You are on page 1of 36

CAN SELF-CONTROL CHANGE

SUBSTANTIALLY OVER TIME? RETHINKING


THE RELATIONSHIP BETWEEN SELF- AND
SOCIAL CONTROL∗

CHONGMIN NA
School of Human Sciences and Humanities
University of Houston–Clear Lake

RAYMOND PATERNOSTER
Department of Criminology and Criminal Justice
University of Maryland

KEYWORDS: self-control, social control/bond, stability and change, self-


selection, social causation

The primary goals of this study were to test the long-term stability
thesis of Gottfredson and Hirschi’s (1990) general theory of crime and
to examine the relationship between self-control and social control over
time. The data come from a field experiment where the “treatment” con-
sisted of an intentional effort to improve the childrearing behaviors of a
sample of caregivers whose children were at high risk of criminal behav-
ior. Caregivers in the control condition were given no such training. The
intervention occurred when all subjects were in the first grade (mean age:
6.2 years old), and we have measurements on self-control and the social
control/bond for each subject from grades 6 to 11 (mean ages: 12 to

Additional supporting information can be found in the listing for this article
in the Wiley Online Library at http://onlinelibrary.wiley.com/doi/10.1111/crim.
2012.50.issue-2/issuetoc.
* An earlier draft of this article was presented at the annual meeting of the
American Society of Criminology, San Francisco, CA, November 2010. We are
grateful to the Johns Hopkins Prevention Intervention Research Center for pro-
viding the data necessary to undertake this study. We also wish to thank Denise
Gottfredson, Gary Gottfredson, Jeffery Harring, Thomas Loughran, and Terence
Thornberry for their helpful comments and suggestions on previous drafts of this
article. Please direct correspondence to Chongmin Na, School of Human Sciences
and Humanities, University of Houston–Clear Lake, 2700 Bay Area Blvd., Hous-
ton, TX 77058 (e-mail: nachon@uhcl.edu).


C 2012 American Society of Criminology doi: 10.1111/j.1745-9125.2011.00269.x
CRIMINOLOGY Volume 50 Number 2 2012 427
428 NA & PATERNOSTER

17 years old). Both a hierarchical linear model and a second-order latent


growth model identified meaningful differences in the growth pattern of
self-control among individuals in the pooled sample and a difference
in the growth parameters for self-control and the social control/bond
over time between the treatment and control groups. Both findings are
inconsistent with Gottfredson and Hirschi’s stability of self-control hy-
pothesis. The same patterns persisted when different analytic techniques
and model specifications were applied, which suggests that the results are
not an artifact of measurement error, model specification, or statistical
methods. Structural equation modeling using the panel design of the data
was better able to disentangle the long-term relationship between self-
and social control—a relationship that was found to be more dynamic
than previously hypothesized.

For the past decade or so, criminologists have begun to address explicitly
the question of what drives stability and change in an individual’s pat-
tern of criminal offending. Most theoretical perspectives have focused on
either time-stable individual characteristics (such Gottfredson and Hirschi’s
[1990] general theory of crime) or the role played by social institutions
and structural environments (such as Sampson and Laub’s [1993] age-
graded informal social control theory). Although some recent theoretical
perspectives have emphasized the dynamic interaction between individual
and social structural factors in promoting persistent offending or desistence
from a criminal career (e.g., Le Blanc, 2006; Wikström, 2006; Wikström
and Treiber, 2007, 2009; Wikström et al., 2010), for the most part, the
mutual role played by internal and external factors in criminal offending has
been neglected and unexplored.1 As an example, although both can trace
their origins to a common source (Hirschi’s [1969] Causes of Delinquency),
recent expressions of social control theory in Sampson and Laub’s (S&L)
theory of age-graded informal social control and Gottfredson and Hirschi’s
(G&H) self-control theory have been portrayed as competing theoretical
perspectives. In fact, some scholars have even argued that there is an
inherent contradiction between self- and social control theory sufficient to
preclude the possibility that both types of control may mutually affect crime
or each other (Akers, 1991; Taylor, 2001).

1. An exception is the important work of Wikström and colleagues whose situational


action theory is a comprehensive general theory that explains crime with reference
to the interaction among structural, personal/individual, and situational factors.
Their work is a much more general multilevel theory of individual offending than
the more modest integration of social control and self-control theory described in
this article.
CAN SELF-CONTROL CHANGE? 429

This ongoing controversy between self- and social control theories well
exemplifies current criminological thinking on the role of internal and
external factors in crime. G&H’s (Gottfredson, 2006; Gottfredson and
Hirschi, 1990; Hirschi, 2004) theorizing on continuity and change in of-
fending over the life course does contrast sharply on one important point
with Hirschi’s (1969) earlier social control theory and with the more recent
age-graded version.2 This point pertains to variation in the social bond
over time and to the relationship of the social bond early in life to of-
fending in later life stages. In Causes of Delinquency (1969) and subse-
quent writings up to 1980, Hirschi was clear that variations in offending
over the life course were directly related to variations in the strength of
the social bond (Hirschi, 1969; Hirschi and Rudisill, 1976: 21 emphasis
added):

Unlike more purely psychological theories, it [social control theory]


emphasizes the significance of variation in group membership across
the life cycle. With this latter emphasis, the theory can attempt to
account for one of the major facts of crime ignored by other theories:
the considerable variation in criminal behavior over the life of the
“offender.”

This position is virtually identical to the age-graded version of social


control theory later developed by S&L (Laub and Sampson, 2003; Laub,
Sampson, and Sweeten, 2006; Sampson and Laub, 1993). Both early social
control theory and the newer, age-graded version posit that variations in
the strength of the social bond are related to variations in offending over
the life course, and that the relationship between the social bond and
offending over time is symmetrical—a reshuffling of the good and the bad is
possible as those who previously were successfully socialized can have this
undone and those who previously were unsuccessful in being socialized can
dramatically improve.
It is the possibility of the reshuffling of the good and the bad that Hirschi
abandons in his 1990 and later work with Michael Gottfredson. In their self-
control theory, there is absolute but not relative change in self-control in

2. Gottfredson (2006: 78) in his redefinition and reoperationalization of self-control


emphasized that “self-control theory is not an implicit rejection of the earlier
social control theory upon which it is founded.” He further argued that they have
become reconciled by their empirical treatment because “they are very difficult to
discriminate empirically and, under some circumstances, may amount to the same
thing” (2006: 86). We still believe that G&H’s (1990) original conceptualization of
self-control is distinct even if not a departure from Hirschi’s (1969) original control
perspective for the reasons that follow.
430 NA & PATERNOSTER

individuals over time. An allowance is made for an absolute change in self-


control because “socialization continues to occur throughout life” (1990:
107), but not relative change because “desocialization is rare” (1990: 107).
The implication is that the good are not likely to become bad. Furthermore,
because “the traits composing low self-control are also not conducive to the
achievement of long-term individual goals . . . they impede educational and
occupational achievement, destroy interpersonal relations, and undermine
physical health and economic well-being” (1990: 96), the bad are not likely
to become good.3 G&H (1990: 107) also claimed that “differences between
people in the likelihood that they will commit criminal acts persist over
time,” which again suggests that relative rankings of self-control between
individuals remain stable over time, and the dynamic social factors over the
life course that are central to earlier social control and current age-graded
social control theory can have only trivial effects on the propensity to offend
(Hirschi and Gottfredson, 1995).
The issue of stability and change also is deeply embedded in ongo-
ing debates about contemporary criminological theories (e.g., population
heterogeneity vs. state dependence or self-selection vs. social causation)
and questions about the effectiveness of prevention/intervention efforts
that explicitly target change in self-control after the formative period of
early childhood. In sum, although there is some important common ground
between the theories of G&H and S&L, there are two critical points
of divergence. First, G&H implied that there should be a great deal of
both short-term and long-term relative stability in an individual’s level of
self-control—a more pure population heterogeneity theory. In contrast,
S&L’s more dynamic state dependent view would be friendly to the no-
tion that self-control varies over time and varies in response to fluctua-
tions in the social bond (see the important paper by Hay and Forrest,
2006).
The latter point brings us to the second point of convergence. G&H
were fairly clear that although a strong social bond is positively related to
the creation of self-control, this relationship should greatly dissipate and

3. At various points, G&H go to great pains in their 1990 book to argue against
any reshuffling of the good and bad. They argue (1990: 118) that “[o]ur stability
postulate asserts the people with high self-control are less likely under all circum-
stances throughout life to commit crime. Our stability notion denies the ability of
institutions to undo previously successful efforts at socialization”; “socialization
is a task that once successfully accomplished, seems to be largely irreversible”
(1990: 107); and “[o]ur theory clearly argues, however, that it is easier to develop
self-control among people lacking it than to undermine or destroy self-control
among those possessing it” (1990: 106, note 3). Although the bad may become
“better,” their socialization and self-restraint will never surpass those whose initial
experiences with socialization were more successful.
CAN SELF-CONTROL CHANGE? 431

perhaps even disappear after 10 years of age or so.4 A corollary implication


of G&H’s view is that self-control is malleable before 10 years of age when
there may be both absolute and relative change; after that age, however,
the window closes and one’s relative position on self-control is fixed.5
S&L, however, would adopt the view expressed by Hirschi in 1969 that
self-control is something that is much more malleable for a longer period
of time, that the social bond is an important source of socialization that
strengthens self-control throughout the life course, and that there should
be a reciprocal relationship between the two over time.
If one important dimension of research is deducing as many of the em-
pirical implications of theory as possible and subjecting them to test, these
divergent theoretical implications of G&H and S&L should be subject to
empirical examination. The current study contributes to this task in several
ways. First, it provides a rigorous test of the long-term stability of self-
control and examines the extent to which an early preventive intervention
(a randomized field experiment deliberately designed to increase caregiver
effectiveness) can alter trajectories of both self- and social control over the
short and long term. Second, by employing structural equation modeling
(SEM) with a longitudinal panel design that explicitly tests the causal
directionality of both theoretical constructs over time, we hope to disen-
tangle the causal mechanisms underlying stability and change in self- and
social control. Following Gottfredson’s (2006) recent position, we would
expect that as socialization improves self-control, the greater attentiveness
to social relationships characteristic of those with self-control would in turn
positively affect the social bond. Finally, the models provide an empirical
base for theoretical modification and integration under a unified control
perspective (a general control perspective that combines social and self-
control or internal and external sources of restraint; see Longshore, Chang,
and Messina, 2005; Nagin and Paternoster, 1994; Wikström, 2006; Wikström
and Treiber, 2007, 2009). Before discussing the data and methods, extant
theories and research are briefly summarized in the following section.

