Download Exploring Self-Control, Social Bonds, and Crime: A Sociological View and more Study Guides, Projects, Research Criminology in PDF only on Docsity!
LOW SELF-CONTROL, SOCIAL BONDS, AND
CRIME: SOCIAL CAUSATION, SOCIAL
SELECTION, OR BOTH?*
BRADLEY R. ENTNER WRIGHT
University of Connecticut
AVSHALOM CASPI
TERFUE E. MOFFIIT
Institute of Psychiatry, University of London and University of
Wisconsin-Madison
PHIL A. SILVA
University of Otago Medical School, New Zealand
This article examines the social-selection and social-causation
processes that generate criminal behavior. We describe these processes
with three theoretical models: a social-causation model that links crime
to contemporaneous social relationships; a social-selection model that
links crime to personal characteristics formed in childhood; and a
mixed selection-causation model that links crime to social relationships
and childhood characteristics. We tested these models with a longitudi-
nal study in Dunedin, New Zealand, of individuals followed from birth
through age 21. We analyzed measures of childhood and adolescent
low self-control as well as adolescent and adult social bonds and crimi-
nal behavior. I n support of social selection, we found that low self-
control in childhood predicted disrupted social bonds and criminal
offending later in life. In support of social causation, we found that
social bonds and adolescent delinquency predicted later adult crime
and, further. that the effect of self-control on crime was largely medi-
ated by social bonds. In support of both selection and causation, we
* This research was supported by a fellowship given to B. Wright by the National
Science Foundation National Consortium on Violence Research, and by grants from the National Institute of Mental Health (MH49414 and MH45070), the William T. Grant Foundation, the William Freeman Vilas Trust at the University of Wisconsin, and the Medical Research Council of the United Kingdom. NCOVR is headquartered at Carnegie Mellon University and receives support from the National Science Foundation in partnership with the National Institute of Justice and the U.S. Department of Housing and Urban Development. The Dunedin Multidisciplinary Health and Development Research Unit is supported by the New Zealand Health Research Council. We are grateful to the Dunedin Unit investigators and staff and to the study members and their families. We thank Ray Paternoster, Robert Brame, and Shawn Bushway for helpful discussions about this article. We thank HonaLee Harrington for research assistance.
CRIMINOLOGY VOLUME 37 NUMBER 3 1999 479
480 WRIGHT ET AL.
found that the social-causation effects remained significant even when
controlling for preexisting levels of self-control, but that their effects
diminished. Taken together. these firtdirigs support theoretical models
that incorporate social-selection and social-causation processes.
Criminological theories, especially those from the sociological perspec- tive, have traditionally explained crime in terms of social causation-that social relationships promote or prevent criminal behavior. As supportive evidence they reference the frequently observed correlations of crime with school, work, family ties, peer delinquency, and prior delinquency. Recently, however, several prominent sociologists have incorporated childhood characteristics-such as low self-control and childhood antiso- cial behavior-into their theories of crime (Gottfredson and Hirschi, 1990; Sampson and Laub, 1990, 1993). This approach accepts the correlations between crime and social relationships, but explains them as being, at least in part, the outcome of social selection-that preexisting individual char- acteristics influence the development of social relationships and criminal
behavior, and so the observed correlations between crime and social rela-
tionships may be spurious. This article examines selection and causation issues in light of two ques-
tions: Do social relationships cause crime? Does prior delinquency cause
later crime? Specifically, we test to what extent, if any, the correlations of
crime with social bonds and of crime with prior delinquency attenuate when controlling for levels of childhood self-control. Social-causation
models, in their pure form, predict no attenuation: social-selection models,
in their pure form, predict complete attenuation: and mixed selection-cau- sation models predict partial attenuation. We also test to what extent the effect of self-control on crime is mediated by social bonds. While previous studies have explored these issues, this article presents a direct and com-
pelling test of these questions by (a) employing measures of self-control
collected in childhood from multiple sources using multiple measurement instruments and (b) following individuals over time in order to measure social relationships and criminal participation in adulthood.
