ABSTRACT 

Title of Thesis: REDUCING THE EFFECT OF PATERNAL 
INCARCERATION ON JUVENILE 
DELINQUENCY: THE ROLE OF PROSOCIAL 
FRIENDSHIP NETWORKS 
 
Michael Jacob Lebron, Master of Arts, 2024 

Thesis Directed By: Dr. Jean M. McGloin, Professor 
Department of Criminology and Criminal Justice 

 

Paternal incarceration is associated with a wide array of negative developmental 

outcomes for children who are reared in this context, thus perpetuating intergenerational patterns 

of cumulative disadvantage. Recent scholarship from Giordano et al. (2019) has called for 

research investigating factors associated with intergenerational discontinuities in the life-course 

trajectories of children with incarcerated parents. There is reason to believe that prosocial peers 

may serve as a potential protective factor capable of ameliorating the negative developmental 

consequences of paternal incarceration. This thesis uses data from the National Longitudinal 

Study of Adolescent Health to explore whether prosocial friends can attenuate the elevated risk 

of delinquency and substance use which are often associated with paternal incarceration. The 

results suggest that prosocial friends are generally related to decreased propensity for 

delinquency and substance use, but they do not buffer the effect of paternal incarceration on 

adolescent delinquency and substance use. In the end, prosocial friends show promise as a 

universal protective factor among adolescents, which has meaningful implications for future 

research and interventions designed to set youth on more positive developmental trajectories.



  

 
 
 
 
 

REDUCING THE EFFECT OF PATERNAL INCARCERATION ON JUVENILE 
DELINQUENCY: THE ROLE OF PROSOCIAL FRIENDSHIP NETWORKS 

 
 
 

by 
 
 

Michael Jacob Lebron 
 
 
 
 
 

Thesis submitted to the Faculty of the Graduate School of the 
University of Maryland, College Park, in partial fulfillment 

of the requirements for the degree of 
Master of Arts 

2024 
 
 
 
 
 
 
 
 
 
 
 
 
 

Advisory Committee: 
  Dr. Jean M. McGloin, Professor, Chair 
  Dr. Bianca E. Bersani, Associate Professor 
  Dr. Wade C. Jacobsen, Associate Professor 
  



  

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

© Copyright by 
Michael J. Lebron 

2024 
 
 
 
 
 
 
 
 
 
 
 
 
 



ii 
 

Dedication 

I would like to dedicate this thesis to my friends, family, and my wonderful companions, 

Apples and Cinnamon. Thank you for keeping me going. 



iii 
 

Acknowledgements 

 There are several faculty members who have provided invaluable guidance and support 

without whom this thesis would not have been possible. First and foremost, I would like to 

extend my sincerest gratitude to Dr. Jean McGloin for not only serving as my thesis chair, but 

also as an exemplary mentor and advocate. You have played an integral role in my professional 

and academic development and without you I would not be where I am today. I am also 

incredibly grateful for the comments and guidance of Dr. Bianca Bersani and Dr. Wade Jacobsen 

as members of my committee and individuals who I greatly respect. I would also be remiss to not 

acknowledge the contributions of Dr. Robert Brame for providing invaluable methodological 

guidance and mentorship. Lastly, I would like to thank Dr. Maria Vélez and Dr. Laure Brooks 

for their support throughout the course of my graduate education.  



iv 
 

Table of Contents 
 
Dedication ................................................................................................................................................................................ ii 
Acknowledgements ............................................................................................................................................................... iii 
List of Tables ............................................................................................................................................................................ v 
List of Figures ........................................................................................................................................................................ vi 
Chapter 1: Introduction ......................................................................................................................................................... 1 
Chapter 2: Paternal Incarceration and the Benefits of Peer Influence ........................................................................ 5 

Collateral Consequences of Paternal Incarceration ......................................................................................................................... 5 

Mechanisms Linking Paternal Incarceration and Adolescent Deviance ................................................................................. 8 

The Shifting Salience of Developmental Bonds: Parents to Peers ............................................................................................ 12 

The Protective Role of Peers ...................................................................................................................................................................... 16 

Current Study ................................................................................................................................................................................................... 23 
Chapter 3: Data and Methods ........................................................................................................................................... 25 

Data and Sample.............................................................................................................................................................................................. 25 

Measures ............................................................................................................................................................................................................. 30 
Dependent Variables ........................................................................................................................................................................... 30 
Independent Variable .......................................................................................................................................................................... 33 
Moderator ................................................................................................................................................................................................ 34 
Controls .................................................................................................................................................................................................... 37 

Analytic Strategy .............................................................................................................................................................................................. 42 
Chapter 4: Results ................................................................................................................................................................ 45 

OLS Regression Models: Delinquency as the Outcome ................................................................................................................... 45 

Negative Binomial Regression Models: Substance Use as the Outcome .................................................................................. 48 

Robustness Checks .......................................................................................................................................................................................... 52 
Chapter 5: Discussion .......................................................................................................................................................... 54 
Appendices ............................................................................................................................................................................. 65 

Appendix A ......................................................................................................................................................................................................... 65 

Appendix B ......................................................................................................................................................................................................... 66 
References .............................................................................................................................................................................. 67 
 

  



v 
 

List of Tables 
  

Table 1. Descriptive Statistics (N=6,470) ..................................................................................... 31 
Table 2. Results of OLS Regression Models Showing the Relationship between Propensity for 
Juvenile Delinquency and Paternal Incarceration (Imputations = 20) .......................................... 47 
Table 3. Results of Negative Binomial Regression Models Showing the Relationship between 
Juvenile Substance Use and Paternal Incarceration (Imputations = 20) ....................................... 51 
Table 4. Descriptive Statistics for Delinquency Items (N = 6,470) .............................................. 65 
Table 5. Pre-imputation Missingness Across Covariates ............................................................. 66 
 
  



vi 
 

List of Figures 
 
Figure 1. Attrition Analysis .......................................................................................................... 30 
Figure 2. Marginal Effects of Paternal Incarceration on Substance Use ...................................... 52 
 
 
 
 
 
 
 
 
 

 



1 
 

Chapter 1: Introduction  

The collateral consequences of mass incarceration in the U.S. are widespread and far-

reaching. Despite reductions in the overall crime rate, the U.S. still maintains an incarceration 

rate roughly five times greater than any other developed, western nation (Carson, 2022; Western 

& Wildeman, 2009; Zeng, 2022). As a result, the collateral consequences of mass incarceration 

have received a considerable degree of attention amongst criminologists, but substantially less 

research has explored potential points of intervention for mitigating the detrimental effects of 

mass incarceration (Kirk & Wakefield, 2018). In essence, we know a great deal about the costs 

of mass incarceration, but relatively little about what can be done to remedy the enduring 

consequences that come with excessive reliance on incarceration as a response to crime. Some 

might say that is less important to identify factors poised to ameliorate the harm caused by mass 

incarceration than it is to address the core problem—by reducing mass incarceration itself we 

reduce the extent of its consequences. This is a laudable goal, but not necessarily one that is 

readily practical. Comprehensive sentencing reform is one potential avenue for change, but 

political challenges and the inability of courts to defer lower-level infractions outside of the 

criminal legal system severely limits our ability to directly confront mass incarceration (Beckett, 

2018). Therefore, until comprehensive sentencing reform is enacted, it is essential to seek an 

understanding of protective factors which safeguard at-risk populations from engaging in acts of 

deviance which amplify their risk of contact with the criminal justice system (Barnert et al., 

2021).  

Prior research examining the collateral consequences of mass incarceration spans across 

several domains, many of which are focused on the life-course outcomes of those who 



2 
 

experience incarceration first-hand (e.g., employment, legal earnings, recidivism, etc., see Kirk 

& Wakefield, 2018 for a review). However, mass incarceration is not only driven by the same 

individuals cycling in and out of penal institutions, but further exacerbated and maintained via 

the justice involvement of disadvantaged children (Mallett, 2017; Sampson & Loeffler, 2010). 

Many adults facing incarceration are also parents. According to recent estimates, approximately 

2.6 million children currently have a parent who is incarcerated, whether that be in prison or jail 

(Sykes & Petit, 2014; Poehlmann-Tynan & Turney, 2021). The number of children experiencing 

parental incarceration is likely far greater, however, as information on parental incarceration 

status is not routinely collected by institutions such as schools, prisons, jails, or social service 

programs (Poehlmann et al., 2010). Recent data suggests that roughly 5 million children in 

America will experience parental incarceration at some point in their lives (Annie E. Casey 

Foundation, 2023). Therefore, the scope of parental incarceration and its subsequent 

consequences for millions of families and children is a much more expansive issue than it might 

appear at first glance.  

Prior research documents the relationship between parental incarceration, specifically the 

incarceration of fathers, and adverse life outcomes for children (Murray, Farrington & Sekol, 

2012). This thesis goes a step beyond demonstrating the link between parental incarceration and 

juvenile delinquency and explores the ability of a youth’s friends to ameliorate the heightened 

risk of deviance that often comes with parental incarceration. In this thesis, I focus exclusively 

on paternal incarceration rather than maternal incarceration in large part because most 

incarcerated parents are fathers and there is mixed evidence concerning the effect of maternal 

incarceration on child wellbeing (Dallaire, 2007; Turney & Wildeman, 2015; Wildeman & 



3 
 

Turney, 2014a).1  However, we still know very little about the mechanisms through which 

children develop resiliency in the face of paternal incarceration, and subsequently resist 

participation in crime and other forms of deviance.  