4. Scholars have traditionally held that self-control is relatively fixed by 10 years of


age. It is conceivable that the window of socialization closes well before that age,
however, because G&H (1990: 229) strongly hinted that the development of self-
control occurs as a result of caregiver activity that largely takes place before a child
enters school: “In our view, research interested in the impact of family factors on
self-control would attempt to determine what the family was like when differences
in self-control were established. The crucial period, research indicates is prior to
the elementary school grades, since important differences are usually present by
that time.”
5. Although “socialization continues to occur throughout life” (Gottfredson and
Hirschi, 1990: 107), the rate at which this occurs is the same for everyone, fixing
one’s relative position.
432 NA & PATERNOSTER

THE RELATIONSHIP BETWEEN SELF- AND SOCIAL


CONTROL: INCOMPATIBLE THEORETICAL
PERSPECTIVES?
Criminologists have taken two distinctive and largely separate ap-
proaches in explaining the source of stability and change in offending over
time (see Nagin and Paternoster, 1991, 2000 for more details). Theories
of population heterogeneity emphasize enduring individual characteristics
that emerge early in life and predispose the probability of criminal engage-
ment throughout the life course. Although such theories do not preclude
the possibility of change, they claim that any change observed does not re-
sult from life events and experiences that are exogenous to the individual’s
propensity to offend, but from other factors that cannot be explained by
existing criminological theories (e.g., aging, maturation). Theories of state
dependence, however, emphasize life circumstances and events that are
external and temporally proximate to individuals. Although such theories
do not discount the role that time-stable individual characteristics play, they
assert that observed stability in offending reflects the fact that committing
a crime alters the probability of subsequent criminal engagement by trans-
forming the offender’s external and proximate factors. Similarly, positive
(and negative) change can occur through the same process when persons
strengthen (or weaken) conventional relationships and make conventional
commitments.
In their largely population heterogeneity theory,6 G&H (1990) implied
that self-control is relatively stable, that socialization should diminish after
approximately 10 years of age, and that intentional interventions directed
at improving self-control also should diminish in effectiveness after 10 years
of age. In sharp contrast, S&L (Laub and Sampson, 2003; Sampson and
Laub, 1993) weighed more heavily on the state dependence account in
their theory of stability and change in offending patterns over time (Samp-
son and Laub, 1997).7 This approach is especially true when they argue
that salient life events and social experiences in adulthood can counteract
the negative consequences of early failures in socialization independent
of prior individual differences in criminal propensity, that negative life
events can undo previously successful socialization, and that the process

6. The general theory of crime describes a causal process that is consistent with state
dependence before approximately 10 years of age, but one more consistent with
population heterogeneity after that.
7. S&L acknowledge, however, that in addition to later life events, prior in-
dividual differences in the propensity to offend are also important factors
in the explanation of crime. So like G&H (1990) theory, theirs is a mixed
model.
CAN SELF-CONTROL CHANGE? 433

of change is dynamic, occurring throughout the life course. In this dynamic


conceptualization, the relationship between social control and self-control
would not be expected to diminish after 10 years of age, and intervention
efforts to improve self-control should be effective at any point in the life
course.
Although some scholars view social and self-control as competing, even
contradictory, theoretical perspectives (Akers, 1991; Taylor, 2001), another
possibility is that they are distinct but nonetheless closely related to each
other. The view we adopt is that a strong social bond continues to in-
fluence the development of self-control well into adolescence and that
social and self-control mutually influence each other over time. Our posi-
tion more closely resembles that of S&L than either G&H or Akers and
Taylor.

PRIOR RESEARCH ON THE STABILITY AND MALLEABILITY


OF SELF-CONTROL
The Stability Thesis
Several studies have examined the stability of self-control over the
short term. For example, among a sample of college students, Arneklev,
Cochran, and Gainey (1998) found that self-control was considerably sta-
ble (a stability correlation of .82) over an academic semester. Although
supportive of G&H’s (1990) stability hypothesis, we should not be too
surprised because the stability of self-control was measured only over 4
months. Employing both an attitudinal and a behavioral measure of self-
control from a national probability sample (the NLSY79), Turner and
Piquero (2002) examined the stability of self-control within a subsample
of youth between 15 and 19 years of age. They found only moderately
stable test–retest correlations (ranging from .33 to .68) during this 4-year
period of adolescence when self-control is theoretically expected to have
been established. Within a sample of incarcerated adults, Mitchell and
MacKenzie (2006) examined the relative stability of self-control over just
a 6-month period. They found only modest Spearman rank-order corre-
lation coefficients between the earlier and later measures of self-control,
ranging in magnitude from .27 to .48. Given such a short time interval,
one would have expected higher relative stability coefficients, particularly
within a sample whose average age was 23 years. In an analysis of quar-
tile rank scores, Mitchell and MacKenzie also found that of those in the
lowest quartile in self-control at the beginning of the 6-month period, only
51 percent remained in the lowest quartile at 6 months. Burt, Simons, and
Simons (2006) examined the stability of self-control within a sample of
African American children and contrary to the stability hypothesis found
434 NA & PATERNOSTER

what they characterized as substantial movement across quartiles of self-


control over only a 2-year period. Beaver and Wright (2007) estimated the
stability of self-control for a sample of children from kindergarten to first
grade and found a great deal of stability even during a time period (before
10 years of age) when self-control is still developing.
Although insightful, all but one of these studies (Turner and Piquero,
2002) have examined the stability of self-control over the relatively short
term (a college semester, 6 months, or 1 or 2 years). As any reading of the
general theory of crime will reveal, G&H (1990: 91) clearly argued that self-
control also would be relatively stable over the long term: “[These] differ-
ences [in self-control] observed at 8 to 10 years of age tend to persist from
then on.” Fortunately, a few studies have examined the stability hypothesis
with a longer time frame. In one of the most rigorous of these studies,
Hay and Forrest (2006) examined the stability of self-control with several
different analytic strategies within a national sample of youth (the NLSY79)
from 7 to 15 years of age. First, they found that the test–retest correlation
between self-control at 7 and 15 years of age was only moderately large
(r = .43).8 Hay and Forrest also estimated semiparametric group trajecto-
ries of self-control between 7 and 15 years of age, and in this analysis, there
was considerable absolute and relative stability in more than 80 percent
of the sample. Consistent with G&H’s hypothesis, only a relatively small
percentage of the youths experienced a substantial amount of instability
in self-control. Finally, Higgins et al. (2009) also estimated a group-based
trajectory model from a sample of youth from 22 schools in six cities in the
United States. They tracked the students for a 4-year period from 12 to
16 years of age, a period when self-control is theoretically already formed
and should be stable. The stability coefficient between self-control mea-
sured at 12 years of age and again at 16 years of age was modest
(r = .48), but as in Hay and Forrest’s (2006) earlier study, Higgins et al.
found substantial absolute and relative stability in the trajectories of self-
control over this time period.
It is difficult to draw a definitive conclusion from the extant literature
about the stability hypothesis of G&H’s (1990) general theory. Although
there is some evidence that self-control is relatively stable over the short
term, there are inconsistent findings. Ironically, somewhat more evidence

8. It was, however, much more stable over the short term. The correlation between
self-control at 7 and 9 years of age was .64. Although this stability coefficient is not
large, keep in mind that 7–9 years of age are within the window when self-control
is still forming and is more dynamic. In opposition to this, however, the correlation
between self-control at 13 and 15 years of age was only .65; this during a time when
the window of development is presumably closed and greater stability should be
observed.
CAN SELF-CONTROL CHANGE? 435

for relative (and absolute) stability is found in studies that have used a
longer time reference, and when group-based trajectory models have been
estimated. Less consistent evidence is found in studies with a short time
period and in those that employ some type of correlation coefficient. We
would finally note that only one study to date has used an appropriately long
time period—Hay and Forrest (2006), who examined the stability of self-
control for 8 years (7–15 years of age). To test G&H’s stability hypothesis,
it is important that other long-term studies be undertaken. In the study
reported herein, we follow Hay and Forrest’s example and examine the
stability of self-control over a 5-year period—when the subjects were in
the 6th grade (mean age of 12 years) until the 11th grade (mean age of
17 years).9 We trace the self-control of subjects during a period when self-
control is theoretically expected to be relatively stable.10

The Malleability Thesis


The malleability of self-control is inextricably related to its stability.
According to G&H’s (1990) general theory, self-control is malleable before
approximately 10 years of age, after which the window on its development
in any substantial sense is closed. They argued that institutions of socializa-
tion such as caregivers (primarily) and schools (secondarily) will have little
impact on the development of self-control after 10 years of age.11 This lack
of impact is because it is difficult for those poorly socialized early in life to

