0.318)**
Conduct disorder
0.120
0.048
0.084
0.050
(0.018)***
(0.018)***
(0.026)***
(0.026)*
(0.063)**
N 2252
2252
2252
2252
2252
2252
Twin Pairs
1126
1126
1126
Senior high school
Arrest before 18 0.380
0.297
0.270
0.230
0.206
0.212
(0.055)***
(0.056)***
(0.055)***
(0.064)***
(0.064)***
(0.063)***
Conduct disorder
0.023
0.014
0.017
0.011
(0.004)***
(0.004)***
(0.005)***
(0.005)**
(0.012)***
N 2252
2252
2252
2252
2252
2252
Twin Pairs
1126
1126
1126
Note: All specifications control for gender. Column (1) and (2) control for age, age squared, education of parents, column (2) and (5)
control for conduct disorder, column (3) and (6) also control for early school performance. Standard errors in brackets. ***/**/* significant
at 1%/5%/10%-level.
22
All estimates in table 5.1 suggest that early arrests have a substantial impact on human capital
accumulation. The cross sectional estimates show that those who are arrested before the age of
18 attain 1.0 to 1.5 less years of education and their probability of completing senior high
school is 27 to 38 percentage points lower. The within-twin effects are smaller but remain large.
Early arrests reduce educational attainment with .7 to .9 years and lower the probability of
completing senior high school with 20 to 23 percentage points. Including conduct order reduces
the effect of early arrests.1 It should be noted that conduct disorder is closely related to early
crime as the 21 statements used for measuring conduct disorder include items that can be
considered as criminal (see table A.1). The estimates with the third specification are quite
similar to the effects of the second specification.2 Hence, including early school performance
does not affect the estimates. This indicates that, conditional on conduct disorder, the findings
are determined by early criminal behaviour, rather than differences in early school performance.
Another remarkable finding in table 5.1 is the effect of conduct disorder, which is substantial
for all specifications.
We further investigated the effect of the timing of the first arrest on education by
constructing a second variable for early arrests. This variable measures the number of years
before the age of 18 that the arrest took place (18 minus age first arrest). Table 5.2 shows the
fixed effect estimates for models that include this arrest years variable and the square of this
variable. Column (1), (2) and (3) show the estimates of the effect on years of education, column
(4), (5) and (6) show the effect on completing senior high school. We use similar controls as in
table 5.1.
Table 5.2 Estimates of the effect of the timing of the early arrest on educational attainment
Years of education Senior high school
FE
FE
FE
FE
FE
FE
(1)
(2)
(3)
(4)
(5)
(6)
18 minus age first arrest 0.080
0.652
0.616
0.037
0.141
0.126
(0.090)
(0.252)***
(0.243)**
(0.018)**
(0.049)***
(0.048)***
(18 minus age first arrest)
squared
0.099
0.097
0.018
0.016
(0.041)**
(0.039)**
(0.008)**
(0.008)**
Conduct disorder
0.051
0.012
(0.026)**
(0.005)**
N
2252
2252
2252
2252
Twin Pairs
1126
1126
1126
1126
Note: All specifications control for gender. Columns (3) and (6) control for conduct disorder and early school performance. Standard
errors in brackets. ***/**/* significant at 1%/5%/10%-level.
1 In case of missing values on conduct disorder we included the value of the other twin. If both values were missing, we
included the mean of the sample. In total we imputed values for 39 twins. We find similar results for the smaller sample
without imputation.
2 We imputed missing values on early school performance for 5 individuals. The results for the smaller sample without
imputations are similar.
23
The estimates in table 5.2 corroborate the previous findings. The estimates show that the effect
of early arrests also depends on the timing of the arrest, with earlier arrests being more
detrimental for educational attainment. For instance, column (4) indicates that each year reduces
the probability of high school completion with 3.7 percentage points. However, the estimates in
column (5) and (6) suggest that the effect is not linear. Arrests at the age of 13, 14 or 15 are the
most detrimental and reduce the probability of high school completion with more than 25
percentage points. Considering the fact that these arrests at age 13, 14 or 15 took place during
compulsory education, these findings seems in line with our expectation that the causality runs
from early arrests to human capital and not vice versa.
Summarizing, we find a large effect of early criminal behaviour on educational attainment,
even when family fixed effects are taken into account. In addition, the timing of the early arrests
matters, arrests at age 13, 14 or 15 are most detrimental for human capital accumulation.
24
25
6 The effect of human capital on crime
The second aspect of the strong association between education and criminal activity might be
the effect of education on crime. Investments in human capital raise the opportunity costs of
crime and may also alter preferences and discount rates. Previous studies for the US and the UK
find evidence for a negative effect of education on crime (Lochner & Moretti, 2004; Machin &
Vujic, 2006).
In this section, we analyze the effect of human capital on crime. The previous section
showed that reverse causality cannot be ignored, as we found substantial effects of early
criminal behaviour on educational attainment. We therefore include various controls in our
model that are informative on criminal behaviour before the age of 18. First, the ‘early arrests’
variable (arrests before the age of 18) can be used as an obvious control. Second, we can also
include the ‘conduct disorder’ variable, which is likely to precede investments in human capital.
We use the senior high school completion variable as our main measure of human capital.
Senior high school can be completed at the age of 17 or 18. This brings the advantage that we
can estimate the effect of completing senior high school on criminal activities since this age.
The distinction between the investment in human capital and the timing of criminal activity
would be less clear if we would use years of education as a measure of human capital instead. A
second argument for using senior high school completion as a measure of human capital is that
the effect of human capital on crime seems to be non linear (see table 4.2).
We investigate the effect of human capital on three self reported measures of crime:
incarceration, arrests since the age of 18, and number of arrests. Unfortunately, our data do not
contain information on the age of incarceration. However, statistics on incarceration in
Australia show that the probability of being incarcerated before the age of 18 is very small.3
Arrests since the age of 18 are derived from the age of the last arrest. For the number of arrests
we constructed a variable which has 4 categories (0; 1; 2; 3). All individuals that reported more
than three arrests were include in the last category (52 individuals reported at least three arrests
of which 22 reported exactly three arrest). The data only contain information on the age of the
first and the age of the last arrest. Hence, for the other arrests it is not clear whether they took
place after the completion of high school. Considering the evidence on reverse causality from
the previous section we expect that this will give a downward bias for the estimates (more
negative estimates).
Table 6.1 shows the estimates of the effect of completing senior high school on the three
measures of crime, using linear probability models. The first three columns show OLS-
estimates, the last three columns show estimates of fixed effect models using different controls.