There is an extensive body of criminological research discussing the role of peer 

influence in the production and amplification of deviance, but a relative dearth of criminological 

literature focused on the equally beneficial nature of peer influence (McGloin & Thomas, 2019; 

Warr, 2002) even though peers are heavily involved in promoting normative psychosocial 

development (Ryan, 2001; Stanton-Salazar & Spina, 2005; Youniss & Smollar, 1985). I address 

this gap in literature by investigating the role of peers in mitigating the negative developmental 

consequences of paternal incarceration. If a youth’s naturally occurring friends can serve as 

social anchors to reduce the negative consequences of paternal incarceration and diminish 

deviant outcomes, then program and policy interventions based on leveraging social networks to 

reduce crime and adverse developmental outcomes for children, should be an area of further 

research and attention (Giordano et al., 2019). By ‘naturally occurring,’ I refer to friendships that 

adolescents form with peers over the natural course of their development as they interact with 

other adolescents with varying attitudes and beliefs (Youniss, 1982). Moreover, I specifically 

focus on the influence of paternal incarceration and peers during adolescence rather than early 

childhood. The reasoning for this is twofold. First, most prior work on paternal incarceration 

focuses exclusively on young children, neglecting the equally important developmental period of 

adolescence during which juveniles are most likely to engage in deviant behavior (Murray & 

Farrington, 2010; Wildeman, Goldman, & Turney, 2018). Secondly, it is commonly understood 

that peer influence peaks during adolescence around age 14 (Berndt, 1979; Warr, 1993). 

 
1 The maternal incarceration rate has increased by over 600% since the 1980s making it an important area for future 
inquiry (Shlafer et al., 2021), but an examination of maternal incarceration is outside the scope of the present thesis.  



4 
 

Therefore, if peers do serve as a protective factor for youth exposed to paternal incarceration, we 

should be most likely to observe an effect during adolescence as opposed to early childhood or 

early adulthood.  

I begin by situating increased adolescent propensity for delinquency and substance use 

resulting from paternal incarceration in the context of cumulative disadvantage across 

generations, and then transition to an integrated developmental framework rooted in social 

learning and control theories to explore the potential of peers to reduce these effects. Using data 

from the National Longitudinal Study of Adolescent Health (Add Health), I assess the ability of 

prosocial friends (i.e., those peers who possess conventional social bonds and endorse attitudes 

and behaviors unfavorable to violation of the law) to moderate the effect of paternal 

incarceration on juvenile delinquency and substance use. This approach mirrors those taken from 

positive criminology, a unified theoretical perspective which explores the potential of positive 

social processes to alter life course trajectories for individuals predisposed to criminogenic risk 

factors (Ronel & Elsiha, 2011).2 Rather than focusing exclusively on the stigmatization and 

social rejection associated with paternal incarceration (Goffman, 1963), this thesis seeks to 

emphasize the benefits derived from the social inclusion and integration of adolescents who 

would otherwise sit on the social margins (Bogdan & Taylor, 1987; Bryan, 2017). In the end, this 

thesis offers additional insights on the mechanisms connecting paternal incarceration and deviant 

behavior, while also offering possible avenues for intervention. 

  

 
2 Positive criminology is akin to the study of protective factors and resiliency in psychology. Although, I am 
primarily focused on resistance to negative developmental experiences, disciplines such as sociology and 
psychology typically categorize this line of research under treatment and rehabilitation (Ronel & Elisha, 2011). 



5 
 

Chapter 2: Paternal Incarceration and the Benefits of Peer Influence 

Collateral Consequences of Paternal Incarceration 

Incarcerated individuals are not the only ones who must contend with repercussions of 

contact with the criminal justice system. A substantial body of criminological research has 

documented the extensive consequences of paternal incarceration for juvenile development and 

life-course outcomes (Eddy et al., 2014; Foster & Hagan 2015; Johnson & Easterling 2012; 

Murray, Farrington & Sekol, 2012; Turney, 2017; Wildeman & Western 2010; Wildeman, 

Wakefield, & Turney, 2013). Moreover, paternal incarceration is often linked to shifts in macro-

level factors which subsequently influence micro-level processes. Leibbrand et al. (2019) 

demonstrate that incarceration of a child’s father is positively associated with residential 

mobility, movement of families to more economically disadvantaged neighborhoods, and 

reduced social cohesion in the neighborhood context. Furthermore, the separation of fathers from 

the family via incarceration perpetuates socioeconomic disadvantage within urban areas, 

resulting in long-term consequences for the labor and marriage markets in these communities, 

which translates into shifting social networks and family structures that persist across generations 

(Rose & Clear, 1998). The degradation of familial bonds and the funneling of disadvantage from 

incarcerated individuals down to their partners and children is a pervasive consequence of mass 

incarceration, one that often contributes to the onset of deviant behaviors such as delinquency 

and substance use (McLeod & Tirmazi, 2017; Parke & Clarke-Stewart, 2002). 

 Many studies report a robust positive correlation between paternal incarceration and 

negative externalizing behaviors among offspring, such as violence and aggression (Craigie, 

2011; Turney & Goodsell, 2018). Using a national probability sample, Roettger and Swisher 



6 
 

(2011) found that paternal incarceration is highly associated with an increased risk of 

incarceration before the age of 25 among offspring in part due to their increased propensity for 

crime and delinquency, which often persists into adulthood (also see Burgess-Proctor, Huebner, 

& Durso, 2016). Indeed, numerous studies find that paternal incarceration is positively 

associated with the onset of childhood behavioral problems and delinquency even after 

accounting for relevant controls (see Antle, Gibson, & Krohn, 2020; Del Toro, Fine, & Wang, 

2023; Geller et al., 2009; Haskins, 2015; Murray & Farrington, 2005; Porter & King, 2015; 

Turney, 2017; Turney, 2022; Wakefield & Wildeman, 2013; Wildeman, 2010).3 There is also 

some evidence suggesting a positive correlation between paternal incarceration and substance 

use. In the same year as their examination of incarceration effects, Roettger and Swisher, with 

the addition of Kuhl and Chavez (2011), reported findings utilizing data from Add Health 

demonstrating that adolescents experiencing paternal incarceration were significantly more likely 

to use marijuana and other illicit drugs. Foster and Hagan (2013) echoed this finding utilizing 

data from AddHealth, reporting that the development of substance use related problems among 

juveniles was more likely to occur in the context of paternal incarceration, but not maternal 

incarceration. However, based on a sample of roughly 1,000 youth drawn from the Pittsburgh 

Youth Study, Murray, Loeber, and Pardini (2012) concluded that youth in the context of parental 

incarceration were no more likely to use marijuana than youth with no history of parental 

incarceration. However, this null finding may be an artifact of their decision to group paternal 

and maternal incarceration together in their analyses, which they acknowledge. Altogether, 

evaluations of the relationship between paternal incarceration and substance use are much more 

 
3 Most of these studies are focused on early childhood and therefore tend to examine behavioral problems rather 
than delinquency per se. 



7 
 

sparse than paternal incarceration and delinquency, which suggests the need for more research 

examining the relationship.  

Some may question the focus on parental incarceration in relation to adolescent deviance 

when there are other factors that may confound the relationship. Indeed, parental incarceration is 

often preceded by an array of risk factors, such as economic inequality, exposure to adult 

substance use, risky behaviors, and neighborhood disadvantage which lend themselves to the 

likelihood that an adolescent will engage in deviance (Arditti, 2012). However, it is important to 

recognize that parental incarceration is both a consequence of structural disadvantage as well as a 

contributing factor that further compounds pre-existing inequalities (Western & Pettit, 2010). 

Giordano and Copp (2015) refer to these intertwined developmental risk factors as ‘packages of 

disadvantage,’ which are difficult to disentangle from one another due to their typically 

interdependent nature.  

The fact that risk factors linked to deviance often precede parental incarceration does not 

negate the potential role of parental incarceration in exacerbating these risk factors, significantly 

increasing the likelihood of adolescent deviance. Furthermore, it does not preclude the possibility 

that parental incarceration maintains a direct, independent effect on adolescent deviance even 

after controlling for potential confounders (Giordano & Copp, 2015). In a meta-analysis and 

narrative review of 16 studies, Murray and colleagues (2009) found evidence that children of 

incarcerated parents were roughly twice as likely to engage in antisocial behavior and display 

poor mental health outcomes as children without incarcerated parents, even after controlling for 

related risk factors. However, they cautioned that causal conclusions could not be drawn from 

the available literature as some studies failed to control for parental criminality. Nevertheless, in 

a more expansive meta-analysis conducted just a few years later, Murray, Farrington, and Sekol 



8 
 

(2012) once again found evidence that parental incarceration was tied to increased risk of 

antisocial behavior, but not substance use. Notably, only 12 of the 40 studies included a measure 

of substance use and the characteristics of the measure were variable. Thus, it appears that most 

prior research provides evidence that paternal incarceration more than likely retains a direct 

effect on antisocial behavior under most contextual conditions, but the evidence regarding 

substance use is less consistent. To provide additional context to these findings, it is important to 

further consider the mechanisms linking paternal incarceration to adolescent delinquency and 

substance use.  

Mechanisms Linking Paternal Incarceration and Adolescent Deviance 

In an extension of their age-graded theory of informal social control, Sampson and Laub 

(1997) developed a theory of cumulative disadvantage. Cumulative disadvantage combines 

aspects of informal social control and labeling theories to explain how the societal stigma 

associated with formal sanctions leads to weakened social bonds and subsequent continuity in 

antisocial behavior over the life-course (Lemert, 1951; Laub & Sampson, 2003; Sampson & 

Laub, 1993). In accordance with this theory, individuals who are formally sanctioned during 

adolescence should be more likely to experience a breakdown of conventional social bonds and 

informal social controls, freeing them to engage in future deviance from societal norms (Lemert, 

1951). Lemert (1951) refers to such behaviors as secondary deviance. Similar logic can be 

applied to understand the effects of paternal incarceration on adolescent delinquency and 

substance use via the complementary life course perspective of “linked lives” (Elder, 1985; 

Thornberry et al., 2003). Linked lives have traditionally been understood as referring to the 

“interlocking trajectories” of consecutive family generations in which the salient life events of 

one generation (i.e., divorce of a parent) provide important context for the life-course trajectories 



9 
 

of other generations (i.e., their children) (Elder, 1985, p.39). As such “misfortune and 

opportunity across these pathways become intergenerational” (Elder, 1985, p.40).  

