ArticlePDF Available

Does Marriage Reduce Crime?


Abstract and Figures

The “marriage effect” is one of the most widely studied topics of life course criminology. The contemporary consensus is that marriage promotes desistance from crime. Most of the 58 studies reviewed here find a negative longitudinal association between marriage and crime. The results are more consistent among men. Studies that attend to relationship quality, such as the level of marital attachment, tend to produce particularly strong associations. Critical scrutiny of the evidence regarding the causal nature of the reported associations suggests, however, that claims about the restraining influence of marriage are overstated. None of the studies demonstrates evidence of direct (counterfactual) causality; no study has served a causal estimate unbiased by selection processes. Moreover, only a few studies address time ordering, and some of those show that desistance precedes rather than follows marriage. Evidence in support of the theoretical mechanisms responsible for the marriage effect is also mixed and insufficient. The criminological literature has been insensitive to the reality that entering a marital union is increasingly unlikely to signify the point at which a committed, high-quality relationship is formed.
Content may be subject to copyright.
Torbjørn Skardhamar, Jukka Savolainen,
Kjersti N. Aase, and Torkild H. Lyngstad
The marriage effect is one of the most widely studied topics of life
course criminology. The contemporary consensus is that marriage promotes
desistance from crime. Most of the 58 studies reviewed here nd a negative
longitudinal association between marriage and crime. The results are more
consistent among men. Studies that attend to relationship quality, such as
the level of marital attachment, tend to produce particularly strong associations.
Critical scrutiny of the evidence regarding the causal nature of the reported
associations suggests, however, that claims about the restraining inuence
of marriage are overstated. None of the studies demonstrates evidence of direct
(counterfactual) causality; no study has served a causal estimate unbiased by
selection processes. Moreover, only a few studies address time ordering, and
some of those show that desistance precedes rather than follows marriage.
Evidence in support of the theoretical mechanisms responsible for the
marriage effect is also mixed and insufcient. The criminological literature
has been insensitive to the reality that entering a marital union is increasingly
unlikely to signify the point at which a committed, high-quality relationship
is formed.
Some 25 years ago, criminologists engaged in a vigorous debate concern-
ing the value of longitudinal research on criminal careers (Gottfredson and
Hirschi 1987; Blumstein, Cohen, and Farrington 1988; Rowe, Osgood,
Electronically published July 30, 2015
Torbjørn Skardhamar is a senior researcher at Statistics Norway and an associate pro-
fessor at the Department of Sociology and Human Geography at the University of Oslo.
Jukka Savolainen is a research scientist at the Institute for Social Research, University of
Michigan, where he serves as the director of the National Archive of Criminal Justice
Data. Kjersti N. Aase is a former junior researcher at Statistics Norway and currently
advisor for the Telemark county council. Torkild H. Lyngstad is a professor at the Depart-
ment of Sociology and Human Geography at the University of Oslo.
q 2015 by The University of Chicago. All rights reserved.
and Nicewander 1990; Greenberg 1991). Whichever our view of the
analytical merits of the debate, it is clear that, as a matter of practical con-
sequence, the position favoring the longitudinal approach has won the
hearts and minds of academic criminologists. Today, the eld known as
life course criminology is stronger than ever. In 2012, the American Society
of Criminology (ASC) established the Division of Developmental and
Life Course Criminology. This is the most recent of the eight specialty
areas formally recognized by the ASC. On a more international scale, a
large number of the recipients of the Stockholm Prize in criminology
can be classied as life course criminologists. Between 2006 (when the
award was established) and 2014, a total of 16 individuals received the
prize, and at least eight were awarded for work that relied on evidence
from analysis of longitudinal life course data.
The early work in life course criminology was focused on the onset of
criminal careers. The basic goal was to address early childhood ante-
cedents of serious criminality. Although this line of inquiry has remained
central, it became less dominant following the publication of Sampson
and Laubs Crime in the Making (1993), which brought the topic of desis-
tance to the forefront. To illustrate, a Google Scholar search using the
term desistance crime generated 219 results for the period 198090,
compared with 5,390 in 20002010. Although much of this reects the
overall growth of academic criminology, the corresponding rate of in-
crease for the search term age of onset criminal was only a fth as large.
Life course criminology passed a turning point when, on the basis of
a reanalysis of the Gluecks rich longitudinal data tracking males born
between 1924 and 1935, Sampson and Laub (1993) introduced their
age-graded theory of informal social control. The main point of this
work was to show that life course matters, that the possibility of
change, including desistance, is omnipresent in the human life course.
The most inuential take-home from the 1993 monograph was evidence
suggesting that transitions to good marriages and stable employment
can set in motion the process of desistance from crime. Although both
marriage and work have received considerable attention in subsequent
research (Siennick and Osgood 2008; Uggen and Wakeeld 2008),
Alfred Blumstein, David Farrington, Friedrich Lösel, John Laub, Terrie E. Moftt,
Daniel Nagin, David Olds, and Robert J. Sampson. John Hagan, who received the award
in 2009 for his research on genocide, has also made signicant contributions to life course
386 Torbjørn Skardhamar et al.
the marriage effect seems to have emerged as primus inter pares. The pri-
ority for marriage is subtle yet clear in Laub and Sampsons follow-up
monograph Shared Beginnings, Divergent Lives (2003). The critically im-
portant chapter 6, Why Some Offenders Stop, includes a separate
section for Marriage as Turning Point but no equivalent subheading
for employment. Instead, employment is discussed in fewer than two
pages under Unpacking the Desistance Process, where the marriage
effect is discussed over six additional pages (Laub and Sampson 2003,
pp. 134 41).
Judging from the program of the 2013 annual meeting of
the ASC, the topic of marriage and crime is strong in the eld at large.
Nearly one-quarter of the papers in panels dedicated to life course crim-
inology examined the association between marriage or romantic relation-
ships and offending. By contrast, only two of the papers in these panels
were focused on employment.
We suspect that the appeal of the marriage hypothesis reects its
greater empirical success in the research literature. As we demonstrate,
a number of high-prole studies report evidence of strong effects of mar-
riage on desistance (e.g., Sampson, Laub, and Wimer 2006; King, Mas-
soglia, and Macmillan 2007). By comparison, the track record for studies
of employment is considerably weaker and more mixed (Bushway and
Apel 2012; Skardhamar and Savolainen 2014). As observed in a recent
article in Criminology, [A] consensus has grown in the literature that mar-
riage holds the potential to promote desistance from crimecommonly
known as the marriage effect. This relationship seems to be fairly robust
and consistent across method, sample, gender, race, and socio-historical
context. Whereas much of the research has demonstrated that a marriage
effect exists, what is less clear is how marriage reduces ones criminal
involvement (Bersani and Doherty 2013, p. 400; our emphasis). Bersani
and Doherty argue that the marriage effect has largely been established,
and it is time to focus on explicating the mechanisms accounting for it.
Similar claims can be found in other published studies, and a recent re-
view concluded that there is evidence of an overall protective effect of
As an additional indication from Shared Beginnings, Laub and Sampson wrote that our
inferences are strongest for marriage (2003, p. 272).
If we include papers on parenthood, studies of family processes covered 43 percent of
all the papers presented in these panels. These statistics are based on the records kept by
one of the authors, who was the relevant area chair of the 2013 ASC meeting (available
from authors).
Does Marriage Reduce Crime? 387
marriage on subsequent criminal desistance (Craig, Diamond, and
Piquero 2014, p. 34).
Over the last few decades, there has been considerable progress both
within life course criminology in general and in the marriage-crime liter-
ature in particular. One facet of this has to do with increasingly frequent
collection and use of longitudinal data. Quite a few data sets allow for life
course analyses of long-term change. The geographical coverage of high-
quality datasources has increased beyond the United States and the United
Kingdom. Another element of progress is the increasing sophistication
of statistical methods that allow researchers to exploit longitudinal data.
These include group-based trajectory models (Nagin 2005), which have
been extremely important in criminal careers research, and the use of
panel data models that allow controls for unobserved heterogeneity.
Despite these major leaps forward, there is room for improvement in
the quality of research and the interpretation of the evidence. Our deci-
sion to write this essay was motivated, in part, by skepticism regarding
the assumption that the causal effect of marriage on crime is as strong
and robust as the criminological consensus suggests. The bar for making
causal inferences in the social sciences has increased over the past de-
cades (Morgan and Winship 2007). Central to this development is the
increasing inuence of the counterfactual model of causality, sometimes
referred to as the potential outcomes framework (Rubin 2005), as an
efcient analytical approach for specifying and estimating causal effects.
Given that the marriage effect has emerged as one of the focal concerns
of contemporary research in life course criminology, we think it is im-
portant to scrutinize the relevant evidence more carefully and critically
than has been done in the past.
Although prior reviews of this literature have been informative and
exceedingly helpful, our approach differs. First, we consider evidence of
causal effects only if it holds up to contemporary methodological stan-
dards (see, e.g., Morgan and Winship 2007; Angrist and Pischke 2009;
Berk 2010). Regardless of what any given study claims about causality,
we do not consider an association causal unless the methodological ap-
proach provides a truly counterfactual causal estimate. We nd consistent
evidence of a negative association between marriage and crime, but this
evidence does not rule out social selection. Second, as another aspect of
causal inference, we pay close attention to evidence on time order in
the association between transition to marriage and the onset of desis-
tance. It is clear that marriage cannot be treated as a cause of desistance
388 Torbjørn Skardhamar et al.
if it occurs after rather than before desistance. Most studies of marriage
and crime have ignored this basic point. The few studies that have ad-
dressed the timing issue yield contradictory results. Third, we discuss
the evidence in support of the hypothesized mechanisms in more detail
than has been done in the past. We agree with Bersani and Doherty
(2013) that there should be more focus on how marriage reduces involve-
ment in crime. Once again, we identify major gaps in the literature:
Many of the processes considered important as mediators of the mar-
riage effect have received little or no attention. However, we do nd con-
sistent support for the expectations that high-quality romantic relation-
ships are related to desistance more strongly than low-quality ones.
This essay is organized in ve major sections. We begin by laying out
the theoretical foundations of the marriage effect on crime in Section I.
In Section II we describe themethods and criteria used to select the 58 stud-
ies on the relationship between marriage or cohabitation and crime, pub-
lished between 1990 and 2014, that we reviewed. Section III offers a de-
tailed discussion of the nature and extent of the evidence regarding the
longitudinal association between romantic unions (marriage, cohabitation,
and related states) and criminal offending. Section IV focuses on the quality
of this evidence from the perspective of causal inference. Drawing on the
prevailing standards of causal inference in the social sciences, we evaluate
claims about causal effect presented in prior work. We also discuss the
evidence regarding the timing issue and review the evidence regarding
the mechanisms expected to produce the marriage effect on crime (peer
processes, marital quality, etc.). We conclude in Section V with a general
discussion of the literature on marriage and crime and suggest directions
for the next generation of studies. The criminological literature has been
insensitive to the reality that entering a marital union is increasingly un-
likely to signify the point in which a committed, high-quality relationship
is formed.
