What Protects Those at High Risk from Criminal Justice Contact Despite the Odds? A Negative Case Analysis

Abstract

Criminal justice contact is a prevalent, if not expected, life event for many high-risk individuals with deleterious consequences; yet, many individuals at high risk are able to avoid this contact (i.e. negative cases exist). In this study, we draw on the life course framework and utilize negative case analysis to (1) estimate the prevalence of criminal justice avoidance within a sample of structurally high-risk Black men and (2) explore the individual, familial and contextual factors in childhood and adolescence that distinguish these negative cases. One’s own ‘on-time’ and one’s siblings’ education emerge as particularly strong protective factors suggesting that the presence of unique protection, as opposed to the absence of risk, may be most salient. Theoretical implications are discussed.

The premise that early risk sets a child on a trajectory that shapes his or her future is a prominent theme in developmental and life course research. Life course theories, in particular, are grounded in the principle that one’s past plays a significant role in one’s future (Elder 1985). According to this principle, high-risk children are expected to experience an array of negative life events that detriment later life course outcomes. One ‘expected’ life course event among high-risk youth that can be particularly consequential to positive development is contact with the criminal justice system. Mounting evidence demonstrates that justice system contact, even minor forms of contact (Harris 2016; Natapoff 2018), reduces educational achievement (e.g. Legewie and Fagan 2019) and employment (e.g. Pager 2003), detriments general and mental health (e.g. Sugie and Turney 2017), increases substance use disorders and perpetuates offending and criminal justice contact (e.g. Doherty et al. 2016). Whether these detrimental outcomes result from the stigma of contact with the criminal justice system and identification as a criminal (Lemert 1951; Becker 1963), by limiting opportunities for attaining conventional social bonds (e.g. Sampson and Laub 1997; Bernburg and Krohn 2003) or by facilitating deviant peer associations (e.g. Bernburg et al. 2006), it is clear that criminal justice contact perpetuates a negative trajectory that is rooted in childhood risk.

Race/ethnicity, gender and structural disadvantage constitute primary risk factors that are associated with criminal justice contact, whereby minority males (Brame et al. 2014) and those living in disadvantaged neighbourhoods (Kirk 2008) disproportionately experience this life event. A recent study using a nationally representative sample of the United States revealed that close to 50 per cent of Black males were arrested by age 23 compared to less than one-fifth of Black females and just over one-third of White males (Brame et al. 2014). Moreover, macro-level forces (e.g. hot spots policing and localized aggressive policing practices) contribute to the concentration of criminal justice presence in disadvantaged neighbourhoods, making the experience of arrest a more common event in these neighbourhoods (Kirk 2008). This disproportionality results in criminal justice interventions approaching the status of expected experiences for Black men from structural disadvantage (Pettit and Western 2004) and a ‘normalization’ of mass arrests and incarceration in disadvantaged communities (Hirschfield 2008: 597; see also Liberman et al. 2014).

Despite the pervasive and persistent reach of the criminal justice system in disadvantaged communities of colour, many youth predicted to be at high risk of becoming enmeshed in the criminal justice system are able to avoid this contact. The recognition that not all at-risk individuals exhibit negative outcomes has provoked research across many disciplines that focuses on the identification of protective factors that ‘diminish the likelihood of negative health and social outcomes’ (Resnick et al. 1997; see also Furstenberg et al. 1998; Luthar et al. 2000). Yet, with respect to arrest, it is not clear how prevalent this group of individuals is, nor what, despite the odds, predicts the avoidance of arrest among multiply at-risk youth. To address this gap, we examine criminal history and interview data from a sample of Black men who grew up in a disadvantaged, crime-prone community in the United States who have been followed longitudinally from age 6 through adulthood.

The current research makes three contributions to the literature. First, we employ a negative case analysis, a strategy that helps reveal divergent patterns of vulnerability and resilience by acknowledging that high-risk children often do not engage in health-compromising behaviour, develop mental illness or experience other negative ‘expected’ outcomes. This type of analysis broadens our inquiry beyond a focus on risk factors to include the identification of what protects these high-risk men from experiencing the expected (Rutter 1987; Resnick et al. 1997, 2000; Fergus and Zimmerman 2005). Second, to inform our study, we draw on prior research that demonstrates that resiliency in the face of adversity is not merely due to an absence of risk but is linked to unique benefits of protective resources and assets (Luthar et al. 2000). Third, separate from offending behaviour, we focus attention on justice system involvement, measured by arrest history, which is a particularly potent distinguisher of positive and negative life course outcomes.

Importance of Negative Cases

Criminologists have predominantly been concerned with the identification of risk factors that predict the ‘positive’ or ‘probabilistic’ case (i.e. those who commit crime) with a focus on the amelioration of risk in order to reduce the likelihood of offending. Given the consequential and disproportionate nature of criminal justice contact, the near-exclusive focus on risk to understand the pathways that lead to involvement in crime is unsurprising. This framework, however, directly adopts or implies a deficit model of development, which assumes that identifying and reducing risk factors will result in positive outcomes, a particularly common framework applied to communities of colour (Resnick 2000). Underlying this strategy are the concepts of risk and vulnerability where vulnerability is the ‘increased likelihood of a negative outcome, typically as a result of exposure to risk’ and is contrasted with protection and resilience where ‘[r]esilience refers to the avoiding of problems associated with being vulnerable’ as a result of exposure to protection (Fergus and Zimmerman 2005: 400). Applying the concepts of risk and vulnerability to this study, the assumption is that the likelihood of arrest is lowest among those at low risk (Figure 1, Group 1A) and highest among those at high risk (Figure 1, Group 2B). Common risk factors for offending and violence stem from several domains, such as one’s individual traits (e.g. aggression and risk taking), family (e.g. parental criminality and residential mobility), school and peers (e.g. academic performance and sibling and peer delinquency) and community (e.g. economic deprivation; Hawkins et al. 1998; Farrington 2000).

