Introduction

Violent victimization—particularly when it happens to young people—can inflict a wide array of negative personal and social consequences (Agnew 2006; Jennings et al. 2012; Turanovic and Pratt 2015). One of the more severe consequences is that being victimized once increases the likelihood of being victimized again (Farrell 1995; Lauritsen and Quinet 1995; Ruback et al. 2014). And while the victimization-repeat victimization link has been clearly established in the criminological literature (Farrell and Pease 2014; Menard 2000; Ybarra and Lohr 2002), what is far less clear is why this relationship exists. Common explanations have traditionally specified either risk heterogeneity or state dependent processes, where a victimization event either “flags” an enduring risk or changes the victim in such a way that “boosts” the odds of repeat victimization (Fisher et al. 2010; Johnson et al. 2009; Tseloni and Pease 2003). Yet despite their popularity in situational crime prevention circles (Hirschfield et al. 2010; Johnson and Bowers 2004), these perspectives have received, at best, mixed empirical support in the broader criminological literature (Osborn et al. 1996; Ousey et al. 2008; Wittebrood and Nieuwbeerta 2000).

Perhaps a more promising explanation may come from the body of research indicating that victimization and repeat victimization can often be the consequence of engaging in certain “risky lifestyles,” such as frequently participating in various forms of crime and analogous behaviors, that tend to bring people into close proximity with potential offenders (Berg et al. 2012; Lauritsen et al. 1991; Posick and Zimmerman 2015). Of course, victimization does not require engaging in such risky behaviors (see, e.g., Cohen and Felson 1979; Felson and Boba 2010), yet doing so certainly increases the probability of it happening (Cops and Pleysier 2014; Schreck et al. 2002; Turanovic et al. 2015). At the same time, recent work suggests that victims who make changes to their behavioral routines (i.e., stop doing risky things) are better able to avoid being victimized again in the future (Turanovic and Pratt 2014). Thus, changes in risky lifestyles are fundamentally linked with repeat victimization.

The problem, however, is that one’s ability to make changes to such behavioral routines after being victimized could be limited by what Hindelang et al. (1978) referred to as “structural constraints.” To be sure, those who reside in impoverished communities face structural conditions that narrow the range of economic and social options that are available to them (Harding 2010; Sampson et al. 2002; Wilson 1987). In addition, scholars have long noted that cultural responses to structural disadvantage can also place constraints on the kinds of behaviors that community residents feel are acceptable to engage in (Anderson 1999; Clark 1965; Matza 1964). These kinds of criminogenic structural conditions (e.g., economic deprivation/concentrated disadvantage) could therefore get in the way of the ability of some victims to make the kinds of behavioral changes that would be helpful in reducing the risk of repeat victimization.

Accordingly, in the present study we examine two interrelated research questions: (1) do those who reside in communities characterized by structural constraints (measured as concentrated disadvantage) continue to engage in risky behavioral routines after being victimized? And (2) does continuing to do so increase the likelihood of repeat victimization? To address these questions, we use ten waves of data (spanning nearly 7 years) from the Pathways to Desistance study, a two-site longitudinal study of serious adolescent offenders (Mulvey et al. 2014). By examining these questions with this unique data source, our broader purpose is to gain a better understanding of the social ecology of risky lifestyles and repeat victimization.

Structural Disadvantage and the Context of Victimization

Life in disadvantaged communities is often fraught with hardship. Dating back nearly 100 years, research indicates that impoverished urban neighborhoods are places where residents suffer many health problems, including disproportionately high rates of mental illness, physical abuse, infant mortality, and poor nutrition (Bradley and Corwyn 2002; Shaw and McKay 1942; Sampson 2012); where people live in dilapidated housing, face bleak employment prospects, receive substandard schooling, and experience residential segregation (Kozol 2006; Peterson and Krivo 2010; McInerney and Smyth 2014); and where discrimination, hopelessness, and violence are routine features of daily life (Anderson 1999; Bolland 2003; Wilson 2009). Importantly, these are also areas where residents face particularly high risks of being violently victimized—often more than once.

Violent victimization is high in disadvantaged neighborhoods in no small part because of the lifestyle patterns and routine activities that community residents take part in (Cohen and Felson 1979; Lauritsen 2001; Miethe and McDowall 1993). Indeed, variations in the lifestyles and routine activities that people engage in carry different levels of risk for putting them in particular places at particular times, and of coming into contact with particular people (Hindelang et al. 1978; Garofalo 1987). Due to economic strains, subcultural norms, and the weakening of formal and informal social controls, residents of disadvantaged communities will also face greater pressures to engage in certain lifestyles that are “risky,” in that they expose individuals to situations conducive to crime and violence. These situations increase contact with potential offenders, which, in turn, increases the likelihood of victimization (Berg and Loeber 2011; Sampson et al. 1997; Wolfgang and Ferracuti 1967).