The top panel shows the effects on the probability of incarceration, the middle panel shows the
3 The rate of non-indigenous persons aged 10-17 in juvenile detention between 1994 and 2003 was between 16 and 26 per
100,000 of relevant population (Charlton and McCall, 2004). This is on average approximately 0.02 % of the population.
26
effect on the probability of being arrested since the age of 18 and the bottom panel shows the
effect on the number of arrest (0-3).
Table 6.1 Estimates of the effect of high school completion on crime
Incarceration
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.041
0.020
0.013
0.038
0.027
0.023
(0.010)***
(0.008)***
(0.008)*
(0.011)***
(0.011)***
(0.010)**
Arrest before 18
0.310
0.281
0.213
0.202
(0.057)***
(0.056)***
(0.023)***
(0.023)***
Conduct disorder
0.008
0.008
(0.002)***
(0.002)***
N 2246
2246
2246
2246
2246
2246
Twin pairs
1123
1123
1123
Arrested since the age of 18
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.085
0.065
0.047
0.037
0.031
0.020
(0.016)***
(0.015)***
(0.015)***
(0.022)*
(0.022)
(0.022)
Arrest before 18
0.304
0.234
0.125
0.103
(0.059)***
(0.057)***
(0.048)**
(0.048)**
Conduct disorder
0.021
0.018
(0.003)***
(0.004)***
N 2252
2252
2252
2252
2252
2252
Twin pairs
1126
1126
1126
Number of arrests
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.214
0.104
0.072
0.108
0.048
0.029
(0.034)***
(0.024)***
(0.023)***
(0.037)***
(0.034)
(0.033)
Arrest before 18
1.645
1.522
1.223
1.183
(0.111)***
(0.106)***
(0.073)***
(0.072)***
Conduct disorder
0.037
0.033
(0.006)***
(0.006)***
N 2250
2250
2250
2250
2250
2250
Twin pairs
1125
1125
1125
Notes: All columns control for gender, columns (2) and (3) control for age, age squared and education of parents
27
The OLS estimates show that education has a negative association with all three measures of
crime. This association reduces substantially when including arrest(s) before 18 and conduct
disorder. All fixed effect estimates in column (4) are statistically significant. Controlling for
early arrests and conduct disorder substantially reduces the size of the estimates. This confirms
the earlier findings on reverse causality. Only the estimates for the effects on incarceration
remain statistically significant. Completing senior high school reduces the probability of
incarceration with 2.3 percentage points. Hence, the fixed effects estimates suggest that the
effect of human capital on crime is only moderate4.
The estimates for the effect of early arrests on the three measures of crime in table 6.1 are
striking. The fixed effect estimates suggest that an early arrest increases the probability on
incarceration with more than 20 percentage points and increase the probability of getting
arrested since the age of 18 with 10 to 12 percentage points. In addition, the average number of
arrests increases with approximately 0.2. The size of these effects is much larger than the
estimated effect of completing senior high school. For instance, the estimated impact of being
arrested before the age of 18 on incarceration is almost ten times higher than the estimated
effect of high school completion. We also estimated the same models as in table 6.1 with years
of education in stead of completing senior high school. The findings are quite similar to those in
table 6.1 and suggest a small effect of human capital on crime after controlling for early arrests
and conduct disorder (see table A.2 in the appendix).
We conclude that this section provides evidence for a negative but moderate effect of human
capital on crime. Completing senior high school reduces the probability of incarceration with at
least 2 percentage points. Human capital also reduces the probability of being arrested since the
age of 18 and the number of arrests. Including early arrests and conduct disorder in the models
substantially reduced the effect of human capital on crime. This confirms that reverse causality
is an important issue. The most remarkable findings are the large effect of early arrests on all
three measures of crime. These effects are substantially larger than the estimated effects of
human capital.
4 As in the previous tables, we imputed values for 39 twins with missing data on conduct disorder. The estimation results on
the smaller sample without the imputed values are similar.
28
29
7 Robustness
In this section, we investigate the robustness of the findings by addressing two issues. First, we
test the sensitivity of the results by imputing missing values on criminal outcomes which are
due to the routing of the questionnaire. Second, we address the issue of measurement error
which is likely to bias the estimates downward.
7.1 Missing values due to the routing of the questionnaire
Due to the rooting of the questionnaire, twins with a conduct disorder score of zero, which
means that they reported negative on all 21 statements on conduct disorder before the age of 18,
did not answer questions about arrests and incarceration. This may bias the estimates because
this involves a large fraction of our sample (approximately 3000 observations). It seems likely
that individuals that report no conduct disorder behaviour will be less involved in crime than
those that have a positive conduct disorder score. For instance, the arrest (incarceration) rate of
those with a conduct disorder score of 3 is 7.4 (0.6) against 2.5 (0.3) for those with a conduct
disorder score of 1. We checked the sensitivity of the results by imputing zeros for twins with
missing values on being arrested and incarceration. Tables 7.1 and 7.2 show the estimation
results for the main models of the previous sections. Table 7.1 shows the results for the effect of
early crime on educational attainment.
Table 7.1 Estimates of the effect of early arrests on educational attainment after imputations for missing
values on early arrests
OLS
OLS
OLS
FE
FE
FE
Years of education (1)
(2)
(3)
(4)
(5)
(6)
Arrest before 18 1.597
0.876
0.759
0.803
0.624
0.668
(0.215)***
(0.211)***
(0.205)***
(0.289)***
(0.291)**
(0.279)**
Conduct disorder
0.148
0.071
0.079
0.038
(0.014)***
0.0(13)***
(0.018)***
(0.018)**
N 5332
5332
5332
5332
5332
5332
Twin pairs
2666
2666
2666
Senior high school
Arrest before 18 0.363
0.254
0.225
0.189
0.162
0.163
(0.052)***
(0.051)***
(0.050)***
(0.053)***
(0.054)***
(0.053)***
Conduct disorder
0.022
0.014
0.012
0.006
(0.003)***
(0.003)***
(0.003)***
(0.003)
N 5332
5332
5332
5332
5332
5332
Twin pairs
2666
2666
2666
Note: All specifications control for gender. Columns (1) and (2) control for age, age squared, education of parents, columns (2) and (5)
control for conduct disorder, columns (3) and (6) also control for early school performance. Standard errors in brackets. ***/**/* significant
at 1%/5%/10%-level.
30
The estimates in table 7.1 are somewhat smaller but quite similar to those in table 5.1. After the
imputation of the missing values for being arrested we still find a large effect of early arrests on
educational attainment.
Table 7.2 shows the estimates for the effect of high school completion on crime. The pattern
of findings in table 7.2 is similar to the pattern in table 6.1. However, the estimates of the effect
of high school completion on crime in the fixed effects model that uses all controls (column 6)
becomes statistically insignificant. This suggests that the effect of educational attainment might
be even smaller than indicated in table 6.1.
