INTIMATE PARTNER VIOLENCE AND DEPRESSIVE SYMPTOMS DURING ADOLESCENCE AND YOUNG ADULTHOOD

Abstract

Using longitudinal data from the Toledo Adolescent Relationships Study (TARS), we examine the relationship between intimate partner violence (IPV) and depressive symptoms during adolescence and young adulthood (N = 1, 273) while controlling for time-stable and time-varying correlates. Results show temporal changes in depressive symptoms, such that increases in depressive symptoms correspond to IPV exposure. While prior work has theorized that certain populations may be at increased psychological vulnerability from IPV, results indicate that both perpetration and victimization are associated with increases in depressive symptoms for both men and women and irrespective of whether IPV exposure occurred in adolescence or young adulthood. Cumulative exposure to IPV does not appear to increase depressive symptoms beyond the effect observed for the most recent IPV exposure, but physical maltreatment by a parent does appear to diminish the effect of IPV perpetration on depressive symptoms for a small subset of the sample.

Intimate partner violence (IPV) is a serious public health issue affecting millions of women and men in the United States (Straus and Gelles 1990; Tjaden and Thoennes 2000). IPV victimization results in a host of deleterious outcomes including poor mental health (Fergusson, Horwood and Ridder 2005), physical injury (Tjaden and Thoennes 2000), and early mortality (Catalano et al. 2009). In this study, we examine self-reports of both IPV victimization and perpetration, and consider whether changes in IPV exposure are related to increases in depressive symptoms. We consider the individual’s role in the violence (e.g., whether violence is mutual or experienced only as a victim), and earlier victimization (early family and peer victimization, prior IPV) as moderating influences on the IPV-depressive symptoms association.

Much prior work examining IPV and depressive symptoms is cross-sectional (e.g., Banyard and Cross 2008; Holt and Espelage 2005; O’Campo et al. 2006). Such studies have identified potentially confounding factors in the intimate partner violence-depressive symptoms relationship including prior coercive parenting, socioeconomic status while growing up and preexisting pathology (Fergusson et al. 2005). While cross-sectional studies controlling for such factors are useful for addressing issues of selection, they do not account for underlying predispositions toward depressive symptoms, nor contextual changes that affect variations in depressive symptoms over time. Thus, questions remain regarding the degree to which IPV influences depressive symptoms, or whether such associations reflect individual characteristics that account for both IPV and a propensity to experience depressive symptoms.

Using longitudinal survey data from the Toledo Adolescent Relationships Study (TARS), we assess how changes in self-reports of intimate partner violence victimization and perpetration correspond to changes in depressive symptoms from adolescence to early adulthood, while controlling for other known correlates associated with both IPV and depressive symptoms. We subsequently account for the form of involvement (mutual, perpetration only, or victimization only) in assessing changes in depressive symptoms. We investigate the following: (1) the influence of IPV (including the form of IPV) on depressive symptoms over time and across relationships; (2) the influence of IPV on depressive symptoms by gender and age; and (3) the moderating influence of prior victimization on the IPV-depressive symptoms relationship.

Background

Intimate Partner Violence and Depressive Symptoms Over Time

From adolescence through young adulthood individuals increasingly are involved in intimate relationships (Scott et al. 2011). This involvement increases risk for intimate partner violence. Examining data from the National Longitudinal Study of Adolescent Health (Add Health), Halpern et al. (2009) found that 8% of respondents reported IPV victimization in adolescence and by young adulthood 22% of men and 27% of women did so. While intimate relationships during young adulthood, on average, are of longer duration than those in adolescence (Furman and Shaffer 2003), a great deal of relationship turnover also occurs (Arnett 2004). Thus, it is important to model change in depressive symptoms across adolescence and young adulthood while simultaneously accounting for changes in exposure to IPV. Prior work, for example, shows that childhood depressive symptoms often precede depressive symptoms during adolescence and young adulthood (DuBois et al. 1995; Emslie 2012). Conversely, some evidence points to change in depressive symptoms over time (Galambos, Baker and Krahn 2006; Garber, Keiley, and Martin 2002). Other work provides evidence for both stability and change in depressive symptoms (Wickrama et al. 2008). Although useful for understanding stability and change in depressive symptoms, these studies have not considered a potentially important source of depressive symptoms, IPV.

Longitudinal studies that have examined the link between IPV and depressive symptoms generally emphasize victimization, including whether current IPV victimization leads to depression after controlling for prior victimization and preexisting pathology. Examining the first and third interviews of the Add Health, Fletcher (2010) finds that while both current and prior IPV increase depressive symptoms, such effects are stronger for current victimization (see also Bonomi et al. 2006). These results suggest that while prior victimization may cast lingering effects, more recent victimization has a stronger effect on current depressive symptoms. Yet this research does not examine how changes to exposure to IPV may correspond to changes in depressive symptoms over time. Changes in exposure are especially important during the young adult years when relationships are more fluid.