Of course, risky lifestyles come in various forms (e.g., Mustaine and Tewksbury 1998; Jensen and Brownfield 1986), but the ones most closely linked to victimization are those that involve crime, violence, and substance use (Pratt and Turanovic 2016; Schreck and Stewart 2011). For example, individuals who engage in crime may put themselves at risk for retaliation from former victims (Jacobs and Wright 2006; Singer 1986; Stewart et al. 2006), and may be more likely to come into contact with violent persons while they commit offenses or use drugs and alcohol during high-risk times (e.g., after dark) and in high-risk settings (e.g., in the absence of capable guardianship). It is therefore not surprising that prior work has found various forms of crime and delinquency to be some of the strongest correlates of victimization, independent of factors such as gang membership, peer associations, leisure activities, and demographic characteristics (e.g., age, sex, and race) (Haynie and Piquero 2006; Pauwels and Svensson 2011; Peterson et al. 2004; Turanovic et al. 2015). These risky lifestyles have also been closely linked to repeat victimization. For instance, Schreck et al. (2006) found that youth who engaged in delinquency (e.g., violent crime, property crime, and drug use) after being victimized were more likely to be victimized again, and Ruback et al. (2014) found that violent offending largely accounted for the relationship between victimization and repeat victimization among adolescents.

It makes sense, then, that victims of violence who make changes to their risky lifestyles can reduce their likelihood of being victimized again in the future—what Hindelang et al. (1978: 129) referred to as the “once bitten, twice shy” hypothesis (see also Cook 1986; Liska and Warner 1991). Although the full body of research is somewhat mixed with respect to this proposition (e.g., Averdijk 2011; Bunch et al. 2014; Ousey et al. 2008), there is some key evidence to support it. Miethe et al. (1990), for example, examined individual lifestyle patterns over two time points and found that the odds of repeat victimization were highest among individuals who continually engaged in risky activities. They concluded that victimization was explained largely by “changes and stability in lifestyles over time” (Miethe et al. 1990: 367) and that victims who frequently engaged in nighttime activities outside of the household (their indicator of risky lifestyles) were most likely to be victimized repeatedly. More recently, using two waves of panel data, Turanovic and Pratt (2014) found that adolescents who failed to make changes to their risky lifestyles after being victimized (e.g., by continuing to engage in risky socializing, substance use, violent offending, and hanging out with violent friends) were those most likely to experience repeat victimization.

Collectively, these studies tell us that making changes to risky lifestyles is an important determinant of repeat victimization. The problem, however, is that not all victims of violence are able to make meaningful changes to their risky lifestyles—especially to their involvement in antisocial behavior. And while prior research has focused on the individual-level factors that shape whether these changes are made (e.g., low self-control; Turanovic and Pratt 2014), there are likely key ecological and structural factors that play an important role in this process. Within communities characterized by severe economic deprivation, victims may not be able to make the kinds of changes to their risky lifestyles that might be necessary to reduce the likelihood of repeat victimization.

Linking Structural Constraints to Changes in Risky Lifestyles and Repeat Victimization

Hindelang et al. (1978) were among the first to put forth the idea of “structural constraints” in relation to personal victimization. They theorized that lifestyle patterns manifest as individual- and group-level adaptations to role expectations and to various aspects of the social structure. People learn attitudes and behaviors in response to their social environment and, once learned, they are incorporated into their routine activities. To be sure, Hindelang et al. (1978: 242) stated the following with respect to structural constraints:

Structural constraints originating from [the social structure] can be defined as limitations on behavioral options that result from the particular arrangements existing within various institutional orders, such as the economic, familial, educational, and legal orders. For example, economic factors impose stringent limitations on the range of choices that individuals have with respect to such fundamentals as area of residence, nature of leisure activities, mode of transportation, and access to educational opportunities.

Accordingly, within economically deprived communities, there are structural constraints that shape daily life in important ways. These constraints can encourage criminal attitudes and beliefs, and they can limit the extent to which individuals are able to avoid coming into contact with risky people and risky settings (Bjarnason et al. 1999). For victims of crime, structural constraints can also shape responses to victimization by influencing the degree to which victims are able (or even willing) to adopt the kinds of protective behaviors that might reduce their exposure to potential offenders. And due to the nature of these structural constraints, it is likely that they restrict the ability of victims to alter their deviant lifestyles in ways that reduce the risk of repeat victimization.

More specifically, structural constraints in disadvantaged communities can limit victims’ ability to make changes to their risky lifestyles in two ways. First, there is an objective component to these structural constraints in that victims who reside in such contexts often do not have the power to change where they live, or where they go to school—factors that alone may be the source of victimization risk for some people (Carson et al. 2013; Foster and Brooks-Gunn 2013; Schwartz et al. 2013). Furthermore, structural constraints on economic opportunities, coupled with poor access to public transportation, may limit both the legitimate and illegal avenues of employment that are available to people in the area (Bellair and Kowalski 2011; Sullivan 1989). Dealing drugs, stealing things, and selling stolen property, for example, may be some of the most accessible ways to generate income in economically deprived neighborhoods (Rose and Clear 1998). Thus, choices among behavioral alternatives are constrained by their availability.

