Family Matters: Moving Beyond “If” Family Support Matters to “Why” Family Support Matters during Reentry from Prison

Abstract

Objectives:

Informed by social control and differential coercion and social support theories, we examine how multiple theoretically and methodologically distinct factors of family support relate to reincarceration, substance use, and criminal offending during prison reentry.

Method:

Using four waves of data from the Serious and Violent Offender Reentry Initiative, we identified three separate factors of family support—interactional (e.g., providing guidance and support), instrumental (e.g., providing housing and transportation), and emotional (e.g., providing love and belongingness). A series of mixed-effects models examined how each form of family support related to reincarceration, substance use, and criminal offending.

Results:

Findings demonstrated that instrumental, but not interactional or emotional, support related to significantly lower odds of reincarceration and lower levels of substance use and criminal offending. Interaction terms revealed that the effect of instrumental family support is almost entirely independent, and not interactive, on each outcome.

Conclusions:

Family support appears to relate to prosocial reentry outcomes not because of emotional or interactional bonds, but because families provide for the basic needs of returning individuals. Instrumental familial support mechanisms such as providing housing and financial support appear more salient in promoting prosocial reentry outcomes than mechanisms of emotional or interactional support.

The process of release from prison—called “reentry”—has received increasing attention from researchers and policymakers alike (Lynch and Sabol 2004; Travis 2014). Despite a small decrease in the incarcerated population over the past two years, the United States continues to incarcerate more individuals than any other country in the world (Carson 2018). Because the vast majority of individuals who spend time in prison will be released (Carson and Anderson 2016; Travis 2005), understanding factors that promote successful reintegration has become an important task for the development of both theory and policy (see Seiter and Kadela 2003). Although research has highlighted a number of dimensions as important for successful reintegration including employment (Seiter and Kadela 2003; Uggen 2000), mental health care (Mallik-Kane and Visher 2008), treatment and reentry preparation (Prendergast 2009; Robbins, Martin, and Surratt 2007), and desistance from criminal peers (Boman and Mowen 2017), perhaps more so than any other factor, family support has been highlighted as an extraordinarily vital component for reentry success (see, generally, Arditti and Few 2006; Braman 2004; Breese, Ra’el, and Grant 2000; Naser and La Vigne 2006; Naser and Visher 2006; Nelson, Deess, and Allen 1999; Shapiro and Schwartz 2001; Visher and Courtney 2007; Western et al. 2015).

The importance of family support has been highlighted in both qualitative and quantitative research efforts using a variety of samples across the United States. Existing research has shown that family support relates to decreased recidivism (Boman and Mowen 2017; Shollenberger 2009; Visher and Courtney 2007), increased odds of employment (Berg and Huebner 2011; Visher, Debus, and Yahner 2008), and better mental health outcomes (Grieb et al. 2014; Wallace et al. 2016) during reentry. As Naser and Visher (2006:20) find, “for most former prisoners, relationships with family members are critical to successful reintegration.” In addition to a moderate level of empirical support, these findings can be understood via criminological theory. From the perspectives of both social control (Hirschi 1969) and differential coercion and social support (DCSS; Colvin, Cullen, and Vander Ven 2002) theories, family members provide both strong affectionate bonds (like emotional support and attachment) and important mechanisms of social support (like housing, transportation, and financial support) to the returning individual that serve to reduce recidivism and promote successful reentry outcomes.

Given the established empirical and theoretical support for the importance that family plays during the reintegration process, it is imperative for researchers to not just ask if family support matters but also examine why family support makes a difference during reentry. With a small handful of exceptions (Martinez 2006; Taylor 2016), the specific mechanisms through which family matters during this process remain unclear. Is family support important for reintegration because of emotional support characterized by love, warmth, and affection? Alternatively, does family support matter due to the presence of instrumental support–like financial resources, transportation, and housing? Or do both perspectives carry explanatory power? Consequently, a shift in focus to understanding why family matters is an important step to take to enhance theoretical clarity concerning the specific mechanisms through which family support matters and to provide evidence for scholars to make sound and empirically grounded policy recommendations.

Accordingly, the current study seeks to accomplish two goals. Using longitudinal panel data from the Serious and Violent Offender Reentry Initiative (SVORI), we first seek to examine how a variety of family dynamics that broadly represent family support may constitute a series of theoretically and empirically distinct subfactors of family support (e.g., emotional or instrumental family support; see Martinez 2006). Second, we seek to examine whether these different forms of family support independently or interdependently relate to reincarceration as well as self-reported substance use and criminal offending during the reentry process.

Why Family Matters: Theoretical Orientations

Social Control Theory

Drawing from the classic formulation of social control theory, family provides important bonds that prevent offending (Hirschi 1969). According to Hirschi, attachment refers to the close affectional ties an individual has to significant others and particularly parental and other family figures. These affectionate ties often include emotional bonds and feelings of warmth and companionship. Although some studies find mixed support for the theory (e.g., Agnew 1991), research tends to support the notion that individuals with strong bonds to conventional others report lower levels of deviance than individuals with weaker bonds (Agnew 1993; Baier and Wright 2001; Durkin, Wolfe, and May 2007; Krohn and Massey 1980; Loeber and Stouthamer-Loeber 1986; Sampson and Laub 1990).

Within the research on prison reentry, the importance of family as a central mechanism that provides strong bonds has been steadily realized. For example, men and women tend to report high levels of perceived emotional family support both prior to release and during the reentry process (Visher et al. 2007). Family members are far more likely to overlook the negative stigma attached to “having a record” than other members of society (Ekland-Olson et al. 1983), and prior work has demonstrated that family members report wanting to emotionally support their returning family member (Grieb et al. 2014). In fact, families tend to mark a relative’s release with a celebration to express their affectionate support (Western et al. 2015). Individuals who lack affectionate family support, or who experience decreases in affectionate support during the reentry process, may be placed at risk of reoffending.

DCSS

Sharing some similarities to social control theory, the social support component of DCSS refers to the feelings of support one receives from another individual (Colvin et al. 2002; Cullen 1994). The presence of social support creates strong bonds among the individual and significant others that reduce criminal offending (Colvin et al. 2002). According to Colvin, Cullen, and Vander Ven (2002), social support promotes noncriminal coping strategies in the presence of adversity, and individuals who receive consistent social support should, therefore, desist from criminal offending. On the other hand, a lack of consistent social support may result in less desirable outcomes. When individuals fail to receive consistent social support from others, they “are more or less being set adrift to fend for themselves as best they can” (Colvin et al. 2002:25).

When applied to the context of reentry, DCSS suggests that individuals who experience decreases in family support across time should report increases in offending, substance use, and reincarceration risk. Moreover, while social control theory tends to focus primarily on affectionate bonds (see Akers, Sellers, and Jennings 2012:123), DCSS allows for social support to also be provided through instrumental supports such as housing, employment, and financial support. Indeed, prior research demonstrates that this instrumental role is routinely filled by family members via providing transportation (Hlavka, Wheelock, and Jones 2015), housing (Western et al. 2015), and financial resources to returning individuals.

Overall, both social control theory and DCSS suggest that families provide important mechanisms of support during the process of reintegration. Whether the support is emotional/affectionate (e.g., attachment; Hirschi 1969) or instrumental (e.g., material social support; Colvin et al. 2002), families are perhaps the most significant source of support during reentry. Unfortunately, it remains unclear how family support matters. Stated differently, criminologists do not fully understand whether emotional/affectionate support, instrumental support, or some other combination of family support reduces recidivism the most. To explore this issue further, we now turn to the research on family support during reentry.