Fig. 1.

Categorization of positive and negative cases based on level of risk and presence of outcome.

Yet, the relationship between risk and outcome is not perfectly linear. Despite the knowledge of one’s level of risk, ‘negative cases’¹ can occur in two ways: youth at high risk evade contact with the criminal justice system (Figure 1, Group 2A) and youth at low risk become ensnared in the criminal justice system (Figure 1, Group 1B). Comparatively less is known about these negative cases whose experiences differ from the empirically or theoretically derived expected outcome, yet these cases provide an important point of departure for testing, refining and broadening theories by highlighting those ‘whose patterns of responses do not fit neatly with our hypotheses’ (Giordano 1989: 261; see also Emigh 1997; Sullivan 2011).

While the infusion and refinement of negative case analysis remains on the periphery of the field of criminology, important examples show the utility of this form of analysis (e.g. Furstenberg et al. 1998; Laub and Sampson 1998). For instance, more than six decades ago, Reckless et al. (1957) argued for the importance of understanding ‘good boys in high delinquency places’. In short, the authors argued that systematic inquiry into cases that challenge expected patterns of behaviour (i.e. high-risk behaviour among those residing in high delinquency places) offers the opportunity to advance our understanding of the complex processes inherent in criminological hypotheses and theory and, consequently, allows for the opportunity to enrich explanation. For Reckless et al. (1957: 18), this focused inquiry on boys who managed to ‘steer a course away from delinquent behavior’ despite exposure to traditional correlates of delinquency provided an opportunity to identify factors that insulated boys from criminal justice contact. Grounding this study as a negative case analysis not only guides us in our examination of the divergence in who is arrested among those at high risk but also directs us to draw on the vast resilience literature, which provides a clear example of how focusing on the negative case broadens scientific inquiry beyond risk and vulnerability.

Importance of Protection

In contrast to the deficit model, ‘the resilience paradigm seeks to identify protective, nurturing factors in the lives of those who would otherwise be expected to be characterized by a variety of adverse outcomes’ (Resnick 2000: 159) and focuses on the need to foster strengths for encouraging healthy development rather than the mitigation of deficits. Findings from resiliency research emphasize that both individual assets (e.g. competence and motivation) and external resources (e.g. social support and parental capital) can maximize resilience and positive development in adolescents and young adults despite risk factors (see e.g. Rutter 1987; Jessor 1993; Masten 2014).

Although resilience has been studied extensively in several disciplines, the explicit investigation of uniquely defined protective factors within high-risk samples, conceptualized here as factors that ‘predict a low probability of [arrest] among persons exposed to high risk’ (Farrington et al. 2012: 47), has been more limited within criminology (see Loeber et al. 2008). Key protective factors have been identified from contemporary criminological longitudinal studies, such as being shy or withdrawn, having parents interested in a child’s education, high IQ and high educational attainment (e.g. Ensminger et al. 1983; Farrington et al. 2016; see Craig et al. 2017 for a review). Traditionally, criminologists have conceptualized protection along a risk continuum where one end captures risk and the other end captures protection. For instance, low parental supervision is risky, whereas high parental supervision is protective. The resiliency paradigm reveals that the nature of risk and protection is more nuanced. While vulnerability (risk) and resilience (protection) may be opposite poles of the same concept (Stouthamer-Loeber et al. 1993), the relationship is not always linear (Rutter 1987, Fergus and Zimmerman 2005) and the absence of risk does not always imply protection or vice versa (Luthar et al. 2000; Farrington et al. 2016). This level of consideration into the nuanced nature of risk and protection remains on the periphery in criminological research. Thus, we draw on the individual, family, and school domains to not only identify protection as the absence of risk but also as the presence of unique protection.

Importance of Criminal Arrest

Prior criminological work is also limited in that it has focused on factors that protect individuals from offending behaviour leaving little attention directed at another significant part of the story—the non-trivial proportion of similarly high-risk individuals who may (or may not) offend but who avoid criminal justice contact. In light of the mounting evidence linking seemingly minor contact with the criminal justice system as a particularly potent event in the negative trajectory from childhood risk to detrimental life course outcomes, it is critical to know more about the factors that protect against contact with the criminal justice system. This need is even more pressing for high-risk individuals who experience disproportionate, but not determinant, levels of criminal justice contact. By doing so, research can shed insight into within-group heterogeneity and the value gained by examining negative cases.

Current Study

The current research, which is rooted in the conceptualization of negative case analysis, begins by identifying individuals deemed to be at high risk. Specifically, we begin with a high-risk cohort of Black men who grew up in neighbourhood disadvantage in the United States. We further identify those who had additional structural risks during childhood in order to focus on the ‘most-at-risk’ for contact with the criminal justice system among this cohort. Among these men, we identify the prevalence of negative cases by comparing high-risk individuals who experience arrest (i.e. positive cases; Figure 1, Group 2B) with those who do not experience arrest (i.e. negative cases; Figure 1, Group 2A). Next, we explore the characteristics of these negative cases by examining individual, familial and contextual factors in childhood and adolescence that represent the absence of risk, as well as have the potential to be uniquely protective against criminal justice involvement in adulthood.