I. Theoretical Foundations
Given that Sampson and Laub were students of (early) Travis Hirschi,
it is not surprising that their age-graded theory (AGT, henceforth) is
rooted in social bonding theory (Hirschi 1969). In the original version
of the AGT (Sampson and Laub 1993), the marriage effect is described
in terms of attachment, that is, the strength of the emotional investment
in the romantic partner. Subsequent analysts argued that Sampson and
Does Marriage Reduce Crime? 389
Laubs study overplayed the role of bonding theory at the expense of
processes more consistent with differential association and routine activ-
ities perspectives. Warr (1998) proposed (and found empirical support
for the contention) that marriage limits offending by restructuring asso-
ciations away from crime-prone peers. Following marriage, men are more
likely to spend time in family settings (with relatives, other couples, friends
with children) and less time with young, single men. These changes have
the potential to disrupt routine activities in ways that reduce criminal
temptations and opportunities.
The updated version of AGT articulated in Shared Beginnings (Laub
and Sampson 2003), although still rooted in control theory, is consider-
ably more pluralistic, embracing multiple pathways from marriage to
criminal desistance. In addition to peer processes, routine activities, and
emotional attachment, Laub and Sampson recognize the capacity of the
spouse to serve as a situational agent of day-to-day social control. For
example, the monograph describes a case in which the wife monitors
the pace and timing of the husbands drinking and makes sure he gets
up the next morning to go to work (p. 136). Drawing on such theorists
as Maruna (2001) and Giordano, Cernkovich, and Rudolph (2002), Laub
and Sampson incorporate changes in self-concept as an additional mech-
anism accounting for the inuence of marital bonding on reduced crim-
inal offending. These symbolic-interactionist perspectives argue that de-
sistance from crime is not complete until the person has undergone a
conversion process in which criminal identity has been replaced with that
of a law-abiding family man.
Figure 1 is our attempt to summarize the key processes expected to
contribute to the marriage effect on desistance. This model assigns priority
FIG.1.A causal model for the effect of marriage on desistance from crime
390 Torbjørn Skardhamar et al.
to attachment over the other intervening factors. This decision reects
our contention that any institutional bond, including marriage, is un-
likely to exert restraining inuence unless it is perceived as valuable by
the actor. Although marriage without love is known to happen, in mod-
ern societies the very act of marriage is expected to signify emotional at-
tachment to the partner: people are unlikely to marry without profess-
ing love for each other.
One might even argue that attachment is
more likely to precede than to follow the transition to marriage. In that
case, of course, we should not refer to the process as the marriage effect.
Consistent with this insight, a number of studies have applied this theo-
retical framework to examine the effects of cohabitation and having a ro-
mantic partner on offending behavior. It should be obvious that the as-
sumed mechanisms are not contingent on the legal status but on the
strength of the relationship as perceived by the subject. We nd it rea-
sonable to assume that married couples, on average, are more commit-
ted than are cohabitants or other romantic couples (Rhule-Louie and
McMahon 2007; Craig and Foster 2013, p. 33). From this perspec-
tive, the variable marriage in gure 1 could be understood as a formal
marker of relationship quality rather than a causal agent in itself.
As denoted by the multiple reciprocal arrows, the model assumes
spousal attachment to develop over time as a function of relationship
dynamics (Laub, Nagin, and Sampson 1998). For example, emotional
bonding is expected to inuence partners motivation to monitor and
supervise spousal behavior, and this, in turn, affects the attachment be-
tween the two, either positively or negatively. Men who are attached to
their wives and enjoy spending time with them are likely to modify
their daily and weekly routines away from delinquent peers and criminal
temptations. If these decisions are experienced as rewarding, they rein-
force the attachment; but if the new lifestyle is not working, the emo-
tional bond is likely to diminish. Over time, if the day-to-day processes
are conducive to change, a new (noncriminal) identity starts emerging
We can imagine situations in which emotional attachment is not an important source
of marital bonding. For example, a person facing economic hardship may be motivated to
marry and stay in a relationship for nancial reasons. However, we expect these situations
to be rare in modern Western societies, and especially rare among the criminally active.
Consistent with this idea, Laub and Sampson (2003, p. 137) recognize economic benet
as one additional bond that may help keep crime-prone men in check. They argue that
most men in their sample married up and had more to lose from straying from the
straight and narrow.
Does Marriage Reduce Crime? 391
with enduring effects on desistance (Paternoster and Bushway 2009; Ber-
sani and Doherty 2013).
Note that this account of the marriage effect includes both situa-
tional (contemporaneous) and developmental processes of inuence.
The model also recognizes that not any marriage is expected to contrib-
ute to desistance as the effect depends on what it does to the relation-
ship and how it is perceived by the actors. Although we believe this to be
a faithful characterization of the mechanisms behind the marriage effect,
we acknowledge that this is our interpretation of the theoretical litera-
ture. To our knowledge, there is no authoritative organization of the
various factors believed to underlie the marriage effect.
We hope g-
ure 1 serves as a helpful starting poi nt.
As important as it is to understand the mechanisms producing the ef-
fect, such discussions are predicated on the presence of a causal associa-
tion. The literature on marriage and crime offers three distinct per-
spectives on causality. First, the AGT treats marriage as an exogenous
turning point with the potential to trigger desistance. Laub and Sampson
(2003, pp. 27879) use the term desistance by default to describe this pro-
cess: Many men made a commitment to go straight without even realiz-
ing it. Before they knew it, they had invested so much time in a marriage
or a job that they did not want to risk losing their investment.
Second, other theories, such as cognitive transformation (Giordano,
Cernkovich, and Rudolph 2002) and the subjective-social model (LeBel
et al. 2008), argue that changes in objective life circumstan ces are un-
likely to facilitate change without a preexisting willingness to go
straight. From this perspective, desistance should already have
startedat least psychologically if not behaviorallybefore transitions
to work, marriage, or parenthood can become inuential. Marriage is
understood as a potential hook for change (Giordano, Cernkovich,
and Rudolph 2002) with the capacity to assist in the process of desis-
tance, but it would be unrealistic to expect marriage to function as a trig-
gering event. Third, it is possible for marriage to have no causal effect on
offending. The negative association between marital bonding and crim-
inal involvement may be a spurious function of maturation, that is, the
age-variant but inevitable process of aging out and settling down (Mas-
Bersani and Dohertys (2013) g. 1 is a helpful way to distinguish between situational
and developmental effects of marriage, but their graph is not intended as a causal model of
marriage effect on desistance. Our g. 1 is consistent with theirs.
392 Torbjørn Skardhamar et al.
soglia and Uggen 2010). According to the maturation perspective, if there
is any causality in the association, it is expected to ow from desistance to
marriage, as criminally active individuals are unlikely to make marital
transitions; and to the extent that they are, they are unlikely to have
partners with characteristics conducive to desistance (Hirschi and Gott-
fredson 1983; Morizot and Le Blanc 2007). This perspective expects sub-
stantial and sustained reductions in criminal offending as a precondition
for marrying a person with the capacity to generate the kind of social cap-
ital described in gure 1.
Figure 2 offers a visual summary of the key differences between these
three different accounts of the longitudinal association between marriage
and desistance. Note that these graphs are ideal-typical representations
of complex theoretical arguments. For example, we do not claim that
the turning point hypothesis exhibits rigid stability prior to marriage or
that the maturation perspective implies no change in offending following
FIG.2.Three ideal-typical offending trajectories around the time of transition to marriage
Does Marriage Reduce Crime? 393
marriage. The purpose of gure 2 is to highlight the difference between
the maturation perspective and the two hypotheses that assume a causal
effect of marriage on crime (turning point and hook for change). It is cru-
cial to recognize that despite clear differences in etiological accounts, each
of these three perspectives expects the average rates of offending to be
lower during periods of marriage. The only difference has to do with
the timing of change in the criminal trajectory vis-à-vis the transition to
marriage. The maturation perspective expects desistance to precede mar-
riage, the turning point hypothesis predicts the opposite pattern, and the
hook for change hypothesis allows the onset of desistance to precede mar-
ital transition but expects additional reductions thereafter. Thus, as g-
ure 2 shows, evidence of lower average offending rates during states of
marriage is equally consistent with social selection and causation explana-
tions of the marriage effect.
II. Inclusion Criteria
We used multiple methods to search the criminological literature. The
studies included were selected by scanning lists of references from pub-
lished studies and by searching two electronic databases, Web of Sci-
ence and Google Scholar, using combinations of the following search
terms: marriage, married, marital status, family formation, cohabitation,
cohabitant, romantic relationship, desistance, criminal career, life course
crime, criminal trajectories, offending trajectories, antisocial behavior,
deviance, and developmental criminology. The initial search produced
380 articles. We performed our nal search update in March 2014; stud-
ies published after that date are not included.
We included only articles published in 1990 or later as older studies
have been reviewed by Wright and Wright (1992). While many of the
studies were primarily interested in marriage/cohabitation/romantic re-
lationships and desistance, we also include studies with a broader scope in
which relationship status is but one of several factors of empirical interest.
Some of the studies on romantic relationships use young samples, but we
exclude studies of adolescent romances, limiting our focus to relationships
in adulthood. Thus, we exclude studies of romantic relationships among
teenagers (19 or younger). In cases in which subjects were tracked from
youth to adulthood, we included the study as long as the data extend to
observations beyond age 20 years. We examine only studies featuring
criminal or delinquent offending as the outcome variable. Some measured
394 Torbjørn Skardhamar et al.
offending with ofcial statistics, such as convictions or arrests; others
used self-reported measures. We ignored studies of alcohol use but in-
cluded studies of illicit drug use. Some included outcomes that were both
within and outside the scope of this essay. In those cases we discuss the
results pertaining to criminal or delinquent offending.
We included only individual-level studies, excluding studies based on
aggregate data (e.g., Edlund et al. 2013). Unpublished studies, even if
available as working papers, were not included (e.g., Andersen, An-
dersen, and Skov 2010). Working papers, by denition, are not the nal
versions, and results are likely to change following the peer-review pro-
cess. We also excluded institutional or internal research reports because
it is difcult to conduct a systematic search (e.g., Broidy and Cauffman
2006; Bersani and DiPietro 2013). Finally, we excluded theses and dis-
sertations. Because we have read all the studies cited in this paragraph,
however, they inform our perspective on the literature. Our decision to
exclude a study does not mean we have ignored it.
Most of the studies that turned up in our literature searches are quan-
titative. This speaks to the nature of the literature and does not reect
our selection criteria. To make sure we did not miss important qualita-
tive studies, we considered all relevant studies covered by Veysey, Mar-
tines, and Christians (2014) review of qualitative research on desistance.