Table 7.2 Estimates of the effect of high school completion on crime after imputations for missing values
on the crime variables
Incarceration
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.021
0.010
0.005
0.016
0.011
0.008
(0.005)***
(0.004)**
(0.004)
(0.006)***
(0.006)*
(0.006)
Arrest before 18
0.291
0.260
0.204
0.185
(0.052)***
(0.051)***
(0.016)***
(0.016)***
Conduct disorder
0.007
0.008
(0.001)***
(0.001)***
N 5326
5326
5326
5325
5326
5326
Twin pairs
2663
2663
2663
Arrested since the age of 18
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.054
0.042
0.025
0.023
0.020
0.011
(0.009)***
(0.009)***
(0.008)***
(0.012)*
(0.012)*
(0.012)
Arrest before 18
0.315
0.218
0.140
0.094
(0.054)***
(0.052)***
(0.033)***
(0.033)***
Conduct disorder
0.021
0.021
(0.002)***
(0.002)***
N 5332
5332
5332
5332
5332
5332
Twin pairs
2666
2666
2666
Number of arrests
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.127
0.065
0.037
0.158
0.027
0.013
(0.020)***
(0.013)***
(0.013)***
(0.020)***
(0.018)
(0.018)
Arrest before 18
1.651
1.491
1.261
1.184
(0.101)***
(0.096)***
(0.050)***
(0.049)***
Conduct disorder
0.035
0.036
(0.004)***
(0.003)***
N 5330
5330
5330
5330
5330
5330
Twin pairs
2665
2665
2665
Notes: All columns control for gender, columns (2) and (3) control for age, age squared and education of parents.
31
We conclude that the estimates of the previous section are robust for imputing missing values of
individuals with a conduct disorder score of zero. However, the estimated effect of high school
completion on crime becomes statistically insignificant in models that control for early crime
and conduct disorder.
7.2 Measurement error
A well-known concern in the literature using within-family models is measurement error
(Griliches, 1979). By taking a within-family perspective, measurement error may exacerbate,
which in turn is likely to bias the estimates towards zero. A solution for this problem has been
proposed by Ashenfelter and Krueger (1994) in their study on the returns to schooling using
data on twins. They suggested using a second independent measure of education as an
instrument for educational attainment. In their study, they asked each sibling to report on both
their own and their twin’s schooling and used this information as independent measures of
schooling. They constructed two instruments for the difference in education within twins
depending on the assumptions about measurement error. Let
1
1
S
refer to the self-reported
education level of the first twin,
2
1
S
to the sibling-reported education level of the first twin,
2
2
S
to the self-reported education level of the second twin and
1
2
S
to the sibling-reported
education level of the second twin. The first instrument uses the difference in the twin’s reports
on the schooling of their sibling as an instrument for the difference in the report on the own
schooling. Hence,
1 2
1 2
S S
is instrumented with
2 1
1 2
S S
. The second instrument assumes that
the measurement error of respondent’s report on the own schooling and the schooling of their
sibling is correlated. In the estimation, the difference in the reports of twin A about the own
schooling and the sibling’s schooling is instrumented with the difference in the reports of twin
B on the sibling’s schooling and the own schooling. Hence,
1 1
1 2
S S
is instrumented with
2
2
2
1SS .
In our study, we can follow this approach in the models that estimate the effect of education
on crime because our data include the same questions on the sibling’s schooling. The
correlation between the self-reported level of education and the sibling-reported education level,
which indicates the reliability ratio, is 0.80. For high school completion this correlation is 0.63.
It should be noted that this approach produces consistent estimates when the measurement error
is classical. However, since our main variable (senior high school completion) is a binary
indicator, the measurement error is non-classical. It has been shown that the IV-estimate will
then be upward biased (Aigner, 1973, Kane et al. 1999). The within-family estimate from the
previous analysis will then provide a lower bound and the IV estimate an upper bound of the
true (negative) effect.
Table 7.3 shows the IV-estimates for the effect of high school completion on the three
measures of crime. Columns (1), (3) and (5) show the estimation results for the first instrument
32
described above. Columns (2), (4) and (6) show the results for the second instrument. All
specifications use early arrest, conduct disorder and gender as controls.
Table 7.3 IV-estimates of the effect of senior high school completion on crime
Incarceration Arrest since 18 Number of arrests
IV1
IV2
IV1
IV2
IV1
IV2
(1)
(2)
(3)
(4)
(5)
(6)
Senior high school 0.199
0.064
0.231
0.081
0.401
0.115
(0.117)*
(0.024)***
(0.228)
(0.050)
(0.348)
(0.075)
N 2243
2243
2249
2249
2247
2247
Twin pairs 1123
1123
1126
1126
1125
1125
Notes: All columns control for gender, early arrest and conduct disorder. Standard errors in brackets. ***/**/* significant at 1%/5%/10%-
level.
The estimates in table 7.3 suggest that measurement error in education might be important. All
estimates increase and most estimates are statistically significant. The estimates with the first
instrument are very large but also have large standard errors. The estimates with the second
instruments are also larger than the estimates in table 6.1 but more precise. These results
suggests that the findings in table 6.1 might underestimate the true effect of human capital on
crime. We find a similar pattern when using years of education instead of completion of senior
high school. However, the estimates are smaller (see table A.3 in the appendix). In addition, we
re-estimated the models from table 7.3 after imputing the missing values for individuals with a
conduct disorder score of zero (see table A.4 in the appendix). The size of the estimates is
smaller after the imputation but the pattern of findings is quite similar.
Unfortunately, our data do not contain sibling reports on criminal behaviour. As such we
can not use this approach for the models that investigate the effect of early crime on education.
However, we can make a tentative assessment using external information on the reliability of
self-reported crime and the intra class correlation in early crime measured in our sample of
twins. Assuming classical measurement error Grilliches (1979) shows that within-family
estimation increases the bias by measurement with 1 / (1
c
ρ
) with c
ρ
as the intra class
correlation in early crime within families. Thornberry and Krohn (2000) report that many
studies find a reliability ratio of self reported crime well above 0.8. The intra class correlation in
early crime in our data is 0.22. This means that the bias of the OLS-estimator is -0.2*β and the
bias of the fixed effect estimator is -0.2/(1-0.22)*β =-0.26* β. This calculation suggests that the
additional downward bias of the within estimator is quite modest.
