Negative Cases in the Nexus Between Self-Control,Social Bonds,and Delinquency

The General Theory of Crime and the Spuriousness Thesis

According to Gottfredson and Hirschi (1990), self-control should emerge in the first decade of life when parents monitor child behavior and recognize and discipline deviance when it occurs. This teaches children that deviance has consequences and that they must self-regulate. Alternatively, when parents fail to do these things, children are placed in a precarious position—they recognize the immediate pleasures associated with crime and deviance, but do not prioritize the negative consequences of such acts. Thus, they should have low self-control and engage in many acts—criminal or noncriminal—that carry short- and long-term consequences.

These arguments have largely been empirically supported. Regarding the causes of self-control, parental socialization is a significant predictor, even in studies that account for key sources of spuriousness, such as the child’s earlier problem behavior (Gibson, Sullivan, Jones, & Piquero, 2010; Hay & Forrest, 2006). With respect to its consequences, low self-control is one of the most reliable predictors of crime and deviance (Pratt & Cullen, 2000), and this finding has emerged across different types of self-control measures (Piquero, 2008; Piquero & Bouffard, 2007). Self-control also affects various different forms of crime and deviance (Junger & Tremblay, 1999).

This support has led many scholars to consider ancillary arguments of the theory. This includes the spuriousness thesis, which holds that many variables thought to cause crime—including social bonds like attachment to others and commitment to conventional goals—are spurious correlates of offending. From this perspective, they are correlated with one another and with crime because all are the result of self-control. This position is revealed in various passages of A General Theory of Crime in which Gottfredson and Hirschi critique the findings from prior research. Key to their perspective is the idea that social bonding is highly reliant upon self-control—those low in self-control “do not tend to make good friends” (p. 157), “tend to have unstable marriages” (p. 89), and have difficulty “satisfying the academic. . . requirements of school” (p. 163). These aspects of bonding involve long-term commitments that require diligence, thoughtfulness, and persistence, and these are the very qualities missing in those with low self-control: “The traits composing low self-control are not conducive to the achievement of long-term individual goals” (p. 96).

From this perspective, those with high self-control self-select into circumstances of strong social bonding, those with low self-control do the opposite, and the correlations between bonds and crime largely are spurious because each is a function of self-control. This perspective is captured in the assertion that “many of the traditional causes of crime are in fact consequences of low self-control—that is, people sort themselves and are sorted into a variety of circumstances that are as a result correlated with crime” (Gottfredson & Hirschi, 1990, p. 119). Similarly, they argue that difficulties in the areas of education, employment, or interpersonal relationships are “without causal significance” because they “are simply different manifestations of the personal factors assumed to cause crime in the first place [low self-control]” (p. 251).

These arguments reflect a specific testable hypothesis: The effect of social bonds on crime should be reduced or should disappear once self-control is introduced as a control variable. Several studies have examined this hypothesis directly. For example, Wright and his colleagues (1999) analyzed data from a New Zealand birth cohort to provide a long-term longitudinal test with a diverse sample. They found that controlling for comprehensive, multi-method measures of self-control during childhood (ages 3–11) and adolescence (ages 15–18) reduced the effects of social bonds on young adult crime by roughly 25%. However, the effects of social bonds involving the family, school, and peers often remained significant. Similarly, in an analysis of adolescent convicted offenders, Longshore, Chang, and Messina (2005) found that when controlling for prior self-control (as measured by the Grasmick et al. scale), family attachment significantly affected self-reported offending 6 months later. In contrast, in a cross-sectional analysis with a representative sample of adults in a Midwestern city, Evans and his colleagues (1997) found that social bonds had little effect on crime whether self-control was accounted for or not; self-control, on the other hand, had relatively large effects on crime.

Several other studies have examined this hypothesis indirectly—in examining other issues, they estimated equations in which crime was the dependent variable and the list of predictors included self-control and social bonds. These studies revealed bonds–crime relationships that often persist even when controlling for self-control. This is the case when assessing family bonds among adolescents (Burt, Simons, & Simons, 2006; Felson & Staff, 2006; Hay & Forrest, 2008), and Burt and her colleagues found this to be true as well for school bonds. Additionally, Doherty (2006) found that a composite scale of adult social bonding (including military service, a cohesive marriage, and stable employment) had a persistent effect on adult offending. Another line of research that indirectly assesses the spuriousness thesis comes from examinations of social bonds that control for persistent unobserved heterogeneity (which includes differences in self-control). These studies tend to find that changes in life circumstances (including changes in social bonds) significantly affect changes in crime (Horney, Osgood, & Marshall, 1995; Piquero, Brame, Mazerolle, & Haapanen, 2002; Sampson, Laub, & Wimer, 2006).

The research above suggests that the spuriousness thesis may be overstated, as bonds–crime associations often persist when accounting for various different measures of self-control. We should emphasize, however, that this conclusion does not invalidate the general theory of crime. Virtually all of the tests above confirm an important effect of self-control. Moreover, any suggestion that self-control is the only true cause of crime (Gottfredson & Hirschi, 1990, p. 232) should not be prioritized in assessments of the general theory of crime—no criminological theory could fare well under such a standard.

A persisting effect of social bonds does, however, point to the presence of negative cases that depart from the theory’s logic and that can inform the theory. Most notably, there may be cases in which social bonds maintain an independent link to later crime because their development is not as reliant upon self-control as we may expect. Some individuals may develop crime-reducing social bonds even without having the expected level of self-control, whereas others individuals with high self-control may struggle to develop strong bonds. These possibilities have not been considered because no previous empirical test of the general theory of crime has examined the interplay of self- and social control within a negative case perspective. Thus, little is known about the key qualities of cases that deviate from the expected theoretical pattern. As we discuss below, an examination of these negative cases can yield insight into key ways to advance the general theory of crime and the broader work in this area.