9. We speak of grades in school that our subjects were in rather than of years of age.
Our subjects mean age in the first grade was 6 years old, and assuming that they
are “on-grade,” they will be one year older at each subsequent grade. Because
most of our analyses are done when the students are in the 6th–11th grade, this
covers a mean age of approximately 12–17 years.
10. To distinguish our study more clearly from theirs, Hay and Forrest (2006) exam-
ined the stability of self-control from 7 to 15 years of age so that approximately
one third of their time period captures ages when self-control is still presumed to
be in its formative stages and would be expected to be less stable. Our time period,
from the 6th to the 11th grade, covers a period after which self-control is presumed
to have been developed. Furthermore, we employ a latent growth model of self-
control over time rather than semiparametric group-based modeling.
11. Two possible, and we think equally defensible, positions are available with respect
to G&H’s view on the matter of the effect of caregivers, schools, and other
institutions in creating self-control in mid-adolescence and beyond. A “hard”
position would be that those who are not effectively socialized by the primary
caregivers are unlikely to develop self-control later in life because they either
fail to perceive these opportunities or misplay them when they arrive. A “soft”
position is that although primary caregivers (G&H, 1990: 97) are the “major
cause” of self-control, it is still possible that other institutions can make up for
their failure. They noted that (1990: 105) “[t]hose not socialized sufficiently by the
family may eventually learn self-control through the operation of other sanction-
ing systems or institutions. The institution given principal responsibility for this
436 NA & PATERNOSTER

improve later on—because they are initially poorly socialized, they misplay
opportunities to improve later in life. S&L (1993; Laub and Sampson, 2003)
adopted a contrary position with respect to the capacity of social control
to affect self-control or inhibitions to commit crime in adolescence and
beyond. Their more dynamic position is that there would be an ongoing
relationship between social control and self-control well beyond 10 years
of age and, more importantly, that intentional efforts (such as schools or
intervention programs) to improve self-control can be effective at any age
they are implemented.
Since their original statement of the general theory (1990), G&H have
continually expressed profound skepticism about the causal effect of insti-
tutions and interventions to improve self-control after childhood (Gottfred-
son, 2005, 2006; Hirschi and Gottfredson, 1995). In fact, one of the most
contentious debates in criminology today is whether events and experiences
later in life, such as changes in the social bond, have a causal effect on
criminal offending. We will not review that abundant evidence here, but we
will restrict our attention to the few studies that have examined the more
specific question as to whether efforts to improve self-control after 10 years
of age can be productive. The meager evidence suggests the possibility that
it may.
A partial answer to this question comes from Hay and Forrest’s (2006:
757) research. They argued that if G&H’s (1990) theory is correct, then
“parental socialization during adolescence should do little to explain dif-
ferences in self-control at that time.” They examined the relationship of
parental socialization at 11 (and 13) years of age on a youth’s level of self-
control at 13 (and 15) years of age, controlling for both prior socialization
and self-control. Consistent with a more dynamic view of both social and
self-control, parental socialization still has an effect on self-control in mid-
adolescence long after the window of effective socialization is presumed
by G&H to have closed. Although it did not address the question as to
whether other institutions or nonfamily interventions could affect self-
control in adolescence, Hay and Forrest’s conclusion clearly leaves open the
possibility that there may be a reshuffling of self-control over time (2006:
76): “Contrary to what self-control theory predicts, changes in parental so-
cialization during adolescence continue to produce changes in self-control,

task in modern society is the school.” We adopt the “hard” position here because
we believe it is more consistent with the collective landscape of their theory and
because even they (1990: 105–6) seem to have little faith in the school or other
institutions to rectify a failure of family socialization: “The evidence suggests,
however, that in contemporary American society the school has a difficult time
teaching self-control. A major reason for this limited success of the modern school
seems to stem from the lack of cooperation and support it receives from families
that have already failed in the socialization task.”
CAN SELF-CONTROL CHANGE? 437

which necessarily affects an individual’s position in the self-control distribu-


tion.” As in our current stability analysis, we build on and further expand
Hay and Forrest’s important study by examining whether a school-based
intervention implemented during the first grade can impact subsequent self-
control, and whether changes in social control generally from the 6th to
the 11th grade can influence the level of self-control over that same time
period.12

THE PRESENT STUDY


The primary goal of this study is to test G&H’s (1990) stability of self-
control hypothesis directly and to investigate further the causal mechanisms
underlying stability and change of both self- and social control using data
collected by the Johns Hopkins University Prevention Intervention Re-
search Center (JHU PIRC). This data set is uniquely suited to the purpose
of this study as it includes multiple indicators of key theoretical constructs
from the same individuals over a relatively long period of time. A key fea-
ture of the intervention in this research study was a randomized experiment
directed at improving parents’ interactions with their children and their
child management techniques. Parents of children in the treatment group
were provided training in effective child behavior management strategies
essentially designed to create better self-discipline and organization in their
children. This intervention effort can be considered an intentional effort to
socialize children and establish greater self-control.13 Parents in the control
group were provided no such training. As we will discuss, the training
was conducted in a series of workshops that was led by the first-grade

12. Turner, Piquero, and Pratt (2005) found that in addition to parental supervision,
attempts by both a youth’s neighborhood and school to socialize and improve
self-control were effective. More generally, Piquero, Jennings, and Farrington
(2010) recently published a meta-analysis of some 34 studies that examined the
effectiveness of school- or clinic-based programs designed to improve self-control
up to 10 years of age. They reported that most of the effect sizes from these studies
were positive, indicating that self-control was improved as a result of the deliberate
intervention. Moreover, the intervention also was generally effective in reducing
delinquent offending. It should be noted that this body of research does not speak
directly to the issue as to whether self-control is malleable in the long term because
they examined the effect of family, neighborhood, and school socialization on self-
control by 10 years of age. The question we address in this article is whether an
intervention can be effective in improving self-control after family socialization is
presumed to have been established.
13. There were other dimensions to the intervention as well. A classroom-based
component was provided for teachers to promote reading and math skills as well
as a program to enhance positive problem solving for their students. The family–
school partnership intervention also included a component for teachers designed
to enhance communication with parents.
438 NA & PATERNOSTER

teacher of the students and either the school psychologist or the social
worker.
The first hypothesis of the present study involves a rigorous and straight-
forward test of self-control theory’s stability hypothesis by employing hi-
erarchical linear model (HLM) and second-order latent growth model
(LGM) approaches: whether subjects exhibit significant variation in the
developmental pattern of self-control over time. G&H’s stability postu-
late implies that, although the absolute level of self-control may change
over time, the relative level of self-control (e.g., “relative ranking” or
“observed differences”) should remain stable over time (Hirschi and Got-
tfredson, 2001: 90). A significantly different rate of change in self-control
across individuals and a different rate of change between individuals
within two initially equivalent groups (the treatment and control group)
would provide concrete evidence against the strict stability postulate of
G&H’s (1990) theory because if slopes are not parallel across individuals
or groups, the relative ranking of individuals on the self-control scores
varies over time. This study goes beyond prior studies by directly address-
ing issues of measurement error and measurement invariance within the
second-order LGM framework, which has been less appreciated in past
research.
A second hypothesis involves the role that the social bond plays dur-
ing adolescence in explaining the changing level of self-control. Although
G&H (1990) argued that parental socialization is the primary source
of self-control in the formative period of early childhood, this study
provides a rigorous test of whether parental socialization during ado-
lescence still has an impact on self-control in a cumulative fashion
by adopting a reciprocal causation model within a longitudinal SEM
framework.

DATA
The data used in this study are part of the second generation of the JHU
PIRC field trials, which involve both “classroom-centered” and “family-
school partnership” interventions directed at improving school achieve-
ment and reducing conduct problems at home and in the school. The inter-
vention design involved 678 first-graders and their families recruited from
27 classrooms in nine Baltimore City public elementary schools. Of these
678 children, 53.2 percent were male, 86.8 percent were African American,
and 63.4 percent of the children were on free or reduced-cost lunch. At
the entrance into the first grade in 1993, the age of the children ranged
from 5.3 to 7.7 years with a mean age of 6.2 years (standard deviation
[SD] = .34). This study examines the family–school partnership interven-
tion that was designed to provide parents with effective teaching and child
CAN SELF-CONTROL CHANGE? 439

behavior management strategies via a series of workshops led by the first-


grade teacher and school psychologist or social worker. Accordingly, our
analyses use only 448 individuals assigned to either the family–school part-
nership intervention (treatment) or the control. The final sample size is
reduced to 399 after removing 49 cases with missing data. No significant
differences were found between remaining and missing cases in terms of in-
tervention conditions or key sociodemographic characteristics (see section
S.1 in the supporting information).14
A randomized block design was employed, with schools serving as the
blocking factor. Three first-grade classrooms in each of nine elementary
schools in Baltimore City were randomly assigned to the treatment or
the control condition. Although the intervention was limited to grade 1,
subjects were interviewed at 10 subsequent time points in the spring of
grades 1–3 and again in grades 6–12. A pretest or baseline assessment was
conducted in the early fall of grade 1. Unfortunately, self-control subscales
created in grades 1–3 had different indicators and were measured by differ-
ent data collection methods (interview vs. checklist format) from those in
grades 6–12. Because of these substantive and practical concerns,15 it was
difficult to use the scores from these two time frames in conjunction. In
addition, other key variables of interest in this study were measured only
during grade 1 and then during grades 6–11 (social control). As a result
of these inherent data limitations, the analyses in this study are conducted
based on the data from grades 6–12 focusing solely on the long-term effect
of an early prevention/intervention program administered during grade 1.

MEASUREMENT
Because the characteristics of respondents, such as their self-control,
affect the validity of responses to questionnaires, it is advisable to seek
measures that are assessed and collected independently of the respondent
(Gottfredson, 2006: 94). In this study, self-control was measured by the
child’s school teacher and social control was measured by the parent or
caregiver. Descriptive statistics can be found in section S.2 in the supporting
information.