Second, for those who reside in economically deprived communities there is a perceptual component to these structural constraints as well. In places where a “code of the street” value system prevails, there are intense social pressures on community residents to behave in certain ways, such as getting even with someone who you think has wronged you in order to avoid any loss of respect (Anderson 1999; Lindegaard et al. 2013; Stewart and Simons 2010). Residents will engage in these behaviors because they perceive that they do not have much of a choice in the matter. And yet behaving in such a way, which is a response to structurally induced and culturally proscribed social processes (Wilson 2009), elevates the odds of being victimized (Berg et al. 2012; Stewart et al. 2006). Berg et al. (2012), for example, found that the relationship between victimization and offending is more pronounced in neighborhoods where street culture predominates. We therefore have good reason to believe that it might be difficult for some victims to make certain lifestyle changes to avoid being victimized again, especially when their autonomy to do so is constrained by their social context. Nevertheless, this remains an open, yet certainly important, empirical question.

Current Focus

In light of these issues, in the current study we examine two interrelated research questions. First, are those who reside in communities characterized by concentrated disadvantage more likely to continue to engage in risky lifestyles (e.g., offending, drug use, and alcohol use) after being victimized? And second, does continuing to do so increase the likelihood of being victimized again? Answering these questions is a critical step toward identifying victims that are most vulnerable to experiencing repeat victimization. In carrying out this study, our broader purpose is to reach a better understanding of how the consequences of victimization are structurally embedded.

In addition, we carry out our work using data from a sample of serious adolescent offenders who have extensive experience with antisocial behavior as well as victimization—much more so than general population samples. This is critical since investigating the factors that may be related to the linkages between risky lifestyles and victimization can help inform the development of programs to intervene (and prior to that, prevent) these adverse experiences from occurring in the first place and persisting thereafter. On a larger level, understanding the factors that are related to changes in antisocial behaviors among a sample of serious youthful offenders is consistent with Laub and Sampson’s (2001) argument that such a sample, much like their Glueck sample, is of utmost relevance for policy discussions.

Methods

Data

The current study is based on a subset of data from Pathways to Desistance (Mulvey et al. 2014), which is a multi-site 7-year longitudinal study of serious juvenile offenders. Between November 2000 and January 2003, 1354 adjudicated youths from the juvenile and adult court systems in Phoenix, AZ (n = 654) and Philadelphia, PA (n = 700) were enrolled in the Pathways Study. All enrolled youth had been found guilty of a serious offense (predominantly felonies, with a few exceptions for misdemeanor property offenses, sexual assaults, and weapons offenses) and were between the ages of 14 and 17 at the time of their offense. The study participants completed follow-up interviews in 6-month intervals for the first 3 years of the study, and then in 12-month intervals for the next 4 years. Cumulative retention rates in the Pathways Study were high, and approximately 86% of participants completed at least 8 of the 10 follow-up interviews. Further details on the Pathways Study, including participant enrollment, study design, and sample characteristics, can be found in Schubert et al. (2004) and Mulvey et al. (2004, 2014).Footnote 1

The Pathways Study was designed to describe the role of social context in promoting either desistance or continuation of antisocial behavior (Schubert et al. 2004), and these data are well suited to our research questions for several reasons. First, since the youth in the Pathways data are serious offenders, they are more likely than members of the general population to reside in structurally disadvantaged communities and to engage in the kinds of risky behaviors that increase their odds of victimization (Gottfredson 1981). Indeed, youthful offenders are among those most likely to be victims of street violence (Jennings et al. 2012; Sickmund and Puzzanchera 2014; Stewart et al. 2006), and as mentioned previously, it is critical for prevention and intervention strategies that we reach a better understanding of the processes that lead to repeat victimization among this population. Second, the data span nearly 7 years, meaning a good portion of youth are transitioning out of adolescence and are entering a stage of the life course when rates of offending should start to decrease (Sweeten et al. 2013a, b). Accordingly, the kinds of changes to risky behaviors that we are interested in should start to take place within this age range (youth are between 14 and 19 years old in wave 1, and between 21 and 26 years old in wave 10). And third, the short, 6- to 12-months time periods between waves of data, and the use of consistent measures, lend themselves well to our research objectives since we wish to examine changes to risky lifestyles and repeat victimization over time.

During each interview, beginning with wave 1, youth were asked to report the community locations where they had lived during the recall period (i.e., in the past 6 or 12 months), and their main location represented the place where they had lived the longest during that time. The address for this main location was then geocoded by Pathways investigators and linked to block group data from the 2000 decennial Census. In general, block groups (i.e., statistical divisions of census tracts) have between 600 and 3000 residents, and are designed to be as homogenous as possible with regard to population characteristics, living conditions, and economic status (U.S. Bureau of the Census 2000). Address information was not recorded for youth who were housed in correctional facilities, residential treatment centers, or homeless shelters/group homes during each interview. The data contain an average of 1.1 respondents per block group.