INTERPRETING THE CORRELATION BETWEEN SOCIAL BONDS AND CRIME
In this section we examine the causal linkage between social bonds and crime from the perspective of models of social causation, social selection, and mixed social selection/causation. We discuss the implications of these models, and we review previous empirical studies of them.
482 WRIGHT ET AL.
fit poorly into conventional society, and so they end up in weakened or
broken social relationships (Gottfredson and Hirschi, 1990157). According to self-control theory, since criminal behavior and social bonds share their origins in childhood self-control, the correlation between them is causally spurious. After childhood ends, social relationships with peers, school, work, family, and marriage have no causal impact on crimi- nal behavior. Gottfredson and Hirschi (1990) are clear on this controver- sial point: “Lack of perseverance in school, in a job, or in an interpersonal relationship is simply different manifestations of the personal factors assumed to cause crime in the first place. Taking up with delinquent peers
is another example of an event withour causal significance” (p. 251, italics
added; see also pp. 154-168). Several theories of crime incorporate processes of social selection and social causation. Notably, the theory of age-graded informal social control
links childhood antisocial behavior to adult crime through two causal
mechanisms (Sampson and Laub, 1990:609-612. 1993:123-138, 1995:145-148). First, in childhood individuals develop an underlying crim- inal propensity that expresses itself as antisocial behavior and carries on into adulthood as criminal behavior. Second, childhood antisocial behav- ior disrupts the formation of later social bonds (i.e., social selection). Nonetheless, these social bonds are not determined fully by childhood characteristics, and they have unique, causal effects on adult crime independent of individuals’ preexisting characteristics (i.e., social causa- tion). In other words, social bonds to school, work, and family in part reflect preexisting criminal propensity and in part cause crime. Developmental theories of crime offer a similar mix of social selection and causation, albeit for separate groups of people (Moffitt, 1993; Moffitt et al., 1996; Patterson and Yoerger, 1993; Patterson et al., 1989). In Mof- fitt’s theory, neuopsychological impairments in childhood can extend into adulthood and cause criminal behavior. These impairments can also ensnare individuals in delinquent friendships, broken family ties, failed schooling, and unemployment, and these social factors in turn cause crime. Individuals without impairments enter crime through a different pathway. They reach biological maturity-puberty-well before they reach social maturity-adult statuses such as a driver’s license, the right to buy alcohol, and marriage. Caught in this maturity gap, they will observe that their delinquent peers have already acquired the “forbidden” resources and
privileges that they desire, and so, by means of rationally motivated social
mimicry, they turn to delinquency. Patterson’s theory makes a similar case with “early starters,” who exhibit childhood problem behaviors such as aggression and temper tantrums, and “later starters,” who in adolescence
experience decreased parental supervision and peer delinquency. Thus,
SELF-CONTROL AND SOCIAL BONDS 483
the theories of Sampson and Laub, Moffitt, and Patterson share a common emphasis on including social selection and social causation.
IMPLICATIONS AND PREVIOUS EVIDENCE
Wide-ranging implications arise from the possibility that selection processes alone generate the observed correlation between social bonds and criminal behavior. From a conceptual perspective, this spurious corre-
lation undermines traditional sociological theories, and so if selection the-
ories are to be accepted, causation theories are to be abandoned (Evans et al., 1997:479). Also, selection theories are highly parsimonious in that they explain crime with primarily one theoretical construct-low self-control or criminal propensity. In contrast, traditional social-causation theories ref- erence multifaceted, multilevel social relationships, and social-selection/ causation theories incorporate both propensity and social relationships. Therefore, all else being equal, the parsimony of social-selection theories makes them the favored explanation of crime (Paternoster et al., 1997).