Empirical Research on Family Support and Reentry

The knowledge that family support is a key contributor to prosocial reentry outcomes is well established in the literature. In fact, the titles of research projects examining family as an important bond often use the metaphor “ties that bind” in both the broader research (Bersani and Doherty 2013; Herreros 2015; Wade 2008) and reentry research specifically (see Berg and Huebner 2011; Boman and Mowen 2017). These titles reflect the wide range of criminological research that has shown that ties to family relate to decreased criminal offending and delinquency (Berg and Huebner 2011; Bersani and Doherty 2013; Boman and Mowen 2017; Canter 1982; Sampson and Laub 1990, 1993; Savolainen 2009; Wright, Cullen, and Miller 2001).

At the same time, there remain both theoretical and empirical limitations on understanding how and why family matters. For example, existing research finds that high levels of family support do not necessarily decrease the link between victimization and recidivism during reentry (Taylor 2015b). Prior work has also suggested that family support may have an indirect—but not a direct—effect of recidivism (Berg and Huebner 2011). At least one study also demonstrates that supportive relationships may also be marked by conflict (Mowen and Boman 2018a), implying that conflictual family relationships may undermine the prosocial ties provided by family support. Although more research is warranted to examine the intricacies of the role of family support during reentry, research tends to support the notion that family support matters.

Despite the understanding that family support decreases a variety of deviant outcomes, scholars have recently called for increased specificity in understanding the mechanisms through which family support provides these necessary ties. In perhaps the first—and most poignant—call for researchers to consider how different forms of family support relate to reentry outcomes reentry, Martinez (2006:34, emphasis in original) implored scholars to examine specific dimensions “[b]ecause support has not been clearly defined or used similarly across studies.” Martinez suggests that researchers must begin to examine a variety of different types of family support mechanisms to provide “clarity as to the exact nature of familial support” (p. 34) on reentry outcomes.

To date, only a handful of studies have grappled with examining how different dimensions of family support play important roles during the reentry process, though a number of projects have implicitly explored multiple dimensions of family support. For example, in a study of individuals returning in the Boston area, Western and colleagues (2015) find that families provide housing and income support to returning family members. In fact, the vast majority of returning individuals had contact with their family on their first day out, and many lived with these family members immediately following release. Detailing the experience of one individual named Nick, Western and colleagues found that Nick’s sister found him a job in construction prior to his release and, upon release, invited him to live with her family in their apartment. Beyond these forms of practical support, Nick’s sister was also reported as being emotionally supportive (p. 1524). While this narrative paints a portrait of the importance of family, it does not directly provide an understanding of how these different forms of support (instrumental or emotional) relate to reentry success.

In a study using the Returning Home data, Mowen and Visher (2016) recognize that multiple dimensions of family support are affected by incarceration. For example, the authors show that some experiences during incarceration can change the amount of instrumental and emotional support an individual receives at release. However, the authors do not demonstrate how specific forms of family support relate to reentry outcomes. In interviewing returning men in multiple states, Visher and Courtney (2007) find that the most commonly cited factor keeping individuals out of prison was family support. Likewise, Nelson, Deess, and Allen (1999:10) find that “self-defined family support was the strongest predictor of individual success.” Yet it is not clear in this study whether individuals are referring to emotional support, instrumental support, some combination of the two, or something else entirely (see also Grieb et al. 2014). Despite this shortcoming, these findings are important as they demonstrate that family support is a multifaceted construct that does impact reentry outcomes.

Only a few studies have examined how multiple forms of family support relate to reentry outcomes. Using the SVORI data, Taylor (2016) summed 10 items encompassing a variety of affectionate-type supports (e.g., my family loves me, I have someone in my family who listens to me) labeled “emotional support.” The author also summed five items capturing instrumental types of support (e.g., someone in my family can provide transportation, support on substance abuse) labeled “instrumental support.” Using logistic regression, Taylor found that individuals with higher emotional support reported lower offending than those with lower emotional support. Instrumental family support did not significantly relate to offending between individuals. Another study published by Taylor (2015a) using the same data and measures found that emotional support was more strongly associated with desistance for females than males. Finally, using a sample of 255 returning individuals, Barrick, Lattimore, and Visher (2014) used identical questions as Taylor (2016) to examine how family emotional support and instrumental support related to reincarceration. As emotional support only marginally associated with reincarceration, the findings were mixed. Overall, these studies provide an excellent place to start as they collectively demonstrate that different family support mechanisms may relate to reentry outcomes, albeit in different ways.

Despite Taylor’s (2015a, 2016) and Barrick et al.’s (2014) important findings, two important questions remain unanswered. First, as each of these studies used the same data and identical measures, the factor structure among these items may demonstrate that family support is comprised of more than two factors. In particular, the 10-item measure of emotional support in all of these studies combined a variety of measures of support (e.g., my family loves me) as well as conflict (e.g., I fight a lot with my family). Drawing from the perspectives of social control and DCSS theories, prior work has demonstrated that family support and conflict are theoretically and empirically distinct (e.g., Mowen and Boman 2018a; Mowen and Visher 2015), and the presence of conflict is not the same as the absence of support. Further, it is possible that the remaining items may represent other forms of support (e.g., emotional, interactional—see Martinez 2006). Thus, a more rigorous analysis is necessary to explore whether family support may be comprised of distinct factors.

Second, and perhaps more importantly, both social control theory and DCSS would expect that changes in family support relate to reentry outcomes. Decreases in particular forms of family support may relate to significant increases in offending. This is particularly important as prior projects have demonstrated that family support changes across time during the reentry process in consequential ways for behavior (see, generally, Boman and Mowen 2017; Stansfield et al. 2017; Wallace et al. 2016). For example, Western and colleagues (2015) document the experiences of people returning to the Boston, MA, area. The researchers carefully document how individuals experience changes in the frequency to which they interact with family, live with family, and receive some forms of support (like financial assistance) from family. Likewise, using the Returning Home data, Mowen and Visher (2015) note that family relationships and dynamics change during the reentry period as individuals work to reintegrate. Similarly, analyzing data from returning individuals in Baltimore and Chicago, Naser and La Vigne (2006) found that prerelease expectations of family support did not necessarily match up to postrelease family support experiences, again highlighting that family support levels, whether perceived or actual, change over the course of reentry. In sum, these studies have demonstrated that family relationships and dynamics—like support—change within persons over time. However, existing research has tended to focus exclusively on between-person differences (e.g., an individual “with” support compared to an individual “without” support). Given the significant amount of change that occurs postrelease as individuals negotiate postrelease identities, roles, and relationships, it is likely that within-person changes—in addition to between-person differences—in family support carry important meaning for postrelease behaviors. These two important limitations draw attention to the goal of the current study.

Current Study

While family support is clearly related to reentry success, the specific means through which family support provides the “ties that bind” (e.g., Boman and Mowen 2017) has proven to be particularly difficult to pin down. To enhance clarity on this issue, we use four waves of data from the SVORI to examine the extent to which different forms of family support relate to reincarceration, self-reported substance use, and self-reported offending over time. Because prior work has largely failed to distinguish between different forms of family support (Martinez 2006), we make only three broad hypotheses. First, we expect that multiple individual items capturing family support will represent distinct factors of family support (Hypothesis 1). Second, we expect that all factors of family support will relate to decreased odds of reincarceration and lower levels of substance use and offending during the reentry time frame (Hypothesis 2). Third, we expect that family support measures will be interdependent—and not independent—in their effects on each outcome measure via multiplicative interactions (Hypothesis 3).

Method

Data

Data for this project come from the SVORI. SVORI, a federally funded initiative, was designed to examine how enhanced reentry services related to criminal justice (CJ), employment, education, housing, and health outcomes among returning individuals. The National Institute of Justice provided funding to the Urban Institute to collect panel data on the SVORI project (Lattimore and Visher 2009) to examine whether enhanced reentry services improved reintegration outcomes. Specifically, the SVORI was a quasi-experimental evaluation study encompassing a total of 1,697 males across 12 different SVORI sites. Approximately half of the individuals received enhanced reentry programming, including a variety of practical services (like needs assessment, reentry planning, and reentry classes) as well as individual change services (like mental health treatment, anger management, and educational resources; see Visher et al. 2017). Although findings on the efficacy of SVORI programming are mixed (Lattimore and Visher 2009; Visher et al. 2017), the breadth of information collected has provided researchers with ample data to examine a variety of topics related to prison reentry.

