critical appraisal

O R I G I N A L P A P E R

‘‘Because You’re Mine, I Walk the Line’’? Marriage, Spousal Criminality, and Criminal Offending Over the Life Course

Marieke van Schellen • Robert Apel • Paul Nieuwbeerta

� Springer Science+Business Media, LLC 2012

Abstract Objectives This study is an analysis of the relationship between marriage and crime in a high-risk sample of Dutch men and women. Marriages are classified as to whether the spouse

had been convicted of a crime prior to the marriage, in order to ascertain if one’s criminal

career after marriage unfolds differently depending on the criminal history of one’s spouse.

Methods Data are from the Criminal Career and Life-Course Study, a random sample of all individuals convicted of a criminal offense in the Netherlands in 1977 (N = 4,615). Lifetime criminal histories for all subjects are constructed from age 12 to calendar year

2003. Official marriage records are also consulted, and the criminal history of all spouses

are similarly constructed. Fixed-effects Poisson models are estimated to quantify the

relationship between marriage, spousal criminality, and conviction frequency, controlling

for age, parenthood, prior conviction, and prior incarceration.

Results Among men, marriage reduces the frequency of criminal conviction, but only if the marriage is to a non-convicted spouse. Marriage to a convicted spouse, on the other

hand, is indistinguishable from singlehood—it neither discourages nor promotes criminal

behavior. Among women, marriage has a crime-reducing effect, regardless of the criminal

history of the spouse. A set of preliminary follow-up analyses suggests further that men

with more extensive criminal histories, and with more stable marriages, benefit in a more

pronounced way from marriage to a non-convicted spouse. However, even unstable

marriages to non-convicted spouses appear to reduce conviction frequency while they last.

Conclusions Marriage is indeed a salient transition in the criminal career, but there are important differences depending on the characteristics of the offender (gender, criminal

history), the characteristics of the spouse (criminal history), and the characteristics of the

M. van Schellen (&) Department of Sociology, Utrecht University, Heidelberglaan 2, 3584 CS Utrecht, The Netherlands e-mail: m.vanschellen@uu.nl

R. Apel School of Criminal Justice, Rutgers University, 123 Washington Street, Newark, NJ 07102, USA

P. Nieuwbeerta Department of Criminology, Leiden University, Steenschuur 25, 2311 ES Leiden, The Netherlands

123

J Quant Criminol DOI 10.1007/s10940-012-9174-x

marriage (duration). The authors conclude that while marriage matters, it does not nec-

essarily mean the end of a criminal career, and that processes of both partner selection and

partner influence deserve close attention by marriage-crime researchers. Qualifications of

the study’s findings include the use of conviction data from official sources, the use of a

sample of men and women who were all convicted of a crime at some point in their lives,

the study of legal marriage in the Netherlands, and the inability to measure potential

mechanisms for the observed marriage effects.

Keywords Marriage � Spousal criminality � Criminal convictions � Life-course criminology � Panel models

Introduction

Marriage has long been correlated with a variety of beneficial outcomes for the involved

individuals. Married individuals appear to have a larger network of help and support, show

less risky and unhealthy behavior, earn higher income, and possess more assets and wealth.

Accordingly, married persons tend to be happier, healthier, and better off financially

(Waite 1995; Waite and Gallagher 2000).

The idea of marriage as a protective institution takes a prominent place in criminology

as well. Marriage is considered to be an important transitional event that can reduce

criminal activity and even lead to desistance from crime (Sampson and Laub 1993). The

decline in criminal behavior is often attributed to the social bond that forms and

strengthens as a result of marriage: Spouses monitor each other’s behavior and do not want

to endanger their marital relationship by committing crime (Laub and Sampson 2003).

Recent empirical studies show that marriage is indeed associated with lower offending

levels (Bersani et al. 2009; Blokland and Nieuwbeerta 2005; Farrington and West 1995;

Horney et al. 1995; King et al. 2007; Laub and Sampson 2003; Piquero et al. 2002;

Sampson et al. 2006; Theobald and Farrington 2011).

Despite its prominence, the idea that marriage reduces crime is less straightforward than

assumed. Although on average the effect might be protective, the benefits of marriage

might not be homogenous and are likely to depend, among other things, on the criminal

history of the spouse. Marriage to a criminal partner could sustain or even stimulate an

offender’s engagement in criminal activities over time (Rhule-Louie and McMahon 2007).

For example, offenders probably have similar views on the appropriateness of criminal

offending, learn from each other, and pass on their criminal skills (Giordano et al. 2007;

Leverentz 2006; Simons et al. 2002). At the very least, marriage to a criminal spouse could

result in persistence in criminal offending, and at worst, escalation.

While empirical studies of marriage effects on criminal behavior have been accumu-

lating, the impact of spouses’ criminal careers has received far less attention (Rhule-Louie

and McMahon 2007). This is surprising, simply because attachments to unconventional

persons are considered to be among the most important predictors of delinquent behavior

during adolescence: Adolescents who have delinquent friends are more likely to become

delinquent, and commit more crimes than adolescents without deviant connections (Haynie

et al. 2005; Simons et al. 2002). Although it has been argued that adolescents are more

sensitive to social influence processes (Warr 2002), this finding does clearly demonstrate

that social ties are not necessarily protective, but can stimulate criminal behavior as well.

To gain more knowledge about the development of individual criminal careers, insight

is needed into the criminal behavior of spouses. One of the reasons for the lack of empirical

J Quant Criminol

123

research on the effects of partners’ criminal history is that the requirements for the design

of these studies are substantial. First, longitudinal information on the development of

criminal behavior is necessary. Second, this information is needed for both marriage

partners. Third, very long periods of observation are required in order to examine research

subjects beyond adolescence into adulthood. Thus far, virtually no study meets these

requirements.

In this study, we estimate quasi-experimental models of the relationship between

marriage and criminal conviction, and investigate whether this relationship varies by the

spouse’s criminal history at the time of marriage. We employ data from a unique long-term

study of a Dutch conviction cohort and their marriage partners: The Criminal Career and

Life-Course Study (CCLS; Nieuwbeerta and Blokland 2003). The CCLS contains data on

the officially registered criminal careers of 4,615 Dutch offenders and their marriage

partners, covering ages 12–72. The aim of this study is to contribute to the current literature

in several ways. First, this study will be the first to investigate the life-long criminal careers

of a large number of offenders and their marriage partners. Second, the CCLS contains data

on the exact timing of marriages, convictions, and periods of incarceration over the entire

life span. Therefore, we can clearly distinguish partner influences from partner selection

processes that took place before marriage. Third, the data allow us to investigate the

criminal careers of male as well as female offenders.

It should be noted that the CCLS does not contain data on never-convicted individuals

and their spouses. This research thus concerns the effects of marriage for a sample of

individuals who have been convicted at some point during their lives. Within this group of

convicts, individuals differ in the timing, number, and seriousness of crimes committed.

Because the CCLS does not provide information about other relationship types than

marriage (e.g., cohabitation), the focus of this study is on the effects of marriage and not on

partner relationships in general.

Marriage, Spousal Criminality, and Crime: Desistance or Persistence?

With the rise of developmental and life-course criminology (Piquero et al. 2003), a

growing number of studies have focused on the unfolding of individual criminal careers

over the life span. These studies show that offending rates tend to increase gradually during

childhood, rise more sharply during adolescence and then begin to decline steadily as

individuals enter adulthood. Although there appears to be a group of persistent offenders

that continues committing crimes far into adulthood, most persons tend to stop their

criminal careers after adolescence (Blokland et al. 2005; Laub and Sampson 2003).

Desistance from crime is often explained by the fact that persons experience crime-

inhibiting life-course transitions as they navigate the bridge from adolescence to adulthood.

The formation of marital relationships has been argued to play a key role in this desistance

process (Laub and Sampson 2003).

Several theoretical mechanisms have been proposed to explain why the transition to

marriage would reduce criminal offending (see also Sampson et al. 2006: 467–468). These

mechanisms can be grouped in four different categories: Social bond, routine activities,

social learning, and cognitive transformation. Originally, all explanations have centered on

the development of individual offending trajectories, and neglected the criminal careers of

partners. This is unfortunate, since several studies have shown that offenders have a higher

chance to form relationships with partners who are criminally active as well, i.e., they mate

assortatively (Moffitt et al. 2001; Simons et al. 2002). Although the institution of marriage

J Quant Criminol

123

might be protective in itself, it could very well be the case that the effects of marriage

depend on the criminal history of the spouse (Giordano et al. 2002). Marriage could

explain desistance as well as persistence in crime—depending on the spouse’s involvement

in crime.

Below we discuss the main theoretical mechanisms, and derive hypotheses on the

effects of marrying a non-criminal spouse versus criminal spouse. In doing so, we pay

explicit attention to gender differences in the effects of marriage and spousal criminality.