From a methodological perspective, if childhood characteristics deter-
mine crime and if these characteristics persist into adulthood (i.e., keep the same ranking relative to the self-control levels of other individuals), there is no need to assess them over time with longitudinal studies. “Iden- tification of the causes of crime at one age may suffice to identify them at
other ages as well-if so, cohort or longitudinal studies of crime are
unnecessary” (Hirschi and Gottfredson, 1995131). Further, if postchild- hood social relationships have no causal effect on crime, they too need not be assessed at all. “Since such ‘events’ [such as school, work, interpersonal relationships, and delinquent peers] are predictable consequences of the
causes of crime, there is little point in monitoring them” (Gottfredson and
Hirschi, 1990:251). In the end, selection theories require only that researchers collect cross-sectional studies of enduring individual characteristics.
From a policy perspective, if the causes of crime are set in childhood and
if they are immune to later social influence, social programs that attempt
to rehabilitate criminal offenders are misguided and unworkable. “Clearly, the general thrust of the public policy implications of our theory is to counter the prevailing view that modifications of the criminal justice system hold promise for major reductions in criminal activities” (Gottfred- son and Hirschi, 1990:255). Also, “[self-control] accounts for the failure of efforts to treat delinquents or to deter them by the threat of punishment” (Hirschi and Gottfredson, 1995:140). Consequently, public policy should attempt not to rehabilitate current offenders, but only to prevent future offenders by cultivating self-control in childhood, a developmental period when psychological characteristics are assumed to be still malleable.
SELF-CONTROL AND SOCIAL BONDS 485
THEORETICAL MODELS
Does previous delinquency cause later crime? Various social-causation
theories answer yes (Nagin and Paternoster, 1991:166). For example, prior
criminal involvement might weaken conventional social bonds, thereby damaging those relationships that once helped deter criminal behavior
(Agnew, 1985; Hirschi, 1969). Likewise, criminal acts, and the formal
sanctions that they might solicit, can give individuals greater exposure to and affinity for other law violators. Through these processes of reinforce- ment and modeling, criminal acts can thus increase criminal participation
(Akers, 1985; Sutherland, 1947).
In contrast, self-control theory casts the correlation between delin- quency and crime over time as spurious. Criminal behavior at any age is determined by childhood self-control, which is stable by adolescence, and therefore delinquent or criminal acts at one time have no causal bearing
on those acts at another time (Sampson and Laub, 1995147).
The theory of age-graded informal social control explains criminal sta- bility as resulting from both social selection and social causation (Sampson
and Laub, 1995:147-148). As social selection, differences in criminal pro-
pensity span from childhood into adulthood (Caspi et al., 1996a). As
social causation, delinquent behavior at one time can jeopardize social
relationships, such as at work and in marriage. These, in turn, diminish
individuals’ life chances, which leads to criminal behavior. Moffitt’s (1993;
Moffitt et al., 1996) theory posits a similar, life-course persistent trajectory
in which delinquent behavior produces later criminal behavior via processes of contemporary continuity (i.e., enduring neuropsychological impairments) and also via processes of cumulative continuity (i.e., social bonds are disrupted in the course of growing from child to adult).
IMPLICATIONS AND PREVIOUS EVIDENCE As with the social bond-crime correlation, the possibility of spurious correlation between repeated measures of criminal behavior over time has wide-ranging implications. Conceptually, social-selection models are again the more parsimonious for the same reason-they explain criminal stability with one theoretical construct. Methodologically, selection mod- els again question the need for longitudinal studies. If prior criminal behavior does not affect later criminal behavior, why record both? For policy, selection models discount the existence of secondary deviance
(Becker, 1963), and so, if they are correct, policymakers need not worry
about social sanctions having latent negative effects. Various studies have tested for individuals’ underlying heterogeneity in criminal tendencies, and here we reference several recently published examples. Each of these studies used statistical methods, such as random-
