Social Learning, Self-Control, and Offending Specialization and Versatility among Friends

Abstract

While it is generally understood that people tend not to specialize in specific types of deviance, less is understood about offending specialization and versatility in the context of friendships. Using a large sample of persons nested within friendship pairs, this study’s goal is to explore how self-control and social learning theories contribute to an explanation for specialization and versatility in offending among friends. We estimate a series of multilevel, dyadic, mixed-effects models which regress offending versatility onto measures of perceptual peer versatility, self-reported peer versatility, attitudinal self-control, behavioral self-control, and demographic controls. Results indicate that higher amounts of perceptual peer versatility and peer self-reported versatility are both related to increases in versatility among friends. Lower levels of the target respondent’s attitudinal and behavioral self-control are also related to higher amounts of offending versatility. However, the peer’s self-control shares no relationship with offending versatility – a point which both supports and fails to support self-control theory’s expectations about how peer effects should operate. Learning and self-control perspectives both appear to explain offending versatility among friends. However, self-control theory’s propositions about how peer effects should operate are contradictory. The concept of opportunity may help remediate this inconsistency in Gottfredson and Hirschi’s theory.

Introduction

Although social learning and self-control theories both explain crime (e.g., Pratt & Cullen, 2000; Pratt et al., 2010), the theories are fundamentally at odds with each other. Underpinning both theoretical approaches is a fundamentally different understanding of deviant behavior. To the learning perspective, deviance – like all things – is learned. However, according to control theories, deviance is somewhat ingrained and must be restrained by the proper development of self-control. Despite what may be irreconcilable differences between the two theories, their strong and consistent ability to explain crime has led to a newer debate that focuses not on whether the theories explain crime, but instead on which theory explains crime the best. Considering research on this newer learning/control issue yields mixed results (cf. Pratt et al., 2010; Vazsonyi, Mikuška, & Kelley, 2017), criminologists must continue to develop innovative ways of exploring the predictive ability of the competing theories.

At the heart of the learning/control tension is a theoretical disagreement in the causal pathways that lead to specialization and versatility in offending. Gottfredson and Hirschi’s (1990) perspective suggests that individuals with low self-control should be versatile in offending, meaning that they should commit a wide range of different types of deviant behavior. That is, individuals with low self-control should not ‘specialize’ by repeatedly committing the same type of deviant behavior. Like most issues surrounding control and learning theories, however, Sutherland’s (1947) differential association theory and Burgess and Akers’ (1966; also Akers, 2009) social learning theory share a quite different prediction regarding the causes of specialization and versatility in offending. To a learning perspective, individuals who specialize in only one type of crime do so because they have learned values, skills, and definitions that promote committing that one type of crime. On the other hand, an individual who is versatile in their offending patterns should have learned definitions favorable to versatility. As such, offending specialization and versatility, like all forms of behavior in a learning context, is the result of a learned process.

Intertwined in the disagreement between learning and control theories is a very different set of assumptions made about friends and friendships. While both theories recognize that friends are of importance, self-control argues that any link between the behavior of friends generally, or versatility and specialization specifically, should be the result of similarity in the friends’ levels of self-control (e.g., Gottfredson & Hirschi, 1990, pp. 157–159). Social learning theory, on the other hand, would argue that the offending versatility or specialization patterns of friends should directly influence an individual’s patterns of behavior (see Akers, 2009; Sutherland, 1947; Thomas, 2016). As such, the theories expect a much different pathway through which specialization (akin to social learning theory) and versatility (akin to either learning or self-control theory) form.

Although friendships are theoretically intertwined into the larger specialization and versatility issue, researchers have had considerable difficulty in developing a knowledge base that recognizes friendships as central to specialization and versatility. Despite this, we do know from studies on specialization and versality that most people do not specialize in their patterns of deviance (see Farrington, 2003; Jennings, Zgoba, Donner, Henderson, & Tewksbury, 2014; Mazerolle, Brame, Paternoster, Piquero, & Dean, 2000; Piquero, 2000; Piquero, Farrington, & Blumstein, 2003; Wright, Pratt, & DeLisi, 2008; also see Armstrong, 2008; Ha & Andresen, 2017; Harris, Smallbone, Dennison, & Knight, 2009; McGloin, Sullivan, Piquero, & Pratt, 2007; Sullivan, McGloin, Pratt, & Piquero, 2006). Although this might seem to support self-control on first glance due to the theory’s adamant stance against specialization, this relatively consistent finding could in fact support either the control or learning perspective when viewed from the perspective of friendships. Further developing a theoretical clarity and understanding of the relationship between specialization, versatility, and peer characteristics is the broad goal of this study. Using data from a large study of people nested within friendship pairs (or friendship ‘dyads’), we apply the primary tenets of self-control theory and social learning theory to versatility and specialization in offending among friends. Before discussing the specifics, however, we begin by reviewing each theory’s approach to explaining offending specialization and versatility and discussing research relevant to the competing theories’ expectations.

Self-Control, Specialization, and Versatility

As a general theory of crime, self-control theory posits that the root cause of deviance is an individual’s level of self-control. Developed in early childhood through effective parental punishment (Gottfredson & Hirschi, 1990), self-control refers to the ability to restrain one’s behavior. Individuals with low self-control tend to be impulsive, insensitive, risk-taking, and tend to not consider the long-term consequences of their behavior. Gottfredson and Hirschi contend that self-control is not specific to any particular set of offending behaviors and, instead, the theory purports to explain “all crimes, at all times” (p. 117). While research on the efficacy of self-control in explaining all types of crime is mixed (e.g., LaGrange & Silverman, 1999; however, see DeLisi, Hochstetler, Higgins, Beaver, & Graeve, 2008), studies tend to support the notion that those with low self-control offend more than those with greater levels of self-control (e.g., DeLisi, 2001). Highlighting this stance, Pratt and Cullen’s (2000) meta-analysis on the general theory of crime concludes that “self-control [is] one of the strongest known correlates of crime” (p. 952; see also Vazsonyi et al., 2017 for a more recent meta-analysis).