The data collection provided four waves of panel data. Wave 1 was collected from the respondent approximately 30 days prior to their scheduled date of release, wave 2 data collected at three months postrelease, wave 3 at about nine months postrelease, and wave 4 data collected 15 months from the day of initial release. Individual sites largely determined how individuals were selected into the SVORI program (see Lattimore and Steffey 2010). Although individuals identified as SVORI participants were more likely to receive reentry services, non-SVORI participants also reported receiving some assistance (Visher et al. 2017). Finally, though the SVORI data also included a subsample of adult females and juvenile males, we use only the adult male sample due to the small sample size and high rate of attrition in the female and juvenile subsamples.

At each wave, respondents were asked about a variety of dynamics and experiences including criminal offending and substance use, family relationships and experiences, treatment and programming, mental and physical health, and a variety of additional measures (see Lattimore and Steffey 2009, for an overview). In addition to data collected from each respondent, data were collected from the state departments of corrections for each site as well as the National Center for Crime Information (NCIC), providing official measures of recidivism (i.e., reincarceration) and criminal history. In the forthcoming analysis, we use data from all four waves but concentrate primarily on the postrelease waves as the outcome measures (reincarceration, substance use, and criminal offending) could only occur following release (waves 2, 3, and 4). Our total sample size encompasses 1,002 individuals (sample attrition is discussed at length in the subsequent section).

Dependent Measures

In this study, we use three time-variant measures encompassing reincarceration, substance use, and criminal offending. Common measures of recidivism include both rearrest and reincarceration. Given that rearrest may include events where charges are dropped, changed, or an individual is acquitted, reincarceration represents a more conservative and consistent measure across jurisdictions (Lyman and LuBoglio 2007). This measure captures whether the respondent had been reincarcerated at each postrelease wave (this includes being reincarcerated for a technical violation). Overall, a total of 23.4 percent of respondents were recorded as having been reincarcerated sometime within the four waves of the SVORI data collection. However, this measure is time variant as individuals could be incarcerated at one wave and not in another. Overall, the within-individual standard deviation (0.246) is slightly smaller than the between individual standard deviation (0.374). Summary statistics for all measures are shown in Table 1.

Table 1.

Descriptive Statistics of the SVORI Sample.

Variable	Mean	SD	Range	SD Within	SD Between
Dependent measures
Reincarceration	0.234	0.424	0, 1	0.246	0.374
Substance use	0.751	0.628	0–2.336	0.462	0.484
Criminal offending	1.422	1.407	0–3.515	0.968	1.097
Family support measures
Interactional	9.739	1.98	3–12	1.079	1.762
Instrumental	12.922	2.419	4–16	1.221	2.194
Emotional	6.878	1.177	2–8	0.621	1.057
Family and demographic controls
Family contact	5.988	2.655	0–12	—	—
Family CJ history	2.217	1.099	0–3	—	—
Number of children	1.404	1.678	0–18	—	—
Married/stable partner	0.109	0.311	0, 1	0.141	0.408
Family conflict	3.905	1.304	2–8	0.648	1.198
Employment	0.657	0.475	0, 1	0.280	0.409
Depressive symptoms	7.843	3.872	5–25	2.002	3.531
Criminal peers	2.089	1.152	0–3	0.579	1.028
Race (White contrast)
Black	0.533	0.498	0, 1	—	—
Other	0.127	0.333	0, 1	—	—
Age	29.203	7.287	18–73	—	—
Criminal history, treatment, and programming
Primary conviction type (Other contrast)
Violent	0.154	0.367	0, 1	—	—
Drug	0.222	0.415	0, 1	—	—
Property	0.118	0.323	0, 1	—	—
Prior arrest	14.515	20.626	1–300	—	—
Prior convictions	5.967	8.268	1–90	—	—
Length of incarceration	918.297	932.353	44–6,486	—	—
Change in criminal behavior	0.438	0.496	0, 1	—	—
Substance abuse treatment	0.224	0.417	0, 1	—	—
Religious support	0.368	0.482	0, 1	—	—
SVORI participant	0.508	0.499	0, 1	—	—

Note: SVORI = Serious and Violent Offender Reentry Initiative; SD = standard deviation; CJ = criminal justice.

To create a measure capturing substance use, we draw from questions at each wave asking the individual if he had used marijuana, hallucinogens, cocaine, heroin, inhalants, sedatives, amphetamines, alcohol methadone, stimulants, or methadone. As some of these substances can be prescribed by a doctor, respondents were asked about illicit use and could respond yes (coded as “1”) or no (coded as “0”). Because these substances represent differing levels of severity, we apply severity weights outlined by Boman and Mowen (2018) and originally developed by Pandina, White, and Yorke (1981). As a result, more severe substances like heroin are not counted equally to less severe substances like marijuana.¹ To create a composite measure, we multiplied each substance by the appropriate weight and summed each. Given the skew in this measure, we use the natural logarithmic transformation (Zumel and Mount 2014). This logged measure has a mean of 0.751, standard deviation of 0.623, and ranges from 0 (no substance use) to 2.336 (a great deal of substance use). As respondents could report differing levels of substance use across the postrelease waves, the within-individual standard deviation is 0.462, which is about the same as the between-individual standard deviation (0.484).

The final dependent measure encompasses criminal offending. To create this measure, we draw self-reported data that asked the respondent (yes = 1, no = 0) if they had (1) threatened to harm someone, (2) physically harmed someone, (3) carried a gun, (4) sold any drugs, or (5) committed any property crimes. Again, because these acts represent very different levels of severity, we apply the weights developed by Wolfgang and colleagues (1985) and summed the items to create a composite index. This strategy that has been used in prior work using the SVORI data (see Boman and Mowen 2017). As this measure is also skewed, we use the natural logarithm in the forthcoming analysis. The logged measure has a mean of 1.422, an overall standard deviation of 1.407, and ranges from 0 (no offending) to 3.515 (a great deal of offending). This measure is time variant as individuals could report different behaviors in each wave. The time-variant within-individual standard deviation is slightly less than the between-individual standard deviation (0.968 and 1.097, respectively).

Independent Measures

The independent measures in this study capture family support. To identify distinct forms of family support, we draw data from a total of 15 questions in the SVORI questionnaire that asked the respondent about a variety of family dynamics/relations. These questions asked the respondent along a four-point Likert-type scale (1 = strongly disagree, 2 = disagree, 3 = agree,4 = strongly agree) whether or not he agreed with the following statements: (1) I feel close to my family; (2) I want family involved in my life; (3) I am a source of support for my family; I have someone in my family to (4) talk to about myself or my problems, (5) to turn to for suggestions, (6) who understands my problems, (7) to love me and make me feel wanted, (8) to provide help and advice finding a place to live, (9) to provide help of advice finding a job, (10) to provide support for a substance abuse problem, (11) to provide transportation to work or other appointments, and (12) to provide financial support; and to what extent, he (13) fights a lot with family members, (14) feels like a disappointment to family, and (15) is criticized a lot by family.

To determine the extent to which these individual items capture specific factors, we performed an orthogonal varimax rotated factor analysis, the results of which are shown in Table 2. Based upon factor loadings (those exceeding 0.600) and Eigenvalues (above 1.00), this factor analysis detected four total factors, three of which capture family support (Kim and Mueller 1978). The first factor captures the extent to which individuals believed they could draw upon their family to talk about their problems, turn to for suggestions, and rely on family for understanding. This is our measure of interactional support. The second factor captures the extent to which individuals could draw upon family for housing, employment, transportation, and financial support. This factor is referred to as instrumental support. The third factor encompasses the extent to which the individual feels close to their family and wants his family involved in his life. This is labeled emotional support. To create measures capturing each of these three factors, we summed the respective items, so that higher values indicate greater levels of family support. The fourth factor identified—which we discuss at greater length as a control measure—captures the extent to which individuals fight with and are criticized by their family members. This measure represents family conflict.