486 WRIGHT ET AL.
effect models, to model heterogeneity, but nonetheless, they found con- trary results. As evidence of social causation, Nagin and Paternoster (1991) analyzed a longitudinal study of high school sophomores. They modeled unobserved propensity for crime with self-reported measures of theft, and they found that previous offending had a positive causal effect on later offending. Paternoster and Brame (1997) modeled criminal pro- pensity with measures of general delinquent activity, and they too found previous delinquency to have a causal effect. As evidence of social selec- tion, Nagin and Farrington (1992) analyzed a longitudinal study of S-year- old working-class boys. They modeled unobserved criminal propensity
with measures of criminal convictions between ages 10 and 31. They
found that the positive association between earlier and later criminal behavior was “largely attributable to stable, unmeasured individual differ-
ences” (p. 235). As evidence of both social causation and social selection,
Paternoster et al. (1997) studied releasees from youth training schools, and they found evidence of both change and continuity in criminal offending and that the change could not be attributed solely to processes of self- selection. The prior evidence on this issue supports both social-selection and social-causation processes, implying again that pure selection or causal models may not be workable as comprehensive theories.
A TEST OF SOCIAL SELECTION AND SOCIAL CAUSATION
Missing from previous studies is perhaps a more direct test of social-
selection versus social-causation models. This test would observe in child-
hood those personal characteristics, such as low self-control, thought by
selection theories to spawn later criminal behavior. It would then obseilre
in adolescence or adulthood those social bonds and delinquent behaviors
thought by causation theories to bring about crime. It would then test how
much, if at all, childhood characteristics diminish the associations between social bonds and crime and between early delinquency and later crime. For such a study, social-causation theories predict no attenuation: social- selection theories predict complete attenuation: and social-selectiodcausa- tion theories predict partial attenuation. The need for this more direct test has been recognized by proponents of both selection and causation theories. Gottfredson and Hirschi (1990) wrote that it is difficult to assess “criminal tendencies independent of opportunity to commit criminal acts.” One solution, they proposed, is that “tendencies may be assessed before crime is possible: that is, the measure
of criminality, i.e., propensity, is constructed from information available in
488 WRIGHT ET AL.
(Junger-Tas et al., 1994; van Dijk and Mayhew, 1992). Our own replica- tion studies suggest that the predictors of problem behavior are the same between the Dunedin study and a similar longitudinal sample of black and white youth collected in Pittsburgh (Moffitt et al., 1995). The Dunedin study has collected multiple measures from multiple sources about study members' levels of self-control, social bonds, and
criminal behavior. We describe these variables in Appendix I, presenting
each variable's content, the age at which it was collected, from whom it was collected, univariate statistical descriptions, and citations of previous published studies that have used the measurement instrument and provide information about its reliability and validity.
SELF-CONTROL MEASURES
The self-control variables measured in childhood include Lack of Con-
trol-Irritability-Distractibility, Impulsivity I, Impulsivity 11, Lack of Persis-
tence, Inattention I, Hyperactivity I, Hyperactivity 11, Hyperactivity 111.
and Antisocial Behavior. These nine variables comprise over 150 separate test items in the Dunedin study. These items were collected from eight sources-study members, parents, two trained observers, and four teach-
ers-at five ages-ages 3, 5. 7, 9, and 11.
The self-control variables measured in adolescence include Impulsivity
111, Impulsivity JY, Hyperactivity IV, Inattention 11, Inattention 111, Physi-
cal Response to Conflict, and Risk Taking. These seven variables com-
prise over 50 separate test items from three sources-study members,
parents, and informants-at two ages-ages 15 and 18. The measurement of self-control has been a contentious issue in previ- ous studies (e.g., Hirschi and Gottfredson, 1993; Longshore et al., 1998;
Piquero and Rosay, 1998), so we now discuss at length the self-control
measures collected in the Dunedin study.
To begin with, it should be recognized that self-control itself is not diffi-
cult to measure, but rather, it simply has not been measured often in the data sets most often used by criminologists. This is because the conceptual importance of childhood characteristics for theories of crime is just now being widely accepted. In contrast, developmental psychologists have long studied and measured childhood antisocial behavior and low self-control,
but they have rarely collected data on later criminal activity and social ties.