Applying this argument to versatility in offending, self-control theory would posit that low self-control – as the root cause of all deviant and antisocial behaviors – should be predictive of a significant degree of variation in offending. This is due in part to the fact that most crime is highly opportunistic and requires little in the way of planning (Pratt, Barnes, Cullen, & Turanovic, 2016). The result is that individuals with low self-control will engage in a variety of offending behaviors with little consistency or stability in the types of crimes they choose to commit. In this vein, Gottfredson and Hirschi (1990) hypothesize that low self-control will produce “much versatility among offenders in the criminal acts in which they engage” (p. 91). Summarizing their opinions of offending and specialization, Gottfredson and Hirschi succinctly state that “offenders do not specialize” (p. 94).

Extending this argument into the realm of friendships, Gottfredson and Hirschi (1990) are clear that people with low self-control should form friendships with one another. That is, people with low self-control tend to befriend individuals who also have low self-control, thus resulting in peer groups containing deviant members who should all have low self-control. While these friendships should be of poor quality, Gottfredson and Hirschi (p. 234) state that self-control levels should nonetheless cause deviant individuals to “flock together”. Self-control, therefore, is the basis for friendship formation and offending alike. Going one step further, self-control is also the basis for offending versatility. As such, those with low self-control should come together and coalesce into friendships which contain members who are 1) marked by low self-control, 2) are highly deviant, and 3) highly versatile in their offending patterns. Accordingly, levels of self-control among friends should be related to, and consequential for, versatility in offending among members of a friendship.

Extant research on self-control and friendship formation has, to an extent, explored Gottfredson and Hirschi’s (1990) expectations about friendships and self-control. While two studies suggest that self-control is not attributing to friendship selection preferences (Boman, 2017; Young, 2011), other studies find that selection processes are important and may be linked to self-control (Baron, 2003; Chapple, 2005; McGloin & Shermer, 2009; Simons, Wu, Conger, & Lorenz, 1994). However, the natural complexity of studying friendships has limited the ability of such research to examine how versatility in offending varies across friends’ levels of self-control. Despite this, research does demonstrate that individuals who choose to offend tend not to specialize (e.g., Brame, Bushway, Paternoster, & Apel, 2004; see also DeLisi, 2005). As Pratt et al. (2016) assert, “The reality is that offenders are not all that picky when it comes to their misbehavior” (p. 838). Supporting this notion, low self-control is related to numerous, versatile antisocial outcomes and offenses (e.g., see Hirtenlehner & Kunz, 2017). In exploring the relationship between self-control and versatility in deviance, Pratt et al. (2016) found that individuals with low self-control are likely to experience instability in employment, be held back in school, and drop out of school. They also are much more likely than those with high self-control to report experiencing problems with alcohol dependency. Other research exploring versatility and self-control has established important linkages between self-control and being involved in accidents (Junger & Tremblay, 1999), academic cheating (Bolin, 2004), binge eating (Tangney, Baumeister, & Boone, 2004), victimization (Schreck, 1999; Turanovic & Pratt, 2014), alcohol-induced sexual assault (Franklin, 2011), and bullying (Unnever & Cornell, 2003). Research has even tied low self-control to ‘drunk dialing’ and the use of profanity in public (Reisig & Pratt, 2011). In short, research using self-control theory demonstrates that “offending is versatile instead of specialized” (Farrington, 2003, p. 223; cf. Higgins & Makin, 2004).

Social Learning, Specialization, and Versatility

Rooted in Sutherland’s (1947) theory of differential association, Akers’ (e.g., 2009) social learning theory posits that deviance is learned through interactions with one’s differential associates. In these interactions, people develop definitions that are either favorable or unfavorable to crime, and crime occurs when the collective weight of definitions favorable to crime exceeds the weight of definitions unfavorable to crime. Overall, research strongly supports the notion that differential association and social learning explain crime in the manner purported by Sutherland (1947) and Burgess and Akers (1966; see Pratt et al., 2010).

Applying this argument to versatility in offending, learning perspectives immediately point to a critical factor to the development of definitions – one’s differential associates. Through interactions with differential associates, people learn and model the behavior of those with whom they share meaningful relationships. In the case where a person has meaningful ties to those who are specialized on one type of deviance, one’s definitions are likely to become favorable for that one type of behavior. As such, the theory in this case would expect specialization as there is little reason to believe that this person would be versatile in their offending patterns. On the other hand, if one has meaningful ties to others who commit a wide variety of offenses, then the person will probably be versatile in their offending patterns. The implication, then, is that social learning theory’s element of differential association provides the context through which to understand how social learning theory may explain both versatility and specialization in offending.

The literature on social learning theory and specialization/versatility is surprisingly underdeveloped. Summarizing the lack of research on the topic, Thomas (2016) recently, and correctly, stated that “little is known about the role peers play in promoting offending versatility” (26). Of the research that does exist, we know that peers who are isolated tend to offend less on the whole (Demuth, 2004) and in more specialized patterns (Thomas, 2016), although a small portion of isolates offend extensively and in versatile patterns (Kreager, 2004). For the most part, these findings play into the notion that those with friends tend to offend in versatile ways. Warr (1996; also 2002, pp. 38–39) has also noted that group-level behavior tends to be more specialized than individual-level behavior. That is, offending patterns of people are diverse even though some groups only tend to engage in certain types of behaviors. This premise has received empirical support in the research. Studying egocentric networks (‘send’ networks), McGloin and Piquero (2010) found a strong positive relationship between the extent to which people are integrated into their social networks and specialization at the group, but not individual, level.

