Learning disabilities and delinquency: The (non-)mediating and (non-)moderating role of peer deviance
Etmanski, B., Ryan, A. L., & Gallupe, O. (2024). Learning disabilities and delinquency: The (non-)mediating and (non-)moderating role of peer deviance. Canadian Journal of Criminology and Criminal Justice, 65(4), 24-50. https://doi.org/10.3138/cjccj-2023-0047
by Brittany Etmanski, Ashley L. Ryan, and Owen Gallupe
Published onMar 18, 2024
Learning disabilities and delinquency: The (non-)mediating and (non-)moderating role of peer deviance
·
Abstract
This study assesses the role of peer deviance in the relationship between learning disabilities and violence and property crime. Two possibilities are explored: a) that youths with a learning disability tend to have more deviant peers which in turn increases delinquent involvement (a mediating effect); b) that youths with a learning disability are more strongly influenced by the deviance of their peers (a moderating effect). We draw on the causality literature and employ a causal directed acyclic graph. Using data from the first two waves of the National Longitudinal Study of Adolescent to Adult Health (n=6,391), we find results that are not in line with either possibility. While adolescents with learning disabilities are shown to exhibit higher levels of violence (but not property crime), peer deviance is not found to play either a mediating or moderating role. We recommend future work test alternative mediating pathways, such as through victimization and self-control.
This research uses data from Add Health, funded by grant P01 HD31921 (Harris) from the Eunice Kennedy ShriverNational Institute of Child Health and Human Development (NICHD), with cooperative funding from 23 other federal agencies and foundations. Add Health is currently directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01 AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Add Health was designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill.
Introduction
Ample evidence has found that youths with disabilities tend to be more heavily involved in delinquent behaviour than youths without a disability (e.g., Mallett 2014; Quinn, Rutherford, Leone, Osher, and Poirier 2005), but why this is the case remains unclear. Scholars have called for the link between disability and delinquency to be examined further, pointing at the possibility of mediating or moderating factors that may account for the association between disability and delinquency (see Evans, Clinkinbeard, and Simi 2015). Assuming there is a link, there are reasons to suspect that peer relations play an important role. Using panel data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) and drawing on social learning theory, we address the following question: do deviant peer associations mediate or moderate the relationship between learning disabilities and offending?
Learning Disabilities and Delinquency
Disability is a term that has been deeply contested. Medical sociological approaches to disability, according to Thomas (2007: 12), approach disability as a product of an injury (developmental, illness, accident) which results in “disablement” for which science ideally offers “corrective solutions.” The social model of disability (Oliver 1996) argues that exclusion is a product of social construction rather than impairment and is characterized by experiences of social oppression (Thomas 2007).1 Despite occasionally heated arguments on both sides, Thomas (2007: 13) notes that both approaches recognize that people with a disability (however it is defined) often experience discrimination. This common thread is reflected in Barclay's (2019: 16) definition of disability (which draws on the International Classification of Functioning, Disability and Health - WHO 2001: 16) as “an impairment associated with some disadvantage, including, but not limited to, loss of functioning either with respect to the ability to execute certain tasks or to participate fully in some aspect of social life.” However, Barclay acknowledges that every aspect of that definition (impairment, disadvantage, function loss, task execution) is vague, though she suggests that this is ideal given the variability of experiences of disability. More precise definitions would necessarily miss people who should be included. Moreover, Barclay makes the case that what is more important than an agreed upon definition is having a reasonable conception of who members are of the population experiencing conditions associated with disability. From an empirical perspective, that requires a sensible working definition that will, with as little error as possible, identify people experiencing these conditions. Unfortunately, classification error is an intractable problem. The result is likely that the more seriously a person is affected, the more likely they are to be classified as having a disability (in the case of our work, a learning disability). And yet, this accords with the “some disadvantage” element of Barclay's definition. The flip side is that the dynamics we present are likely to be less relevant to those who are not as disadvantaged by their learning issues (regardless of the reason for the lesser disadvantage).
Over the last 20 years, studies employing varying methods have generally shown that persons with learning disabilities tend to be more heavily involved in offending behaviour than persons without. For example, Quinn et al. (2005) attempted to collect a census of state juvenile corrections systems to assess the extent to which youth with disabilities populate the justice system. They noted that 33% of justice-system-involved youths had a disability; 39% of those (13% of the total) had a learning disability. This is higher than the 9% of the general population at the time that had any disability. Barrett, Katsiyannis, Zhang, and Zhang (2014) matched juvenile justice records to a non-delinquent sample in South Carolina and found that the presence of a learning disability was a predictor of membership in the delinquency group. Expanding on this work, Barrett and Katsiyannis (2016) demonstrated that a learning disability as an adolescent was a risk factor for arrest as an adult. Mallett’s (2014) examination of youth court records in a midwestern US state found that youths with learning disabilities were more likely to be suspended from school, receive a criminal record at a younger age, and appear in court more frequently. Shandra and Hogan (2012), using self-reports from the National Longitudinal Survey of Youth, found that youths with learning/emotional disabilities were more likely to steal, carry a weapon, and physically attack others. and commit a variety of violent acts.
The major caveat to this trend is the fact that Simpson and Hogg’s (2001: 394) systematic review offered little evidence to suggest that criminal activity is higher among persons who have intellectual disabilities compared to those without, though, of course, intellectual disabilities are not synonymous with learning disabilities. However, Malmgren, Abbott, and Hawkins (1999) show similar non-effects for learning disabilities.
How, then, do we reconcile this? One potential avenue is that the effect of learning disabilities on delinquency is conditional upon, or mediated, by other factors. There is some evidence in existing literature that the relationship is more complex than a simple direct effect. For example, Savage and Bouck (2017), examining intellectual disabilities, found that the severity of the disability moderated the relationship such that youth with more mild disabilities were more actively involved in delinquent behaviour. Fogden, Thomas, Daffern, and Ogloff (2016), also focusing on intellectual disabilities, showed that the way disability related to criminal behaviour was dependent on the type of offence (less likely to have an official record, but more likely to commit a violent or sexual offence). In a similar vein, Evans et al. (2015) demonstrated that youth with learning disabilities were more likely to report participating in a violent act but they found no significant difference between these groups on property crime, or tobacco and marijuana use (210). Mendoza, Blake, Marchbanks, and Ragan (2020) highlighted complex interactions between race, gender, and learning disabilities in predicting juvenile justice system contact (see also Mallett, Quinn, Yun, and Fukushima-Tedor forthcoming). Drawing on learning theories of criminal behaviour, we suggest that probing the role of peers might help to clarify the uncertain link between learning disabilities and delinquency.