In addition to victimization, there are reasons to expect that perpetration may increase depressive symptoms. First, any IPV likely reflects low relationship quality (Lawrence and Bradbury 2007), which in itself is related to decreased emotional well-being (Umberson et al. 2006; Williams 2003). Second, the perpetrator may understand the use of violence as a failure experience. Although some theoretical conceptualizations link perpetration with the enforcement of dominance and control (Dutton and Goodman2005; Komter 1989), the use of violence may be understood as an attempt to gain control rather than emanating from a position of power (Stets and Burke 2005). Thus, IPV perpetration may reflect frustrated control attempts. Furthermore, negative reactions may further reinforce the notion of IPV involvement as a personal failure with associated declines in emotional well-being. Studies show that while some forms of violence are seen as appropriate or even status enhancing, survey items asking about the legitimacy of hitting a partner receive low endorsement (Markowitz 2001; Simon et al. 2010). Consequently, perpetrators may lack either the psychological resources or social support that protect against demoralization and guilt that likely follow violent encounters.

We hypothesize that changes (that is, within individual-variations) in depressive symptoms will be associated with changes in IPV experiences. Consistent with cross-sectional and longitudinal studies, we hypothesize that IPV victimization will be associated with increases in depressive symptoms. Additionally, for the reasons outlined, we hypothesize that IPV perpetration will also be positively related to changes in depression.

Scholars have recently paid increasing attention to mutual violence rather than focus on just victims or perpetrators. Studies using community-based samples consistently find perpetration and victimization experiences to be highly correlated (Archer 2000; Caetano et al. 2008; Tillyer and Wright, forthcoming). Using data from The Family Life Project, Gustafsson and Cox (2012) find that among the 37% of respondents reporting IPV, approximately 70% reported the violence as mutual. Melander et al. (2010) using nationally representative data from Wave III of Add Health, find that among the 25% of respondents reporting IPV, more than half report the violence as mutual (13%) compared to perpetration only (7%) and victimization only (5%). Furthermore, Whitaker et al. (2007) find that injury rates were highest among those reporting mutual IPV. “In fact, men in relationships with reciprocal violence, were reportedly injured more often (25.2%) than were women in relationships with nonreciprocal violence (20.0%); this is important as violence perpetrated by women is often seen as not serious” (Whitaker et al. 2007: 945). Such findings do not indicate that there is ‘gender symmetry’ (Dobash et al. 1992) in the experience of IPV, but suggest the possibility that some incidents of reciprocal or female-only violence may not be defined by couples as trivial or unimportant.

IPV that is reciprocal in nature may reflect an overarching interactive pattern characterized by negativity and conflict (Anderson, Umberson, and Elliott 2004). Thus, mutual violence may operate as an indicator of a relationship that is troubled overall and represent a source of chronic stress. Furthermore, individuals in relationships with reciprocal IPV experience the stress associated with being both a victim and a perpetrator. For these reasons, mutual violence may be particularly detrimental to emotional well-being and correspond to increases in depressive symptoms over time.

Effects of IPV on Depressive Symptoms by Gender

Prior work focuses largely on women’s victimization and their depressive symptoms (Beydoun et al. 2012; Bonomi et al. 2006; Foster, Hagan, and Brooks-Gunn 2008; O’Campo et al. 2006). Such emphasis is understandable given that in general, women report higher levels of depressive symptoms (Nolen-Hoeksema, Larson and Grayson 1999) and are more vulnerable to physical injury (Straus 2008). Studies using survey-based samples, however, find rates of female perpetration are significant, suggesting the importance of examining whether men’s emotional well-being is influenced by victimization (Hines and Malley-Morrison 2001). Among Wave III Add Health respondents reporting nonreciprocal violence, both women (67.7%) and men (74.9%) more often identified women as the perpetrator (70.7%) (Whitaker et al. 2007). Tillyer and Wright (2013) found similar patterns using Wave IV data from Add Health. Recent work examining patterns of dating violence among a school-based sample in New York, New Jersey and Pennsylvania, found boys reported higher rates of physical victimization compared to girls across categories of mild, moderate and severe physical violence (Zweig et al. 2013). This is consistent with Archer’s (2000) meta-analysis revealing rates of female perpetration that were higher in younger compared to older samples. These findings demonstrate that IPV exposure among boys and young men is more than minimal. Thus, attention to both women and men’s outcomes is worthy of exploration in examining IPV and depressive symptoms as individuals transition from adolescence to young adulthood.

Prior work has theorized that adolescent girls’ greater reliance on relationships for connection and identity formation (Miller 1990) render them more vulnerable to IPV (e.g., Callahan, Tolman and Saunders 2003). Since adolescent boys are socialized to be less reliant on relationships, IPV should be less detrimental to their sense of well-being. Yet recent work examining romantic relationships finds that adolescent boys report similar levels of intimate engagement as their female counterparts (Giordano, Long more and Manning 2006). Furthermore, aggression among boys and young men may be subject to greater sanctions both formal (Hamilton and Worthen 2011) and informal (Akers et al. 2011). Consequently, male youths may not be as immune to psychological distress associated with IPV victimization or perpetration as previously suggested during adolescence and early adulthood. We hypothesize that IPV will demonstrate a positive relationship with changes in depressive symptoms for male youths, but this relationship will be stronger for female youths.