Some research also highlights that social learning approaches are consistent when people do learn specialized behaviors, such as in the case of sexual offending against adolescents (Felson & Lane, 2009) and, to a lesser extent, stalking (Fox, Nobles, & Akers, 2011). As such, learning theories have been found to explain both offending versatility and specialization. Since social learning is theoretically equipped to explain both behaviors, the theoretical tenets of the theory appear to be supported. Despite this, there is a need for further research on the topic in the context of friendships. This observation raises attention to the goals of the current study.

Current Study

Using a large dataset consisting of individuals nested within friendship pairs, the current study has three primary research questions. First, drawing on social learning (e.g., Akers, 2009) and self-control (e.g., Gottfredson & Hirschi, 1990) theories, is versatility in offending more related to a peer’s offending versatility or levels of self-control among members of friendships? Second, does the method of measurement of peer offending versatility change the understanding of the strength of social learning measures on versatility in offending? Specifically, we interchange an indirect, perceptual measure of peer versatility with a measure of self-reported versatility directly from the friend him/herself. Third, and in a similar mindset, does the means through which self-control is measured – through either an attitudinal or behavioral measure – change the understanding of how self-control relates to versatility in deviance? In lieu of offering hypotheses, we employ an exploratory approach to these theoretically-driven research questions and avoid making specific predictions about the relationships we seek to explore.

Methods

Data

The data for this project come from a large sample of persons nested within self-identified friendship pairs (dyads). The data, which consists of 2154 undergraduate college students nested within half as many friendships (1077 dyads), was collected in 2009 at a large university in the southeastern United States. To collect the dyadic sample, the lead investigator contacted faculty teaching the 50 highest enrollment classes offered by the university. The instructors were asked whether they would be interested in providing extra credit to students for the completion of a survey that involved dyads. Two dozen instructors said that they would compensate their students with varying amounts of extra credit for participation in the study.

The chief investigator then prepared documentation for each course’s website and made in-class visits to notify potential participants about the opportunity. Respondents were asked to come to a university building where the study was headquartered during set operating hours. Instead of coming to the study alone, however, they were asked to attend with one of their five best friends in undergraduate studies. The procedure of asking for one of the respondent’s five best friends, which drew from precedent developed by the Add Health data (e.g., Haynie, 2002; Haynie & Osgood, 2005) and the NSCR’s (e.g., Weerman & Smeek, 2005) and Kandel’s (e.g., 1978) friendship selection procedures, was designed to attract very close friends (i.e., ‘best’ friends) as well as more ‘regular’ friends. Capturing both ‘best’ and ‘regular’ friends aligns with prior research which finds that best friends may exert a stronger behavioral influence on persons than regular friends (Weerman & Smeek, 2005). Upon arrival to the study’s headquarters, dyad members provided informed consent and were then sent to different locations to complete the study.

Once separated, members of the research team provided the friends with identical paper surveys that were pre-coded with a matching dyadic identification number to link them as being members of one dyad. Each survey contained questions about the respondent, the friend, and the friendship. Research team members monitored respondents during survey administration and were instructed to eliminate any potential avenue for communication (e.g., texting) between the friends during the time the survey was being taken. Following completion of the surveys, each dyad member was individually debriefed and exited the study’s headquarters via separate exits.

Many classes in the project were very large, with several classes carrying enrollments of hundreds of students. Although the 24 classes combined to a total enrollment of 5000 persons, the sampling frame’s size cannot be calculated since it is unknown how many of each respondent’s five friends he/she would have considered bringing to the project. Due to the very large number of potential respondents, about one in five ‘friends’ also received extra credit from a selected course. No significant differences were identified between those who did and did not receive extra credit. Despite all demographic characteristics closely matching the target population, females (66% of the sample; 59% of the population) were slightly overrepresented in the sample. The sample is comprised of 1152 women nested within female-only dyads, 444 men nested within male-only dyads, and 558 people nested within split gender dyads. All dyads are independent, meaning that no person was nested within more than one friendship pair. All procedures were approved by the institution’s review board.

Dataset Structure

In the criminological context, dyadic data are unique because models can be estimated that explore how the characteristics of two individual people relate to the target respondent’s behavior. The current project uses a double-entry file (see Kenny, Kashy, & Cook, 2006), a type of datafile structure common in dyadic data analysis, to maximize the inferences which can be made. In the double-entry file, the units of analysis are individuals nested within dyads (n = 2154 persons within 1077 friendships). This creates a situation where each person in the dyad has his/her own line of data and serves as the focal person of interest (called the ‘actor’) whose behavior may be dependent on characteristics of him/herself and another person (the ‘friend’). Stated differently, Person A is the actor who has Person B as a friend, and, inversely, Person B is the actor who has Person A as a friend. For more information on this file structure, see Kenny et al.’ (2006) and Campbell and Kashy’s (2002) work.

Dependent Variable

Self-Reported Versatility in Offending

The outcome variable in this project, which is designed to capture the extent of versatility in an acto’s offending, required several steps to construct. The data contained 20 different self-reported deviance items that asked the actor about crime over the past twelve months. Each question was worded, “In the past 12 months, how often have you item” and was measured on the National Youth Survey’s metric of 0 (“never”) to 8 (“two to three times a day”; see Elliott, Huizinga, & Ageton, 1985). These 20 self-reported items load onto five distinct constructs of behavior – theft (4 items), vandalism (4 items), violence (4 items), alcohol use and related behaviors (4 items), and drug use and related behaviors (4 items). Confirmatory factor analyses (CFAs) showing evidence of very close fit (all Confirmatory Fit Indices [CFIs] ≥ .95, Tucker-Lewis Indices [TLIs] ≥ .95, and root mean square errors of approximation [RMSEAs] ≤ .06) confirmed the five-factor solution (unreported).