The Present Study

The purpose of this study is to assess the negative cases that are implied by prior tests of the spuriousness thesis. The independent effects of social bonds point to the presence of negative cases in which self-control does not translate into the expected level of social bonding. We see three questions that, if answered, can yield a more nuanced understanding of the nexus between self-control, social bonds, and crime. First, how common are cases in which self-control does not produce the expected level of social bonding? ¹ Do these cases represent a minority of less than 10% of all cases, a strong majority (e.g., greater than 60%), or something in between? Second, and related, what form do negative cases take? Individuals marked by incongruence between self-control and social bonding logically can be divided into two types: (1) Those with high self-control who, for some reason, do not develop strong social bonds and (2) those with low self-control who unexpectedly develop strong bonds. Our question is this: Are both types well represented? If the first type is common, the clear implication is that having high self-control—with its advantages regarding forethought, attentiveness, and determination—is not enough to produce strong social bonding. Other unknown factors can come into play to block the expected benefits of self-control. If the latter type also is common, the implication is that perhaps through good fortune or the efforts of others, a person can develop strong bonds even when they lack the “diligence, tenacity, and persistence” (Gottfredson & Hirschi, 1990, p. 89) thought to be preconditions for effective bonding.

Our third research question focuses on how delinquency varies across these cases in which there is a divergence between one’s self-control and social bonding. Specifically, when the level of self-control pulls an individual in one direction, and social bonds are pulling him or her in the opposite direction, which source of control holds sway? Gottfredson and Hirschi’s theory sees such incongruence as unlikely, but the spirit of the theory suggests that if it were to occur, self-control should be most consequential. This is captured by Gottfredson and Hirschi’s (1990, p. 87) view that those with high self-control “avoid criminal acts whatever the circumstances in which they find themselves.”

A different but not necessarily contradictory position can be found in Sampson and Laub’s (1993) age-graded social control theory. This perspective is quite open to the general theory’s view that latent traits such as low self-control lead individuals to self-select into statuses and relationships that increase offending. They also emphasize, however, that incongruence between self-control and social bonds will sometimes emerge—individuals with a history of antisociality can experience unexpected increases in bonds that initiate a crime-free trajectory. Improvements in bonding can come from good fortune (“there is an element of luck, randomness, or chance operating throughout the course of life,” Laub & Sampson, 1993, p. 317) or from the efforts of others (such as parents or teachers) whose investments in a struggling child “trigger a return investment” from the child (Laub & Sampson, 1993, p. 311). In either case, the improved bonding could override the earlier established criminal propensity (e.g., low self-control), and thus lead to lower than expected involvement in crime.

Our analysis addresses these three questions, and in so doing, provides the first investigation to date of the negative cases that deviate from the expected logic of the spuriousness thesis. Thus, although prior research suggests that negative cases exist, little is known about how common they are, what exact forms they take, and how delinquency varies across different types of negative cases. Our approach of identifying the distribution of these negative cases and showing their association with delinquency offers an opportunity to extend research on the general theory of crime and direct attention to potential areas for refinement. As others have noted (Giordano, 1989; Sullivan, 2011), this approach is an essential tool for theoretical advancement. In Lindesmith’s (1968, p. 20) words, “The progressive refinement of theory that is brought about by the necessity of taking negative instances seriously . . . makes for a closer articulation of theory with the empirical evidence.” Similarly, Sullivan (2011, p. 907) recently emphasized that by offering insight into the contradictions that inevitably emerge between hypotheses and real-world data, studying negative cases contributes to the “essential ‘dialogue’ between theory and research” that must guide scientific efforts. Our goal for the present analysis, therefore, is to bring this vital perspective to research on the nexus between self-control, social bonds, and delinquency.

Measures

Self-control

Dialogue on the measurement of self-control has long focused on the distinction between attitudinal measures in which respondents rate themselves on indicators of self-control (Grasmick, Tittle, Bursik, & Arneklev, 1993) and behavioral measures, which assess involvement in noncriminal acts thought to be manifestations of low self-control. Hirschi and Gottfredson (1993, p. 48) argued for the latter, emphasizing that self-reported self-control might have significant measurement error (Piquero, MacIntosh, & Hickman, 2000). However, prior research indicates that the two types of measures have similar effects (Pratt & Cullen, 2000; Tittle, Ward, & Grasmick, 2003).

The measure in this study relies on behavioral indicators that correspond closely to the elements and manifestations of self-control (e.g., impulsivity, temper, and self-centeredness) that have been prioritized in attitudinal measures. Specifically, we focused on behaviors that Gottfredson and Hirschi identified as likely outcomes of low self-control. (The items are listed in the Appendix A, along with the items used to measure social bonds and delinquency). These behaviors reflect such things as impulsivity (those low in self-control have “a concrete here and now orientation,” p. 89), self-centeredness (those with low self-control are “self-centered, indifferent, or insensitive to the needs of others,” p. 89), a strong temper (those low in self-control are prone to a “loss of temper in response to . . .minor frustrations,” p. 93), inattentiveness and hyperactivity (those low in self-control will struggle when they are expected to be “quiet, physically inactive, and attentive,” p. 162), and difficulty getting along well with others (those low in self-control “have difficulty acquiring and retaining friends,” p. 94). While there is some disagreement in the literature about the best items for measuring self-control (Arneklev, Grasmick, Tittle, & Bursik, 1999; Piquero & Rosay, 1998), the items used in our investigation are similar or identical to those used in other recent studies that have revealed the expected relationships between self-control and antisocial behavior (Chapple, 2005; Hay & Forrest, 2006; Pratt, Turner, & Piquero, 2004).