Self-Control
Cognitive-based measures of self-control often are preferred to
behavioral-based measures in testing G&H’s (1990) theory because they

14. Additional supporting information can be found in the listing for this
article in the Wiley Online Library at http://onlinelibrary.wiley.com/doi/10.
1111/crim.2012.50.issue-2/issuetoc.
15. Please refer to the JHU PIRC website: http://www.jhsph.edu/prevention/Data/
Cohort3/index for more details.
440 NA & PATERNOSTER

are less vulnerable to the issue of tautology. Some scholars, including


Hirschi and Gottfredson (1993: 49) themselves, however, have argued
that behavioral measures better capture the theoretical construct of self-
control because they have greater construct validity and accordingly are
more consistent with the theoretical propositions articulated by self-control
theory. More recently, Gottfredson (2006) once again clearly articulated
that, because self-control itself likely affects survey responses, behavioral
measures—either from the respondent or an informant—are preferable to
attitudinal survey responses. The current study uses behavioral measures
of self-control taken from teachers. In addition, in response to the criticism
that behavioral measures are inherently tautological, only specific measures
that capture behavioral manifestations of self-control itself but inherently
do not involve force or fraud for self-gratification are used. Most of all,
this study uses a sufficient number of indicators of theoretical elements
of self-control to increase measurement reliability. Finally, these indicators
were administered only by youths’ English/language arts and mathematics
teachers for consistency. The responses are scored in such a way that higher
values reflect more self-control.
The Teacher Report of Classroom Behavior Checklist (TRCBC). Five
domains of self-control are assessed in the TRCBC: 1) impulsivity,
2) hyperactivity, 3) concentration problems, 4) oppositional-defiant behav-
ior, and 5) helplessness. These domains are measured at yearly intervals
with identical items employed from grades 6 to 12. The five domains of
self-control used in this study have strong face validity because they cap-
ture the behavioral manifestations of some combination of the defining
elements of self-control from G&H’s (1990) original theory (see section
S.3 in the supporting information). For each grade from grade 6 to grade
12, a summated scale of self-control was created with items measuring
these five domains. The coefficient alphas over grades 6–12 ranged from
.65 to .79 for impulsivity, from .76 to .88 for hyperactivity, from .90 to .93 for
concentration problems, from .87 to 93 for oppositional-defiant behavior,
and from .83 to .86 for helplessness (see also section S.3 in the supporting
information). Scores range from 0 to 5 with higher scores indicating greater
self-control.

Social Control/Bonds
G&H (1990: 96) explicitly argued that people are not born criminals, do
not inherit a propensity for criminality, or anything of the sort, but that
individual differences in offending propensity seem to be largely products
of ineffective or incomplete socialization by primary caretakers. Although
they clearly argued that a mutual level of emotional attachment between
caregiver and child is a necessary condition for effective socialization, they
primarily emphasized the control dimension of socialization by arguing that
CAN SELF-CONTROL CHANGE? 441

much parental action is in fact geared toward the suppression of impulsive


behavior, toward making the child consider the long-range consequences of
their acts (1990: 96–7, emphasis added).
In recent work, they also have argued (Gottfredson, 2006; Hirschi, 2004:
543) that at least empirically in adolescence, “social control and self-control
are the same thing” in the sense that a strong attachment between child
and caregiver is both necessary and sufficient for the successful creation of
self-control. In midadolescence, therefore, although it may be difficult to
distinguish self- from social control, an attentive and involved caregiver is
an indispensible antecedent to successful socialization and self-control. In
view of this theoretical position (and out of curiosity driven by Hirschi’s
1969 view about the instability of the social bond), we also measure ele-
ments of the social bond over time.
In this study, caregivers in the treatment group received interventions
targeting elements of the social bond, specifically the improvement of
disciplinary practices such as monitoring conduct, recognizing bad con-
duct, and properly dealing with bad behavior—all the elements of which
are consistent with what G&H (1990) emphasized as the source of self-
control. In addition, caregivers also learned how to strengthen their re-
lational attachment to and involvement with their children to improve
on academic achievement and behavioral outcomes through enhancement
of parent–teacher communication. We measure the social bond with the
Structured Interview of Parent Management Skills and Practices (SIPMSP).
Scores ranged from 1 to 5 with higher scores indicating a stronger social
bond.
The SIPMSP was intended by the original research team to assess the
major constructs included in Patterson, DeBaryshe, and Ramsey’s (1989)
model of the development of antisocial behavior in children. Our examina-
tion of its content convinced us that it is a reasonably accurate analog of
the elements of parental attachment in the social bond of Hirschi (1969).
SIMPSP includes parent disciplinary practices and practices associated
with the development of antisocial behavior. The relevant parental disci-
plinary practice constructs are 1) parental monitoring, 2) discipline, 3) re-
inforcement, 4) rejection, and 5) problem solving. In collaboration with the
Oregon Social Learning Center Prevention Center, JHU PIRC also modi-
fied the SIMPSP to include items assessing parent–teacher communication
and involvement and support for the child’s academic achievement. Based
on extant theories and research (Hay and Forrest, 2006; Hirschi, 2004;
Tittle, Ward, and Grasmick, 2004), we created five subscales that repre-
sent the key elements of the caregiver–child attachment component of the
social bond that would function as a source of self-control: 1) monitoring,
2) punishment, 3) attachment, 4) involvement, and 5) support (see section
S.3 in the supporting information for more details). A summated scale
442 NA & PATERNOSTER

of the social bond was created at each grade from grade 6 to grade 11.
The coefficient alphas for the POCA subscales ranged from .25 to .67
(monitoring), from .75 to .80 (punishment), from .59 to .85 (attachment),
from .33 to .59 (involvement), and from .50 to .72 (support) (see section S.3
in the supporting information).

ANALYTIC STRATEGY
Investigation of the patterns and sources of stability and change in
self-control requires making decisions about the statistical model to be
employed. Currently, HLMs (Raudenbush and Bryk, 2002) and group-
based trajectory models (GBTMs; Nagin, 2005) are two statistical model-
ing techniques that are most commonly used for modeling developmental
trajectories in the field of criminology (Kreuter and Muthén, 2008: 2).
These approaches are based on different assumptions,16 and the choice
of model should primarily be driven by a strong theoretical rationale,
rather than by some practical or other concern (Nagin, 2005; Nagin and
Piquero, 2010: 109; Sampson and Laub, 2005: 911). G&H’s (1990) theory is
claimed to be a “general” theory of crime and therefore is strongly opposed
to offender taxonomies. They never predicted the existence of groups with
distinct etiological implications, but they made a prediction about a more
uniform developmental commonality in the population.
This study employs both HLM and LGM (Bollen and Curran, 2006), ap-
proaches assuming that all subjects in the population are growing according
to a common developmental pattern, but the growth parameters (HLM) or
growth factors (LGM) may vary in their magnitude across individuals. In
particular, we believe these approaches are more relevant than GBTM for
the purpose of this study because 1) the current data involve a homoge-
neous sample with the subjects sharing similar individual and environmen-
tal characteristics; 2) data are limited to the relatively early stages of life
when individuals tend to share relatively similar life events and experiences
and therefore follow a similar developmental pattern; and 3) an additional
substantive interest of this study is to advance current understanding in the
field by explicitly comparing the results from conventional HLM (that use

16. In HLM and LGM, the joint distribution of either the observed or latent outcome
variables of interest are assumed to be normally distributed and individual vari-
ation is expressed as random coefficients or growth factors that are allowed to
vary across individuals. GBTM does not rest on such distributional assumptions
but attempts to explain variations in offending patterns by group membership
with distinct developmental pathways (but there is no further variation within the
groups).
CAN SELF-CONTROL CHANGE? 443

composite scales as in GBTM) with those from second-order LGM (that


retain the original indicators of some latent construct).17
In HLM and second-order LGM, random intercepts and random slopes
permit each individual in the sample to have a unique trajectory over time.
In HLM, individual change is analyzed by a two-level hierarchical model.
At level 1, each individual’s level of self-control is represented by an indi-
vidual growth trajectory that depends on a unique set of parameters (e.g.,
intercept and slope)18 . Although conceptually taking a similar approach,
the main difference between HLM and LGM is that, in the latter, such ran-
dom coefficients are incorporated into an SEM framework by considering
both the intercept and slope growth factors as latent variables.19 In addition,
second-order LGM can go farther than traditional LGM by combining
longitudinal confirmatory factor analysis (CFA) into a structural model that

17. Nonetheless, the same patterns observed in this study are replicated when GBTM
is employed. We found that both approaches have their own advantages and
limitations in understanding developmental trajectories, which is beyond the scope
of this study and will be presented in a subsequent article.
18. That is, the observed self-control score at time t for individual i is a function of
a systematic growth trajectory plus random error. The level 1 model assuming a
polynomial of degree P could be represented as Yti = π0i + π1i ati + π2i ati2 + · · · +
π Pi atiP + eti . For i = 1, . . . , n subjects, where ati is the grade at time t for subject i, π pi
is the growth parameter p for subject i associated with the polynomial of degree
P, and eti represents a deviation of an observation from the individual’s growth
prediction. The level 2 model with P + 1 individual growth parameters could be
represented as

QP

π pi = β p0 + βpq Xqi + r pi
q=1

where Xqi is either a measured characteristic of the individual’s background or


of an experimental treatment, βpq represents the effect of Xq on the pth growth
parameter, and r pi is a random effect with mean of zero. The set of P + 1 random
effects for individual i are assumed to be multivariate normally distributed with
full covariance matrix, T, dimensioned (P + 1) × (P + 1) (see Raudenbush and
Bryk, 2002: 160–204 for more details).
19. If ηt is denoted as a latent construct indicated at time t by J measured variables ϒt j .
The level 1 measurement model could be represented as y = τ + η + ε, where y
is a vector that contains T sets of values across time for J measured variables Y,
τ is a vector of intercepts of measured variables,  is a matrix of factor loadings
relating each ηt construct to its measured variables, and ε is a vector of random
normal errors. The level 2 structural model could be represented as η = ξ + ζ ,
where is a matrix of second-order factor loadings reflecting hypothesized growth
patterns underlying ηt constructs (e.g., for linear function, [1 1 1 . . . 1] in the first
column and [0 1 2 . . . 6] in the second column), ξ is a vector of growth factors (e.g.,
intercept, slope), and ζ is a vector of random normal disturbances in the first-order
ηt constructs.
444 NA & PATERNOSTER

allows for an assessment of the tenability of the hypothetical measurement


model over multiple time points.
After determining the distinct patterns and sources of stability and
change in self-control, we directly examine the longitudinal relationship
between self-control and social control/bonds over 5 years. We first build
two latent constructs, one for self-control and another for the social bond.20
Then, by employing longitudinal SEM with a panel design, we examine
whether there is a time-lagged bidirectional relationship between self- and
social control over time. In particular, we compare fit indices and parameter
estimates from the unidirectional and bidirectional models to assess which
model fits the data better.