Although the different theoretical perspectives are not fundamentally incompatible, they

differ in their central focus. Some explain the crime-reducing effect of marriage by pro-

cesses external to the individual (e.g., changes in opportunities to commit crime), while

others focus on internal factors (e.g., changes in preferences). Although the data do not

allow us to test the various underlying mechanisms, they give us more insight in why a

relationship between marriage, spousal criminality, and crime can be expected.

Social Bond

Marriage may change criminal offending because it strengthens social bonds to conven-

tional society. This idea has a prominent place in Sampson and Laub’s (1993) age-graded

theory of informal social control. Spouses monitor and attempt to control each other’s

behavior, and tend to discourage activities that do not pay off in the long run, like hanging

out with deviant friends. Also, especially if ties are strong and stable, marital relationships

create obligations and restraints that increase the costs of offending. Over time, as com-

mitment and investment in the relationship grows, there are fewer incentives to commit

crime, because more is at stake (Laub et al. 1998).

Whether the effects of marriage are protective may, however, strongly depend on the

criminal history of the spouse to whom one is attached. We nuance Sampson and Laub’s

theoretical ideas in two ways. First, marriages are not necessarily strong if both spouses are

involved in crime (Simons et al. 2002). The idea that marital ties are of higher quality and

less likely to dissolve if spouses resemble each other, is prominent in family sociology

(Brines and Joyner 1999; Kalmijn 1998). It is, however, questionable whether this also

applies when it concerns similarity in criminal behavior. For example, if both partners are

involved in crime, they are both likely to have personal traits and to be involved in situ-

ations that undermine the quality and stability of the marriage (Western 2006: 5). Second,

conventional behavior is not necessarily encouraged if both partners are criminally

involved. If their spouses are not involved in crime, offenders might indeed risk their

relationship by violating the law. If spouses have a criminal history as well, offending is

likely to be a conventional way of behaving, which is less likely to be discouraged and

does not threaten the continuation of the relationship.

Routine Activities

A different interpretation of marriage’s role has been given by Warr (1998) (see also Laub

and Sampson 2003). Warr (1998) emphasizes the role of peers in criminal offending. The

decline in crime following marriage is caused by a decrease in time spent with (delinquent)

friends and an accompanying reduction in opportunities and reinforcement for criminal

behavior. Married people spend more time in each other’s company, and stay home

together more often. Although non-criminal partners can indeed be expected to promote a

conventional lifestyle (Simons et al. 2002: 404), this is less likely to apply to criminal

partners. Criminal partners are likely to be enmeshed in a criminal network themselves,

J Quant Criminol

123

and may therefore stimulate contact with other antisocial individuals and bring one to risky

places at risky moments.

Socialization

Another mechanism that may underlie the relationship between marriage and criminal

offending can be derived from differential association and social learning theories. These

theories state that behavior is learned through social interaction within a cohesive and

intimate group, where criminal norms, values, and knowledge are passed on through

ongoing socialization processes (Akers 1973; Warr 2002). Although these theories have

traditionally been used to explain the influence of delinquent peers, the same ideas can be

applied to marital relationships (see also Haynie et al. 2005; Simons et al. 2002). In

contrast to the earlier discussed mechanisms, socialization theories make explicit that

intimate associations can have a positive or negative influence on offending depending on the normative orientation of others. Although marrying a non-criminal partner is likely to

lead to socialization in a conventional law-abiding environment, marrying a criminal

partner is likely to sustain or stimulate an individual’s criminal activities over time (Rhule-

Louie and McMahon 2007). Offenders probably have similar views on the appropriateness

of criminal behavior, learn from each other, pass on their criminal skills, or may even start

co-offending and become real ‘‘partners in crime.’’ It has been suggested that this form of

socialization may be more powerful and important for females. As they would be more

oriented toward relationships, their behavior would be more frequently determined by the

behavior of their partners (Moffitt et al. 2001; Steffensmeier and Allan 1996).

Cognitive Transformations

Finally, marriage may also lead to changes in criminal offending, because it changes one’s

sense of self through cognitive transformations (Giordano et al. 2002; Sampson et al. 2006:

468). In contrast to the earlier mentioned mechanisms, processes internal to the offender

are emphasized. Marriage—if accompanied by an openness to and readiness for mean-

ingful change—can lead to desistance, because it fosters pro-social role modeling. The

accompanying cognitive transformations result in a change in the meaning and salience of

criminal behavior: Criminal behavior is no longer seen as positive, viable, or personally

relevant (Giordano et al. 2002). Although marrying a non-criminal spouse may indeed lead

to a pro-social and responsible lifestyle, criminal partners are less likely to function as

positive role models and therefore may undermine conventional identity change.

Hypotheses

Although we are not able to directly test the underlying mechanisms, all aforementioned

mechanisms lead to the expectation that marriage in general reduces criminal offending. It

not only reduces the preference to commit crimes, but also leads to fewer opportunities to

offend. We nuance this hypothesis in two ways. First, we assume that the effect of marriage

depends on the criminal behavior of the spouse. Being married to a non-criminal spouse is

expected to lead to a decrease in the number of offenses, and being married to a criminal

spouse to persistence in crime. On the one hand, we may find no change in the level of

offending compared to singlehood. On the other hand, partners may influence each other in

such a way that they stimulate criminal activity, resulting in an increase in the number of

J Quant Criminol

123

offenses. Second, we expect gender differences in the effect of marriage and spousal

criminality. It has been argued that men would benefit more from marriage in general,

because they are more likely to marry non-criminal spouses than women are. In other

words, ‘‘men marry up’’ (Laub and Sampson 2003). However, according to the sociali-

zation perspective, women might be more influenced by the behavior of their partners.

Therefore, we expect the crime-reducing effects of marrying a non-criminal spouse and the

crime-stimulating effect of marrying a criminal spouse to be even larger for women.

Earlier Studies and Their Limitations

Although a growing number of studies have investigated the relationship between marriage

and crime, only a few of them considered the impact of partners’ criminal behavior. Using

data from Glueck and Glueck’s classic study of criminal careers, Sampson et al. (2006)

investigated 226 delinquent men followed prospectively from adolescence to age 32.

Although men with criminal or deviant wives displayed higher criminal offending rates,

within-individual estimates of the effects of marriage showed that it nevertheless signifi-

cantly reduced criminal involvement, controlling for duration of the marriage, marital

attachment, and spousal criminal record.

Moffitt et al. (2001) investigated the effects of partner relationships on antisocial

behavior among a birth cohort of 360 individuals followed from age 13 to 21. The analyses

revealed that women were more likely to persist in crime (measured as self-reported

antisocial behavior at age 21) when they formed unions with antisocial men. However,

antisocial men continued to be antisocial, regardless of whether their female partner was

antisocial at the time of the relationship.

Simons et al. (2002) simultaneously tested the impact of delinquent friends and partners

on delinquent behavior among 236 young adults. Results showed that having an antisocial

romantic partner was related to higher levels of criminal behavior both directly as well as

indirectly, through its effect on the quality of the romantic relationship and involvement

with deviant friends. These relationships were significant for both men and women,

although having an antisocial partner was associated with criminality more strongly for

women. Using data from a nationally representative sample of school-going adolescents,

Haynie et al. (2005) reached similar conclusions. Romantic partners’ deviance was more

strongly related to females’ involvement in minor deviance, although gender did not

condition the strength of the relationship between the romantic partners’ serious delin-

quency and the respondents’ serious delinquency.

The study by Woodward et al. (2002) was unique in that it included single as well as

romantically involved individuals. Individuals involved with a non-deviant partner had

lower rates of offending at age 21 than those with no partner, while those without a partner

had lower rates of offending at age 21 than those involved with a deviant partner. Similar

results were found for men and women.

The most recent empirical study has been conducted by Capaldi et al. (2008). Using a

sample of at-risk men (N = 191), the results showed that a partner’s antisocial behavior was related to both onset and persistence of arrests, even when controlling for deviant peer

associations. In contrast to earlier studies, respondents’ and partners’ offending behavior

were not measured at the same time, enabling stronger conclusions about the direction of

influence. Unfortunately, data on partners’ criminal history before marriage was lacking.

Another drawback is that the analysis only investigated respondents’ arrests in the first year

after the relationship was formed.

J Quant Criminol

123

In sum, most studies have found that a partner’s offending is associated with an increase

in the referent subject’s own offending. In some cases, this relationship is stronger for

women (Capaldi et al. 2008; Haynie et al. 2005; Moffitt et al. 2001; Simons et al. 2002).

Other studies have found that single individuals are even better off than those involved

with a delinquent or criminal partner (Woodward et al. 2002). This finding is inconclusive,

however, because there is other evidence that marriage has protective effects irrespective

of the criminal history of the spouse (Sampson et al. 2006).