In essence, cumulative disadvantage may not only characterize the life-course trajectories 

of individuals who directly engage in deviance and experience sanctioning, but also the life-

course trajectories of their children who are exposed to stigma and disadvantage via association 

with the sanctioned adult (Gilligan, Karraker, & Jasper, 2018). Therefore, paternal incarceration 

may not only be a turning point in the life-course of a father, but also their child (Sampson & 

Laub, 1997). Traditionally, turning points have been thought of as salient life events that 

promote desistance from antisocial behavior, things such as stable employment and marriage 

(Bersani & Doherty, 2018). However, Sampson and Laub (1997) reasoned that just as salient life 

events can alter life trajectories in favor of conventionally adaptive behaviors, so too can they 

alter life trajectories in favor of continuity in maladaptive behaviors, such as delinquency and 

substance use.  

The convergence of these theoretical ideas can be seen in Jacobsen’s (2019) extension of 

cumulative disadvantage to what he deems the intergenerational stability of punishment. Put 

simply, the intergenerational stability of punishment hypothesis is based on the idea that the 

consequences of formal sanctions are experienced across generations. Furthermore, Jacobsen 

(2019) reasons that the effects of formal sanctions, namely paternal incarceration, on juvenile 

outcomes operate via social control and labeling mechanisms. Essentially, paternal incarceration 

and its accompanied risk factors, result in the weakening of a child’s social bonds, which 

promote the development of childhood behavioral problems, otherwise known as 

intergenerational secondary deviance (Hagan & Palloni, 1990; Jacobsen, 2019). However, 

because of the stigma associated with having an incarcerated father, children are also vulnerable 



10 
 

to intergenerational secondary sanctioning, which refers to punishment from formal institutions 

because of an individual’s status characteristics and not their behavior (Jacobsen, 2019; 

Liberman, Kirk, & Kim, 2014). Jacobsen (2019) finds support for both of these mechanisms in 

his examination of the relationship between paternal incarceration and school punishment. 

One way in which paternal incarceration leads to weakened informal social control is 

through the dissolution of familial bonds. Consistent with the assumptions of cumulative 

disadvantage, incarcerated males are significantly less likely to be married, but are no less likely 

to be fathers (Western, 2006; Shaw, 2016).  Lewis, Garfinkel, and Gao (2007) found that 

formerly incarcerated fathers earned 28% less annually than those who had never been 

incarcerated. Geller, Garfinkel, and Western (2011) report that these fathers are also less likely to 

financially contribute to their families following their incarceration, with those who do 

contribute providing significantly less than they would, had they never been incarcerated. This is 

not only due to limited employment opportunities, but the higher likelihood that these fathers 

will end up living separately from their children due to the erosion of the relationship between 

the child’s mother and father (Haskins & Jacobsen, 2017; Geller, Garfinkel, & Western, 2011; 

Turney & Wildeman, 2013). Although, many families reside below the poverty line prior to 

incarceration of a parent, according to the Bureau of Justice Statistics, 54% of fathers in state 

prison report that they were the primary financial provider for their children prior to 

incarceration, suggesting that these fathers were involved in their children’s lives (Glaze & 

Maruschak, 2010). In their absence, 88% of fathers reported that primary care responsibilities 

were left to the child’s mother (Glaze & Maruschak, 2010). In these cases, the remaining 

caregiver must contend not only with financially supporting their children, but shouldering the 

primary responsibility for supervision, socialization, and the education of their children (Arditti, 



11 
 

2016; Arditti & Few, 2006). As such, the ability of mothers to exercise informal social control 

and maintain positive attachment to their children is significantly diminished (Turney, 2014a). 

 Prior research also finds that stigmatization from other adults and peers, internalized 

feelings of shame (self-labeling), and embarrassment are all likely to follow incarceration of a 

parent for both children and adolescents (Boswell & Wedge, 2002; Cochran, Siennick, & Mears, 

2018; Lowenstein, 1986; Nesmith & Ruhland, 2008). Depending on familial circumstances, the 

consequences of incarceration may immediately prove more impactful for those who are left 

behind than for the incarcerated individual themselves. Although the father is temporarily 

removed from society via incarceration, their family remains in the community. Thus, their 

families, and particularly their children, are exposed to potential judgment and devaluation by 

other members of their community and formal social institutions, such as schools and 

prospective employers (Braman, 2004; Shaw, 2016). Thus, the stigmatization applied to 

adolescents with incarcerated parents may act as an additional barrier to the formation 

conventional social ties among adolescents via social exclusion (Bryan, 2017; Cochran, 

Siennick, & Mears, 2018). Coupled with a breakdown of informal social control mechanisms 

rooted in the family, juveniles with incarcerated fathers have greater chances of engaging in 

antisocial behavior and experiencing subsequent criminal justice system contact, advancing 

intergenerational patterns of secondary deviance and punishment (Harris, 2006; Jacobsen, 2019). 

Jacobsen’s (2019) extension of cumulative disadvantage shows evidence of being a promising 

lens through which to better understand the deleterious effects of paternal incarceration. 

However, I aim to demonstrate that the utility of this theoretical frame in addressing the 

consequences of paternal incarceration may be amplified through acknowledgement of salient 

social bonds outside the family context. 



12 
 

The Shifting Salience of Developmental Bonds: Parents to Peers 

In the words of Coleman (1980, p. 409), “an integral feature of adolescence is the gradual 

severance of early emotional ties with parents,” and a shift in adolescents’ focus toward peer 

relationships. This sentiment is especially prominent among developmental psychologists 

influenced by the works of Sullivan (1953) and Piaget (1965). In his synthesis of the ideas 

advanced by Sullivan and Piaget, Youniss (1980) identifies the shift in salience from youth’s 

parents to their peers as a critical point in the development of social maturity. Warr (2002) 

echoes this sentiment with his declaration that sociologists and developmental psychologists are 

in near unanimous agreement that peers play a critical role in the transition to adulthood as 

juveniles seek to sever ties to their childhood identity. From the perspective of Sullivan and 

Piaget, social maturity can be defined as the development of interpersonal understanding 

(Youniss, 1980). They argue that during early childhood, children are primarily focused on 

adopting social meaning directly from adults, namely their parents through unilateral 

communication. Infants and young children are naturally inclined to seek order and adults are 

equipped to supply them with the rules to which they are expected to conform (Sullivan, 1953). 

Inevitably this renders children incapable of distinguishing between “what is good in [their] 

parents, and what is open to criticism” (Piaget, 1965, p.192).   

This changes when children begin school around age 5 and begin interacting with their 

peers on an equal footing. Unlike socialization via adults, socialization via peers is characterized 

by symmetrical reciprocity in which children challenge each other’s views of the world and work 

toward mutual understanding, learning from one another (Piaget, 1962). This process of 

cooperative interaction comes to a head in adolescence when youth are no longer content to 

unconditionally accept the unilateral authority of adults (Youniss, 1980). Instead, adolescents 



13 
 

begin to actively seek out relationships with both peers and adults who will engage in 

cooperative relationships. This explains in part why youth’s relationships with adults become 

less salient during adolescence; once reaching this developmental stage youth seek symmetrical 

reciprocity, which many adults are unwilling to engage in because they have already cemented 

their own beliefs and understanding of the social order (Youniss, 1980). As a result, adolescents 

primarily associate with peers who are willing to exchange ideas and air differences in opinion 

so that they may work toward a common view or position. These cooperative interactions form 

the building blocks of friendship through which adolescents with very different backgrounds and 

beliefs may come to form meaningful interpersonal relationships (Youniss, 1980). This helps 

explain how adolescents in disadvantaged contexts may still come to associate with prosocial 

peers who endorse conventional values and beliefs regarding the law. The ideas advanced by 

Sullivan and Piaget demonstrate that homophily, although related to, is not a prerequisite to the 

development of meaningful friendships. Thus, although most research suggests that adolescents 

with incarcerated fathers will go on the associate with deviant others and subsequently engage in 

delinquency (Bryan, 2017; Kjellstrand, Reinke, & Eddy, 2018), Iit is also reasonable to conceive 

that some will go on to associate with supportive non-deviant peers who expose these 

adolescents to attitudes and beliefs in alignment with prosocial behavior (Brechwald & Prinstein, 

2011; Paternoster & Iovanni, 1989). In other words, some adolescents in the context of paternal 

incarceration will naturally befriend deviant peers, while others will naturally befriend non-

deviant peers who embody adherence to conventional social norms and values.  

The ideas behind the work of Sullivan and Piaget have received considerable empirical 

support among developmental scholars who consistently find that relationships with peers 

become more salient as children age into adolescence (Brown & Larson, 2009). Buhrmester and 



14 
 

Furman (1987) conducted a study exploring the development of companionship and intimacy 

among a sample of second, fifth, and eighth graders. They found that family members were an 

important source of companionship for second and fifth graders, but less so for eighth graders. 

On the other hand, same-sex peers were rated as important providers of companionship and 

intimacy across all three grade levels, contrary to researcher expectations. There were no 

significant differences across age groups in regard to desire for intimacy, implying that 

adolescents still seek to form meaningful, fulfilling relationships even as they shift their focus 

away from their parents. Nickerson and Nagle (2005) conducted a similar study looking at 

fourth, sixth, and eighth graders’ self-reported attachment to parents and friends. They found that 

older adolescents more frequently relied on peers to fulfill their attachment needs, as compared 

to those in the sample who were younger. These adolescents were also more likely to seek out 

peers to fulfill their attachment needs when they perceived their relationship with their parents to 

be less secure, which may be particularly likely for children with incarcerated parents. 

Additionally, in a longitudinal study of 655 high-school aged adolescents over the course of 18 

months, Hay and Ashman (2003) found that adolescents increasingly transferred their emotional 

attachment from their parents to their peers as they progressed through adolescence.    