RESULTS
HLM ANALYSES
Although more complicated functional forms of a model may better
capture meaningful patterns of variation, a simpler functional form can still
provide an easy-to-understand, good approximation of the general pattern
of growth trajectories of interest. Considering the primary goal of this study
is to investigate different rates of change between individuals to verify
whether the relative rank ordering of the values on the outcome variable
changes over time, a simplified model with only a linear growth parameter
is employed.21
The HLM results predicting the general pattern of self-control devel-
opment from grades 6–12 are shown in table 1. The fixed-effect results
provide an estimate of the initial level and growth rate of self-control in the
population, whereas the random-effect results allow us to assess whether
subjects exhibit significant variation around the estimated mean trajectory.
The fixed-effects results suggest that 1) children have a mean self-control
score of 3.16 points at grade 6 (average age of 12 years), and 2) the level
of self-control increases on average by approximately .04 points with each
increasing grade (p < .001). This result is consistent with G&H’s (1990)
prediction that, although the relative rankings or differences between-
individuals’ self-control remain stable over time, within-individual self-
control may change (most likely increase) over time by the direct influence

20. We think that our measure of self-control captures the conceptual common ground
among the diverse meanings given this term in the field (see Gottfredson, 2006;
Gottfredson and Hirschi, 1990; Hirschi, 2004; Tittle, Ward, and Grasmick, 2004;
Wikström and Treiber, 2007).
21. A nonlinear model (Self-controlti = πoi + π1i Gradeti + π2i Grade2ti + eti ) produced
almost identical patterns to those of the simpler linear-only model.
CAN SELF-CONTROL CHANGE? 445

Table 1. HLM Results Predicting the Pattern of Self-Control


Development: Grades 6–12 (N = 399)
Fixed Effect Coefficient SE t ratio df p value
For π0
β 00 3.160 .084 37.687 398 <.001
For π1
β 10 .040 .008 4.954 398 <.001

Random Effect SD Var. χ2 df p value


r0 1.197 1.4337 828.7 379 <.001
r1 .094 .0088 637.2 379 <.001
Level 1, e .549 .3006

of aging or socialization that continues to occur. In this sample, the within-


individual level of self-control continues to increase in general although its
magnitude is minimal (.04). The random-effects table indicates that both
the intercept and the slope (linear growth parameters) vary significantly
among individuals (p < .001), which suggests that the growth pattern of self-
control in the current sample is characterized by substantial instability. That
is, because slopes are not parallel, an individual’s standing on self-control
varies significantly over time—a finding that contradicts G&H’s prediction
that changes in the rankings of self-control over time should be minimal.
The covariance between the initial level of self-control and the rate of
change is also an important characteristic to assess the relative stability
hypothesis of the theory. The negative value of the estimated covariance
between the intercept and slope parameters (–.096) suggests that individ-
uals with lower levels of self-control at grade 6 tend to gain it at a faster
rate,22 which opens the possibility of a reshuffling of self-control trajectories
among individuals over time. Under the assumption of a normal distribu-
tion of slope parameters, the estimated fixed effect (mean growth rate) and
random effect (standard deviation) for the slope parameter can be used
to create the distributional representation for the growth pattern of self-
control over time for these youths. Figure 1 indicates that approximately
95 percent of youths change annually between –.145 and .224 points along
the self-control continuum. Interestingly, almost 42 percent of individuals
manifested a decreasing level of self-control over time, which was not
predicted by G&H (1990). In addition, this finding opens the possibility that
exogenous social variables continue to explain such a changing pattern of
self-control over time.

22. We note that the correlation between initial status and rate of change may vary
depending on the specific time point selected for the initial status (Raudenbush
and Bryk, 2002: 167).
446 NA & PATERNOSTER

Figure 1. Distribution of Growth Pattern (Slope Parameter)


of Self-Control: Grades 6–12

41.5

-.145 0 .040 .224

95

NOTE: The higher values of self-control scale represent more self-control.

Figure 2 shows that, although no significant difference occurs in the


initial level of self-control, the level of self-control increases with time at
a significantly higher rate (p < .10) for the youths in the treatment group
compared with those in the control group. Combined with the evidence of
a negative correlation between initial status and growth rate, the significant
interaction effect observed between treatment and growth represents more
direct evidence for the possibility of a reshuffling in the relative rankings
of self-control over time than the magnitude and significance of a bivariate
correlation between self-control scores across different time points (e.g., the
relative ranking of self-control for the youths that belong to the treatment
group—especially those with a lower level of self-control at the initial
time point—will significantly and substantially increase at a higher rate).
Interestingly, after incorporating a changing level of social control/bond as
a time-varying covariate at level 1, the observed different rates of change
between the two groups become negligible in both magnitude and signif-
icance.23 That is, by controlling for both time specific, within-individual

23. Because the variance of social control/bond in this model represents the combi-
nation of two sources (within- and between-person), treating this time-varying
predictor as a single variable—instead of differentiating the impact of more stable,
between-individual sources of variation from the impact of time-specific, within-
individual sources of variation—will usually result in biased parameter estimates
(for more details, see Curran and Bauer, 2011; Hoffman and Stawski, 2009). We
reestimated the model using a person-mean-centered social control/bond in the
level 1 model and added its between-person mean in the level 2 model. Even
after employing this method for achieving a disaggregation of effects within the
Figure 2. Different Rate of Change in the Level of Self-Control between Control and Treatment
Groups—Before (Left Panel) and After (Right) Controlling for Time-Varying Covariate
of Social Control/Bond: Grades 6–12

3.73 3.67
Treatment Treatment
Control Control
3.64 3.59

3.55 3.52
SC SC

3.47 3.45

3.38 3.37
6.00 7.50 9.00 10.50 12.00 6.00 7.25 8.50 9.75 11.00
GRADE GRADE
CAN SELF-CONTROL CHANGE?

β11 = .027; p = .087 (two-tailed) β11 = .013, p = .524 (two-tailed)


NOTE: The higher values of self-control scale represent more self-control.
447
448 NA & PATERNOSTER

variation of the social bond in the level 1 model and more stable, between-
individual variation of social control/bond in the level 2 model after dis-
entangling the within- and between-individual variance components, the
observed distinct pattern of self-control development among the youths
that belong to the treatment and control groups becomes negligible. This
observation suggests that the within-individual change of self-control over
time is related to the within-individual change of social control/bond even
after the formative period of early childhood, which opens the possibility of a
malleable nature of self-control and a dynamic relationship between the two
control mechanisms.

SECOND-ORDER LGM ANALYSES


At the level of the measurement model of the second-order LGM, the
expected value of the observed items is not only a function of the factor
loadings but also depends on the intercept and error variance (y = τ +
η + ε). The “growth of composite” models (e.g., HLM and first-order
LGM) that use the sums or means of items measuring the same concept
assume that the factor loadings, intercepts, and error variances of the
indicators measuring the same latent constructs are equivalent at different
time points. If these very restrictive and seemingly unrealistic assumptions
are not met, however, the composite scales are contaminated and the
estimated growth parameters may be biased, which makes it difficult to
distinguish between mere changes in the measurement structure and true
longitudinal change in the latent construct (Leite, 2007: 582). In this section,
the factor loadings, measurement errors, and intercepts of multiple items
are estimated using a “second-order LGM” (Hancock, Kuo, and Lawrence,
2001) or “curve of factor model” (McArdle, 1988) to verify whether the
results from the conventional HLM continue to hold after accounting for
the potential biases resulting from measurement error and violation of
factorial invariance assumptions.24 This direct effort addresses the concerns
of G&H (1990: 108), who asserted that any changes in the relative rankings

multilevel model, the substantive meaning of our key findings did not change
(before: β11 = .013, p = .524; after: β11 = .012, p = .533).
24. An alternative approach to HLM using composites of multiple items—to account
for the different amount of measurement error in a set of items—is estimated by
fixing the error variances of the composites using the reliability estimates (e.g.,
Cronbach’s alpha: Bollen, 1989; Jöreskog and Sorbom, 1996). Although the factors
represent the estimated true scores at each time point after removing the estimated
error variances, Leite (2007: 586–7), through Monte Carlo simulation, found that
both approaches produce the same parameter estimates, standard errors, chi-
square statistics, and fit indexes. Most of all, this approach also has an inherent
limitation in assessing factorial invariance assumptions because factor loadings
and intercepts of indicators are not estimated.
CAN SELF-CONTROL CHANGE? 449

of self-control over time should be “sufficiently small to suggest that they


may be accounted for in large part by misidentification or measurement
errors” (emphasis added).
After fixing a measurement model that fits the data reasonably well:
χ 2 = 1,114.791, p = .000 (df = 420), standardized root mean residual
(SRMR) = .072, root mean square error of approximation (RMSEA) =
.064, comparative fit index (CFI) = .936,25 the second-order LGM was fit
to the data using maximum likelihood (ML) estimation in Mplus version
6.1 (Muthén and Muthén, 1998–2010). Conceptually, this model replicates
what we have already found in table 1 and figure 1 using HLM. The
unconstrained second-order LGM with mean structure still fits the data
reasonably well: χ 2 = 1,155.252, p = .000 (df = 443), SRMR = .078,
RMSEA = .063, CFI = .934. The results suggest that, even after accounting
for the measurement error of observed indicators, the level of self-control
is found to be increasing on average by approximately .09 points over time
(p < .01) with evidence of significant variability across individuals (.01,
p < .01)—findings similar to those of the HLM (β11 = .04, r 1 = .01, both
at p < .001).
To assess the tenability of factorial invariance assumptions, a series of
likelihood ratio tests were conducted with increasingly more restricted
models, which compared the fit indices of the more restricted solution with
those of a comparable solution without constraints (see section S.4 in the
supporting information for more details). Although a marginally nonin-
variant factor structure was observed in the current data,26 the substantive
meaning of our findings were not affected because the parameter estimates
of primary interest (e.g., fixed and random effects of slope) in the second-
order LGM are almost identical across different model specifications when
different levels of invariance constraint and reference indicators are used.27