Early Victimization and Depressive Symptoms

Child abuse and exposure to community violence are often linked to IPV victimization, and perpetration (Taylor et al. 2008). Coercive parenting practices, in addition to providing models of behavior similar to those received through witnessing parental physical violence (Kinsfogel and Grych 2004), are theorized to produce negative affect that prompts aggressive behavior in children (Kim et al. 2001). Studies examining these mechanisms in tandem find that coercive parenting exerts stronger effects on depressive symptoms relative to those of interparental violence (Ulman and Straus 2003). Prior work theorizes that coercive parenting affects depressive symptoms indirectly by undermining parents’ ability to effectively teach children active problem-solving strategies that safeguard against depression (Nolen-Hoeksema 1991). Similarly, youths who experience neighborhood or school violence report higher levels of depression (Turner et al. forthcoming).

Less well understood is how victimization background influences the negative sequelae of intimate partner violence. A background of early abuse may reflect an accumulation of adversity, further reinforcing the negative affect (e.g. depressive symptoms) associated with IPV. Alternatively, an early background surrounded by adversity in effect “normalizes” violence and victimization such that the individual may be less influenced by intimate partner violence. A similar line of reasoning may hold for prior IPV victimization as well (see e.g., Bonomi et al 2006; Fletcher 2010). We examine these competing hypotheses about the role of prior victimization experiences (parent and peer victimization, prior IPV) as a moderating influence of proximal IPV experiences on depressive symptoms.

The Current Study

Using growth curve analyses we assess whether changes in IPV exposure correspond to changes in individual levels of depressive symptoms. This study builds on previous studies in three key ways. First, distinguishing between-person from within-person change permits us to address unobserved heterogeneity within our sample that could explain selection into either IPV or an increasing trajectory of depressive symptoms. This effort is further enhanced by controlling for both time-stable characteristics, and a number of time-varying factors that likely influence depressive symptoms and selection into relationships characterized by violence. Second, we consider the influence of gender on the nature of the IPV-depressive symptoms relationship. As described above, we expect IPV to demonstrate a significant association with depressive symptoms for both male and female youths. However, the relationship may be stronger for girls and young women. Finally, we explore the potential role of prior victimization (parent and peer victimization, prior IPV) to assess whether these earlier experiences accumulate and lead to higher levels of depressive symptoms.

Method

Data

The TARS sample (n = 1,321) was drawn from the year 2000 enrollment records of all seventh, ninth, and eleventh graders in Lucas County, Ohio. The initial sample universe encompassed records elicited from 62 schools across seven school districts. All schools complied with data requests, as this information is legally available under Ohio’s Freedom of Information Act. The sampling frame consists of 15,188 eligible students, and is divided into 18 strata by grade, race/ethnicity (non-Hispanic other, non-Hispanic black, and Hispanic), and gender. Random subsamples were selected from each strata to achieve a total sample of 2,273 students. Of these students, we contacted 1,625, with 304 refusals, resulting in a total sample of 1,321, or 81.3% of the original 1,625 students who were contacted. The stratified, random sample, devised by the National Opinion Research Center, oversampled Black and Hispanic adolescents. Unlike school-based studies, school attendance was not required for sample inclusion. We conducted interviews in respondents’ homes using preloaded laptops to maintain privacy. At the time of the first interview, a separate questionnaire was administered to parents that included items related to the family’s socioeconomic status, including mothers’ educational attainment.

The longitudinal design of TARS is an asset enabling us to draw on interviews 1 through 4 (2001, 2002/2003, 2004/2005, 2006/2007) to assess changes in depressive symptoms from adolescence to young adulthood. Respondents’ ages range from 12 to 19 at the time of the first interview collected in 2001, to 17 to 24 at interview 4 collected in 2006/2007. Retention rates from the first interview were 89.1% for the second interview, 84.4% for the third interview, and 82.8% for the fourth interview. The data reflect a 6-year accelerated longitudinal design across four periods with three overlapping cohorts, allowing assessment of developmental patterns across the ages 12 to 24.

Regarding missing data, an advantage of growth curve analysis is that it allows us to include respondents with some missing data on the within-subject measures. Respondents with missing data on the between-subject variables are deleted (n= 22). Additionally, we deleted respondents who reported their race/ethnicity as other than White, Black, or Hispanic since there are too few to be analyzed (n = 26). The final analytic sample constitutes 1,273 respondents and 4,363 person-period observations with sample sizes that range from 1,054 to 1,122 respondents across the four interviews.¹ Seventy-eight percent of respondents participated in every data collection interview. We conducted t-tests comparing depressive symptom scores at interview 1 and those missing at any interview to those who participated and found no significant differences. An examination of correlations between attrition rates and our key variables (IPV victimization and perpetration, and non-IPV victimization), as well as on basic demographics (age, gender and race/ethnicity), reveals that respondents with higher participation rates are slightly younger and less likely to be Black. Neither our focal variables, gender, nor Hispanic ethnicity are significantly correlated with participation rates.