Considering the well-documented risk associated with parental incarceration, it is 

reasonable to question if naturally occurring peer relationships during adolescence may be 

capable of diminishing criminogenic risk by promoting normative development. If peer 

relationships challenge adolescents to change their understanding of the world and their attitudes, 

then peers may offer an avenue for change over the early life-course of youth facing the 

packages of disadvantage which so often accompany parental incarceration (Giordano & Copp, 

2015; Youniss, 1980). According to Giordano et al. (2019, p.418) exploration of prosocial 



15 
 

connections may prove to be a promising avenue for altering the trajectory of youth exposed to 

parental incarceration and “additional research on factors associated with such discontinuities (in 

effect “bucking the trend”) should be a high priority.” The line of research that Giordano and 

colleagues (2019) advocate for strongly aligns with the perspective taken by positive 

criminologists.  

Positive criminology focuses on leveraging positive experiences and social processes as a 

means of coping with risk factors (Ronel & Elisha, 2011). The ideas underlying positive 

criminology have long been acknowledged by developmental psychologists but have received 

considerably less attention amongst criminologists whose research is typically centered around 

the etiology of crime and the exploration of criminogenic risk (Ronel & Elisha, 2011). 

Nevertheless, research aligning with the ideas advanced by positive criminology shows promise. 

For instance, some qualitative work finds that interactions between at-risk youth and outreach 

workers helps to expose youth to ideas of altruism which in turn reduce youth endorsement of 

self-interested orientations associated with problem behaviors (Ronel, 2006). Similarly, using a 

cross-sectional sample of adolescent offenders from disadvantaged communities, Smith, Faulk, 

and Sizer (2016) found that promoting ties to community resources (i.e., recreational, faith-

based, school, etc.) not only improved family functioning in disadvantaged communities, but 

also connected youth with more positive peer influences. This thesis adds to the body of 

literature adopting the perspective of positive criminology and answers the call of Giordano and 

colleagues (2019) regarding the need for additional research gauged at better understanding the 

factors that allow adolescents to resist the intergenerational cycle of incarceration and 

criminality. Due to the nature of adolescent development, naturally occurring peer relationships 



16 
 

may provide a potential avenue for blunting the effect of paternal incarceration on adolescent 

delinquency and substance use. 

The Protective Role of Peers  

The ability of one’s social ties to blunt the risk of deviance has traditionally been 

understood from the perspective of control theories, prime among them being Hirschi’s (1969) 

social bonding theory. Hirschi (1969) argued that social bonds to peers and conventional social 

institutions exert social control preventing individuals from engaging in delinquency. Under the 

social control framework proposed by Hirschi (1969), attachments are inherently prosocial in 

nature in so far as they restrain individuals from participating in crime and adjacent forms of 

deviance. According to Hirschi’s “cold and brittle” hypothesis, the relationships and social 

network characteristics of delinquent youth are markedly different from youth who do not 

engage in delinquency. Notably, their relationships are short-lived and lacking in intimacy. 

Essentially, Hirschi reasoned that delinquents are incapable of forming strong attachments to 

others and are thus socially isolated. Under this assumption, Hirschi’s theory rejects the idea that 

delinquent behavior is learned through interaction within intimate personal groups, a core 

proposition of Sutherland’s (1947) differential association theory.  

A substantial body of research indicates that Hirschi’s “cold and brittle” hypothesis does 

not have empirical support. Early work by Claes and Simmard (1992) found no differences 

between delinquent and non-delinquent adolescents in their self-reported levels of attachment or 

degree of intimacy with their friends. Similarly, using data from the NYS, Vasquez, and 

Zimmerman (2014) found no evidence that prior delinquency has a significant effect on 

friendship quality or attachment to peers as measured by perceived closeness. However, they did 

find evidence that prior delinquency is associated with more time spent amongst peers. 



17 
 

Additionally, among adolescents who report associations with delinquent peers, peer social 

support is positively associated with delinquency, contrary to the core belief held by control 

theorists that social support is consistently related to positive outcomes in the context of 

adolescent development (Brezina & Azimi, 2017). These findings underscore the idea that 

attachment is not always a protective factor. Rather, the effects of attachment and peer support 

are conditional on the direction of the attitudes and behaviors held by one’s peers in relation to 

the law. Thus, just as Piaget (1965) and Sullivan’s (1953) concepts of cooperation and 

reciprocity would lead us to believe, who one’s peers are matters; not all attachments are 

oriented in favor of prosocial norms and values.  

Instead, the relationship between the orientation of one’s peer attachments and 

subsequent behavior can be better understood from the standpoint of Sutherland’s (1947) 

differential association (DAT) and Akers’ (1998) social learning theory (SLT). Sutherland 

(1947) outlined his theory of differential association across nine key propositions, which can be 

condensed into two central ideas. The first being that learning is built upon through 

communication within intimate personal groups, in which individuals adopt the definitions of 

others based on the frequency, duration, intensity, and priority of the interactions they have with 

members of these groups. This leads to the second idea: criminal behavior results when an 

individual possesses an excess ratio of definitions favorable to violation of the law as opposed to 

definitions unfavorable to violation of the law. Additionally, Sutherland reasoned that the content 

of what is learned alongside these definitions contains the specific motives, drives, and 

rationalizations for crime. Thus, individuals learn not only attitudes favorable to crime, but also 

the techniques for engaging in criminal behavior and justifications for this behavior through 

interaction with intimate others. The foundations of SLT begin with Sutherland’s concept of 



18 
 

differential associations, being the exposure of individuals to definitions both favorable and 

unfavorable to violation of the law, which is moderated by the frequency, duration, priority, and 

intensity of the interactions one has with others. However, Akers (1998) specified that these 

associations can be direct or indirect and verbal or nonverbal. Akers also retained Sutherland’s 

concept of definitions but clarified that ‘definitions’ are equivalent to one’s attitudes towards the 

law and beliefs governing behavior. According to Akers, definitions are inclusive of, but not 

limited to, the techniques, motivations, and rationalizations for criminal behavior. In addition to 

these principles, Akers (1998) also integrated the concepts of differential reinforcement, from his 

work with Burgess and modeling, which he borrowed from developmental psychology (Bandura, 

1969). Differential reinforcement is a component of operant conditioning and refers to the 

introduction of positive stimuli or removal of negative stimuli to encourage a given behavioral 

response (Burgess & Akers, 1966) The concepts of modeling and imitation stipulate that 

individuals are more likely to engage in a behavior when they have observed the behavior. 

Furthermore, this is especially the case when the observed behavior is reinforced. Imitation is 

also more likely when the model is someone who the individual values or holds some form of 

attachment (Akers, 1998). Thus, individuals should be more likely to engage in behaviors and 

endorse certain attitudes when these attitudes and behaviors are modeled and reinforced by 

valued peers. SLT has received extensive empirical support in the context of both delinquency 

and substance use (Akers & Jensen, 2017; Kruis, Seo, & Kim, 2020; Pratt et al., 2010). 

At the heart of both DAT and SLT is the idea that both conforming and non-conforming 

behavior can be explained via the same learning process. Akers (1998) explained that it is not 

simply a matter of associations with “bad companions” or kinds of people that results in crime, 

rather learning criminal behavior involves the communication of both criminal and non-criminal 



19 
 

definitions (p. 26). To Akers (1998), the only difference between learning deviant behavior as 

opposed to conforming behavior is the direction of one’s differential associations, definitions, 

reinforcement, and models which they imitate. Following this logic, one is likely to have an 

increased propensity for prosocial behavior when they disproportionately associate with others 

who express prosocial behavior or hold attitudes favorable to such behavior, have greater 

exposure to valued peers who model conforming behavior, do not view criminal behavior 

themselves as justifiable, and have received or anticipate receiving greater rewards for 

conforming rather than deviating. If this is indeed the case, social learning mechanisms may be 

capable of complementing social control mechanisms to diminish the odds that adolescents 

engage in intergenerational secondary deviance.  

Both labeling and learning theories are rooted in the process of socialization, which can 

be understood in part utilizing the framework of symbolic interactionism (Blumer, 1956; Mead, 

1934; Paternoster & Iovanni, 1989). Essentially, symbolic interactionism posits that individuals 

interact with elements of their social environment based on the meaning that they assign to these 

elements. These meanings then inform the ways in which an individual constructs or perceives 

their definitions, attitudes, beliefs, and position in life. Finally, it is because of the variation in 

assigned meaning to these elements and self-perceptions that individuals respond differently to 

similar situations and contexts (i.e., why some adolescents engage in delinquent behavior 

whereas others do not). Matza (1969) referred to this process as signification, which is what 

allows for the interaction between theories of affinity and theories of affiliation (McGloin & 

Thomas, 2017).  

Affinity is the idea that, “persons, either individually or in aggregates, develop 

predispositions to certain phenomena, say, delinquency, as a result of their circumstances” 



20 
 

(Matza, 1969, p. 90-91). In the context of the present research, parental incarceration and 

cumulative disadvantage can be understood as circumstances which increase an individual’s 

affinity, or propensity, for engaging in crime and adjacent forms of deviance, such as substance 

use. Affiliation “describes the process by which the subject is converted to conduct novel for him 

but already established for others” (Matza, 1969, p. 101). Sutherland’s (1947) theory of 

differential association and Akers’ (1998) social learning theory are both theories of affiliation as 

they seek to explain deviant behavior through associations with others who endorse attitudinal 

orientations favorable to deviance from conventional behavior. Therefore, paternal incarceration 

and cumulative disadvantage establish individual affinity for deviance, which is typically 

reinforced or activated via affiliation with peers who possess attitudinal orientations conducive to 

deviance.  