Although earlier studies have made important contributions to the marriage-crime lit-

erature, they have several limitations. First, they lack longitudinal information on partners’

criminal histories. As relationship status and partner criminality trend to be measured at the

same time, there are limits to causal inferences: Any association between partners’

criminal behavior could result from selection processes that take place before relationship

formation. Second, most studies limit their focus to adolescence and early adulthood. Yet

partnerships are particularly salient during adulthood, not to mention that long-term effects

of marriage are impossible to study. Third, earlier studies investigated relationships of

varying durations, and at various stages of attachment (e.g., married, cohabiting, unmarried

but committed relationship). In addition, the (marital) relationships under study might be

the first relationship, but also the second or even the third. Although these different types of

relationships might very well have different effects, they are not analyzed separately

(partly because of small sample sizes).

Data

In this study, we use data from the CCLS (Nieuwbeerta and Blokland 2003). The CCLS

subjects were selected by taking a four-percent sample of all cases of criminal offenses

tried in the Netherlands in 1977, with an oversampling of less common—mostly serious—

offenses (e.g., murder, rape, drug offenses) and an undersampling of common offenses

(e.g., drunk driving). This resulted in a total sample of 4,615 offenders (4,191 men and 424

women).

Extracts (‘‘rap sheets’’) from the General Documentation Files (GDF) of the Criminal

Records Office were used to reconstruct the entire criminal careers of all 4,615 research

subjects from the age of 12—the minimum age of criminal responsibility in the Nether-

lands—until calendar year 2003. In the Netherlands, individuals are not given a ‘‘clean

slate’’ upon becoming an adult. The extracts thus contain information on both juvenile and

adult offenses. Although the GDF contain information on all offenses that led to any type

of judicial action, here we use only information on those offenses that were either followed

by a conviction or a prosecutorial disposition because of policy reasons. 1

We therefore

exclude cases that resulted in an acquittal or a prosecutorial disposition because of

1 In the Dutch criminal justice system, the public prosecutor has the discretionary power not to prosecute all

cases forwarded by the police. First, the public prosecutor may decide to drop the case if prosecution would probably not lead to conviction due to lack of evidence, or for technical considerations (procedural or technical waiver). Second, the public prosecutor is authorized to waive prosecution ‘‘for reasons of public interest’’ (waiver for policy considerations). The Board of Prosecutors-General has issued national prose- cution guidelines under which a public prosecutor may decide to waive a case for policy reasons. In some cases measures other than penal sanctions are preferable or more effective, or prosecution would be dis- proportionately unjust or ineffective in relation to the nature of the offense or the offender, or prosecution would be contrary to the interest of the state or the victim (Tak 2003).

J Quant Criminol

123

insufficient evidence. This means that the criminal offenses which are studied have most

likely been committed by the offenders.

To measure the unfolding of life circumstances, the judicial data were supplemented

with data from the population registration records. These records contain information on

the exact timing of marriage, divorce, fertility, and mortality. All data are derived from

official sources, which means that they are of high quality and have very few missing

values. Because the mean age of the research subjects was 27.8 years in 1977 (med-

ian = 25 years), we have data on convictions and life circumstances that reach far into

adulthood for a large portion of the sample. For example, while the sample was followed

until a mean age of 43.6 years (median = 44 years), the age of follow-up for almost six

percent of the sample is 70 or older. 2

In preparation for this study, the CCLS was supplemented with data on the complete

criminal careers of all of the marriage partners of the research subjects from age 12 to

calendar year 2007. The population registration records revealed that 74.5 % (N = 3,437) of the original 4,615 research subjects married on at least one occasion, to a total of 4,409

partners. This supplement to the CCLS allows us to determine the exact timing of marriage

and, for all research subjects and their married partners, the exact timing of criminal

offenses, the type of offenses committed, and periods of prison confinement.

To estimate the empirical models of conviction frequency, we eliminate the ‘‘criterion

conviction,’’ referring to the conviction that brought each subject into the CCLS sampling

frame. Because of the way that the CCLS data were collected, all subjects are convicted at

least once during their lives. For most offenders (96.8 %), this conviction was in either

1976 or 1977. If subjects who marry have no convictions prior to their first marriage then

they must, by construction, have at least one conviction during or after their first marriage.

This could result in the estimation of a criminogenic effect of marriage that is artifactual.

To avoid this, we exclude the criterion conviction altogether.

We also focus the analysis on the first marriage. 3

By focusing on first marriages, we

avoid having to account for feedback effects between marriage and crime whereby (a

criminal) marriage affects the likelihood of crime, which in turn affects the likelihood of (a

criminal) marriage. Although the focus on first-time marriage limits the generalizability of

our findings, we eliminate this kind of endogeneity bias as a source of confounding of the

empirical estimates (see also Nieuwbeerta et al. 2009: 232).

Analytic Strategy

We are interested in what effect marriage to a non-convicted or convicted spouse has on an

individual’s post-marriage conviction frequency. One empirical challenge is that marriage

and partner selection are not randomly determined. Individuals who marry are likely to

have different characteristics than persons who do not marry, and individuals who marry

2 The mean age in 1977 is older than the peak of the well-known age-crime curve. This is because the

CCLS contains data on criminal convictions rather than arrests, which will lead to a slightly older sample. On average, the CCLS offenders have been followed for 32 years (min = 1, max = 60). The sample size varies across ages. For example, at age 12, the data contain information on all 4,615 individuals. The sample drops at age 22 (4,605), age 32 (4,547), age 42 (4,255), age 62 (788), and age 72 (245). 3

Most of the subjects in our sample—76 %—marry only once.

J Quant Criminol

123

convicted spouses are likely to have different characteristics than offenders who marry

non-convicted spouses. This is known as the selection problem, and it can cloud causal

interpretations of correlations between marriage and criminality, since differences in crime

risk probably exist even in the absence of marriage. One solution to the selection problem

is to attempt to adjust away these differences by including as many control variables as

possible in a regression or propensity score model. This represents a ‘‘selection on

observables’’ approach to causal effect estimation (see Heckman and Hotz 1989). Although

the CCLS data are unusually rich with regard to offenders’ marital and criminal histories,

as with most official sources of data, information on other variables known to be correlated

with marriage and crime (e.g., personality characteristics, educational attainment, socio-

economic status) is unfortunately unavailable. Yet even with an exhaustive set of such

control variables, the selection problem would persist because differences between indi-

viduals are always partly unobserved. A unique strength of the CCLS data, with lifetime conviction histories on all offenders, is the ability to estimate the effect of marriage and

spousal criminality on conviction frequency in the presence of ‘‘selection on unobserva-

bles’’ (Heckman and Hotz 1989). The most rigorous way to do so is through the use of a

fixed-effects model.

Fixed-effects models adjust for so-called ‘‘unobserved heterogeneity’’ by restricting

attention to within-individual change in marriage and crime over time. The model thus

eliminates biases that are attributable to any source of variation in criminality that

remains constant over time, for example, biological or genetic differences (Halaby

2004; Johnson 1995). In other words, any estimate of the ‘‘marriage effect’’ on crime is

purged of enduring differences between individuals. Yet fixed-effects models still

produce inconsistent estimates in the presence of ‘‘dynamic selection,’’ or omitted time-

varying regressors that are correlated with joint changes in marriage and crime. We

explicitly measure two such characteristics (age, parenthood), and control as rigorously

as possible for prior offending (convictions, incarceration). By using fixed-effects

models in this way, we aspire to take advantage of the strengths of the CCLS data (i.e.,

the unique longitudinal data on time-varying variables) and compensate as much as

possible for the weaknesses (i.e., the lack of relevant time-stable confounding vari-

ables).We return later to a more thorough discussion about causal identification in

marriage-crime studies.

To estimate the fixed-effects models, a person-year file is constructed with records

containing information on each individual in each calendar year. For every person the

records begin at age 12 and end in the year 2003 (the end of data collection), in the last

year of the first marriage (divorced subjects are excluded in all years after their first

marriage), or in the year prior to death (in this way we account for ‘‘false desistance’’

caused by mortality). The fully constructed data file contains information on 150,315

person-years for the 4,615 CCLS subjects.

In our empirical models the dependent variable, Yit, is a discrete random variable representing a count of the number of convictions received by subject i (i = 1,…,N) in calendar year t (t = 1,…,Ti). It is distributed Poisson with density:

fðYit Xit; Sitj Þ¼ expð�SitkitÞðSitkitÞYit

Yit !

where Sit represents the inverse of a subject’s exposure, or the ‘‘street time’’ in a given calendar year, measured as the proportion of the year not confined in a correctional

J Quant Criminol

123

institution. 4

By controlling for the opportunity to commit crimes, we eliminate the pos-

sibility of false desistance attributable to incarceration (see Piquero et al. 2001).