The behaviors were assessed with survey items answered by a caretaker (the mother in more than 95% of cases) who indicated whether various statements were not true, somewhat true, or very or often true of their child. We recognize that mothers (or any other nonchild rater) could report inaccurately or may not know the behaviors of their offspring. On the other hand, the use of mother reports is well validated in the literature on the measurement of children’s self-control and early problem behavior (Hoyle & Davisson, 2011; Piquero, 2008; Tremblay, Boulerice, Arsenault, & Junger, 1995). Moreover, a principal alternative—using children’s own assessment of their self-control—presents established problems with same-source bias when measures of outcome variables also are drawn from data provided by the child (which is largely the case in this study). Piquero and colleagues (2000, p. 923) therefore recommend external ratings by teachers and parents that can “bypass the problem uncovered . . . regarding the level of self-control influencing survey response.”

To construct identical measures of self-control at ages 10, 12, and 15, the items were recoded (so that higher values reflect greater self-control) and then averaged to create a 17-item scale at each age period (α = .86 at age 10; α = .85 at age 12; α = .87 at age 15). Factor analyses revealed that all items load significantly on a single factor with an eigenvalue greater than 1.00 (with average loadings across the three waves of .50 to .54). The scale has high construct validity as revealed by the significant correlations (p < .01) between age 10 self-control, age 12 family bonds (r = .31), and age 12 school bonds (r = .24). Thus, this scale is reasonably strong with respect to key properties of measurement (including face validity, construct validity, and internal reliability), although we believe that self-control can be measured well in alternative ways.

Social bonds

We focus on two types of social bonds at ages 12 and 15: Family bonds and school bonds. These were chosen because of their central position in social bonding theory as well as their relevance to the spuriousness thesis. With respect to family bonds, those with low self-control at one age should report weaker family bonding at a later age because their impulsive behavior should provoke lower parent–child attachment and greater harshness from parents. This is in line with Gottfredson and Hirschi’s position that those with low self-control tend to have poor relationships with others (pp. 154–168). Similarly, low self-control should reduce school bonding because schools are organized to punish those who cannot show restraint (pp. 162–163). Thus, those with low self-control “will tend to avoid . . . school in favor of less restricted environments” (p. 163). By extension, they should be less bonded to school.

The SECCYD data provide strong measures for both types of bonds. Family bonds are measured with 14 items (six from the mother and eight from the child) that focus particularly on parent–child attachment and the parents’ avoidance of erratic and harsh forms of authority. The items measuring attachment asked mothers to indicate their agreement with such statements as “I share affection and have a warm relationship with my child” and “My child openly shares feelings and experiences with me.” Response categories for the items ranged from “definitely does not apply” (=1) to “definitely applies” (=5). The items on erratic and harsh authority were taken from the child and capture the inverse of what is expected among those with strong parent–child bonds (Sampson & Laub, 1993). Study children were asked how often their mother did such things as “shout or yell at you,” “strike or hit you,” “criticize your ideas,” or “boss you around a lot.” Response categories ranged from “never” (=1) to “always” (=4). The items on erratic and harsh authority were recoded such that high values indicate the absence of these incidents, and these items were combined with the attachment items to form a 14-item family bonds scale (α = .78 at age 12; α = .80 at age 15). (All items were standardized and then averaged, given the differences in scale between the attachment and hostility items). These scales also have strong construct validity, as indicated, for example, by the significant correlations between the measure of age 12 family bonds and age 10 self-control (r = .31), age 12 school bonds (r = .28), and age 15 delinquency (r = −.14).

School bonds at ages 12 and 15 are measured with 19-item scales (α = .82 at age 12; α = .85 at age 15) in which study children responded to statements regarding how much they enjoy school and are committed to doing well. Items included such statements as “I feel close to others at school,” “I work hard in school so I will be able to go to college,” and “I care about what my teachers think of me.” Response categories for the items ranged from “not at all true” (=1) to “very true” (=4). This measure also fared well in terms of construct validity, as indicated by significant correlations between age 12 school bonds and age 12 family bonds (r = .28) and age 15 delinquency (r = −.20). All items were coded such that high values indicate high school bonds, and then averaged.

Finally, in light of a pattern in which family and school bonds exerted similar effects in our age 15 contemporaneous models, we also constructed a global age 15 social bonds measure that includes all of the family and school bonding items noted above. The items were standardized and averaged to create a 33-item scale (α = .87) that is significantly correlated with age 15 delinquency (r = −.41).

Delinquency

SECCYD data contain information on a variety of delinquent acts reported by study children at age 15. Specifically, in reference to 27 different acts, respondents were asked “How many times in the past year have you [committed a specific act]?” The response categories were “not at all” (=0), “once or twice” (=1), and “more than twice” (=2). The items pertained to various violent, property, and substance offenses, including threatening to beat someone up, breaking into a building to steal or take something, and using marijuana. In line with Gottfredson and Hirschi’s (1990, p. 91) argument that “there will be much versatility among offenders in [their] criminal acts,” we used these items to create a variety index of delinquency. Each item was recoded to reflect whether or not the child had engaged in that behavior during the past year, and these items were then summed to produce a 27-item delinquency scale (α = .85).

Control variables

With respect to concerns about spuriousness, our primary control is the self-control variable described earlier. However, to better specify the association of delinquency with self-control and social bonds, controls were included for sex (1 = male, 0 = female), race (1 = non-White, 0 = White), age (measured in years down to tenths), family structure (1 = both biological parents, 0 = all others), and total household size. Table 1 provides descriptive statistics and correlations for the sample analyzed in this study. ⁴

OPEN IN VIEWER

Table 1.

