The Criminal Victimization–Depression Sequela

Abstract

Drawing on three waves of survey data, the authors examined the effects of criminal victimization on depression. First, the authors developed a structural equation model to determine whether criminal victimization predictsdepression. Second, recognizing that victimization is contingent on background factors, they tested whether victimization, conceptualized as an assigned treatment, has significant effects on depression, using inverse probability of treatment weighting (IPTW). In structural equation modeling equations, victimization predicts initial levels of depression and change in depression across waves. In the IPTW regression models, victimization had significant effects on levels of depression. There is considerable evidence to suggest that victimization influences depression, and investigators must be cautious when examining the temporal and selection issues surrounding the effects.

Keywords

criminal victimization depression violence mental health propensity

Introduction

Criminal victimization is a distressing experience. Yet it is not rare. According to the National Crime Victimization Survey (NCVS), there were more than 21.3 million violent and property victimization cases in the United States during 2008 (Rand, 2009), and these victimizations produce tremendous financial and social cost to the nation. These costs are not distributed evenly across social categories (Doerner & Lab, 2005; Macmillan, 2000, 2001; Rand, 2009). With the exception of sexual victimization, intimate partner violence, and stalking, males have higher rates than females (Catalano, 2006; Rand, 2009; Ruback & Thompson, 2001). Victimization is related inversely to age, with younger individuals suffering higher rates than older adults. By race, the robbery victimization rate for African Americans is nearly three times higher than the rate for Whites. By income, the burglary victimization rate for households with income less than $7,500 is nearly four times higher than the rate for households with incomes more than $75,000 (Rand, 2009). The connection between disadvantaged social status and high rates of victimization has implications for the connection between social status, stress, and distress (Aneshensel, 2009).

In a stress process model, Pearlin and colleagues (Pearlin, 1999; Pearlin, Schieman, Fazio, & Meersman, 2005) articulated how social structure comes to influence mental health, focusing on the connection between disadvantaged social status and high rates of psychopathology. In part, disadvantaged social status generates elevated levels of psychological distress because it exposes people to stress. In addition, disadvantaged social status is seen as limiting access to mediators and moderators of stress, so that lack of psychosocial resources such as social support, personal mastery, and coping skills is the route through which social status affects mental health (Aneshensel, 2009; Lazarus, 1966; Menaghan, 1983; Pearlin, 1999).

One of the most devastating consequences of criminal victimization is psychological trauma, and the links between criminal victimization and various traumatic effects are well known (Swan, Gambone, Fields, Sullivan, & Snow, 2005). A diverse literature (e.g., Kilpatrick & Acierno, 2003; Pearlin et al., 2005; Wheaton, 1994) indicates that victimization is a potent event that has the capacity to trigger temporary and prolonged distress, exhibited by diverse symptoms. Exposure to trauma jeopardizes health in many ways (Kessler & McGee, 1993; Pearlin, Schieman, Fazio, & Meersman, 2005), and the health and mental health consequences may be observed long after the occurrence of the trauma (Horwitz, Widom, McLaughlin, & White, 2001). There are empirical and theoretical reasons why victimization might be correlated with psychological symptoms. We hypothesized that incidents of criminal victimization are traumatic stressors that contribute to depression. Using a sample of older U.S. adults, we longitudinally examined the relationship of victimization to depression over three waves of data with an eye toward eliminating possible biases attributed to previous depression and selection for victimization.

The Victimization–Depression Interrelationship

Victimization correlates with many psychological and related problems, including depression, anxiety, posttraumatic stress disorder, suicidal behavior, substance abuse, avoidant behavior, fear of crime, self-blame, poor quality of life, social isolation, and reduced socioeconomic status (SES; Britt, 2001; Campbell, 2002; Lurigio, 1987; Macmillan, 2000, 2001; Ruback & Thompson, 2001), but the link to depression is particularly strong (Hawker & Boulton, 2000). Although the links between victimization and depression may be established, there is still basic empirical work to be done. Part of this work pertains to further clarifying the feedback between psychological variables and victimization over time and at different stages of life (Clements, Sabourin, & Spiby, 2004; Macmillan, 2001; Orva, McLeod, & Sharpe, 1996). For instance, the outcomes of victimization can be conceptualized as causes of unhealthy attitudes and behaviors that contribute to continued vulnerability and victimization (Kleinman, Schonfeld, Gould, Klomek, & Marrocco, 2007; Lauritsen, Laub, & Sampson, 1992; Manasse & Ganem, 2009; Menard, 2002; Nurious, Macy, Bhuyan, Holt, Kernic, & Rivara, 2003; Schreck, Stewart, & Fisher, 2006). Further complicating the causal questions concerning victimization and its psychological outcomes is the fact that many of the concurrent correlates of victimization are also short-term consequences (Hodges & Perry, 1999).

Even further complexity is added by the fact that victimization often reoccurs. For example, prior burglary victimization is an important predictor of future burglary, presumably because certain homes, and by extension persons, are vulnerable to burglary and because offenders may strike more than once in the same place (Ellingsworth, Farrell, & Pease, 1995; Pease, 1998; Polvi, Looman, Humphries, & Pease, 1990). Similarly, assaults are also more likely to occur among those who already have been assaulted, and this pattern may extend across childhood and adult sequences of victimization (Lloyd, Farrell, & Pease, 1994; Maker, Kemmelmeier, & Peterson, 2001; Messman-Moore, Long, & Siegfried, 2000; Noll, 2003).

Much of the research on the effects of victimization draws on data from youth and young adults (Caspi et al., 2002; Dodge, Bates, & Pettit, 1989; Finkelhor, 1995; Finkelhor & Dziuba-Leatherman, 1994; Widom, 1989). Less is known about the longitudinal consequences of victimization on psychological development in general adult populations, and studies of older adults are rare, although recent research indicates significant links between criminal victimization and psychological well-being in this population as well (DeLisi, Jones-Johnson, Johnson, & Hochstetler, 2010).