33
8 Decomposing the association between crime and
education
The two main findings from the previous sections are that early criminal behaviour is
detrimental to investment in human capital and that human capital has a negative effect on
crime. In this section we try to assess the importance of these two effects for the association
between crime and education. We estimated within-twin models of the effect of education on
‘ever being arrested’ and inspect how the estimated effect of education changes after including
early crime and conduct disorder. Including ‘early arrests’ in the estimation controls for the
effect of early criminal behaviour on educational attainment and ‘explains’ all arrests before the
age of 18, leaving only crime since 18 to be explained. Table 8.1 shows the estimation results
using years of education or high school completion as explanatory variables.
Table 8.1 Fixed effect estimates of the effect of education on the probability of being arrested
(1)
(2)
(3)
Years of education 0.010
0.004
0.003
(0.005)**
(0.004)
(0.004)
N 2252
2252
2252
Pairs 1126
1126
1126
Senior high school 0.067
0.026
0.018
(0.024)***
(0.021)
(0.021)
N 2252
2252
2252
Pairs 1126
1126
1126
Controls
Early arrest No
Yes
Yes
Conduct disorder No
No
Yes
Note: All columns control for gender. Standard errors in brackets. ***/**/* significant at 1%/5%/10%-level.
The estimates in the first column show that one year of education is associated with a reduction
of the probability of being arrested with 1 percentage point. Completion of high school is
associated with a reduction of the probability of being arrested with 6.7 percentage points. The
estimates of the effect of human capital reduce dramatically after the inclusion of ‘early arrest’
(column (2)). The estimated effect of one year of education reduces to 0.4 percentage points and
the estimated effect of high school completion to 2.6 percentage points. Including conduct
disorder further reduces the estimated effects to 0.3 and 1.7 percentage points (column (3)). In
other words, controlling for early arrests and early behaviour problems reduces the estimated
effect of human capital on crime to less than one third of the previously estimated association.
From this, we conclude that early criminal behaviour explains most of the association between
human capital and crime.
34
35
9 Conclusions and discussion
This paper aims at disentangling the strong association between human capital and crime by
investigating whether crime reduces investment in human capital or whether education reduces
criminal activity. Heretofore, we exploit two aspects of the Australian survey data on education
and crime we use. First, as the data are obtained from twins, we are able to control for many
unobserved characteristics affecting both criminal behaviour and the schooling decisions.
Second, as criminal behaviour is measured over different periods of time – prior to and after
senior high school completion – we can address the causality between crime and education as
well. As early criminal behaviour may affect human capital formation, and human capital may
influence criminal behaviour in later stages of life, we follow a two step analysis.
First, we address the effects of early criminal behaviour on educational attainment. The
estimates suggest that early criminal behaviour is detrimental to investment in human capital.
Within pairs of twins we find that early arrests (before the age of 18) reduce educational
attainment with .7 to .9 years and lower the probability of completing senior high school with
20 to 23 percentage points. In addition, the timing of the early arrest matters, arrests at age 13,
14 or 15 are most detrimental for educational attainment. These estimates are found after
controlling for conduct disorder and early school performance.
Second, we focus on the effect of human capital on crime. As early criminal activity might
be an important confounder, we control for early arrests. The estimates suggest that human
capital has a negative effect on crime. Completing senior high school reduces the probability of
incarceration with 2 to 3 percentage points. We find similar but statistically insignificant effects
on the probability of being arrested since the age of 18 and on the number of arrests. The size of
these estimates might be downward biased because of measurement error in schooling. IV-
estimates using a second independent measure of schooling suggest that the effect of human
capital might be larger. Lochner and Moretti (2004) report IV-estimates of the effect of high
school completion on imprisonment of 8 percentage points for blacks and 0.9 for whites.
When combining these findings, it seems that the causality between human capital and
crime runs in both directions. Still, the impact of early criminal behaviour on human capital
formation dominates the impact of human capital formation on future crime behaviour.
Controlling for early arrests and early behaviour problems reduces the estimated effect of
human capital on crime to less than one third of the previously estimated association. From this,
we conclude that early criminal behaviour explains most of the association between human
capital and crime.
The strong detrimental effects of early criminal behaviour become also transparent if we
consider the estimated effects of early arrests on all three measures of crime. Early arrests
increase the probability of incarceration with 20 percentage points and the probability of being
arrested since the age of 18 with 10 percentage points. These effects are much larger than the
estimated effects of human capital. For instance, the estimated effect of being arrested before
36
the age of 18 on incarceration is almost ten times higher than the estimated effect of completing
high school.
In line with previous studies (Lochner and Moretti, 2004, Machin and Vujic, 2006) our
findings suggest that policies that succeed in raising investment in human capital might reduce
crime. However, the (direct) returns to polices that succeed in preventing early criminal
behaviour might be much larger. The estimated effects of early criminal behaviour and conduct
disorder stress the importance of the early stages of life for preventing crime. Programs that
keep children on ‘the right track’ not only may yield high private returns but also may yield
high social returns through their impact on crime reduction. Studies on the effects of effective
early schooling programs in the US show that these program have large social returns mainly
through their impact on preventing crime (Carneiro, et. al, 2003).
Our main conclusion is that the strong association between human capital and crime is
mainly driven by the effect of early criminal behaviour on educational attainment. This finding
based on within-twin estimation confirms one of the main conclusions from a synthesis of the
literature on the causes of crime: ‘We must rivet our attention on the earliest stages of the life
cycle, for after all is said and done, the most serious offenders are boys who begin their criminal
careers at a very early age.’ (Wilson and Hernstein, 1985, cited in Dilulio, 1996).
37
References
Aigner, D., 1973, Regression with a binary independent variable subject to errors of
observation, Journal of Econometrics 1 (1973), pp. 49–60.
Ashenfelter, O. and A. Krueger, 1994, Estimates of the Economic Return to Schooling from a
New Sample of Twins, The American Economic Review, Vol. 84, No. 5, p. 1157-1173.
Australian Institute of Criminology, 2007, Australian Crime, Facts & Figures 2007, Canberra,
Australia.
Behrman, J. R., P. Taubman and T.J. Wales, 1977, Controlling for and Measuring the Effects of
Genetics and Family Environment in Equations for Schooling and Labour Market Success, in
Paul Taubman, ed., Kinometrics: The Determinants of Socioeconomic Success within and
between Families, Amsterdam: North Holland, p. 35-96.
Bound, J. and G. Solon, 1999, Double trouble: on the value of twins-based estimation of the
return to schooling, Economics of Education Review, Vol. 18, p. 169-182.
Borghans, L., A.L. Duckworth, J.J. Heckman and B. ter Weel, 2008, The Economics and
Psychology of Personality Traits, Journal of Human Resources, Vol. 43, No. 4, forthcoming.
Borghans, L, H. Meijers and B. ter Weel, 2008, The role of non-cognitive skills in explaining
cognitive test scores, Economic Inquiry, Vol. 46, No. 1, p. 2-12.