The analysis begins with the baseline model that controls for characteristics that have

been demonstrated to influence the development of criminal behavior:

ln kit ¼ a1f Ageitð Þþ a2Childit þ a3Coni;t�1 þ a4AccumConi;t�2 þ a5Inci;t�1 þ a6AccumInci;t�2 þ ui

Ageit is modeled as a cubic to capture age-related changes in the rate of conviction for the entire sample. Childit is a time-varying dummy variable for whether the subject has one or more children under the age of 18. We also include several time-varying measures of

criminal history in the models. Two measures of prior convictions are added, including a

dummy indicator for having been convicted in the previous calendar year (Coni,t-1) as well as the total number of convictions accumulated as of 2 years ago (AccumConi,t-2). Two measures of imprisonment are also added, including a dummy indicator for having been

incarcerated in the previous calendar year (Inci,t-1) and the total accumulated time spent in prison as of 2 years ago (AccumInci,t-2).

5 The individual effect, ui, captures unobserved

heterogeneity in conviction risk, or that portion of the total variation in conviction that is

unobserved (and unmeasured) but is stable over time. The individual effect or error

component, ui, is modeled as fixed in this analysis. To the variables that comprise the baseline specification, henceforth denoted for the

purpose of economy as akXitk, the first model of substantive interest adds a time-varying indicator for marriage. The model is thus specified as follows:

ln kit ¼ XK

k¼1 ak Xitk þ bMarriedit þ ui ð1Þ

Marriedit is coded ‘‘1’’ in each year that subjects are married and ‘‘0’’ in all earlier years. If marriage promotes desistance from crime, we expect b to be negative. Recall that, since all person-years after the first marriage ends are excluded, this indicator quantifies the

effect of one’s first marriage on conviction frequency.

In our second model we take into account the criminal history of the spouse by adding a

second marriage indicator in the following manner:

ln kit ¼ XK

k¼1 ak Xitk þ bMarriedit þ cMarriedConvictit þ ui ð2Þ

The indicator Marriedit is coded as before. The new indicator MarriedConvictit is coded ‘‘1’’ in the years that subjects are married to a spouse who had a criminal conviction

preceding the marriage. The paired marital states—Marriedit and MarriedConvictit—are not mutually exclusive. Therefore, b represents the effect of being married relative to

4 The fixed-effects Poisson model proceeds by maximizing the conditional likelihood, where conditioning is

achieved by summing across each individual’s Ti observations on the dependent variable. This technically makes it a conditional fixed-effects model. Therefore it necessarily excludes individuals whose observations (here, total number of convictions) sum to zero during the period of observation, resulting in the loss of degrees of freedom. In our model, 3,356 of the 4,191 men (80 %) are retained, whereas 173 of the 424 women (41 %) are retained. 5

The first order-lags, Coni,t-1 and Inci,t-1, capture recency in criminal offending and are binary while the second-order lags, AccumConi,t-2 and AccumInci,t-2, capture the accumulated criminal history and are non- binary. By including both types of variables we can distinguish the short-term, state-dependent effects of criminal conviction and incarceration from the long-term effects.

J Quant Criminol

123

singlehood and c represents the additional effect of being married to a convicted spouse relative to being married to a non-convicted spouse. In other words, c is a contrast. In order to recover the impact of marriage to a convicted spouse relative to singlehood, these two

coefficients must be summed together and tested against zero.

If marriage to a non-convicted spouse leads to desistance from crime, then b will be negative and significant. If, however, marriage to a convicted spouse leads to persistence in

crime, then we expect c to be positive and significant. Additionally, if marriage to a convicted spouse exacerbates crime relative to remaining single, then b ? c will be positive and significant. If, on the other hand, marriage to a convicted spouse does not

differ from singlehood, then b ? c will not be significantly different from zero. In a third and final set of models, we evaluate whether the effects of marriage and

spousal criminality depend on the offender’s own criminal history as well as on the length

of marriage. These analyses are limited to the male CCLS offenders. We provide a more

extensive description of these models in later sections.

Results

The aim of this paper is to investigate the extent to which the effects of first marriage

depend on the criminal history of the spouse. Before we turn to our panel models, we begin

with descriptive statistics, provided in Table 1. It appears that three in four male offenders

(73.2 %) and six in seven female offenders (86.5 %) marry before calendar year 2003. Of

these married offenders, only one in 20 males (5.5 %) but one in four females (26.7 %)

marry a convicted spouse. The fact that women are far overrepresented among individuals

who marry convicted spouses is consistent with the observation that ‘‘women marry down’’

while ‘‘men marry up’’ when it comes to crime (Laub and Sampson 2003). This may

indicate that there are simply more men with criminal records in the marriage market, and

therefore the chances are higher for a woman to marry an offender. Also, there might be

stronger selection processes at work for female offenders compared to male offenders.

Because of the relative rarity of female criminal behavior, a convicted female is likely to

be more deviant or ‘‘pathological’’ with respect to the unobserved characteristics correlated

with crime (e.g., lower self-control, lower socioeconomic status, personality disorders).

Finally, it might be the case that some of the female partners actually do have a criminal

history, but are convicted less often (e.g., because they commit less serious crimes).

Examining criminal backgrounds, male subjects who marry a convicted spouse in their

first marriage accumulate twice as many convictions over their lifetime, on average, than

subjects who marry a non-convicted spouse (21.1 vs. 10.2 convictions). Correspondingly,

they are more likely to have received an early first conviction (35.5 vs. 24.8 %) and to have

ever been incarcerated (65.7 vs. 43.9 %). The same pattern holds true for female subjects.

Females who marry a convicted partner in their first marriage have more lifetime con-

victions (4.0 vs. 3.4), a higher risk of early first conviction (12.2 vs. 5.9 %), and a higher

incarceration risk (18.4 vs. 13.4 %).

Although the mean age of first marriage differs by no more than a couple of years (and

only for male subjects), the length of the first marriage is substantially shorter for subjects

who marry a convicted spouse compared to a non-convicted spouse, indicative of greater

marital instability (males: 13.4 vs. 17.1 years; females: 14.7 vs. 20.2). But note that

marriages are quite durable, irrespective of the criminal behavior of the spouse. Most of the

CCLS offenders grew up in an era in which divorce was less common than today. As a

result, individuals were more likely to stay married. Subjects who marry a convicted

J Quant Criminol

123

T a

b le

1 D

e sc

ri p ti

v e

st a ti

st ic

s b y

g e n d e r

V a ri

a b

le M

a le

su b

je c ts

(N =

4 ,1

9 1

) F

e m

a le

su b

je c ts

(N =

4 2

4 )

F ir

st m

a rr

ia g

e to

a …

F ir

st m

a rr

ia g

e to

a …

N o

n -c

o n v

ic te

d sp

o u se

C o n

v ic

te d

sp o

u se

N e v

e r

m a rr

ie d

N o

n -c

o n v

ic te

d sp

o u se

C o n v ic

te d

sp o u se

N e v e r

m a rr

ie d

N 2

,9 0 1

1 6

9 1

,1 2 1

2 6

9 9

8 5

7

% o

f sa

m p

le 6

9 .2

4 .0

2 6

.7 6

3 .4

2 3

.1 1

3 .4

Y e a rs

in st

u d

y 4

0 .9

(9 .4

) [3

9 ]

4 0

.2 (9

.3 )

[3 7

] 3

5 .6

(8 .6

) [3

5 ]

4 6

.2 (1

0 .5

) [4

6 ]

4 1

.9 (9

.3 )

[4 2

] 3

8 .2

(9 .8

) [3

6 ]

Y e a rs

in a n

a ly

si s

sa m

p le

a 3

1 .6

(1 2

.5 )

[3 1

] 3

0 .5

(1 2

.2 )

[3 1

] 3

5 .6

(8 .6

) [3

5 ]

3 2

.2 (1

5 .0

) [3

1 ]

2 6

.4 (1

2 .7

) [2

4 ]

3 8

.2 (9

.8 )

[3 6

]

P e rs

o n a l

b a c kg

ro u n d

Y e a r

o f

b ir

th 1

9 4

8 .4

(1 0

.9 )

[1 9

5 1 ]

1 9

4 9

.6 (1

0 .2

) [1

9 5

3 ]

1 9

5 3

.0 (8

.0 )

[1 9

5 5

] 1

9 4

2 .3

(1 1

.9 )

[1 9

4 4 ]

1 9

4 7

.9 (9

.3 )

[1 9

4 8 ]

1 9

5 0 .6

(9 .5

) [1

9 5

4 ]

N o

n -D

u tc

h (%

) 1

1 .1

9 .5

1 7

.0 1

7 .6

8 .2

2 1

.1

D e a th

p ri

o r

to c a le

n d a r

y e a r

2 0 0 3

(% )

1 5

.0 1

4 .8

1 9

.2 1

6 .5

8 .2

1 7

.5

C ri

m in

a l

b a c kg

ro u n d

A g

e o

f fi

rs t

c o

n v

ic ti

o n

2 1

.5 (8

.8 )

[1 9

] 1

9 .2

(7 .6

) [1

7 ]