Measures

The dependent variable, depressive symptoms, is measured across interviews 1 through 4 using a seven-item version of the Center for Epidemiological Studies’ depressive symptoms scale (CES-D) (Radloff 1974). Respondents are asked how often each of the following statements was true during the past seven days: (1) “you felt you just couldn’t get going;” (2) “you felt that you could not shake off the blues;” (3) “you had trouble keeping your mind on what you were doing;” (4) “you felt lonely;” (5) “you felt sad;” (6) “you had trouble getting to sleep or staying asleep;” and (7) “you felt that everything was an effort.”Responsesare 1 (never) to 8 (every day). This is a mean scale of the seven items with alpha reliability scores ranging from .83 to .84. Due to skewness, we use the logarithm of the scale.

Within-Subject Predictors

All of our within-subject predictors are assessed using data from all four interviews. Intimate Partner Violence - IPV victimization assesses any victimization (victimization only and mutual). Using items from the Conflict Tactics Scale (Straus and Gelles 1990) respondents are asked how often their current or most recent partner has done the following: “thrown something at you;” “pushed, shoved or grabbed you;” “slapped you in the face or head with an open hand;” and “hit you.” Responses were 1 (never), 2 (hardly ever), 3 (sometimes), 4 (often), and5 (very often). Those responding “never” to all items are coded as 0, and others are coded as 1. Alpha reliability scores range from .89 to .91. To assess IPV perpetration we ask how often respondents have committed these acts against their current or most recent partner. We code any perpetration as 1 and 0 if the respondent reported “never” having committed any of these acts. Alpha reliability scores range from .89 to .92. To assess the differential influence of IPV by type, we create dichotomous variables that distinguish between respondents reportingno violence, victimization only, perpetration only, and mutual violence.

We measure IPV and depressive symptoms concurrently, rather than using a lagged measure of IPV. With IPV assessed for any time within the current or most recent relationship, however, and the reference period for depressive symptoms being the previous week, we infer that IPV likely occurred prior to the experience of depressive symptoms, thereby establishing the time order. Additionally, it is possible that some respondents reported IPV at later interviews that actually occurred in previous interviews if they were with the same partner, although only a small portion of our sample report having the same partner at two or three interviews (n = 82). No respondents report the same partner at all four interviews. Supplemental analyses (available on request) that exclude these respondents yield substantively similar results as those presented in the current study.

Romantic Context

We include two measures of the romantic context. Relationship type assesses whether the current or most recent relationship reported by respondents is a dating, cohabiting or marital relationship. Single respondents are used as the referent. Current relationship is a dichotomous variable measured at each interview and coded as 1 for those in a current relationship (52%), and 0 for those reporting on their most recent relationship.

Time Varying Controls

We include five variables that capture changes as youth transition from adolescence to adulthood. Gainful activity is coded as 1 (92%) if the respondent is attending school or employed full-time. Respondents who are not engaged in either activity are coded as 0.

In the absence of a standardized and widely used scale, we measure social support from peers using the mean of a three-item scale that reflects similar measures of social support received from parents (Hirschi, 1969). We ask respondents the following question: “How much do you agree or disagree with the following things about your friends?” Items were (1) “I can tell them private things and know they won't tell other people;” (2) “They care about me;” and (3) “My friends make me feel good about myself.” Responses range from 1 (strongly disagree) to 5 (strongly agree). Alpha reliability scores range from .70 to .81.

Receipt of public assistance, is coded 1 if the respondent reports familial receipt of public assistance including TANF, WIC, public housing, and supplemental security income (SSI) and 0 otherwise.

Delinquency is measured using nine items adapted from the 26-item inventory by Elliot and Ageton (1980). Items assess how frequently respondents engage in various antisocial behaviors including drug use, theft (minor and major), breaking and entering, assault and battery, property damage, selling drugs, public drunkenness, and carrying a hidden weapon. The responses for each item range from 0 (never), to 8 (more than once a day). Alpha scores range from .74 to .88.

Self-esteem is assessed using the mean of Rosenberg’s (1995) six-item self-esteem scale. Respondents are asked how much they agree with the following items: (1) “I am able to do things as well as other people;” (2) “I feel that I have a number of good qualities;” (3) “I feel I do not have much to be proud of” (reverse coded); (4) “At times I think I am no good at all” (reverse coded); (5) “I feel that I am a person of worth, at least on an equal basis with others;” and (6) “I take a positive attitude toward myself.” Responses range from 1 (strongly disagree) to 5 (strongly agree). Alpha reliability scores range from .71 to .77.