Given that our research questions focus exclusively on victims, our study sample consists of 217 youth who reported violent victimization at wave 1 and who completed at least one follow-up interview in the community.Footnote 2 Over the course of 10 waves, respondents in the victim subsample were interviewed an average of 9.1 times, for a total of 1943 person-periods.

Measures

Repeat Victimization

Repeat victimization is our key dependent variable, captured at waves 2 through 10 of the data. This outcome is measured as a variety score (possible range 0–6) that reflects whether each respondent was a victim of the following violent acts during the recall period (i.e., since the date of last interview): “you were shot,” “you were shot at,” “you were attacked with a weapon, such as a knife or bat,” “you were chased when you thought that you could get really hurt,” “you were hit, slapped, punched, or beaten up,” and “you were sexually assaulted, molested, or raped.” These six items originate from the Exposure to Violence Inventory (ETV; Selner-O’Hagan et al. 1998), which was modified for the Pathways Study. Since all participants in the sample reported violent victimization at wave 1, subsequent violent victimization over the course of data collection (from waves 2 to 10) reflects “repeat victimization.” Most victims in the sample (72.4%, n = 157) experienced repeat victimization, where 47.5% of victims (n = 103) reported multiple repeat victimizations (i.e., at two or more waves of data), and 24.9% of victims (n = 54) reported repeat victimization only once (i.e., at one wave of data). Patterns of repeat victimization for the study sample across all waves of data are presented in Table 1.

Table 1 Longitudinal patterns of violent victimization in the study sample
Full size table

Risky Lifestyles

Changes in income offending, aggressive offending, illicit drug use, and getting drunk are our intervening “risky lifestyle” variables that are treated as outcomes in a series of regression models. Income offending reflects non-violent crimes that have been linked to victimization in prior research (Berg and Loeber 2015; Frederick et al. 2013). Respondents at each wave were asked whether they did the following since the date of the last interview: “sold drugs,” “broke into a car to steal something,” “entered a building to steal something,” “shoplifted,” “bought or sold stolen property,” “stole a car or motorcycle,” and “used checks or credit cards illegally.” These seven items were adapted from Huizinga et al.’s (1991) Self-Reported Offending Scale, and were summed to create variety scores (ranging from 0 to 7).

Aggressive offending reflects youths’ aggressive crimes that are known to increase the risk of victimization (Piquero et al. 2005; Sampson and Lauritsen 1990). We capture aggressive offending separately from income offending since violent and aggressive behaviors may carry different risks for violent victimization (Lauritsen et al. 1991; Posick 2013; Pratt and Turanovic 2016). Aggressive offending at each wave is measured using a variety score that reflects whether respondents had “shot at someone,” “set fire to a house, building, car, or vacant lot,” “purposely destroyed someone else’s property,” “robbed someone with a weapon,” “robbed someone without a weapon,” “beat someone up badly enough that they probably needed a doctor,” or “had been in a fight” since the date of last interview (possible range 0–7). These items were similarly drawn from Huizinga et al. (1991).

Illicit drug use is measured using a variety score that reflects whether each respondent used cocaine, methamphetamine, or marijuana during the recall period (range 0–3). Getting drunk captures how often youth “got drunk” from drinking alcohol during each recall period, where closed-ended responses ranged from 0 (never) to 8 (everyday). Indicators of illicit drug use and getting drunk are included given their associations to concentrated disadvantage, victimization, and repeat victimization among youth and young adults (Leventhal and Brooks-Gunn 2000; Shorey et al. 2016; Turanovic and Pratt 2014). Individuals who use drugs and get drunk regularly may spend more time in risky settings at risky times, and be viewed as vulnerable targets for victimization by potential offenders (Bjarnason et al. 1999; Testa et al. 2010).

Concentrated Disadvantage

Our key independent variable and indicator of structural constraints is concentrated disadvantage. In keeping with prior research (Sampson et al. 1997; Sampson et al. 2005), concentrated disadvantage is measured using a weighted factor score of four items from the 2000 Census, measured at the block group-level: the proportion in poverty, the proportion unemployed, the proportion on welfare, and the proportion of single-parent households (eigenvalue = 2.87; factor loadings >.79; alpha = .84). Factor scores range from −3.29 to 1.70 standard deviations, where higher values indicate more concentrated disadvantage. Consistent with our research questions and theoretical framing, we focus on the average level of disadvantage that each respondent experiences across the 10 waves of data collection (i.e., the person-level mean of data collection). This measure is consistent with that of previous research using the Pathways data (Wright et al. 2014).

Since respondents can move to new areas and experience changes in neighborhood conditions over time, we conducted supplemental analyses that included a measure of within-person changes in concentrated disadvantage.Footnote 3 The results of these supplemental analyses showed that changes in concentrated disadvantage were linked to repeat victimization, but not to changes in risky lifestyles. While looking at the changes in structural conditions that youth experience is beyond the scope of this study, an important avenue for future research will be to explore further how victimized youth who move to more blighted neighborhoods face elevated risks of repeat victimization, and we revisit this issue in the Discussion Section.