25. Error variances (indicator uniqueness) for the corresponding items measuring the
same constructs were allowed to covary across different time points, which is
justified on the basis of method effects that result from the application of common
assessment methods over time.
26. This observation also accounts for the method effects originating from rater
unreliability when different raters provide the observations for the same subjects
each year, which was a concern raised by one reviewer.
27. At this point, further models could be examined to improve understanding of
the source of invariance in the context of the “partial invariance” model (Byrne,
Shavelson, and Muthén, 1989), followed by a set of additional sensitivity checks
to verify whether there are meaningful changes in the estimates of important
parameters under different versions of partial invariance specifications (e.g., using
only some combination of a subset of constrained factor loadings and item inter-
cepts found to be invariant over time). Given the consistent patterns of parameter
estimates across different levels of factorial invariance, however, we decided not
to investigate further the partial noninvariance patterns within each level.
450 NA & PATERNOSTER

Figure 3. Second-order LGM with Time-Invariant and


Time-Variant Covariates (with Strong Factorial
Invariance Constraints)
Group
.05 ( p = .526) .02 ( p = .231)

-.040**

Intercept Slope
1 1 0
1 1 1
1 1 2 5
3 4

SC6 SC7 SC8 SC9 SC10 SC11

.778** .796** .790** .802** .841** .861**

SB6 SB7 SB8 SB9 SB10 SB11

NOTE: Measurement model using longitudinal CFA is not shown for the
interest of simplicity.
∗∗
p < .01 (two-tailed).

The distributional representation for the growth pattern of self-control


estimated by a second-order LGM also is similar to that estimated by HLM.
Approximately 95 percent of youths change annually between –.119 and
.209 points along the latent self-control continuum, and almost 30 percent
of individuals manifest decreasing levels of self-control over time, which
contradicts G&H’s (1990) prediction about a steady rate of increase in
socialization.
In the second-order LGM, the path from group membership (0 = control,
1 = treatment) to the slope factor describes the sign and magnitude of the
effect of group membership on the changing rate of self-control. Similar to
HLM results (β11 = .027, p < .10), the second-order LGM also confirms
that individuals who received treatment during grade 1 had faster increas-
ing rates than those who did not by approximately .03 points each year
(p < .10) and the same pattern persists when different factorial invari-
ance constraints and reference indicators are used. In figure 3, we show
the second-order LGM with both time-invariant (group membership) and
CAN SELF-CONTROL CHANGE? 451

Table 2. Goodness-of-Fit Indices of the Longitudinal SEM


(N = 395)
Model χ2 (df)  χ2 (df) SRMR RMSEA CFI
Unidirectional
SEM-0 2,870.6 (1,541) .181 .047 .899
SEM-1 3,044.7 (1,581) 174.1∗∗ (40) .232 .048 .889
SEM-2–1 3,415.9 (1,621) 371.2∗∗ (40) .111 .052 .870
Bidirectional
SEM-0 2,588.8 (1,536) .084 .042 .920
SEM-1 2,738.7 (1,576) 149.9∗∗ (40) .094 .043 .911
SEM-2–1 3,131.1 (1,616) 392.4∗∗ (40) .108 .049 .885
∗∗ p < .01 (two-tailed).

time-variant (social control/bond) covariates. After incorporating changing


levels of social control/bond as a time-varying covariate, the different rate
of change becomes smaller and nonsignificant (.02, p = .231), which also
is consistent with the HLM result (β11 = .013, p = .524). Moreover, the
paths from the construct for the social bond (SB) to self-control (SC) are
relatively strong, consistent in magnitude, and significant across all time
points.28 The consistency of the findings from different analytic techniques
and model specifications allows for further investigation of the dynamic
relation between self-control and social control over time.

LONGITUDINAL SEM ANALYSES


Both the HLM and the second-order LGM results suggest that there is
substantial variability in the growth rate of self-control between individuals,
especially those between the two study groups, and that such variability
is accounted for by the changing level of the social bond triggered by
the treatment condition. To disentangle the causal mechanism underlying
stability and change of self- and social control over time, a longitudinal
SEM with a panel design is employed. The results for the longitudinal SEM
model are reported in table 2 and figure 4.
As in the second-order LGM, table 2 shows that the more constrained
models fit the data significantly worse than the less constrained models.29
Nonetheless, a distinct pattern is found across different model specifications

28. The same pattern persists when a time-lagged model is adopted.


29. Modification indexes suggest that establishing strong factorial invariance for the
social bond construct is the primary source of ill-fit, which means that the mea-
surement structure of the social bond construct has changed significantly across
the different developmental periods. Although we are not overly enthusiastic with
the fit indices presented in table 3—especially those of constrained models—we
decided to add further invariance constraints in view of the prevailing rationale in
452 NA & PATERNOSTER

Figure 4. SEM with Longitudinal Latent Variables


(with Strong Factorial Invariance Constraints):
Unidirectional (Upper Panel) and Bidirectional
(Lower Panel)
1.028** .969** 1.008** 1.030** 1.013**
     
SC6 SC7 SC8 SC9 SC10 SC11

.009 .003 .049* -.035 .018

     
SB6 SB7 SB8 SB9 SB10 SB11
.970** .985** .983** 1.021** .976**

.618** .683** .681** .576** .746**


     
SC6 SC7 SC8 SC9 SC10 SC11

.014 -.003 .051** -.049* .037


.312** .246** .262** .376** .223**

     
SB6 SB7 SB8 SB9 SB10 SB11
.979** .995** .947** 1.028** .940**


p < .05; ∗∗ p < .01 (two-tailed).

that support a social causation model over a self-selection model, where


social causation is reflected in the path from SB (the social bond) to SC
(self-control) and self-selection in the path from SC to SB in figure 4.
In G&H’s (1990) theoretical model, those low in self-control should sub-
sequently self-select themselves into environments that offer little in the
way of effective socialization, whereas in S&L’s (1993, 2003) theoretical

measurement theory (see section S.4 in the supporting information) and consis-
tency with previous analyses. Although error covariances for the same indicators
across different time points were specified in this study (on the basis of a method
effect that results from the repeated application of common assessment methods),
error covariances across multiple indicators measuring the same constructs at
a given time point or measuring different constructs were not added. Doing
so would improve overall fit indices because such a model could explain more
variance/covariance among variables (and modification indexes also suggest this).
However, we did not do this for the interest of parsimony (which is related to
RMSEA) at the sacrifice of the other fit indices and to avoid the overparameteriza-
tion of the model given our relatively small sample size. Most of all, the parameters
of these error covariances are theoretically less interesting and specification of too
many error covariances might affect the variances/covariances of primary interest
in this study (the covariance between two theoretical constructs in the “structural”
part of the model).
CAN SELF-CONTROL CHANGE? 453

Table 3. The Comparison of Unstandardized Parameter


Estimates (N = 395)
Path All Control Group Treatment Group
Self-Selection
SC6 → SB7 .014 −.006 .046
SC7 → SB8 −.003 −.011 .013
SC8 → SB9 .051∗∗ .048 .066∗
SC9 → SB10 −.049∗ .012 −.183∗∗
SC10 → SB11 .037 −.002 .130∗
Social Causation
SB6 → SC7 .312∗∗ .266∗∗ .365∗∗
SB7 → SC8 .246∗∗ .245∗∗ .265∗∗
SB8 → SC9 .262∗∗ .203∗∗ .335∗∗
SB9 → SC10 .376∗∗ .306∗∗ .482∗∗
SB10 → SC11 .223∗∗ .220∗∗ .249∗∗
∗p < .05; ∗∗ p < .01 (two-tailed).

model, social control can continue to have an effect on self-control well into
adolescence and beyond. The deletion of five directional paths from the
social bond to time-lagged self-control leads to a significant deterioration
in model fit relative to changes in the degrees of freedom. This outcome
is not surprising considering the magnitude and significance of the path
parameters that were included in the bidirectional model but omitted in
the unidirectional model, all of which are consistently strong and significant
across different model specifications.
Interestingly, no meaningful pattern is observed in the directional paths
from self-control to the time-lagged social bond measure. The upper panel
in figure 4 shows that only one path (SC8 to SB9), although small in
magnitude, is significant (p = .049). Nonetheless, we conclude that it is the
exception to the rule because the unidirectional model with the deletion of
all the directional paths from self-control to the time-lagged social bond did
not lead to a significant deterioration in the fit of model: χ 2 (d.f.) = 15.355
(40), p > .05. In sum, contrary to G&H’s (1990) assertion but consistent
with Hay and Forrest (2006) and S&L (1993, 2003), our data suggest that
social causation processes continue to occur during adolescence, whereas
the magnitude and significance of the self-selection process is negligible
during the same period.
To verify further whether the same process explains both stability and
change within a unified theoretical framework (Caspi, Robert, and Shiner,
2005; Longshore, Chang, and Messina, 2005; Nagin and Paternoster, 1994;
Sampson and Laub, 2003), multiple-group analyses were conducted within
a longitudinal SEM framework. A multiple-sample analysis estimates a
model across two study groups where the systems of equations are solved
for all groups together. Table 3 shows that the same directional pattern of
454 NA & PATERNOSTER

effects persist across both the control and the treatment group, supporting
the social causation model over the self-selection model. In particular, the
pattern of parameter estimates observed in both groups combined becomes
more pronounced in the treatment group.