Social Learning Theory
The main premise of social learning theory is that individual actions are shaped through interactions with others in which mechanics, motivations, and rationalisations for the behaviour are learned (Akers 2009; Sutherland 1947). Social learning theory has been used to explain a wide set of offenses (e.g., substance use, domestic violence, theft) (Gallupe, Nguyen, Bouchard, Schulenberg, Chenier, and Cook 2016; Øygard, Klepp, Tell, and Vellar 1995; Sellers, Cochran, and Branch 2005). The primary criminological form of social learning theory was first introduced in Sutherland’s differential association theory (1947) in which he argued that criminal behaviours and the motivations and attitudes underlying those behaviours are transmitted via social interactions. The influence of these social interactions are thought to be stronger among more important relationships (e.g., friends compared to strangers) (Akers, Krohn, Lanza-Kaduce, and Radosevich 1979: 638).
Akers’ (2009) version of social learning theory expanded on Sutherland by adding the elements of imitation, definitions (attitudes), and reinforcement to the theory. Imitation is the idea that people learn how to behave in social situations by watching others act. Seeing others offend not only shows individuals how to act to successfully perpetrate the act but also sends the message that others are likely to view them positively if they commit similar acts. The definitions (attitudes) element states that the meanings and perspectives an individual applies to a behaviour are informed by interacting with others. For example, a person is more likely to internalize the perception of drug use as positive to the extent that they spend time around people who use drugs. The final element of Akers’ social learning theory is reinforcement. This element suggests that the likelihood of a person repeating a learned action is influenced by the type of reinforcement they receive. Receiving negative reactions from others for committing a particular act makes it less likely they will continue behaving this way in the future. Getting caught or having to pay a fine constitute unfavourable reactions that tend to reduce the likelihood of future acts. Positive reinforcements (e.g., excitement, goods), on the other hand, may contribute to the continuation of the behaviour. Reinforcements may also be social (e.g., social rejection or acceptance).
The differential association element is essentially an umbrella under which the other components operate. By associating with peers who are more heavily involved in delinquent behaviour, one is more likely to witness behavioural models of delinquency, receive reinforcements that support delinquency, and develop pro-delinquent attitudes. There is no reason to assume that social learning dynamics will operate in a different way for persons with and without learning disabilities. However, we hypothesize that having a learning disability may influence either the form of, or reaction to, peer associations which shapes delinquent involvement.
Learning Disability, Delinquency, and Peer Dynamics
If deviant peer associations shape the relationship between learning disability and delinquency, it must be the case that, in the aggregate, youths with and without learning disabilities differ in their peer relations. There is some evidence to suggest this may be the case. Stone and La Greca (1990) examined the type of social difficulties students with learning disabilities face. In a sample of 547 4th to 6th graders, their results indicated that students with learning disabilities had lower social status, were less popular, and were more rejected and neglected compared to students without learning disabilities. Bruefach and Reynolds (2022) demonstrated that students with learning disabilities had fewer friends; Hoffmann, Wilbert, Lehofer, and Schwab et al. (2021) found similar findings regarding students with special education needs. Additionally, Estell, Jones, Pearl, Van Acker, Farmer, and Rodkin (2008) analyzed the isolation and social standing of students with learning disabilities. Using a sample of 1,361 students in grades 3 to 6, they found that those with learning disabilities were generally part of peer groups that were similar in size to other students’ peer groups but were viewed as having a lower social standing compared to students without disabilities. This effect held over time. Jones, James, and Johnson (2023) demonstrated that students with learning disabilities had the same social acceptance and popularity goals as others but tended to be members of less popular groups. Hogan, McLellan, and Bauman (2000) noted that Australian school students with a disability (e.g., learning, physical, sensory) were just as likely as non-disabled students to have friends but felt lonelier and more isolated and were more subject to peer pressure.
Given the evidence that having a learning disability can be socially deleterious, the question remains how that translates to delinquency levels. One possible mechanism is that having a disability may constrain the available peer association options to others who also have constrained peer choices. Some literature suggests that certain personality characteristics are associated both with reduced breadth in peer choices as well as with increased involvement in deviant behaviour. For example, Gottfredson and Hirschi (1990) argued that low self-control tends to make a person a bad friend which reduces the pool of people willing to associate with them. Their limited friendship options lead to a situation where the friends they make are with other people who also have few friends. While Gottfredson and Hirschi made this point to suggest that people with low self-control tend to hang out together, if disabilities are socially detrimental then a similar limitation on friendships may be placed on people with learning disabilities. That is, whether it is due to low self-control, a learning disability, or some other reason, people who have constrained friendship choices are likely to draw friends from others with similar constraints. Because of this, having a disability may increase the likelihood of having friends with low self-control. Having friends with low self-control is likely to mean greater exposure to deviant peers given the widely acknowledged link between self-control and offending (e.g., Pratt and Cullen 2000; Vazsonyi, Mikuška, and Kelley 2017). It seems possible, then, that people with learning disabilities may have more delinquent peer groups. If this is the case, deviant peers are likely to mediate the relationship between having a learning disability and delinquency (i.e., learning disabilities delinquent peers delinquency).
However, it is also possible that adolescents with a learning disability are more susceptible to peer influence. As suggested by R. Pearl (2002), the fact that adolescents with learning disabilities tend to have low social status makes them particularly receptive to peer messages for fear of losing friends. This echoes both Baumeister and Leary’s (1995) well-established need to belong theory as well as criminological research on the criminogenic influence of social status motives (e.g., Faris and Ennett 2012; Gallupe and Bouchard 2015). Research on peer conformity among youths with learning disabilities has a long history (e.g., Bryan, Werner, and Pearl 1982; Bryan, Pearl, and Fallon 1989), however not much appears in the criminological literature and there is even less that draws explicitly on the causality literature. There does appear to be sufficient justification to expect that peer deviance will moderate the effect of learning disabilities on delinquency (i.e., learning disabilities X delinquent peers delinquency). Overall, it appears that the role of peers in the link between learning disabilities and delinquency is not a settled question.