We beneted from two recent book chapters evaluating the evidence on
the marriage effect on crime. Focused on evidence from the Nether-
lands, Bersani and van Schellen (2014) reviewed nine studies. Craig, Di-
amond, and Piqueros (2014) review was not limited by geography and
covered 31 studies.
Table 1 lists and describes the studies included in our analysis: 58
publications met our selection criteria. In addition, we came across a cou-
ple of relevant additional studies (Bersani and DiPietro 2014; Siennick
et al. 2014). Because they are too recent to have appeared in our system-
atic search, we do not include them among the 58 studies. Studies not
yet in print but made available online (e.g., Craig 2015) are included.
III. Empirical Evidence
To our knowledge, the rst ever review of the marriage-crime literature
was authored by Wright and Wright (1992). The majority of studies
included were cross-sectional; only six were longitudinal (West 1982;
Gibbens 1984; Shavit and Rattner 1988; Caspi, Bem, and Elder 1989;
Does Marriage Reduce Crime? 395
Overview of Studies of the Marriage Effect
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Barnes and
Beaver (2012)
United States Add Health
(waves 13)
Men and
4,568 Logistic regression,
sibling differences
Benecial 11
Barnes et al. (2014) United States Add Health
(waves 14)
Men and
13,000 Crossed-lagged
model (Granger
causality test) and
reciprocal model
between marriage
and criminal
Beaver et al. (2008) United States Add Health
(waves 13)
Men and
1,994 Logistic regression,
sibling differences
Benecial 52
Beijers, Bijlevel,
and van Poppel
Netherlands Men only 971 Fixed- and random-
effect Poisson
regression analysis
Benecial if married
after 1970
Bersani and
Doherty (2013)
United States NLSY97 Men and
2,838 Hierarchical linear
Benecial 4
Bersani, Laub, and
Netherlands CCLC Men and
4,615 Hierarchical linear
Benecial 79
Blokland and
Netherlands (A) CCLC;
(B) Dutch
Crime Survey
Men and
(A) 4,615
(B) 2,951
trajectory modeling
Benecial 167
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Burt et al. (2010) United States Minnesota
Twin Family
Men only 289 twin pairs Multilevel
modeling (twin
Benecial 16
Capaldi, Kim, and
Owen (2008)
United States OYS and the
Men only 110 Zero-inated
Poisson modeling
Benecial (only for
stability among
men with prior
Craig (2015) United States Add Health
(waves 14)
Men and
3,327 OLS regression Marriage is found to
lead to changes in
levels of offending
among whites and
Hispanics but not
Craig and Foster
United States Add Health
(waves 13)
Men and
3,082 Survey-adjusted
OLS regression
Benecial for both
men and women
Doherty and
United States Woodlawn
Men and
965 Hierarchical linear
Benecial (at least
for men)
Wilkerson, and
England (2006)
United States NLSY79 Men and
depending on
Logistic OLS and
Tobit regression
Benecial (but only
for married men
and marijuana use)
Farrington and
West (1995)
Men only 378 (73 for
OLS and logistic
Benecial 168
TABLE 1 (Continued)
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Forrest and Hay
United States NLSY79 Men and
2,325 Logistic regression Benecial for
marriage, not
signicant for
not signicant
estimate for
married when
controlling for
Cernkovich, and
Rudolph (2002)
United States Ohio Life-
Course Study
Men and
197 Regression analysis No association
between spousal
and criminal
Giordano et al.
United States TARS Men and
1,066 Hierarchical linear
No signicant
Godfrey, Cox, and
Farrall (2007)
(end of 19th
century) data
from Crewe,
Men and
101 No signicant
difference in
conviction rates
between married
and unmarried;
convictions went
up after marriage
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Herrera, Wiersma,
and Cleveland
United States Add Health Men and
1,267 romantic
partners pairs
Positive impact
of relationship
quality and
Horney, Osgood,
and Marshall
United States Survey of
sentenced to
Dept. of
Men only 617 Hierarchical linear
Living with
signicantly more
likely to commit
any crime or drug
crime; living with
wife: signicantly
less likely to
commit an assault;
otherwise not
Jaffee, Lombardi,
and Coley (2013)
United States Add Health
(waves 14)
Men only 4,149 Four different:
logistic regression,
propensity score
matching models,
models, and
models of sibling
Benecial 1
Kerr et al. (2011) United States OYS Men only 206 Multilevel modeling Benecial for
marriage, not
signicant for
TABLE 1 (Continued)
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
King, Massoglia,
and Macmillan
United States NYS, 1976 Men and
1,725 Propensity score
Benecial (only
for men [all]
and women
[with moderate
propensity to
Matsueda, and
Erosheva (2010)
United States Denver Youth
567 Fixed-effect logistic
No signicant
between marriage
and delinquency;
negative association
between marriage
and drug use
Uggen, and
Shelton (2000)
United States Minnesota
based sex
Men and
556 Cox regression No signicant
at sentencing and
reoffense or union
dissolution and
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Laub and Sampson
United States Glueck and
study on
Men only 52 (ages 1770) and
419 (ages 1732)
Poisson model,
Benecial 1,310
Laub, Nagin, and
Sampson (1998)
United States Glueck and
study on
Men only 480 Semiparametric
Poisson mixture
model and
hierarchical linear
Benecial 659
Leverentz (2006) United States Women in
halfway house
in Chicago
49 Qualitative
Lyngstad and
Norway Norwegian
register data
Men only 90,919 Logistic regression Benecial only in
the years before
Massoglia and
Uggen (2007)
United States Youth
Men and
1,000 Logistic regression Relationship quality
benecial for all
four outcomes
Maume, Ousey,
and Beaver
United States NYS (waves 5
and 6)
Men and
593 Logistic and
bivariate probit
Benecial, but only
for those with
high marital
McGloin et al.
Netherlands CCLC Men and
4,612 Random- and
effect models
Benecial 7
McGloin et al.
United States Survey of
sentenced to
Dept. of
Men only 658 Hierarchical linear
TABLE 1 (Continued)
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Terovan, and
van der Geest
Netherlands 17Up Men and
540 Trend vector model Benecial 0
Dimitrieva, and
Cauffman (2014)
United States Pathways to
Men and
354 Conditional growth
No impact of
current romantic
length benecial
only for boys,
opposite for girls
OConnell (2003) United States Ongoing studies
project for
those at risk
for drug use
Men and
576 Structural equation
modeling (SEM)
Marriage not
Burgers, and
Reppucci (2014)
United States GAP Women
114 Regression Negative association
between relationship
length and violent
offending; no
between other
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
and Piquero
Netherlands CCLC Men and
5,000 Two-part latent
growth curve
Benecial on both
prevalence and
MacDonald, and
Parker (2002)
United States California
Men only 524 Negative-binomial
model with
disturbance term
Marriage benecial
for total arrests
and nonviolent
arrests; when
stratied by race:
signicantly more
violent arrests
among whites
after they marry
Ragan and Beaver
United States Add Health
(waves 13)
Men and
1,884 Logistic regression Benecial 11
Salvatore and
United States Add Health
(wave 3)
Men and
4,880 Negative binomial
Benecial 1
Sampson, Laub,
and Wimer
United States Glueck and
study on
Men only 440 to age 32,
52 to age 70
Inverse probability
of treatment
weighting and
Poisson modeling
Benecial (both
marriage and
Sampson and Laub
United States Glueck and
study on
Men only 622 at the most Regression analysis,
event history
Signicant negative
between marital
attachment and all
measures of crime
and deviance
TABLE 1 (Continued)
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Sampson and Laub
United States Glueck and
study on
Men only 622 at the most Regression analysis Signicant negative
between marital
attachment and all
measures of crime
and deviance
Savolainen (2009) Finland National
(register data)
Men only 1,325 Negative binomial
not signicant
Giordano, and
United States Ohio Life-
Course Study
Men and
152 Regression analysis No signicant
partner happiness
and desistance or
persistance in full
Simons et al. (2002) United States Iowa Youth and
Men and
236 and their
SEM and logistic
Positive association
between having
antisocial romantic
partner and
criminal behavior;
quality of romantic
signicant only
for females
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Simons and Barr
United States FACHS Men and
589 Negative binomial
No signicant
between having a
romantic partner
and criminal
behavior, but
signicantly less
crime for those
having a high-
quality relation-
ship (although not
signicant when
controlling for
and Lyngstad
Norway Norwegian
register data
Men only 80,064 Generalized additive
Benecial in the
years before
marriage, but
potentially also for
those who marry a
criminal woman
Theobald and
Men only 319 Propensity score
matching and lo-
gistic regression
Separated men who
formed a new
(cohabitation or
remarriage) had
a smaller increase
in offending
than those who
TABLE 1 (Continued)
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
Theobald and
Men only 111 Propensity score
matching and
analysis of covari-
Benecial only if
married before
age 25
Theobald and
Study in
Men only 140 Propensity score
Benecial only if
married before
age 25
Thompson and
Petrovic (2009)
United States NYS (waves 5
Men and
1,496 Logistic panel
models with xed
Marriage benecial
(only men);
strength benecial
for women;
increasing drug
use for men
Uggen and
United States National Sup-
ported Work
tion Project
Men and
Illegal earnings:
2,667; arrests:
Cox proportional
hazard model
No signicant
between living
with spouse and
outcome variables
Study Country Data Source Gender
Sample Size Methods
Findings on the
Marriage Effect
according to
March 2014
van Schellen, Apel,
Netherlands CCLC Men and
4,615 Fixed-effect Poisson
For men: benecial
only if married to
spouse; for
women: benecial
Warr (1998) United States NYS (waves 5
and 6)
Men and
1,725 Logistic regression Benecial, but not
associations when
controlling for
having delinquent
friends and time
spent with friends
Terovan et al.
Netherlands 17Up Men and
540 Fixed and random-
effect models
Benecial (signicant
only for men);
No signicant
association between
marriage duration
and offending
Terovan et al.
Netherlands 17Up Men and
93 couples Cox regression 0
Farrington 1989; Sampson and Laub 1990). None of the six suggested
that marriage per se contributed to signicant reductions in crime. For
example, West (1982) found no signicant difference in self-reported
delinquency between married and unmarried men but observed that de-
linquents were more likely to marry delinquent women. Farrington
(1989) found convicted men to be less likely to get along with their
spouses. Gibbens (1984) found the association to be causally ambiguous
because the men were more likely to marry as their lives were becoming
more stable. Sampson and Laub (1990) found that marriage was not asso-
ciated with reduced offending but presented evidence suggesting that
marital attachment reduced offending.
Wright and Wright (1992, p. 54) concluded that no clearly conrm-
ing set of ndings has emerged from research to date that demonstrates
that getting married and having children reduces the likelihood of crim-
inal offense. This is no longer the dominant impression (see g. 3).
There has been a tremendous growth in availability of longitudinal data
FIG.3.Number of studies published by year. The gure for 2014 is for the rst three months
of the year.