DISCUSSION AND CONCLUSION


In contrast to G&H’s (1990) prediction that any observed difference
in self-control among individuals should remain relatively stable after 8
or 10 years of age (Hirschi and Gottfredson, 2001: 90), this study has
found meaningful differences in the growth pattern of self-control during
adolescence across individuals in the pooled sample. In other words, sig-
nificantly different rates of change among individuals suggest that one’s
relative standing on self-control varies over time. A subsample, those whose
caregivers were part of an intervention effort to improve parenting prac-
tices, showed significantly greater gains in self-control than did a control
group. These findings suggest that self-control is malleable, is responsive to
intentional attempts to increase it, and continues to develop in response
to the changing level of social control/bond at least up until grade 12
(17 years of age). Although a recent meta-analytic review of 34 evaluation
studies (Piquero, Jennings, and Farrington, 2010: 829) has already suggested
that programmatic efforts aimed at improving self-control and reducing
delinquency among children and young adolescents up to 10 years of age are
effective, the authors acknowledge that research is limited to the outcomes
before 10/12 years of age and it is still not clear “if the effectiveness of self-
control improvement programs persists over time.”
This study adds to existing scholarship in several ways. First, in direct
response to the call for “theory-driven” approaches to the choice of an-
alytic models and to verify whether the same patterns identified in prior
research persist under alternative modeling strategies, we employed HLM
and LGM approaches to investigate better the variability of individual
trajectories within a common functional form. Second, in direct response
to the measurement concerns of G&H (1990), we employed second-order
LGM and found that the observed differences are not an artifact of sta-
tistical methods, measurement error, or model specification. Third, given
the random assignment of subjects to the treatment or control group, an
assignment that is exogenous to the initial level of self-control, the different
rates of change that we observed between groups can best be explained by
the changing level of social control/bonds within the individual—a change
triggered by treatment conditions. Although we do not exclude the pos-
sibility that genetic/biological factors might substantially influence one’s
offending propensity (Barkeley, 1997; Guo, Roettger, and Cai, 2008; Rowe,
1994; Unnever, Cullen, and Pratt, 2003; Wright and Beaver, 2005), and that
community-level factors also might either directly or indirectly influence
CAN SELF-CONTROL CHANGE? 455

the formation and development of self-control (Lynam et al., 2000; Pratt,


Turner, and Piquero, 2004), random assignment controls not only for such
common confounders but also for other factors such as aging, maturational,
or historical processes that are being shared by subjects within different
study groups. Last, this study was better able to disentangle the long-term
relation between self- and social control, which is found to be more dynamic
than previously hypothesized.
Practically, this study provides convincing evidence of the utility of
prevention/intervention programs by tracking and highlighting the long-
term effects of such efforts, instead of merely comparing before-and-after
mean scores of the outcome variable. We believe such a “process-based”
approach focusing on the question of “how” and “why” will better inform
program developers and policy makers about the elements and conditions
of programs that are related to an observed successful outcome. In addition,
we need to note that any effort targeting self-control during adolescence
(e.g., in the school setting) should still be encouraged considering the mal-
leable nature of self-control observed here (see also Hay and Forrest, 2006).
In this vein, although S&L (1993) originally developed the idea of a “turn-
ing point” to focus on the life events and experiences during the adult stages
of life, this study demonstrates how some good prevention/intervention
programs during relatively earlier stages also might function as a “turning
point” of an individual’s life trajectory. Although G&H (1990: 105–6) did
not completely discount the possibility that an institution such as the school
could improve on previously unsuccessful socialization, they were not opti-
mistic because “[t]he evidence suggests . . . that in contemporary American
society the school has a difficult time teaching self-control.” Our findings do
not entirely contradict this observation. The intervention studied here that
was successful in improving students’ self-control involved a partnership
between caregivers and the school personnel. Such joint ventures may
prove more effective than either institution working alone.
Theoretically, the observed malleable nature of self-control and the dy-
namic relation between internal and external inhibiting factors over time
provide evidence for a “mixed” model of population heterogeneity and
state dependence. Although offending propensity such as that reflected
in self-control may be modestly stable over time, a substantial amount
of variability still exists across individual trajectories, which was in part
a result of the influence of the social contexts in which individuals were
situated. In particular, contrary to the pure population heterogeneity ex-
planation, which claims that time-stable propensities exert their influence
(independent of settings or) by making individuals select into criminogenic
settings that facilitate them to act in an almost predetermined manner, this
study found weak evidence of such pure self-selection processes. Similarly,
contrary to the pure state dependence thesis that emphasizes the predomi-
nant roles played by life events and experiences without any clear reference
456 NA & PATERNOSTER

to the changing nature of offending propensity, this study found some


strong and significant evidence that the improvement in the relationship
between caregiver and child by the treatment intervention continues to
have a direct impact on self-control.
By drawing on emerging evidence of the malleable nature of self-control
(e.g., Burt, Simons, and Simons, 2006; Hay and Forrest, 2006), future ef-
forts could investigate the possibility of the theoretical elaboration and
modification—and eventually, the theoretical integration—of both control
perspectives. Even the most recent efforts to reconceptualize and reopera-
tionalize self- and social control (Gottfredson, 2006; Hirschi, 2004; Piquero
and Bouffard, 2007; Tittle, Ward, and Grasmick, 2004) pay little attention
to the mechanism we have already discussed in this article. We think
Wikström and Treiber’s (2007, 2009) theoretical ideas are consistent with
the “mixed” model, and this line of thinking is useful in understanding
how internal and external sources of social control interact over time. In
particular, they clearly asserted that differences in executive capabilities
continue to develop because one’s internal representation of potential costs
and benefits associated with crime and deviance remains flexible and adapt-
able to changing circumstances.
Although we think our study is informative, it is not without its limita-
tions. First, this study uses a sample of high-risk youth (inner-city minorities
with low socioeconomic status), and the findings cannot be generalized
without caution to more representative, general population samples. In-
deed, G&H’s (1990) theory was developed based on the empirical regulari-
ties observed from a general population across different times and places. In
addition, it is possible that a group-based trajectory approach could identify
distinct group patterns in the general population that cannot be captured in
the HLM/LGM approaches. Although this study uses a relatively homoge-
neous sample that captures the early stages of life, and assumes that the
average causal effect of the intervention program is similar for all members
of the treatment group, this “constant effects” assumption might not hold
if substantial heterogeneity in the treatment effect exists. In this scenario,
a statistical analysis of the average treatment effect (ATE) could mask or
underestimate the meaningful local treatment effect (LTE).30
A second limitation of our study is that the base measurement model
(longitudinal CFA) could not be optimally specified prior to more

30. In a similar vein, G&H’s (1990) predictions might be supported when the same
analyses are replicated using a well-representative sample but focus only on the
ATE considering that most individuals tend to have a high level of self-control
and manifest a relatively stable trajectory. Thus, the same intervention could have
a negligible impact on them, and accordingly, the average change in the relative
ranking or distance observed might be minimal because the aggregate pattern may
obscure some meaningful differences between individuals or groups.
CAN SELF-CONTROL CHANGE? 457

complex analyses (second-order LGM and longitudinal SEM) because the


five subscales of self-control used in this study had already been cre-
ated by the Johns Hopkins research staff and the actual items that com-
prise those subscales were not made available. These subscales of self-
control do not precisely match each defining element of self-control as
presented by G&H (1990). Accordingly, although the final longitudinal
CFA model suggests a reasonable data-model fit, it still fails to reach the
conventional cutoff criteria and there still exists a possibility of model
misspecification.
In sum, although we think we have some provocative findings, corrobo-
ration and extension of our results await additional research from different
samples by other researchers. What we hoped to have accomplished is to
spark criminologists’ interest in thinking about a unified control model
of offending that incorporates insights from both social control and self-
control theories (Longshore, Chang, and Messina, 2005; Wikström, 2006;
Wikström and Treiber, 2007, 2009).

REFERENCES
Akers, Ronald. 1991. Self-control as a general theory of crime. Journal of
Quantitative Criminology 7:201–11.

Arneklev, Bruce J., John K. Cochran, and Randy R. Gainey. 1998. Test-
ing Gottfredson and Hirschi’s low self-control stability hypothesis: An
exploratory study. American Journal of Criminal Justice 23:107–27.

Barkeley, Russell A. 1997. ADHD and the Nature of Self-control. New


York: Guilford Press.

Beaver, Kevin M., and John Paul Wright. 2007. The stability of low self-
control from kindergarten through first grade. Journal of Crime and
Justice 20:63–86.

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New


York: Wiley.

Bollen, Kenneth A., and Patrick J. Curran. 2006. Latent Curve Models: A
Structural Equation Perspective. Hoboken, NJ: Wiley.

Burt, Callie H., Ronald L. Simons, and Leslie G. Simons. 2006. A longitu-
dinal test of the effects of parenting and the stability of self-control:
Negative evidence for the general theory of crime. Criminology 44:
353–96.

Byrne, Barbara M., R. J. Shavelson, and Bengt O. Muthén. 1989. Testing


for the equivalence of factor covariance and mean structures: The issue
of partial measurement invariance. Psychological Bulletin 105:456–66.
458 NA & PATERNOSTER

Caspi, Avshalom, Brent W. Robert, and Rebecca L. Shiner. 2005. Person-


ality development: Stability and change. Annual Review of Psychology
62:543–84.

Curran, Patrick J., and Daniel J. Bauer. 2011. The disaggregation of within-
person and between-person effects in longitudinal models of change.
Annual Review of Psychology 62:583–619.

Gottfredson, Michael R. 2005. Offender classifications and treatment ef-


fects in developmental criminology: A propensity/event consideration.
The ANNALS of the American Academy of Political and Social Science
602:46–56.

Gottfredson, Michael R. 2006. The empirical status of control theory in


Criminology. In Taking Stock: The Status of Criminological Theory,
eds. Francis T. Cullen, John Paul Wright, and Kristie Blevins. New
Brunswick, NJ: Transaction.

Gottfredson, Michael R., and Travis Hirschi. 1990. A General Theory of


Crime. Stanford, CA: Stanford University Press.
Guo, Guang, Michael E. Roettger, and Tianji Cai. 2008. The integration
of genetic propensities into social-control models of delinquency and
violence among male youths. American Sociological Review 73:543–68.

Hancock, Gregory R., Wen-Ling Kuo, and Frank R. Lawrence. 2001. An


illustration of second-order latent growth models. Structural Equation
Modeling 8:470–89.