Descriptive Statistics and Correlation Matrix (N = 835)

	Variable	Mean	SD	Min	Max	1	2	3	4	5	6	7	8	9	10	11	12	13
1	Age	15.10	0.14	14.79	15.97
2	Male	0.51	0.50	0.00	1.00	.00
3	Non-White	0.19	0.39	0.00	1.00	.05	.01
4	Family structure	0.59	0.49	0.00	1.00	−.05	−.02	−.27***
5	Household size	4.19	1.21	2.00	12.00	.01	−.02	.04	.12***
6	Self-control (Age 10)	1.75	0.25	0.35	2.00	.03	−.14***	−.06	.20***	−.10**
7	Self-control (Age 12)	1.79	0.24	0.47	2.00	.02	−.16***	−.05	.20***	−.04	.74***
8	Self-control (Age 15)	1.74	0.25	0.29	2.00	.00	−.10**	−.02	.17***	.01	.60***	.66***
9	Family Bonds (Age 12)	0.00	0.51	−2.74	0.72	.04	−.05	−.11**	.14***	−.05	.31***	.33***	.28***
10	Family Bonds (Age 15)	0.01	0.52	−2.95	0.81	.03	−.09*	−.09**	.12***	−.04	.20***	.19***	.31***	.44***
11	School Bonds (Age 12)	3.33	0.41	1.79	4.00	.01	−.25***	−.09*	.25***	−.02	.24***	.23***	.19***	.28***	.23***
12	School Bonds (Age 15)	3.22	0.45	1.68	4.00	−.00	−.11**	−.12***	.22***	−.00	.19***	.16***	.23***	.21***	.33***	.48***
13	Global Bonds (Age 15)	0.00	0.43	−1.69	0.82	.01	−.13***	−.13***	.21***	−.02	.23***	.21***	.32***	.38***	.75***	.46***	.87***
14	Delinquency (Age 15)	2.31	3.23	0.00	27.00	.03	.15***	.16***	−.18***	−.06	−.11**	−.13***	−.29***	−.14***	−.33***	−.20***	−.33***	−.41***

Note. *p < .05. **p < .01. ***p < .001 (two-tailed).

Results

The Association Between Social Bonds and Delinquency

The first step in the analysis was to conduct our own test of the spuriousness thesis to establish whether the prevailing pattern from prior tests—one in which social bonds are related to delinquency even when accounting for differences in self-control—emerges with our data as well. If this occurs, it suggests the presence of negative cases to be further investigated.

Two things bear emphasizing regarding our approach to this issue. First, we test the spuriousness thesis across different temporal arrangements that range from a longitudinal model with clear temporal ordering to a contemporaneous model sensitive to short-term immediate associations between our variables of interest. As we note below, none of these models are ideal. Together, however, they provide a diverse investigation that matches prior research on the spuriousness thesis, which includes the long-term longitudinal approach taken by Wright and his colleagues (1999) and the contemporaneous approach used by Evans and his colleagues (1997). Additionally, our analysis uses negative binomial regression because our count measure of delinquency has a distribution marked by positive skew and many zero values (37% of individuals reported no delinquency). Under such circumstances, the assumptions of ordinary least squares regression are violated, and negative binomial regression is more appropriate.

We estimated a total of five models, the results for which are shown in Table 2. We began with two models that feature a longitudinal arrangement with temporal lags between the measures of age 10 self-control, age 12 social bonds, and age 15 delinquency. This arrangement is consistent with a theoretical model in which bonds at age 12 have a lagged relationship with delinquency that could be attributable to preexisting differences in self-control that were relatively fixed by age 10. To consider this possibility, Model 1 assesses the baseline relationships between age 12 social bonds and age 15 delinquency while excluding the age 10 measure of self-control, whereas Model 2 adds the age 10 measure of self-control to consider whether the bonds-delinquency relationships are attenuated.

OPEN IN VIEWER

Table 2.

Negative Binomial Regressions of Delinquency at Age 15 on Social Bonds and Self-Control (N = 835)

	Longitudinal I (Ages 10, 12, and 15)				Longitudinal II (Ages 12 and 15)		Contemporaneous (Age 15)
	Model 1		Model 2		Model 3		Model 4		Model 5
Predictors	IRR	RSE	IRR	RSE	IRR	RSE	IRR	RSE	IRR	RSE
Age	1.39	.49	1.39	.48	1.37	.47	1.22	.44	1.14	.37
Male	1.40***	.12	1.39***	.12	1.38***	.13	1.45***	.11	1.41***	.10
Non-White	1.41*	.20	1.41*	.20	1.41*	.20	1.34**	.14	1.34**	.14
Family structure	0.73**	.09	0.73*	.09	0.74*	.09	0.76*	.08	0.78*	.08
Household size	0.95	.03	0.95	.03	0.95	.03	0.95	.03	0.96	.03
Family bonds (Age 12)	.83	.11	0.85	.09	0.86	.08	—	—	—	—
Family bonds (Age 15)	—	—	—	—	—	—	0.59***	.04	0.65***	.04
School bonds (Age 12)	0.65***	.08	0.66**	.08	0.66***	.08	—	—	—	—
School bonds (Age 15)	—	—	—	—	—	—	0.53***	.05	0.55***	.05
Self-control (Age 10)	—	—	0.88	.19	—	—	—	—	—	—
Self-control (Age 12)	—	—	—	—	0.79	.21	—	—	—	—
Self-control (Age 15)	—	—	—	—	—	—	—	—	0.55***	.09
Log α	0.23	.05	0.23	.05	0.23	.05	0.02	.07	−0.01	.08
Log pseudo-likelihood	−1,638		−1,638		−1,638		−1,594		−1,588
AIC	3,295		3,294		3,293		3,205		3,193
BIC	3,337		3,337		3,336		3,248		3,236

Note. IRR = incidence rate ratio; RSE = robust standard error.

*p < .05. **p < .01. ***p < .001 (two-tailed).