To construct the measure of self-reported versatility, each item was first collapsed into a binary measure where the actor either indicated that he/she engaged in the behavior or refrained from the behavior altogether (0 = did not commit; 1 = committed). Second, each binary item was then summed into a variety index for each construct of behavior, making the range of all five constructs ‘0’ to ‘4’ since each behavioral construct carries four items. Third, this project’s main goal lies in investigating the relationship between versatility in offending and theoretical predictors. This issue more closely adheres to the commission of a wide range of behaviors rather than the frequency of such behaviors. As such, each construct’s variety index was collapsed into a binary measure where a score of ‘1’ indicates the actor committed a deviant act within the purview of each behavioral construct (a score of ‘0’ means the actor did not engage in that construct of deviance whatsoever). Fourth, and finally, the binary measures of construct-specific deviance were summed into the dependent variable, self-reported offending versatility. This variety index has a range of ‘0’ – indicating that the actor’s behavior was totally non-versatile since he/she committed no deviance whatsoever – to ‘5,’ which indicates the actor’s behavior was extremely versatile since he/she engaged in all five constructs of behavior. Higher scores represent a greater offending versatility. Since the mean of this item (M = 1.834) is greater than one, the average person was at least somewhat versatile in their patterns of deviance (see Table 1 for descriptive statistics).

Table 1.

Summary characteristics of the dyadic sample (N = 2154 people nested within 1077 friendships)

	M	SD	Min.	Max.
Dependent Variable
Self-reported offending versatility	1.834	1.396	0	5
Independent Variables
Perceptual peer versatility	1.363	1.250	0	5
Peer self-reported versatility	1.834	1.396	0	5
Self-control (attitudinal)	2.878	0.350	1	4
Self-control (behavioral)	5.596	0.687	1	7
Male	0.336	0.472	0	1
Non-white	0.369	0.483	0	1
Hispanic	0.186	0.389	0	1
Age	19.339	1.433	18	42

Descriptive characteristics for the actor and friend are identical due to the structure and nesting within the data

Independent Variables

Perceptual Peer Versatility in Offending

The actor’s perception of the versatility of the friend’s offending serves as the first independent variable. Created to be conceptually similar to the dependent variable, 20 questions asked the actor “In the past 12 months, how often has the friend you attended this study with item?” The behaviors are the same as those in the dependent variable. Measured again on the NYS metric of ‘0’ (never) to ‘8’ (two to three times a day), these items also load on a 5-factor structure where each construct (theft, vandalism, violence, alcohol use and related behaviors, and drug use and related behaviors) has four items and shows evidence of close fit (unreported CFA fit statistics: all CFIs and TLIs ≥ .95, all RMSEAs ≤ .06).

To create this independent variable, the 20 base perceptual items were collapsed into binary measures indicating whether the actor thought his/her friend engaged in the behavior (scored ‘1’) or did not engage in the behavior (‘0’). These items were summed together for each construct and re-dichotomized to indicate whether the respondent thought his/her friend engaged in any deviance within the respective construct (scored ‘1’) or refrained entirely from deviance within the construct (‘0’). Finally, the binary items for each construct were summed to capture perceptual peer versatility in offending. This measure captures the number of constructs of deviance which the actor thought his/her friend had engaged in over the past year (M = 1.363, SD = 1.250, range 0–5). Higher scores represent greater versatility.

Self-Reported Peer Versatility in Offending

Due to the way the study was designed and the double-entry datafile structure, we are capable of providing an alternative measure of peer versatility that relies on reports directly from the peer him/herself. In addition to the authors of self-control theory disliking perceptual measures of offending (e.g., Gottfredson & Hirschi, 1990), research demonstrates that measures of perceptual peer offending (termed ‘indirect’ peer deviance) operate differently in multivariate models than measures of self-reported offending gathered directly from peers themselves (called ‘direct’ peer deviance; see Meldrum, Young, & Weerman, 2009).

Accordingly, we include a measure of direct peer deviance that captures self-reported peer versatility in offending. Although the measure captures the versatility of the friend’s offending, it is constructed in an identical manner to the dependent variable. Due to the double-entry data structure, this measure has the same descriptive statistics as the outcome measure (see Kenny et al., 2006), although its nesting in the dataset makes it a unique variable (also see Campbell & Kashy, 2002).

Attitudinal Self-Control

Two measures of self-control are used in this project. The first, attitudinal self-control, is captured through the frequently-used scale developed by Grasmick, Tittle, Bursik, and Arneklev (1993). This measure, which contains 24 items, captures the six subdimensions of self-control–impulsivity, a preference for simple tasks, a preference for physical activities, risk-seeking, self-centeredness, and temper – which were defined by Gottfredson and Hirschi (1990). This average-item-score scale, which fits the data consistently (Cronbach’s α =.84), is coded so that higher scores measure higher self-control levels (M = 2.878, SD = 0.350, range 1–4). The actor’s and the friend’s attitudinal self-control levels are both utilized as independent variables in the forthcoming analysis.