408 Torbjørn Skardhamar et al.
sets and increased methodological sophistication since the early literature
on marriage and crime. The contemporary opinion is much more con-
dent about the capacity of marriage to exert restraining inuence on
crime (Craig, Diamond, and Piquero 2014). This condence is under-
standable given that of the 58 publications that met our selection criteria,
36 found marriage (or related states) to be associated with desistance from
crime. Nine additional studies found mixed evidence. Thus, as many as
45 (78 percent) studies reported at least some support for the expectation
that marriage is associated with reduced offending.
Because much of this literature is based on a relatively small number
of longitudinal data sets (see g. 4), we grouped the 58 studies under
headings indicating the data source used. The length of the bar in gure 4
shows the number of publications generated by the data set. The num-
ber next to the bars provides the combined citation count from all the
reviewed studies using these data. In terms of citations, the Glueck study
resurrected by Sampson and Laub (1993) is in a league of its own. With
6,441 citations, it is more inuential than the rest of the studies com-
In terms of publication count, the National Longitudinal Study
of Adolescent Health (Add Health) leads the way with nine articles,
followed by the Dutch Criminal Career and Life Course study (CCLC)
and the Glueck study. It is notable, however, that two of the publications
from the Glueck data are book-length monographs. The Cambridge
Study, the National Youth Study (NYS), and the National Longitudinal
Studies (1979 and 1997) of Youth (NLSY) are also prominent. The
Ohio Life-Course Study and the Nebraska Inmate Sample are inuen-
tial in terms of citation counts.
A. Glueck Sample
Sampson and Laubs seminal research program is grounded on re-
analysis and reconstruction of longitudinal data collected by Sheldon
and Eleanor Glueck (1940) when the boys were 14 years of age and ad-
ditional waves at ages 25 and 32. The Gluecks sampled 500 boys born
between 1924 and 1935 from a correctional school in Boston and
matched them with same-aged boys from a normal population sample.
The rst article by Sampson and Laub (1990) examined the association
Citation count should be treated as only a rough indicator of inuence. The mono-
graphs by Sampson and Laub are general treatises that cover multiple topics of life course
criminology, while many of the other studies are more narrowly focused.
Does Marriage Reduce Crime? 409
between spousal attachment and rates of crime and deviant behavior at
ages 1725 and 2532, controlling for confounding factors measured in
childhood and early adulthood. In analysis of the full sample, they found
no association between marriage and offending as measured by both
self-reported delinquency and ofcial arrests. However, they found a
negative association between marital attachment and crime in the anal-
ysis limited to ever-married men. Sampson and Laubs Crime in the
Making (1993) largely replicated the results from the 1990 article with
some adjustments and elaborations. For example, some of the analyses
were extended to ages 3245. Their main conclusion was that child-
hood and adult social bonds in the form of job stability and marital at-
tachment independently explain signicant variations in adult crime
(1993, p. 178).
FIG.4.Number of publications by data set. Sums of citations according to Google Scholar
are to the right of each bar. The gures for CCLC include one study also including the Dutch
national crime survey (Blokland and Nieuwbeerta 2005).
410 Torbjørn Skardhamar et al.
In a subsequent article, Laub, Nagin, and Sampson (1998) analyzed the
Glueck sample from ages 14 to 32. Their theoretical argument was fo-
cused on the quality of marital bonding and the incremental nature of this
effect. Importantly, this study attended to the possibility of a courtship
effect, that is, the potential effect of romantic bonding prior to entering
marriage. They found no evidence of a courtship effect. However, among
those with strong marital attachment, the estimates showed evidence of an
incremental decrease in offending in each annual time point during the
postmarriage period. Consistent with the hypothesis, no such effect was
observed among men with weak spousal attachment.
In order to extend the original data to later points in the life course,
Laub and Sampson (2003) made an effort to track the men from the
Glueck study. They were able to interview 52 individuals and to update
their conviction records up to age 70. On the basis of retrospective data
from life history narratives, the quantitative analyses estimated annual
changes in the number of convictions. The results showed reduced rates
of offending during states of marriage. Importantly, these models ignored
variation in the quality of marital bonding; the results were generalized
to marital unions of any kind. Evidence from qualitative interviews pro-
vided additional evidence in support of marriage as a major turning point.
One of the men stated that his wife straightened him out (p. 121) and that
marriage probably even saved his life (p. 134). The key message of the
2003 monograph was that marriage may lead to desistance because of di-
rect social control effects by spouses (p. 136) but also that in-laws pro-
vided additional support (p. 137). The latter is related to the assumption
that these men usually married up and gained access to additional re-
sources through the extended family of the wife. Overall, the qualitative
evidence was consistent with processes of informal social control, change
in routine activities, identity shift, and social capital formation through
the extended family.
From the perspective of causal inference, Sampson, Laub, and Wimer
(2006) offer the most stringent test of the marriage effect on desistance.
They used inverse probability of treatment weighting as the methodolog-
ical approach to imitate an experimental design.
The idea is to approx-
imate a comparisonwithin the limits of observational databetween in-
We discuss this method further in the next section as we evaluate the evidence on
Does Marriage Reduce Crime? 411
dividuals who are otherwise as similar as possible with the exception of
marital status. They concluded that marriage is associated with a 35 per-
cent average reduction in crime (p. 496). Moreover, this reduction was
greater among those reporting high marital attachment. In an additional
analysis, they report evidence suggesting that a similar association applies
to cohabiting couples. Thus, having a stable partner may promote desis-
tance regardless of union type according to this study.
B. National Youth Study
The NYS is a panel study of 1,725 males and females selected to be
representative of the US population. The study started in 1976 when
the participants were between 11 and 17 years of age, and they have been
followed up so far to ages 3945. The study is now called the National
Youth Survey and Family Study.
Warr (1998) used the NYS data waves 5 and 6 to examine the relation-
ship between marriage and desistance. His theoretical goal was to illumi-
nate the mechanisms responsible for the marriage effect. He advanced
the argument emphasizing the salience of differential association pro-
cesses as an alternative to the social bonding explanation offered by
Sampson and Laub (1993). He argued that marriage is inuential be-
cause it restructures activities away from (delinquent) male peers and to-
ward increased time with the spouse and in the family setting. This
study documented a signicant negative association between marital sta-
tus and three indicators of self-reported delinquencypetty theft, van-
dalism, and marijuana useall measured at wave 6. Consistent with the
hypothesis, this association disappeared when controlling for differences
in exposure to delinquent friends. Owing to data limitations, the more
compelling analysis, based on change scores between waves 5 and 6, was
possible only for marijuana use as the outcome variable. This test was con-
sistent with the ndings from the more comprehensive cross-sectional
analysis. Warr concluded that marriage reduced delinquency because of
its effect on peer associations.
Maume, Ousey, and Beaver (2005) used the NYS to reexamine Warrs
hypothesis but focusing on marital attachment rather than mere marital
status as the measure of romantic bonding. Consistent with the AGT,
high levels of marital attachment were related to desistance, and the
For more information, see
412 Torbjørn Skardhamar et al.
effect persisted even controlling for the inuence of delinquent peer
associations. This analysis suggests that marital attachment has a direct
effect on desistance, with only a small portion being due to delinquent
peers (p. 48).
Ignoring the question of mechanisms, King, Massoglia, and Mac-
millan (2007) used NYS data to estimate the causal effect of marriage
on desistance. They employed propensity score matching on data from
NYS wave 7, when the average age of the participants was 24 years.
Marital status was measured as being married the year before, and a
large number of covariates from previous waves were used to estimate
the propensity to marry. Their main nding was that marriage reduced
crime among men but not signicantly among women. Moreover, they
stratied the sample on the propensity score, interpretable as the prob-
ability of marrying. This analysis suggests that, although the effect of
marriage on crime applies to all men, it is the strongest among those
least likely to marry.
Thompson and Petrovic (2009) found that marriage reduced drug use
among men and women. In addition, they included an index of relation-
ship strength, which covered married as well as nonmarried couples. Con-
trolling for marital status, relationship strength was associated with fur-
ther declines in female drug use but, surprisingly, an increase among men.
C. The Ohio Life-Course Study and the Toledo Adolescent Relationship Study
The sample for the Ohio Life-Course Study was drawn in 1982 from
inmates in state-run institutions for delinquent girls (N p 127) and boys
(N p 127). The participants were contacted again in 1995 and 2003.
The data were analyzed using quantitative as well as qualitative methods.
Focusing on the two rst waves, Giordano, Cernkovich, and Rudolph
(2002) found no signicant association among males or females between
attachment to spouse and criminal involvement. They proposed that it is
the combination of marriage and stable jobs that provides sufcient
bonding and social control to assist in desistance. They refer to this com-
bination as a respectability package. They found that individuals with
a high-quality package (i.e., marital happiness and stable earnings) were
more likely to desist. Evidence from qualitative interviews highlighted
the importance of initial willingness to change as well as structural con-
straints limiting opportunities to achieve life goals consistent with the
motivation to go straight.
Does Marriage Reduce Crime? 413
Giordano, Cernkovich, and Holland (2003) elaborated on these nd-
ings with a special focus on understanding dynamics in romantic rela-
tionships. They argued that marital (or romantic) attachment is unlikely
to reduce crime unless the spouse truly wishes and makes an effort to limit
behaviors and habits conducive to criminal involvement. They argued that
a decision to align with a relatively more conforming partner generally
represents an afrmative, agentic move away from a criminal lifestyle.
And the spouses respectability is often a critical component of the actors
own continuing identity transformation project (p. 306). They also sug-
gested that an intimate partner can be used to invoke a kind of cover
for not participating in criminal opportunities as now being a family
man (p. 305). Under this framework, marriage may help redirect lives
away from crime, but this pathway is preceded by a subjective desire to
change. The relationship quality is not something that happens but
reects a choice to gravitate toward a partner with these characteristics
(p. 321).
Using all three waves of the Ohio Life-Course Study, Schroeder,
Giordano, and Cernkovich (2007) investigated the role of drug use in
the desistance process. This study was informed by Laub and Sampsons
(2003) conjecture that involvement in substance misuse sustained crim-
inality because this lifestyle made it difcult to establish bonds to mar-
riage and work (p. 212). Schroeder, Giordano, and Cernkovich (2007)
did not nd an association between marriage and crime, and the associ-
ation between drug use and crime was not mediated by marital quality.
The Toledo Adolescent Relationships Study (TARS) is based on a
stratied random sample of 1,066 students in seventh, ninth, and elev-
enth grades across 62 schools in seven school districts, followed until
age 21 on average. Marital status was included as a covariate with a pri-
mary focus on parenthood. Giordano et al. (2011) found no signicant
association between cohabitation/marriage and offending. However,
parents (mothers and fathers) who resided with the other biological par-
ent were found to experience marginally signicant reductions in crim-
inal activity (p. 9). Evidence from qualitative interviews suggested that
many of the romantic partners would be unlikely to provide a solid
prosocial anchor (p. 9).