The results from Models 1 and 2 suggest that bonds can have a long-term effect on delinquency that persists when controlling for prior self-control. We interpret our models in terms of the incidence rate ratios (IRRs), and find that in Model 1, the IRRs are below 1.00 for both family and school bonds. This indicates the expected negative association between bonds and delinquency, but for family bonds, this relationship falls short of statistical significance (with an IRR of .83, a one-unit increase in family bonds decreases the expected delinquency rate by roughly 17%). For school bonds, however, the relationship is higher in magnitude and statistically significant—with an IRR of .65, a one-unit increase in school bonds decreases the delinquency rate by roughly 35%. Model 2 then adds the measure of age 10 self-control, which has an insignificant effect on age 15 delinquency (IRR = .88, p < .56). This is consistent with some prior studies that reveal weak long-term effects of self-control in models that include variables that may mediate effects of self-control on delinquency (Wright et al., 1999). Also, controlling for self-control did little to reduce the association between bonds and delinquency—the insignificant effect of family bonds is essentially unchanged (IRR = .85), while for school bonds, the significant relationship in Model 1 remains intact in Model 2 (IRR = .66, p < .001).

Although these models provide a clear specification of temporal ordering, they are limited in ways that prompted us to consider other models. First, given that self-control is not perfectly stable after age 10 (Hay & Forrest, 2006; Turner & Piquero, 2002), measuring self-control at age 10 disadvantages that variable in a model that includes more proximate (age 12) measures of social bonds. We therefore estimated Model 3, which controls for self-control at age 12 instead of age 10. The results in Model 3 largely replicate those from the lagged equation estimated for Model 2—school bonds maintain a significant effect of equal magnitude (IRR = .66), whereas family bonds and self-control have effects that are in the predicted direction but fall short of statistical significance.

Even Model 3, however, is a conservative test of these relationships, given that both self-control and social bonds can fluctuate between ages 12 and 15 in ways that are relevant to delinquency. In short, this 3-year temporal lag may fail to capture more immediate, short-term impacts of self-control and social bonds on delinquency. We therefore estimated Models 4 and 5, which assess contemporaneous relationships between these variables at age 15. Model 4 reveals significant and relatively strong relationships with delinquency for both family (IRR = .59, p < .001) and school (IRR = .53, p < .001) bonds. Model 5 adds a control for age 15 self-control. This variable too has a significant association with delinquency—with an IRR of .55, a one-unit increase in self-control reduces the delinquency rate by 45%. Nevertheless, controlling for self-control only modestly reduces the contemporaneous relationships with delinquency for family bonds (IRR = .65, p < .001) and school bonds (IRR = .55, p < .001.).

None of these models are ideal, but taken together, they produce informative patterns. Most notably, school bonds are consistently associated with delinquency, and this association is not explained by preexisting or contemporaneous differences in self-control. Family bonds, on the other hand, have no significant lagged relationship with delinquency. This perhaps points to a pattern in which family bonds have lagged effects that are quite indirect and that operate through various intervening mechanisms not considered here, including peer associations. However, the contemporaneous relationship between family bonds and delinquency is significant even when accounting for self-control (which itself has a relatively strong significant contemporaneous relationship with delinquency). These results support the finding in prior research that independent effects of social bonds can persist even when accounting for differences in self-control. The principal implication for the present analysis is that it suggests the presence of negative cases to be further examined.

The Negative Cases: Their Prevalence and Form

Our next step was to consider what portion of subjects can be classified as a negative case—a case in which self-control is not producing the expected corresponding level of social bonding. In considering this, we focus on the age 15 measures of self-control and social bonds. As noted, this approach is best for capturing the short-term, immediate associations that are expected to be most important for these variables. Moreover, to simplify our presentation of results and provide a more comprehensive assessment of social bonds, we use a combined family and school bonds measure. As noted above, school and family bonds are significantly correlated with one another (r = .33) and they have a similar contemporaneous relationship with age 15 delinquency (see Model 5 in Table 2).

To identify negative cases, we used an approach similar to that used by Burt and her colleagues (2006) in their study of self-control stability and change. Subjects were coded on the basis of their quartile positions in the self-control and social bonds distributions, and these two distributions were then cross-tabulated. Cases on the diagonal of this cross-tabulation are consistent with the theory—self-control is translating into the expected level of social bonding, such that cases are in the same quartile in the self-control distribution as they are in the social bonds distribution. On the other hand, cases off the diagonal are marked by an incongruence between self-control and social bonds that makes them a negative case.

Table 3 presents the results for this quartile analysis, and as can be seen, there are a considerable number of negative cases: 32.6% of individuals (273 cases of the 835) are found in the cells on the diagonal, leaving roughly 67% who are off the diagonal and are therefore a negative case to some degree. However, given that there may be substantively little difference between cases near the borders of the quartiles, it is important to focus on more extreme cases in which positions in the self-control and social bonds distributions are at least two quartiles removed from one another. Overall, 223 subjects—representing 26.7% of the overall sample—can be classified as this sort of negative case. Moreover, 8% of subjects (67 cases) provide the strongest example of a negative case by appearing in the top quartile for one distribution and the bottom quartile for the other.

OPEN IN VIEWER

Table 3.

Association Between Self-Control at Age 15 and Social Bonds at Age 15 (N = 835)

Four Quartiles of Self-Control	Four Quartiles of Social Bonds
	1c	2	3	4d	Total
Self-Control group 1a	78	57	35	38	208
	37.5	27.4	16.8	18.3
Self-control group 2	67	59	41	42	209
	32.1	28.2	19.6	20.1
Self-control group 3	34	48	67	60	209
	16.3	23.0	32.1	28.7
Self-control group 4b	29	45	66	69	209
	13.9	21.5	31.6	33.0
Total	208	209	209	209	835

Key = frequency/row %.

^aGroup with lowest 25% of self-control.

^bGroup with highest 25% of self-control.

^cGroup with the lowest 25% of social bonds.

^dGroup with the highest 25% of social bonds.