Data and Methods

Sample

This study draws from a cohort of 606 Black male and 636 Black female first graders residing in Woodlawn, a disadvantaged neighbourhood in Chicago, IL. At the start of the study in 1966, when the children were 6 years old, Woodlawn was overcrowded with 90,000 people living in an urban area built to house 45,000.² By the 1970s, when the cohort was around 15 years old, the population of Woodlawn had decreased to 54,000 people and had the eighth highest percentage of families living below the poverty line (27 per cent as compared with 12 per cent in Chicago). Moreover, at that time, Woodlawn had the highest rate of male juvenile delinquency among the 76 community areas of Chicago (Council for Community Services 1975).

In 1966–67, all first graders in Woodlawn were recruited to participate in the study (n = 1,242). Only 13 families declined. First grade teachers reported on the child’s social adaptational status to school, mothers reported on the family social and economic resources, family childrearing practices and ratings of psychological and behavioural aspects of the child (Kellam et al. 1975). In adolescence, mothers and adolescents still in the Chicago area were followed. Mothers again provided extensive information on the family and on the teenagers (n = 939). The teenagers self-reported on family relationships, prosocial and antisocial behaviours and psychological feelings (n = 705; Ensminger and Slusarick 1992).³ The original cohort was followed at ages 32 (n = 952) and 42 (n = 833) with 1,053 men and women completing at least one of the adult assessments (85 per cent of the original cohort) when they were asked about their families, community involvements, drug and criminal behaviour and physical and mental health. The interview data are supplemented by arrest records spanning ages 17–52 (Doherty and Ensminger 2014).

In this study, we identify a subset of men who, on the basis of several childhood structural risk indicators, can be considered at ‘highest risk’ for criminal justice contact among an already at-risk sample. We identified this subsample by creating a summated index of eight dichotomous mother-reported measures assessed in first grade tapping into childhood structural risk: (1) childhood poverty,⁴ (2) mother alone household, (3) low maternal education,⁵ (4) household overcrowding, (5) residential instability, (6) poor supervision, (7) adolescent mother and (8) poor maternal mental health (see Table A1). Among those with at least six valid indicators, we selected men with three or more risk factors on the index, which resulted in selecting the top 50 per cent (n = 233, 52.1 per cent of 447 men with at least six indicators). The 50 per cent cut-off delineates men at ‘highest risk’ while retaining a reasonable sample size. Five men were removed due to conflicting information (e.g. self-reported jail time at age 32 but had no arrest record). Thus, the final analytic sample is 228 high-risk men defined on the basis of race, gender, early neighbourhood disadvantage and childhood structural risk.

Measures

This study utilizes the first grade, adolescent, young adult (age 32) and mid-adult (age 42) self-report data, as well as official criminal history data collected from the Chicago Police Department and Federal Bureau of Investigation (for ages 17–32) in 1993 and from the Illinois Criminal Justice Authority (17–52) in 2012.

Arrest groups

Arrest is measured as the number of official arrest counts incurred between ages 17 and 52 for violent (e.g. homicide, assault, rape and robbery) and non-violent offenses (e.g. burglary, fraud, drug possession and public order).⁶ The men were then categorized into (1) the ‘no involvement’ group, which includes men with no arrests between the ages of 17 and 52 (n = 49); (2) the ‘minor involvement’ group, which includes men with one to four non-violent offenses (n = 40) and (3) the ‘serious involvement’ group, which includes all men arrested for a violent offense and those with five or more non-violent arrests (n = 139; see Table 1).

Table 1.

Descriptive data on the categorization of high-risk men based on seriousness and frequency of arrest and deviance

	Official arrests (ages 17–52)	Self-reported offending (ages 17–32)	Self-reported substance use (ages 17–32)
	n (%)	n (%)	n (%)
No involvement	49 (21.5)	50 (21.9)	86 (37.7)
Minor involvement	40 (17.5)	30 (13.1)	72 (31.6)
Serious involvement	139 (61.0)	148 (64.9)	70 (30.7)
Total	228 (100)	228 (100)	228 (100)

Given the fact that criminal arrests occur later into the life course for Black men (Doherty and Ensminger 2014), the inclusion of arrest information to age 52 ensures that the ‘no involvement’ group truly has no arrests. For the minor involvement group, close to three-quarters of the group (72.5 per cent) was arrested by age 32 with an average of 1.88 offenses, all of which were non-violent by definition. For the serious involvement group, 95.2 per cent of the group was arrested by age 32 with over three-quarters (77.7 per cent) arrested by age 22. On average, this group incurred 15.64 offenses and 4.32 violent arrests with all but 10 of the 139 men arrested at least once for violence.

Self-report groups

Criminal justice contact is closely related to the seriousness and frequency of offending behaviour; yet, recent research exposes evidence of the growing likelihood of criminal justice system contact irrespective of actual behaviour for Black men (Weaver et al. 2019). In order to isolate criminal arrest separately from one’s offending and substance use, we also categorized the 228 high-risk men into groups based on their self-reported offending and substance use in young adulthood (see Table 1).