Although the occurrence of repeat victimization and its correlates are well known, there is limited understanding of the precise nature of the links between victimization and psychological symptoms. To rectify this, the current study examined stability and change in depression and whether they were linked to serious victimization in a panel of American adults. The research goals were (a) to determine whether additional instances of victimization had similar effects on depression as the initial instances and (b) to understand the effects of victimization on depression for members of the general population by accounting for selection effects on victimization.

Method

Participants

We analyzed national longitudinal data spanning from 1986 to 1995 from the Americans’ Changing Lives (ACL) study that were collected by the Institute for Social Research (House, 2002). The sample was selected from a multiple-stage-area probability sampling that oversampled African Americans, persons older than 59 years of age, and married women with husbands older than 64 years. Face-to-face interviews were conducted in 1986, 1989, and 1994, and only those cases with complete data through the last wave were used in the analysis. This yielded a small number of cases of missing data for variables of interest, which were replaced with the serial mean. The sample was weighted to be representative of the U.S. population in the interviewed age groups. Descriptive statistics for the weighted and unweighted samples are presented in Table 1.¹

Table 1.

Wave 1 Descriptive Statistics Weighted and Unweighted.

	Unweighted Range, Mean, Standard Deviation	Weighted Mean, Standard Deviation
Age	(24-96), 54, 17.6	47, 16.5
Race	(0-1 White), 0.64, 0.48	0.83, 0.37
Sex	(0-1 male), 0.38, 0.48	0.47, 0.50
Married	(0-1 married), 0.55, 0.50	0.69, 0.46
Drinks	(0-4), 0.97, 1.19	1.18, 1.23
Verbal IQ	(0-6), 3.35, 1.13	3.53, 1.14
Low SES	(0-1 low), 0.32, 0.47	0.19, 0.40
Weight	(80-380 lb), 162, 36	162, 36
BMI	(11.5-55), 26, 5.1	25.5, 4.7
Buildings kept	(1-4), 1.68, 0.75	1.54, 0.68
Yards kept	(1-4), 1.68, 0.76	1.64, 0.71
Victimization	(0-1), 0.23, 0.42	0.24, 0.43
Felt depressed	(1-3), 1.43, 0.58	1.47, 0.61
Everything an effort	(1-3), 1.61, 0.70	1.62, 0.71
Sleep restless	(1-3), 1.65, 0.70	1.67, 0.70
Felt happy	(1-3), 1.39, 0.59	1.43, 0.62
Felt lonely	(1-3), 1.36, 0.57	1.43, 0.61
People unfriendly	(1-3), 1.27, 0.50	1.27, 0.52
Enjoyed life	(1-3), 1.39, 0.60	1.34, 0.60
Didn’t feel like eating	(1-3), 1.32, 0.56	1.38, 0.60
Felt sad	(1-3), 1.39, 0.57	1.43, 0.60
People dislike me	(1-3), 1.18, 0.42	1.20, 0.46
Couldn’t get going	(1-3), 1.52, 0.58	1.56, 0.61

Note. IQ = intelligence quotient; SES = socioeconomic status; BMI = body mass index.

The analysis was conducted in two parts. First, we examined a number of structural equation models to determine the influence of victimization in each wave on depression and change in depression. Structural equation modeling (SEM) was selected primarily for its ability to specify latent variable models that provide separate estimates of the relations among latent constructs and their manifest indicators (the measurement model) and of the relations among constructs (the structural model). This allows investigators to assess the psychometric properties of measures and to estimate the relations among constructs that are corrected for biases attributable to random error and construct-irrelevant variance (Bollen, 1989).

Measures

Depression

The depression measure was created from items derived from the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, 1977), using four parcels to form a latent depression factor in each wave of data.² The parcels were theoretically driven, based on the original specification of a four-factor solution for the CES-D and on findings concerning the factor structure of the CES-D documented in previous studies and a variety of populations (Radloff, 1977; Yu & Yu, 2007). The items were combined by calculating means to create parcels that represented mood, positive outlook (reverse coded), somatic responses, and interpersonal relations. There were 11 items in the depression measure and answer sets, ranging from 1 = hardly ever to 3 = most of the time (Table 1). These items had acceptable reliability when treated as a single factor (Cronbach’s α = .78). There were also high levels of reliability among the items used in the four parcels: Cronbach’s α = .77 (mood), .65 (somatic), .72 (positive), and .68 (interpersonal). The four parcels then were used as indicators of the latent variable depression. The parcels forming the latent factor were also found to have acceptable reliability (Cronbach’s α = .71). In the propensity score analyses reported in the latter portion of the findings, the measure is the factor score of the four parcels, where items were added to form each parcel, approximating the latent variable measure used in the SEM.

Victimization

The measure of victimization indicates the occurrence of victimization by several forms of serious crime in the past 3 years. A bivariate measure was created from two questions: (a) Have you ever been the victim of a serious attack or physical assault in the past 3 years? (b) Apart from what you have already told me, were you robbed, or was your home burglarized in the past 3 years? If the respondents answered yes to either or both of these items, then they were coded as a victim (1 = yes, 0 = no).

Exogenous variables

Three exogenous control variables were used in the SEM portion of the analysis. Age was continuously measured (M = 53.5 years, range = 25-95). Race was coded as self-reporting as White or another race (1 = White, 0 = other).³ Sex was coded (0 = female, 1 = male).