Note. Row percentages may not add to 100% due to rounding

Pearson X ² = 64.73*** (p < .001); Cramers V = .16.

Table 3 also reveals that these most extreme negative cases come in two different forms. Twenty-nine subjects form a high self-control/low social bonds group. These individuals are in the highest self-control quartile but the lowest social bonds quartile. Thus, they possess the high self-control thought to be prerequisite for developing strong bonds, but they have not experienced this expected benefit of self-regulation. These cases represent 3.5% of the entire sample and 13.9% of those cases who are highest in self-control. The other type of extreme negative case—one in which youths are in the highest quartile of the social bonds distribution but the lowest quartile in the self-control distribution—is represented by 38 cases. These negative cases are showing a pattern in which they develop strong social bonds even in the absence of the high self-control thought to be necessary for this outcome. These cases represent 4.6% of the total sample but nearly one fifth (18.3%) of cases in the highest quartile in the social bonds distribution.

To be clear, these results indicate that negative cases—especially extreme ones—are fairly uncommon. This is perhaps to be expected given the strong theoretical logic predicting a positive association between self-control and social bonds. Nevertheless the prevalence of negative cases is not exactly trivial either, ranging from between eight and 27% of cases depending on the standard that is used. ⁵ To varying degrees, these negative cases represent instances in which the expected link between self-control and social bonds does not exist.

One final analysis on this issue was conducted to confirm this result on a multivariate basis. An ordinary least squares (OLS) regression equation was estimated that regressed age 15 school or family bonds on age 15 self-control and the demographic controls. The results for these equations are shown in Table 4 and reveal significant effects of self-control on school (B = .19, p < .001) and family (B = .29, p < .001) bonds. This pattern supports the contention of self-control theory that self-control influences family and school bonding. Thus, this result affirms the predicted association underlying the spuriousness thesis. However, the level of explained variance (around 10%) points to an association between self-control and bonding that is not especially determinative, even if quite acceptable by normal social science standards. Thus, although self-control does increase social bonds, this relationship is moderate rather than strong, and as the quartile analyses show, negative cases clearly are evident.

OPEN IN VIEWER

Table 4.

OLS Regressions of Social Bonds at Age 15 on Self-Control at Age 15 (N = 835)

	Model 1 (Family Bonds)			Model 2 (School Bonds)
Predictors	b	RSE	B	b	RSE	B
Age	.14	.10	.04	.03	.07	.01
Male	−.06	.03	−.06	−.08*	.03	−.09
Non-White	−.09	.05	−.07	−.08	.04	−.07
Family structure	.06	.06	.06	.15**	.04	.17
Household size	−.02	.01	−.05	−.01	.01	−.02
Self-control	.62***	.09	.29	.34***	.05	.19
Constant	−3.05	1.52		2.24	.99
Adj. R 2		.10			.09

Note. OLS = ordinary least squares; RSE = robust standard error.

*p < .05. **p < .01. ***p < .001.

Implications for Delinquency

Our final focus concerns differences in delinquency across the different types of negative cases. Specifically, when there is an incongruence between self-control and social bonds, which factor has the greatest influence on delinquency? We focus again on contemporaneous relationships at age 15 (but we report alternative analyses below). Based on their positions in the self-control and social bonds distributions, we separated subjects into groups marked by (1) low self-control/low bonds, (2) high self-control/low bonds, (3) low self-control/high bonds, and (4) high self-control/high bonds. Building on the quartile analyses just presented, subjects needed to be in the bottom quartile to be classified as low and in the top quartile to be classified as high.

To map the distribution of offending across these groups, a dummy variable was created for each one, and a negative binomial regression equation was estimated in which age 15 delinquency is the dependent variable and the predictors include the controls and the dummy variables for three groups (low self-control/low bonds is the reference group). Table 5 shows the results for these equations and reveals, as expected, that delinquency is lowest for the high self-control/high bonds group—its IRR of .19 (p < .001) indicates that being high in both self-control and social bonds reduces the delinquency rate by roughly 80% when compared to the low self-control/low bonds group. This provides strong evidence on the way in which self-control and social bonds jointly reduce delinquency.

OPEN IN VIEWER

Table 5.

Negative Binomial Regression of Delinquency at Age 15 on Categories of Self-Control and Social Bonds at Age 15 (N = 214)

Predictors	IRR	RSE
Age	1.59	.76
Male	1.27	.26
Non-White	1.56**	.22
Family structure	0.91	.13
Household size	0.91	.06
Low self-control/high bonds	0.44**	.12
High self-control/low bonds	0.88	.24
High self-control/high bonds	0.19***	.05
Log α	−.01	.16
Log pseudo-likelihood	−435
AIC	888
BIC	919

Note. IRR = incidence rate ratio; RSE = robust standard error.

**p < .01. ***p < .001.

Of special interest, however, are the delinquency rates for our two groups of negative cases. On this issue, we find that the lower delinquency rate is found among those in the low self-control/high bonds group—the IRR of .44 for this variable indicates that membership in this group was associated with a 56% reduction in the delinquency rate. Thus, the high-level bonds allowed this group to exhibit lower delinquency despite its low level of self-control. In contrast, membership in the high self-control/low bonds group was not associated (IRR = .88, p = .64) with a significant reduction in delinquency relative to the low self-control/low bonds group. This coefficient is in the correct direction (indicating a 12% reduction in the delinquency rate for this group) but is not statistically significant.