Carneiro, P. and J.J. Heckman, 2003, Human Capital Policy, in J. J. Heckman, A. B. Krueger
and B. M. Friedman (Eds.), Inequality in America: What Role for Human Capital Policies?
Cambridge, MA: MIT Press.
Charlton, K. and M. McCall, 2004, Statistics on juvenile detention in Australia: 1981-2003,
Australian Institute of Criminology.
Currie, J. and M. Stabile, 2006, Child Mental Health and Human Capital Accumulation: The
Case of ADHD, Journal of Health Economics, Vol. 25, No. 6, pp. 1094-1118.
Currie, J. and M. Stabile, 2007, Mental Health in Childhood and Human Capital, NBER
Working Paper, WP 13217.
38
Donohue, J. and P. Siegelman, 1998, Allocating Resources among Prisons and Social
Programmes in the Battle against Crime, Journal of Legal Studies, Vol. 27, No. 1, p. 1-43.
Farrington, D., 1973, Self-reports of deviant behavior: Predictive and stable?, Journal
of Criminal Law and Criminology, 64(1): 99-110.
Fisher, D.G., G.L. Reynolds, M.M. Wood and M.E. Johnson, 2004, Reliability of Arrest and
Incarceration Questions on the Risk Behaviour Assessment, Crime & Delinquency, Vol. 50, No.
1, 24-31.
Fletcher, J. and B.L. Wolfe, 2007, Child Mental Health and Human Capital Accumulation: The
Case of ADHD Revisited, NBER Working Paper, WP 13474.
Freeman, R. B., 1996, Why Do So Many Young American Men Commit Crimes and What
Might We Do about It?, Journal of Economic Perspectives, American Economic Association,
Vol. 10(1), pp. 25-42.
Garces, E., D. Thomas and J. Currie, 2002, Longer-Term Effects of Head Start, American
Economic Review, Vol. 92, No. 4, pp. 999-1012.
Griliches, Z., 1979, Sibling models and data in economics: beginnings of a survey, Journal of
Political Economy, 87(5), pp. S37-S64.
Hardt, R., & S. Hardt, 1977, On determining the quality of the delinquency self-report
method, Journal of Research in Crime and Delinquency, July: 247-257.
Heckman, J. J., J. Stixrud and S. Urzua, 2006, The Effects of Cognitive and Non-cognitive
Abilities on Labour Market Outcomes and Social Behaviour, Journal of Labour Economics,
Vol. 24, No. 3, p. 411-481.
Heckman, J. J. and D.V. Masterov, 2007, The Productivity Argument for Investing in Young
Children, IZA Discussion Paper 2725.
Horney, J. and I. Marshall, An experimental comparison of two self-report methods for
measuring lambda, Journal of Research in Crime and Delinquency, 29(1): 102-121.
Horney, J. and I.H. Marshall, 1992, Risk perceptions among serious offenders: the role of crime
and punishment, Criminology 30, pp. 575–594.
39
Huizinga, D. and D.S. Elliott, 1986, Reassessing the reliability and validity of selfreport
delinquency measures, Journal of Quantitative Criminology, 2, 293-327.
Kane, T.J., C.E. Rouse and D. Staiger, 1999, Estimating returns to schooling when schooling is
misreported, Working paper 419, Princeton University.
Lally, J. R., P. Mangione and A. Honig, 1988, The Syracuse University Family Development
Research Programme: Long-Range Impact on an Early Intervention with Low-Income Children
and Their Families, In Parent Education as Early Childhood Intervention, Douglas Powell, ed.
Norwood, N.J.: Ablex.
Le, A. T., P.W. Miller, A.C. Heath and N. Martin, 2005, Early childhood behaviors, schooling,
and labour market outcomes: estimates from a sample of twins, Economics of Education
Review, Vol. 24, p. 1-17.
Lochner, L, 2004, Education, Work and Crime: A Human Capital Approach, International
Economic Review, 45, 811-843.
Lochner, L. and E. Moretti, 2004, The Effect of Education on Crime: Evidence from Prison
Inmates, Arrests, and Self-Reports, American Economic Review, 94, 155-189.
Machin, S. and S. Vujic, 2006, Crime and Education: New Evidence From Britain, CEP
Working Paper.
Maddux J. and D. Desmond, 1975, Reliability and validity of information from chronic heroin
users, Journal of Psychiatric Research, 12(2): 87-95.
Mieczkowski, T., 1990, The accuracy of self-reported drug use: An evaluation and analysis of
new data, in R. Weisheit (Ed.), Drugs, crime and the criminal justicesystem, pp. 275-302,
Cincinnati: Anderson.
Miller, P., C. Mulvey and N. Martin, 1995, What Do Twins Studies Reveal about the Economic
Return to Education? A Comparison of Australian and U.S. Findings, American Economics
Review, Vol. 85, No. 3, pp. 586-599.
Miller, P., C. Mulvey and N. Martin, 2006, The return to schooling: Estimates from a sample of
young Australian twins, Labour Economics, 13, pp. 571-587.
40
Nieves, K., J. Draine and P. Solomon, 2000, The Validity of Self-Reported Criminal Arrest
History Among Clients of a Psychiatric Probation and Parole Service, Journal of Offender
Rehabilitation, Vol. 30 (3/4), 2000, pp. 133-151.
Schweinhart, L., H. Barnes and D. Weikart, 1993, Significant Benefits: The High-Scope Perry
Pre-School Study through Age 27, Ypsilanti, Mich.: High Scope Press.
Short, J.F. and I. Nye., 1957, Reported behavior as a criterion of deviant behavior, Social
Problems, 5, 207-213.
Slade, E.P. and L.S. Wissow, 2007, The Influence of Childhood Maltreatment on Adolescents,
Academic Performance, Economics of Education Review, Vol. 26, No. 5, pp. 604-614.
Tauchen, H., A.D. Witte and H. Griesinger, 1994, Criminal deterrence: Revisiting the issue
with a birth cohort, Review of Economics and Statistics, 76(3), pp. 399-412.
Thornberry, T.P. and M.D. Krohn, 2000, The self-report method for measuring delinquency and
crime, Criminal Justice 2000, vol. 4, pp. 33-83.
Vujic, S., H.D. Webbink and P. Koning, 2008, The effect of childhood conduct disorder on
human capital, CPB Discussion Paper.
Weiss, R., L. Najavits, S. Greenfield, J. Soto, S. Shaw and D. Wyner, 1998,Validity of
substance use self-reports in dually diagnosed outpatients, American Journal of Psychiatry,
155, 127-128.