Behavioral Self-Control

Behavioral measures of self-control are preferred by self-control theory’s original authors (see Hirschi & Gottfredson, 1993). To capture behavioral self-control, we use the reduced version of the Retrospective Behavioral Self-Control Scale (the ‘RBS’; see Marcus, 2003). The original RBS contained 67 items which inquired about behaviors relevant to self-control during the age ranges of 8–13, 14–18, and 19–25 years of age. The items are measured on a scale of 1, indicating the respondent ‘never’ behaved in that manner, to ‘7,’ indicating that he/she ‘always’ behaved that way.

Despite having the advantage that the scale asks about behaviors rather than attitudes, many of the original items on the RBS were tautological because they directly asked about crime. Using item-response modeling in conjunction with face validity tests, Ward, Gibson, Boman, and Leite (2010) reduced the 67 item RBS to 18 items that showed both face and internal validity. Ward and colleagues’ items were used to construct an average-item-response scale – the RBS reduced version (RBS-r) – in this study. The RBS-r is coded so that higher scores capture higher self-control (M = 5.596, SD = 0.687, range 0–7). The items scale consistently (α = .82), and actor and friend measures of the RBS-r are used as independent variables in forthcoming analyses. The wording of the items can be seen in Appendix Table 4.

Controls

Models are estimated with demographic controls of the actor and the friend. First, we control for the sex (1 = male [33.6% of the sample]; 0 = female) of the actor and the friend. Second, we control for the race of the actor and friend, which is defined as a dichotomy that compares those who are non-white (coded ‘1’; 36.9% non-white) to those who are white (coded ‘0’). Third, we include a standalone measure of whether the respondent identified as being of Hispanic ethnicity (1 = Hispanic [18.6% of the sample]; 0 = non-Hispanic). Fourth, and finally, we include a measure of the age of the actor and the friend, measured as a count (M = 19.339, SD = 1.433).

Analytical Strategy

With naturally nested data, a multilevel analysis is necessary. Accordingly, this study employs the use of two-level, mixed-effects hierarchical linear models. These models fall under the classification of actor-partner interdependence models, a type of analysis specifically designed for dyadic data (see Kenny et al., 2006). In this case, a mixed model is necessary because there is variation both within dyads (friends are different from each other) and between dyads (friendships are different from each other). The level 1 equation will include characteristics of the actor and the partner and is the focal level of interest in this study. Level 2 in the forthcoming models is called a ‘grouping’ level, as it simply groups the level 1 equation around the dyadic friendship identification variable.

Very minor amounts of missing data (less than 1% missing on all variables) were imputed using a Markov-chain Monte Carlo imputation technique (20 draws from 200 burn-in iterations). Models using listwise deletion (not reported) showed similar results. All models were estimated with Stata version 14.2.

Results

Primary Findings: Social Learning, Self-Control, and Versatility

Results relevant to the first research question are presented in a series of mixed models in Table 2. Model 1 regresses the actor’s versatility in offending onto actor measures of perceptual peer versatility, attitudinal self-control (Grasmick et al., 1993), and controls. The actor’s perception of the peer’s versatility is positive and highly significant (b = .511, SE = .021, p ≤ .001), indicating that actors who perceive their peers are more diverse in offending are more diverse themselves. Additionally, the actor’s attitudinal self-control is negative and also significant (b = −.043, SE = .003, p ≤ .001), suggesting that those who have lower levels of attitudinal self-control tend to have a higher versatility in offending. Although most controls do not approach levels of statistical significance, males (b = .337, SE = .054, p ≤ .001) report significantly more offending versatility than women.

Table 2.

Mixed models regressing the actor’s offending versatility onto characteristics of the actor and the friend (attitudinal self-control results; N = 2154)

	Model 1		Model 2		Model 3		Model 4
	b	SE	b	SE	b	SE	b	SE
Level 1: Person-Level
Actor Effects
Perceptual peer versatility	.511	.021***	.514	.022***	-	-	-	-
Self-control (attitudinal)	−.043	.003***	−.043	.003***	−.056	004***	−.056	004***
Male	.337	.054***	.372	.059***	.425	.060***	.436	.065***
Non-white	−.086	.053	−.074	.060	−.206	.059***	−.158	.067*
Hispanic	.109	.066	.099	.068	.143	.073	.125	.077
Age	−.023	.018	−.022	.022	−.013	.020	−.015	.024
Friend Effects
Self-reported versatility	-	-	-	-	.192	.020***	.186	.022***
Self-control (attitudinal)	-	-	.000	.003	-	-	-.003	.004
Male	-	-	-.088	.059	-	-	-.042	.065
Non-white	-	-	-.029	.060	-	-	-.102	.066
Hispanic	-	-	.040	.067	-	-	.060	.075
Age	-	-	-.002	.021	-	-	.004	.024
Constant	4.420	423***	4.437	.492***	5.498	.463***	5.666	.559***
Level Two: Dyad-Level
τ	.000	.000	.000	.000	.000	.000	.000	.000
σ	1.212	.038	1.210	.038	1.513	.048	1.510	.048
Model Statistic
F	196.34		108.22 ***		94.19		51.99***

^*p ≤ .05

^**p ≤ .01

^***p ≤ .001

Model 2 of Table 2 adds friend measures to the level 1 equation. However, no friend measures reach statistical significance, and the overall patterns of significance in the actor effects from Model 1 remain unchanged. However, results do change in Model 3 of Table 2, which removes the peer effects but adds a measure of self-reported peer versatility captured directly from the peer’s self-reports. This measure reaches high levels of statistical significance (b = .192, SE = .020, p ≤ .001), and the direction indicates that those who have friends who self-report high levels of versatility in their offending are versatile themselves. However, the coefficient strength of the perceptual measure in Model 2 (b = .514) is much stronger than the coefficient in Model 3 (b = .192) with approximately the same standard error, suggesting that the perceptual measure of peer versatility is more impactful than the peer’s self-report. Finally, white (b = −.206, SE = .059, p ≤ .001) male (b = .425, SE = .060, p ≤ .001) actors report significantly higher versatility in offending than non-whites and females, respectively.