Taken together, these results suggest that although it is better to be high in self-control and social bonds, if one is to be high in just one of them, social bonds are the preferred form of control. We assessed the robustness of that conclusion in a variety of ways. First, we considered whether it holds across different types of delinquency (violent, property, and substance use) that are represented in our general delinquency measure. We found that the pattern noted above held for both violent and property delinquency. For substance use, however, neither type of negative case was associated with a delinquency reduction relative to the low self-control/low bonds group—the only group significantly lower in substance delinquency was that marked by both high self-control and high social bonds. Second, we considered whether the pattern reported above was maintained when using a more extreme definition of negative cases in which low and high categories were defined by membership in the lowest or highest quintile (rather than quartile). It bears emphasizing that these more extreme thresholds produce fewer cases categorized as low or high, thus producing an analysis with a lower N (136, as compared to the N of 214 for the analysis presented in Table 5). That said, this analysis replicated the results presented in Table 5. The delinquency rate was most reduced by membership in the high self-control/high bonds group (IRR = .19, p < .001), but membership in the low self-control/high bonds group also produced a significant reduction in delinquency (IRR = .38; p < .01) relative to the low self-control/low bonds group. No such reduction occurred for the high self-control/low bonds group (IRR = .66, p = .10), although this effect approached statistical significance. Finally, we considered whether the results reported in Table 5 were maintained in a longitudinal model that estimated lagged effects of age 12 self-control and social bonds on age 15 delinquency. That analysis revealed that the only significant reduction in delinquency came from being in the high self-control/high bonds group. Thus, with respect to how delinquency rates vary across our negative cases, the most consistent pattern involved the reduction in delinquency for the low self-control/high bonds group, relative to the low self-control/low bonds group. That pattern did not emerge in all analyses, but it was evident for contemporaneous age 15 analyses of general, violent, and property delinquency.

Discussion and Conclusion

This study began by considering the argument that the link between social bonds and crime is spurious and attributable to differences in self-control. Similar to prior tests, we found significant associations between bonds and delinquency that often persisted when accounting for self-control. This was evident for school bonds both in longitudinal and in contemporaneous models and for family bonds in contemporaneous models estimated at age 15.

This pattern of findings gave rise to our central focus on the negative cases that underlie this empirical pattern and that can be used to inform future research (Giordano, 1989; Sullivan, 2011). This focus on negative cases represents the most novel part of the study—it provides the first investigation to date of the negative cases that deviate from the expected logic of the spuriousness thesis. Thus, although prior research suggests that such negative cases exist, little is known about how common they are, what exact forms they take, and how delinquency varies across different types of negative cases. The analysis pointed to several key conclusions on these issues.

First, in terms of prevalence, we found that negative cases were not especially common. Depending on the standard used to identify negative cases, they represent between 8% and 27% of cases. The high estimate of 27% is for cases that were at least two quartiles removed from one another in these distributions. This occurred, for example, for cases that were in the top quartile in self-control but in the third quartile or below (and thus, below the mean) for social bonds. The low estimate of 8% is for cases in opposite quartiles of the self-control and social bonds distributions; thus, their positions in these two distributions were removed by at least 50 percentile points. The fact that these extreme negative cases represent less than 10% of subjects might lead some to ignore them. Prior discussions of negative cases, however, highlight the important ways in which they can inform theory and research “regardless of their frequency” (Lindesmith, 1968, p. 20) . Additionally, given that crime itself—especially serious crime—is prevalent among only a small portion of the population, it seems unwise to ignore extreme patterns simply because they are infrequent (Wolfgang, Figlio, & Sellin, 1972).

A second key finding is that the negative cases were well represented by the two logically possible types: (1) those who possessed high self-control but did not experience the strong social bonds that were expected and (2) those with low self-control who had unexpectedly high social bonds. Thus, high self-control at times did not produce strong bonds, and strong bonds sometimes occurred in the absence of high self-control.

A clear implication of this pattern is its suggestion that factors other than self-control come into play in the development of key social bonds. Empirically considering this issue was beyond the scope of this study, but in thinking about what those other factors may be, it bears emphasizing that social bonding may not be purely a function of one’s own qualities and characteristics (e.g., one’s own self-control). The actions of other individuals could be relevant as well. For example, social support from family members or teachers could compensate for an individual’s low self-control, while aversive, nonsupportive actions could counteract the expected benefits of high self-control (and thus prevent strong bonding). Also, luck or serendipity may come into play, as an individual’s chance experiences impact their social bonds (Laub & Sampson, 1993, pp. 317–318). And one’s own characteristics other than self-control may be important as well. For example, being good-natured, amiable, talented, or charismatic could compensate for low self-control in the development of social bonds, whereas the absence of these things could lead those with self-control to be less bonded than expected. Future research could be usefully directed toward understanding how these factors and others contribute to differences between individuals’ actual and expected levels of bonding.

The final key conclusion involves the implications for delinquency of our negative cases. By definition, these negative cases mean that one form of control—self-control or social control—is absent (relatively speaking) and one is present. We found that being in the low self-control/high bonds group often reduced delinquency relative to the low self-control/low bonds groups, whereas no such reduction occurred for the high self-control/low bonds group. This points to the protective effects of social bonds even for those who are low in self-control; however, high self-control often did not provide this same protective role for those low in social bonds. This conclusion is consistent with those theoretical perspectives (such as Sampson and Laub’s age-graded control theory and “social causation” models in general, Wright et al., 1999) that view positive social relationships and experiences as capable of reducing delinquency, even when they occur in the absence of key crime-reducing individual characteristics (like self-control). As we note below, however, firm conclusions on this issue cannot be reached after just one investigation of these negative cases. Moreover, the greatest reductions in delinquency were for those high in self-control and social bonds, therefore pointing to the important joint effects of the two.