Current Study
Overall, then, the literature to date highlights three things: 1) a link between having a learning disability and offending behaviour; 2) the socially detrimental nature of having a learning disability; and 3) a well-established link between deviant peers and offending. The fact that previous research has not conclusively demonstrated why (1) is the case provides the impetus for investigating the role of peer deviance given (2) and (3). We will examine two possibilities in this study. The first possibility is that peer deviance mediates the causal relationship between having a learning disability and delinquency. If, in fact, peer choices for those with a learning disability are more constrained to others with a predisposition to offending, then the causal effect of learning disabilities on crime may be mediated by associations with deviant peers. The second possibility is that learning disabilities moderate the effect of peer deviance on offending. Again, owing to the negative social impact of having a disability, if peer choices are constrained for people with a disability, it may make them more fearful of losing relationships which are not as easily replenished as they are for others. Youths with learning disabilities may therefore be more receptive to deviant peer influence as a way to maintain relationship harmony. These possible mechanisms can be summarized in two hypotheses:
Hypothesis 1: Youths with learning disabilities will have more deviant peers and will, accordingly, be more involved in delinquency. In other words, the effect of learning disabilities on delinquency will be mediated by peer deviance.
Hypothesis 2: The effect of peer deviance on delinquency will be stronger for people with learning disabilities. In other words, learning disabilities will moderate the effect of peer deviance on delinquency.
The causal assumptions of this analysis are encoded in the causal directed acyclic graph (DAG) shown in figure 1. The focal causal and mediating variables (learning disability, peer delinquency) are embedded within a broader causal model. This ensures proper model specification in that it can be used to determine what controls are needed to close all backdoor biasing pathways and to avoid “bad” controls (Cinelli, Forney, and Pearl forthcoming); i.e., controlling on collider variables that actually open biasing pathways when controlled (Novak, Boutwell, and Smith forthcoming). This approach draws heavily on the causality literature (e.g., Cinelli, Forney, and Pearl forthcoming; McElreath 2020; Morgan and Winship 2014; J. Pearl 2009). Using the “dagitty” tool (Textor, van der Zander, Gilthorpe, Liśkiewicz, and Ellison 2016), we isolated the minimal sufficient adjustment set (MSAS) required to interpret both a total and direct causal effect of learning disabilities on criminal behaviour mediated and moderated by peer delinquency. That causal model is up for debate, but assuming its plausibility, the MSAS for the total effect requires only that race and sex be controlled. The MSAS for the direct effect is more comprehensive; it consists of self-control, social bonds, parenting practices, unstructured socializing, victimization, neighbourhood cohesion, neighbourhood disadvantage, sex, age, socio-economic status, and race/ethnicity. Negative life events and perceptual deterrence were not required to estimate a direct causal path between learning disability and delinquency (i.e., they did not close any backdoor biasing pathways) and were therefore left out of the mediation and moderation models.
Figure 1. Causal directed acyclic graph.
Note: perceptual deterrence is treated as unobserved since there are no suitable measures in the Add Health data.
A graph of dislocations Description automatically generated
Methods
This research employs data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) (Harris 2009). The Add Health sample consisted of 90,118 participants drawn from 80 high schools and 52 middle schools in 1994-95 that, when sample weights are employed, is representative of the US school population on region, school size and type, ethnicity, and urbanicity (Harris, Halpern, Whitsel, Hussey, Killeya-Jones, Tabor, and Dean 2019). Of those eligible to participate in the in-school sample, 20,745 adolescents and 17,670 parents/guardians (from which the learning disability measure was drawn) participated in the more extensive wave 1 in-home survey. Of these, 14,738 also participated at wave 2 approximately a year later; 13,570 were given valid sample weights. As per the recommendations of Chen and Chantala (2014), cases with missing weights were dropped. Of those 13,570, a total of 8,015 had valid peer delinquency scores. The reduction at this stage is attributable to the fact that the constructed network variables provided by Add Health are only available for students in schools where more than 50% of students completed the survey to avoid presenting “misleading images of the school's social structure” (Harris and Udry 2001: 1). It is unclear what impact on results this might have. After listwise deletion, there were 6,391 remaining cases [see dis-02b-valid.do]. This is the final sample used in the analysis.
Measures
Dependent Variables
Two dependent variables were examined: violent and property crime (wave 2). Following the recommendations of Sweeten (2012), these two scales consisted of the sum of binary indicators of any involvement in each of a number of related acts in the past 12 months. Variety scales measure the number of different types of criminal acts reported by respondents. However, they also correlate with frequency and severity such that respondents with higher scores not only report greater diversity in offending, but also tend to be more frequent and severe offenders (Farrington 1973). The 6-item violence variety scale included threatening/use of a weapon, group fights, pulling a knife/gun, shooting/stabbing someone, getting in a physical fight, and hurting someone badly enough to need a doctor (mean=0.50). Property crime consisted of the following 8 items: graffiti, damaging property, taking something from a store, driving a car without the owner’s permission, stealing something worth more than $50; stealing something worth less than $50, going into a house/building to steal, and selling marijuana/other drugs (mean=0.79).
Table 1. Survey-corrected descriptive statistics.
Mean/%
SE
Min
Max
Violence
0.495
0.024
0
6
Property crime
0.793
0.031
0
8
Learning disability
11.3%
Peer deviance
5.940
0.158
0
36
Impulsivity
2.227
0.014
1
5
Attachment to school
3.853
0.020
1
5
Attachment to parents
4.667
0.012
1
5
Unstructured socializing
1.983
0.019
0
3
Victimization
0.657
0.023
0
9
Neighbourhood cohesion
76.921
0.861
0
100
Neighbourhood disadvantage
-0.038
0.044
-2.570
3.221
Socio-economic status
6.222
0.084
2
10
Age
14.826
0.133
11
20
Male
47.7%
White
75.6%
Black
12.4%
Hispanic
3.3%
Other
8.7%
n=6,391
Following the approach of Mallett et al. (forthcoming), learning disabilities were coded 1 if the participants’ parent/guardian indicated that they “have a specific learning disability, such as difficulties with attention, dyslexia, or some other reading, spelling, writing, or math disability” and/or have received special education services in the past 12 months. All others were coded 0. The simplicity of this operationalization is appealing, and it likely does a good job isolating those with more serious learning issues. However, it does not capture the fact that learning disabilities are on a continuum of severity as opposed to being a discrete phenomenon. Those with minor learning acquisition issues are unlikely to be coded as learning disabled following this approach and therefore the peer dynamics being tested here are likely to be applicable only to the more seriously learning-disabled group. Approximately 11% of the sample were reported to have a learning disability.
To create the peer deviance measure, we used the wave I in-school friendship nomination data so that peer reports of deviant involvement could be linked back to each respondent. Self-reported involvement in each deviant act was summed across all of a respondent’s friends then divided by the number of friends to generate an average peer deviance score for each act. The peer group was defined as the people who nominated the respondent as a friend (the receive network). Peer group averages of the following variables were summed to create the peer deviance measure: tobacco use, alcohol consumption, number of times inebriated, number of times raced (e.g., car, bike, boat), participating in dangerous dares, and skipping school without an excuse (all coded 0=never to 6=nearly every day in the past 12 months - mean=5.94).