Control Variables

Several well-known and theoretically-relevant correlates of victimization, concentrated disadvantage, and risky lifestyles are also included in the analyses to control for potential spuriousness.

Goal blockage assesses youths’ perceived opportunities for school and work in their neighborhoods (Agnew 1999, 2006). This construct originates from Eccles et al. (1998), and is measured at each wave using the following six items: “in my neighborhood it is hard to make money without doing something illegal,” “college is too expensive for most people in my neighborhood,” “the chances of getting ahead/being successful are not very good in my neighborhood,” “kids in my neighborhood do not have as much opportunity to succeed as kids from other neighborhoods,” “in my neighborhood it is easy for a young person to get a good job” (reverse-coded), and “most of my friends in my neighborhood will graduate from high school” (reverse-coded). Responses to each item ranged from 1 (strongly disagree) to 5 (strongly agree), and were summed and averaged so that greater scores reflect more goal blockage (α = .70).

Neighborhood disorder assesses physical characteristics of the environment surrounding each respondent’s home (Sampson and Raudenbush 1999). Items from this self-report measure tap physical disorder of the neighborhood (e.g., “cigarettes on the street or in the gutters,” “graffiti or tags”), as well as social disorder (e.g., “adults fighting or arguing loudly,” “people using needles or syringes to take drugs”) at each wave of data. The scale contains 21 items to which participants respond on a 4-point Likert scale ranging from 1 (never) to 4 (often), with higher scores indicating a greater degree of disorder within the community (α = .94). Reponses were summed and averaged so that higher values indicate greater neighborhood disorder.

Temperance is measured at each wave using subscales of “impulse control” and “suppression of aggression” from the Weinberger Adjustment Inventory (WAI; Weinberger and Schwartz 1990). The WAI contains eight items for impulse control (e.g., “I become ‘wild and crazy’ and do things people might not like”) and seven items for suppression of aggression (e.g., “people who get me angry better watch out”). Responses to these 15 items ranged from 1 (True) to 5 (False), and scores were summed and averaged so that higher values indicate greater temperance (α = .83) (Steinberg and Cauffman 1996). We treat temperance as a proxy for self-control, which has been closely linked to victimization, risky behavior, and structural disadvantage in prior literature (Pratt et al. 2014; Pratt et al. 2004; Zimmerman 2010).

Deviant peer influences on risky lifestyle changes and repeat victimization (Akers 1998; Warr 2002) are taken into account using a scale of seven items drawn from Elliott (1990) and Thornberry et al. (1994). These items assess how many of respondents’ friends did the following at each wave of data collection: “suggested you go out drinking with them,” “suggested you sell drugs,” “suggested you steal something,” “suggested you hit or beat someone up,” “suggested you carry a weapon,” “claimed that you have to be drunk to have a good time,” and “claimed that you have to be high on drugs to have a good time.” Responses to each item ranged from 1 (none of them) to 5 (all of them), and were summed and then averaged to create the measure of deviant peer influences (α = .89). We also include measures for whether each respondent is employed in a formal job (1 = yes, 0 = no) or enrolled in school at each interview (1 = yes, 0 = no).

Intelligence is measured using the Wechsler Abbreviated Scale of Intelligence (WASI) from the baseline interview (Wechsler 1999). The WASI produces an estimate of general intellectual ability based on two subtests, Vocabulary (42 total items that require the subject to orally define four images and 37 words presented both orally and visually) and Matrix Reasoning (35 incomplete grid patterns that require the participant to select the correct response from five possible choices). Administered in approximately 15 min, the WASI is a quick estimate of an individual’s level of intellectual functioning, with higher scores indicating greater intellectual ability. The WASI is linked to both the Wechsler Intelligence Scale for Children (WISC-III) and the Wechsler Adult Intelligence Scale (WAIS-III), and has been normed for individuals aged 6–89 years.

Early onset behaviors is a variety score (ranging from 0 to 5) that reflects whether each respondent got into trouble for the following behaviors before age 11: “fighting,” “stealing,” “being drunk or stoned,” “disturbing the class,” and “cheating.” In addition, we include the number of respondents’ prior petitions as a control variable (range 1–12). This measure is based on official records, and indicates the number of juvenile court petitions each respondent had prior to the Pathways baseline interview.

Low parental education reflects the mean of the biological mother and father’s education level at the baseline interview, where higher values reflect lower levels of education. We also include terms for age (which is measured at each wave of data collection and centered at age 18), male (1 = male, 0 = female), black (1 = black, 0 = otherwise), Hispanic (1 = Hispanic, 0 = otherwise), and other racial minority (1 = non-white, 0 = otherwise), where non-Hispanic white serves as the reference category. Lastly, variables for study site (1 = Phoenix, 0 = Philadelphia) and the number of months since the wave 1 interview (ranging from 0 to 78) are included in the analyses. Descriptive statistics for all variables included in the present study are presented in Table 2.