Propensity score covariates

Additional variables were used in the propensity score component of the analysis that were not included in the SEM. Controls included being currently married (0 = no, 1 = yes) and whether the respondent reported having any drinks in the past month (0 = no, 1 = yes). Verbal intelligence, which is the number of correct answers to six statements that require missing words to be inserted in the blanks, was also included as a control. Verbal intelligence data for Wave 3 were taken from Wave 2 as they were not collected in the former. SES was coded as affirmative for low SES based on a combination of family income and respondent education. Low SES (0 = other, 1 = low) was a composite of less than 11 years of education at Wave 1 and current family income less than $20,000. Where income data were missing, the figure from the previous wave was used. Wave 1 education data were used for all waves as they indicated the timely completion of 11 years of education. Two measures of perceptual neighborhood conditions were also included in the propensity score calculation. The respondents were asked how well the structures in the neighborhood and yards in the neighborhood were kept (ranging from 1 = very well to 4 = very poorly). These perceptual neighborhood variables were not included in Wave 3, so we used data from the previous wave after deciding that they were relevant in the calculation of propensity scores. The final two control variables are body weight and body mass index, with the latter squared to control for potential confounding factors of body weight on depression, as body type is linked to victimization in some groups (Bauman, 2008; Man & Cronan, 2002) and an extensive literature links it to depression (Dragan & Akhtar-Danesh, 2007; Zhao et al., 2009).

In the propensity score analysis and the IPTW regression for Waves 1 and 2, we included the victimization variable lagged to previous waves of data. Wave 2 models included victimization data from Wave 1, and Wave 3 models included victimization data from Waves 1 and 2, a method that acknowledged that one indicator of propensity for victimization is previous victimization. In other words, victimization in the previous wave was likely to predict victimization subsequently.

Analytical Strategy

A two-pronged strategy was used to examine the effects of victimization on depression. The first used an SEM model where victimization predicted depression over three waves, using M-plus (Muthén & Muthén, 2007). Figure 1 illustrates the conceptual model for the final SEM analysis with three waves of data.

Figure 1.

Conceptual model for SEM in Waves 1-3.

Exogenous controls included age, race, and sex. In conjunction with victimization in each wave, these are used to predict initial levels of depression and change in depression between waves. The model also examined the correlation between the exogenous controls. In Waves 2 and 3, depression was regressed on lagged depression, resulting in an analysis of change. The goal of this model was to determine how victimization shaped the psychological outcomes net its prior level. SEM models were reported sequentially, where the first step was to report the fit of the measurement model for depression. Next, we reported the results for Wave 1 followed by a change model for Wave 2. Finally, the effects of victimization on depression were examined with controls. The analysis was reported in steps because of attrition of sample members between waves.⁴

The second portion of the analysis acknowledged that the effects of victimization on depression could be biased by the fact that victimization was not occurring randomly (Lauritsen & Quinet, 1995; Pease, 1998; Sorenson & Golding, 1990). This is particularly a matter of concern in that the same variables that predicted victimization might well predict depression. To account for this selection problem, we adjusted for individuals’ exposure propensities for victimization based on several predictor variables, using inverse probability of treatment weighting (IPTW).⁵ We first calculated propensity for victimization in logistic regression analyses for each wave and then adjusted the data with weights for these propensities. IPTW uses the estimated odds of exposure to assign individual weights to all observations, resulting in an altered composition of the study population. We used weights that were the inverse of the propensity score in those exposed and the inverse of 1 minus the propensity score (i.e., the probability of “nonexposure”) in those unexposed. In this way, the investigator can mimic randomization as the chances of exposure are leveled by applying weights. With weights applied, we regressed depression on the predictors in all three waves to determine whether victimization predicted depression. The results reveal the effects of recent victimization on depression net of controls and where victimization had been randomized as a treatment.

Findings

In the findings that follow, we report on the model characteristic and the results of the SEM analysis. We then describe the findings from estimating the propensity of victimization, before proceeding to the presentation of findings for the effects of victimization on depression when accounting for selection.

Structural Equation Modeling

Table 2 reports the coefficients for the depression measurement models. For the sake of brevity, the table presents measurement models for Wave 3 (present data). The findings for the SEM reveal that the measured items for depression load convincingly on the latent factors. In addition, the fit of the model may be interpreted as satisfactory to proceed, although the fit statistics indicate moderate to questionable levels of fit of the model to the data. For example, with two waves of data in the measurement model and the latent variable depression in Wave 2 regressed on the latent variable depression for Wave 1, the fit was adequate, albeit not excellent: χ² = 545, df = 18, p = .00; comparative fit index (CFI) = .93; Tucker–Lewis index (TLI) = .89; root mean square error of approximation (RMSEA) = .10; RMSEA 90% confidence interval (CI) = .09, .11; standardized root mean square residual (SRMR) = .04. The standardized effects of each parcel on its latent variable were significant at p < .000, indicating that the items were indicators of those constructs (Wave 1 depression = interpersonal β = .44, mood β = .92, somatic β = .77, positive β = .67; Wave 2 depression = interpersonal β = .57, mood β = .80, somatic β = .69, positive β = .67). There was stability in the depression measure across waves, and the standardized effect of depression in Wave 1 on Wave 2 depression was significant (β =.57, p = .000).

Table 2.

Standardized Effects for Measurement Models for Waves 1-3.

	Depression₁	Depression₂	Depression₃
Wave 1 data
Interpersonal	.46*
Mood	.87*
Somatic	.66*
Positive	.68*
n = 3,617, CFI = .99, TLI = .97, RMSEA = .07, SRMR = .02, RMSEA 90% CI = .06, .09
Waves 1-2 data
Interpersonal	.44*	.57*
Mood	.92*	.80*
Somatic	.77*	.69*
Positive	.67*	.67*
Depression _t ₋₁		.57*
n = 2,867, CFI = .93, TLI = .89, RMSEA = .10, SRMR = .04, RMSEA 90% CI = .09, .11
Waves 1-3 data
Interpersonal	.45*	.58*	.54*
Mood	.83*	.74*	.90*
Somatic	.70*	.68*	.70*
Positive	.70*	.74*	.71*
Depression _t ₋₁		.63*	.70*

p < .05

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual.