Here I propose an integrated theoretical framework for understanding the relationship 

between paternal incarceration and adolescent deviance. Paternal incarceration leads to the 

weakening of social controls and the intensification of pre-existing structural inequalities 

resulting in cumulative disadvantage (Sampson & Laub, 1997). These conditions lend 

themselves to an individual’s affinity to engage in maladaptive behaviors in the form of 

intergenerational secondary deviance, such as substance use and delinquency (Jacobsen, 2019; 

McGloin & Thomas, 2017). Individuals are then more likely to associate with peers who have 

been similarly labeled or freed of social controls. Through interactions with similarly situated 

peers, individuals are exposed to the attitudes and definitions favorable to violation of the law, 

which they come to adopt as part of their own attitudes and beliefs. Alternatively, individuals 

may be safeguarded from achieving an excess of definitions favorable to violation of the law and 

subsequent perception of themselves as deviant via associations with peers who espouse 



21 
 

prosocial attitudes and beliefs. Coming from a disadvantaged background characterized by 

diminished social control does not always mean an adolescent will go on to associate with 

deviant peers. Through cooperation and symmetrical reciprocity youth from disparate contexts 

may develop a shared understanding of the world rooted in adherence to conventional norms and 

social institutions (Youniss, 1980). Thus, via affiliation with those Goffman (1963) refers to as 

the “wise,” those who are aware of the stigma linked to an individual yet decide to associate with 

them despite not sharing the same status, adolescents may be protected from the deleterious 

effects of paternal incarceration. Those adolescents who are able to naturally form prosocial 

friendships may be able to reinforce their ties to conventional social institutions that would 

otherwise be weakened by their father’s incarceration.  

Prior research suggests that positive social influences can shape life-course outcomes for 

the better. Although brief in their assessment, McKinney and colleagues (1977) were among 

some of the earliest criminologists to acknowledge the seemingly beneficial role of positive peer 

culture in residential treatment programs of adjudicated youth. Similarly, using data from a 

survey of 1,300, 7th-12th grade adolescents, Brown, Eichner, and Petrie (1986) found that peer 

group membership was positively tied to emotional and instrumental support, fostering positive 

social development. In a review of the research on the benefits of peer influence, Bernard (1990) 

cited work linking positive peer influences and peer-based interventions to numerous positive 

developmental outcomes such as academic outcomes (Bukowski & Hoza, 1989; Fantuzzo et 

al.,1989; Glasser,1986; Greenwood et al., 1989; Johnson & Johnson, 1983; Johnson et al., 1981; 

Ladd, 1990), reductions in underage drug and alcohol use (Bangert-Drowns, 1988; Tobler 1986), 

increased self-esteem (Slavin, 1990), and the development of critical social skills associated with 

later life success such as impulse control, problem-solving, and communication (Attili, 1990; 



22 
 

Kellam et al., 1982). More recently, Chun-Hall & Chen (2009) compared the peer group 

orientations of 330 elementary school students and found that students with more prosocial peer 

groups performed better in school and reported more positive self-perceptions of social and 

behavioral competence than their peers in the more aggressive peer groups. Furthermore, in their 

review of crime prevention programs focused on peer influence and mentoring, Sullivan and 

Jolliffe (2012) concluded that recent empirical evaluations of prevention programs rooted in 

prosocial peer influence provide promising avenues for reducing delinquent behavior. Moreover, 

they found that the evidence for peer mentoring programs is more extensive than that for 

interventions solely focused on reducing ties to antisocial peers (Sullivan & Jolliffe, 2012). 

However, it may also be the case that the cumulative disadvantage of paternal 

incarceration and the associated stigmatization is too strong to overcome, even in the presence of 

prosocial peers. For instance, a recent study evaluating the efficacy of Big Brother and Big Sister 

programs among children with incarcerated parents found only temporary reductions in juvenile 

delinquency during the duration of the program and found that delinquency rates returned to pre-

intervention values if not higher after conclusion of the mentoring program (Morris, 2017). This 

may be rooted in the fact parental incarceration is often preceded by ‘packages of risk’ which 

substantially increase the likelihood that youth will interact with the criminal justice system via 

intergenerational secondary deviance or labeling (Arditti, 2012; Giordano & Copp, 2015). Or it 

may be due to the possibility that adolescents revert to their old delinquent peer networks upon 

severing the social bond with their mentor. It is also possible that despite ties to prosocial friends, 

informal and formal labeling processes may initiate a shift in a juvenile’s identity that is too 

strong to overcome. Indeed, Matsueda’s (1992) work on reflected appraisals suggests that 

perception of oneself as a rule violator is not only based on prior delinquency, but also parental 



23 
 

appraisals of adolescent behavior. Additionally, Matsueda, O’Neill and Kreager (2020) find that 

perceptions of oneself as a rule violator are also based on reflected appraisals from teachers and 

peers. Although a youth’s friends may be prosocial in nature, that does not preclude the 

possibility that these friends may still view the youth as deviant despite their association, thus 

reinforcing the youth’s own self-image as one that is consistent with delinquency and substance 

use. Therefore, it is entirely possible that the cumulative disadvantage may be too overwhelming 

for youth to overcome via bonds with prosocial friends. 

Current Study 

This thesis seeks to examine a potential mechanism for ameliorating the negative 

consequences of paternal incarceration on an adolescent’s delinquency and substance use. 

Because of their influential role in adolescent development, friends may serve as a potential 

protective factor for adolescents experiencing paternal incarceration. In particular, establishing 

and maintaining associations with peers holding attitudes unfavorable to violation of the law may 

help to reduce the likelihood that children respond to the labeling effects of paternal 

incarceration by engaging in their own deviant behavior. Essentially, prosocial peer networks 

may act as a sort of “social anchor” (see Giordano et al., 2019), grounding disadvantaged 

adolescents in an environment characterized by increased emphasis on prosocial behavioral 

norms and values. In conducting this research, I am interested in discovering those youth in the 

context of paternal incarceration who are ‘exceptions to the rule.’ This complements prior 

research on the collateral consequences of paternal incarceration by not only shifting the 

developmental focus from early childhood to adolescence, but also investigating a potential 

pathway for change in antisocial behavioral patterns, that may otherwise be characterized by 

continuity (Wildeman, Haskins, & Poehlmann-Tynan, 2017). I also focus my analysis on 



24 
 

paternal incarceration as past research has consistently found a that paternal incarceration is 

positively linked to behavioral issues, whereas the findings for maternal incarceration are less 

consistent (Murray, Farrington, & Sekol, 2012; Turney & Wildeman, 2015). If prosocial peers 

serve as a protective factor, it should be most apparent among children of incarcerated fathers, 

for which there is a well-documented risk of maladaptive behaviors (Emory, 2018). More 

specifically, when prosocial peer orientation is low, delinquency and substance use should be 

higher among those with incarcerated fathers, as compared to those without incarcerated fathers. 

However, when prosocial peer orientation is low, the gap should be reduced such that the 

likelihood of deviance is similar regardless of one's paternal incarceration history. 

Using data from the National Longitudinal Study of Adolescent Health (AddHealth), I 

start by assessing whether adolescents are more likely to engage in delinquency and substance 

use in the context of paternal incarceration. Although I outline a complex theoretical model with 

many intermediating mechanisms between paternal incarceration and delinquency, I only test the 

relationship between paternal incarceration and deviance. I am not testing for evidence of 

mediation via the mechanisms outlined in my theoretical model. Second, I investigate whether 

the magnitude of the effect between paternal incarceration and adolescent delinquency should be 

diminished among adolescents with more prosocial friendship networks. Then I examine both 

hypotheses in relation to substance use as well. If prosocial peers can diminish the criminogenic 

risk associated with paternal incarceration, this would suggest a promising avenue for 

interrupting intergenerational patterns of continuity in criminality and system involvement. Such 

a finding would have important implications for interventions aimed at reducing the incidence of 

delinquency in highly disadvantaged contexts. In short, this thesis takes an important step toward 



25 
 

advancing the field’s understanding of peer influence by examining the ability of prosocial peers 

to diminish delinquency and substance use among stigmatized youth. 

Chapter 3: Data and Methods 

Data and Sample 

 This study uses data from the National Longitudinal Study of Adolescent to Adult Health 

(AddHealth) (Harris & Udry, 1994). Add Health is a nationally representative, school-based 

cohort study of U.S. adolescents in grades 7-12 during the 1994-1995 academic year (Harris & 

Udry, 1994). Data collection began with a school-based survey. A stratified sample of 80 high 

schools across the U.S. was generated with probability of selection proportional to size of a 

school’s student body (Harris, 2013). Schools were stratified by race, urbanicity, size, and type 

(i.e., public, or private). Each high school was then linked with one of its feeder schools to 

include eligible adolescents in each community, usually 7th and 8th graders, who were not yet 

attending these high schools. During the 1994-1995 school year, in school questionnaires were 

administered to over 90,000 students from across the 132 schools included in the final sample. 

These questionnaires captured information on friendship networks, school context, participation 

in school activities, expectations for the future, and an array of student health factors. The 

research team then used the completed questionnaires to choose a sample for in-home interviews. 

Sample selection was stratified by grade level and sex, resulting in a base sample of 12,105 

adolescents who participated in Wave I of the Add Health in-home interview. Supplemental 

samples were also obtained to better represent some ethnic minorities, assess effects of genetic 

relatedness among sibling, adoption, disability, and detailed information on social network 

characteristics, yielding a final sample of 20,745 adolescents who participated in Wave I of the 



26 
 

Add Health in-home interviews. Additionally, researchers attempted to interview one parent 

present in the home of each adolescent participating in Wave I to gain more contextual 

information. Participation was high (N=17,670), although parental respondents primarily 

consisted of mothers (Harris, 2013). 

 Unlike the in-school questionnaire, the in-home interview was repeated across multiple 

waves resulting in a set of panel data constituting the longitudinal component of the Add Health 

study. Wave II of the in-home interviews were successfully completed for 14,738 adolescents 

one year after the initial Wave I in-home interviews. These first two waves contain numerous 

items designed to tap key constructs of interest during adolescent development, such as family 

context, peer networks, substance use, and delinquency, all of which will be used in the current 

thesis.  