Table 2.

Orthogonal Varimax Rotated Factor Analysis for Family Support Measures.

Question	Factor 1: Interactional	Factor 2: Instrumental	Factor 3: Emotional	Factor 4: Conflict
I feel close to my family	.295	.272	.627	.220
I want family involved in my life	.340	.278	.625	.104
I am a source of support for my family	.229	.226	.334	.202
I have someone in my family … to talk to about myself/my problems	.761	.287	.141	.125
to turn to for suggestions	.809	.314	.179	.143
who understands my problems	.701	.325	.233	.181
to love me and make me feel wanted	.510	.520	.318	.154
to provide help/advice finding a place to live	.456	.691	.155	.159
to provide help or advice finding a job	.361	.751	.160	.129
to provide support for a substance abuse problem	.479	.473	.172	.067
to provide transportation to work/appointments	.237	.645	.223	.145
to provide me with financial support	.306	.656	.176	.185
I fight a lot with my family members	−.192	−.172	−.219	−.610
I feel like I disappoint my family	−.125	−.148	−.017	−.539
I am criticized a lot by my family	−.261	−.239	−.204	−.611
Eigenvalue	3.077	2.979	1.351	1.331

Note: Standardized factor loadings reported.

Table 2 shows descriptive statistics for these scales as well as the additional measures in this analysis. Overall, interactional family support has three items and a mean of 9.739, a standard deviation of 1.98, and ranges from 3 (very low interactional support) to 12 (very high interactional support). As support changes overtime, the within-individual standard deviation is 1.079. Instrumental family support has four items, a mean of 12.922, a standard deviation of 2.419, and ranges from 4 to 16. The within-individual standard deviation is 1.221 (between-individual standard deviation = 2.194). The measure of emotional support has two items, a mean of 6.878, a standard deviation of 1.177, and ranges from 2 to 8. The within-individual standard deviation is 0.621 (between-individual standard deviation = 1.057).

Family and Demographic Controls

We include a number of family and demographic controls. First, because prior research has demonstrated the importance of maintaining family contact on family outcomes and offending postrelease, we control for family contact during incarceration. To create this measure, we drew data from three separate questions that asked respondents the extent to which they had received in-person visits, phone calls, and mail from family members while incarcerated. Respondents could answer along a five-point Likert-type scale (0 = never, 1 = a few times, 2 = monthly, 3 = weekly, and 4 = daily). To create a scale capturing family contact during incarceration, we summed these responses. This measure has a mean of 5.988, a standard deviation of 2.655, and ranges from 0 (no contact) to 12 (a great deal of contact).

We also account for family incarceration, criminal, and drug use history by drawing data from three separate questions that asked the respondent whether any of their family members had ever been incarcerated, used drugs, or been convicted of a crime (all items measured 1 = yes, 0 = no). We summed these items together to create a three-position scale capturing family history (labeled “Family CJ History” in Table 2). Overall, this measure has a mean of 2.217, a standard deviation of 1.099, and ranges from 0 (no prior family member contact) to3 (all family contacts reported).

We also include a measure indicating the number of children the respondents had at wave 1. This measure has a mean of 1.404, a standard deviation of 1.678, and ranges from 0 children to 18. As very few men reported having additional children within the SVORI time frame, this measure is time invariant. To account for the influence of marital/relationship status (Bersani and Doherty 2013; Sampson and Laub 1993), we include a measure that asked the respondent if they were married or in a steady relationship (mean = 0.109, standard deviation = 0.311). As individuals could report changes in partnership, we include this as a time-variant measure (within-individual standard deviation = 0.141).

Finally, the previously discussed factor analysis on the questions assessing family relationships revealed a fourth factor capturing family conflict. As prior research has established the importance of the harmful aspect of family conflict (Mowen and Boman 2018a; Mowen and Visher 2015), we include it as a control. Two items asked the extent to which the individual fights with (item 1) and is criticized by (item 2) family members (both items have a range of 1 [never] to4 [a great deal]). Item scores were summed, so that higher scores capture higher conflict. The scale has an overall mean of 3.905, a standard deviation of 1.304, and ranges from 2 (no conflict) to8 (a great deal of conflict). Much like family support, levels of family conflict can change across time. As such, this measure is time variant (within-individual standard deviation = 0.648).

To account for the importance of employment (Skardhamar and Telle 2012; Uggen 2000), we draw from data at each wave that asked respondents whether they were legally employed. Overall, 65.7 percent of the respondents reported being employed. Individuals could—and did—report changes in employment status (within-individual standard deviation = 0.280).

To control for the importance of depression (e.g., Mallik-Kane and Visher 2008), we draw data from five measures encompassing depressive symptoms. Measured along a five-point Likert-type scale (1 = not at all, 2 = a little bit, 3 = moderately, 4 = quite a bit, and 5 = extremely), questions asked the respondent how much they felt: hopeless about the future, blue, lonely, like they had no interest in things, or worthless. Although these items have already been validated (see Wallace et al. 2016), the average Cronbach’s α across each wave is .809, indicating a high level of interitem reliability. This measure has a mean of 7.843, a standard deviation of 3.872, and ranges from 5 (no depressive symptoms) to 25 (high depressive symptoms). As individuals reported changes in depressive symptoms, we include this as a time-variant measure (within-individual standard deviation = 2.002).

To account for peer influence (Thornberry et al. 1994), we summed three items that asked whether the respondent had a close friend who had sold drugs, committed assault, or been incarcerated. Overall, this measure has a mean of 2.089, a standard deviation of 1.152, and ranges from 0 (no peer offending) to 3 (high levels of peer offending). As respondents could report changes in peer crime across time, we treat this measure as time variant (within-individual standard deviation = 0.579). The Kuder–Richardson 21 (KR21) reliability statistic (KR21 = .834) indicates a high level of reliability across these measures (Webster 1960).

We include two additional demographic controls to account for the influence of race and age. Specifically, we include a series of time-invariant binary variables indicating whether the respondent was Black (53.3 percent of the sample), White (32.8 percent of the sample), or Other race/ethnicity (12.7 percent of the sample). We withhold White as the contrast category in the forthcoming analysis. Finally, age is also included as a time-invariant measure (mean = 29.203, standard deviation = 7.287, range = 18–73).

Criminal History, Treatment, and Programming Controls

To account for criminal history, we first include binary measures that capture whether the respondent was convicted of a violent (15.4 percent), drug (22.2 percent), or property crime (11.8 percent) in contrast to any other crime. We also include measures that capture the total number of prior arrests (mean = 14.515, standard deviation = 20.626), prior convictions (mean = 5.967, standard deviation = 8.268), and length of incarceration (mean = 918 days, standard deviation = 932 days). As each of these measures is significantly skewed, we use the natural logarithm in the forth-coming analysis. To account for treatment, we include a binary measure indicating whether the respondent participated in treatment to change their attitudes and behaviors toward crime. Overall, 43.8 percent of respondents reported receiving this treatment. We also include a measure that indicated whether or not the respondent participated in substance abuse treatment (22.4 percent) in contrast to those who did not. As recent work has marked the importance of the social networks of support provided by religion for people in prison (e.g., Stansfield et al. 2017) and since religious involvement can promote family interaction (Wilcox 2002), we include a binary measure that asked the respondent whether he had received religious or spiritual assistance. Overall, about 36.8 percent of respondents reported that they received religious/spiritual support. Finally, as the goal of SVORI was to examine outcomes associated with SVORI program participation, we covary a variable indicating if the respondent was an SVORI program participant (50.8 percent) in contrast to nonparticipant.