1 8

.9 (6

.3 )

[1 7

] 3

0 .5

(1 1

.3 )

[2 9

] 2

4 .3

(8 .0

) [2

3 ]

2 3

.1 (7

.9 )

[2 0

]

A c c u m

u la

te d

c o

n v

ic ti

o n

s 1

0 .2

(1 4

.9 )

[5 ]

2 1

.1 (2

1 .0

) [1

5 ]

1 7

.5 (2

0 .8

) [1

0 ]

3 .4

(6 .5

) [1

] 4

.0 (5

.2 )

[2 ]

6 .4

(1 6

.1 )

[2 ]

E a rl

y fi

rs t

c o n

v ic

ti o

n (\

1 7

y e a rs

) (%

) 2

4 .8

3 5

.5 3

3 .6

5 .9

1 2

.2 5

.3

E v

e r

in c a rc

e ra

te d

(% )

4 3

.9 6

5 .7

6 1

.6 1

3 .4

1 8

.4 2

1 .1

A c c u m

u la

te d

p ri

so n

ti m

e (m

o n th

s) b

1 4

.8 (2

3 .5

) [5

.6 ]

2 1

.2 (2

6 .6

) [1

2 .3

] 2

1 .6

(2 7

.3 )

[1 0

.0 ]

8 .4

(1 4

.0 )

[4 .0

] 6

.0 (1

1 .5

) [1

.9 ]

1 9

.0 (4

5 .9

) [1

.2 ]

E v

e r

c o

n v

ic te

d o

f v

io le

n t

c ri

m e

(% )

5 0

.5 6

6 .9

5 8

.8 1

4 .9

1 9

.4 1

5 .8

E v

e r

c o

n v

ic te

d o

f p

ro p

e rt

y c ri

m e

(% )

6 4

.0 7

9 .9

7 5

.2 6

3 .6

6 6

.3 5

4 .4

E v

e r

c o

n v

ic te

d o

f d

ru g

c ri

m e

(% )

1 2

.9 3

0 .2

3 1

.5 6

.7 1

0 .2

2 2

.8

M a rr

ia g e

b a c k g ro

u n d

A g

e o

f fi

rs t

m a rr

ia g

e 2

6 .0

(5 .9

) [2

5 ]

2 8

.9 (8

.4 )

[2 7

] 2

3 .3

(5 .6

) [2

2 ]

2 3

.1 (6

.4 )

[2 1

]

L e n

g th

o f

fi rs

t m

a rr

ia g

e (y

e a rs

) 1

7 .1

(1 2

.3 )

[1 5

] 1

3 .4

(1 1

.9 )

[9 ]

2 0

.2 (1

5 .2

) [1

7 ]

1 4

.7 (1

2 .1

) [9

]

J Quant Criminol

123

T a

b le

1 c o

n ti

n u e d

V a ri

a b

le M

a le

su b

je c ts

(N =

4 ,1

9 1

) F

e m

a le

su b

je c ts

(N =

4 2

4 )

F ir

st m

a rr

ia g

e to

a …

F ir

st m

a rr

ia g

e to

a …

N o

n -c

o n v

ic te

d sp

o u se

C o n

v ic

te d

sp o

u se

N e v

e r

m a rr

ie d

N o

n -c

o n v

ic te

d sp

o u se

C o n v ic

te d

sp o u se

N e v e r

m a rr

ie d

C o

n v

ic te

d 5

y e a rs

b e fo

re m

a rr

ie d

(% )

4 9

.7 7

0 .4

1 3

.1 2

1 .4

T o

ta l

# c o

n v

ic ti

o n

s a t

m a rr

ia g

e 4

.1 (6

.9 )

[2 ]

1 0

.1 (1

2 .2

) [6

] 0

.6 (2

.1 )

[0 ]

1 .1

(2 .4

) [0

]

M e a n

s o

f b

in a ry

v a ri

a b

le s

a re

sh o

w n

a s

p e rc

e n

ta g

e s.

F o

r ra

ti o -l

e v

e l

v a ri

a b

le s,

st a n

d a rd

d e v

ia ti

o n

s a re

p ro

v id

e d

in p

a re

n th

e se

s a n

d m

e d ia

n s

a re

sh o

w n

in b

ra c k

e ts

a M

a rr

ie d

in d

iv id

u a ls

h a v

e a n

a rt

ifi c ia

ll y

sh o

rt e r

le n

g th

o f

fo ll

o w

-u p

in th

e a n a ly

si s

sa m

p le

b e c a u

se th

e y

a re

e x

c lu

d e d

in a ll

y e a rs

a ft

e r

th e ir

fi rs

t m

a rr

ia g

e d

is so

lv e s.

H o

w e v

e r,

th e

d e sc

ri p ti

v e

st a ti

st ic

s p ro

v id

e d

in th

is ta

b le

u se

a ll

a v a il

a b le

y e a rs

b O

n ly

in d

iv id

u a ls

w h

o w

e re

e v

e r

in c a rc

e ra

te d

a re

in c lu

d e d

J Quant Criminol

123

spouse are also far more likely to have been convicted themselves in the 5 years prior to

marriage (males: 70.4 vs. 49.7 %; females: 21.4 vs. 13.1 %) and to have accumulated more

convictions at the time of their marriage (males: 10.1 vs. 4.1; females: 1.1 vs. 0.6),

implying a substantial degree of assortative mating with respect to criminal behavior.

Interestingly, female subjects who never marry tend to be more crime prone than their

counterparts who marry, and in many instances, more crime prone than those who marry a

convicted spouse. For example, perpetually single females exhibit a younger age of first

conviction, a higher volume of lifetime convictions, and a higher lifetime incarceration

risk. The same is not necessarily true for male subjects, however. Male offenders who

marry a convicted spouse are uniformly more crime prone relative to their married and

never-married counterparts. Thus, first marriage to a convicted spouse tends to be the

deviant marital state for men, whereas singlehood is the deviant marital state for women.

Marriage, Spousal Criminality, and Conviction Frequency

Results from the fixed-effects Poisson models are provided in Table 2. Recall that these

models estimate the effect of marriage and spousal criminality on the number of convic-

tions per year of street time. Model 1 estimates the impact of marriage on conviction

frequency irrespective of the criminality of the spouse, and is equivalent to Eq. 1. Model 2

estimates the impact of spousal criminality on conviction frequency, and is equivalent to

Eq. 2. All models are estimated separately for male and female subjects. Note that in

follow-up models that are not shown, we lagged the marriage indicators by 1 year to ensure

temporal priority of marriage vis-à-vis criminal conviction. The results from these models

were virtually identical to those reported in Table 2.

In Model 1, as expected from a variety of theoretical perspectives, being married is

associated with a significant decrease in conviction frequency relative to being single. This

is true for male as well as female offenders. Exponentiating each of the coefficients (eb)

Table 2 Fixed-effects Poisson models of the impact of first marriage on conviction frequency, by gender

Variable Male subjects (N = 3,356) Female subjects (N = 173)

Model 1 Model 2 Model 1 Model 2

Age .487 (.010)*** .489 (.010)*** .574 (.083)*** .577 (.083)***

Age squared -.013 (.000)*** -.013 (.000)*** -.012 (.002)*** -.012 (.002)***

Age cubed .000 (.000)*** .000 (.000)*** .000 (.000)*** .000 (.000)***

Have a child -.157 (.021)*** -.154 (.021)*** -.026 (.140) -.006 (.141)

Convicted last year .470 (.013)*** .469 (.013)*** .267 (.105)* .257 (.105)*

Accumulated convictions -.013 (.001)*** -.013 (.001)*** -.130 (.015)*** -.129 (.015)***

Imprisoned last year .283 (.015)*** .282 (.015)*** .296 (.161) .316 (.162)

Accumulated prison time .063 (.009)*** .064 (.009)*** .958 (.115)*** .947 (.114)***

Currently married -.314 (.023)*** -.351 (.024)*** -.801 (.169)*** -.617 (.194)***

Currently married convict .296 (.054)*** -.498 (.267)

Coefficients and standard errors are provided. Models adjust for exposure time. Exponentiating the coef- ficient and subtracting one (eb - 1) yields the proportional increase/decrease in the number of convictions associated with a unit increase in the regressor. Italicized coefficients represent those that are significantly different from zero when summed together, yielding the main effect of being married to a convicted spouse relative to being single (p \ .05) * p \ .05; ** p \ .01; *** p \ .001 (two-tailed tests)

J Quant Criminol

123

provides an incident rate ratio (IRR), and subtracting one (eb - 1) yields the proportional increase/decrease in the number of convictions given a state of marriage as opposed to a

state of singlehood. The IRRs for males and females are 0.73 (e-.314) and 0.45 (e-.801), respectively. This can be taken to mean that being married significantly lowers conviction

frequency by 27 % among males and 55 % among females, on average and all else equal.