 Wave III in-home interviews were conducted between August 2001 and April 2002 for 

15,170 participants from the Wave I in-home interview. At this time, participants ranged in age 

from roughly 18-26. During the first two waves of Add Health, respondents were not asked 

about their experience with parental incarceration. This changed in Wave III when respondents 

were asked to report whether their father had ever been to jail or prison, up until the time of 

Wave III data collection. However, Wave III of the in-home questionnaire lacks information on 

the timing of parental incarceration, specifically how old respondents were when their fathers 

were first incarcerated and most recently incarcerated. Therefore, it is impossible to discern 

whether respondents were exposed to paternal incarceration before, or after the Wave I in-home 

questionnaire using the retrospective measures contained in the Wave III in-home questionnaire 

data. This is key to the analysis at hand because outcome measures are taken from Wave II and 

so measurement of the independent variable must occur at Wave I to ensure proper temporal 



27 
 

ordering. Thus, data from Waves I and II will be linked with data from Wave IV of Add Health, 

since additional items were included in this wave which can be used to determine the 

approximate timing of a respondent’s first and most recent experience with paternal 

incarceration.  

Wave IV data collection occurred from 2008-2009 when respondents (N=15,701) were 

aged 24-32. At this time, respondents were also asked if their mother had ever been to jail or 

prison. A relatively small number of respondents indicated experience with maternal 

incarceration and this number is even smaller after accounting for timing of incarceration and 

missingness across other covariates. Therefore, in addition to the reasons noted earlier, the 

present analysis will only assess the effects of paternal incarceration and how they are moderated 

by the attitudinal and behavioral orientation of one’s friendship network. It is also worth noting 

that retrospective reports of life events are never ideal due to potential inaccuracies, especially in 

circumstances where a substantial amount of time has passed since the event occurred, as is 

likely the case for most respondents reporting exposure to paternal incarceration (Blumstein, 

Cohen, & Farrington, 1988). Unfortunately, many longitudinal datasets that contain measures of 

both developmental outcomes and parental incarceration suffer from this issue (Murray, 

Farrington, & Sekol, 2012).  

Final Sample 

 My final analytic sample is substantially smaller than the original Wave I in-home 

sample. Figure 1 details the attrition. The initial Wave I in-home questionnaire yielded an initial 

sample of 20,745 respondents. However, the sample suffers from a sizeable level of attrition by 

Wave IV. Although AddHealth obtained an 80.3% response rate to their Wave IV in-home 

questionnaire, this response rate does not reflect the number of respondents who also provided 



28 
 

responses for each prior wave (Harris, 2013). Indeed, only 11,863 respondents are represented 

across Waves I, II, and IV, meaning that across Waves I and IV nearly 9,000 respondents are 

missing from at least one Wave of the data required for the present analysis. The sample size 

then dropped once again when retaining only those individuals who are also represented in the 

Wave I network data. Over 3,000 respondents in the home sample did not provide information 

for the in-school friendship network data, yielding a sample of 8,137 respondents who are 

represented across all time points. Of the 8,137 respondents, 1,207 did not nominate any in-

school friends who reported their own data, and 392 were missing data on paternal incarceration. 

These respondents were dropped, in addition to 68 respondents missing data on the key 

outcomes of interest, yielding a final sample of 6,470 respondents. Listwise deletion of 

respondents due to missingness across controls would result in the loss of roughly 600 additional 

cases (N=5,884). Missingness across controls is documented in Appendix A. To avoid this 

further data loss, multivariate normal multiple imputation was used to generate values on 

covariates for respondents with incomplete or missing data. Multiple imputation is a statistical 

technique through which missing data in an analytic sample is replaced with estimates of what 

would have plausibly been observed had the data not been missing (Rubin, 1987). This is done 

by generating multiple datasets based on the posterior predictive distribution of the missing 

values (Rubin, 1988). Given that none of the controls used in this analysis were missing more 

than 7% of their observations, I imputed 20 datasets based on recommendations from the 

empirical literature (Graham, Olchowski, & Gilreath, 2007). Regression analyses are run 

separately on each of the imputed datasets and their subsequent results are averaged using 

standard combining rules (i.e., Rubin’s Rules) to produce a single set of estimates intended to 



29 
 

approximate the results that would have been observed in the original dataset had there been no 

missing data (Li, Stuart, & Allison, 2015). 

 The original sample which results from merging Waves I, II, and IV is composed of 

11,863 cases. In comparing this original sample to the final sample used in this thesis, several 

demographic characteristics are statistically different from each other at the α=.05 threshold. 

Compared to the initial sample the final analytical sample prior to imputation is composed of 

significantly more non-Hispanic Whites and non-Hispanic Blacks, but significantly fewer 

Hispanics. My sample is also older by half a year on average and comprised of more female 

respondents. The statistically significant demographic differences were as follows: non-Hispanic 

White (µ=0.561, SD=0.496), non-Hispanic Black (µ=0.21, SD=0.408), Hispanic (µ=0.148, 

SD=0.355), sex (µ=0.457, SD=0.498), and age (µ=14.678, SD=1.570). There were no 

statistically significant differences between the outcome measures contained in the original and 

final samples. 

 



30 
 

Figure 1. Attrition Analysis 

 
 

Measures 

Dependent Variables 

Delinquency: During Wave II of the in-home questionnaire respondents were asked about how 

often they had engaged in an array of violent and non-violent offenses over the course of the past 

12 months. These offenses include (1) graffiti, (2) damaging someone else’s property, (3) taking 

something from a store without paying for it, (4) motor vehicle theft, (5) selling drugs, (6) theft 

of something worth more than $50, (7) theft of something worth less than $50, (8) burglary, (9) 

robbery, (10) participating in a group fight, (11) pulling a knife or gun on someone, (12) 

shooting or stabbing someone, and (13) using a weapon in a fight. The first ten items were 

captured on a four-point scale. Responses options included “never=0,” “one or two times=1,” 

“three or four times=2,” and “five or more times=3.” The latter three items were captured on a 

three-point scale. Response options included “never=0,” “once=1,” and “more than once=2.” 

Dropped respondents with 
missing data on outcomes

Missing Paternal 
Incarceration Information

Droped respondents with 
no send network

Waves I, II, and IV  with 
Network Data

Wave I, II and IV

Wave I In-home Sample N=20,745

N=11,863

N=8,699

N=6,930

N=6,538

N=6,470



31 
 

Overall, delinquency as it is measured across these items is relatively rare among respondents in 

the AddHealth data with most respondents reporting that they have either never engaged in these 

acts or at most committed these acts one or two times (see Appendix A). Of course, this is 

intuitive as Addhealth is not a risk-based sample, essentially meaning that the sample is meant to 

be representative of the general population and not just those who youth who have the highest 

risk of engaging in delinquency. Nevertheless, of the 13 items, 12 loaded onto a single factor 

with an eigen value of 3.216 (individual items had factor loadings ranging from .474 to .630) 

indicating that these 12 measures appear to tap the same underlying construct, in this case, 

propensity for delinquency. However, the item capturing whether the respondent reported motor 

vehicle theft did not load above an acceptable threshold (|.4|). This may be due to the wording of 

the item in the in-home questionnaire. In the in-home questionnaire respondents are asked 

whether they have “drive[n] a car without it’s owner’s permission.” Although the wording of this 

question implies that the car was stolen, and although that may legally be the case, it is possible 

that the car in question belonged to the respondent’s parents given the relatively low likelihood 

that most youth would be capable of accessing and driving a car unless the keys were readily 

available, as would most likely be the case for a household vehicle. As such, it may be that this 

item would fail to load at an acceptable threshold for an underlying construct such as criminal 

propensity. Descriptives for the delinquency factor and all other covariates based on the 

summary results of the imputed data are presented in Table 1.  

 
Table 1. Descriptive Statistics (N=6,470) 

 
Variable Mean SD Min Max 

Outcome Variables     
Delinquency 0 0.919 -0.41 11.63 
Substance Use 1.245 1.285 0 6 



32 
 

Predictor Variable     
Father incarcerated any time before Wave I 0.083 0.275 0 1 

Moderator     
Prosocial Network Characteristics 0 0.941 -6.38 1.42 

Controls     
Black 0.197 0.389 0 1 
White 0.590 0.488 0 1 
Hispanic 0.133 0.340 0 1 
Other Race 0.080 0.271 0 1 
Male 0.433 0.495 0 1 
Respondent Age (Wave I) 15.196 1.543 11 20 
Parent Education 2.584 1.663 0 5 
Maternal Attachment 4.558 0.744 1 5 
School Attachment 11.571 2.499 3 16 
GPA  2.873 0.739 1 4 
Impulsivity 2.930 1.107 1 5 
Parental Supervision 4.200 0.686 1 5 
Unstructured Socializing 1.966 0.986 0 3 
Neighborhood Social Cohesion 0.751 0.318 0 1 
Perception of Neighborhood Safety 0.902 0.297 0 1 

Note: GPA = grade point average; The reported mean and standard deviation are taken from the results of 
the first of 20 imputed datasets. The minimum and maximum values are taken from the unimputed data to 
accurately reflect how all covariates were coded. 

 
Substance Use: During Wave II of the in-home questionnaire respondents were also asked about 

their experience with substance use in the past 12 months. On a binary scale, respondents were 

asked whether or not they had (1) smoked a cigarette, (2) drank alcohol more than two or three 

times, (3) tried or used marijuana, (4) tried or used any form of cocaine, (5) used inhalants, and 

(6) used any other type of illegal drug or pills, without a doctor’s prescription. Responses were 

coded “1” for yes and “0” for no. In the context of this thesis, substance use will be 

operationalized as a count variable with values ranging from 0-6 to reflect the number of 

substances each respondent reportedly used since the time of their last in-home interview. Within 

the final sample respondents reported using an average of 1.25 substances (SD=1.285). 



33 
 

Independent Variable 

Paternal Incarceration: During Wave IV of Add Health, respondents were asked if their 

biological father had ever spent time in jail or prison. Respondents were then asked to report 

their age at the time of their father’s first incarceration, as well as their age at the time of their 

father’s most recent incarceration release in the case that he had been incarcerated more than 

once. Using these three items in conjunction with Wave I data on respondent age at the time of 

the baseline home interviews I constructed a binary indicator of paternal incarceration status.4 

This variable indicates whether the respondent’s father was incarcerated at any time before the 

Wave I in-school questionnaire was administered. Respondents were coded as a “1” if they 

experienced paternal incarceration at any time prior to Wave I of the in-home questionnaire, and 

a “0” if they did not experience paternal incarceration before the Wave I in-home questionnaire. 