A variety of approaches to coding self-control with Add Health have been used; we opted to follow the lead of Haynie, Doogan, and Soller (2014) and Vazsonyi, Cleveland, and Wiebe (2006) by focusing on the impulsivity component of self-control. To do this, we calculated the average score (all measured from 1=strongly agree to 5=strongly disagree) on the following 4 items: “When you have a problem to solve, one of the first things you do is get as many facts about the problem as possible;” “When you are attempting to find a solution to a problem, you usually try to think of as many different ways to approach the problem as possible;” “When making decisions, you generally use a systematic method for judging and comparing alternatives;” and “After carrying out a solution to a problem, you usually try to analyze what went right and what went wrong” (mean=2.23).
Attachment to school was used as a proxy for social bonds more broadly. This measure was created by taking the average score of 3 questions (all 1=strongly disagree to 5=strongly agree) asking participants about feeling close to people at school, feeling part of the school, and being happy at school (mean=3.85). Attachment to parents was used to index parenting practices more broadly under the assumption that good parenting practices are reflected in strong attachments. Participants were asked to indicate how close they are to their mother and father and how much their mother and father care for them; the mean of these four items was taken. For those with a single parent, the 2-item mean was used (mean=4.67).
Unstructured socializing was measured by the following single item: “During the past week, how many times did you just hang out or talk with friends?” (0=not at all to 3=5+ times). While this is a blunt measure of unstructured socializing, other scholars have made the case that it does a reasonable job of capturing the essence of the concept as participants answered it after being asked about more structured activities (Augustyn and McGloin 2013; Haynie and Osgood 2005; Meldrum and Barnes 2017), though it is not explicit about the absence of authority figures (mean=1.98).
Following Haynie and Piquero (2006), the victimization measure focused specifically on violence as it is likely to cause the greatest perceived strain. The variable was created by taking the five-item sum of questions that asked about having a knife/gun pulled on them, being shot, being cut/stabbed, getting into a physical fight, and getting jumped (all 0=never to 2=more than once) (mean=0.66).
Drawing on work by Harding (2009) and Dawson, Wu, Fennie, Ibañez, Cano, Pettit, and Trepka (2019), perceived neighbourbood social cohesion was assessed with the following three items: “In the past month, you have stopped on the street to talk with someone who lives in your neighborhood,” “You know most of the people in your neighborhood,” “People in this neighborhood look out for each other” (all 0=no, 1=yes). We coded this as the percentage of maximum possible for those who responded to any of the three items (mean=76.92).
Neighbourhood disadvantage was measured by standardizing then taking the mean of the following census tract-level variables (Harding 2009): family poverty rate, proportion of single-mother households, median age (reversed), proportion of unemployed males, proportion Black, proportion of population over age 25 who are college graduates (reversed), proportion in managerial/professional occupations (reversed), and proportion of affluent families (yearly income $75,000+ - reversed) (mean=-0.04).
Socio-economic status was calculated by taking the parent score with the highest value on an index combining education and occupation (mean=6.22). Biological sex (0=female, 1=male), age (mean=14.83), and race/ethnicity (0=White, 1=Black, 2=Hispanic/Latino, 3=other) were also controlled.
Analyses
To test the total effect of learning disabilities on delinquency, negative binomial models were estimated that controlled for race and sex as they are the only covariates needed to close all biasing paths. For the mediation analysis, we follow the approach recommended by Zhao, Lynch, and Chen (2010), drawing on the work of Preacher and Hayes (2004; 2008), who suggested estimating indirect effects via simultaneous equations and calculating the significance of indirect effects via bootstrapped, bias-corrected confidence intervals (2000 replications). Simultaneous equations were estimated using sem in Stata (StataCorp 2021). To test for an interaction effect, disability by peer deviance product terms were included in bootstrapped linear regressions (2000 repetitions). This was supplemented by negative binomial interaction models. Notwithstanding some scepticism over their use, multiplicative terms are important for testing moderation hypotheses in nonlinear models (see Kaufman 2019: 18). There are, however, complications related to the fact that “the nonlinearity of the link function can create the appearance of an interaction effect when there is none” (177-178) since graphical displays can confound the nonlinearity of the interaction with the nonlinearity of the model (145). As such, plotting the predicted outcome score can be misleading. Confounded nonlinearity can be addressed by plotting predicted counts on a logarithmic scale. Using Kaufman’s (2019) icalc suite of commands for Stata, we present dual axis plots in the supplementary material which displays the predicted logged count of delinquency for each disability group across the range of peer deviance scores. The units of the left-hand y-axis are in the log of the delinquency variable; the right-hand y-axis shows the corresponding unlogged metric. These lead to the same conclusions as the bootstrapped linear regression interaction results.
Sample weights were used in all models to ensure representativeness of the broader population of American school-based adolescents. Survey design characteristics (students clustered in schools stratified by region) were accounted for using Stata’s (StataCorp 2021) svyset capabilities as per the recommendations of Chen and Chantala (2014)
Results
The survey-corrected negative binomial models estimating the total causal effect of learning disabilities on delinquency showed that learning disabilities are causally related to violence (b=0.229, IRR=1.258, p=.002) but not property crime (b=0.012, IRR=1.012, p=.901).2 Compared to adolescents without learning disabilities, those with learning disabilities are predicted to score 26% higher on the violence scale but at the same level for property crime.3
Mediation Analysis
Interestingly, there appears to be a whole lot of nothing in the results in terms of support of the mediating and moderating hypotheses. While there were positive coefficients for learning disability (relative to no disability) as a predictor of peer deviance, they did not reach traditional levels of statistical significance (b=0.133, p=.553). Furthermore, there were no significant direct effects of learning disability on either violence (b=-0.01, p=.789) or property crime (b=-0.10, p=.242). The indirect effects of learning disability on either form of delinquency via peer deviance were also non-significant (the confidence intervals crossed zero). By the typology of Zhao et al. (2010), this is evidence of no-effect nonmediation. Supplementary causal mediation analyses using the approach described by Imai, Keele, and Tingley (2010) and implemented in the medeff package in Stata (Hicks and Tingley 2011) confirmed the non-mediation results.
Table 2. Mediating effect of peer delinquency in the relationship between learning disability and delinquency.