In this context, the Dunedin study offers a unique opportunity, for it has involved a multidisciplinary team of psychologists, criminologists, and sociologists who have included age-appropriate measures of self-control, crime, and social ties in a longitudinal study spanning from childhood to young adulthood. It is significant that the Dunedin study measured self-control during
SELF-CONTROL AND SOCIAL BONDS 489
childhood, for self-control theory claims that self-control is best measured in childhood. This is because external opportunities in adolescence can alter expressions of later low self-control: therefore, self-control is most
clearly assessed before there exists an abundance of opportunities for
criminal behavior (i.e., in childhood). Additionally, when studies concur- rently measure self-control, delinquency, and social bonds, as is the case with cross-sectional studies, temporal ordering is lost and the resulting findings are causally ambiguous and difficult to interpret. Any empirical relationship observed between self-control and delinquency or social bonds is open to interpretations of reciprocal or reverse causation (Evans et al., 1997:493). The self-control measures in the Dunedin study fit squarely within Gott- fredson and Hirschi’s specification of self-control. They include measures of impulsivity, a lack of persistence, high activity levels, risk taking, and responding to conflict physically. One measure, Antisocial Behavior, in particular fits Hirschi and Gottfredson’s (1993) prescription for measuring
self-control in childhood: “The question becomes, can independent
indicators of self-control be identified? With respect to crime, we have
proposed such items as whining, pushing, and shoving (as a child)” (p. 53).
Polakowski (1994) adds empirical evidence here, finding that self-control “is significantly comprised by early behavioral indicators of aggression and fighting” (p. 41). Almost these exact behaviors are measured in the Rutter Hyperactivity and Antisocial Behavior scales, which as described in Appendix 1, measure fussiness, fighting, and bullying. They were col-
lected in the Dunedin study at ages 5,7,9, and 11, from both parents and
teachers. The measures of self-control are highly intercorrelated, both within and across developmental periods. The intercorrelations between the 16 self- control variables are presented in Appendix 2, and to summarize, of the
120 correlations presented, all but 6 are statistically significant. This inter-
correlation is specified in self-control theory, which holds that the traits that compose self-control “come together in the same person, and... tend to persist through life” (Gottfredson and Hirschi, 199091). The reliability of the 9 childhood self-control measures is alpha = 36, and the reliability of the 7 adolescent measures is alpha = .64. These compare favorably with those found in other studies. For example, the self-control measures used by Evans et al. (1997) had a reliability of .61. In addition to their conceptually relevant content, the self-control meas- ures in the Dunedin study come from multiple sources, including the study members themselves, parents, other family members, friends, teachers,
and trained observers. To convey the importance of having data from
multiple sources, we must consider several possible types of self-control measures. Two distinctions are most relevant here. Data can be self-
SELF-CONTROL AND SOCIAL BONDS 491
observed correlation between them and delinquency is causally spurious and will disappear once levels of prior self-control are taken into account (Gottfredson and Hirschi, 1990154-168). Therefore, the relationship between these social bonds and crime, net of self-control, provides “a cru-
cial test of [self-control] theory vis-a-vis the standard theories of positive
criminology” (Gottfredson and Hirschi, 1990:167). Measures of these bonds include Friends Are Delinquent, Companion For Delinquency, Friends Are Good Citizens, Educational Aspirations, Months Education, Educational Achievement, Did Not Like-Left School, Months Unemploy- ment, Months Full-Time Employment, Occupational Aspirations, Job Desirability, Living with Parents, Involvement with Parents, Intimacy with Parents, Intimacy with Partner, and Companionship with Partner. These 16 variables comprise 46 self-reported interview items collected at ages 18 and 21.