Self-reported offending between ages 17 and 32 is drawn from a series of questions from the young adult interview tapping into violent offending (eight types, e.g. beat up someone to get money and force someone to have sex) and non-violent offending (19 types, e.g. steal something worth at least $10, carry a weapon and sell illicit drugs). The men were then categorized into one of three groups: (1) the ‘no involvement’ group, which includes men with no self-reported offending between the ages of 17 and 32 (n = 50); (2) the ‘minor involvement’ group, which includes men with one to four non-violent self-reported offending types (n = 30) and (3) the ‘serious involvement’ group, which includes all men who self-reported violence and those who self-reported five or more non-violent offending types (n = 148).

For the substance use groups, we measured the severity of substance use between ages 17 and 32 based on whether the respondent reported using marijuana, cocaine or heroin (asked separately) during young adulthood. Substance using groups are (1) the ‘no involvement’ group, which includes men with no marijuana, cocaine or heroin use between the ages of 17 and 32 (n = 86); (2) the ‘minor involvement’ group, which includes men who reported marijuana use only (n = 72) and (3) the ‘serious involvement’ group, which includes all men who used cocaine and/or heroin (n = 70).

Domains of childhood and adolescent protective factors

We examine a variety of childhood and adolescent factors across individual, family, school and neighbourhood domains that have been identified as potentially protective in prior research. As resilience researchers contend (Rutter 1987; Resnick 2000; Fergus and Zimmerman 2005), risk and protection are not necessarily opposite poles of the same construct. Whenever possible, we include factors that represent unique sources of protection, as well as factors that represent the absence of known risk.

Individual factors

Shyness and low levels of hyperactivity, extraversion and neuroticism have been found to be protective against offending (Farrington et al. 2016). First grade teachers rated each child in their classroom on five aspects, including aggressiveness (e.g. fights too much, steals, lies, obstinate and resists authority), restlessness (e.g. fidgets and unable to sit still), underachievement (e.g. does not perform to abilities), immaturity (e.g. cries too much and has tantrums) and shyness (e.g. shy and timid; Kellam et al. 1975). All of these variables were measured by the Teacher’s Observation of Classroom Adaptation scale, which ranges from 0 to 3, coded adapting to severely maladapting. To conceptualize the first four factors as protective (i.e. the absence of risk), adaptive responses were coded as (1) and mild to severely maladaptive were coded as (0). Shyness, which is conceptualized as protective, was coded as (1) if the teacher reported mild to severe shyness versus acceptable levels of shyness (0).

Family factors

Many family structural risk factors are incorporated in the construction of the at-risk sample (e.g. maternal education and maternal mental health). Here, we tap into the protective factor of parental involvement in first grade, measured as the number of times in a week the child was played with or read to, ranging from less than weekly (0) to at least several times a week (1) but primarily focus on siblings as a unique source of protection. Several sibling variables are included with a particular emphasis on older or similar-aged siblings (within two years) under the assumption that the influence of siblings is particularly salient when the siblings are older or close in age (Defoe et al. 2013). To tap into siblings as protective, we draw on details from the household roster administered during the young adult interview (when respondents were aged 32 in 1993) about the adolescent household (when the respondent was 16). Using these data, we coded the number of older or similar-aged siblings in the adolescent household and the highest education level achieved by these older or similar-aged siblings. To assess the level of older or similar-aged sibling education we calculated: (1) the proportion of older or similar-aged siblings with less than a high school degree, (2) the proportion of those with a high school degree and (3) the proportion of those with more than a high school degree.⁷

School factors

One indicator of academic achievement is school readiness, defined here through the metropolitan readiness test (MRT) administered in first grade. The MRT is a standardized test used to measure a child’s readiness for school learning by scoring the child’s initial responses to the cognitive tasks of the classroom (Anastasi 1968). MRT levels of readiness are unready, low normal, average, high normal and superior. We combined high normal and superior scores as protective (1) compared with average or below (0). Each child’s academic achievements were also assessed through the child’s first grade teacher-reported reading and math grades, graded as unsatisfactory, fair, good or excellent in reading and math separately. To operationalize these measures as protective, we dichotomized these factors into excellent (1) and good, fair or unsatisfactory (0). To assess a person’s educational attainment, acknowledging the life course principle of the importance of achieving role transitions ‘on time’ (Elder 1998), we measure educational attainment as no high school (HS) degree or General Education Diploma (GED), a high school degree or GED ‘on time’ (defined as by 1979 when the majority of the cohort was 19 years old) or a high school degree or GED ‘off time’ (defined as after 1979).

Neighbourhood factors

Although all of the men lived in Woodlawn in first grade, some had moved out of Woodlawn by adolescence. In order to control for relative neighbourhood advantage/disadvantage in adolescence, we use an index developed by Ross and Mirowsky (2001), which assigns an objective score to each respondent using US census data (see also Lo 2010). This score is calculated by combining the (1) neighbourhood poverty, (2) female-headed households with children, (3) college-educated heads of households and (4) homeowners. More specifically, the score is calculated by dividing each of these four census level indicators by ten, adding the prevalence of poverty and female-headed households, subtracting the proportion of college-educated residents and home owners and dividing the result by four. The index ranges from negative to positive values where lower scores indicate more advantaged neighbourhoods (i.e. many adults with a college degree and their own homes; few households are poor or female headed) and higher scores indicate more disadvantaged neighbourhoods (i.e. few adults have a college degree and many rent rather than own their homes; many households are poor and female headed). To operationalize this measure as a protective factor, we code negative scores as (1) to represent advantaged neighbourhoods in adolescence and positive scores as (0) to represent disadvantaged neighbourhoods.