Missing Data

Missing data within the SVORI sample have been well-documented (Lattimore and Steffey 2009). In our analysis, we draw data from a total of 1,002 respondents or about 59 percent of the original sample. Although we discuss the specifics of the analytic strategy in the following section, one of the key advantages of a mixed-effects model is that it leverages the ability to include both between-individuals differences and within-individual changes to retain individuals within the sample even if they do not have data on all measures in all waves (Rabe-Hesketh and Skrondal 2012). Specifically, the 1,002 respondents used in our analysis had at least two waves of complete data and thus could contribute to both between-individual differences and estimates of within-individual change.

A variety of prior publications offer in-depth discussions on attrition in the SVORI data (see Boman and Mowen 2018; Stansfield et al. 2017; Wallace et al. 2016). The conclusion that is reached is that patterns of missing data are largely at random. Reports published by the National Institute of Justice report that there are no key predictors of attrition across the four waves of data, further supporting this previous work. However, to confirm that our analyses were not significantly affected by attrition, we performed a sensitivity analysis by dichotomizing each measure and comparing patterns of missing data using t tests (see Brame and Paternoster 2003). No t test reached statistical significance. As a result, we conclude that while missing data are an issue in the SVORI data, our results are not significantly altered by patterns of missingness.

Analytic Strategy

Although there are a variety of methods suitable for the analysis of longitudinal panel data like SVORI, we use a mixed-effects model (Rabe-Hesketh and Skrondal 2012).² Because respondents were sampled at four distinct time points, the assumption of independence made by ordinary least squares regression is violated as individuals tend to be like themselves over time. This correlation requires the use of a random intercept to account for similarity across time (Raudenbush and Bryk 2002). A mixed-effects model introduces a random intercept and models the effects of both between-individual differences and within-individual changes on the outcome measure. As reincarceration is a binary outcome, we used a generalized nonlinear form of mixed-effects model (see Mowen and Culhane 2017) and a linear mixed-effects model for the substance use and criminal offending outcomes.

We note that one important assumption of the mixed-effects model is the assumption of equality (sometimes referred to as the assumption of endogeneity). That is, because a mixed-model examines both between-person differences and within-person changes on the outcome measure, it assumes that between-person differences exert similar effects as within-person changes. To examine whether this assumption is met, we use Hausman tests (see Rabe-Hesketh and Skrondal 2012) throughout the presentation of results to compare the mixed-effects estimates to fixed-effects estimates for each model.

To address our hypotheses, we first examine the bivariate relationships among the differing measures of family support and the three outcome variables used in the study. Then, we move to the multivariate framework to assess the influence of each measure of family support on reincarceration, substance use, and criminal offending while accounting for the control measures discussed above. Additionally, we expect the various dimensions of family support to interact with each other dimension of support in relation to the outcome measures. To examine this possibility, we introduce interaction terms encompassing each combination of family support to determine whether they are independent, or interdependent, in their effects. To properly create the interaction terms, we grand mean centered each variable and then multiplied them together (see Paccagnella 2006).

Results

A bivariate analysis (a t test; not shown) demonstrates that individuals who are reincarcerated during the SVORI time frame report significantly lower levels of family emotional, interactional, and instrumental support than individuals who are not reincarcerated. Similarly, bivariate analyses (two analyses of variance; not shown) show that higher scores on each of the three family support measures are related to significantly lower levels of substance use and criminal offending at the bivariate level. These findings, combined with the results from the factor analysis, demonstrate that each measure of family support is empirically distinct (r = about .40 across all three measures) despite each independently relating to positive reentry outcomes at the bivariate level. The question, however, is whether these results will remain once the measures of support are considered independently and interdependently in a more rigorous set of analyses.

Moving to the multivariate context, mixed-effects models examining the outcome of reincarceration are shown in Table 3. A Hausman test was used to compare whether the within-individual effect of the time-variant measures was equal in magnitude to the between-individual effect and vice versa. The test statistic was nonsignificant, indicating that the assumption of equality was met. Further examination revealed that the differences in the fixed- and mixed-effects coefficients were very small, thus further validating the results.

Table 3.

Results of Mixed-Effects Models Examining Reincarceration.

Variable	Model 1		Model 2		Model 3		Model 4
	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
Family support measures
Emotional	−.004	.061	−.006	.062	−.006	.062	−.005	.062
Interactional	.110	.096	.104	.098	.120	.096	.153	.101
Instrumental	−.136	.052**	−.136	.053*	−.127	.054*	−.127	.054*
Emotional × Interactional	—	—	−.007	.030	—	—	—	—
Emotional × Instrumental	—	—	—	—	.018	.013	—	—
Interactional × Instrumental	—	—	—	—	—	—	.033	.023
Family/demographic measures
Family contact	.322	.221	.322	.221	.321	.221	.326	.222
Family CJ history	−.057	.089	−.057	.089	−.055	.089	−.057	.090
Number of children	.258	.172	.260	.172	.251	.172	.250	.172
Married/stable partner	−.524	.308	−.526	.309	−.515	.308	−.508	.309
Family conflict	.060	.076	.060	.076	.058	.076	.063	.076
Employment	.458	.197*	.460	.197*	.446	.198*	.453	.197*
Depressive symptoms	.244	.028***	.245	.028***	.242	.028***	.245	.028***
Criminal peers	.183	.065**	.182	.065**	.186	.065**	.186	.065**
Race
Black	.275	.228	.275	.228	.282	.228	.287	.229
Other	.243	.306	.242	.306	.241	.306	.246	.307
Age	−.049	.016**	−.049	.016**	−.049	.016**	−.049	.016**
Criminal history, treatment, and programming
Primary conviction
Violent	−.032	.271	−.029	.271	−.038	.271	−.041	.272
Drug	−.465	.261	−.466	.261	−.469	.260	−.475	.261
Property	.230	.293	.233	.294	.218	.294	.221	.294
Prior arrest	.274	.152	.275	.152	.270	.152	.270	.152
Prior convictions	.047	.173	.047	.173	.050	.173	.051	.173
Length of incarceration	.084	.125	.085	.125	.080	.125	.081	.125
Change in criminal behavior	.444	.221	.443	.221	.441	.221	.438	.222
Substance abuse treatment	.088	.217	.091	.217	.082	.217	.072	.218
Religious support	.008	.198	.009	.199	.005	.198	.004	.199
SVORI participant	−.137	.197	−.140	.198	−.130	.197	−.138	.198
Constant	−4.974	1.372***	−4.910	1.396***	−5.245	1.391***	−5.414	1.415***
χ2	98.44***		98.49***		98.24***		97.69***
Intraclass correlation	.418	.075	.419	.075	.417	.076	.421	.076
Random variation	.861	.311	.862	.310	.857	.313	.870	.312

Note: SVORI = Serious and Violent Offender Reentry Initiative; coef. = unstandardized coefficient, SE = standard error; CJ = criminal justice.

^*p ≤ .05.

^**p ≤ .01.

^***p ≤ .001.

As shown in Table 3, the significant χ² statistic indicates good model fit to the data while the intraclass correlation demonstrates that about 41 percent of the variation in reincarceration is within individuals across time. Turning to the results of the coefficients, model 1 demonstrates that emotional and interactional family support are not significantly associated with reincarceration. On the other hand, instrumental support is significantly associated with lower odds of reincarceration. Specifically, individuals with higher levels of instrumental support report lower odds of reincarceration net the effect of the other measures in the models. Converting the mixed-effect coefficient to odds ratios can assist in interpreting the magnitude of the effect. Overall, an increase of one unit on the family instrumental scale is associated with a 13 percent decrease in the odds of reincarceration both between-individuals and within-individuals across time.

Results of the family and demographic control measures indicate that employment is related to lower odds of reincarceration, while those with higher levels of family conflict, depressive symptoms, and criminal peers have significantly greater odds of reincarceration. Finally, older individuals report lower odds of reincarceration than younger individuals.