DELINQUENCY AND CRIME MEASURES
A fourth, and last, category of variables are measures of delinquency and crime. These include Delinquency at Age 15 and Crime at Age 21. Both of these measures are self-reported variety scores that record how many types of illegal acts each study member committed at least once in
the previous year. They are scored from 0 to 29 and 0 to 48 illegal acts,
respectively. Variety scores such as these are endorsed by proponents of
self-control theory. “Indeed, it appears that the best available operational
measure of the propensity to offend is a count of the number of distinct
problem behaviors engaged in by a youth (that is, a variety scale)” (Hir- schi and Gottfredson, 1995:134). Variety scales do not, however, incorpo- rate the frequency of committed criminal acts, a difficult concept to operationalie since its meaning varies by the severity of the act (e.g., 10
drunken drivings versus 10 homicides). (The distribution of the age-
crime measure was skewed somewhat to the right. To test the robustness of our ordinary least squares analyses, we reran our analyses using a trans- formed measure of age-21 crime, obtained by taking its natural logarithm, and the pattern of findings did not change.) In addition to the substantive variables described above, we also used measures of gender and social class as control variables. Gender is a dummy variable coded 1 = male. It correlated with age-21 crime at r = .32, which was statistically significant. The social-class measure averages the socioeconomic status of study members’ families across the first 15 years
of the study (Wright et al., 1999). It correlated with age-21 crime at r = -
.02, which was not statistically significant. By and large there were not many missing data in these measures, with one main exception. The Diagnostic Interview Schedule for Children (DISC), collected at age 11, had about 20% missing cases. This was not a
492 WRIGHT ET AL.
great concern because each of the DISC measures was replicated at other ages with variables containing fewer missing data. For any variable that had missing data, in the regression equations, we created a dummy varia- ble indicating which cases were missing. We then recoded the substantive variable to its mean and included both it and the missing dummy variable in the regression equation (Little and Rubin, 1987).
RESULTS
In this section, we present four sets of analyses that test the social-selec- tion and the social-causation processes of crime.
SOCIAL-SELECTION PROCESSES
Our first set of analyses examined if low self-control correlated with later criminal behavior and disrupted social bonds, as per the social-selec-
tion model, and it did. Table 1 presents the correlations between self-con-
trol and criminal behavior. All 16 of the childhood and adolescent self- control variables, with two exceptions, significantly correlated (p I .05)
with delinquency at age 15 and crime at age 21. The significant correla-
tions ranged from r = .06 to r = .46. To summarize these correlations, we created two summary scales that comprised the self-control variables. One scale averaged the standardized scores of the nine childhood self-con- trol variables. The other averaged the seven adolescent self-control vari- ables. The two summary scales both significantly correlated with the measures of criminal behavior. Childhood self-control correlated at r = .23 with age-15 delinquency and r = .21 with age-21 crime, and adolescent self-control correlated at r = .41 with delinquency and r = .45 with crime. We note that the correlation between adolescent self-control and crime ( r
= .45) slightly exceeded that found between delinquency and crime ( r =
The correlations reported in Table 1 compare favorably in size to those found in previous studies that have related self-control to various foims of
deviance. Such correlations include r = .36 and r = .40 (Gibbs and Giever,
1995:251), r = .32 and r = .34 (Piquero and Tibbetts, 1996:496), r = .15 to
.30 (Wood et al., 1993:119),p = .32 (Burton et al., 1994:228),p = .30 (Evans et al., 1997:489), and p = .18 (Ameklev et al., 1993:234). Importantly, these previous correlations were produced in cross-sectional studies, and yet they did not exceed those found in the Dunedin study across a span of several years ( r = .41 and .45), and in fact, some were even less than found here across a decade ( r = .23 and .21). These correlations in Table 1 give
additional confidence in our measures of self-control, for a measure of
self-control can be “validated by its ability to predict subsequent behav- ior” (Gottfredson and Hirschi, 1990220).