Analysis

The analysis consists of three phases. First, we compare the overlap between the self-reported offending groups, substance use groups and arrest groups to examine the risk profiles of those in each arrest group. This allows for a preliminary check on whether experiencing arrest is largely ‘explained’ by one’s self-reported deviance (i.e. whether the non-arrested group largely abstain from deviance). In the second phase of the analysis, we conduct a series of bivariate tests (e.g. t-tests, analysis of variance and chi-square) to identify the protective factors that distinguish the negative cases from the positive, or expected, cases. The third phase estimates several multinomial logistic regressions for the individual, family, school and neighbourhood factor domains on the arrest groups separately for each domain (Models 1–4) before estimating a full model with the factors that emerge as significant (Model 5). For the multivariate analyses, we control for the self-reported deviance groups to identify protective factors above and beyond a lack of deviant behaviour. For all of the multivariate analyses, we use multiple imputation to accommodate missing data (Schafer and Graham 2002). The primary variable impacted by missing data is neighbourhood advantage during adolescence.⁸

Results

Prevalence: how common are negative cases?

As described previously in Table 1, the majority of these high-risk men had serious involvement in the criminal justice system with 61 per cent arrested for a violent offense and/or incurring five or more non-violent arrest counts (i.e. the serious arrest group, n = 139). However, arrest is not a foregone conclusion as a non-trivial proportion were never arrested in all of adulthood (n = 49, 21 per cent) and 18 per cent (n = 40) were arrested only a few times for non-violent offenses. Moreover, as shown in Table 2, although adult offending and drug use behaviour are significantly related to arrest, these self-reported groupings do not perfectly overlap with the seriousness of arrest groups. In fact, while 35 per cent of the no arrest group self-reported no offending and 45 per cent self-reported no drug use, 47 per cent of the no arrest group are in the serious self-reported offending group (n = 23) and 20 per cent self-reported serious drug use. The finding that many of those in the no arrest group are not ‘abstainers’, sparks the question: What factors are related to avoiding criminal arrest among men with similar childhood structural risk that is not explained by differences in self-reported deviance?

Table 2.

Bivariate comparisons of risk and protective factors

		Arrest groups
	Total sample (n = 228)	No involvement (n = 49)	Minor involvement (n = 40)	Serious involvement (n = 139)	Test statistic	p-value
Childhood structural risk
Structural risk scale	4.26	4.12	4.30	4.30	F = 0.465	0.628
Self-reported deviance
Offending
Serious offending	64.9%	46.9%	52.5%	74.8%	X  2 = 15.858	0.003
Minor offending	13.2%	18.4%	17.5%	10.1%
No offending	21.9%	34.7%	30.0%	15.1%
Drug use
Serious use	30.7%	20.4%	17.5%	38.1%	X  2 = 9.686	0.046
Minor use	31.6%	34.7%	35.0%	29.5%
No use	37.7%	44.9%	47.5%	32.4%
Individual-level factors
Low aggressiveness	58.8%	67.3%	62.5%	54.7%	X  2 = 2.679	0.262
Low immaturity	56.6%	61.2%	52.5%	56.1%	X  2 = 0.714	0.700
Low restlessness	60.5%	69.4%	70.0%	54.7%	X  2 = 5.104	0.078
Low underachievement	48.7%	49.0%	60.0%	45.3%	X  2 = 2.680	0.262
Shy/withdrawn	37.3%	26.5%	47.5%	38.1%	X  2 = 4.251	0.119
Family-level factors
Family involvement—first grade (n = 223)	58.3%	59.6%	45.0%	61.8%	X  2 = 3.613	0.164
Number of older or same-aged siblings in adolescent household	2.27	1.88	1.73	2.56	F = 6.276	0.002
Older or similar-aged siblings’ education (n = 189)
Mean proportion with less than HS	0.276	0.131	0.232	0.333	F = 4.447	0.013
Mean proportion with HS	0.456	0.432	0.527	0.448	F = 0.505	0.604
Mean proportion with more than HS	0.266	0.437	0.232	0.220	F = 4.385	0.014
School-level factors
School readiness (n = 203)	8.4%	14.6%	8.6%	6.3%	X  2 = 2.808	0.246
First grade math—excellent (n = 210)	5.7%	6.4%	8.8%	4.7%	X  2 = 0.920	0.631
First grade reading—excellent (n = 213)	7.0%	6.3%	14.3%	5.4%	X  2 = 3.397	0.183
High school education
No high school degree/GED	29.4%	20.4%	20.0%	35.5%	X  2 = 18.589	0.001
High school degree/GED ‘on time’	54.4%	75.5%	65.0%	43.5%
High school degree/GED ‘off time’	16.2%	4.1%	15.0%	21.0%
Neighbourhood-level factors
Neighbourhood disadvantage index—adolescence (n = 111)	36.9%	44.0%	26.3%	37.3%	X  2 = 1.460	0.482

Protective factors: distinguishing negative cases from positive ‘expected’ cases

Not surprisingly, given the fixed social context among respondents in childhood, Table 2 shows no significant bivariate differences in childhood structural risk across arrest groups. Interestingly, none of the individual factors emerge as protective, yet factors from the family and school domains do distinguish the arrest groupings. From the family domain, these bivariate analyses show that having fewer siblings who are older or of similar age (within two years) is a significant protective factor for arrests and having siblings with high educational attainment is a strong protective factor. Among those with no arrests, on average, 0.131 of their older or same-aged siblings earned less than a high school diploma and 0.437 earned more than a high school diploma. A higher proportion of the older and similar-aged siblings in both the minor and serious involvement groups earned less than a high school degree (0.232 and 0.333, respectively) and a lower proportion earned more than a high school degree (0.232 and 0.220, respectively).