In sum, the work of Giordano and colleagues challenges the idea that
marriage or even marital attachment promotes desistance from crime.
Rather, marriage (to a conventional partner) in combination with stable
income is more likely to reect the individuals decision to change and
414 Torbjørn Skardhamar et al.
ability to capitalize on opportunities to do so. Importantly, the results
indicate that a large number of individuals in their sample were aficted
with disadvantages that made it very difcult for them to realize their
motivation to change.
D. Criminal Career and Life Course Study
Several important studies of marriage and crime have been based on
the Dutch CCLC study, which is a random 4 percent sample of all
persons convicted in the Netherlands in 1977 (N p 4,164). The CCLC
contains information about the entire conviction history up to 1977 and
all subsequent convictions through 2002. In the rst publication using
these data, Blokland and Nieuwbeerta (2005) found that marriage was
associated with a within-individual reduction in crime among offenders
of low to moderate frequency, but not for sporadic offenders.
The CCLC is not a birth cohort study but follows a random cross sec-
tion of convicted offenders who vary in age and who married at different
periods of time. The oldest members of the sample were born in 1907
and the youngest in 1965. Sensitive to this issue, Bersani, Laub, and
Nieuwbeerta (2009) stratied the sample into three groups based on
age in 1977: those aged 32 and up (born before 1945), those aged 22
31 (born 194655), and those aged 1221 (195665).
They found that
marriage was associated with a reduction in crime across each cohort,
but more strongly among males and among the more recent cohorts.
McGloin et al. (2011) used the CCLC to examine the effect of marriage
on versatility of criminal offending. They found marriage to be asso-
ciated with a decline in versatility, and this was not fully explained by reduc-
tions in the frequency of offending. Moreover, they report evidence sug-
gesting that the declining trend in offending started before the year of
marriage (p. 370). Petras, Nieuwbeerta, and Piquero (2010) found mar-
riage to be associated with a decline in both the prevalence and frequency
of offending. Another study investigated whether the marriage-crime
association depends on the criminal history of the spouse (van Schellen,
Apel, and Nieuwbeerta 2012). The results showed that those who married
a criminal spouse did not experience any decline in offending. Additional
nuances are suggested by van Schellen, Poortman, and Nieuwbeerta
In a footnote, the authors discuss the concern that the sample is selected on being
convicted in 1977, making some by denition adult offenders (or serious offenders).
Does Marriage Reduce Crime? 415
(2012), who found that convictio ns, and particularly recent convictions,
lowered the chances of getting married and increased the chances of
marrying a criminal partner, and for the se reasons, offenders are less
likely to experience protective effects of marriage.
E. National Longitudinal Studies of Youth
There are three studies included in our review based on the National
Longitudinal Studies of Youth (NLSY79 or NLSY97). These are panel
surveys of nationally representative samples of men and women born in the
years 195764 and 198084, respectively. In each survey, the respondents
were 1217 years of age at the time of the rst interview.
Both studies
include a main sample and two supplementary samples. In the rst sup-
plementary sample, Hispanics or Latinos, blacks, and economically disad-
vantaged, nonblack/non-Hispanics were oversampled. The second supple-
mentary sample consists of persons serving in the military. The total sample
sizes of the NLSY79 and NLSY97 are 12,686 and 8,984, respectively.
Duncan, Wilkerson, and England (2006) examined changes in mari-
juana use across 11 waves from the NLSY79. The authors selected a sub-
sample of everyone married for the rst time and operationalized the
marriage period as the 5 years before and after the year of entering mar-
riage, so that each person contributed up to 11 person-years to the anal-
ysis. To isolate the potential marriage effect from other dynamic factors,
they specied a piecewise linear spline distinguishing the periods before
marriage (periods 25 through 22), surrounding marriage (periods 21
through 1), and after marriage (periods 11 through 15). Under this
design, the difference between the rst and second splines constitutes
the marriage effect (pp. 69596). They applied the same approach to
cohabitation. The ndings showed that marriage but not cohabitation
reduced marijuana use among men, but neither union type had any effect
among women.
Forrest and Hay (2011) proposed that objective life course transitions,
such as marriage, may inuence levels of self-control and that this may
be a mechanism that contributes to the effect of marriage on desistance.
Drawing on NLSY79 waves 19962004, Forrest and Hay examined this
hypothesis focusing on changes in marijuana use. The models included
a dummy predictor representing the state of marriage and another
For more information, see; for
NLSY97, see
416 Torbjørn Skardhamar et al.
dummy for cohabitation. These effects were estimated controlling for
a wide range of potentially confounding variables. Both marriage and
cohabitation were found to be associated with increased levels of self-
control, and marriage (but not cohabitation) was associated with re-
duced levels of marijuana use, even controlling for differences in self-
control. They concluded that changes in self-control mediate some of
the relationship between marriage and desistance.
Using 13 waves from the NLSY97, Bersani and Doherty (2013) exam-
ined whether there was a longitudinal association between marriage and
the risk of arrest, and what happens to this association when marriage
dissolves. In an attempt to devise a crucial test between two theories,
they argued that if identity transformation is the key mechanism of in-
uence, there should be little or no change following divorce; but if so-
cial bonding and changes in routine activities are the driving force, we
should expect an increase (or relapse) following divorce. They rst es-
tablished the presence of a negative association between state of mar-
riage and risk of arrest. To estimate the effect of divorce, the next step
of the analysis was limited to those who got married and, thus, were at
risk of divorce. The results showed strong increases in offending among
those who divorced.
F. National Longitudinal Study of Adolescent Health
Add Health is a nationally representative study of US adolescents
sampled initially during 199495. So far, there have been four waves
of data collection up to ages 2432. A total of 20,745 students were in-
terviewed in the rst wave. Add Health includes a wide range of health-
related information, including biomarkers, such as saliva samples for
genotyping. Participants in the genetic subsample (N p 2,574) had a
twin or sibling participating in the Add Health study.
Focusing on the genetic subsample, Beaver et al. (2008) selected one
sibling from each twin pair (N p 1,994) to examine genetic and environ-
mental correlates of desistance. They found that married persons were
more likely to desist from crime. The serotonin transporter gene was
associated with desistance for females, and there were signicant inter-
action effects between marriage and the transporter genes DRD2 and
DRD4 and monoamine oxidase A (MAOA). They concluded that the
effect of marriage is heterogeneous as a function of genetic differences.
Barnes and Beaver (2012) used a sibling and twin design to shed
light on to what extent genetic confounders account for the frequently
Does Marriage Reduce Crime? 417
observed marriage effect. The study capitalized on information about
the zygosity of twins, relatedness of siblings (half vs. full siblings), and
even cousin relationships. This information was used to decompose the
variances reecting genetic similarities and thereby isolating the inuence
of genetic factors on desistance from crime. The ndings showed a nota-
ble genetic component to the association between marriage and desis-
tance: the association was reduced by more than 50 percent once cleaned
from the inuence of genetic relatedness.
Jaffee, Lombardi, and Coley (2013) used four complementary meth-
ods of analysis to assess the effect of marriage on crime: lagged multivar-
iate regression models, propensity score matching, xed-effects models,
and sibling analysis. Each methodological approach found marriage to
be associated with reduced levels of delinquent offending, suggesting
that the nding is robust.
Another study based on Add Health found that marriage was associ-
ated with a reduction in crime among both males and females. Focusing
on racial differences, Craig (2015) found support for the marriage effect
among Caucasians and Hispanics, but not among African Americans.
Salvatore and Taniguchi (2012) estimated the effects of several indicators
of social bonding and turning points on criminal offending, nding that
marriage was associated with reduced offending. Ragan and Beaver (2010)
found marriage to be associated with desistance from marijuana use be-
tween waves 1 and 2 of Add Health data.
Barnes et al. (2014) noted that prior studies of the marriage effect have
failed to address the possibility that the association between marriage
and crime may be reciprocal, that is, that crime may affect marriage.
They estimated cross-lagged structural equation models between mari-
tal status and criminal behavior in waves 3 and 4 of Add Health. They
found that being married was associated with reductions in crime but
also that crime was negatively related to the probability of marriage. When
accounting for reciprocal effects, the direct effect of marriage on crime
was no longer statistically signicant. This study suggests that ignoring
reciprocal pathways in the association results in overstated estimates of
the marriage effect.
G. The Cambridge Study in Delinquent Development
The Cambridge Study is one of the most cited in the criminal careers
literature. The data include 411 boys from a working-class neighbor-
hood in South London, rst interviewed in 196162 at the ages of 8
418 Torbjørn Skardhamar et al.
and 9. The sample includes all boys in this age group who lived in that
area at the time. The data collection consists of nine waves, the last of
which occurred when the sample participants were age 48.
Farrington and West (1995) estimated a set of regression models
predicting convictions between ages 21 and 32, convictions in the 5 years
before the interview at age 32, and self-reported offending in the last
5 years up to age 32. They found stable marriage to be negatively associ-
ated with crime. They also conducted a within-individual analysis com-
paring the number of convictions in the 5 years before marriage and
the 5 years after marriage and by matching each married man with an
unmarried man on the basis of the number of convictions in the 5 years
before marriage. The results showed that men who married accrued sig-
nicantly fewer convictions after marriage.
Two recent studies focused on a subsample of 162 convicted men
from the Cambridge Study, excluding those whose marriages lasted less
than 5 years (Theobald and Farrington 2009, 2011). These men got mar-
ried at different ages and were matched using propensity score methods to
those not (yet) married at the same age. Theobald and Farrington (2009)
concluded that early marriage (1824 years of age) was associated with
reduced offending, but they did not observe a similar pattern among those
who married later (251 years of age). In a later study, they explored the
characteristics of those who married early versus late (Theobald and
Farrington 2011) and found that those who married late were different
from those who married early. They were more likely to have grown
up in a broken home, to be more nervous, and to have married a woman
older than themselves (p. 153).
H. Nebraska Inmate Sample
Horney, Osgood, and Marshall (1995) interviewed a sample of 658
newly convicted men sentenced to prison in Nebraska in 198990. The
inmates were interviewed using a life history calendar covering 25
36 months leading up to the arrest leading to imprisonment. The instr u-
ment recorded monthly changes in offe nding b ehavio r and local life
circumstances, such as marriage and employment. Using a hierarchical
linear model (HLM ) with covariates centered at individuals mean
values, the study estimated within-individual changes in rates of of-
fending in response to changes in life circumstances. They found that
living with a wife was negatively associated with offending, while cohab-
itation increased the risk of crime. In a reanalysis of the same data,
Does Marriage Reduce Crime? 419
McGloin et al. (2007) showe d that marriage was also negatively associ-
ated with offending versatility.