Disability (=1) peer deviance (mediator)
Disability (=1) delinquency
Indirect effect of disability delinquency through peer deviance
b
SE
b
SE
ab
95% BCCI
Violence model
0.133
0.224
-0.011
0.041
0.002
-0.004 to 0.011
Property crime model
0.133
0.224
-0.102
0.087
0.004
-0.010 to 0.020
* p<.05 ** p<.01; n=6,391
Note: Statistical significance of indirect effects are based on bias-corrected standard errors and confidence intervals estimated with 2,000 bootstrap samples. Models control for impulsivity, attachment to school, attachment to parents, unstructured socializing, victimization, neighbourhood cohesion, neighbourhood disadvantage, sex, socio-economic status, age, and race/ethnicity.
Moderation Analysis
As displayed in table 3, the effect of peer deviance on either form of delinquency does not appear to be moderated by learning disability (p>.05 in both the violence and property crime models). Figures 2 and 3 show a positive slope for the relationship between peer deviance and delinquency, but with overlapping confidence intervals between those with and without a learning disability (note that in all models, the direct effect of peer deviance on delinquency is positive and statistically significant, as we would expect). The interaction plots derived from survey-corrected negative binomial models and plotted using Kaufman’s (2019) icalc tools do little to change the interpretation that peer deviance has a positive effect on delinquency that is unmoderated by learning disability (see supplementary material for these additional plots).
Table 3. Interaction between peer deviance and learning disability on delinquency.
Violence model
Property crime model
b
SE
b
SE
Peer deviance
0.016**
0.005
0.029**
0.005
Learning disability (=1)
0.031
0.089
-0.153
0.128
Interaction: peer deviance by learning disability (=1)
-0.007
0.013
0.008
0.019
* p<.05 ** p<.01; n=6,391
Note: Linear regression model with bootstrapped standard errors (2,000 replications). Models control for impulsivity, attachment to school, attachment to parents, unstructured socializing, victimization, neighbourhood cohesion, neighbourhood disadvantage, sex, socio-economic status, age, and race/ethnicity.
Figure 2. Interaction between learning disability and peer delinquency on violence.
A graph of a graph showing the difference between disability and peer delinquency Description automatically generated
Figure 3. Interaction between learning disability and peer delinquency on property crime.
Discussion
Drawing on social learning theory (Akers 2009; Sutherland 1947), the goal of the current study was to investigate why, in the context of research showing that learning disability status is related to involvement in deviant behaviour (e.g., Barrett et al. 2014; Quinn et al. 2005; Shandra and Hogan 2012) and that having a learning disability tends to be socially detrimental (Estell et al. 2008; Hogan et al. 2000; Stone and La Greca 1990), youths with learning disabilities tend to be more involved in delinquency. There were plausible arguments pointing both to a mediation effect and to a moderation effect. The basic idea behind the mediation hypothesis is that the social ‘penalty’ associated with having a learning disability might be that peer choices are constrained which leads to having friends who are more heavily delinquent. This in turn was thought to lead to higher levels of delinquent involvement via traditional influence processes. For the moderation hypothesis, the argument was that peer deviance might exert more of an influence on the delinquent behaviour of youths with learning disabilities. The social penalty associated with having a learning disability may have resulted in these youths being more fearful of losing friendships, and therefore more susceptible to peer deviance.
Overall, there was a significant total effect of learning disabilities on violence (but not property crime) such that adolescents with learning disabilities were predicted to score 26% higher on the violence scale. However, there was little support for either the mediating or moderating hypotheses connecting the learning disabilities effect on delinquency to peer deviance. The link between learning disability status and delinquency, either as a direct effect or mediated or moderated by peer deviance, did not hold when closing the biasing backdoor paths outlined in the causal DAG. So, learning disabilities do not appear to increase adolescents’ exposure to peer deviance, nor does it make them more susceptible to peer deviance. And yet, the association remains (at least for violence).
This begs the question of what to make of it all. A substantial amount of research has shown that having a learning disability is related to higher levels of deviance (Barrett et al. 2014; McNamara and Willoughby 2010; Shandra and Hogan 2012), though there is work that casts doubt on the trend (Malmgren et al. 1999). Our work falls somewhere in the middle. The fact that we find only a total effect of learning disabilities on violence, but no direct effect, suggests other mediating pathways might be helpful in explaining the association in ways that peer deviance does not. If the causal DAG presented in figure 1 is valid, victimization or self-control are plausible mediators and sensible next steps would be to investigate those. Agnew’s (1992) general strain theory and Gottfredson and Hirschi’s (1990) general theory of crime could be used to generate testable hypotheses that expand our understanding of the link between learning disabilities and delinquency. If adolescents with learning disabilities are more likely to be victimized and/or react differently to it than others, this could help explain the fact that learning disabilities are related to delinquency, but not in a direct way. Similarly, if people with learning disabilities are more likely to develop low self-control, then self-control would act as a mediator of the learning disabilities/delinquency relationship.4
Limitations
The measurement of learning disabilities places it at risk of misclassification. Asking parent’s if their child has a learning disability or have received special education services in the past has been used in prior research (e.g., Evans et al. 2015; Mallett et al. forthcoming), but it is not a formal diagnosis. It’s hard to say what the implications might be for our work, but it is likely to inject some degree of uncertainty into the findings.
A secondary limitation is that the Add Health sampling strategy would have missed youths with learning disabilities who are homeschooled or who attend specialized schools. If adolescents with such severe learning disabilities that they are unable to attend mainstream schools are also those whose disability manifests itself in behavioural problems, then the effect of learning disabilities on delinquency would be underestimated in this analysis. Similar dynamics are likely at play regarding peer deviance where adolescents with more serious learning disabilities who do not appear in the Add Health data will have more deviant peers while others will have less depending on their capacity to attend school and the type of school they attend. Whether peer deviance influences delinquency differently for people in these groups remains an open empirical question.
Furthermore, it is conceivable that using data that is nearly 30 years old may result in us elucidating dynamics which no longer exist. While social media has given adolescents more options to connect than in the mid-1990s, they still spend significant amounts of time together at and outside school as they did then. Moreover, it seems unlikely that peer status, delinquency, and their relation to learning disabilities will have changed in ways that invalidate the current study. Nonetheless, empirical work using more recent data would be welcome.
Additionally, the ability to generate causal estimates of learning disabilities on delinquency are dependent on the causal assumptions built into the DAG. To the extent that there are other important variables at work that were not included in the DAG, they effectively act as unobserved variables. Depending on how those unobserved variables fit into the causal model, they may open other backdoor paths which, through their omission, result in biased estimates of the hypothesized cause on the hypothesized effect. Litigating the state of the field and implementing changes to the DAG based on the evolving state of knowledge, and therefore the set of controls required to infer causality, is key to pushing the field forward.