Between-Subject Predictors

We create multiple measures to assess the potential cumulative influence of IPV. Number of victimizations is the total number of interviews in which the respondent reported IPV victimization, which ranges from 0 (none) to 4 for victimization reports at every interview. Number of perpetrations assesses perpetration across the four interviews. Alternatively, a series of dichotomous variables assesses whether the respondent reports no victimization, one victimization, or multiple victimizations. We create corresponding measures for perpetration.

Non-IPV Victimization

We include two measures that assess non-IPV victimization. Physical maltreatment by a parent is measured from an item that asks respondents: “When you and your parents disagree about things, how often do they do the following:” “push, slap, or hit you?” Responses range from 1 (never) to 6 (two or more times a week). Physical assault by peers is assessed by asking respondents how often a friend had done the following: “thrown something at you;” “pushed, shoved or grabbed you;” “slapped you in the face or head with an open hand;” and “hit you.” Respondents who answer “never” are coded as 0, while respondents who report any victimization at the hands of a peer are coded as 1. Alpha reliability is .84.

Demographic Characteristics

Gender is a dichotomous variable with male as the reference. Race and ethnicity is composed of White non-Hispanic, Black non-Hispanic, and Hispanic. We create dichotomous variables for each race/ethnic category, with White as the reference. Mother’s education is assessed with the following question from the parent questionnaire: “How far did you go in school?” If the father answered the questionnaire and was married or cohabiting, we ask: “How far did your partner go in school?” Classifications are “less than 12 years,” “high school graduate or equivalency,” and “more than 12 years.” Family structure is composed of four dichotomous variables indicating the household type in which the adolescent reported living at the time of the first interview. Classifications are “two-biological-parent,” “single parent,” “step-parent” and “other” with “two biological-parent” households as the reference.

Analytic Strategy

As previously discussed, prior research efforts suggest that the current romantic context is most salient when thinking about the link between IPV and depressive symptoms. Within our current analyses, we do not presume to untangle the issues related to causal ordering. Rather our primary aim is to undertake an examination of how changes in IPV correspond to changes in depressive symptomatology. However, to the extent that we can address issues of unobserved heterogeneity that may inform selection into either IPV or depressive symptoms we choose to do so. To that end, we employ growth-curve modeling coupled with a group-mean centering approach that allows each individual to serve as his or her own control. Thus, any effect of IPV on depressive symptoms over time reflects change in each individual’s own propensity towards depressive symptomatology.

We begin by assessing how best to model growth in our sample by running an unconditional growth model that includes the intercept and age as random effects. The coefficient for age demonstrates a modest effect (b = .005, p = .048); however the variance component is significant (b = .115, p < .001) indicating that while the average trajectory for the sample as a whole appears to be flat, individual trajectories demonstrate variability. Prior work generally documents increases in depressive symptoms during early and middle adolescence (Garber, Keiley, and Martin 2002; Ge et al. 1994). In contrast, late adolescence and young adulthood are periods when depressive symptoms generally level off or decline (Galambos et al. 2006; Wickrama et al. 2008). There are several approaches to modeling growth patterns reflecting these kinds of discontinuities. One possibility is to operationalize age as a polynomial function of time. However, a disadvantage of this operationalization is the increased complexity and accompanying difficulty in interpretation (Singer and Willett 2003). Another possibility is to use a spline function to model more than one slope (i.e., one slope for adolescence, and another for young adulthood). This approach has the advantage of capturing discontinuities in growth, while maintaining simplicity in interpretation (Fitzmaurice, Laird, and Ware 2004). Comparison of the polynomial trend and spline models revealed little difference between the two regarding their fit to our data². Given its simplicity, we elect to use the spline model that includes an intercept (initial level) and two slope segments – one for early to middle adolescence (ages 12 – 16) and one for late adolescence to young adulthood (ages 17 – 24).

While the age term for adolescence demonstrates sufficient variance to include as a random effect, the young adult age term does not. Thus, our unconditional growth model includes a random intercept and a random effect for the adolescent slope, and fixed effects for the intercept, and both adolescent and young adult slopes. This unconditional growth model reveals a mean initial level of depressive symptoms of .688 (p < .001), and significant results for both slope coefficients. The significant coefficient (b = .019, p < .001) for the adolescent slope indicates that depressive symptoms increase with age at a rate that is significantly different from zero. The significant negative coefficient (b = −.023, p < .001) for the young adult slope indicates that the change in slope beginning at age 17 is significantly different from the adolescent slope. The actual slope for young adults is the sum of these two coefficients (.019 – .023 = .004) and an F-test reveals this to not be significantly different from zero. These results indicate that while depressive symptoms increase during early to middle adolescence, beginning in late adolescence (age 17), depressive symptoms level off such that they do not appear to significantly change during young adulthood.