I. Studies Based on Nordic Administrative Register Data
Desistance scholars in Finland and Norway have taken advantage of
the systems for administrative records and ofcial statistics available in
the Nordic countries. These data sets typically cover the total popula-
tion, with longitudinal, individual-level information on marital events,
crime, and a wide range of other characteristics, such as educational at-
tainment and socioeconomic background. The sources have been used
widely in demography and public health but remain relatively untapped
in criminology (Lyngstad and Skardhamar 2011).
Savolainen (2009) examined a national sample of felony offenders
from Finland. They were convicted in 1996 and tracked through 2001
using data from interlinked population registries. The study focused
on a subsample of high-rate offenders who were unemployed, single
(neither married nor cohabiting), and childless at the start of the track-
ing for new offenses (beginning of 1996). The study sought to examine
whether ch anges in offending were related to tran sitions to adult social
roles (work and family). With respect to marriage, Savolainen found
that living in a romantic union (i.e., marriage or cohabitation) was as-
soci ated with signicant reductions in the rate of offending . However,
when this association was disaggregated by marital status, it turned out
that this effect was generated by changes among men who were co-
habiting; the marriage effect was no longer statistically signicant. As a
possible explanation of the unexpected ndings, he proposed that, in
Finland, marriage is typically preceded by several years of cohabitation.
In this normative context, women who are relatively quick to marry a
man with an intense criminal history may be more approving of the crim-
inal lifestyle than women who delay marriage. In other words, in this sit-
uation, men in the cohabitation category may be more likely to live with a
prosocial partner.
Lyngstad and Skardhamar (2013) used Norwegian data on all men
who got married between 1997 and 2001 and examined their annual
rates of ofcially recorded offending 5 years before and after the year
of marriage. The results demonstrated a declining trend in offending
during the period leading up to marriage but no additional decline after
marriage. As in Finland, the suggested explanation was that marriage is
typically preceded by several years of cohabitation and that entering
420 Torbjørn Skardhamar et al.
marriage may be a poor proxy for when the potential effects of romantic
bonding take place. Skardhamar, Monsbakken, and Lyngstad (2014)
used a similar design to investigate whether results varied by the criminal
history of the wife. Unsurprisingly, men who married a criminal woman
were more involved in crime themselves. However, these men also expe-
rienced the greatest declines in offending, and some of that occurred
after marriage.
J. 17Up
Another Dutch study, 17Up, sampled 270 males and 270 females
released from a juvenile treatment center in the Netherlands. Recorded
crimes were collected from age 12 onward and linked to information on
marital status and other demographic variables available from the Munic-
ipal Population Registration database. The data le also contains infor-
mation gleaned from the treatment les of the juvenile institution. Studies
using these data show that marriage was associated with reduced offend-
ing in property crime among males and females but with an increase in
violent crime among men (and a decrease among women). No signicant
association between marriage and drug crimes was observed for either
gender (Mercer, Zoutewelle-Terovan, and van der Geest 2013). The
duration of marriage had no additional effect over and above marital sta-
tus (Zoutewelle-Terovan et al. 2012), but having a deviant spouse was
shown to support a criminal lifestyle (Zoutewelle-Terovan et al. 2013).
K. Additional Studies
Some data sources were associated with only one relevant publication.
The Youth Development Study features a sample of 1,000 students from
public schools in St. Paul, Minnesota. Massoglia and Uggen (2007)
showed that relationship quality was consistently associated with desis-
tance across multiple measures of offending. The Denver Youth Study
is a probability sample of households from disadvantaged neighbor-
hoods in Denver, Colorado. Using a subsample of 567 women, Kreager,
Matsueda, and Erosheva (2010) found no association between marriage
and delinquency, with the exception of reduced marijuana use. In a sam-
ple of 965 rst-grade children from the nine public and three parochial
schools located in a disadvantaged Chicago neighborhood, marriage
was a consistent correlate of desistance among men, but the results varied
by type of offending among women (Doherty and Ensminger 2013).
Does Marriage Reduce Crime? 421
OConnell (2003), studying recidivism in a sample of persons with serious
drug problems serving a prison sentence in Delaware (N p 576), found
marriage to be unrelated to either rearrest or drug use. Uggen and Krutt-
schnitt (1998) used a sample consisting of participants in a supported em-
ployment experiment to explor e factors related to desistance. Living
with a spouse was not a statistically signican t correlate. In a study of
206 me n from a medium-sized metropolitan area, marr iage wa s asso-
ciat ed with reductions in self-reported crime, fewer arrests, and reduc ed
marijuana use, while cohab itation was not statistically signicantly re-
lated to these outcome s (Kerr et al. 2011). In a 7-year follow-up of
524 parolees from California, Piquero, MacDonald, and Parker (2002)
found that marriage was negatively associated with nonviolent arrests
for bot h whites and nonwhites but not for violent arre sts. Common-
law marriages were associated with increased offending among non-
whites but not among white s.
A number of studies have focused on attachment or other aspects of
relationship quality. Using data from the Oregon Youth Study (OYS),
a community sample of 206 high-risk men, Capaldi, Kim, and Owen
(2008) identied relationship stability as a key predictor of desistance, while
attachment to a partner was not inuential. Simons and Barr (2012) used
data from the Family and Community Health Study (FACHS), consisting
of 100 African Americans living in Iowa or Georgia. The results showed
that a nonmarital relationship in early adulthood was associated with
desistance only if it was warm and supportive and that change in peer
associations explained a marginal proportion of the association. Simons
et al. (2002) used a sample of 236 persons across eight counties in Iowa
interviewed in ninth grade and then again 6 years later. Having a conven-
tional romantic partner and a high-quality relationship were associated
with a reduction in adult crime. Among males, this association was insig-
nicant once exposure to delinquent friends was taken into account.
Several studies have focused on the effect of partner characteristics,
especially the criminality of the spouse. On the basis of a sample drawn
from juvenile courts in two US metropolitan areas, Monahan, Dmitrieva,
and Cauffman (2014) found evidence suggesting that having a prosocial
partner was associated with a decrease in antisocial behavior. The Gender
and Aggression Project (GAP) sampled 141 girls from a juvenile correc-
tion facility and followed them up to 2023 years of age (Oudekerk,
Burgers, and Reppucci 2014). This research suggests that living in a stable
romantic relationship tempers offending only among those with a non-
422 Torbjørn Skardhamar et al.
criminal partner. Leverentz (2006) conducted qualitative interviews of
14 women living in a halfway house after release from prison. Many had
an antisocial partner unlikely to curb drug use and offending. In some
cases the couple sought treatment together, providing mutual support
for recovery.
Some of the studies stand out because of the unique character of the
data or the research design. Burt et al. (2010) analyzed 289 male twin
pairs from the Minnesota Twin Family Study and found that the within-
pair association between marriage and crime was signicant for mono-
zygotic twins, suggesting an association independent of a clear selection
effect into marriage. Kruttschnitt, Uggen, and Shelton (2000) studied
a sample of sex offenders placed on probation in Minnesota in 1992
(N p 556). In a 5-year follow-up, they found that marital status exerted
almost no effect on recidivism. Focusing on a cohort of men who were
criminally active in late nineteenth- and early twentieth-century En-
gland, Godfrey, Cox, and Farrall (2007) found no evidence of desistance
following transition to marriage. They suggest that wives were econom-
ically too dependent during this period to exercise effective social con-
trol. Finally, Beijers, Bijleveld, and van Poppel (2012) examined an inter-
generational sample consisting of descendants of males placed in a Dutch
reform school between 1911 and 1914. They found that the restraining
effect of marriage was limited to cohorts entering marriage after 1970.
L. Overview of Patterns
We conclude with a brief summary of main patterns. One important
theme has to do with the salience of marriage versus cohabitation,
inasmuch as cohabitation has become increasingly common in recent
decades. Whether marriage affects both women and men has been dis-
cussed frequently. So has the universality or specicity of the marriage
effect across historical periods and cultural contexts.
1. Cohabitation. Some scholars have suggested that marriage may be
qualitatively different from other romantic relationships: taking the
conscious step to get married demonstrates some non-negligible commit-
ment, being married appears to be more protective of future offending
than cohabiting (Craig,Diamond,andPiquero2014,p.33).Although
we nd this assumption plausible, it is worth emphasizing that it under-
scores systematic selection into marriage as an important mechanism of
inuence. Perhaps marriage is associated more strongly with desistance
Does Marriage Reduce Crime? 423
than cohabitation because tying the knot is more likely to occur after
sustained behavioral change.
Some studies nd no signicant association between cohabitation and
crime (Duncan, Wilkerson, and England 2006; Schroeder, Giordano,
and Cernkovich 2007; Forrest and Hay 2011; Giordano et al. 2011; Kerr
et al. 2011), and two studies found that living with a girlfriend increased
offending (Horney, Osgood, and Marshall 1995; Piquero, MacDonald,
and Parker 2002). Contrary to these, other studies have found cohabita-
tion to have effects similar to those of marriage (Sampson, Laub, and
Wimer 2006) or even more important than marriage (Savolainen 2009).
One recent study found cohabitation to be associated with a reduction
in substance abuse, but the association was not statistically signicant
for criminal offending (Siennick et al. 2014). Irrespective of the formal
status of the union, the evidence on relationship quality is fairly consistent.
Virtually all the studies that include measures of good marriage or
cohabitation report signicant effects in the expected direction (Simons
et al. 2002; Massoglia and Uggen 2007; Capaldi, Kim, and Owen 2008;
Simons and Barr 2012; Oudekerk, Burgers, and Reppucci 2014).
2. Gender Differences. The literature has been dominated by studies
of male-only samples. The question has been raised whether the associ-
ation holds for women. Women have lower crime rates than men, and
their role in the family remains different. If Laub and Sampson (2003)
are correct that many male offenders marry up in terms of resources
as well as criminal propensity, it follows that women marry down.
Whether this is true for offenders has not been empirically established,
as none of the studies investigated the spouses socioeconomic char-
acteristics. Evidence from the demographic literature suggests that it
is women, not men, who marry up (Birkelund and Heldal 2003).
A number of studies have found the marriage effect to be stronger for,
or even limited to, men (e.g., Duncan, Wilkerson, and England 2006;
King, Massoglia, and Macmillan 2007; Beaver et al. 2008; Bersani, Laub,
and Nieuwbeerta 2009; Zoutewelle-Terovan et al. 2012; Doherty and
Ensminger 2013; Mercer, Zoutewelle-Terovan, and van der Geest 2013).
Others found no gender difference in the association between marriage
or marital attachment and crime (Giordano, Cernkovich, and Rudolph
2002; Simons et al. 2002). Van Schellen, Apel, and Nieuwbeerta (2012)
found greater reductions in crime among females following marriage
(see also Giordano et al. 2011). Thompson and Petrovic (2009) found
no gender difference in the association between state of marriage and
424 Torbjørn Skardhamar et al.
illicit drug use, but the strength of marital attachment reduced drug use
among women and increased it among men.