Interestingly, a comparison of the impact of marriage for men and women indicates that

women benefit significantly more from marriage than do men. Since the models are

estimated independently, we can conduct a test of the difference in marriage coefficients, a

test that yields a z-statistic of 2.86 (p \ .01) (for details on this test, see Brame et al. 1998). Before proceeding to Model 2, we first draw attention to the remaining regressors. First,

the expected age-crime relationship is observed, as the coefficients imply an inverted-U

shape to the mean number of convictions per year of street time. Second, fertility status

appears to have an inverse relationship with conviction frequency among male subjects,

but no relationship with conviction frequency among female subjects. Third, past con-

viction strongly influences current conviction frequency. For both males and females,

having been convicted in the previous calendar year increases one’s conviction rate in the

current year, although the accumulated number of convictions is inversely associated with

conviction. Because multicollinearity is not a problem in this model, the effect can be

interpreted as a ‘‘slowing down’’ of the effect of accumulated convictions over time, net of

age. Fourth and finally, past imprisonment influences current conviction risk. Having been

confined in the last calendar year is associated with significantly more convictions only for

male offenders, while the accumulated time served in prison is also positively associated

with conviction frequency among both males and females.

Model 2 examines the differential effects of marriage to non-convicted and convicted

spouses on the number of convictions. In these models, the coefficient for marriage actually

represents the effect of being married to a non-convicted spouse, while the coefficient for

marriage to a convicted spouse represents a contrast with marriage to a non-convicted spouse.

Recall that, to recover the effect of marriage to a convicted spouse relative to singlehood,

these coefficients must be summed together. Italicized coefficients in Table 2 indicate those

for whom the summed coefficients are significantly different from zero.

For men, the coefficient for marriage is negative and significant, which means that

marriage to a non-convicted spouse is associated with a significant decline in conviction

frequency relative to singlehood (IRR = .70). On the other hand, the contrast for marriage

to a convicted spouse is positive and significant, meaning that the decline in conviction

frequency is not as pronounced for these individuals. To test the effect of marriage to a

convicted spouse compared to remaining unmarried, the two marriage coefficients summed

together yields a coefficient of -0.055 (s.e. = .052) and an IRR of 0.95, an effect that is

not statistically significant. In sum, being married to a non-convicted spouse reduces

conviction frequency by 30 percent relative to being unmarried, while being married to a

convicted spouse is statistically indistinguishable from singlehood.

The findings are somewhat different for female subjects. As expected, the coefficient for

marriage is negative and statistically significant (IRR = .54), implying that being married

to a non-convicted spouse leads to significantly fewer convictions relative to remaining

single. While the contrasting coefficient for marriage to a convicted spouse is negative, it is

not statistically significant. It should thus not be interpreted from this that marriage to a

convicted spouse is more protective than marriage to a non-convicted spouse, and close

inspection reveals that this contrast is driven by a comparatively small number of females

(N = 44), which introduces instability. On the other hand, the sum of the two marriage coefficients does yield a significant contrast of marriage to a convicted spouse relative to

J Quant Criminol

123

singlehood of -1.115 (s.e. = .238), with an IRR of 0.33. This can be taken to mean that,

for female offenders, marriage per se is the most salient transition, with no predictive

influence of the criminal history of the husband. Thus, for both males and females, mar-

riage slows the pace of criminal conviction. Yet for male offenders, only marriage to a non-

convicted spouse is protective. On the other hand, for female offenders, marriage to a

convicted or non-convicted spouse exerts a similar protective influence. 6,7

The Moderating Effect of Criminal History

The influence of a (non-) convicted spouse may depend on the offender’s own criminal

history at the time of marriage. We expect the effects of marriage to become smaller if

individuals committed more offenses before marriage. Chronic offenders would be more

present-oriented and self-centered, and would not have developed the capacity and desire

to invest in social relationships. Therefore, they would be less likely to be affected by

social ties like marriage (Moffitt 1993; Nagin and Paternoster 1994; Rhule-Louie and

McMahon 2007). To investigate the degree to which this is the case, we modeled the

interaction between the subject’s current marital status, spousal criminality, and the sub-

ject’s conviction history at the time of marriage. Conviction history is modeled as a series

of mutually exclusive dummy indicators (0 convictions, 1–3, 4–6, 7–9, 10–12, 13–15,

16?). Among female subjects, the effect of marriage did not differ by the criminal history

of the spouse, nor did it differ by the number of convictions at the time of marriage. As

described above, marriage per se is the relevant transition for female offenders. But we

should note that small cell sizes limit this part of the analysis.

Among male subjects, the impact of marriage does indeed differ by the offender’s

criminal history. In order to facilitate interpretation of this model, we summarize the results

in Fig. 1, which plots the IRRs and 95 % confidence intervals for the interaction between

marital status, spousal criminality, and the number of prior convictions. Because the

6 Note that the estimates yield average reductions over the entire span of marriage. However, some subjects

(i.e., those who were older in 1977) are in the analysis for more years than others. To test the sensitivity of the results, we estimated the models by limiting attention to discrete post-marriage intervals: the first 1, 5, and 10 years of the first marriage. To be able to compare married and unmarried persons, singles were followed until the mean age of first marriage plus respectively 1, 5, or 10 years. Importantly, for males and females alike, the findings are replicated when a limited number of post-marriage years are considered. 7

At the request of an anonymous reviewer, we also investigated cohort effects. Cohabitation has become much more widespread over the last decades. Therefore, the effects of marriage may have changed over time. We limited this investigation to male subjects, as the results for females did not exhibit sensitivity to birth cohort. We began by stratifying the men into one of three cohorts based on their birth year (1907–1945, 1946–1955, 1956–1965), and then constructed separate marriage indicators for each cohort to include into the fixed-effects Poisson model. Interestingly, for the earliest cohort, the coefficient for marriage was positive and statistically significant, while it was negative and significant for the last two cohorts. This suggests that, relative to singlehood, marriage to a non-convicted spouse increases in salience and desistance potential over time. In the earliest cohort, in fact, these marriages appear to be criminogenic. However, the contrasting coefficient for marriage to a convicted spouse was positive and significant for all three cohorts, indicating that marriage to a non-convicted spouse is more beneficial compared to marriage to a convicted spouse, irrespective of cohort. An additionally interesting result was that, for the latest cohort only, marriage to a convicted spouse was associated with a significant reduction in convictions compared to being single. This suggests that even marriage to a convicted spouse possesses desistance potential in later cohorts (but not as much potential as marriage to a non-convicted spouse). The finding that the crime-reducing impact of marriage becomes stronger over time is in line with the study of Bersani et al. (2009). They argue that the quality and stability of recent marriages may be higher, because these marriages are often preceded by cohabitation. Cohabitation is considered to be a testing phase, and marriage a further investment in the relationship.

J Quant Criminol

123

distinction between marriage to a non-convicted spouse and marriage to a convicted spouse

remained relevant, two fitted IRR curves are displayed. Notice first that the effect of

marriage to a non-convicted spouse was estimated very precisely (the 95 % error bars are

comparatively tight around the fitted IRR). Additionally, marriage to a non-convicted

spouse appears to grow significantly stronger in the number of pre-marriage convictions. In

other words, men with lengthy conviction histories benefited in a quite pronounced way

from these marriages. Although the results are not conclusive, to some degree, the same

can also be said of men who marry convicted spouses, but only if they have accumulated at

least ten prior convictions. Men with fewer than ten convictions at the time of marriage are

generally indistinguishable from singlehood when they marry a convicted spouse. How-

ever, it is worth noting that the fitted IRR is unstable and the confidence intervals are quite

wide, suggesting that the analysis has limited power to detect any moderating influences of

criminal history for men who marry convicted spouses.

The Moderating Effect of Marital Stability

The impact of marriage and spousal criminality on criminal conviction might also be

influenced by characteristics of the marriage itself. Sampson and Laub (1993), for example,

suggest that higher marital quality and stability increase the protective effect of marriage

on criminal behavior. This is of extra relevance since individuals marrying a convicted

spouse are more likely to have unstable marriages, as we saw in Table 1. We test this

expectation in a set of additional analyses. We created interactions between the marriage

indicators and a set of mutually exclusive indicators for the total length of the first marriage

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

1-3 4-6 7-9 10-12 13-15 16+

Number of Convictions at the Time of Marriage

E ff

e c t

o f

M a rr

ia g

e o

n C

o n

v ic

ti o

n F

re q

u e n

c y

Married to Convicted Spouse

Married to Non-Convicted Spouse

Fig. 1 Effect of marriage and spousal criminality on the number of convictions per year of street time (male subjects only). The estimates shown are exponentiated coefficients (i.e., incident rate ratios) from a fixed-effects Poisson model in which the effect of marriage is interacted with the total number of convictions at marriage. Error bars correspond to 95 % confidence intervals. An exponentiated coefficient of 1.0 implies no relationship between marriage and conviction

J Quant Criminol

123

in years (1–5, 6–10, 11–15, 16?) until divorce, death, or the end of the observation period

(calendar year 2003). As before, we limit this analysis to the male subjects, because of

power problems in the model for female subjects. These results are reported in Table 3.