Wilson, J.Q., and R.J. Herrnstein, 1985, Crime and human nature, New York: Simon and
Schuster.
41
Appendix
A1. Variable Definitions
Table 9.1 Conduct disorder statements from the TWIN89 questionnaire
Variable Question
misbehaved L3 Did you frequently get into a lot of trouble with the teacher or principal for
misbehaving in school? (primary or secondary school)
wagged school L4 Before age 18, did you ever wag school for an entire day at least twice in 1 year?
suspended/expelled L5 Were you ever suspended or expelled from school?
stay out late L6 As a child or a teenager, did you often stay out much later than you were supposed
to?
sneak out at night L6A Did you often sneak out of the house at night?
run away overnight L6C Before age 18, did you ever run away from home overnight?
lied, used false name L7 Before 18, did you ever tell a lot of lies or use a false name or alias?
outsmarted, conned others L7B Before age of 18, was there ever a period when you often outsmarted others and
“conned” them?
stole from home or family L8 Before age 18, did you steal money or things from your home or family more than
once? If yes, did you only steal things of trivial value, like loose change or things like
that?
shoplifted L8A Before age 18, did you steal or shoplift from shops or other people (without their
knowing) more than once? If yes, did you only steal things of trivial value like comics or
lollies?
forged signature L8B Before age 18, did you forge anyone’s signature on a cheque or credit card more
than once?
damaged property L9 Have you ever damaged someone’s property on purpose?
started physical fights L10 Before age 18, did you start physical fights (with persons other than your brothers
or sisters) 3 or more times?
used a weapon L11 Before age 18, did you ever use a weapon like a bat, brick, broken bottle, gun or a
knife (other than in combat, when hunting, or as part of your job) to threaten or harm
someone?
physically injured someone L12 Before age 18, (other than fighting or using a weapon) did you ever physically injure
anyone on purpose?
bullied others L13 Before age 18, were you often a bully, deliberately hurting or being mean to others?
mean to animals L14 Before age 18, were you ever mean to animals including pets or did you hurt
animals on purpose?
lighted fires L15 Before 18, did you ever deliberately light any fires you were not supposed to?
broke into someone’s
car/house
L16 Before 18, did you ever break into someone’s car or house or anywhere else (not
because you were locked out)?
forcefully stole money or
property
L17 Before age 18, did you ever take money or property from someone else by
threatening them or using force, like snatching a purse or robbing them?
forced someone into sexual
activity
L20 Before age 18, did you ever force anyone into intercourse or any other form of
sexual activity?
42
Table A.2 Estimates of the effect of years of education completed on crime
Incarceration
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Years of education 0.006
0.003
0.001
0.004
0.002
0.001
(0.002)***
(0.001)**
(0.001)
(0.002)*
(0.002)
(0.002)
Arrest before 18
0.313
0.284
0.218
0.205
(0.057)***
(0.056)***
(0.023)***
(0.023)***
Conduct disorder
0.009
0.009
(0.002)***
(0.002)***
N 2246
2246
2246
2246
2246
2246
Twin pairs
1123
1123
1123
Arrest since 18
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Years of education 0.012
0.009
0.006
0.006
0.005
0.003
(0.003)***
(0.003)***
(0.003)**
(0.004)
(0.004)
(0.004)
Arrest before 18
0.314
0.242
0.128
0.104
(0.059)***
(0.057)***
(0.048)***
(0.048)**
Conduct disorder
0.021
0.018
(0.003)***
(0.004)***
N 2252
2252
2252
2252
2252
2252
Twin pairs
1126
1126
1126
Number of arrests
OLS Within twin estimates
(1)
(2)
(3)
(4)
(5)
(6)
Years of education 0.032
0.016
0.010
0.019
0.011
0.007
(0.006)***
(0.004)***
(0.004)**
(0.007)***
(0.007)
(0.007)
Arrest before 18
1.660
1.532
1.225
1.184
(0.111)***
(0.106)***
(0.073)***
(0.072)***
Conduct disorder
0.038
0.032
(0.006)***
(0.006)***
N 2250
2250
2250
2250
2250
2250
Twin pairs
1125
1125
1125
Notes: All columns control for gender, columns (2) and (3) control for age, age squared and education of parents.
Table A.3 IV-estimates of the effect of years of education on crime
Incarceration Arrest since 18 Number of arrests
IV1
IV2
IV1
IV2
IV1
IV2
(1)
(2)
(4)
(5)
(1)
(2)
Senior high school 0.005
0.003
0.006
0.007
0.008
0.008
(0.005)
(0.003)
(0.010)
(0.007)
(0.015)
(0.011)
N 2243
2243
2249
2249
2247
2247
Twin pairs 1123
1123
1126
1126
1125
1125
Notes: All columns control for gender, early arrest and conduct disorder. Standard errors in brackets.
43
Table A.4 IV-estimates of the effect of high school completion on crime after imputating missing values
Incarceration Arrest since 18 Number of arrests
IV1
IV2
IV1
IV2
IV1
IV2
(1)
(2)
(4)
(5)
(1)
(2)
Senior high school 0.106
0.030
0.117
0.061
0.283
0.094
(0.066)
(0.013)**
(0.130)
(0.026)**
(0.202)
(0.039)**
N 5322
5322
5328
5328
5326
5326
Twin pairs 2663
2663
2666
2666
2665
2665
Notes: All columns control for gender, early arrest and conduct disorder. Standard errors in brackets. ***/**/* significant at 1%/5%/10%-
level.
44
... Our study is most closely related to the study of Webbink et al. (2013), which estimates a twin fixed effects model using data on twin pairs from Australia. If the family and regional environment is similar for fraternal twins, the twin fixed effect model controls adequately for this unobserved heterogeneity. ...
... Another advantage of the twin fixed effect approach is that twins are prevalent across all educational levels, which contrasts with instrumental variable approaches or regression discontinuity designs that estimate effects at a very specific margin. For the sample of fraternal twins, Webbink et al. (2013) find that arrest before the age of 18 reduces educational attainment by up to 0.99 education years and lowers the probability of completing senior high school by up to 24 percentage points. These results are statistically significant, but relatively imprecise, because there are only 28 fraternal twin sets with variation in arrest status in the sample. ...
... Studies that use a sampled population of twins can suffer from lack of statistical power (e.g. Webbink et al. 2013). ...