Table 2 Descriptive statistics (N × T = 1943)
Full size table

Analytic Strategy

Using a subsample of respondents who reported being violently victimized at the wave 1 interview, we assess the extent to which structural disadvantage influences changes in risky lifestyles, and whether such changes are linked to repeat victimization over nearly a 7 year period. We use multilevel modeling since the data have a two-level hierarchical structure, where repeated measures are nested within persons (Raudenbush and Bryk 2002). Level 1 of the data corresponds to within-person effects, and level 2 corresponds to between-person effects. Level 1 variables are also considered “time varying” since they reflect within-person changes over time (Hox 2010). Since our outcome variables of risky lifestyles (i.e., offending, illicit drug use, and getting drunk) and repeat victimization are skewed and zero-inflated, we use hierarchical negative binomial regression. This modeling approach is consistent with prior studies using the Pathways data (see, e.g., Sweeten et al. 2013a).

At level 1, our regression models contain the following time-varying covariates: temperance, deviant peers, being employed, being enrolled in school, age, and time in months since the wave 1 interview. In models predicting repeat victimization, risky lifestyle variables are also included as time-varying covariates. These covariates are specified as time-varying since they can change dramatically between adolescence and early adulthood. Each time-varying variable is disaggregated into time-stable and time-varying components. The time-stable component is the person-specific mean of each variable, and the time-varying component is created by subtracting the person-specific mean from each observation. This partitions the between- and within-person variance for all time varying covariates, and transforms level 1 of each model into a fixed effects analysis.

All time-stable components are measured at level 2, along with person-level variables that include concentrated disadvantage, goal blockage, neighborhood disorder, intelligence, early onset problem behaviors, prior petitions, low parental education, male, race/ethnicity, and the study site. Because these variables are measured as person-specific means, their coefficients reflect non-wave-specific effects on risky lifestyles and repeat victimization. By including covariates at level 1 and level 2, our analytic strategy guards against both static and dynamic forms of selection bias (Bjerk 2009).

Our analyses proceed in three stages. First, we estimate a series of hierarchical negative binomial models to determine whether concentrated disadvantage is related to changes in victims’ risky lifestyles (i.e., income offending, aggressive offending, illicit drug use, and getting drunk). Second, models are estimated to determine whether concentrated disadvantage and risky lifestyle changes are related to repeat victimization. Lastly, we conduct sensitivity analyses to assess the robustness of the findings to alternate specifications. All analyses are carried out using Stata 14 (StataCorp, College Station, TX).

Results

The multilevel models predicting changes to risky lifestyles and repeat victimization are presented in Tables 3 and 4. Turning to our first research objective, models 1 through 4 in Table 3 present equations predicting changes to risky lifestyles among the subsample of victims.Footnote 4 These findings show that concentrated disadvantage is positively and significantly related to increases in most of the risky lifestyles that we assess. Indeed, victims who live in neighborhoods characterized by higher levels of concentrated disadvantage are more likely to engage in aggressive offending (b = .161, p = .037), income offending (b = .242, p = .013), and use illicit drugs (b = .196, p = .023). The incident rate ratios (IRRs) calculated for these effects indicate that a unit increase in concentrated disadvantage increases the rate of aggressive offending by 17.5% (IRR = 1.175), the rate of income offending by 27.4% (IRR = 1.274), and the rate of illicit drug use by 21.7% (IRR = 1.217). However, concentrated disadvantage was not related to changes in how often victims reported getting drunk (b = .062, p = .233, IRR = 1.064). Put simply, youth who live in structurally disadvantaged communities engage in more aggressive offending, income offending, and illicit drug use post-victimization. These findings lend support to the notion that community-level structural constraints impose limits on victims’ abilities to reduce their involvement in risky lifestyles—specifically for aggressive offending, income offending, and illicit drug use.

Table 3 Select parameter estimates for models predicting changes in risky lifestyles among victims
Full size table
Table 4 Select parameter estimates for models predicting repeat victimization
Full size table

Consistent with our second research objective, five additional models are presented in Table 4 that assess whether concentrated disadvantage and risky lifestyle changes are related to repeat victimization. The findings presented in model 1 of Table 4 show that, net of controls, concentrated disadvantage is positively and significantly related to repeat victimization (b = .209, p = .016). This means that victims who live in disadvantaged neighborhoods are more likely to be victimized again, and the incident rate ratio for this coefficient suggests that a unit increase in concentrated disadvantage increases the rate of repeat victimization by 23.2% (IRR = 1.232). In addition, models 2 through 5 of Table 4 indicate that victims who engage in aggressive offending (b = .337, p < .001, IRR = 1.401), income offending (b = .196, p < .001, IRR = 1.217), use illicit drugs (b = .378, p < .001, IRR = 1.459), and get drunk more often (b = .103, p < .001, IRR = 1.108) are at increased risk of experiencing repeat victimization. The inclusion of changes to aggressive offending (model 2) and changes to income offending (model 3) also fully mediate the effects of concentrated disadvantage on repeat victimization seen in model 1.Footnote 5