If respondents reported during Wave IV that they were not yet born at the time of their father’s 

first incarceration or most recent release from incarceration, they were counted as having 

experienced paternal incarceration pre-Wave I. Unfortunately, if respondents at Wave IV 

reported that they were the same age as they were at the time of the Wave I in-school 

questionnaire when their father was first incarcerated or most recently released, there is no way 

to determine if the incarceration was experienced just before Wave I in-school data collection or 

just after Wave I in-school data collection. In an effort to reduce the likelihood of a type 1 error, 

these respondents were coded as youth who did not experience paternal incarceration prior to 

Wave I. The proportion of the final sample experiencing paternal incarceration prior to Wave I is 

 
4 Add Health only asks about incarceration of the respondent’s biological father. Exposure to the incarceration of a 
non-biological father is not recorded in the data, and therefore cannot be accounted for in this thesis. 



34 
 

.083.5 Put more simply, I have 534 respondents with a history of paternal incarceration and 5,936 

with no history of paternal incarceration.  

Moderator 

Prosocial Orientation of Friends: During the Wave 1 in-school questionnaire, respondents were 

asked to identify five of their closest male friends and five of their closest female friends for a 

maximum of 10 friendship nominations. Adolescents were then linked to their self-nominated 

peers (send network) who were also administered the Wave I in-school questionnaire The 

network data contains averages of nominated friends’ responses for most of the items 

administered in the in-school questionnaire. These averages were calculated by summing the 

responses of a respondent’s self-nominated friends and dividing by the total number of friends 

the respondent nominated. Notably, this excludes nominated peers who were not administered 

the in-school questionnaire, either because they had already dropped out, attended another 

school, or refused to participate. Fortunately, respondents only made an average of 1.25 

friendship nominations that fell outside of the school sample, with an overall average of 5.7 

friendship nominations. Furthermore, by linking individuals to their nominated friends, measures 

can be constructed to reflect the nature of these social ties specifically, whether or not an 

individual’s social network is primarily comprised of peers who are oriented in favor of prosocial 

attitudes and behaviors. For the purposes of this thesis, prosocial peers will be measured by 

tapping into performance and behavior in school, future aspirations for conventional social 

bonds, and (lack of) participation in deviant activities.  

 
5 Less than 1% of my final sample was exposed to paternal incarceration prior to birth, precluding analysis of the 
effects of paternal incarceration that are a result of differential timing.  



35 
 

During Wave 1 of the in-school questionnaire, respondents’ friends were asked on a five-

point scale how often they have had trouble “getting along with teachers,” and “getting along 

with other students,” in the past 12 months. Responses to these items were: (0) never, (1) just a 

few times, (2) about once a week, (3) almost every day, and (4) every day. Both items were then 

reverse coded so that higher values indicate lower levels of interpersonal conflict and higher 

levels of adherence to prosocial norms of social cohesion. Friends were also asked how hard they 

try to do well on their schoolwork (school effort). Responses were recorded on a four-point 

scale: (1) “I try very hard to do my best;” (2) “I try hard enough, but not as hard as I could;” (3) 

“I don’t try very hard;” or (4) “I never try at all.” To stay consistent with the prior items, this 

item was also reverse coded so that higher values indicate higher levels of commitment to 

academic achievement, and thus a prosocial value orientation. The average of the friendship 

group responses on all three items is included in the Wave I network data. The network data also 

includes a measure of the averaged GPA across a respondent’s self-nominated friends to account 

for peer school performance, which are based on the same measures used to construct my control 

for academic performance.  

Respondents were also asked to report their perception of the likelihood that they would 

graduate from college on a nine-point scale with higher values indicating a greater perceived 

likelihood of the event happening. This item reflects peer aspirations for conventional social 

bonds. Because education is often linked to increases in informal social control, it stands to 

reason that youth with aspirations to graduate from college are generally oriented toward more 

prosocial values and developmental goals that are inconsistent with antisocial behavior (Nielsen, 

1999). Friendship group responses to this item were also averaged for each respondent in the 

Wave I network data. 



36 
 

Lastly, peer deviance is measured using peer self-reports of delinquent behavior. Unlike 

alternative data sources, Add Health provides peer self-reports of behavior, rather than relying on 

respondent perceptions of their peers’ behaviors. During Wave I of the in-school questionnaire, 

respondents’ friends were asked to report on their participation in the following deviant activities 

over the past 12 months: (1) smoking cigarettes; (2) drinking alcohol; (3) getting drunk; (4) 

doing something dangerous because they were dared; (5) lying to their parents or guardian; (6) 

skipping school without an excuse; and (7) getting into a physical fight. The first six items were 

measured on a seven-point scale: 0 (never); 1 (once or twice); 2 (once a month or less); 3 (2-3 

days a month); 4 (once or twice a week); 5 (3–5 days a week); and 6 (every day). The seventh 

and final item was measured on a five-point scale: 0 (never); 1 (1 or 2 times); 2 (3 to 5 times); 3 

(6 to 7 times); and 4 (more than 7 times). All items were reverse coded so that higher values are 

indicative of more prosocial peer networks.  

Because the value-orientation of one’s friends cannot be directly measured or observed, 

the degree to which one’s friends are prosocial can be understood as a latent construct (Byrne, 

1998). In order to capture this construct, this thesis relies upon a principal component factor 

analysis to confirm which of the aforementioned items reflect a prosocial network orientation 

given the available data. School effort, school performance, college aspirations, not smoking, not 

drinking, not getting drunk, not engaging in dangerous behavior, and not skipping school all 

loaded onto a single factor. Peer bonds with teachers, other students, and not fighting did not 

load onto the same construct and so these measures were dropped in favor of the majority of 

items, which loaded as expected. All retained items loaded on a single factor with an eigen value 

of 3.07 (individual item factor loadings ranged from .432 to .872). Using these results, a factor 

score was generated using the regression scoring method for all respondents based on the 



37 
 

retained network variables. Higher values are indicative of peer groups characterized by more 

prosocial value orientations. 

Controls 

Socioeconomic Status: Because the relationship between parental incarceration and adolescent 

deviance may be spurious due to socioeconomic status (SES). In the present thesis, higher 

educational attainment is used as a proxy for higher levels of SES. Wave I of Add Health 

includes items in the in-home questionnaire which capture the highest level of educational 

attainment among one’s parents with whom they reside. Add Health contains 12 possible levels 

of educational attainment that respondents can report for their residential father and residential 

mother. For simplicity, educational attainment for each parent was recoded into an ordinal 

variable with five levels, where 0=some or no high school, 1=high school graduate or GED, 2= 

vocational or trade school, 3=some college, 4=college graduate, and 5=postgraduate education. 

Finally, the highest level of educational attainment between the two parents was taken to 

construct a proxy measure of respondent SES. In the case of missing data for one parent, the 

educational attainment score of the non-missing parent was counted as the highest level of 

educational attainment. The average level of parental educational attainment in the final sample 

before multiple imputation was 2.59 (SD=1.662). 

 

Maternal Attachment: Hirschi’s (1969) social control theory suggests that deviation from societal 

norms is negatively related to attachment and research demonstrates that this holds true for 

children of incarcerated fathers (Wakefield, 2014). To address this possibility, a control is 

included for maternal attachment as mothers are often the primary caregivers for youth with 

incarcerated fathers (Glaze & Maruschak, 2010). During Wave I of the in-home interview, 



38 
 

respondents were asked to rate the degree of closeness between themselves and their mothers on 

a 1-5 scale with higher values indicating a greater degree of closeness between the respondent 

and their mother. The respondent’s rating of closeness to their residential mother, whether that 

happened to be their biological mother or not, was used. In the final sample before multiple 

imputation, the average level of maternal attachment was 4.56 (SD=.744) 

 

School Attachment: To be sure that respondents’ relationships with their friends moderates the 

relationship between paternal incarceration and delinquency net of school bonds, a control is 

included for respondent level of school attachment at Wave I. School attachment was measured 

in relation to each respondent’s level of agreement with the following three statements: (1) “[I] 

feel close to people at [my] school,” (2) “[I] feel like [I am] a part of [this] school,” and (3) “[I 

am] happy to be at [my] school These responses were captured using a five-point Likert scale 

ranging from strongly agree=1 to strongly disagree=5 and subsequently recoded so that higher 

values are consistent greater degrees of school attachment. Responses to these three items were 

summed to create a scale measuring each respondent’s level of school attachment (α=0.81). In 

the final sample prior to multiple imputation respondents reported an average school attachment 

of 11.57 (SD=2.637). 

 

Academic Performance: To account for respondent commitment to prosocial norms, academic 

performance during Wave I is controlled for using respondent self-reports of their grades in 

language arts, mathematics, social studies, and science (Hirschi, 1969) For each subject, 

respondents were asked to indicate whether they had received an A=1, B=2, C=3, D=4. These 

response options were then reverse coded and averaged to be consistent with how schools 



39 
 

calculate grade point averages (GPA). A metric that is frequently used as the primary means for 

evaluating academic commitment (Hirschi, 1969; Porter & King, 2015). The final sample before 

multiple imputation had an average GPA of 2.87 (SD=0.746). 