Conclusion
The current study examined whether the link between delinquency and having a learning disability was mediated or moderated by having deviant friends. To date, this has been a gap in the literature. Although the total effect of learning disabilities on violence (but not property crime) was significant, this study found no support for the hypothesis that the relationship between learning disabilities and delinquency is mediated by peer deviance. We also found no evidence for the possibility that peers might exert more of an influence on the behaviour of youths with a learning disability. There are other plausible causal pathways, such as victimization and self-control, that we suggest might be helpful in elucidating why the link between learning disabilities and violence exists.
More broadly, this study calls into question both the “common knowledge” around the criminogenic effect of learning disabilities as well as the relevance of peer influence theories for a substantial portion of the population. As Altarac and Saroha (2007) note, a lifetime prevalence rate of 9.8% among US children equated to nearly 3 million children at that time. And a significant proportion of these children have co-occurring internalizing and externalizing problems (Cristofani et al. 2023) which layers on additional complexity given the varying relationships of those issues to delinquency (Haapasalo, Tremblay, Boulerice, and Vitaro 2000). Comparative tests, such as this one, across relevant subpopulations should drive theoretical refinements since they have the capacity to highlight the conditions under which those theories start to lose explanatory power.
References
Agnew, Robert. 1992. Foundation for a general strain theory of crime and delinquency. Criminology 30(1): 47–88. https://doi.org/10.1111/j.1745-9125.1992.tb01093.x.
Akers, Ronald L. 2009. Social Learning and Social Structure: A General Theory of Crime and Deviance. New Brunswick, N.J.: Transaction.
Akers, Ronald L., Marvin D. Krohn, Lonn Lanza-Kaduce, and Marcia Radosevich. 1979. Social learning and deviant behavior: a specific test of a general theory. American Sociological Review 44(4): 636–55.
Augustyn, Megan Bears, and Jean Marie McGloin. 2013. The risk of informal socializing with peers: considering gender differences across predatory delinquency and substance use. Justice Quarterly 30(1): 117–43. https://doi.org/10.1080/07418825.2011.597417.
Barclay, Linda. 2019. Disability with Dignity: Justice, Human Rights and Equal Status. New York: Routledge.
Barrett, David E., and Antonis Katsiyannis. 2016. Juvenile offending and crime in early adulthood: a large sample analysis. Journal of Child and Family Studies 25(4): 1086–97. https://doi.org/10.1007/s10826-015-0304-6.
Barrett, David E., Antonis Katsiyannis, Dalun Zhang, and Dake Zhang. 2014. Delinquency and recidivism: a multicohort, matched-control study of the role of early adverse experiences, mental health problems, and disabilities. Journal of Emotional and Behavioral Disorders 22(1): 3–15. https://doi.org/10.1177/1063426612470514.
Baumeister, Roy F., and Mark R. Leary. 1995. The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin 117(3): 497–529. https://doi.org/10.1037/0033-2909.117.3.497.
Bruefach, Tyler, and John R. Reynolds. 2022. Social isolation and achievement of students with learning disabilities. Social Science Research 104: 102667. https://doi.org/10.1016/j.ssresearch.2021.102667
Bryan, Tanis, Ruth Pearl, and Patrick Fallon. 1989. Conformity to peer pressure by students with learning disabilities: a replication. Journal of Learning Disabilities 22(7): 458–59. https://doi.org/10.1177/002221948902200713.
Bryan, Tanis, Mara Werner, and Ruth Pearl. 1982. Learning disabled students’ conformity responses to prosocial and antisocial situations. Learning Disability Quarterly 5(4): 344–52. https://doi.org/10.2307/1510916.
Chen, Ping, and Kim Chantala. 2014. Guidelines for Analyzing Add Health Data. Carolina Population Center, University of North Carolina at Chapel Hill. http://www.cpc.unc.edu/projects/addhealth/documentation/guides/wt_guidelines_20161213.pdf.
Cinelli, Carlos, Andrew Forney, and Judea Pearl. Forthcoming. A crash course in good and bad controls. Sociological Methods & Research, 1-34. https://doi.org/10.1177/00491241221099552.
Cristofani, Paola, Maria Chiara Di Lieto, Claudia Casalini, Chiara Pecini, Matteo Baroncini, Ottavia Pessina, Filippo Gasperini, Maria Bianca Dasso Lang, Mariaelisa Bartoli, Anna Maria Chilosi, and Annarita Milone. 2023. Specific learning disabilities and emotional-behavioral difficulties: phenotypes and role of the cognitive profile. Journal of Clinical Medicine 12(5): 1882. http://dx.doi.org/10.3390/jcm12051882
Dawson, Christyl T., Wensong Wu, Kristopher P. Fennie, Gladys Ibañez, Miguel Á. Cano, Jeremy W. Pettit, and Mary Jo Trepka. 2019. Perceived neighborhood social cohesion moderates the relationship between neighborhood structural disadvantage and adolescent depressive symptoms. Health & Place 56(March): 88–98. https://doi.org/10.1016/j.healthplace.2019.01.001.
Estell, David B., Martin H. Jones, Ruth Pearl, Richard Van Acker, Thomas W. Farmer, and Philip C. Rodkin. 2008. Peer groups, popularity, and social preference: trajectories of social functioning among students with and without learning disabilities. Journal of Learning Disabilities 41(1): 5–14. https://doi.org/10.1177/0022219407310993.
Evans, Mary K., Samantha S. Clinkinbeard, and Pete Simi. 2015. Learning disabilities and delinquent behaviors among adolescents: a comparison of those with and without comorbidity. Deviant Behavior 36(3): 200–220. https://doi.org/10.1080/01639625.2014.924361.
Faris, Robert, and Susan Ennett. 2012. Adolescent aggression: the role of peer group status motives, peer aggression, and group characteristics. Social Networks 34(4): 371–78. https://doi.org/10.1016/j.socnet.2010.06.003.
Farrington David P. 1973. Self-reports of deviant behavior: predictive and stable? Journal of Criminal Law & Criminology 64(1): 99–110.
Fogden, Billy C., Stuart D. M. Thomas, Michael Daffern, and James R. P. Ogloff. 2016. Crime and victimisation in people with intellectual disability: a case linkage study. BMC Psychiatry 16(1): 170. https://doi.org/10.1186/s12888-016-0869-7.
Gallupe, Owen, and Martin Bouchard. 2015. The influence of positional and experienced social benefits on the relationship between peers and alcohol use. Rationality & Society 27(1): 40–69. https://doi.org/10.1177/1043463114546316.
Gallupe, Owen, Holly Nguyen, Martin Bouchard, Jennifer L. Schulenberg, Allison Chenier, and Katie D. Cook. 2016. An experimental test of deviant modeling. Journal of Research in Crime and Delinquency 53(4): 482–505. https://doi.org/10.1177/0022427815625093.