Sensitivity Analyses

Thus far, the data suggest that the changes victims make (or do not make) to their risky lifestyles—specifically to aggressive offending, income offending, and illicit drug use—are influenced by the structural conditions of their communities, and that engaging in risky behaviors increases the likelihood of repeat victimization. It is important to note that these findings are not sensitive to the inclusion of theoretically relevant variables such as neighborhood goal blockage, neighborhood disorder, changes in temperance, changes in deviant peer influences, or changes in employment and school enrollment over time. Nevertheless, several additional models were estimated to determine the robustness of the results (not shown in table form). First, since black and Latino/a youth are more likely to reside in disadvantaged communities and to respond to these structural conditions differently (Peterson and Krivo 2010; Sampson and Bean 2006; Wright et al. 2016), analyses were conducted to determine whether the effects of structural constraints vary across racial and ethnic groups. To do so, interaction terms with race/ethnicity (i.e., black, Hispanic, and other racial minority) and concentrated disadvantage were specified. None of the interaction terms were statistically significant in models predicting changes to risky lifestyles or repeat victimization (p > .05), indicating that the effects of structural disadvantage appear to be similar across racial and ethnic groups in this sample of serious adolescent offenders.

Second, to determine whether the effects of structural constraints vary across male and female victims, interactions between male and concentrated disadvantage were estimated. No statistically significant interaction effects emerged (p > .05). These findings lend support to the notion that structural constraints tend to have similar effects on male and female victims of violence. The results also remained the same as those presented in Tables 3 and 4 (in terms of sign and significance) when models were re-estimated without females. The consistency of findings across all specifications gives added confidence that the results we present are robust and stable. In sum, victims of violence in economically deprived communities are more likely to continue to engage in risky lifestyles post-victimization, and are thus more likely to experience repeat victimization.

Discussion

The idea that our behavior is subject to structural constraints has been around for a long time. Hindelang et al. (1978) were perhaps the most explicit about it and made it a core part of their theory. And yet before them it was Cohen (1955) and Wolfgang and Ferracuti (1967) who discussed how delinquent subcultures—which are themselves a product of structural conditions—place constraints on youths’ behavioral choices. Our concern is that these works are no longer being read or empirically investigated with the idea of structural constraints in mind. And while the sources of victimization are complex (Pratt et al. 2014), it is clear that an appreciation for the behavioral boundaries imposed by social structure, while not completely absent from the current generation of victimization research, is certainly rare (for exceptions see Berg et al. 2012; Berg and Loeber 2011, 2015; Stewart et al. 2006). We would like to see a lot more of it. Accordingly, the purpose of the present study was to examine whether structural constraints (measured as concentrated disadvantage) limit victims’ ability to make changes to their risky behavioral routines which, in turn, would make it harder for them to avoid repeat victimization. Based on the analyses presented here, three primary conclusions are warranted.

First, our results highlight the importance of noting that the kinds of risky lifestyles that serve as precursors to victimization and repeat victimization are socially and structurally embedded. Put simply, structural constraints are real. We need to take them into account if we want to gain a better understanding of the causal processes underlying why some people get victimized and others are able to avoid it (see, e.g., Pratt and Turanovic 2016). Indeed, those who live their lives in communities plagued by economic deprivation will not only be exposed to greater levels of criminogenic structural conditions, but they will also be subjected to—and will therefore have to be sensitive to—the cultural expectations that emerge in response to those very conditions (Harding 2010; Turanovic and Pratt 2016). What this means in the present context is that victims of crime who live in economically deprived communities may be less able to change their risky behavioral routines, even if they really want to.

The key theoretical implication of these findings is that we clearly need a revised theory of risky lifestyles and victimization. While theories of offending have always been in mass production in criminology (see Lilly et al. 2015), theoretical advances in the study of victimization have largely stalled, and little in the way of new theoretical headway has been made for well over a decade. This is problematic since, just like we know that context matters when it comes to offending (Bursik and Grasmick 1993; Pratt and Cullen 2005; Sampson et al. 1997), we know that context matters for victimization as well (Pinchevsky and Wright 2012; Rountree et al. 1994; Thompson and Gartner 2014). But what we do not yet have is a theoretical perspective that recognizes the broader structural context that shapes victimization, the changes that people make (or do not make) to their antisocial behaviors, and their likelihood of repeat victimization. Developing one would be immensely useful for the next generation of victimization research.