We examine the influence of IPV victimization and perpetration, and IPV by type(victimization, perpetration, mutual, versus none) on changes in depressive symptoms using a two-level model that nests 4,363 observations in our sample of 1,273 individuals over the four time periods. Since a primary aim is to differentiate between-person effects from within-person change, we follow the recommendation of Horney, Osgood, and Marshall (1995) by group-centering the values for time-varying covariates in the level-one equation. That is, we transformed responses for each of our time-varying covariates into deviations from each individual’s mean as calculated across all periods of observation. Additionally, in the level-two model, a person-level mean for each time-varying measure is included as a means of controlling for our other between-subject variables (Allison, 2005:36). However, given that the between-subject coefficients of our time-varying variables do not really contribute anything of any real substantive interest, and may in fact be confounded with the effects of other unobserved variables, we have chosen to eliminate them from the tables. The group-centering procedure allows individuals to serve as their own control, thereby increasing confidence that between-person differences are not contaminating the assessment of within-individual change (Singer and Willett 2003; Zhang, Zyphur, and Preacher 2009). While our modeling approach of parsing between- and within-subject variation helps to address the problems of unobserved heterogeneity, this does not preclude the need to control for additional time-varying factors that could potentially influence selection (Bjerk 2009) into either IPV or increased depressive symptoms. Thus, we include time-varying measures of several factors that prior work has identified as potential risks for IPV and depressive symptoms. These include gainful activity, social support from peers, receipt of public assistance, adolescent delinquency and self-esteem.

We assess the relationship between IPV victimization and depressive symptoms, followed by an examination of IPV perpetration and depressive symptoms. We then examine whether IPV victimization or perpetration interact with age, gender, or non-IPV victimization experiences (i.e., parental or peer victimization). Next, we assess whether cumulative victimization or perpetration are significantly associated with depressive symptoms by considering both additive and multiplicative models. Finally, we explore whether the relationship between IPV and depressive symptoms varies depending on whether the individual experiences victimization only, perpetration only, or mutual violence.

Results

Descriptive Results

Table 1 provides the distributions of depressive symptoms for each of the four interviews. Depressive symptoms appear stable over the time period. However, partner victimization experience increases by 50% over time from 21.3% during adolescence (first interview) to 32.2% when respondents reach young adulthood (fourth interview). Similarly, violence perpetration increases by nearly 40% from 18.1% to 25.2%. Examining IPV by type reveals that the largest proportion of those who experienced IPV reported mutual violence. Thus, at the time of the fourth interview among the 36% who report some form of IPV, 46% report mutual violence, compared to 36% who report only victimization, and 18% who report perpetration only. Intimate partner violence demonstrates fluidity over time with few respondents reporting perpetrating violence (.5% of male and 1.5% of female respondents) or being victimized (2.4% of male and .8% of female respondents) at all four interviews. Rather, it is common for those who ever experienced intimate partner violence to report violence in only one relationship suggesting the utility of considering changes in the relationship context and using a time-varying measure of partner violence for examining long-term influence on depressive symptoms. Approximately 23% report victimization by a parent during adolescence, and 43% report victimization by peers. Regarding gender, 52% of respondents are female and 48% male.

Table 1.

Descriptive Statistics by Wave, Means/Proportions and Standard Deviations (N = 1,273 subjects, 4,363 observations).

	Wave 1	Wave 2	Wave 3	Wave 4
	Mean/Pct.	Mean/Pct.	Mean/Pct.	Mean/Pct.
Depressive symptoms	2.319 (1.142)	2.437 (1.274)	2.432 (1.241)	2.414 (1.275)
Depressive symptoms (log)	.744 (.443)	.780 (.468)	.778 (.458)	.768 (.476)
Intimate Partner Violence
Any IPV Victimization	.213	.223	.271	.322
Any IPV Perpetration	.181	.180	.209	.252
Intimate Partner Violence Type
None	.806	.810	.709	.637
Victimization only	.061	.063	.104	.130
Perpetration only	.037	.033	.049	.065
Mutual	.097	.094	.138	.168
Romantic Context
Relationship Type
Not dating	.263	.272	.100	.068
Dating	.737	.625	.741	.622
Cohabiting	.000	.000	.078	.162
Married	.000	.000	.011	.054
Current Relationship	.441	.584	.652	.482
Time-Varying Controls
Gainful activity	1.000	.772	.853	.770
Social support from peers	4.183 (.585)	4.124 (.565)	4.066 (.655)	4.024 (.651)
Receipt of public assistance	.372	.104	.155	.196
Delinquency	1.202 (.555)	1.264 (.554)	1.330 (.592)	1.383 (.619)
Self-esteem	3.944 (.611)	3.954 (.593)	3.988 (.606)	4.060 (.580)
Age	15.222 (1.730)	16.375 (1.761)	18.171 (1.768)	2.331 (1.786)
Non-IPV Victimization
Physical maltreatment by a parent	.226
Physical assault by peers	.430
Demographic Characteristics
Female	.518
Race/Ethnicity
White	.640
Black	.247
Hispanic	.113
Mother’s Education
Less than 12 years	.122
H.S. graduate	.354
More than 12 years	.524
Family Structure
Two biological parents	.494
Single parent	.260
Step-parent	.177
Other living arrangement	.069

Note: Numbers in parentheses are standard deviations.