From the school domain, one’s own education is also a key factor with graduating ‘on time’ as opposed to ‘off time’ being protective. The vast majority of the no arrest group graduated ‘on time’ (75.5 per cent) compared with 65.0 per cent of the minor involvement group and less than half of those in the serious involvement group (43.5 per cent). In contrast, only 4.1 per cent of those with no criminal arrest graduated ‘off time’ compared with 15.0 and 21.0 per cent of the minor and seriously involved groups, respectively.

As displayed in Table 3, the multivariate analyses of the arrest groups again reveal that most of the significant findings centre on the family and school factors—namely the number of siblings, sibling education and one’s own education. Model 1 of each arrest group comparison shows only one significant difference, yet this is in the opposite than expected direction with those who are shy being less likely to be in the no arrest group compared to the minor arrest group.⁹ With respect to family factors, in Model 2 of each comparison, those with fewer older or similar-aged siblings are more likely to be in the never arrested group or the minor arrest group compared with the serious arrested group. Moreover, those with a higher proportion of older or same-aged siblings who have earned more than a high school degree are more likely to be in the never arrested group compared with the minor arrest group or serious arrested group. One’s own education emerges as significant in Model 3 of the comparison of the never arrested group compared with the serious arrested group. For the neighbourhood domain, Model 4 shows no significant protective relationship between living in neighbourhood advantage in adolescence and arrest group.

Table 3.

Coefficients and standard errrors from multinomial logit regressions of protective factors on arrest group

	No arrest group vs. serious arrest group					No arrest group vs. minor arrest group					Minor arrest group vs. serious arrest group
Model	1	2	3	4	5	1	2	3	4	5	1	2	3	4	5
Constant	−0.201 (0.518)	0.352 (0.552)	0.106 (0.403)	−0.192 (0.398)	0.868 (0.531)	1.120 (0.684)	−0.312 (0.583)	0.340 (0.459)	0.174 (0.456)	0.466 (0.583)	−1.322 (0.621)	0.664 (0.550)	−0.233 (0.430)	−0.366 (0.424)	0.402 (0.554)
Individual-level factors
Low aggressiveness	0.237 (0.440)				–	0.332 (0.557)				–	−0.095 (0.475)				–
Low immaturity	−0.149 (0.482)				–	0.269 (0.611)				–	−0.418 (0.515)				–
Low restlessness	0.584 (0.464)				–	−0.102 (0.598)				–	0.687 (0.512)				–
Low underachievement	−0.184 (0.423)				–	−1.004 (0.538)				–	0.819 (0.456)				–
Shy/withdrawn	−0.648 (0.445)				−0.464 (0.412)	−1.184 (0.557)				−0.925 (0.480)	0.537 (0.460)				0.460 (0.397)
Family-level factors
Family involvement		−0.049 (0.387)			–		0.532 (0.455)			–		−0.581 (0.390)			---
Number of siblings		−0.401   (0.154)			−0.354  (0.132)		−0.055 (0.193)			0.014 (0.164)		−0.356 (0.164)			−0.368   (0.140)
Proportion of sibs < HS		−0.752 (0.755)			–		0.019 (0.873)			–		−0.771 (0.689)			–
Proportion of sibs with HS		0.491 (0.619)			–		0.393 (0.703)			–		0.099 (0.613)			–
Proportion of sibs > HS		1.852 (0.691)			1.558   (0.534)		1.659 (0.813)			1.491   (0.682)		0.193 (0.769)			0.067 (0.646)
School-level factors
School readiness			0.881 (0.664)		–			0.521 (0.779)		–			0.360 (0.772)		–
First grade math			−0.171 (0.959)		–			0.180 (1.100)		–			−0.350 (0.988)		–
First grade reading			0.264 (0.876)		–			−0.890 (0.948)		–			1.154 (0.824)		–
High school degree
HS/GED ‘on time’			Ref		Ref			ref		Ref			Ref		Ref
No HS/GED			−0.966 (0.431)		−0.692 (0.453)			−0.168 (0.559)		0.180 (0.577)			−0.798 (0.475)		−0.872 (0.485)
HS/GED ‘off time’			−1.992 (0.436)		−1.875 (0.803)			−1.508 (0.867)		−1.298 (0.886)			−0.484 (0.533)		−0.578 (0.548)
Neighbourhood-level factors
Neighbourhood advantage				0.156 (0.433)	–				0.376 (0.533)	–				−0.220 (0.468)	–

Bold values = p < 0.05. HS = High School, GED = General Education Diploma.

Finally, Model 5 reports the results of the full model. From this full model, graduating ‘on time’ is protective compared with graduating ‘off time’ as is sibling’s education. Figure 2 displays the relative risk ratios for the full models (Model 5) of Table 3, which are interpreted as increasing one’s risk of being in the never arrested group when the value is greater than 1 and reducing the risk of no arrest when the value is less than 1. In Figure 2, the relative risk ratio of being in the never arrested group compared to the serious arrested group is expected to increase by 6.52 for those who graduated ‘on time’ compared with those who graduated ‘off time’.¹⁰ Moreover, Table 3 and Figure 2 show that, while having more older or similar-aged siblings is a risk factor, sibling educational success is a strong protective factor; for those with a higher proportion of siblings with a high school degree, the relative risk increases by a factor of 4.75 (exp^(1.558)) for being in the never arrested group compared to the serious arrested group and by a factor of 4.44 (exp^(1.491)) compared to the minor arrested group.