494 WRIGHT ET AL.
Table 2. Correlations Between Self-Control and Social
Bonds
Childhood Adolescent Self-Control Scale Self-Control Scale
Peer Group Friends Are Delinquent (Self-Report, Ages 18. 21) Companion for Delinquency (Self-Report. Age 21) Friends Are Good Citizens (Self-Report, Ages **18,** 21) School Educational Aspirations (Self-Report. Ages **15.** 18) Months Education (Self-Report, Age 21) Educational Achievement (Self-Report, Ages 18, 21) Did Not Like, Left School (Self-Report, Age 18) Months Unemployment (Self-Report, Age 21) Months Full-Time Employment (Self-Report, Age 21) Occupational Aspirations (Self-Report, Ages 18, 21) Job Desirability (Self-Report, Age 21) Job Family Living with Parents (Self-Report, Age 21) Involvement with Parents (Self-Report, Age 18) Intimacy with Parents (Self-Report, Age 18) Partner Intimacy with Partner Companionship with Partner (Self-Report, Age 21) (Self-Report, Age 21) **.29*** _-._ - 77s **_-. 06_** - 12" -.11* -.08* .33* .-I^ **_35"_** -.38* -.18* - 39* -.30* **.17*** .21* **_-._** - .*75* -.11* - **_75*_** - .- **_75*_** **.I** -.17* -.14* **NOTES:** N = 773-778 for partner variables, N = 920-1.012 for remaining variables. Cells present simple correlation coefficients. Data from the Dunedin (New Zealand) Multidisciplinary Health and Development Study. **_* p_** < _.05_ (two-tailed tests). ### SELF-CONTROL AND SOCIAL BONDS 495 #### Table 2 presents the correlations between self-control and social bonds. The two self-control summary scales significantly correlated with nearly eveiy measure of social bonds. Study members with low self-control had more delinquent peers, diminished bonds to school, lesser work achieve- ments, and weakened family and partner ties. The significant correlations #### between self-control and social bonds ranged from r = .07 to r = .39, mag- nitudes consistent with those found in previous studies (Evans et al., #### 1997:490; Sampson and Laub, 1990615). Taken together, the correlations #### presented in Tables 1 and 2 are evidence of social-selection processes in #### the generation of criminal behavior, for low self-control prospectively predicts disrupted social bonds and criminal behavior. Whether these selection effects are strong enough to preempt social causation is the criti- cal question, and we take it up in the following analyses. #### SOCIAL-CAUSATION PROCESSES Our next set of analyses examined if social bonds correlated with crimi- nal behavior, as per the social-causation model, and we found that they #### did. The first column in Table 3 presents the correlations between social bonds and delinquency. To facilitate the remaining analyses, we present these correlations as standardized regression coefficients estimated by #### regressing crime at age 21 on each social bond measure separately plus #### gender and social class (for a total of 16 regression equations). As shown in column one, the social bond variables correlated with crime at levels ranging from **p** = **.129** (with educational aspirations) to p = **.479** (compan- ion for delinquency), and all were statistically significant. The statistical significance of these correlations is important for the pur- poses of this study. Both selection and causation models assume that social bonds correlate with crime; they simply differ in their causal inter- pretation of these correlations. These countering interpretations can be distinguished empirically only if social bonds and crime are first corre- lated. Put differently, a spurious correlation must first be a correlation. A lack of initial correlation between social bonds and crime would argue equally strongly against both models. We make this point because it iden- #### tifies a problem found in previous research. Evans et al. (1997) adopted a #### similar analytic strategy to the one we use in this