As shown in Table 2, using three waves of data for depression with Wave 3 regressed on Wave 2 and Wave 2 regressed on Wave 1 (n = 2,345) worsened the fit considerably in a measurement model where no predictors were used for the three wave measurement models (CFI = .88; TLI = .85; RMSEA = .05; RMSEA 90% CI = .06, .07; SRMR = .05) The preceding two models suggested that the fit of the model with two waves was acceptable, but adding a third wave resulted in a moderately poor fit with which to begin the analyses.⁶ As in Waves 1 and 2, the parcels in Wave 3 were significantly related to the depression factor, p < .000 (interpersonal β = .54, mood β = .90, somatic β = .70, positive β = .71). Depression in Wave 2 also predicted depression in Wave 3 (β = .70, p = .000).

The remaining SEM results are shown in Table 3. The next step of the analysis was to examine the predictors of depression in Wave 1 (n = 3,617). The fit of this model was adequate (CFI = .93; TLI = .88; RMSEA = .07; RMSEA 90% CI = .07, .08; SRMR = .04). The effects of all independent variables and all parcels on the latent factor were significant at p < .001. Victimization predicted depression (β = .29, p < .000). Age (β = −.10, p = .000), race (β = −.06, p = .001), and sex (β = −.12, p = .000) also were significant predictors. Exogenous variables were correlated in all the theoretical SEM models, and these variables also were correlated with each wave’s victimization variable. Two of the three relationships between the controls were significant, suggesting that females and Whites were significantly older in the weighted data, as is true in the population (age with race β = .14, p = .000; age with sex β = −.05, p = .000).

Table 3.

SEM Standardized Effects for Measurement Model and Predictors of Depression Waves 1-3.

	Depression₁	Depression₂	Depression₃
Wave 1 data
Interpersonal	.45*
Mood	.91*
Somatic	.63*
Positive	.52*
Victimization	.29*
Age	−.10*
Race	−.06*
Sex	−.12*
n = 3,617, CFI = .93, TLI = .88, RMSEA = .07, SRMR = .04, RMSEA 90% CI = .07, .08
Waves 1-2 data
Interpersonal	.44*	.50*
Mood	.92*	.82*
Somatic	.69*	.65*
Positive	.49*	.55*
Depression _t ₋₁		.54*
Victimization _t	.06*	.08*
Age	−.10*	−.02
Race	−.17*	−.07
Sex	−.30*	−.17*
n = 2,867, CFI = .88, TLI = .84, RMSEA = .07, SRMR = .05, RMSEA 90% CI = .07, .08
Waves 1-3 data
Interpersonal	.48*	.52*	.62*
Mood	.85*	.78*	.82*
Somatic	.65*	.64*	.72*
Positive	.52*	.56*	.53*
Depression _t ₋₁		.64*	.59*
Victimization _t	.07*	.07*	.06*
Age	−.18*	−.03	−.09*
Race	−.12*	.00	−.14*
Sex	−.11*	−.07*	.00
n = 2,345, CFI = .88, TLI = .85, RMSEA = .06, SRMR = .046, RMSEA 90% CI = .06, .07

p < .05

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual.

By adding Wave 2 to this model, we can assess whether victimization subsequent to Wave 1 predicted change in depression between the waves (n = 2,867). We modeled initial levels of depression and change in depression between Waves 1 and 2 and predicted each with exogenous variables and current-wave victimization (CFI = .88; TLI = .84; RMSEA = .07; RMSEA 90% CI = .07, .08; SRMR = .05).⁷ There was stability across waves in the depression measure (β = .54, p <.000). Victimization predicted both Wave 1 depression (β = .06, p = .001) and change in depression between Waves 1 and 2 (β = .08, p = .007). All the controls were predictors of initial levels of depression (age β = −.10, p = .000; race β = −.17, p = .000; sex β = −.30,p = .000). Sex predicted change in depression (sex β = −.17, p = .000). Victimization between Waves 1 and 2 was correlated significantly (β = .08, p = .000).

For the final part of the SEM analysis, we examined an extension of the model with all cases where Wave 3 data were present (n = 2,345). This allowed us to determine whether victimization continued to predict change in depression between Waves 2 and 3. The fit for Wave 3 was poor and was not improved greatly by the addition of predictor variables, but an acceptable to poor fit was achieved (CFI = .88; TLI = .85; RMSEA = .06; RMSEA 90% = .06, .07; SRMR = .05). Wave 3 predictors of depression included depression lagged to Wave 2 (β =.59, p = .000), victimization between Waves 2 and 3 (β = .06, p = .000), and controls for age and race (age β = −.09, p = .028; race β = −.14, p = .000). The results for Waves 1 and 2 were substantively similar to those reported in previous analyses, with victimization proving significant in all three waves.

In sum, the SEM analyses supported a confident assertion that victimization predicted both initial levels of depression and change in depression over time. Clearly, victimization was related across waves. Therefore, the selection effect operating on victimization may influence the results of victimization’s effect. To illustrate this, we need only point to the persons who reported being victimized in Wave 2. Of these, 31% were also victimized in Wave 1.

It is important to note that the SEM models included only the most fundamental control variables: age, race, and sex. The use of many variables in SEM becomes cumbersome and generally reduces the fit if they are not correlated strongly with outcomes according to some measures. Moreover, the odds of victimization are unlikely to be evenly distributed in a population; this can be seen but is not adjusted in the SEM. Therefore, uneven exposure propensities lead to correlations that do not reflect what the effects of victimization would be if it could be applied randomly to members of the population of interest, but instead, they reflect the probability that the effects found in these data can be generalized to similar populations.

IPTW Results

Table 4 displays the results of a logistic regression to determine the propensity for victimization. In the 3 years preceding Wave 1, 13% of the participants were victimized. We predicted victimization based on age, race, sex, drinking, marital status, weight, body mass index, low SES, perceptions of how yards are kept in the neighborhood, and perceptions of how structures are kept in the neighborhood. The model explained about 7% of the variation in victimization (−2 log likelihood [LL] = 2579.33, Wald = 1,439, p = .000, Nagelkerke R² = .065). As seen in Table 4, which gives the logistic regression coefficients that accompany the following percent change figures, age (%Δ = .02), drinking (%β = 16.9), marital status (%Δ = .52), and low SES (%Δ = .14) were all significant predictors of victimization in Wave1.