Article
Full-text available
Objective We examine the relationship between early criminal involvement and school dropout, and analyze which factors underlie this relationship, making use of administrative data from the Netherlands. Methods We start by determining the unconditional correlation between early criminal involvement and school dropout, using a basic ordinary least squares model. As this association is likely to be driven by different factors, we proceed by including an extensive set of observable family and individual characteristics into the estimation model. We further proceed to models that account for the influence of unobservable heterogeneity by estimating school, class, sibling and twin fixed effects. ResultsCriminal involvement is associated with an 11 percentage point higher probability of school dropout. The magnitude of this relationship decreases gradually when we account for larger shares of observed and unobserved heterogeneity. The coefficient in the same-gender twin fixed effects model indicates a 3 percentage point higher probability of school dropout, which is statistically significant at a 10 % level. We also find that the association between criminal involvement and school dropout is stronger if juveniles are involved in severe criminal activities. Conclusions We conclude that the observable and unobservable factors for which we account explain around 73 % of the unconditional correlation between criminal involvement and school dropout. The remaining variation likely reflects individual-specific characteristics that are different between same-gender twins. A true treatment effect, if existing, is likely to be relatively small. At the same time, serious criminal behavior appears to causally affect school dropout.
... According to the Office of Justice Programs, an estimated 1.4% of individuals who had contact with law enforcement had force used or threatened against them in 2008. 79 Younger people (ages [16][17][18][19][20][21][22][23][24][25] were twice as likely (2.6%) to experience a threat or the use of force by police than people 26 years or older (1.3%). 8 People of color experience disproportionate use of force by police compared to their White counterparts. ...
... Youth who are first arrested at a young age -13, 14, or 15 -experience the most detrimental effects. They are 25% more likely to drop out of high school, 20% more likely to be incarcerated, and 10-12% more likely to be arrested again after the age of 18. 23 Overall, youth who are arrested are nearly twice as likely to drop out of high school, 101 and they average 1 to 1.5 fewer years of education. 22 23 ...
... When employment options are limited for adults (ages 25+) such as during recessions, adults will take jobs usually filled by youth and young adults (ages [16][17][18][19][20][21][22][23][24], resulting in higher unemployment rates for youth. 108 ...
Research
Full-text available
Being arrested as an adolescent can impact a person’s health and life trajectory profoundly. Social policies like zero-tolerance school discipline and neighborhood gang injunctions have led to a proliferation of police surveillance and punishment of common youth misbehaviors. These policies disproportionately impact Black and Latinxi boys and girls living in low-income communities. In this report, we evaluate the health and equity impacts of youth arrest (for kids under the age of 17) in Michigan, with a focus on the city of Detroit, and Wayne and Washtenaw Counties.
... However, enforcing penalties may also come with additional costs on society. In addition to the costs of the judicial system, traditional punishment also comes with other societal costs, such as negative effects on educational outcomes (see Hjalmarsson, 2008;Webbink et al., 2013;Aizer and Doyle, 2015). This is especially relevant considering the crime-reducing effect of educational attainment (Machin et al., 2011). ...
... In addition to estimating the effects of programme participation on criminal involvement, we also estimate the effects of programme participation on tertiary educational attainment. There is evidence that criminal involvement at a young age is not only positively associated with future criminal involvement (see Ellis et al., 2009;Loeber et al., 2013) but that it can also lead to lower educational outcomes (see Hjalmarsson, 2008;Webbink et al., 2013;Aizer and Doyle, 2015). ...
Article
Full-text available
Objectives This study examines the effects of a restorative justice programme in the Netherlands on educational outcomes and repeated delinquent behaviour of young people who have committed a criminal offence. Methods We use data from a field experiment, in which the participants are randomly assigned to a restorative justice programme. We link the data from the field experiment to longitudinal administrative data on education and criminal records and correct for selective attrition by implementing an instrumental variable approach. Results The results show that participation in the restorative justice programme increases the probability of recidivism one year after the programme by 39.3 percent from a base rate of 17.1 percent and decreases tertiary educational attainment by 29.1 percent from a base rate of 30.9 percent. Conclusions Altogether, the results of this study suggest that despite the design backed by criminological and sociological theory, a restorative punishment from the Halt programme does not succeed in reducing criminal involvement and improving the educational outcomes.
... Few studies examine the effect of juvenile delinquency on educational attainment. Hjalmarsson (2008) investigates the effects of juvenile arrest, charge, conviction, and imprisonment on high school graduation; Kirk and Sampson (2013) and Webbink et al. (2013) study the impact of being arrested on graduating from high school; while Aizer and Doyle (2015) investigate the effect of incarceration on high school graduation. These studies conclude that interactions with the criminal justice system in the form of arrests and incarceration decrease educational attainment. ...
... For example, Hjalmarsson (2008) finds that it is incarceration that reduces the probability of high school graduation, which she interprets as possibly reflecting a stigma effect or the impact of absence from the classroom. Webbink et al. (2013) and Kirk and Sampson (2013) find that it is arrest that leads to high school dropout, which may suggest a stigma effect. Similarly, offenders who are members of gangs may be more embedded in a criminal peer culture in which the returns to education are less valued (Akerlof & Yellen 1993;Williams & Sickles 2002). ...
Article
This paper investigates the effect of delinquency in youth on subsequent educational attainment. To do so, we focus on delinquent acts committed by age 16 and examine their impact on two measures of educational attainment: high school graduation and college graduation. Using information on males from the extremely rich National Longitudinal Survey of Youth 1997, we find plausible evidence that delinquency by age 16 reduces the likelihood of graduating from high school and college. This effect is driven by early initiators, those who offend intensely, and by those whose delinquent activities involve income generating acts. Importantly, the impact of delinquency on education is not confined to those who have interaction with the criminal justice system, or gang members. Further analysis suggests that a mechanism through which delinquency impacts on education is expected returns to crime, as reflected by subjective beliefs about the probability of arrest for a property crime. This effect is stronger for those of higher ability and is robust to accounting for attitude to risk.
... Arrest first occurs at older ages and is less common. Nonetheless, youthful interactions with the justice system are far from harmless, with several recent studies demonstrating that arrest for delinquent behaviour leads to early school leaving, adversely impacting on both high school completion and college attendance (Hjalmarsson, 2008;Webbink et al., 2013;Aizer and Doyle, 2015;Rud et al., 2018). 1 These findings have important policy implications given the central role of education in lifetime economic well-being, as well as the increased likelihood of future arrest and imprisonment that results from early school leaving (Becker, 1962;Lochner and Moretti, 2004;Oreopoulos and Salvanes, 2011;Merlo and Wolpin, 2015;Buonanno and Leonida, 2009; JEL Classification numbers: C4, I2, K4, D0. *The authors are grateful to the Department of Economics, University of Melbourne for supporting this research. We thank Stephen Machin, Olivier Marie, Sarmistha Pal and other participants in the Surrey-UGPN Conference on Youth Crime and Public Policy Interventions 2017, Kevin Staub and two anonymous referees for helpful feedback on this research. ...