Hay, Carter, and Walter Forrest. 2006. The development of self-control:


Examining self-control theory’s stability thesis. Criminology 44:739–74.

Higgins, George E., Wesley G. Jennings, Richard Tewksbury, and Chris L.


Gibson. 2009. Exploring the link between low self-control and violent
victimization trajectories in adolescents. Criminal Justice and Behavior
36: 1070–84.

Hirschi, Travis. 1969. Causes of Delinquency. Berkeley: University of


California Press.

Hirschi, Travis. 2004. Self-control and crime. In Handbook of Self-


Regulation: Research, Theory, and Applications, eds. Roy F. Baumeister
and Kathleen D. Vohs. New York: Guilford Press.

Hirschi, Travis, and Michael R. Gottfredson. 1993. Commentary: Testing


the general theory of crime. Journal of Research in Crime and Delin-
quency 30:47–54.
CAN SELF-CONTROL CHANGE? 459

Hirschi, Travis, and Michael R. Gottfredson. 1995. Control theory and the
life course perspective. Studies on Crime and Crime Prevention 4:131–
42.

Hirschi, Travis, and Michael R. Gottfredson. 2001. Self-control theory. In


Explaining Criminals and Crime, eds. Raymond Paternoster and Ronet
Bachman. Los Angeles, CA: Roxbury.

Hirschi, Travis, and David Rudisill. 1976. The great American search:
Causes of crime, 1876–1976. The Annals 423:14–22.
Hoffman, Lesa, and Robert S. Stawski. 2009. Persons as contexts: Evaluat-
ing between-person and within-person effects in longitudinal analyses.
Research in Human Development 6:97–100.

Jöreskog, Karl G., and Dag Sorbom. 1996. LISREL 8: User’s Reference
Guide. Chicago, IL: Scientific Software International.

Kreuter, Frauke, and Bengt Muthén. 2008. Analyzing criminal trajectory


profiles: Bridging multilevel and group-based approaches using growth
mixture modeling. Journal of Quantitative Criminology 24:1–31.

Laub, John H., and Robert J. Sampson. 2003. Shared Beginnings, Divergent
Lives: Delinquent Boys at Age 70. Boston, MA: Harvard University
Press.

Laub, John H., Robert J. Sampson, and Gary A. Sweeten. 2006. Assessing
Sampson and Laub’s life course theory of crime. In Taking Stock:
The Status of Criminological Theory, eds. Francis T. Cullen, John Paul
Wright, and Kristie Blevins. New Brunswick, NJ: Transaction.

Le Blanc, Marc. 2006. Self-control and social control of deviant behavior


in context: Development and interactions among the life course. In
The Explanation of Crime: Context, Mechanisms, and Development,
eds. Per-Olof H. Wikström and Robert J. Sampson. Cambridge, U.K.:
Cambridge University Press.

Leite, Walter L. 2007. A comparison of latent growth models for constructs


measured by multiple items. Structural Equation Modeling 14:581–610.

Longshore, Douglas, Eunice Chang, and Nena Messina. 2005. Self-control


and social bonds: A combined control perspective on juvenile offending.
Journal of Quantitative Criminology 21:419–37.

Lynam, Donald R., Avshalom Caspi, Terrie E. Moffitt, Per-Olof Wikström,


Rolf Loeber, and Scott P. Novak. 2000. The interaction between im-
pulisivity and neighborhood context on offending: Effects of impulsivity
460 NA & PATERNOSTER

are stronger in poorer neighborhoods. Journal of Abnormal Psychology


109:563–74.

McArdle, John J. 1988. Dynamic but structural equation modeling of re-


peated measures data. In Handbook of Multivariate Experimental Psy-
chology, 2nd ed., eds. John R. Nesselroade and Raymond B. Cattell.
New York: Plenum.

Mitchell, Ojmarrh, and Doris Layton MacKenzie. 2006. The stability and
resiliency of self-control in a sample of incarcerated offenders. Crime &
Delinquency 52:432–49.

Muthén, Bengt O., and Linda K. Muthén. 1998–2010. Mplus (v.6.1). Los
Angeles, CA: Muthén & Muthén.

Nagin, Daniel S. 2005. Group-based Modeling of Development Over the Life


Course. Cambridge, MA: Harvard University Press.

Nagin, Daniel S., and Raymond Paternoster. 1991. On the relationship of


past to future participation in delinquency. Criminology 29:163–89.

Nagin, Daniel S., and Raymond Paternoster. 1994. Personal capital and
social control: The deterrence implications of a theory of individual
differences in criminal offending. Criminology 32:581–606.

Nagin, Daniel S., and Raymond Paternoster. 2000. Population heterogene-


ity and state dependence: State of the evidence and directions for future
research. Journal of Quantitative Criminology 16:117–44.

Nagin, Daniel S., and Alex R. Piquero. 2010. Using the group-based trajec-
tory modeling to study crime over the life course. Journal of Criminal
Justice Education 21:105–16.

Patterson, Gerald R., Barbara D. DeBaryshe, and Elizabeth Ramsey. 1989.


A developmental perspective on antisocial behavior. American Journal
of Psychiatry 44:329–35.

Piquero, Alex R., and Jeff A. Bouffard. 2007. Something old, something
new: A preliminary investigation of Hirschi’s redefined self-control.
Justice Quarterly 24:1–27.

Piquero, Alex R., Wesley G. Jennings, and David P. Farrington. 2010.


On the malleability of self-control: Theoretical and policy implications
regarding a general theory of crime. Justice Quarterly 27:803–34.

Pratt, Travis C., Michael G. Turner, and Alex R. Piquero. 2004. Parental
socialization and community context: A longitudinal analysis of the
CAN SELF-CONTROL CHANGE? 461

structural sources of low self-control. Journal of Research in Crime and


Delinquency 41:219–43.

Raudenbush, Stephen W., and Anthony S. Bryk. 2002. Hierarchical Linear


Models, 2nd ed. Thousand Oaks, CA: Sage.
Rowe, David C. 1994. The Limits of Family Influence: Genes, Experience,
and Behavior. New York: Guilford Press.

Sampson, Robert J., and John H. Laub. 1993. Crime in the Making: Path-
ways and Tuning Points though Life. Cambridge, MA: Harvard Univer-
sity Press.

Sampson, Robert J., and John H. Laub. 1997. A life-course theory of


cumulative disadvantage and the stability of delinquency. In Develop-
mental Theories of Crime and Delinquency, ed. Terence P. Thornberry.
Brunswick, NJ: Transaction.

Sampson, Robert J., and John H. Laub. 2003. Desistance from crime over
the life course. In Handbook of the Life Course, eds. Jeylan T. Mortimer
and Michael Shanahan. New York: Kluwer Academic/Plenum.

Sampson, Robert J., and John H. Laub. 2005. Seduction of method: Rejoin-
der to Naign and Tremblay’s “Developmental trajectory groups: Fact or
fiction?” Criminology 43:905–13.

Taylor, Claire. 2001. The relationship between social and self-control: Trac-
ing Hirschi’s criminological career. Theoretical Criminology 5:369–88.

Tittle, Charles R., David A. Ward, and Harold G. Grasmick. 2004. Capacity
for self-control and individuals’ interest in exercising self-control. Jour-
nal of Quantitative Criminology 20:143–72.

Turner, Michael G., and Alex R. Piquero. 2002. The stability of self-control.
Journal of Criminal Justice 30:457–71.

Turner, Michael G., Alex R. Piquero, and Travis C. Pratt. 2005. The school
context as a source of self-control. Journal of Criminal Justice 33:327–
39.

Unnever, James D., Francis T. Cullen, and Travis C. Pratt. 2003. Parental
involvement, ADHD, and delinquent involvement: Reassessing
Gottfredson and Hirschi’s general theory. Justice Quarterly 20:471–500.

Wikström, Per-Olof H. 2006. Individuals, settings, and acts of crime: Situ-


ational mechanisms and the explanation of crime. In The Explanation
of Crime: Context, Mechanisms, and Development, eds. Per-Olof H.
Wikström and Robert J. Sampson. New York: Cambridge University
Press.
462 NA & PATERNOSTER

Wikström, Per-Olof H., and Kyle Treiber. 2007. The role of self-control in
crime causation: Beyond Gottfredson and Hirschi’s general theory of
crime. European Journal of Criminology 4:237–64.

Wikström, Per-Olof H., and Kyle Treiber. 2009. What drives persistent
offending? The neglected and unexplored role of the social environ-
ment. In The Development of Persistent Criminality, ed. Joanne Savage.
Oxford, U.K.: Oxford University Press.

Wikström, Per-Olof H., Vania Ceccato, Beth Hardie, and Kyle Treiber.
2010. Activity fields and the dynamics of crime. Journal of Quantitative
Criminology 26:55–87.

Wright, John Paul, and Kevin M. Beaver. 2005. Do parents matter in


creating self-control in their children? A genetically informed test of
Gottfredson and Hirschi’s theory of low self-control. Criminology
43:1169–202.

Chongmin Na is an assistant professor in the School of Human Sciences


and Humanities at the University of Houston—Clear Lake. His research in-
terests include causal mechanism underlying stability and change in offend-
ing behaviors, school-based crime prevention and control, and quantitative
methodology.

Raymond Paternoster is a professor in the Department of Criminology


and Criminal Justice at the University of Maryland. His research interests
include offender decision making, offending over the life course, adolescent
employment and crime, and capital punishment.

SUPPORTING INFORMATION
The following supporting information is available for this article:

Section S.1. Missing Data Pattern for Self- and Social Control/Bond
Section S.2. Descriptive Statistics for Key Variables
Section S.3. Subscales of Self- and Social Control
Section S.4. Technical Details on the Assessment of Factorial Invariance

Supporting Information may be found with the online version of this article.

Please note: Wiley-Blackwell is not responsible for the content or func-


tionality of any supporting information supplied by the authors. Any
queries (other than missing material) should be directed to the correspond-
ing author for the article.

You might also like