Table 4.

Unstandardized Logistic Regression Coefficient and Odds Ratio (Ψ) for Victimization Waves 1-3.

	Victimization₁	Victimization₂	Victimization₃
Age	−0.02* (Ψ = 0.98)	−0.01* (Ψ = 0.99)	−0.02* (Ψ = 0.98)
Race	0.09 (Ψ = 1.10)	−0.53* (Ψ = 0.59)	−0.03 (Ψ = -0.97)
Sex	−0.05 (Ψ = 0.95)	0.15 (Ψ = 1.17)	0.13 (Ψ = 1.14)
Married	−0.74* (Ψ = .48)	−0.51* (Ψ = 0.60)	−0.55* (Ψ = 0.58)
Drinks	0.13* (Ψ = 1.14)	0.00 (Ψ = 1.0)	0.02 (Ψ = 1.03)
Verbal IQ	0.00 (Ψ = 1.00)	0.07 (Ψ = 1.08)	0.03 (Ψ = 1.03)
Low SES	−0.03* (Ψ = .97)	0.02 (Ψ = 1.02)	0.09 (Ψ = 1.10)
Weight	0.00 (Ψ = 1.00)	0.00 (Ψ = 1.00)	0.00 (Ψ = 1.0)
BMI	0.000 (Ψ = 1.00)	0.00 (Ψ = 1.00)	0.00 (Ψ = 1.00)
Buildings kept	0.15 (Ψ = 1.16)	0.23 (Ψ = 1.25)	−0.47* (Ψ = 0.62)
Yards kept	0.05 (Ψ = 1.05)	0.12 (Ψ = 1.13)	0.51 (Ψ = 1.66)
Victimization₁		1.02* (Ψ = 2.77)	0.59*(Ψ = 1.80)
Victimization₂			0.99* (Ψ = 2.7)
	−2 Loglikelihood = 2,579, NagelkerkeR² = .07, n = 3,617	−2 Loglikelihood = 1,852, NagelkerkeR² = .08, n = 2,867	−2 Loglikelihood = 1,852, NagelkerkeR² = .08, n = 2,345

p < .05

Note. IQ = intelligence quotient; SES = socioeconomic status; BMI = body mass index.

In the model with propensity-weighted data constructed from the Wave 1 results in the preceding paragraph, and reported in Table 5, we regressed the depression factor on victimization, race, sex, marital status, drinking, verbal intelligence, low SES, weight in pounds, the perception that yards are kept well in the neighborhood, the perception that structures are well kept in the neighborhood, and body mass index squared. In the IPTW regression model, victimization was a significant predictor of depression in Wave 1 (β = .10, p = .000). The model was significant and explained 12% of the variation in depression (adjusted R² = .12, F = 76, p = .000). Additional variables that achieved significance at p < .05 when propensities for victimization were weighted include age (%Δ = −.13, p = .000), race (β = −.09, p = .000), marital status (β = −.18, p = .000), verbal intelligence (β = −.08, p = .000), low SES (β = .13, p = .000), weight (β = −.03, p = .000), and body mass index squared (β = .00, p = .000).

Table 5.

Propensity-Adjusted (IPTW) Unstandardized and Standardized Effects for Predictors of Depression Waves 1-3.

	B _t ₁	(β) _t ₁	B _t ₂	(β) _t ₂	B _t ₃	(β) _t ₃
Victimization _t _-1	.22 (.025)	.10*	.17 (.03)	.07*	.16 (.06)	.05*
Age	−.01 (.00)	−.13*	−.01 (.00)	−.07*	−.01 (.00)	−.06*
Race	−.20 (.03)	−.09*	−.25 (.04)	−.10*	−.43 (.06)	−.13*
Sex	.03 (.04)	.01	−.09 (.05)	−.04*	−.06 (.05)	−.03
Married	−.39 (.03)	−.18*	−.36 (.03)	−.15*	−.28 (.05)	−.11*
Drinks	−.02 (.01)	−.02	.00 (.00)	−.02	.00 (.05)	−.00
Verbal intelligence	−.08 (.01)	−.08*	−.05 (.02)	−.04*	−.03 (.02)	−.03
Low SES	.31 (.03)	.13*	.44 (.04)	.16*	.53 (.06)	.17*
Weight	−.00 (.03)	−.03*	−00 (.00)	−.04*	.00 (.00)	.03
BMI	.01 (.10)	.00*	.00 (.00)	.08*	.00 (.00)	.04
Buildings kept	−.00 (.00)	−.03	.047 (.07)	.01	.08 (.10)	.02
Yards kept	.05 (.03)	.03	.25 (.07)	.08*	.14 (.07)	.04
Adjusted R²	.12		.10			.09

p < .05

Note. SES = socioeconomic status; BMI = body mass index.

Not surprisingly, the logistic regression model used to create the propensity score explained more of the variation (though only 8%) in Wave 2 victimization when a lagged victimization variable was included (−2 LL = 1852.04, Wald = 1272.62, p = .000, Nagelkerke R² = .080). As seen in Table 4, age (%Δ = 1.2), race (%Δ = 40.9), marital status (%Δ = 39.9), and Wave 1 victimization (%Δ = 1.9) were all significant predictors of victimization in Wave 2.

IPTW results for Wave 2 are shown in Table 5. The IPTW model for Wave 2 was significant (adjusted R² = .10, F = 52, p = .000). Victimization was significant again (β = .07, p = .000). Race (β = −.10, p = .000), sex (β = −04,p = .036), low SES (β = .16, p = .000), marital status (β = −.15, p = .000), verbal intelligence (β = −.04, p = .002), weight (β = −.04, p = .016), how well yards were kept (β = .08, p = .000), and body mass index (β = .08, p = .004) all were significant predictors. Wave 2 data buttress confidence that victimization significantly predicts depression when it occurs at random in a population.