To be sure, our effort was limited in some respects, and these limitations offer directions for subsequent inquiry. First, although we examine the interrelationships between self-control, social bonds, and delinquency after the period of time that Gottfredson and Hirschi expect self-control to be relatively established (i.e., by age 10), we only covered a small swath of time until middle adolescence. It is important to examine whether our findings are replicable at other stages of the life course (late adolescence, early and middle adulthood) and with other types of social bonds (involving peers, romantic relationships, or children). Second, we operationalized self-control as a trait in a manner that is a consistent with Gottfredson and Hirschi (1990). Recently, Hirschi (2004) argued that self-control is a more situational variable that should be considered at the point of the decision to offend when one considers the costs and salience of a criminal act (Piquero & Bouffard, 2007). A vibrant dialogue regarding the exact nature of self-control and how it should be conceptualized continues through the present (Piquero, 2008), and the insights that ultimately emerge should influence future tests of the spuriousness thesis. Additionally, our study was unable to consider how patterns with respect to negative cases may play out differently across different demographic subgroups of the population. It would be informative, for example, to consider whether the spuriousness thesis is supported as much for males as it is for females or as much for whites as it is for racial and ethnic minorities. Our analysis also cannot claim to have captured the definitive causal ordering of our hypotheses, even in those models that were longitudinal and contained temporal lags.

An additional issue that bears highlighting is same source bias. Findings on the degree of congruence and incongruence between self-control and social bonds likely are sensitive to the decision that researchers make regarding the distinction between a same-source and mixed source design. When measures of self-control and social bonds are taken from the same source, this may overstate the degree of convergence between these two variables. On the other hand, a mixed-source approach could understate this relationship if there are questions about the validity of data provided by alternative sources. We took largely a mixed-source approach (with self-control data coming from the mother and school bonds data coming from the child) because of the established concerns about same source bias as well as the favorable evidence on the validity and reliability of our measures. Nevertheless, we do not see our approach as the only acceptable one; most importantly, in moving forward, the degree of same source bias should be an important consideration when evaluating research on the convergence between self-control and social bonding.

We also should highlight a few limitations regarding our data. Most notably, our data came from a nonprobability sample drawn from 10 U.S. cities, and so confirmation of our findings in other samples is important. Additionally, although our analyses of these longitudinal data revealed minimal problems with selective attrition, attrition analyses are necessarily limited to measured variables. Thus, the possibility of selective attrition cannot be ruled out. And like many samples that do not overrepresent high-risk groups, the typical respondent had relatively high self-control and social bonds and relatively low delinquency. The use of samples with a greater number of high-risk subjects (and thus, variables with greater variation) should be studied in future research, in part because greater variability can enable a more precise identification of the causes of delinquency. It bears emphasizing, however, that this study nevertheless was able to identify negative cases despite these limits on variability.

With these limitations in mind, we do not offer our empirical test as the only or even the most definitive way to examine the spuriousness thesis or the negative cases that depart from its logic. On the contrary, we expect that future tests with alternative samples, measures, and analytical approaches can significantly inform work in this area. We do, however, see the findings presented here as having key implications for theory. Most notably, our results undermine any pure version of the spuriousness thesis in which social bonds are largely a function of self-control and have no independent association with delinquency. To be clear, we do not see the spuriousness thesis as especially pivotal to the general theory of crime, and so failure to support it does not invalidate that theory. That said, we see clear ways in which the theoretical understanding of the link between self-control and social bonds should be amended if similar results emerge in other studies. Theory in this area—and the general theory of crime in particular—will be more consistent with empirical evidence if it emphasizes a more probabilistic association between self-control and social bonds. Thus, although self-control (and the diligence, thoughtfulness, and persistence that it includes) may contribute to social bonding, exceptions to that pattern (i.e., negative cases) will emerge, and these cases direct theoretical attention to the factors other than self-control that can influence social bonding. These include such things as the efforts of significant others who are personally invested in an individual, individual characteristics other than self-control that may facilitate strong bonding (e.g., charisma or amiability), and even the luck that comes from being in the right situation at the right time in ways that foster bonds to family, school, or employment. Any of these factors may reveal why the expected association between self-control and social bonds sometimes does not emerge. Thus, they may point to areas in which existing perspectives on the association between self-control and social bonds may be theoretically elaborated.

We should emphasize that these findings also have important policy implications for responding to youth delinquency and antisocial behavior. These implications are most notable in reference to children with significant deficits in self-control. Given the disruptive nature of their behavior and their potential for later delinquency, such children attract much attention from school officials, the juvenile justice system, and social service agencies. Our findings, however, reveal the positive developments that are possible for these children in spite of their relative deficits in self-control. They often were able to develop strong social bonds, and when this occurred, delinquency was lower. Importantly, this pattern emerged in the absence of any systematic intervention offered to SECCYD subjects. This suggests the gains that can come from explicit programmatic efforts to increase social bonds among high-risk youths. Many successful programs have been identified for increasing bonds to family, the school, or to community mentors, and the list of such programs includes empirically validated interventions such as multisystemic therapy (Henggeler, Schoenwald, Borduin, Rowland, & Cunningham, 1998), the Seattle Social Development Model (Hawkins, Kostermann, Catalano, Hill, & Abbott, 2005), the Triple P Positive Parenting Program (Piquero, Jennings, & Farrington, 2010; Sanders, Turner, & Markie-Dadds, 2002), and the Big Brothers and Big Sisters mentoring program (Grossman & Tierney, 1998). These programs may work in part by increasing self-control, but in line with the results presented here, they may also work by enhancing the social bonds of children with potentially enduring deficits in self-control. Our results support the continued and expanded use of such programs for those marked by strong deficits in self-control.

Of more general importance, our results suggest that children with low self-control are far from a lost cause—other redeeming qualities that they may have, the efforts of significant others, and even interpersonal luck may allow them to develop strong social bonds that protect them from their propensity for crime and delinquency. When this occurs, these instances represent negative cases for the spuriousness thesis—cases in which the expected congruence between self-control and social bonds does not emerge. Our hope is that the continued study of such negative cases can provide important new insights for theory, research, and policy efforts relevant to the interconnections between self-control, social bonds, and delinquency.