In order to examine the extent to which these measures of family support are conditionally related to reincarceration, we introduce interaction terms encompassing each combination of family support in models 2, 3, and 4. While the substantive results of the main effects and control measures remain substantively identical across each model, the interaction terms fail to reach significance, thus indicating that levels of family support are largely independent in their effects on reincarceration. Stated differently, the effect of instrumental support on reincarceration does not appear to be moderated by emotional or interactional family support.

Models in Table 4 examine the relationship between family support and self-reported substance use. Again, we invoked a Hausman test to ensure the models were robust to the assumption of equality and not affected by endogeneity across level 1 (within) and level 2 (between) effects. The Hausman test statistic was nonsignificant, indicating the assumption was met and the models unbiased by effect levels.³ Model 1 fits the data well and accounts for about 24.4 percent of the variation in self-reported substance use. Similar to the prior analysis, the mixed-effects regression in model 1 demonstrates that instrumental support is related to significant decreases in substance use. Individuals with high levels of instrumental support—or those who experience increases in instrumental support across time—report lower levels of substance use than those with lower instrumental support. More specifically, a one-unit change in family instrumental support is related to a .017 reduction in substance use. As the measure of substance use has a mean of 0.751, this effect is modest in size. Like the prior analysis, emotional and interactional support appear unrelated to substance use net the effect of the covariates in the model.

Table 4.

Results of Mixed-Effects Models Examining Self-Reported Substance Use.

	Model 1		Model 2		Model 3		Model 4
Variable	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
Family support measures
Emotional	1.42E-04	.007	5.57E-04	.007	−7E-05	.007	−4E-05	.007
Interactional	0.006	.010	0.008	.010	0.006	.010	0.008	.010
Instrumental	−0.017	.005**	−0.017	.005**	−0.017	.005**	−0.016	.005**
Emotional × Interactional	—	—	0.003	.003	—	—	—	—
Emotional × Instrumental	—	—	—	—	−0.001	.001	—	—
Interactional × Instrumental	—	—	—	—	—	—	0.002	.002
Family/demographic measures
Family contact	−0.017	.025	0.016	.025	−0.017	.025	−0.017	.025
Family CJ history	−0.002	.010	0.002	.010	−0.002	.010	−0.002	.010
Number of children	−0.027	.020	0.027	.020	−0.026	.020	−0.027	.020
Married/stable partner	−0.074	.030*	0.074	.030	−0.074	.030	−0.074	.030
Family conflict	0.035	.008***	0.035	.008***	0.035	.008***	0.035	.008***
Employment	−0.050	.019**	0.050	.019**	−0.049	.019**	−0.050	.019**
Depressive symptoms	0.018	.002***	0.018	.003***	0.018	.003***	0.018	.003***
Criminal peers	0.044	.006**	0.044	.006**	0.044	.006**	0.044	.006**
Race	−0.083		−0.083		−0.083		−0.082
Black		.026**		.026**		.026**		.026**
Other	−0.036	.037	−0.036	.037	−0.036	0.037	−0.036	.037
Age	−0.005	.001**	−0.005	.001**	−0.005	.001**	−0.005	.001**
Criminal history, treatment, and programming
Primary conviction	−0.013		−0.014		−0.013		−0.014
Violent		.032		.032		.032		.032
Drug	0.057	.029	0.056	.029	0.057	.029	0.056	.029
Property	0.071	.036*	0.071	.036*	0.072	.036*	0.071	.036*
Prior arrest	0.063	.017***	0.062	.017***	0.063	.017***	0.062	.017***
Prior convictions	0.051	.021	0.051	.021	0.051	.021	0.051	.021
Length of incarceration	−0.033	.013*	−0.033	.013*	−0.032	.013*	−0.033	.013*
Change in criminal behavior	−0.044	.026	−0.044	.026	−0.044	.026	−0.044	.026
Substance abuse treatment	0.038	.026	0.037	.026	0.038	.026	0.037	.026
Religious support	−0.016	.023	−0.016	.023	−0.016	.023	−0.016	.023
SVORI participant	−0.020	.023	−0.020	.023	−0.020	.023	−0.020	.023
Constant	0.541	.147***	0.524	.149***	0.550	.148***	0.521	.148***
χ2	448.73***		449.18***		449.43***		427.96***
R2	.244		.244		.244		.224
Intraclass correlation	.450		.450		.449		.474
Random variation	.271		.272		.272		.283

Note: SVORI = Serious and Violent Offender Reentry Initiative; coef. = unstandardized coefficient; SE = standard error; CJ = criminal justice.

^*p ≤ .05

^**p ≤ .01.

^***p ≤ .001.

Results of the control measures demonstrate that individuals in a steady relationship and Black respondents report lower levels of substance use. Individuals convicted of a property crime report higher levels of substance use than those convicted of another offense, and length of incarceration is related to decreased substance use. The effects of employment, depression, criminal peers, age, and family conflict are consistent with the prior models.

To again examine the independent or interdependent nature of emotional, interactional, and instrumental support on the outcome measure, we introduce interaction terms encompassing each combination of family support on offending. Like the prior analysis examining reincarceration, no interactions reach statistical significance in any model. As such, differing forms of family support appear to be independent in their effects on self-reported substance use.

The last series of models examines criminal offending as an outcome (see Table 5). The Hausman test statistic was not significant, indicating that the assumption of equality was met. Results demonstrate that high levels of instrumental support are significantly associated with decreased criminal offending. Specifically, a one-unit change in instrumental support is associated with a reduction of .037 on the offending scale. As the offending scale has a mean of 1.422, this effect is modest in size. Emotional and interactional family support is again not significantly associated with self-reported offending. Results of the controls closely mirror the prior results as depression, criminal peers, prior arrest history, and family conflict relate to high levels of self-reported crime. Age, on the other hand, is significantly associated with decreased criminal offending.

Table 5.

Results of Mixed-Effects Models Examining Self-Reported Criminal Offending.

	Model 1		Model 2		Model 3		Model 4
Variable	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
Family support measures
Emotional	.037	.020	.036	.020	.038	.020	.035	.020
Interactional	.007	.030	.007	.031	.009	.030	.024	.031
Instrumental	−.037	.017*	−.037	.017*	−.035	.017*	−.035	.016*
Emotional × Interactional	—	—	−.001	.010	—	—	—	—
Emotional × Instrumental	—	—	—	—	.005	.004	—	—
Interactional × Instrumental	—	—	—	—	—	—	.020	.008*
Family/demographic measures
Family contact	−.029	.073	−.029	.073	−.029	.073	−.027	.073
Family CJ history	.015	.031	.015	.031	.016	.031	.016	.031
Number of children	−.013	.059	−.013	.059	−.014	.059	−.016	.059
Married/stable partner	−.181	.091*	−.181	.092	−.180	.092	−.177	.092
Family conflict	.153	.024***	.152	.024***	.152	.024***	.154	.024***
Employment	−.072	.058	−.072	.058	−.075	.058	−.075	.058
Depressive symptoms	.070	.007***	.070	.007***	.069	.007***	.069	.007***
Criminal peers	.177	.019***	.177	.019***	.178	.019***	.178	.019***
Race					−.091		−.089
Black	−.092	.078	−.092	.078	.078			.078
Other	−.025	.108	−.025	.108	−.026	.108	−.025	.108
Age	−.025	.004***	−.025	.004***	−.025	.004***	−.025	.004***
Criminal history, treatment, and programming
Primary conviction
Violent	−.188	.096	−.187	.096	−.190	.096	−.196	.096
Drug	−.066	.086	−.066	.086	−.068	.086	−.073	.086
Property	.028	.106	.028	.107	.024	.106	.018	.106
Prior arrest	.140	.052**	.140	.052**	.139	.052**	.138	.052**
Prior convictions	.013	.061	.013	.061	.014	.061	.015	.061
Length of incarceration	−.015	.041	−.014	.041	−.016	.041	−.015	.041
Change in criminal behavior	−.010	.075	−.010	.075	−.011	.075	−.014	.075
Substance abuse treatment	−.033	.075	−.033	.075	−.034	.075	−.041	.075
Religious support	−.056	.069	−.056	.069	−.057	.069	−.058	.069
SVORI participant	−.068	.068	−.068	.068	−.067	.068	−.068	.068
Constant	.268	.441	.275	.445	.225	.442	.085	.446
χ2	452.00***		451.60***		453.57***		459.24***
R2	.233		.233		.234		.236
Intraclass correlation	.402		.402		.401		.403
Random variation	.766		.767		.765		.766

Note: SVORI = Serious and Violent Offender Reentry Initiative; coef. = unstandardized coefficient; SE = standard error; CJ = criminal justice.