The final model in Table 2, Model 4, once again adds in the friend effects to examine their relationship on the direct measure of versatility in peer offending. Namely, while attitudinal actor self-control is still significant (b = −.056, SE = .004, p ≤ .001), the peer’s level of self-control is unrelated to the actor’s versatility in offending. Besides the peer’s self-reported offending versatility (b = .186, SE = .022, p ≤ .001), no other peer measures reach significance and the actor effects are similar to those in Model 3.

Table 3 presents a similar set of mixed models as reported in Table 2, but instead removes the attitudinal measure of actor and friend self-control and replaces it with the behavioral measure (the RBS-r). InModel1 of Table 3, the actor’s perception of the peer’s versatility is significant and positively related to the actor’s offending versatility (b = .519, SE = .021, p ≤ .001). The actor’s behavioral level of self-control is negative and highly significant (b = −.456, SE = .043, p ≤ .001), suggesting that actors with lower behavioral self-control are much more likely to be versatile in their offending. Additionally, males (b = .398, SE = .054, p ≤ .001) and those of Hispanic descent (b = .133, SE = .065, p ≤ .05) are more likely to be versatile. Non-whites (b = −.121, SE = .053, p ≤ .05) and younger (b = −.049, SE = .018, p ≤ .01) actors are less likely to demonstrate versatility.

Table 3.

Mixed models regressing the actor’s offending versatility onto characteristics of the actor and the friend (behavioral self-control results; N = 2154)

	Model 1		Model 2		Model 3		Model 4
	b	SE	b	SE	b	SE	b	SE
Level 1: Person-Level
Actor Effects
Perceptual peer versatility	.519	.021***	.518	.022***	-	-	-	-
Self-control (attitudinal)	−.456	.043***	−.457	.043***	−.623	.004***	−.626	.047***
Male	.398	.054***	.410	.060***	.501	.060***	.487	.066***
Non-white	−.121	.053*	−.085	.060	−.256	.059***	−.179	.068**
Hispanic	.133	.065*	.126	.069	.175	.073*	.161	.077*
Age	−.049	.018**	−.044	.022*	−.049	.020*	−.046	.024
Friend Effects
Self-reported versatility	-	-	-	-	.200	.020***	.184	.022***
Self-control (attitudinal)	-	-	.009	.042	-	-	-.061	.050
Male	-	-	-.035	.060	-	-	-.032	.066
Non-white	-	-	-.076	.061	-	-	-.168	.067*
Hispanic	-	-	.022	.068	-	-	.037	.076
Age	-	-	-.010	.021	-	-	.009	.023
Constant	4.525	469***	4.689	.568***	5.814	.518***	6.344	.650***
Level Two: Dyad-Level
τ	.000	.000	.000	.000	.000	.000	.000	.000
σ	1.240	.040	1.238	.040	1.547	.049	1.540	.049
Model Statistic
F	182.33***		99.95***		83.50***		46.48***

^*p ≤ .05

^**p ≤ .01

^***p ≤ .001

Peer measures are added into Model 2 of Table 3. No peer measures are significantly related to the actor’s versatility in offending, including the peer’s behavioral self-control. However, the inclusion of these measures does reduce the actor’s race and ethnicity to levels of non-significance, although the actor’s perception of the peer’s versatility (b = .518, SE = .022, p ≤ .001) and behavioral self-control levels (b = −.457, SE = .043, p ≤ .001) remain significant at nearly identical levels as in Model 1.

When the actor’s perception of the peer’s versatility is removed and replaced by the peer’s self-reported offending versatility (Table 3‘s third model), the direct measure reaches high levels of statistical significance (b = .200, SE = .020, p ≤ .001) in a positive direction with the actor’s self-reported versatility. The actor’s behavioral self-control coefficient (b = −.623) also increases substantially in magnitude over the prior model (b = −.457). Finally, when peer covariates are added back into the model (Model 4, Table 3), the coefficient of direct peer versatility (b = .184) decreases slightly, but maintains high levels of statistical significance. Additionally, lower levels of the actor’s behavioral self-control (b = −.626, SE = .047, p ≤ .001) remain strongly related to versatility in offending. Male (b = .487, SE = .066) and Hispanic actors (b = .161, SE = .07, p ≤ .05) are more likely to offend in a versatile manner, whereas non-white (b = −.179, SE = .068, p ≤ .01) actors and those with non-white friends (b = −.168, SE = .067, p ≤ .05) are less likely to offend in a versatile manner.

Discussion and Conclusions

Drawing from social learning and self-control theories and using data from a large sample of people nested within friendship pairs, this study explored the relationships between specialization and versatility in offending among friendships. Results from a series of multilevel models demonstrated that versatility in offending is related to the actor’s self-control as well as the actor’s perception of his/her friend’s versatility. When friend effects were entered into the equations, the peer’s self-reported versatility also positively related to the actor’s versatility, although the peer’s self-control did not. This same basic pattern of significant findings was observed when the commonly used attitudinal measure of self-control developed by Grasmick et al. (1993) was interchanged with a behavioral measure of self-control developed by Marcus (2003) and refined by Ward et al. (2010). Despite the similarities in the significance patterns, the actor’s attitudinal self-control had a substantially weaker relationship with offending versatility than the actor’s behavioral self-control. In this section, we discuss the implications of our findings for social learning, self-control, and the broader context of offending specialization and versatility in the context of friendships.