Fig. 2.

Relative risk ratios from multivariate final models. HS = High School, GED = General Education Diploma.

Discussion

The saying ‘give me a child until he is seven and I will give you the man’ exemplifies the notion that childhood risk dictates one’s future; yet, despite significant early life course adversity, many ‘manage to make it’ (Furstenberg et al. 1998). In this study, we employed a negative case analysis to examine the divergence in histories of criminal justice contact, beyond self-reported offending and substance use, among a sample of high-risk Black men raised in a socially disadvantaged neighbourhood in the United States. Our findings suggest that avoiding arrest is a common outcome and that exposure to protective factors may best account for divergent trajectories among those at greatest risk. Findings from this study point to several methodological and theoretical implications that can guide future research.

Negative cases and protective factors

First, this study highlights the value of negative case analysis as a mechanism to explore unanticipated patterns in data. The widespread focus on the identification and amelioration of risk factors in order to reduce the likelihood of negative outcomes has resulted in limited progress on understanding how high-risk youth avoid contact with the criminal justice system despite the odds. Similar to the wealth of evidence that there is divergence in health and social outcomes among those at high risk (e.g. Resnick et al. 1997; Furstenberg et al. 1998), in this study, we find that not only do negative cases exist but also that they comprise a sizeable and consequential group, whereby nearly a quarter of the most structurally disadvantaged young Black men in this cohort avoided arrest across their life course. Counter to the traditional probabilistic orientation predicting arrest, the finding that many at greatest risk manage to avoid arrest confirms that these men are not merely statistical noise and redirects attention to individual and situational contingencies that impede arrest (see also Sullivan 2011).

Second, investigation into these negative cases revealed protective factors unique from the absence of risk factors (e.g. Rutter 1987; Luthar et al. 2000; Fergus and Zimmerman 2005), highlighting the complexity of the risk and protection relationship. For instance, whereas aggression is a common risk factor for crime and criminal arrest, in these data, the absence of aggression demonstrates no protective effect. Instead, educational factors emerge as paramount in protecting youth from criminal justice contact throughout the life course, beyond a host of individual, family, school and neighbourhood factors and involvement in general risky behaviours. The pattern of findings regarding protective factors is consistent with prior research revealing the complex nature of risk and protection in predicting outcomes. Whereas one’s own education may represent opposite poles of risk and protection (i.e. graduating from high school ‘on time’ represents a protective factor and graduating ‘off time’ represents a risk factor), the influence of sibling education appears to be one sided, whereby sibling advanced educational attainment is protective in nature, but sibling low educational attainment does not represent a risk. Similar to the life course recognition that factors responsible for the onset of and desistance from offending may be asymmetrical (Uggen and Piliavin 1998; see also Farrall et al. 2014), theoretical development on avoidance of criminal offending and criminal justice contact could benefit from bringing the potential asymmetry of relationships between risk, vulnerability, protection and resilience to the forefront (see Rutter 1987; Fergus and Zimmerman 2005).

While previous work has signalled education as a key conduit to success, this work largely conceptualizes education as a control variable. In line with recent work highlighting the dynamic and central role of education in the development of offending (Kennedy-Turner et al. 2020), we utilize a life course framework (Elder 1998) to capture features of education to illuminate processes that may be driving its relationship with criminal justice system avoidance. Incorporating the principle of timing that suggests that the same event, experienced at different times/ages, may yield different effects, we distinguished between ‘on-time’ and ‘off-time’ education. We found that one’s own ‘on-time’ educational progression is protective with the majority of the no arrest group completing high school ‘on time’. Completing school on time may set the stage for successful life transitions clustered in young adulthood, such as enrolment in college and attaining employment. We also incorporated the principle of linked lives, which underscores that lives are lived interdependently. Whereas prior research has focused on the educational attainment of focal respondents and/or their parents (see e.g. Farrington et al. 2012), given that the majority of children under 18 in the United States (77 per cent in 1995) lived with at least one sibling under the age of 25 while only 56 per cent lived with both biological parents (Hernandez 1997), we expanded the sphere of influence beyond parents (see Cleveland et al. 2012; Bersani and Doherty 2018) to include siblings. Among those with same-aged or older siblings, nearly half of the no arrest group had siblings with advanced education. While extant research has focused on the risk of sibling deviance on the development of offending (e.g. Lauritsen 1993; Huijsmans et al. 2019), future research should also examine the positive protections garnered from siblings (see McHale et al. 2012).

Potential protective mechanisms

At the core of resilience research is the importance of shifting from identification of protective factors to understanding the underlying mechanisms associated with the protective factor. While the current research contributes to the literature on factors that distinguish divergent trajectories among a sample at high risk of arrest, we can only speculate on the mechanisms by which siblings and education relate to positive outcomes among those at highest risk. Two potential and interrelated frameworks for situating and explicating the findings are cumulative advantage and capital. First, the protective nature of one’s own ‘on-time’ and one’s siblings’ education could facilitate or contribute to the process of cumulative advantage (O’Rand 1996; Dannefer 2003) leading to improved chances for residential and social mobility. Among those exposed to multiple disadvantages, incremental educational advantages may accumulate over time, redirecting trajectories away from disadvantage. Moreover, sibling education may produce protective benefits through the diffusion of resources and the shaping of an individual’s expectations of themselves that add to one’s accumulation of advantage. Post hoc analyses of the men in this sample indicate that those who were in the never arrested group were significantly more likely to own their own home or apartment at age 32 than the minor arrested or seriously arrested groups (35, 18 and 11 per cent, respectively), more likely to live outside of the Chicago Metro area (31, 20 and 9 per cent, respectively) and less likely to be living in poverty at this age (30, 41 and 52 per cent, respectively). The benefits of the accumulation of advantage may be most apparent in the most disadvantaged contexts where divergence is stark. That is, for the most disadvantaged men, even minor improvements in one’s social circumstances may serve to redirect one’s life course for continued accumulations of relative advantage.