^*p ≤ .05.

^**p ≤ .01.

^***p ≤ .001.

Models 2, 3, and 4 introduce interaction terms encompassing the various combinations of the different factors of family support. Interestingly, the interaction encompassing interactional and instrumental family support emerged as significant despite the main effect of interactional support remaining nonsignificant. To further explore this relationship, we split the interactional support coefficient into two separate effects capturing the within-person change in interactional support (b₁) and the between-person difference in interactional support (b₂). After reestimating with the two separate coefficients, results of these models (not shown) demonstrated that only the interaction encompassing the between-person effect of interactional support reached significance. As a result, this interaction demonstrates that instrumental support is especially important for individuals with lower levels of interactional family support. Put differently, people with comparatively low levels of interactional support report lower levels of offending when they are provided with ample amounts of instrumental support. Although this interaction emerged as significant, results of all other models demonstrated that instrumental support is largely independent in its effect on reincarceration, substance use, and criminal offending.

Discussion and Conclusions

The goal of this project was to move beyond the question of if family support matters during reentry and ask how family support matters. Using data from the SVORI, we sought to examine whether family support items represented distinct factors and, if so, how these factors related to reincarceration, self-reported substance use, and criminal offending. We now turn back to our hypotheses and prior literature to understand the findings of this study.

Our first hypothesis proposed that multiple items capturing family support would represent fundamentally distinct factors of family support. This expectation is confirmed. The results of a factor analysis on a series of items capturing support in the SVORI data demonstrated that these items encompassed three support factors—interactional family support, instrumental family support, and emotional family support. As Martinez (2006) suggests, the broad conceptualization of “family support” as a unidimensional construct is theoretically and methodologically imprecise. Instead, family support is comprised of a variety of distinct factors encompassing more affectionate-type bonds (emotional and interactional support) as well as more utilitarian bonds (instrumental support). In fact, it is likely that there are other forms of family support not captured in this study. By way of example, Martinez suggests that families may also provide support in the form of validation by providing feedback on the appropriateness of behavior, a notion consistent with tenets of social learning theory. Future research should explore how family support may be comprised of more than the three factors we capture in the current study.

The second hypothesis premised that all forms of family support would relate to prosocial reentry outcomes received mixed support. Through a series of mixed-effects models, results demonstrated that higher family instrumental support related to lower odds of reincarceration and lower levels of substance use and criminal offending. Yet, contrary to expectations, interactional and emotional support did not significantly relate to any outcome. Results relevant to our third hypothesis further help understand this phenomenon. Although we expected that the measures of family support would interact in their effects on each outcome measure, a complete lack of significant interaction terms leads us to reject the third hypothesis. Overall, findings relevant to Hypotheses 2 and 3 demonstrate that instrumental family support is independent, and not interdependent, in its effect on reincarceration, substance use, and criminal offending.

From the perspective of social control theory (Hirschi 1969), attachment should prevent offending since it is a form of an affectionate bond (it is both emotional and interactional). Our findings, however, suggest that affectionate-type bonds do not significantly relate to reincarceration, substance use, or offending during the reentry process. Although this contrasts with the primary tenets of the theory (see Krohn and Massey 1980), this work does support other research which shows that emotionally based family support fails to reduce substance use or offending for returning men (see Mowen and Visher 2015). It could be that supportive family relationships are also characterized by stress and emotional turmoil. For example, Berg and Huebner (2011) find that while individuals report high levels of emotional support from their families, these relationships are also characterized by fighting, upheaval, and emotional stress (see similar findings from Breese et al. 2000). As a consequence, while individuals report high levels of emotional and interactive support, the stress that accompanies a returning individual may reduce the effectiveness of these bonds in promoting desistance.

On the other hand, we find support for the notion that instrumental forms of family support promote successful reentry outcomes. From the perspective of DCSS (Colvin et al. 2002), individuals who experience instrumental social support may feel connected to others, and utilitarian measures of social support may provide individuals with the important ties they need to desist. Emotional and interactional support may both be undermined by the stress that accompanies the reentry process (see Breese et al. 2000). Even though the protective effects of emotional and interactional family support may be lost amid the strains of family life during reentry, instrumental support may be more tangible and predictable. In other words, an individual may continue to receive instrumental family support even if the relationship is emotionally strained.

When cast against the backdrop of research highlighting the ubiquity and importance of family during the reentry process (Visher and Courtney 2007; Naser and Visher 2006; Nelson et al. 1999; Western et al. 2015), our findings shed light onto the specific mechanisms and underlying reasons behind why family support matters. While prior research has shown that family relationships are central to reentry success, existing studies have overwhelmingly tended to treat family support as a single and monolithic characteristic. Results from this study clearly demonstrate that although higher reported levels of any family support improved reentry outcomes at the bivariate level, once specific types of family support were delineated and considered within a broader statistical framework, only instrumental support significantly related to prosocial outcomes. It appears, then, that instrumental support, relative to emotive types of support, is far more significant during the process of reentry.

This finding meshes with prior research conducted by Western and colleagues (2015) who told the story of reintegration of “Nick.” Upon returning from prison, Nick’s sister provided him with housing, financial support, and employment—all types of instrumental support. While the narrative does not discuss any other form of support (e.g., emotional or interactive) Nick may have received, the reality is that instrumental family support was highlighted as the key factor for Nick’s reentry success. Similarly, other studies that have highlighted family support as a key source of desistance (e.g., Visher and Courtney 2007) may also be tapping instrumental types of support instead of emotional types of support. At the same time, researchers should not simply discard the role of other forms of support as emotional and interactional support may play an important role in other reentry outcomes. However, our findings clearly suggest that the broad characterization that family support “matters” (e.g., Nelson et al. 1999; Western et al. 2015; Visher and Courtney 2007) may be overstated and simplified. Our findings, instead, provide evidence that it may be the instrumental component of family support—and not the emotional or interactive components—that motivates such strong and robust findings regarding family within the research on reentry.

In addition, findings from our study suggest that mechanisms of family support are independent, but not interdependent, in their effects on reincarceration, substance use, and criminal offending. This finding is important as additional models (not shown) combining all measures of family support in one metric demonstrated significant results for a combined and seemingly unidimensional measure of family support across each model. This once again demonstrates a clear lack of theoretical clarity and methodological precision as separating the various types of family support into methodologically informed factors uncovers the key differences.

Interestingly, our findings are nearly the opposite of findings reported by Taylor (2016) who used the same data to try to determine the impact of different types of family support during reentry. Our results also diverge from the findings of Barrick et al. (2014). We believe there are multiple reasons for these differences. First, our use of a factor analysis to examine the multiple dimensions of family support resulted in fundamentally different measures than those used by Taylor (2015a, 2016) and Barrick et al. (2014). Whereas the emotional support measure in these prior studies included 10 items, our analysis demonstrated that several of those items composed two methodologically distinct constructs (emotional and interactive support), while some items did not load consistently on any factor at all. Similarly, our instrumental measure was comprised of only four items relative to five by Taylor (2016) and Barrick et al. (2014). We believe the use of a factor analysis to construct multiple markers of family support (e.g., Martinez 2006) is a powerful tool which has helped determine which items are, and—equally importantly—are not, useful at helping inform why family matters during the reentry process.