3. National Context. Bersani, Laub, and Nieuwbeerta (2009) argued
that the association between marriage and crime is universal and robust
across sociohistorical context, but other studies have provided evidence
to the contrary (e.g., Godfrey, Cox, and Farrall 2007; Savolainen 2009;
Lyngstad and Skardhamar 2013). The meaning of marriage may vary
among the United States, southern and northern Europe, and especially
the Nordic countries, where cohabitation is widespread and in many
ways comparable to marriage. As shown in gure 5, the vast majority
of the studies on marriage and crime are based on data from the United
States. The 17 publications from outside the United States represent a
relatively narrow selection of nations: the Netherlands, England, Fin-
land, and Norway. It is premature to make strong conclusions about
the universality of the association between marriage and crime.
FIG.5.Number of publications by the country from which the data are collected
Does Marriage Reduce Crime? 425
IV. So, Is There a Marriage Effect?
Belief in the marriage effect is strong in the criminological literature.
Most of the studies we examined frame their empirical ndings as con-
sistent with theories predicting a causal effect of marriage on crime.
Much of this evidence, however, is equally consistent with noncausal
theories, such as the maturation perspective. In order to demonstrate
causality, the evidence should rule out social selection as the alternative
account of the negative association between marriage and crime. The
social selection hypothesis posits that marriage is associated with desis-
tance because individuals who have cleaned up their acts, that is, ma-
tured out of crime, are more likely to marry than those who persist in
crime. Evidence that fails to rule out this interpretation does not dem-
onstrate causality.
Most research on the marriage-crime relationship relies on some tech-
nique of multivariate regression to control for the inuence of observed
confounders. It is increasingly well understood that this method can
establish causality only under the highly unrealistic assumption that all
the relevant factors inuencing the association have been perfectly mea-
sured and included in the model. If this assumption is not met, the results
are biased by unobserved heterogeneity: To the extent that marriage is
inuenced by individual self-selection, the marriage-crime relationship
is potentially spurious. . . . The bottom line seems to be that whereas
causal claims are frequent, albeit often ambiguously stated or rendered
implicit, the strategies that are used to support them often fall well short
(Sampson, Laub, and Wimer 2006, p. 70).
Although xed-effects models (and closely related hierarchical mod-
els) of within-individual change are more sophisticated than the tradi-
tional approach of regression with covariate adjustment, this analytic
approach is still limited as a method for showing a causal effect. The rea-
son is that xed-effects models cannot rule out the inuence of un-
observable time-varying confounders (Sampson, Laub, and Wimer
2006, p. 71; Bjerk 2009). King, Massoglia, and Macmillan (2007), who
advocated propensity score matching as a superior alternative, observed,
As typically used, [covariate adjustment models] give mean offending
differences between those who are married and those who are not mar-
ried, which are adjusted for any number of control variables. Under very
specic conditions, such as when respondents are randomly assigned to a
particular status, the results generated from regression models may rep-
resent the actual treatment effect of marriage on crime. As marriage is
426 Torbjørn Skardhamar et al.
not a randomly occurring event, we use propensity score matching esti-
mators to assess the treatment effect of marriage on adult offending
(p. 43). In this section, we critically examine the evidence in support of
the causal interpretation of the marriage-crime association. First, we
discuss two inuential studies aiming at direct causal estimates. Next,
we discuss two kinds of evidence that speak to the causal interpretation,
albeit more indirectly: evidence on the timing of change and evidence
on proposed theoretical mechanisms.
A. Evidence of Counterfactual Causal Effects
A counterfactual causal effect is dened as the difference in the out-
come under treatment condition Y
and the outcome without treatment
. Under random assignment, any pretreatment characteristics that
vary across the individuals, X, are unrelated to the treatment, D. Thus,
if we could study the marriage effect experimentally, those in the con-
trol group would allow us to observe the counterfactual, that is, what
would have happened to the treatment group had they not been exposed
to marriage. Formally, the average treatment effect can be expressed as
2 Y
) p E(Y jD p 1, X) 2 E(Y jD p 0, X ).
For events that are not randomly allocated, D depends on X, and
therefore, any difference between the two groups reects systematic dif-
ferences in who receives the treatment as opposed to the treatment
effect. This is a formal denition of selection effect.
Propensity score matching (henceforth PSM) has sometimes been
described as a method that overcomes this problem in observational
(nonexperimental) data. There are two particularly inuential publica-
tions in the marriage-crime literature that rely on this approach, and both
make causal claims. King, Massoglia, and Macmillan (2007) employ PSM
as the main analytical technique. Sampson, Laub, and Wimer (2006) ex-
ploit a more advanced method of inverse probability of treatment weight-
ing (IPTW), which involves the use of propensity scoring as an element
of HLM of within-person change.
King, Massoglia, and Macmillan (2007) use a nearest-neighbor PSM
to mimic an experimental design. This procedure involves estimating
the probability of marrying to construct a comparison group of non-
married people by matching each married person to a nonmarried person.
Does Marriage Reduce Crime? 427
Exact matching on a large number of variables is impossible in practice,
but as shown by Rosenbaum and Rubin (1985), matching on the propen-
sity score can achieve balance on all covariates. The propensity score is the
probability of treatment, D. This quantity has to be estimated, typically
with a logistic regression model predicting the treatment outcome, such
as marriage. Each married person is then matched with a nonmarried per-
son with a similar propensity score. Using this approach, a quasi control
group is created, and the treatment effect can be estimated by calculating
the difference in the observed outcomes between the treatment and the
quasi control groups, Y
2 Y
The goal of PSM is to achieve balance on all covariates, so that all
variables included in X are identically distributed in the treatment and
the control groups. In other words, the goal is to make the two groups
equal on all pretreatment characteristics. It is critical to understand that
matching on the set of observable characteristics X does not necessarily
provide balance on unobserved characteristics. Thus, not unlike tradi-
tional regression analysis, propensity score methods yield causal effect
estimates insofar as all the relevant variables are observed, measured,
and included in the matching modeland assuming these variables
are balanced after matching.
Thus, the problem of omitted variable
bias is omnipresent. We have a truly causal estimate only if we assume
the absence of omitted variable biasan assumption that can never be
fully evaluated. As Heckman and Pinto (2013, p. 10) put it, The liter-
ature on matching . . . solves the problem of confounders by assumption.
It postulates that a set of observed pre-program variables, say X, spans
the space generated by unobserved variables V although it offers no
guidance on how to select this set (our emphasis). Although there are
several kinds of matching methods, they all rest on this basic assumption
for making causal inferences. It should be noted that matching methods
have useful justications beyond the estimation of causal effects (Morgan
and Winship 2007, p. 142).
The balance criterion must be further checked as balance is a necessary (but not suf-
cient) condition for causal interpretation. While balance ideally is assessed by comparing
the joint empirical distribution of all covariates X between the matched and control
groups, this becomes infeasible with high-dimensional X (Imai, King, and Stuart 2008,
p. 495). In practice, many studies do what King, Massoglia, and Macmillan (2007) did,
comparing the average values of each covariate, reporting statistics such as t-tests and stan-
dardized bias. This is less than thorough. Other balance checks that compare the distribu-
tion of each variable should be considered (see, e.g., Diamond and Sekhon 2013).
428 Torbjørn Skardhamar et al.
Sampson, Laub, and Wimer (2006) used the related IPTW approach
based on propensity scores. The basic procedure is as follows. First, the
probability of being married is estimated for each time point in the lon-
gitudinal data set. Using this propensity score, a weight variable is con-
structed for each individual at each point in time, which represents the
inverse of the probability of being in the state (married or unmarried)
that the individual is in at that point in time (for details, see Sampson,
Laub, and Wimer [2006, p. 73]). In the next stage, the observations are
weighted by the weight variable in subsequent regression models.
This approach does not yield unbiased and consistent estimates of
causal effects unless important assumptions are met, such as the assump-
tion of sequential ignorability, which states that the treatment assign-
ment of unit i at time j is exogenous given treatment and covariate his-
tory of the same unit up to that point in time (Imai and Ratkovic,
forthcoming, p. 2). In other words, it assumes no time-varying unmea-
sured confounders. If time-varying confounders are missing from the
propensity score model, the results may be biased.
The problem of confounders is well understood in the literature and
acknowledged by both Sampson, Laub, and Wimer (2006) and King,
Massoglia, and Macmillan (2007). Indeed, the key argument by Sampson,
Laub, and Wimer against omitted variable bias is precisely that they have
included a large number of relevant variables in the propensity score
model, so that any effect of an omitted variable would be implausibly
large (p. 499). However, increasing the number of covariates is hardly
a persuasive approach to ruling out potentially important confounders,
as it is unlikely that one can adequately measure all such confounders
These studies fail to demonstrate a causal effect of marriage on crime
as dened by prevailing methodological standards. Given the nature of
the research questionthe effect of marriage on crimethe absence of
studies demonstrating counterfactual causality is not surprising. While
it is possible to experiment with some life course transitionssuch as
employment (see, e.g., Uggen 2000; Bushway and Apel 2012)a ran-
domized study of the effect of marriage on crime is implausible. Neither
are methodological substitutes such as instrumental variables likely to
work very well, as Sampson, Laub, and Wimer (2006) have noted: In
practice, however, nding plausible instruments that can pass the neces-
sary identifying restrictions has proven very difcult. . . . We are not
aware of any plausible instrument for identifying causal effect of mar-
riage on crime in the literature to date (p. 473).
Does Marriage Reduce Crime? 429
We recognize that the criticism we have laid against these sophisti-
cated studies may seem unreasonable to scholars accustomed to evalu-
ating causal hypotheses with observational data. If it is implausible to
conduct randomized experiments and unrealistic to identify valid instru-
mental variables, why should we not pursue the next-best approach? We
believe that research interested in the real thing should not settle for an
imitation. If the true interest is the causal effect of marriage on crime,
these methodological realities should not be ignored. We should not re-
lax our criteria of causality simply because our data do not allow demon-
stration of causal effects.
It is possible to learn about causal mechanisms without estimating causal
effects (Berk et al. 2015), but we should avoid overstating our conclu-
sions. Instead, we should be clear on what we can and what we cannot
learn from the data. We believe that one source of confusion about causal
effects in the social sciences is the pervasive use of words such as ef-
fect and impact when these are not warranted by the data and methods
used. In most cases, authors are quite aware of the limitations but use
these terms as loose alternatives to coefcient and association. By
being more precise and explicit, one could avoid some of this confusion.
B. Time Order: Does Desistance Happen Before or After Marriage?
On a more optimistic note, a direct test of causality is not the only
way to evaluate causal claims empirically. According to one classic def-
inition, causality is a function of three components: the presence of an
association, the absence of spuriousness (confounders), and correct time
order. Much of the discussion thus far has been concerned with evidence
of the (negative) association between marriage and crime and methods
of eliminating spuriousness (selection effect). We now turn to the third
component, namely, the timing issue, which has been largely ignored in
this literature. In order for marriage to qualify as a causal factor in desis-
tance, reductions in offending should occur after marriage, not before.