article. They started with #### 13 social bond measures, but only 3 of them were significantly associated #### with crime before they controlled for self-control (Table 3, p. 492). n u s , it #### is not clear how well the data used by Evans et al. (1997) could distinguish selection versus causation models. #### SOCIAL CAUSATION NET OF SOCIAL SELECTION Our third set of analyses tested the strength of the social-causation Table 3. Regression of Crime at Age 21 on Social Bonds and Prior Delinquency, With and Without Controls for Self-Control ~ Dependent Variable^ =^ Crime at Age^21 _( 5 )_ Amount of Coefficient, Attenuation (4) **_( 3 )_** Partial Regression Amount of Coefficient, Attenuation Controlling for Introduced by Coefficient, Controlling for Introduced by Childhood and Childhood and (2) Partial Regression Independent Variables = Not Controlling Childhood Childhood Adolescent Adolescent Social Bonds and Delinouencv for Self-Control Self-Control Self-Control Self-Con trol Self-Control (1) Full Regression Delinquent Peer Group Friends Are Delinquent .464 (16.6)" .470 (16.8)" -1% .387 (13.4)" 17% Companion for Delinquency ,479 (17.7)* .481 (17.9)" 0 **_Yo_** ,413 (15.6):i. 14% (Self-Report, Ages 18, 21) (Self-Report, Age 21) (Self-Report, Ages 18, 21) Friends Are Good Citizens -.432 (-15.2)" -.424^ (-14.5)^2^96 -.338^ (-11.3)*^ 22% Educational AsDirations -.129 (-4.08)" -.I **17** (-3.66)* 9 ?k -.OM (-2.78):g **35%** School (Self-Repoit,'Age 18) (Self-Report. Age 21) Months Education -.236 (-7.19)* -.232 (-6.79)* 2 _Yo_ -.I47 (-4.29)* 3896 Educational Achievement -.208 (-6.31)* -.I99 (-5.67)* 4 **yo** -.126 (-3.64)' 39% (Self-Report, Ages 18, 21) (Self-Report, Age 18) Did Not Like, Left School ,177 (5.84)* .169 **_(5.55):'_** 5% .I00 (3.41)" 44 _Yo_ Job Months Unemployment .I75 (5.66)*^ **,155**^ (4.78)"^ 11%^ .I09^ (3.60)"^ 38% Months Full-Time Employment -.I29 (-4.21)* **-.I15** (-3.73)* 11% -.099 (-3.45)" 23 **70** Occupational Aspirations -.I93^ (-6.16)"^ -.I87^ (-5.85)*^^3 %^ **-.I35**^ (-4.46)"^ 30% Job Desirability -.156 (-5.15)'' -.147 (-4.86)" **_6 Yo_** -.114 (-4.06)" 27 % (Self-Report, Age 21) (Self-Report, Age 21) (Self-Report, Ages 18, 21) (Self-Report, Age 21) ## K r 7 r 3 Z U F r W ## v, 3 **Table 3.** (continued) (^) % _00_ Dependent Variable = Crime at Age 21 **(4)** _( 5 )_ Amount of Coefficient, Attenuation _( 3_ **1** Partial Regression Amount of Coefficient , Attenuation Controlling for Introduced by Coefficient, Controlling for Introduced by Childhood and Childhood and (2) **( 1 )** Partial Regression Full Regression Independent Variables = Not Controlling Childhood Childhood Adolescent Adolescent Social Bonds and Delinquency for Self-Control Self-Con trol Self-Con trol Self-Con trol Self-Con trol Family Living with Parents (Self-Report, Age 21) Involvement with Parents (Self-Report, Age 18) Intimacy with Parents (Self-Report, Age 18) Partner Intimacy with Partner Companionship with Partnei (Self-Report, Age 21) (Self-Report. Age 21) (Self-Report. Age 15) Delinquency Delinquency _-.320_ (-11.0)" -.312 (-10.6)" 2% -.227 (-7.79)" 29% -.I49 (-4.91)* -.138 (4.52)" 7 _Yo_ **-.055** (-1.87):'. 63% < -.153 _(-5.02):'_ -.140 (4.60)" 8 **_Yo_** -.073 (-2.52):' _5296_ **_E_** -220 **(-6.69)*** -.201 (-6.08)" 9 % -.157 (-5.07)" 2996 _8_ m #### -.129 (-3.86)''. -.I 12 (-3.33):'. 1306 -.081 (-2.60):' 37 96 4 ### k **,388 (13.7)"** _,375_ (12.9)" **_3 40_** _,285_ (8.90)* -77% r ~~~ ~ ~ ~ ~ **NOTES:** N = 776 for partner variables. N = 956 for remaining variables. Cells present standardized OLS regression coefficients with t-values in parentheses. Data from the Dunedin (New Zealand) Multidisciplinary Health and Development Study. **All** equations control for gender and social class. Dummy variables used in equations to control for missing cases. **_* p_** < _.05_ (two-tailed tests).