For men, the pattern suggests that the beneficial impact of marriage to a non-convicted

spouse generally increases in the total length of marriage, whereas marriage to a convicted

spouse is indistinguishable from singlehood no matter the length of marriage (the summed

marriage coefficients are never significant). A notable finding, however, is that even

marriages to non-convicted spouses which dissolve after a few years have desistance

potential. In other words, even unstable marriages (to conventional women) reduce crime

while the marriages last.

Discussion

The aim of this study has been to investigate the relationship between marriage, spousal

criminality, and the subsequent development of criminal offending. Although numerous

studies have shown that marriage is a transformative life-course event that reduces criminal

offending, the criminal history of the spouse has largely been neglected. We thus used data

from the CCLS, a long-term study of a conviction cohort of Dutch offenders. The data have

a number of advantages for studying the effects of marriage on crime. In contrast to the few

existing studies in this area, the CCLS contains data on the criminal careers of both

offenders and their spouses that reach far into adulthood. Moreover, we had information on

the exact timing of convictions and marriages. Therefore, we were able to clearly distin-

guish partner selection from partner influences during marriage. Furthermore, the data

enabled us to investigate the criminal careers of male as well as female offenders.

The results show that marriage is indeed a salient transition in the criminal career, but

there are important qualifications to this conclusion that have to do with the characteristics

of the offenders (gender, criminal history), the characteristics of the spouses (criminal

history), and the characteristics of the marriages (duration). Among men, being married to

a non-convicted spouse uniformly reduces criminal involvement. On the other hand, being

married to a convicted spouse is indistinguishable from singlehood, and thus sustains

criminal involvement. Although ‘‘criminal marriages’’ are thus indeed not protective, we

do not observe an increase in criminal behavior as has been found in earlier studies

(Woodward et al. 2002). One explanation might be that these earlier studies had no

Table 3 Fixed-effects Poisson models of the impact of first marriage on conviction frequency, by gender and total length of marriage (male subjects only)

Variable Total length of marriage

1–5 years 6–10 years 11–15 years 16? years

Currently married -.107 (.040)** -.357 (.042)*** -.528 (.049)*** -.445 (.034)***

Currently married convict .149 (.097) .184 (.149) .463 (.134)*** .324 (.086)***

Coefficients and standard errors are provided. Fully specified models are estimated as in Table 2. Expon- entiating the coefficient and subtracting one (eb - 1) yields the proportional increase/decrease in the number of convictions associated with a unit increase in the regressor. Italicized coefficients represent those that are significantly different from zero when summed together, yielding the main effect of being married to a convicted spouse relative to being single (p \ .05) * p \ .05; ** p \ .01; *** p \ .001 (two-tailed tests)

J Quant Criminol

123

longitudinal information on spouses’ criminal behavior, and have not been able to clearly

distinguish partner selection from partner influences during the relationship. Similarity in

criminal behavior could also result from the fact that partners already resemble each other

before relationship formation.

In contrast to our expectation, the effect of being married to a non-convicted spouse is

especially pronounced for men with extensive criminal involvement prior to marriage. In

line with this finding, it has been suggested that the crime-reducing effects of relationships

would be stronger for individuals with a higher propensity to commit crimes, simply

because they have more potential criminal behavior in need of deterrence (Wright et al.

2001). Additionally, we find that the impact of a non-convicted spouse increases when

males’ marriages are more stable (i.e., of longer duration). However, even the marriages

that dissolve after a few years appear to lower crime risk while they last.

Women who marry also benefit from their union, but interestingly, this relationship

holds up irrespective of the conviction history of the spouse. Thus, the institution of

marriage per se tends to promote desistance among high-risk female subjects. Remarkably,

we do not find support for the idea that women are more strongly influenced by the

criminal behavior of their partners than men. The fact that we do not find any effect of a

convicted spouse might be partly attributed to the birth of children during marriage. The

birth of a child might have a more pronounced impact on females’ lives (both practically

and emotionally) and reduce the preferences and opportunities to commit crimes even

more than for men (Giordano et al. 2002; Uggen and Kruttschnitt 1998). The crime-

reducing effect of childbearing might thus outweigh the crime-stimulating effect of a

convicted husband. And in fact, in our data we do observe a modest tendency for the

presence of children to contribute to the marriage effect (results of which are not shown).

Untangling the complex interactions between marriage and parenthood is an important task

for future research. In addition, less contact with peers might explain the finding that

marriage reduces women’s criminal behavior irrespective of the criminal background of

the spouse: Married women might prioritize family responsibilities over friends.

Although the data used in this paper are unique—they stem from a large-scale, pro-

spective, longitudinal study with a very long observation period—a number of limitations

have to be taken into account when interpreting the results. These limitations offer several

guidelines for future research. First, our sample consists of persons who were in contact

with the criminal justice system in 1977—all individuals in the sample have committed an

offense at least once during their lives. The sampling frame influences the generalizability

of our results in two ways. First, our results speak to the effect of marriage on criminal

convictions among convicted and to-be-convicted individuals, and not necessarily among

the population at large. Second, the results pertain to a particular time and place. The

Netherlands was characterized by a lenient penal climate until the 1990s. This means that

offenders were less easily convicted in 1977 than today. The fact that the CCLS offenders

were convicted during this era therefore means that they were relatively serious offenders.

Studies in different contexts will therefore be needed to test the generality of the findings

from this sample.

In addition, due to the use of official data, we cannot rule out the possibility that we

underestimate the total number of criminal acts. Not all offenses are recorded by the police

or are prosecuted. The underestimation may be selective, meaning that the probability of

being convicted is not equal for all persons. For example, some offenders may be more

intensely monitored by the police, while others (e.g., those with a higher intelligence) may

adopt more effective strategies to keep out of the arms of the law. However, it should be

noted that the use of official data has important advantages as well. It enables us to

J Quant Criminol

123

examine a great variety of criminal acts that differ in severity, e.g., violent offenses,

property offenses, drug offenses, weapon offenses, and offenses against the public order.

Moreover, it enables us to investigate the development of criminal behavior over the entire

life course.

Second, our measure of marriage includes only legal marriage. From the 1980s onward,

it has become more and more common to cohabit, and cohabitation has even become a

substitute for marriage in the Netherlands (Liefbroer and Dykstra 2000). This development

is less of a problem for our analyses, because the CCLS contains data on a cohort of

individuals convicted in 1977. The largest share of these persons already reached mar-

riageable age before this time. Although nowadays the Netherlands is known for its high

cohabitation rate, marriage patterns were comparable to other countries (e.g., the US)

during most of the period under study. The current increase in cohabitation rates is not

unique to the Netherlands and has taken place in other countries as well (Kalmijn 2002).

Therefore, future studies should also focus on the impact of other relationship types on

criminal offending (see also Bersani et al. 2009).

Third, we are unable to gain insight into the intermediate mechanisms underlying the

relationship between marriage, spousal criminality, and criminal behavior. The question

why marriage to a non-convicted spouse reduces criminal conviction, while marriage to a

convicted spouse does not change conviction risk (at least for men), ultimately remains

unanswered. It is unclear whether the changes in the former case are caused by social

bonds, restructured routine activities, social learning processes, or cognitive transforma-

tions. Untangling these mechanisms and determining their relative importance is an

important task for future research.

Fourth, given that analyses of the marriage-crime relationship must be limited to quasi-

experimental designs, we must maintain a healthy skepticism about the degree to which

our models provide estimates of the causal impact of marriage and spousal criminality on criminal convictions. To date, no single study can claim to have reliably estimated the true

causal impact of marriage on crime, our study included. Because marriage cannot be

randomly assigned, even in principle, researchers must resort to quasi-experimental

designs that can at least narrow the boundaries of plausible causal estimates. The fixed-

effects models employed here produce consistent estimates of the ‘‘marriage effect’’ in the

presence of confounding by time-stable unobservables, but inconsistent estimates if there

are time-varying unobservables which are correlated with marriage and crime (see Bjerk

2009). The models are thus not a panacea to the selection problem, although they do

restrict the potential sources of confounding to time-varying rather than time-stable ones.

As observed by an anonymous reviewer, the ‘‘marriage effects’’ that we estimate in this

study reflect a sum of the ‘‘true effect’’ of marriage and spousal criminality on conviction

and the ‘‘dynamic selection effect’’ that persists due to lack of exhaustive controls for

relevant time-varying confounders.

Finally, as pointed out by an anonymous reviewer, it will be important moving forward

to further untangle, and theorize about, the selection mechanisms that are at work in the

processes of marriage and partner selection. Most analyses (the present one included) treat

the selection process as one sided, and only from the perspective of the sampled individual.