Gottfredson, Michael, and Travis Hirschi. 1990. A General Theory of Crime. Stanford, CA: Stanford University Press.
Haapasalo, Jaana, Richard E. Tremblay, Bernard Boulerice, and Frank Vitaro. 2000. Relative advantages of person-and variable-based approaches for predicting problem behaviors from kindergarten assessments. Journal of Quantitative Criminology 16(2): 145-168. https://doi.org/10.1023/A:1007512521780
Harding, David J. 2009. Collateral consequences of violence in disadvantaged neighborhoods. Social Forces 88(2): 757–84. https://doi.org/10.1353/sof.0.0281.
Harris, Kathleen M. 2009 The National Longitudinal Study of Adolescent to Adult Health (Add Health), Waves I & II, 1994–1996; Wave III, 2001–2002; Wave IV, 2007-2009 [machine-readable data file and documentation]. (Ed.) University of North Carolina at Chapel Hill Carolina Population Center. https://doi.org/10.3886/ICPSR27021.v9.
Harris, Kathleen M., Carolyn T. Halpern, Eric A Whitsel, Jon M Hussey, Ley A Killeya-Jones, Joyce Tabor, and Sarah C Dean. 2019. Cohort profile: The National Longitudinal Study of Adolescent to Adult Health (Add Health). International Journal of Epidemiology 48(5): 1415–1415k. https://doi.org/10.1093/ije/dyz115.
Harris, Kathleen M., and J. Richard Udry. 2001. National Longitudinal Study of Adolescent Health (Add Health), Network Variables Code Book. Chapel Hill, NC: Carolina Population Center, University of North Carolina at Chapel Hill.
Haynie, Dana L., Nathan J. Doogan, and Brian Soller. 2014. Gender, friendship networks, and delinquency: a dynamic network approach. Criminology 52(4): 688–722. https://doi.org/10.1111/1745-9125.12052.
Haynie, Dana L., and D. Wayne Osgood. 2005. Reconsidering peers and delinquency: how do peers matter? Social Forces 84(2): 1109–30. https://doi.org/10.1353/sof.2006.0018.
Haynie, Dana L., and Alex R. Piquero. 2006. Pubertal development and physical victimization in adolescence. Journal of Research in Crime and Delinquency 43(1): 3–35.
Hicks, Raymond, and Dustin Tingley. 2011. Causal mediation analysis. The Stata Journal 11(4): 605–19. https://doi.org/10.1177/1536867X1201100407.
Hoffmann, Lisa, Jürgen Wilbert, Mike Lehofer, and Susanne Schwab. 2021. Are we good friends?–Friendship preferences and the quantity and quality of mutual friendships. European Journal of Special Needs Education 36(4): 502-516. https://doi.org/10.1080/08856257.2020.1769980
Hogan, Anthony, Lyndall McLellan, and Adrian Bauman. 2000. Health promotion needs of young people with disabilities - a population study. Disability and Rehabilitation 22(8): 352–57. https://doi.org/10.1080/096382800296593.
Imai, Kosuke, Luke Keele, and Dustin Tingley. 2010. A general approach to causal mediation analysis. Psychological Methods 15(4): 309–34. https://doi.org/10.1037/a0020761.
Jones, Martin H., Morgan T. James, and Chalon Johnson. 2023. Social status, popularity goals and students with learning disabilities: an initial investigation. British Journal of Special Education 50(1): 83–103. https://doi.org/10.1111/1467-8578.12445.
Kaufman, Robert L. 2019. Interaction Effects in Linear and Generalized Linear Models. Thousand Oaks, CA: SAGE. http://us.sagepub.com/en-us/nam/interaction-effects-in-linear-and-generalized-linear-models/book253602.
Mallett, Christopher A. 2014. The ‘learning disabilities to juvenile detention’ pipeline: a case study. Children & Schools 36(3): 147–54. https://doi.org/10.1093/cs/cdu010.
Mallett, Christopher A., Linda Quinn, Jinhee Yun, and Miyuki Fukushima-Tedor. Forthcoming. The ‘learning disabilities-to-prison’ pipeline: evidence from the Add Health national longitudinal study. Crime & Delinquency: 1-35. https://doi.org/10.1177/00111287221081024.
Malmgren, Kimber, Robert D. Abbott, and J. David Hawkins. 1999. LD and delinquency: rethinking the ‘link.’ Journal of Learning Disabilities 32(3): 194–200. https://doi.org/10.1177/002221949903200301.
McElreath, Richard. 2020. Statistical Rethinking: A Bayesian Course with Examples in R and STAN. 2nd ed. New York: Chapman and Hall/CRC. https://doi.org/10.1201/9780429029608.
McNamara, John K., and Teena Willoughby. 2010. A longitudinal study of risk-taking behavior in adolescents with learning disabilities. Learning Disabilities Research & Practice 25(1): 11–24. https://doi.org/10.1111/j.1540-5826.2009.00297.x.
Meldrum, Ryan C., and J. C. Barnes. 2017. Unstructured socializing with peers and delinquent behavior: a genetically informed analysis. Journal of Youth and Adolescence 46(9): 1968–81. https://doi.org/10.1007/s10964-017-0680-x.
Mendoza, Martin, Jamilia J. Blake, Miner P. Marchbanks III, and Kelsey Ragan. 2020. Race, gender, and disability and the risk for juvenile justice contact. The Journal of Special Education 53(4): 226–35. https://doi.org/10.1177/0022466919845113.
Morgan, Stephen L., and Christopher Winship. 2014. Counterfactuals and Causal Inference: Methods and Principles for Social Research. 2nd ed. New York: Cambridge University Press. https://doi.org/10.1017/CBO9781107587991.
Novak, Abigail, Brian B. Boutwell, and Thomas Bryan Smith. Forthcoming. Taking the problem of colliders seriously in the study of crime: a research note. Journal of Experimental Criminology: 1-10. https://doi.org/10.1007/s11292-023-09565-x.
Oliver, Michael. 1996. Understanding Disability: From Theory to Practice. London, Macmillan.
Øygard, Lisbet, Knut-Inge Klepp, Grethe S. Tell, and Odd D. Vellar. 1995. Parental and peer influences on smoking among young adults: ten-year follow-up of the Oslo youth study participants. Addiction 90(4): 561–69. https://doi.org/10.1046/j.1360-0443.1995.90456110.x.
Pearl, Ruth. 2002. Students with learning disabilities and their classroom companions. In The Social Dimensions of Learning Disabilities: Essays in Honor of Tanis Bryan, ed. Bernice Y. L. Wong and Mavis L. Donahue. Mahwah, N.J.: Routledge.