Article
Full-text available
Boys typically initiate delinquent behaviour during their teenage years, and many go on to be arrested. We show that engaging in delinquency and being arrested in youth are each associated with early school leaving. The effect of delinquency on school leaving is largely driven by crimes that produce a monetary return, and the increase in school leaving is greater when onset of these types of crime, and arrest, occur at younger ages. The sizeable impact of delinquency on school leaving highlights the need for crime prevention efforts to extend beyond youth who come into contact with the justice system.
... Scholars have long argued that delinquency impacts educational outcomes including academic achievement, high school dropout, and educational attainment. While the focus of previous research has been on establishing the relationship between criminal justice contact (e.g., arrest, charge, conviction, and incarceration) and education (e.g., Hirschfield 2009;Hjalmarsson 2008;Kirk and Sampson 2013;Monk-Turner 1989;Webbink et al. 2013), several studies have suggested that juvenile delinquency per se is related to educational outcomes. Tanner and colleagues (1999) found that both male and female adolescents who engaged in delinquent activities, especially skipping school and drug use, were less likely to receive a high school diploma and obtain a college degree. ...
Article
Given large variations in the etiology and developmental trajectories of violent and nonviolent delinquency, this study examines whether educational outcomes of violent and nonviolent offenders might differ. In particular, this study attempts to remove environmental influences such as family background and neighborhood effects from the effects of delinquency because these factors are likely to differentially confound the effects of violent and nonviolent delinquency on educational attainment. By exploiting variation within sibling pairs, this study finds that the effects of engagement in violent delinquency on education is driven spuriously by shared family background, whereas the effects of nonviolent delinquency are quite robust to adjustment for family fixed effects. Moreover, relying on fixed effects estimates, this study finds that the effects of engagement in nonviolent delinquent activity on educational attainment occur in part through disruption of educational progress, rather than through institutional responses to student delinquency and social-psychological processes.
... Adolescent criminal activities and delinquency have received substantial recognition as a pressing social problem since they are known to have short-and long-term consequences at the individual level (Moffitt et al. 2002;Nilsson and Estrada 2011) as well as widespread social and monetary burden for society (Welsh et al. 2008). At the individual level, in particular, adolescent delinquency is known to be negatively associated with human capital accumulation, such as lower academic achievement, school dropout, and fewer years of schooling completed (Hirschfield 2009;Hjalmarsson 2008;Kirk and Sampson 2013;Monk-Turner 1989;Nilsson and Estrada 2011;Ou et al. 2007;Webbink et al. 2013). These early exits from school and lower educational attainment are, in turn, closely linked with a range of long-term detrimental effects, including poor health status, recidivism, and lower occupational attainment (Baert and Verhofstadt 2015;Moffitt et al. 2002;Monk-Turner 1989;Sewell and Hauser 1975). ...
Article
This article examines the effect of delinquent peers on an individual’s criminal activity by leveraging quasi-experimental variation in exposure to peers, separating confounding and causal effects. In particular, we examine the role of wider peer networks (i.e., classmates) as a critical source of influence on adolescents’ delinquent behavior. Using a combined instrumental variables/fixed effects methodology, we address important methodological challenges in estimating peer effects. Results suggest that increasing the proportion of peers who engage in criminal activities by 5% will increase the likelihood an individual engages in criminal activities by 3 percentage points.
Article
Using administrative Danish Register Data to identify all twins born 1965–1982, this paper estimates heterogeneous effects of education on crime. Controlling for genetic and environmental factors, the completion of upper secondary education significantly lowers the probability of conviction for total, property, and violent crimes for males. Family factors matter—education lowers crime earlier in the life cycle for children of low educated parents and later in life for children of high educated parents. Exposure to crime during childhood similarly impacts the dynamics of the crime reducing effects of education across the life cycle. Examining different educational programs reveals completing high school is important for crime reduction, while vocational education has a smaller impact on crime. Results are robust to controlling for differences in early health, directly estimating reverse causality between education and crime, and using data on prison sentences instead of convictions.
Conference Paper
Full-text available
In the Netherlands dropping out from school is experienced as a serious problem. Not only for students in personal life but also for the society in which the students participate. Four schools for second vocational education develop curricula, based on design research, for students at risk within a top sport context. For developing new curricula teachers participate in ‘teacher design teams’ and schools cooperate with top sport organizations to develop the curriculum. The central goal of the design research is to develop a curriculum with content and teaching methods that encourage students to contribute student’s engagement, motivation to provide entering the labour market or starting with a new educational trajectory in VET. The general objective of this study is to contribute a sustainable curriculum in top sport context for students at risk, and to determine effects on students’ motivation and engagement. Therefore during 18 months, 150 students and 16 teachers are asked to join the study. In this research we use a mix method perspective as well qualitative as quantitative data will be collected. In this paper we report about the tension between the motives and objectives from the researcher and the believes, policy, ownership, and organizational power form the manager. We discuss their perspectives and experienced boundaries.
Article
Full-text available
We examined 48-hour test-retest reliability of the arrest and incarceration questions on the Risk Behavior Assessment (RBA; National Institute on Drug Abuse, 1993). Participants were 229 street-drug users recruited in 11 cities throughout the United States. Results revealed that lifetime arrest and incarceration items demonstrated good to excellent reliability. The 30-day arrest and incarceration items provided such poor reliabilities that they would yield unreliable data with limited research or clinical use. Future research needs to identify alternative items that can yield reliable data regarding recent arrest history; until then, it is recommended that the recent arrest items be used with caution.
Article
This article establishes that a low-dimensional vector of cognitive and noncognitive skills explains a variety of labor market and behavioral outcomes. Our analysis addresses the problems of measurement error, imperfect proxies, and reverse causality that plague conventional studies. Noncognitive skills strongly influence schooling decisions and also affect wages, given schooling decisions. Schooling, employment, work experience, and choice of occupation are affected by latent noncognitive and cognitive skills. We show that the same low-dimensional vector of abilities that explains schooling choices, wages, employment, work experience, and choice of occupation explains a wide variety of risky behaviors.
Article
The abstract for this document is available on CSA Illumina.To view the Abstract, click the Abstract button above the document title.
Article
The abstract for this document is available on CSA Illumina.To view the Abstract, click the Abstract button above the document title.
Article
Increasing recognition that official statistics on juvenile delinquency have serious limitations when used in comparative analyses of juvenile misbehaviors has led to the growing reliance on the use of the self-report technique. Using data from a self-report study of adolescents, the quality of the results of using this technique is appraised by several procedures. Evidence is presented which attests to the generally high validity of the reports and suggests the utility of the technique in both etiological and evaluation studies.