The logistic regression equation for Wave 3 given in Table 4 predicted the binary victimization variable. Of the total respondents to Wave 3, 11% were victimized between Waves 2 and 3. Of these victims, 22% also had been victimized in Wave 2. The model explained 8% of the variation in victimization (−2 LL = 1852.19, Wald = 1190.81, p = .000, Nagelkerke R² = .080). Age (%Δ = 2.0), marital status (%Δ = 44), how structures are kept (%Δ = −66), victimization in Wave 1 (%Δ = 82), and victimization in Wave 2 (%Δ = 170) were significant predictors of victimization in Wave 3.

Table 5 presents the IPTW results for Wave 3. The explained variation improves slightly (adjusted R² = .09, F = 23, p = .000). Victimization between Waves 2 and 3 was a significant predictor of depression in Wave 3, confirming that victimization affects change in depression (β = .05, p = .013). Significant control variables include age (β = −.06, p = .002), race (β = −.13, p = .000), marital status (β = −.11, p = .000), and low SES (β = .17, p = .000).

Discussion

What at first appears to be a straightforward empirical question often becomes complex when selection effects are likely and randomization cannot feasibly be achieved. This is the case when attempting to evaluate the effects of criminal victimization on depression. One problem is that understanding the relationship between victimization and depression in real-world distributions of their occurrence may not provide solutions to the question of what victimization can be expected to do to the average person or in the general population. This study attempted to show how victimization predicts depression over three waves of nationally representative data for older Americans. Three types of questions were addressed: (a) What is the initial relationship of victimization to depression? (b) Do future instances of victimization affect the change in depression? (c) Does any relationship of victimization hold when selection effects, conceptualized as the propensity for victimization, are controlled?

The results of the analysis were remarkably consistent, although none of the effects explain a great deal of the variation in depression. Depression, of course, has many social antecedents and also varies according to the conditions of life. Effect sizes are seldom large in predicting it, especially in studies that contain few control variables. Based on these data, we can say with confidence that criminal victimization predicts levels of depression and change in depression among a population of adults. The results are impressive given that in the SEM portions of the analysis—where stability in depression was apparent—victimization influences it nevertheless. In the propensity-scoring portion of the analysis, we used previous victimization to develop the propensity scores in Waves 2 and 3 (effectively randomizing victimization), but victimization still maintained a significant effect. The only other variables that were significant across all three waves were age, race, marital status, and low SES. All these variables are important for determining a wide range of life outcomes and are often found to have significant effects on the psychological variables (Cloninger, 2004; House et al., 1994; Mirowsky & Ross, 2003). These findings support the expansion of the stress process model to include social contexts (Pearlin, 1999). Given the caution in our methodology and the fact that victimization might have preceded the measurement of depression by some time, we were surprised at the consistent confirmation of the hypothesis.

In providing multiple waves of victimization and psychological outcomes over a long period of time for a nationally representative sample, the current data suit the purposes at hand. As is often the case with secondary analysis, however, many of the limitations of this study are located in deficiencies of the data for the question of interest and in the necessity of selecting a manageable focus for a research question. It has not escaped us, for example, that the measures of criminal victimization are dichotomous and entail several forms of crime that are almost certain to have varying effects, with little attention to the specifics of certain offenses such as domestic violence. Moreover, there are long periods between waves of data, which introduces problems for self-report information, including but not limited to telescoping of events. In addition to these problems, we have devoted no attention to gendered interactions or other interaction effects in this study by electing to focus on the problem of selection into victimization. Nevertheless, the data suit the primary purpose at hand, which is to identify and correct for stability in depression and selection into victimization.

Although the answers to some questions appear apparent before the research is done, simple problems often entail empirical complexity. The current study was inspired by the desire to show that the links between criminal victimization and depression shown by simpler, extant strategies—especially cross-sectional designs—can be called into question. Vulnerable persons may exhibit their vulnerability across a range of variables. Therefore, the reciprocal relationships between multiple psychological conditions, criminal victimization, and lifestyles deserve greater inquiry to unpack the causal directions as well the characteristics that moderate and mediate the relationships. Moreover, group differences in outcomes and resiliency to depression also deserve careful analysis, to account for the propensities for being harmed and exhibiting manifestations of that harm. It is hoped that the current empirical study and its specific application of techniques for examining the criminal victimization–depression sequela will inspire others intent on pursuing these questions with newer data precisely designed for that purpose.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the authorship and/or publication of this article.

Funding

The author(s) received no financial support for the research and/or authorship of this article.

1.

In each wave, we used only those respondents who participated in the current interviews. This was the most conservative approach, with substantial attrition, and followed precedent with these data (House et al., 1994). Much of the data in ACL are not missing at random, but victimization risk does not differ between those Wave 1 participants who are missing and those who are present in subsequent waves (we compared those who were missing and present at wave 2 using wave 1 data; t = .50, p = .62).

2.

The depression measure was not identical across waves as the original researchers were keeping the scale to date. Only two questions changed. It was decided to use all available information from the scale in each wave to create the parcels.

3.

In the unweighted data, the sample declared in nonexclusive categories as 64% White, 32% Black, 3% American Indian, 0.8% Asian, and 0.2% other.

4.

Wave 1 used all cases with missing data replaced with serial means. Subsequent waves used only data with current participation, with additional missing data replaced with the serial mean.

5.

In calculating the propensity score, we used the sample weights included in the data so that the data were weighted, first, to generalize to the U.S. population in the interviewed age groups and, second, to account for the differential odds of exposure in IPTW. Sampling weights were applied in SEM analysis.

6.