Pratt, Travis C., and Francis T. Cullen. 2000. The empirical status of Gottfredson and Hirschi’s general theory of crime: a meta-analysis. Criminology 38(3): 931–64. https://doi.org/10.1111/j.1745-9125.2000.tb00911.x.
Preacher, Kristopher J., and Andrew F. Hayes. 2004. SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behavior Research Methods, Instruments, & Computers 36(4): 717–31. https://doi.org/10.3758/BF03206553.
Preacher, Kristopher J., and Andrew F. Hayes. 2008. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods 40(3): 879–91. https://doi.org/10.3758/BRM.40.3.879.
Quinn, Mary Magee, Robert B. Rutherford, Peter E. Leone, David M. Osher, and Jeffrey M. Poirier. 2005. Youth with disabilities in juvenile corrections: a national survey. Exceptional Children 71(3): 339–45. http://doi.org/10.1177/001440290507100308
Savage, Melissa N., and Emily C. Bouck. 2017. Predictors of risky behavior and offending for adolescents with mild intellectual disability. Intellectual and Developmental Disabilities 55(3): 154–66. https://doi.org/10.1352/1934-9556-55.3.154.
Sellers, Christine S., John K. Cochran, and Kathryn A. Branch. 2005. Social learning theory and partner violence: a research note. Deviant Behavior 26(4): 379–95. https://doi.org/10.1080/016396290931669.
Shandra, Carrie L., and Dennis P. Hogan. 2012. Delinquency among adolescents with disabilities. Child Indicators Research 5(4): 771–88. https://doi.org/10.1007/s12187-012-9135-9.
Simpson, M. K., and J. Hogg. 2001. Patterns of offending among people with intellectual disability: a systematic review. part I: methodology and prevalence data. Journal of Intellectual Disability Research 45(5): 384–96. https://doi.org/10.1046/j.1365-2788.2001.00345.x.
StataCorp. 2021. Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC.
Stone, Wendy L., and Annette M. La Greca. 1990. The social status of children with learning disabilities: a reexamination. Journal of Learning Disabilities 23(1): 32–37. https://doi.org/10.1177/002221949002300109.
Sutherland, Edwin H. 1947. Principles of Criminology. 4th ed. Philadelphia: J. B. Lippincott.
Textor, Johannes, Benito van der Zander, Mark S Gilthorpe, Maciej Liśkiewicz, and George TH Ellison. 2016. Robust causal inference using directed acyclic graphs: The R package ‘Dagitty.’ International Journal of Epidemiology 45(6): 1887–94. https://doi.org/10.1093/ije/dyw341.
Thomas, Carol. 2007. Sociologies of Disability and Illness: Contested Ideas in Disability Studies and Medical Sociology. New York: Palgrave Macmillan.
Vazsonyi, Alexander T, H Harrington Cleveland, and Richard P Wiebe. 2006. Does the effect of impulsivity on delinquency vary by level of neighborhood disadvantage? Criminal Justice and Behavior 33(4): 511–41. https://doi.org/10.1177/0093854806287318.
Vazsonyi, Alexander T., Jakub Mikuška, and Erin L. Kelley. 2017. It’s time: a meta-analysis on the self-control-deviance link. Journal of Criminal Justice 48: 48–63. https://doi.org/10.1016/j.jcrimjus.2016.10.001.
World Health Organization (WHO). 2001. International Classification of Functioning, Disability and Health (ICF). Geneva.
Zhao, Xinshu, John G. Lynch, and Qimei Chen. 2010. Reconsidering Baron and Kenny: myths and truths about mediation analysis. Journal of Consumer Research 37(2): 197–206. https://doi.org/10.1086/651257.
Supplementary material
Appendix 1. Dagitty code used to produce figure 1.
Figure S1. 95% confidence intervals for factor effect of peer delinquency (Δ=1) moderated by learning disability for violence.
Figure S2. 95% confidence intervals for factor effect of peer delinquency (Δ=1) moderated by learning disability for property crime.
Figure S3. Violence by the interaction of peer delinquency and learning disability, dual outcome axes.
Figure S4. Property crime by the interaction of peer delinquency and learning disability, dual outcome axes.
Appendix 3. Supplementary analyses.
Table S1. Total effect of learning disabilities on delinquency controlling for verbal intelligence, race, and SES – survey-corrected negative binomial model.
An anonymous reviewer suggested that we should briefly note how our work fits with criminological research on academic performance, intelligence, and executive functioning. As McGloin, Pratt, and Maahs (2003) point out, academic performance is important for its relation to school-based social bonds. The fact that we included school-based social bonds in the DAG means that it is effectively accounted for. The executive function research has significant overlap with the self-control literature (i.e., much of the executive function research seeks to explain the neuropsychological basis of self-control). If we assume, therefore, that school-based social bonds are a proxy for academic performance and that self-control is a proxy for executive function, these factors are accounted for in the modeling strategy. Intelligence, however, is not. While learning disability is not synonymous with intelligence (e.g., Siegel 1990), it does suggest a potentially relevant variable that we have not included in the DAG. If we incorporate it in the ways proposed by McGloin et al. (2003) as well as plausible paths leading to intelligence (the code to produce this DAG is included in the supplementary material), it suggests that intelligence is one of the few necessary controls required to estimate an unbiased effect of learning disabilities on delinquency. We estimated this model using the standardized version of the Picture Vocabulary Test (a measure of verbal intelligence – see Boccio et al. 2018); results were substantively identical for the direct effect of learning disabilities on both property and violent crime.
Boccio, Cashen M., Kevin M. Beaver, and Joseph A. Schwartz. 2018. The role of verbal intelligence in becoming a successful criminal: Results from a longitudinal sample. Intelligence 66: 24-31.
McGloin, Jean Marie, Travis C. Pratt, and Jeff Maahs. 2004. Rethinking the IQ-delinquency relationship: A longitudinal analysis of multiple theoretical models. Justice Quarterly 21(3): 603-635.
Siegel, Linda S. 1990. IQ and learning disabilities: R.I.P. In Learning disabilities: Theoretical and research issues, ed. H. L. Swanson & B. Keogh. Hillsdale, NJ: Erlbaum.
Violence [dis-05g-total_viol_w_intel]
Property crime [dis-05h-total_prop_w_intel]
b (SE)
b (SE)
Learning disabilities
0.168* (0.078)
0.037 (0.100)
* p<.05
Dagitty code used to derive the minimal sufficient adjustment set for the models in table S1 (intelligence, race, sex):
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC-BY-NC-ND 4.0)