Assessment of the Influence of Any IPV Victimization and Any IPV Perpetration

Table 2 presents the multilevel analyses examining the association between IPV victimization (model 1) and IPV perpetration (model 2) on depressive symptoms. Experiencing IPV victimization at any time increases depressive symptoms (b = .065, p < .001) indicating that the association is not due to underlying differences in depression between respondents. We find several other time-varying indicators affect depressive symptoms. Gainful activity is significant at the within-person level (b = −.079, p < .001) suggesting that employment or educational pursuits may help reduce levels of depressive symptoms. Similarly, increases in self-esteem further protects against depressive symptoms as well (b = − .128, p < .001). With respect to factors that positively influence depressive symptoms, receipt of public assistance is associated with increases in depressive symptoms (b = .033, p < .05). Similarly, youths who continue to engage in delinquent behavior see additional increases in depressive symptoms over time (b =.098, p < .001). The results for age remain substantively the same as those reported in the unconditional growth model.

Table 2.

Multilevel Models for the Influence of IPV Victimization and Perpetration on Depressive Symptoms (N = 1,273 subjects, 4,363 observations).

	Model 1 Any IPV Victimization		Model 2 Any IPV Perpetration
	Coefficient	SE	Coefficient	SE
Intercept	1.556***	(.126)	1.558***	(.125)
Time Varying Variablesa
Intimate Partner Violence	.065***	(.017)
Any IPV Victimization
Any IPV Perpetration			.047*	(.019)
Romantic Context
Relationship Type
Dating	.018	(.020)	.024	(.020)
Cohabiting	−.015	(.034)	−.011	(.035)
Married	−.020	(.055)	.027	(.055)
Current Relationship	−.012	(.016)	−.012	(.016)
Time-Varying Controls
Gainful activity	−.079***	(.018)	−.080***	(.018)
Social support from peers	−.006	(.012)	−.006	(.012)
Receipt of public assistance	.033*	(.016)	.032*	(.016)
Delinquency	.098***	(.014)	.100**	(.014)
Self-esteem	−.128***	(.013)	−.130***	(.013)
Age (12 – 16 yrs.)	.013*	(.006)	.013*	(.006)
Age (17 – 24 yrs.)	−.019*	(.008)	−.019*	(.008)
Time Stable Variables
Non-IPV Victimization
Physical maltreatment by a parent	.067**	(.022)	.062**	(.021)
Physical assault by peers	.051**	(.019)	.051**	(.019)
Demographic Characteristics
Female	.137***	(.020)	.116***	(.021)
Race/Ethnicity
Black	.023	(.024)	.020	(.024)
Hispanic	−.003	(.030)	−.004	(.030)
Mother’s Education
Less than 12 years	−.017	(.031)	−.026	(.030)
More than 12 years	.032	(.019)	.027	(.019)
Family Structure
Single parent	−.007	(.023)	−.008	(.023)
Step-parent	.051*	(.025)	.053*	(.025)
Other living arrangement	−.025	(.042)	−.018	(.042)
Likelihoodχ2		569.95***		559.85***

^*p < .05,

^**p < .01,

^***p < .001

Note: Referents are non-daters, white, high school graduate, two biological parents.

^aBetween-subject predictors using the individual means for each time-varying variable are included in the model, but not shown.

With respect to time-stable predictors that distinguish between trajectories, non-IPV victimization experiences of physical maltreatment by a parent (b = .067, p < .01) and physical assault by peers during adolescence (b = .051, p < .01) are positively associated with depressive symptoms. This suggests that prior victimization experiences exert unique effects on depressive symptoms beyond any indirect effect through increased risk of IPV involvement since we are controlling for between-subject effects of IPV involvement by including the individual means in the model. Consistent with prior research findings, female respondents report higher levels of depressive symptoms (b = .137, p < .001) as do those from step-parent families (b = .051, p < 05).

Next we examine the relationship of IPV perpetration on depressive symptoms. Model 2 shows that similar to victimization, IPV perpetration is positively related to depressive symptoms at the within-person level (b = .047, p < .05). Thus, IPV perpetration, like IPV victimization leads to a corresponding increase beyond the individual’s own average level of depressive symptoms. Results for the remaining covariates are substantively the same as those for IPV victimization.

We test a series of interactions with IPV victimization or IPV perpetration (results not shown). IPV victimization and perpetration operate similarly among both women and men on depressive symptoms. Thus, while women more often experience depressive symptoms relative to men, the influence of IPV on depressive symptoms is similar for men and women. Furthermore, these findings are robust whether we consider variations in victimization or perpetration. Similarly, an interaction with age is also not significant indicating that irrespective of whether IPV victimization or perpetration occurs during adolescence or young adulthood, IPV’s influence is similar in its positive effect on depressive symptoms. Finally, the influence of IPV on depressive symptoms does not vary based on prior victimization by parents or peers. This suggests that regardless of how the individual arrives at a relationship characterized by IPV, once involved in such a relationship, the influence of IPV on depressive symptoms is substantively the same.