Results from this study carry important implications for self-control theory. While findings demonstrated that the actor’s self-control related to offending versatility as self-control theory would clearly expect (Gottfredson & Hirschi, 1990), the friend’s level of self-control was not significantly associated with the actor’s versatility in offending. There are two ways this can be interpreted. First, the lack of a significant peer effect supports self-control theory because self-control is an intra-individual trait that should seemingly be uninfluenced by external forces. Accordingly, any peer effect should be either inconsequential, spurious, or an artifact of measurement error. As a consequence, the peer’s self-control should not have an effect on actor’s behavior – a theoretical tenet which our results strongly support.

A second interpretation of these results is also possible. The lack of a significant peer self-control effect on the actor’s offending versatility is, in a way, paradoxical. Gottfredson and Hirschi (1990) clearly outline that individuals with low self-control will form friendships with others who also have low self-control. Due to self-control causing a dual process of both friendship formation and offending versatility, the friend’s self-control should be related to characteristics of the actor’s offending because of a strong correlation that should exist between the friends’ levels of self-control. From this point of view, it is theoretically counterintuitive to expect that peer self-control – which should be virtually identical to the actor’s self-control – to be unrelated to offending. Accordingly, the results regarding peer self-control from the first perspective support self-control theory. However, from the second perspective, support is not provided to the theory. It appears that self-control theory’s hypotheses regarding peer effects compete with one another. Stated differently, the theory appears to contradict itself in regard to the meaning of peer influence.

Interestingly, Gottfredson and Hirschi do offer a potential explanation for the peer effect that is based in the concept of opportunity. According to Gottfredson and Hirschi (1990, p. 92), any specialized offending that occurs among individuals with low-self-control is “determined by convenience and opportunity.” It could be the case that individuals within friendships do have similar levels of self-control but experience varying levels of opportunity to commit crime and/or specialize in it. Indeed, the variation in opportunity among friends (see Osgood & Anderson, 2004) could very well explain the lack of a relationship between friend self-control and the actor’s level of offending versatility – a point which future research should explore. The opportunity construct will be imperative to resolving the contradictory expectations from Gottfredson and Hirschi’s (1990) theory regarding how the peer’s self-control should relate to an actor’s behavior.

Following prior literature (e.g., Boman, 2017; Evans, Cullen, Burton, Dunaway, & Benson, 1997), we included a measure of attitudinal self-control derived from Grasmick et al.’ (1993) work as well as a behavioral measure (from Marcus, 2003 and Ward et al., 2010; see Appendix Table 4 for more information). Despite similarity in the overall (in)significance patterns of the peer self-control effect for both the attitudinal and behavioral measures, there were differences for the actor’s self-control that were measurement dependent. While both the attitudinal and behavioral measures were significant, the coefficients of the actor’s behavioral self-control were much stronger than the coefficients of attitudinal self-control. As such, our findings demonstrate empirical support to Hirschi and Gottfredson’s (1993) preference for behavioral measures of self-control while also highlighting the utility of different operationalizations of the general theory’s primary construct.

In addition to carrying new and unique findings for how self-control relates to versatility in deviance, our results also carry implications for Burgess and Akers’ (1966; also Akers, 2009) theory of social learning. Results across all models clearly demonstrated that the construct of differential association is of importance in relation to offending versatility. Specifically, models using a perceptual measure of the versatility of the peer’s offending demonstrated that the differential association-informed construct was positive in direction and highly statistically significant. As such, the models suggest that actors tend to offend in more versatile ways when they think their peers are also versatile and, inversely, that actors tend to be more specialized when they believe their friends to be highly specialized. When the measure of differential association was changed from a perception of the peer’s deviance to the peer’s self-reported deviance, the results were largely the same: The measure was significantly related to offending versatility and positive in direction. Regardless of how the versatility construct is operationalized, differential association and social learning theories are supported (see, generally, Boman, Stogner, Miller, Griffin, & Krohn, 2012; Meldrum & Boman, 2013; Young, Rebellon, Barnes, & Weerman, 2014).

Overall, findings from the perceptual peer and peer’s self-reported versatility measures demonstrate a considerable amount of support for social learning and differential association theories. Despite the measurement technique, the measure of differential association related strongly to the actor’s offending versatility. While social learning theory (Akers, 2009) would expect the perceptual measure of peer versatility to be the most meaningful on the outcome, the theory would also hypothesize that the peer’s self-reported versatility would be influential as well (Akers, 2009, pp. 117–120). Differences in the coefficient sizes between the two measurement methods could easily be interpreted as being supportive of this very specific hypothesis from Akers. As such, social learning theory has received strong support as differential association relates to offending versatility regardless of how the construct is measured.

As research on social learning and specialization and versatility moves into the future, an important avenue of inquiry may lie in recognizing that some behaviors are more ‘groupy’ than others. That is, some behaviors are more likely to be committed in groups than other types of behavior. The term ‘groupy’ was developed by Warr (2002) and has become used in conjunction with the related term of ‘non-groupy’ (see Boman & Gibson, 2016) that completes Warr’s taxonomy. Examples of groupy behavior include vandalism, substance use, and status crimes. On the other hand, examples of non-groupy behavior include theft, certain types of serious assault, and sexual battery (see Warr, 2002). Research on the groupy and non-groupy issue has proven to be imperative in other contexts in criminology (e.g., Boman & Gibson, 2016; Boman & Mowen, 2018; also see Beaver et al., 2011 for a related study), and there is ample reason to believe that the same issue may also be of importance in the specialization and versatility context. The groupy/non-groupy taxonomy may also help research segue into the notion of friendship-level specialization, an important theoretical and policy-based issue that – unfortunately – has remained largely unexplored by criminologists to date (for the exceptions, see McGloin & Piquero, 2010; Thomas, 2016; Warr, 1996). There are also measurement issues surrounding perceptual peer specialization and versatility that are immediately relevant to the difference in the coefficient magnitude between the perceptual and direct peer specialization measures. Similar in concept to issues surrounding peer deviance, there has been no literature to date investigating how peer specialization could be or should be measured. This literature would be very useful, and it could incorporate Warr’s groupy and non-groupy taxonomy to determine how the nature of specialization impacts our understanding of specialization and versatility in general and in a friendship context specifically.