Yet marriage is clearly a two-sided affair, as a sample subject chooses his (her) partner,

while the spouse-to-be must likewise choose the sample subject as her (his) partner. To the

extent that the partner’s unobservables, which jointly influence his/her marriage and crime

decisions, are highly correlated with the focal individual’s unobservables, and both are

fairly time stable, the results from the analysis will remain robust. However, the veracity of

this assumption is ultimately unknown (and unknowable), because these processes are

J Quant Criminol

123

poorly understood (and understudied). We would add further that this will be true of the

study of any ‘‘market’’ behavior where outcomes depend on decisions made by two or

more parties. Most notably, studies of the employment-crime relationship rarely

acknowledge that a job applicant’s decisions are determined, to an unknown degree, by the

tastes and preferences of potential employers.

Conclusion

Notwithstanding the foregoing limitations and knowledge gaps, this study extends our

knowledge about the role of marriage in the criminal career in important ways. Marriage

matters for the development of criminal behavior, but its impact depends in systematic

ways on gender, criminal history, length of marriage, and spousal criminality. With regard

to the latter observation, getting married does not necessarily mean the end of a criminal

career. For men, the ‘‘good marriage effect’’ clearly depends on the criminal history of the

spouse whom one marries. Our conclusions have significant implications for criminolog-

ical theories emphasizing the protective effects of marriage and give important guidance

for future research. On the basis of the current findings, we believe it is important to

provide more nuance to the prominent idea that marriage uniformly reduces criminal

behavior (e.g., Laub and Sampson 2003). Future studies of the marriage-crime relationship

would be well advised to devote attention to partner selection processes and the way in

which partners influence each other during marriage.

References

Akers RJ (1973) Deviant behavior. A social learning approach. Wadsworth Publishing Co, Belmont Bersani BE, Laub JH, Nieuwbeerta P (2009) Marriage and desistance from crime in the Netherlands: do

gender and socio-historical context matter? J Quant Criminol 25:3–24 Bjerk D (2009) How much can we trust causal interpretations of fixed effects estimators in the context of

criminality? J Quant Criminol 25:391–417 Blokland AAJ, Nieuwbeerta P (2005) The effects of life circumstances on longitudinal trajectories of

offending. Criminology 43:1203–1240 Blokland AAJ, Nagin DS, Nieuwbeerta P (2005) Life span offending trajectories of a Dutch conviction

cohort. Criminology 43:919–954 Brame R, Paternoster R, Mazerolle P, Piquero A (1998) Testing for the equality of maximum-likelihood

regression coefficients between two independent samples. J Quant Criminol 14:245–261 Brines J, Joyner K (1999) The ties that bind: principles of cohesion in cohabitation and marriage. Am Sociol

Rev 64:333–355 Capaldi DM, Kim HK, Owen LD (2008) Romantic partners’ influence on men’s likelihood of arrest in early

adulthood. Criminology 46:267–299 Farrington DP, West DJ (1995) Effects of marriage, separation, and children on offending by adult males.

In: Blau ZS, Hagan J (eds) Current perspectives on aging and the life cycle, vol 4. JAI Press, Greenwich, pp 249–281

Giordano PC, Cernkovich SA, Rudolph JL (2002) Gender, crime, and desistance: toward a theory of cognitive transformation. Am J Sociol 107:990–1064

Giordano PC, Schroeder RD, Cernkovich SA (2007) Emotions and crime over the life course: a neo- Meadian perspective on criminal continuity and change. Am J Sociol 112:1603–1661

Halaby CN (2004) Panel models in sociological research: theory into practice. Annu Rev Sociol 30:507–544 Haynie DL, Giordano PC, Manning WD, Longmore MA (2005) Adolescent romantic relationships and

delinquency involvement. Criminology 43:177–210 Heckman JJ, Hotz VJ (1989) Choosing among alternative nonexperimental methods for estimating the

impact of social programs: the case of manpower training. J Am Stat Assoc 84:862–874

J Quant Criminol

123

Horney JD, Osgood W, Haen Marshall I (1995) Criminal careers in the short-term: intra-individual vari- ability in crime and its relation to local life circumstances. Am Sociol Rev 60:655–673

Johnson DR (1995) Alternative methods for the quantitative analysis of panel data in family research: pooled time-series models. J Marriage Fam 57:1065–1077

Kalmijn M (1998) Intermarriage and homogamy: causes, patterns and trends. Annu Rev Sociol 24:395–421 Kalmijn M (2002) Sociologische analyses van levensloopeffecten: een overzicht van economische, sociale

en culturele gevolgen. [Sociological analyses of life course effects: an overview of economic, social, and cultural consequences]. Bevolking en Gezin 31:3–46

King RD, Massoglia M, MacMillan R (2007) The context of marriage and crime: gender, the propensity to marry, and offending in early adulthood. Criminology 45:33–65

Laub JH, Sampson RJ (2003) Shared beginnings, divergent lives. Delinquent boys to age 70. Harvard University Press, Cambridge

Laub JH, Nagin DS, Sampson RJ (1998) Trajectories of change in criminal offending: good marriages and the desistance process. Am Sociol Rev 63:225–238

Leverentz AM (2006) The love of a good man? Romantic relationships as a source of support or hindrance for female ex-offenders. J Res Crime Delinq 43:459–488

Liefbroer AC, Dykstra PA (2000) Levenslopen in verandering: een studie naar ontwikkelingen in de levenslopen van Nederlanders geboren tussen 1900 en 1970. [Life courses in change: a study of the developments in life courses of the Dutch born between 1900 and 1970]. Sdu Uitgevers, Den Haag

Moffitt TE (1993) Life-course-persistent and adolescence-limited anti-social behavior: a developmental taxonomy. Psychol Rev 100:674–701

Moffitt TE, Caspi A, Rutter M, Silva PA (2001) Sex differences in antisocial behaviour. Cambridge University Press, Cambridge

Nagin DS, Paternoster R (1994) Personal capital and social control: the deterrence implications of a theory of individual differences in criminal offending. Criminology 32:581–606

Nieuwbeerta P, Blokland AAJ (2003) Criminal careers of adult Dutch offenders (codebook and docu- mentation). NSCR, Leiden

Nieuwbeerta P, Nagin DS, Blokland AAJ (2009) Assessing the impact of first-time imprisonment on offenders’ subsequent criminal career development: a matched samples comparison. J Quant Criminol 25:227–339

Piquero AR, Blumstein A, Brame R, Haapanen R, Mulvey EP, Nagin DS (2001) Assessing the impact of exposure time and incapacitation on longitudinal trajectories of criminal offending. J Adolesc Res 16:54–74

Piquero AlexR, Brame Robert, Mazerolle Paul, Haapanen Rudy (2002) Crime in emerging adulthood. Criminology 40:137–169

Piquero AR, Farrington DP, Blumstein A (2003) The criminal career paradigm. Crime Justice Rev Res 30:359–506

Rhule-Louie DM, McMahon RJ (2007) Problem behavior and romantic relationships: assortative mating, behavior contagion, and desistance. Clin Child Fam Psychol Rev 10:53–100

Sampson RJ, Laub JH (1993) Crime in the making: pathways and turning points through life. Harvard University Press, Cambridge

Sampson RJ, Laub JH, Wimer C (2006) Does marriage reduce crime? A counterfactual approach to within- individual causal effects. Criminology 44:465–508

Simons RL, Stewart EA, Gordon LC, Conger RD, Elder GH Jr (2002) A test of life-course explanations for stability and change in antisocial behavior from adolescence to young adulthood. Criminology 40:401–434

Steffensmeier D, Allan E (1996) Gender and crime: toward a gendered theory of female offending. Annu Rev Sociol 22:459–487

Tak PJP (2003) The Dutch criminal justice system: organization and operation. Boom Legal Publishers, Meppel

Theobald D, Farrington DP (2011) Why do the crime-reducing effects of marriage vary with age? Br J Criminol 51:136–158

Uggen C, Kruttschnitt C (1998) Crime in the breaking: gender differences in desistance. Law Soc Rev 32:339–366

Waite LJ (1995) Does marriage matter? Demography 32:483–507 Waite LJ, Gallagher M (2000) The case for marriage: why married people are happier, healthier, and better

off financially. Doubleday, New York Warr M (1998) Life-course transitions and desistance from crime. Criminology 36:183–215 Warr M (2002) Companions in crime. Cambridge University Press, Cambridge Western B (2006) Punishment and inequality in America. Russell Sage Foundation, New York

J Quant Criminol

123

Woodward LJ, Fergusson DM, Horwood JL (2002) Deviant partner involvement and offending risk in early adulthood. J Child Psychol Psychiatry 43:177–190

Wright BRE, Caspi A, Moffitt TE, Silva PA (2001) The effects of social ties on crime vary by criminal propensity: a life course model of interdependence. Criminology 39:321–350

J Quant Criminol

123