A related framework to situate our results and examine how siblings and education initiate cumulative advantage may be understood through the concept of capital. Capital has been conceptualized as a resource that translates into expectations and opportunities that can influence action, both prosocial (Coleman 1988) and criminal (McCarthy and Hagan 1995) in nature. Our findings suggest a transitive property of capital—what we refer to as ‘collateral capital’—whereby sibling educational competencies may shape others’ expectations of one’s potential, independent of one’s own achievements. For instance, examining racial variation in the likelihood of police contact, Crutchfield et al. (2012) draw upon Anderson’s (2002) ethnographic descriptions of youth from ‘street’ families and find that Black youth are more likely to have police contacts if they had a parent who had been arrested or a sibling involved in crime as it signalled a ‘risk-producing element’. Thus, police contact stemmed not only from one’s actual criminal behaviour but also from officers’ expectations of behaviour based on the knowledge of who belongs to which ‘type’ of family. This rationale corresponds with evidence that teachers’ knowledge of a student’s parental incarceration shapes the teacher’s behavioural and competency expectations for that student (Dallaire et al. 2010; Wildeman et al. 2017).

Using this same logic, we ask might the transference of expectations occur among ‘decent’ families (Anderson 2002), ones viewed as valuing education and respecting authority? We posit that knowledge of one’s membership in an educationally oriented family may function to reduce police contact by signalling a ‘risk-reducing element’. In this vein, similar-aged sibling achievements may provide a source of collateral capital, whereby authorities’ (e.g. police and teachers) perceptions of one’s siblings shape the expectations of behaviour and potential of the individual, beyond a person’s own actual behaviour. Though the data does not capture police perceptions, our finding that the no arrest group were not crime abstainers yields preliminary support for this assertion. Future research should directly examine this notion of collateral capital and the potential of sibling achievement as a risk-reducing element and potential contributor to cumulative advantage to further our understanding of how those at high risk avoid arrest despite the odds.

Conclusions

Our findings support the assertion that the tendency of criminological research to neglect negative cases introduces the concern that ‘the field might be overlooking or discarding information that is potentially useful in explaining crime’ (Sullivan 2011: 916). Negative cases should be viewed as more than statistical outliers or anomalies in the data as the wide adoption of the negative case framework could help move the field towards a greater understanding of the contingencies and conditions inherent in offending and desistance (see Bersani and Doherty 2018). Whereas the focus of this research was on the prevalence of high-risk/favourable outcome ‘negative cases’ in one high-risk cohort who came of age in the late 1970s and 1980s from one city in the United States, additional research on a variety of high-risk groups, broadly defined, from contemporary and/or international samples could speak to the widespread prevalence of this type of negative case and to the variability of protective processes over time, location and groups. While we expect that general theoretical processes underlying cumulative advantage and collateral capital are not bound by socio-historical context, the specific ways it takes shape may vary.

As Jessor (1993: 121) notes, a ‘social policy agenda should be concerned not only with the reduction of risk but with the strengthening of protection as well’. Our findings suggest the potential value in shifting the lens away from a widespread reliance on a deficit model of development to incorporate a strengths-based approach to better understand the process of avoiding the negative effects of risk exposure (Fergus and Zimmerman 2005) and inform strengths-based interventions aimed at fostering pathways towards successful outcomes.

Funding

This work was supported in part by the National Institutes of Health (R01 DA042748 to Kerry Green) and the Harry Frank Guggenheim Foundation (to E.E.D).

Appendix

Table A1.

Childhood structural characteristics

Childhood poverty	Family household income in 1967 was at or fell below the poverty threshold for the household size (coded 1) or not (coded 0)
Mother alone	A dichotomous variable of whether a child was living in a ‘mother alone’ household, meaning no other adult present, (coded 1) or not (coded 0)
Low maternal education	Respondent’s mother completed less than 12 years of education (coded 1) or completed 12 or more years of education (coded 0)
Household overcrowding	Based on the US Department of Housing and Urban Development’s (2014) definition of severe overcrowding, respondents living in a household with more than 1.5 people per room (coded 1) versus those living in a household with fewer than 1.5 people per room (coded 0)
Residential instability	Respondent moved four or more times between birth and first grade (coded 1) and those with three or fewer moves between birth and first grade (coded 0)
Poor supervision	The ratio of number of children in the household to the number of adults, dichotomized at the top quartile, such that more than three kids per adult in the household indicates poor supervision (coded 1) as opposed to three kids or fewer per adult (coded 0)
Adolescent mother	Mother was younger than 18 years at her first birth (coded 1) or mother was 18 or older at her first birth (coded 0)
Poor maternal mental health	Mean scale of two measures indicating how often the mother feels sad or blue and how often she feels nervous or tense—both ranked from 1 (hardly ever) to 4 (very often), dichotomized to identify the top quartile as experiencing poor maternal mental health