The notion that risky lifestyles and victimization are socially and structurally embedded also has important policy implications. On the one hand, research indicates that victims from structurally disadvantaged communities are less likely to report their victimization to the police (Berg et al. 2013), and are therefore missing a critical link to receive victims’ services. But on the other hand, even if victims are able to get access to services, those services may be ill-equipped to meet the needs of many victims of crime. For example, a recent policy paper on the state of victims’ services in California is getting much attention for its call for “trauma-informed services” (Warnken 2014: 4; see also Long et al. 2016). This approach entails a heavy focus on individual coping strategies—which is certainly not a bad thing—but it does so without recognizing that victims’ coping resources and ability to put them into action are influenced in significant ways by the structural conditions that surround them (see, e.g., Agnew 2006; Kaufman et al. 2008; Skogan and Maxfield 1981). Until the importance of structural context is recognized by those with an influence over how victims’ services are delivered, some of the most socially vulnerable members of the population may not get the help they need.

Our second conclusion is that our results demonstrate the importance of modeling risky lifestyles as dynamic processes. This is rarely done in the victimization literature, even in longitudinal research designs. The typical approach instead is to measure some form of risky lifestyles (e.g., offending or substance use) at one point in time and see if it predicts victimization a year (or more) later. The implicit assumption being made is that those risky behaviors are highly stable for everyone, often over really long stretches of time. But while there is certainly evidence of short-term stability in criminal and deviant behaviors (McGloin et al. 2007; Sullivan et al. 2006, 2009), people change all the time over the life course in ways that will not be captured by the typical methodological approach employed in this literature. People can stop staying out late at night drinking in bars and picking fights they might lose; they can hang out with more prosocial friends and spend more time with a romantic partner; and they can learn to be less aggressive interpersonally so that others might be less inclined to behave violently toward them. Of course, we have shown here that making those changes is a lot easier for some people than it is for others (see also Turanovic and Pratt 2014), but it is clear that those changes have consequences and future research needs to model these changes accordingly whenever possible.

Relatedly, it is important to recognize that victims are a mobile group (Xie and McDowall 2008), and therefore they can move to different neighborhoods and experience changes in their community conditions over time. In supplemental analyses, we found that changes in concentrated disadvantage were linked to repeat victimization, but not to changes in risky lifestyles. Thus, additional casual processes likely need to be examined when considering why those who move to more disadvantaged neighborhoods might be at an increased risk of repeat victimization. It is possible that individuals who are new to more blighted areas lack the community-based social ties necessary to aid in their own protection (Kendrick et al. 2012), or that they are unfamiliar with how to safely navigate the streets in their new neighborhoods (Sharkey 2006). Either way, residential mobility is often the consequence of structure (particularly structural disadvantage, South and Deane 1993; see also Porter and Vogel 2014), and future work could continue to explore how moving to a new place elevates one’s risk of victimization even if they avoid engaging in traditional risky lifestyles like offending (Lindegaard et al. 2013).

Third, while our analyses allowed us to address key questions, our study also reveals several opportunities for future research in this area. For example, one crucial element that we could not assess directly concerns cultural pressures and how they are intertwined with structural constraints (Wilson 2009). It would be useful for future studies to examine the kinds of individual and contextual protective factors that might serve as buffers against the effects of criminogenic cultural elements and would keep people resilient to victimization and its consequences, especially in disadvantaged communities. In addition, and relatedly, future research could examine the extent to which these potential protective factors are age-graded. Victimization scholars have only just started to ask questions like this (Turanovic 2015; see also MacMillan 2001), and assessing whether (and how) individuals’ susceptibility or resilience to structural constraints might change as they age further into adulthood would add an important dimension to how we approach the study of victimization over the life course. Future work could also identify behavioral constraints imposed by things other than structural disadvantage. These constraints can come in many forms, such as those imposed by demographic factors (e.g., age and gender), family factors (e.g., family structure and process), and institutional factors (e.g., school characteristics, conditions of confinement in correctional facilities).

And finally, a critical—and yet still remaining—empirical question concerns which ecological unit of analysis is most responsible for imposing the kinds of structural constraints we have discussed here. While we examined rather small geographic spaces (block groups), there is still a lot of variability within such units in terms of where criminal events are most concentrated (Weisburd 2015). To be sure, even on the same block there are some specific places that are far more dangerous than others (Groff et al. 2010)—a pattern that also characterizes even smaller units of analysis like street segments (Weisburd et al. 2012). Nevertheless, scholars have long argued that no unit of analysis—however large or small—can be fully understood as if it were divorced from those next to it, above it, or below it (Peterson and Krivo 2010; Raudenbush and Sampson 1999; Sampson 2013). As such, future work should examine how the effect of structural constraints on risky lifestyles and victimization might be conditioned in important ways by the full spectrum of ecological contexts that surround peoples’ lives.

In the end, a more complete understanding of the sources and consequences of victimization will only be reached by shedding our allegiance to any particular unit of analysis (see Sampson 2012). Yes, individual traits and personality factors matter when it comes to victimization (Schreck 1999), but they only matter to the extent that they influence social behavior, and their effects are absolutely shaped by social context (Pratt 2016). In short, victimization does not occur in a vacuum because people do not live their lives in a vacuum. Context, as Sampson (2011: 83) noted, “is everything.” And good victimization research—research that makes a difference to how we think about victimization and the harm it can cause—will need to take context seriously as we move forward. We trust that the field is up to the task.