Despite some valuable findings and necessary future directions regarding how social learning and self-control relate to specialization and versatility in offending among friends, the current study has some notable limitations. First and foremost, the current data come from a cross-section of students attending only one American university. Not only does this limit our ability to determine causal order between our key variables of interest, our findings are likely to be of limited generalizability to those in other populations. An overrepresentation of females may also have limited the variance in our dependent variables. Second, while a strength of the current study is the dyadic design, dyads are an inherently limited unit of analysis. People typically have several friends, meaning they are nested within multiple dyads. Unfortunately, our data contain only one of a person’s potentially many dyadic friendships. A similar analysis using social network data would be valuable as it would build upon recent work by Thomas (2016) and McGloin and Piquero (2010). Such a study should seek to explore other methods of capturing specialization, including the methods developed by Osgood and Schreck (2007) as well as the method by DeLisi et al. (2011). Third, our outcome and key predictor measures are constructed from only twenty items spanning five types of offending. We highlight that there are different types of control (DeLisi & Vaughn, 2014) and many more types of offenses, meaning the variability in these measures is limited. Although our study has some limitations, the findings nonetheless further develop the state of knowledge as to how offending versatility relates to constructs derived from two of our field’s most prominent theories.

Findings from this study offer strong support to both self-control (Gottfredson & Hirschi, 1990) and social learning (Akers, 2009) perspectives’ expectations on offending specialization and versatility. However, what is most apparent is the many future directions of necessary exploration to which findings from this study raise attention. Future research should seek to explore how self-control and opportunity (Osgood & Anderson, 2004) interact for versatility in the friendship context. Future research on this topic also has the ability to remediate a notable inconsistency regarding how peer effects should operate within the context of self-control theory. This study highlights that the lack of a peer effect supports the theory. But, at the same time, low self-control theoretically causes friendships to form while also causing both offending and versatility in offending, forming a strange paradox in how Gottfredson and Hirschi view friends and friendships. Additionally, certain behaviors are more (or less) likely to be committed in groups than others (Warr, 2002), and the notion of friendship-level specialization is emphasized (see McGloin & Piquero, 2010; Warr, 1996; 2002). Perhaps people engage in specialized behaviors when around only certain friends – that is, a person may only consume alcohol with one friend, and may only use opioids with another friend (see Warr, 2002, p. 38). If this were the case, then offending may appear versatile overall (e.g., Farrington, 2003) even though friendship-level contexts may carry overlooked, but nonetheless important, patterns of specialization. If this were true, the current understanding that people ‘do not specialize’ in offending may be incorrect at worst or misleading at best. Either way, the notion of friendship-level specialization must take precedence in the research on specialization and versatility due to the established importance of peers and the applicability of friendship-level events to many different types of theory.

Biography

John Boman is an Assistant Professor at Bowling Green State University in the Department of Sociology. His research focuses on the roles of interpersonal influences, and particularly peers and friends, on crime, deviance, and substance use over the life-course. His recent work appears in Criminology, the Journal of Criminal Justice, and the Journal of Youth and Adolescence.

Thomas J. Mowen is an Assistant Professor in the Department of Sociology at Bowling Green State University. His research broadly examines the consequences of school security and punishment as well as the process of prison reentry. His recent research has appeared in Criminology, Justice Quarterly, Journal of Quantitative Criminology, and Journal of Research in Crime and Delinquency.

George E. Higgins is Professor in the Department of Criminal Justice at the University of Louisville. He received his Ph.D. in Criminology from Indiana University of Pennsylvania in 2001. His most recent publications appear or are forthcoming in Journal of Criminal Justice, Criminal Justice and Behavior, Justice Quarterly, Deviant Behavior, and Youth and Society.

Appendix

Table 4.

Wording of items in the behavioral self-control RBS-r scale

Item #	Item Wording and Retrospective Timeframe of Question
	During the ages of 8–13 …
1	I was well prepared for school exams.
2	I asked my parents for more pocket money because I had already spent my regular allowance.
3	I couldn’t follow lessons in class because I was busy doing other things.
4	I did my homework on time.
5	I spent my allowance long before I got my next one.
	During the ages of 14–18 …
6	There was something else when the time came to do my homework.
7	I tired of hobbies quickly.
8	I have been late for school or at work because I stayed out too late the night before.
9	I have made private dates or appointments and failed to show up.
10	I have ended a close friendship because of a romance.
	During the ages of 19–25 …
11	I have been late for important appointments.
12	I could have saved myself a lot of trouble if I had watched what I said.
13	I have bought things on the spur of the moment, which I really did not need.
14	I have said things to my partner in an argument which hurt her or him badly.
15	On vacation, I have spent all my money before the vacation was over.
16	In a bad mood, I have insulted people without any particular cause.
17	I have not been exactly tactful in disagreements with my boss or other people in authority.
18	I have bought something of considerable value without comparing prices beforehand.

All items measured on a scale of 1–7 (1 = never; 2 = once; 3 = two or three times; 4 = fairly many times; 5 = often; 6 = very often; 7 = always). For more information, see Marcus (2003) and Ward et al. (2010)