If desistance occurs before marriage, marriage should be understood ei-
ther as a consequence of desistance or as a downstream effect of what-
ever is causing desistance, such as cognitive transformation or maturation
(see g. 2). The majority of studies compare average offending rates in
each state, but this approach does not capture changes occurring within
each state.
A widely cited study by Laub, Nagin, and Sampson (1998) addressed
the timing issue with the Glueck data up to age 32. In the rst analytic
430 Torbjørn Skardhamar et al.
step, they estimated group-based trajectories (Nagin and Land 1993) for
the entire observational period from age 7 to 32 and then assigned each
person to a group on the basis of posterior probabilities. In the second
stage, they regressed the number of arrests in each 2-year interval from
age 17 to 32 on changes in marital status and quality, controlling for
group membership and other characteristics. In addition, they conducted
separate regression analyses by trajectory group membership. The
results showed that, among those whose unions progressed into good
marriages (measured at the last wave), arrest rates were stable during
the courtship period (the two periods preceding marriage), followed by
a gradual decline in offending during the state of marriage. They found
no evidence of a courtship effect and thus concluded that good marriages
reduce crime gradually after marriage. We are somewhat hesitant to
endorse this conclusion for methodological reasons. Because the tra-
jectories were estimated over the entire periodincluding the marital
periodcontrolling for group membership implies controlling for post-
marriage offending outcomes as well. We expect this aspect of the analytic
strategy to bias the results, but further efforts are needed to assess the
substantive implications of this methodological approach.
Duncan, Wilkerson, and England (2006) used a ne methodological
approach to examine the timing of change in substance misuse with
respect to marriage. Focusing on only those who got married, they used
data covering 5 years before and 5 years after the year of marriage, thus
creating a panel structure with 11 observations per person. Using a
piecewise linear spline, they separated the changes in the premarriage
period, around the time of marriage, and the period following the tran-
sition to marriage. They found a pattern that resembles the hook for
change trajectory described in gure 2: a slight declining trend during
the premarriage period followed by accelerated decline during the period
of marriage.
While Duncan, Wilkerson, and England (2006) used piecewise splines,
Lyngstad and Skardhamar (2013) used a similar approach to estimate
parameters for single-year dummy variables one for each time point.
The main advantage of this approach is that it avoids imposing paramet-
ric restrictions (e.g., linearity) on the curve, permitting a more accurate
analysis of the timing of change in offending relative to the timing of mar-
riage. (As a potential downside for most data sets, this method requires a
large number of data points.) The results showed that the decline in
offending began long before the transition to marriage, and the period
Does Marriage Reduce Crime? 431
after marriage wa s associated with an increase (a rebound)inof-
fending r ather than further declines. This pattern contradicts the ex-
pected marriage effect. Lyngstad and Skardhamar acknowledge that
romantic bonding, including cohabitation, preceding the marital transi-
tion may be responsible for some of this pattern.
Finally, there are two Dutch studies that provide information about
the timing of change as a by-product of the main analysis. McGloin
et al. (2011) employed a set of dummies representing each year before
and after marriage, with nonmarried as the reference category. The
marriage coefcients changed from 2.01 2 years before marriage to
2.02 at the time of marriage and the year after and then back to 2.01
the second year after marriage. Each of these effects is signicant with
respect to the reference category, but the authors do not report whether
these coefcients were different from each other, that is, if the changes
in the marriage effects were signicant. The study by Beijers, Bijleveld,
and van Poppel (2012), which found evidence of historical contingency
in the marriage effect, includes a descriptive trend chart showing levels
of criminal offending in the 10 years preceding and following the year of
marriage (g. 3, p. 435). In this study the negative association between
marriage and crime is limited to those born after 1970; there was no
signicant association among the older cohorts. Focusing on those born
after 1970, it is clear that offending started to decline sharply about 3 years
before the year of marriage. The overall level of crime continues to drop
following marriage, but at a slower rate with occasional periods of increase.
In conclusion, the evidence on timing of desistance relative to marriage
is addressed by only a handful of publications. Although the results are
mixed, and to some extent difcult to compare because of methodologi-
cal variation, the evidence suggests that marriage is more likely to occur
as a consequence of desistance in northern Europe. The results from
American studies are mixed. The most certain conclusion one can draw
is that there is no clear evidence showing time order consistent with the
turning point hypothesis as described in gure 2. The available evidence
is more consistent with the maturation or hook for change perspectives.
C. Evidence on Causal Mechanisms
Theories of the marriage effect assume specic causal mechanisms,
such as informal social control, to produce the effect. In this section,
we evaluate the empirical evidence pertaining to the mechanisms or pro-
cesses suggested by theories of the marriage effect.
432 Torbjørn Skardhamar et al.
Laub and Sampsons account involves a number of theoretical mech-
anisms. The primary process of inuence is social control through stakes
in conformity. Attachment to the spouse is expected to restrain crime-
prone individuals from offending. In addition, romantic partners have
the capacity to serve as effective monitors and guardians. In relation
to these processes, marriage is expected to change routine activities in
ways that increase time spent in prosocial settings and reduce time spent
with antisocial peers. These environmental effects are expected to facilitate
psychological change in self-concept to a point at which personal identity
becomes inconsistent with criminal behavior. Finally, Laub and Sampson
(2003, p. 137) observe that some of the men in their sample gained con-
crete economic benets by marrying women with some wealth and income
or women from families that were able to offer nancial security.
In this section we discuss direct empirical evidence pertaining to these
mechanisms. We focus on research that has attempted to measure one or
more of the processes explicated in the theoretical model.
1. Marital Quality and Spousal Attachment. Sampson and Laubs key
argument is that marriage assists in desistance because of the social
bonds it creates. Accordingly, we should expect the marriage effect to
materialize in the presence of strong attachment. Despite this clear as-
sumption, many studies ignore the quality of marriage, probably be-
cause the data do not permit appropriate measures. We found 17 pub-
lications that feature direct measures of attachment to a spouse or a
romantic partner. The operationalization of this concept varies from
psychological indicators of attachment to the mere length of marriage.
The most prominent support for the bonding process is documented
in research testing Sampson and Laubs age-graded theory of informal
social control (Sampson and Laub 1990, 1993; Laub, Nagin, and
Sampson 1998; Laub and Sampson 2003). All these studies nd evidence
that marital attachment is negatively related to criminal offending. This
pattern has been replicated in other studies (Massoglia and Uggen 2007;
Capaldi, Kim, and Owen 2008; Herrera, Wiersma, and Cleveland 2011).
Some studies have found mixed support for the hypothesized associ-
ation. For example, marital attachment may be important for some types of
offending or only for males (or females) (Simons et al. 2002; Oudekerk,
Burgers, and Reppucci 2014). Other studies have failed to nd any evidence
in support of the attachment/bonding process (Giordano, Cernkovich, and
Rudolph 2002; Schroeder, Giordano, and Cernkovich 2007; Simons and
Barr 2012). It has been observed that if the romantic partner is antisocial,
Does Marriage Reduce Crime? 433
that is, does not disapprove of a criminal lifestyle, attachment, however
strong, is unlikely to assist in desistance (Giordano, Cernkovich, and
Holland 2003). We would add that a measure of relationship quality (sta-
bility, warmth, attachment) may be related to the individual charac-
teristics of the partner. For example, unions with an antisocial partner
should be more stormy than unions with a prosocial partner. It is pos-
sible that the evidence consistent with the marital quality hypothesis
reects the normative orientation of the romantic partner.
A number of studies nd marriage to be associated with desistance
without paying attention to the quality of the union. Among the 41 stud-
ies with no measures of marriage quality, 31 found marriage to be asso-
ciated with desistance for at least one outcome measure. In sum, there is
some evidence that good marriages are associated with reduced offend-
ing, but there is also ample evidence from studies that ignore the qualitative
aspects of the relationships. Perhaps most marriages are so good (or good
enough) that the average association tends to be statistically signicant.
2. Direct Social Control by Spouse. Although the idea that a prosocial
spouse may act as a guardian of crime-prone adults is frequently discussed,
we were unable to nd a single study testing this hypothesis using direct
measures of relevant behavior. One way spouses may exert social control
is by limiting time spent with peers and by modifying routine activities.
However, as these are not direct measures of spousal control, we discuss
evidence on routines and peer association in a separate section below.
The idea of spousal control is based on an assumption that the spouse
disapproves of antisocial behavior. Owing to assortative mating by anti-
social behavior (Krueger et al. 1998), this assumption is questionable.
If crime-prone men are likely to marry antisocial women, the processes
of social control are less likely to emerge (Knight 2011; van Schellen,
Poortman, and Nieuwbeerta 2012). As a special case of the peer effect,
antisocial partners may even encourage or at least enable criminal behav-
ior. A broader review of research on romantic relationships and problem
behaviors suggests that the nature of the inuence depends on the char-
acteristics of the romantic partner (Rhule-Louie and McMahon 2007, p. 91).
We found 13 studies attending to the criminality of the partner. The
majority support the assumption that having an antisocial partner pro-
motes rather than curbs antisocial behavior (Simons et al. 2002; Schroeder,
Giordano, and Cernkovich 2007; Herrera, Wiersma, and Cleveland 2011;
Simons and Barr 2012; Monahan, Dmitrieva, and Cauffman 2014; Oude-
kerk, Burgers, and Reppucci 2014). Maume, Ousey, and Beaver (2005)
434 Torbjørn Skardhamar et al.
and Sampson, Laub, and Wimer (2006) found that the restraining effect
of marriage was attenuated but remained statistically signicant, control-
ling for spousal criminality. Farrington and West (1995, p. 275) found
that reductions in crime were similar among men who married women
with and without a criminal record, noting the caveat that there were very
few convicted wives. Although individuals who marry someone with a
criminal record tend to be more criminal themselves, Skardhamar, Mons-
bakken, and Lyngstad (2014) found more substantial declines in offending
among those individuals compared with those who married a noncrimi-
nal. Leverentz (2006) observed that deviant couples may provide support
for each other if both are oriented toward desistance.
3. Acquiring Social Capital through Spouse and Extended Family. Laub
and Sampson (2003) report that the partnership helped the men in their
sample in organizing and managing their affairs as adults and that
marrying a socioeconomically stable partner generated many concrete
benets such as housing, employment, and other material goods (p. 137).
Outside the qualitative observations from the Glueck men, we did not nd
any other studies that systematically investigated these processes.
4. Routine Activities and Peer Associations. We found six studies focus-
ing on peer associations and routine activities as potential mediators. In a
seminal article, Warr (1998) presented differential association mecha-
nisms as an alternative to the social bonding account of the marriage
effect. He found support for the hypothesis that changes in time spent
with peers were responsible for reduced offending associated with mar-
riage. Similarly, Simons et al. (2002) found that having a prosocial roman-
tic partner was unrelated to adult crime once associating with deviant
peers was