 

Impulsivity: Prior research has demonstrated that there is a link between impulsivity and 

delinquent behavior, especially among juvenile in disadvantaged contexts (Loeber et al., 2012; 

Lynam et al., 2000). Impulsivity is often implicated as a component of the scales typically used 

to measure Gottfredson and Hirschi’s (1990) conception of self-control. Furthermore, Hirschi 

(2004) later clarified that low self-control is the inability of individuals to fully consider the 

consequences of their actions. Therefore, I account for impulsivity using a single item, “when 

making decisions, you usually go with your ‘gut feeling’ without thinking too much about the 

consequences,” as it holds true to Hirschi’s reconceptualization of self-control, without relying 

on additional items contained in AddHealth, which may weaken construct validity. This item 

was coded on a five-point scale and reverse coded so that higher values indicate higher levels of 

impulsivity. Respondents forming the final sample prior to multiple imputation scored an 

average of 2.93 (SD= 1.108) on the impulsivity measure.  

 

Parental Monitoring Several studies have demonstrated that parental monitoring is another 

mechanism of social control under Hirschi’s (1969) principal of attachment, which inhibits 

juvenile delinquency (Farrington, 2011; Flanagan, Auty, & Farrington, 2019; Walters, 2021). 

Respondents were asked the following questions with respect to their residential mother’s 

presence in the home: (1) “How often is she at home when you leave for school?” (2) “How 

often is she at home when you return from school?” and (3) “How often is she at home when you 



40 
 

go to bed?”6 These three items are measured on a five-point scale with 1=always and 5=never. 

Responses to these items were reverse coded so that higher values indicate higher levels of 

maternal supervision. Similar items, using ‘he’ in place of ‘she,’ were asked about the 

respondent’s residential father or father figure, the items referencing the father were used to 

create a measure of paternal supervision. Responses to these three items for both mothers and 

fathers were then averaged to reflect the average level of maternal supervision and paternal 

supervision for each respondent. The highest average level of supervision between the two 

parents was used to represent the overall level of parental supervision. In the case of missing data 

for one parent, the supervision score of the non-missing parent was counted as the highest level 

of educational attainment. The average level of parental supervision in the final sample before 

multiple imputation was 4.20 (SD=0.686).   

 

Unstructured socializing: The relationship between unstructured socializing and juvenile 

delinquency is well-documented in criminological literature. Increases in unstructured 

socializing are typically associated with higher levels of juvenile delinquency (see Haynie & 

Osgood, 2005; Hoeben, Osgood, & Siennick, 2021; Osgood & Anderson, 2004; Osgood & 

Anderson, 2005; Osgood, Wilson, & O’Malley, 1996). Unstructured socializing is captured in 

this thesis through the following item from Wave I of the in-home interview: “During the past 

week, how many times did you just hang out with friends?” Responses were coded 0 (not at all); 

1 (1 or 2 times); 2 (3 or 4 times); or 3 (5 or more times). Higher values indicate higher levels of 

 
6 Respondents in Add Health are asked these items in reference to their residential mother and father, meaning the 
male and female figures in the household who they consider to be their parents regardless of whether these 
individuals are their biological parents.  



41 
 

unstructured socializing. In the final sample prior to multiple imputation, respondents report an 

average level of unstructured socializing of 1.966 (SD=0.986). 

 

Neighborhood Cohesion and Safety: Research has consistently found that collective efficacy 

reduces crime rates in the neighborhood context, meaning it is theoretically correlated with the 

predictor and outcome measures contained in this analysis (Morenoff, Sampson, & Raudenbush, 

2001; Sampson, Raudenbush, & Earls, 1997). Social cohesion, a component of collective 

efficacy, is accounted for using responses to three items taken from Wave I of the in-home 

questionnaire: (1) “you know most people in your neighborhood,” (2)  “in the past month, you 

have stopped on the street to talk with someone who lives in your neighborhood,” and (3) 

“people in this neighborhood look out for each other.” Respondents were asked to respond to 

these items with “true” or “false”. ‘True’ responses to these items were set equal to “1” and 

‘false’ responses to these items were set equal to 0. Responses to all three items were then 

averaged to generate a measure of neighborhood social cohesion for each respondent. A control 

is also included for perception of neighborhood safety as a proxy for neighborhood violence, as 

neighborhood violence partially mediates the relationship between neighborhood disadvantage 

and developmental outcomes (Harding, 2009). Perceptions of neighborhood safety are measured 

via a single item taken from Wave I of the in-home questionnaire where respondents were asked: 

“do you usually feel safe in your neighborhood?” Respondents were asked to answer with ‘yes’ 

or ‘no’. Affirmative responses were coded as ‘1’ and negative responses were coded as ‘0’. In 

the final sample before multiple imputation, the average level of neighborhood social cohesion is 

.751 (SD=0.318) and the average perception of neighborhood safety was 0.903 (SD=0.297). 

 



42 
 

Demographics: Lastly, I control for demographic characteristics such as age at the time of the 

Wave I in-home questionnaire, race, and gender. Age was measured using respondent reported 

age at Wave I and a quadratic term was also included to account for non-linear trends associated 

with age. Here I report the demographics of the final sample before multiple imputation to offer 

a clearer picture of my sample characteristics prior to any wide-scale manipulation of the data. 

The mean age was 15.19 (SD=1.560). Race was captured through an array of items where 

respondents were asked to report their race and ethnicity. Race was recoded into four dummy 

variables: non-Hispanic White (µ=0.590, SD=0.492), non-Hispanic Black (µ=0.198, SD=0.398), 

Hispanic (µ=0.133, SD=0.339), and ‘other’ race (µ=0.080, SD=0.271). The other category is 

inclusive of those reporting another race, such as Asian/Pacific Islander, Native American, or 

any other ethno-racial identity not contained within Add Health’s pre-generated racial categories. 

Gender is also a dummy variable with male respondents receiving the value of “1” and female 

respondents receiving the value of “0”. The mean sex of respondents in the final sample before 

imputation was 0.433 (SD=0.495), meaning slightly less than half the sample pre-imputation was 

male.7 All of these sample characteristics remain nearly identical to those resulting after running 

the multiple imputation. 

Analytic Strategy 

 There are two outcomes of interest for this thesis. The first is delinquency, which is 

measured utilizing a factor score. Factor scores are normally distributed with a mean of “0” and a 

standard deviation just under “1,” making the delinquency outcome a suitable candidate for 

analysis via ordinary least squares (OLS) regression, since it is approximately, normally 

 
7 Demographics are reported on the analytical sample prior to multiple imputation for transparency as to how the 
final sample relates to the original sample prior to data manipulation. 



43 
 

distributed. This also makes interpretation of interaction effects relatively straightforward. Model 

1 estimates the bivariate relationship between paternal incarceration and propensity for 

delinquency. Model 2 includes relevant controls. Model 3 includes an additional covariate for 

prosocial network orientation. Model 4 includes an interaction term between paternal 

incarceration and having prosocial peers to determine whether the latter moderates the effect of 

the former on delinquent tendencies. The second outcome measure is substance use, which is 

measured using a count variable. Because substance use is a count outcome with overdispersion, 

analysis of this outcome will use negative binomial regression. Model 5 estimates the bivariate 

relationship between paternal incarceration and substance use. Model 6 includes relevant 

controls. Model 7 includes an additional covariate for prosocial network orientation. Lastly, 

Model 8 replicates Model 7, but with the addition of an interaction term between paternal 

incarceration and peer network orientation. Unfortunately, interpretation of interaction effects 

utilizing a negative binomial model is not as straightforward as it is when relying on OLS (Mize, 

2019). Therefore, the interaction effect will also be probed via graphs to aid in visual 

presentation of the interaction between paternal incarceration and prosocial peers.  

Notably, it is also important to acknowledge that my results may be downwardly biased 

as some time varying controls may be measured after a father’s initial incarceration. This is the 

case because respondents in the data may have had their father incarcerated at any point before 

Wave I, but covariate measures are only available at the time of Wave I data collection. Put more 

simply, I cannot account for contextual factors at the time of an adolescent’s first exposure to 

paternal incarceration because the measures do not exist in the data. There are no baseline 

measures available at the respondent’s birth in the AddHealth data. As such, measurement of all 

controls across my models must take place at Wave I, which means that control values I observe, 



44 
 

may have been directly influenced by the adolescent’s exposure to paternal incarceration (i.e., 

respondent’s SES, neighborhood characteristics, school attachment, etc.).8   

 
8 A check for multicollinearity was performed for all models presented in the analysis and none had a VIF greater 
than 2. 



45 
 

Chapter 4: Results 

OLS Regression Models: Delinquency as the Outcome  

 Table 2 presents OLS regression results showing the association between paternal 

incarceration and propensity for juvenile delinquency. Bivariate results in Model 1 indicate that 

there is a substantively small but statistically significant (p<.001) relationship (b= .140) between 

paternal incarceration and the propensity for juvenile delinquency. After accounting for relevant 

controls (Model 2), paternal incarceration retains a positive, albeit diminished, direct association 

with adolescent propensity for delinquency (b= .103, p<.05). In Model 3, I include the prosocial 

orientation of one’s friendship network as an additional covariate. Once again, I find that 

paternal incarceration is positively related to adolescent propensity for juvenile delinquency 

(b=.089, p<.05). More specifically, paternal incarceration is associated with a .089 standard 

deviation increase in adolescent propensity for juvenile delinquency. The model also shows that 

prosocial network orientation is in fact negatively associated with adolescent propensity for 

delinquency, as would be expected under my theoretical framework (b= -.086, p<.001). In other 

words, a one standard deviation increase in prosocial network orientation is associated with a 

.086 standard deviation decrease in adolescent propensity for juvenile delinquency. Model 4 

contains an interaction term of paternal incarceration and prosocial network orientation to test 

my hypothesis concerning the ability of prosocial friends to specifically reduce the magnitude of 

the association between paternal incarceration and adolescent delinquency. Although the 

interaction term is negative, it is not statistically significant.  

 Across the multivariate OLS models, most, but not all controls are statistically significant 

predictors of adolescent propensity for delinquency. Using whites as a comparison, Hispanics 



46 
 

and non-black minorities are significantly more likely to engage in juvenile delinquency. 

Interestingly, blacks appear less likely to engage in delinquency, but this result is not statistically 

significant. Covariates representing conventional social bonds, such as maternal attachment, 

parental supervision, school attachment, and GPA are all negative