We decided that it was worthwhile to present the three wave analyses with a warning and caveat to the reader. It should be noted that when no controls were included and victimization Wave 1, Wave 2, and Wave 3 were the only predictors, the model achieved a much better fit (CFI = .89, TLI = .88, RMSEA = .068, RMSEA 90% CI = .06, .08, SRMR = .05).

7.

CFI and TLI are disadvantaged by low correlations and penalized for each path added to the models; values of .95 or higher indicate a very good fit, but values close to .90 generally are viewed as acceptable. SRMR is advantaged by large samples and many paths; values of .08 or lower indicate a good fit. RMSEA assesses the model against the population; .05 is a good fit, and .10 indicates a poor fit. Before the addition of control variables, using only victimization and depression measures, the fit was better for this model (CFI = .94, TLI = .91, RMSEA = .06, RMSEA 90% CI = .06, .07, SRMR = .04).

Author Biographies

Andy Hochstetler is an associate professor of sociology and affiliated with Criminal Justice Studies at Iowa State University.

Matt DeLisi is coordinator of Criminal Justice Studies at Iowa State University and editor-in-chief of the Journal of Criminal Justice.

Gloria Jones-Johnson is a professor of sociology at Iowa State University.

W. Roy Johnson is an associate professor of management at Iowa State University.

References

Aneshensel

C. S.

(2009). Toward explaining mental health disparities. Journal of Health and Social Behavior, 50, 377-394.

Bauman

(2008). Victimization by bullying and harassment in high school: Findings from the 2005 youth risk behavior survey in a southwestern state. Journal of School Violence, 7, 86-104.

Bollen

K. A.

(1989). Structural equations with latent variables. New York, NY: John Wiley.

Britt

C. L.

(2001). Health consequences of criminal victimization. International Review of Victimology, 8, 63-73.

Campbell

J. C.

(2002). Health consequences of intimate partner violence. Lancet, 359, 1331-1336.

Caspi

McClay

Moffitt

T. E.

Mill

Martin

Craig

I. W.

. . .Poulton

(2002). Role of genotype in the cycle of violence of maltreated children. Science, 297, 851-854.

Catalano

S. M.

(2006). Criminal victimization, 2005 (Bureau of Justice Statistics Bulletin). Washington, DC: U.S. Department of Justice.

Clements

C. M.

Sabourin

C. M.

Spiby

(2004). Dysphoria and hopelessness following battering: The role of perceived control, coping, and self-esteem. Journal of Family Violence, 19, 25-36.

Cloninger

C. R.

(2004). Feeling good: The science of well-being. New York, NY: Oxford University Press.

10.

DeLisi

Jones-Johnson

Johnson

W. R.

Hochstetler

(2010). The aftermath of criminal victimization: Race, self-esteem, and self-efficacy. Crime & Delinquency, 60(1), 85-105.

11.

Dodge

K. A.

Bates

J. E.

Pettit

G. S.

(1989). Mechanisms in the cycle of violence. Science, 250, 1678-1683.

12.

Doerner

W. G.

Lab

S. P.

(2005). Victimology (4th ed.). Cincinnati, OH: LexisNexis.

13.

Dragan

Akhtar-Danesh

(2007). Relation between body mass index and depression: A structural equation modeling approach. Biomed Central Medical Research Methodology, 7, 17.

14.

Ellingsworth

Farrel

Pease

(1995). A victim is a victim is a victim: Chronic victimization in four sweeps of the British Crime Survey. British Journal of Criminology, 35, 360-365.

15.

Finkelhor

(1995). The victimization of children. A developmental perspective. American Journal of Orthopsychiatry, 65, 177-193.

16.

Finkelhor

Dziuba-Leatherman

(1994). Children as victims of violence: A national survey. Pediatrics, 94, 413-420.

17.

Hawker

D. S. J.

Boulton

M. J.

(2000). Twenty years’ research on peer victimization and psychological maladjustment: A meta-analytic review of cross-sectional studies. Journal of Child Psychology and Psychiatry, 41, 441-455.

18.

Hodges

Perry

(1999). Personal and interpersonal antecedents and consequences of victimization by peers. Journal of Personality and Social Psychology, 76, 676-685.

19.

Horwitz

Widom

C. S.

McLaughlin

White

H. R.

(2001). The impact of childhood abuse and neglect on adult mental health: A prospective study. Journal of Health and Social Behavior, 42, 184-201.

20.

House

J. S.

(2002). Americans’ changing lives: Waves I and II, 1986 and 1989 (ICPSR version) [Computer file]. Ann Arbor, MI: University of Michigan, Survey Research Center [producer]. Ann Arbor, MI: Inter-University Consortium for Political and Social Research [distributor].

21.

House

J. S.

Lepkowski

J. M.

Kinney

A. M.

Mero

R. P.

Kessler

R. C.

Herzog

A. R.

(1994). The social stratification of aging and health. Journal of Health and Social Behavior, 35, 213-234.

22.

Kessler

R. C.

Magee

(1993). Childhood adventures and adult depression: Basic patterns of association in a U.S. national survey. Psychological Medicine, 23, 679-690.

23.

Kilpatrick

Acierno

(2003). Mental health needs of crime victims: Epidemiology and outcomes. Journal of Traumatic Stress, 16, 2, 119-132.

24.

Kleinman

Schonfeld

I. S.

Gould

M. S.

Klomek

A. B.

Marrocco

(2007). Bullying, depression, and suicidality in adolescents. Journal of the American Academy of Child and Adolescent Psychiatry, 46, 40-49.

25.

Lauritsen

J. L.

Heimer

Lynch

J. P.

(2009). Trends in the gender gap in violent offending: New evidence from the National Crime Victimization Survey. Criminology, 47, 361-400.

26.

Lauritsen

J. L.

Laub

J. H.

Sampson

R. J.

(1992). Conventional and delinquent activities: Implications for the prevention of violent victimization among adolescents. Violence and Victims, 7, 91-108.