To the prior point, a total of four items in the SVORI data failed to clearly load onto any single factor of family support. This includes questions that asked the respondent the degree to which they believed: (1) I am a source of support for my family, (2) I have someone in my family to love me and make me feel wanted, (3) I have someone in my family to provide support for substance abuse problems, and (4) I feel like I disappoint my family. There are potential reasons behind why each of these items failed to load cleanly onto any of the constructs. Item one captures the extent to which the individual wants to be a source of support for the family instead of how much he feels supported. As a result, this question may be capturing a fundamentally different component of support representing support given to others instead of support received from others. The second item’s loadings were split between emotional and interactional support. This is likely because the item taps the two separate dimensions of love (a form of emotional support) and a feeling of belongingness (a form of interactive support).

Capturing substance use support, the third item is somewhat different than the first two due to variability. People in the SVORI sample overwhelmingly have members of their family with substance use and abuse histories. As a result, respondents may have been less confident in receiving help desisting from substance abuse by family members who probably have histories of abuse themselves, thus leading this item to fail to load on any measure. Finally, the fourth question, which partially loaded onto family conflict, taps an emotional response while the remaining two items of family conflict capture actions (fighting and being criticized). Thus, this item captures a negative emotion probably more analogous to a lack of emotional support rather than family conflict, which could explain why family conflict has been shown to relate to reincarceration (Mowen and Boman 2018b). Collectively, this discussion demonstrates that family support measures are theoretically and methodological distinct. Beyond this, however, some individual items may overlap and simultaneously tap multiple distinct dimensions of family support and/or conflict. Future research should consider rigorous, measurement-focused methodological techniques (e.g., item response theory; de Ayala 2009) to more clearly leverage all items into separate dimensions.

A second explanation for our divergent findings from prior literature is likely due to the inclusion of family conflict as a separate construct that is different from simply the absence of family support (e.g., Mowen and Visher 2015). Our analysis included family conflict as a distinct control measure that was robustly related to two of the three outcomes. As a consequence, accounting for family conflict as a separate measure to family support likely explains some of the differences in findings as well. Relatedly, factor analytic techniques demonstrated that one item used in prior research as a measure of family conflict (e.g., Mowen and Boman 2018b) was capturing negative emotions and not necessarily conflict. Again, these findings strongly suggest that researchers need to better consider the complexities and multidimensional nature of measures of family support and family conflict. Third, and finally, our analysis examined both between-person differences and within-individual changes in contrast to Taylor’s (2015a) and Barrick and colleagues’ focus on between-individual estimates only. As a result, our models account for longer term changes across time, an important point which aids in the explanation of the divergence between this study and prior research on this topic.

Outside of the theoretical implications, this research also carries potentially important policy implications. Because instrumental family support is the only form of support linked to reincarceration, substance use, and offending, our findings suggest that policymakers should ensure that the outreach and support offered to families welcoming loved ones home from prison focus on instrumental assistance. With access to public housing and employment restricted by law in addition to amassed fees and other legal obligations, many people may face a prolonged reliance on families for instrumental support. It is thus important that families be kept informed of support resources available to them, including access to nonprofits which can provide food, shelter, and financial assistance (Travis 2005). In the absence of such support and assistance, families may experience tension between meeting the needs of the formerly incarcerated person and other family members (Christian, Mellow, and Thomas 2006). Innovative programs and policies aimed at developing financial planning and assisting persons released from prisons in making steady payments toward fees and other legal obligations may offer several benefits for both the released person and his or her family. While these programs may provide some relief for families, they may be even more significant for individuals who would not otherwise experience instrumental support.

Of course, there are cases where individuals face reintegration in isolation from family. Consequentially, instrumental support may not be able to be derived from family members in all situations. Drawing from the perspective of DCSS theory, other individuals and/or social groups could fill this role and deliver instrumental support. Supporting this proposition, recent studies (e.g., Chamberlain et al. 2018; Vidal, Oudekerk, and Reppucci 2015) show that individuals can derive social support from their probation/parole officer. As such, people who do not have family contact could potentially still benefit from the instrumental support that stems from other important social groups such as supervision officers, supportive peers, or members of a faith community during reintegration.

Despite the contributions of this project, it is not without notable limitations. First, as the SVORI data contain individuals classified as “serious” and “violent,” our findings may not be generalizable to a broader population. Future research should examine how multiple family support factors relate to reentry outcomes across other samples. Our control measure of peer criminality may also suffer from projection (see Boman and Ward 2014; Young et al. 2014), and this could have affected estimates of other variables in the model. If an analysis similar to this one could be performed with dyadic or network data, it would be extremely valuable.

In addition to the limitations above, it is likely that differing forms of family support have effects on outcomes beyond recidivism. It is possible that family emotional support—while perhaps not linked to recidivism— may be linked to employment or educational outcomes which, in turn, promote desistance (Denver, Siwach, and Bushway 2017). As a result, it is likely that different forms of family support may have direct effects on other outcomes and have indirect effects on offending. In a related point, we found that several items did not load cleanly onto any measure of family support. Despite estimating supplemental analyses using the factor loadings as weights, this did not change the substantive results of the bivariate or multivariate models. Future research is needed to explore the additional forms of family support as well as the overlap among the forms of support which people can receive when leaving prison.

Finally, DCSS also posits that social support can be derived from macro-level factors such as communities and social networks (Colvin et al. 2002). Placed within the context of reentry, the role of communities and neighborhoods in affecting social support and recidivism is an important—but often overlooked—factor, especially considering that individuals who return to disadvantaged neighborhoods are far more likely to recidivate than those returning to areas with more resources irrespective of microlevel characteristics (Kubrin and Stewart 2006). From the perspective of DCSS, economically disadvantaged and marginalized communities may have fewer forms of social support to offer and their support services may face greater demand (Hipp, Petersilia, and Turner 2010). Accordingly, neighborhood context may significantly impact a family’s ability to provide support to returning individuals. Prior work supports this notion, as Morenoff and Harding (2014:424) find that communities with fewer resources are “less able to provide supportive environments for those leaving prison.” Consequently, it is entirely possible that a family’s ability to provide instrumental support may be influenced both by the family’s socioeconomic status and community-level resources. Future research should examine how these macro-level processes affect the delivery and impact of family support during reentry.

Overall, this study serves as a reminder and a cautionary note for researchers that the reentry process is multifaceted and complex. Family support—so often considered critical for successful reentry and desistance processes more broadly—is not a monolithic or unidimensional construct. Instead, family support is multidimensional, and particular dimensions of family support may possess anti-criminogenic effects while other forms of family support may not significantly reduce recidivism at all. With specific regard to reentry, our findings suggest that affectionate and emotional types of family support may play a secondary role to the more utility-oriented form of instrumental support in promoting positive reentry outcomes. In other words, families may “matter” during reentry because they provide support in the form of caring for the basic needs of returning loved ones.

Author Biographies

Thomas J. Mowen is an assistant professor in the Department of Sociology at Bowling Green State University. His research examines the effect of punishment on youth and family outcomes and the role of family within the process of reentry from prison. Tom’s recent work has appeared in Criminology, Justice Quarterly, and Journal of Quantitative Criminology.

Richard Stansfield is an assistant professor of criminal justice at Rutgers University - Camden. He teaches and publishes broadly on the topics of violence, risk assessment, recidivism, and reentry with a particular interest in the role of religious services and the individual ways people make meaning in their desistance journey.

John H. Boman, IV is an assistant professor at Bowling Green State University in the Department of Sociology. His research is primarily focused on developmental issues, peers and social relationships through the life-course, and theory. John’s recent work has appeared in Criminology, Justice Quarterly, and Crime & Delinquency.