27.

Lauritsen

J. L.

Quinet

K. F. D.

(1995). Repeat victimization among adolescents and young adults. Journal of Quantitative Criminology, 11, 143-166.

28.

Lazarus

R. S.

(1966). Psychological stress and the coping process. New York, NY: McGraw-Hill.

29.

Lloyd

Farrell

Pease

(1994). Preventing repeated domestic violence: A demonstration project on Merseyside (Police Research Group, Crime Prevention Unit Paper No. 49). London, England: Home Office.

30.

Lurigio

A. J.

(1987). Are all victims alike? The adverse, generalized, and differential impact of crime. Crime & Delinquency, 33, 452-467.

31.

Macmillan

(2000). Adolescent victimization and income deficits in adulthood: Rethinking the costs of criminal violence from a life course perspective. Criminology,38, 553-588.

32.

Macmillan

(2001). Violence and the life course: The consequences of victimization for personal and social development. Annual Review of Sociology, 27, 1-22.

33.

Maker

A. H.

Kemmelmeier

Peterson

(2001). Child sexual abuse, peer sexual abuse, and sexual assault in adulthood: A multi-risk model of revictimization. Journal of Trauma and Stress, 14, 351-368.

34.

Man

D. D.

Cronan

J. P.

(2002). Forecasting sexual abuse in prison: The prison subculture of masculinity as a backdrop for “deliberate indifference.” Journal of Criminal Law and Criminology, 92, 127-186.

35.

Manasse

M. E.

Ganem

N. M.

(2009). Victimization as a cause of delinquency: The role of depression and gender. Journal of Criminal Justice, 37, 371-378.

36.

Menaghan

E. G.

(1983). Individual coping efforts: Moderators of the relationship between life stress and mental health outcomes. In Kaplan

H. B.

(Ed.), Psychosocial stress: Trends in theory and research (pp. 157-191). New York, NY: Academic Press.

37.

Menard

(2002). Short- and long-term consequences of adolescent victimization. Washington, DC: U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention.

38.

Messman-Moore

T. L.

Long

P. J.

Siegfried

N. J.

(2000). The revictimization of child sexual abuse victims: An examination of the adjustment of college women with child sexual abuse, adult sexual assault, and adult physical abuse. Child Maltreatment,5, 18-27.

39.

Mirowsky

Ross

C. E.

(2003). Social causes of psychological distress. Hawthorne, NY: Aldine de Gruyter.

40.

Muthén

L. K.

Muthén

B. O.

(2007). Mplus: Statistical analysis with latent variables, User’s guide (5th ed.). Los Angeles, CA: Author.

41.

Noll

(2003). Revictimization and self-harm in females who experienced childhood sexual abuse: Results from a prospective study. Journal of Interpersonal Violence, 18, 1452-1471.

42.

Nurious

P. S.

Macy

R. J.

Bhuyan

Holt

V. L.

Kernic

M. A.

Rivara

F. P.

(2003). Contextualizing depression and physical functioning in battered women: Adding vulnerability and resources to the analysis. Journal of Interpersonal Violence, 18, 1411-1431.

43.

Orva

T. A.

McLeod

P. J.

Sharpe

(1996). Perceptions of control, depressive symptomology, and self-esteem of women in transition from abusive relationships. Journal of Family Violence, 11,167-186.

44.

Pearlin

L. I.

(1999). The stress process revisited: Reflections on concepts and their interrelationships. In Aneshensel

C. S.

Phelan

J. C.

(Eds.), Handbook of the sociology of mental health (pp. 395-415). New York, NY: Kluwer Academic.

45.

Pearlin

L. I.

Schieman

Fazio

E. M.

Meersman

S. C.

(2005). Stress, health, and the life course: Some conceptual perspectives. Journal of Health and Social Behavior, 46, 205-219.

46.

Pease

(1998). Repeat victimization: Taking stock (Crime Detection and Prevention Series, Paper No. 90). London, England: Police Research Group, Home Office.

47.

Polvi

Looman

Humphries

Pease

(1990). Repeat break and enter victimization: Time course and crime prevention opportunity. Journal of Police Science and Administration, 17, 8-11.

48.

Radloff

L. S.

(1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385-401.

49.

Rand

M. R.

(2009). Criminal victimization, 2008. Washington, DC: U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics.

50.

Ruback

R. B.

Thompson

M. P.

(2001). Social and psychological consequences of violent victimization. Thousand Oaks, CA: Sage.

51.

Schreck

C. J.

Stewart

E. A.

Fisher

B. S.

(2006). Self-control, victimization, and their influences on risky activities and delinquent friends: A longitudinal analysis using panel data. Journal of Quantitative Criminology, 46, 871-906.

52.

Sorenson

S. B.

Golding

J. M.

(1990). Depressive sequelae of recent criminal victimization. Journal of Traumatic Stress, 3, 337-350.

53.

Swan

S. C.

Gambone

L. J.

Fields

A. M.

Sullivan

T. P.

Snow

D. L.

(2005). Women who use violence in intimate relationships: The role of anger, victimization, and symptoms of PTSD and depression. Violence and Victims, 20, 267-285.

54.

Wheaton

(1994). Sampling the stress universe. In Avison

W. R

Gottlieb

I. H.

(Eds.), Stress and mental health (pp. 77-114). New York, NY: Plenum Press.

55.

Widom

C. S.

(1989). The cycle of violence. Science, 244, 160-166.

56.

S. C.

M. N.

(2007). Comparison of internet-based and paper-based questionnaires in Taiwan using multi-sample invariance approach. Cyberpsychology and Behavior, 10, 501-507.

57.

Zhao

Ford

E. S.

Dhingra

Shrine

T. W.

Mokdad

A. H.

(2009). Depression and anxiety among U.S. adults: Associations with body mass index. International Journal of Obesity, 33, 257-266.