We examine the relationship between cumulative IPV exposure and depressive symptoms by addressing both potentially additive and multiplicative properties. The additive indicator that includes the count of IPV (measured as victimization or perpetration) is not statistically significant once the time-varying measure of IPV is included in the model (results not shown). It is the current or most recent experience (victimization or perpetration) that influences within-person change in depressive symptoms rather than the number of IPV exposures. We also examine whether prior IPV exposure operates as an amplifier of current IPV. The interaction between first IPV exposure and subsequent IPV (whether victimization or perpetration) indicates that earlier exposure does not increase the negative relationship between current IPV and depressive symptoms (results not shown).

Assessment of the Influence of IPV by Type

The analyses demonstrate that IPV victimization and perpetration appear to operate similarly in terms of their association with individual change in depressive symptoms. However, a limitation of these models is that they do not differentiate those who only experience victimization or perpetration versus those individuals who are in relationships where both partners engage in violence. To accomplish this, we turn to models in Table 3 that examine IPV experience by type: no violence, victimization only, perpetration only, and mutual violence. Unfortunately, we cannot include an examination by gender as the majority of respondents in the perpetration only category (n = 200) are female (n = 182), and thus we do not have sufficient cell sizes. Consistent with the previous analyses we are able to examine age and early victimization by parents and peers as potential moderators.

Table 3.

Multilevel Models for the Influence of IPV and Physical maltreatment by a parent on Depressive Symptoms (N = 1,273 subjects, 4,363 observations).

	Model 1		Model 2
	Coefficient	SE	Coefficient	SE
Time Varying Variablesa
Intercept	1.556***	(.125)	1.547***	(.126)
Intimate Partner Violence
IPV Type
Mutual violence	.065**	(.021)	.082**	(.026)
Victimization only	.076***	(.023)	.097***	(.027)
Perpetration only	.051	(.032)	.098**	(.038)
IPV×physical maltreatment by a parent
Mutual violence			−.058	(.045)
Victimization only			−.082	(.052)
Perpetration only			−.160*	(.069)
Romantic Context
Relationship Type
Dating	.014	(.020)	.014	(.020)
Cohabiting	−.019	(.035)	−.021	(.035)
Married	.018	(.055)	−.015	(.055)
Current Relationship	−.012	(.016)	−.012	(.016)
Time-Varying Controls
Gainful activity	−.079***	(.018)	−.079***	(.018)
Social support from peers	−.005	(.012)	−.004	(.012)
Receipt of public assistance	.034*	(.016)	.035*	(.016)
Delinquency	.098***	(.014)	.099***	(.014)
Self-esteem	−.129***	(.013)	−.129***	(.013)
Age (12 – 16 yrs.)	.013*	(.006)	.013*	(.006)
Age (17 – 24 yrs.)	−.019*	(.008)	−.019*	(.008)
Time Stable Variables
Non-IPV Victimization
Physical maltreatment by a parent	.065**	(.021)	.085**	(.031)
Physical assault by peers	.048*	(.019)	.048*	(.019)
Demographic Characteristics
Female	.118***	(.021)	.118***	(.031)
Race/Ethnicity
Black	.018	(.024)	.017	(.024)
Hispanic	−.005	(.030)	−.005	(.030)
Mother’s Education
Less than 12 years	−.020	(.031)	−.019	(.030)
More than 12 years	.033	(.019)	.035	(.019)
Family Structure
Single parent	−.009	(.023)	−.009	(.023)
Step-parent	.052*	(.025)	.052*	(.025)
Other living arrangement	−.029	(.042)	−.027	(.042)
Likelihood χ2		563.12***		563.91***

^*p < .05,

^**p < .01,

^***p < .001

Note: Referents are no IPV, non-daters, white, high school graduate, two biological parents. Numbers in parentheses are standard errors.

^aBetween-subject predictors using the individual means for each time-varying variable are included in the model, but not shown.

We introduce our categorical measure of IPV in model 1. Examining within-individual changes in depressive symptoms, mutual violence (b = .065, p < .01) and IPV victimization (b = .076, p < .001) emerge as significant predictors. These findings suggest the possibility that victimization experiences may be more clearly associated with changes in depressive symptoms. However, it bears repeating that mutual violence remains the modal category reported. As such, these within-person results for mutual violence include respondents who report perpetration. Thus, caution should be taken with respect to interpreting these results. As in the prior analyses, IPV does not interact with age or physical assault by peers indicating that even after we decompose by type, the influence of IPV on depressive symptoms is similar for both adolescents and young adults and not conditional on prior victimization by peers. Model 2 presents the results between physical maltreatment by a parent and the type of IPV experiences reported. The results indicate that physical maltreatment by a parent reduces the coefficient for IPV perpetration on depressive symptoms (b = −.160, p < .05). Specifically, for those individuals who report physical maltreatment by a parent, IPV perpetration is not associated with individual change in depressive symptoms³. However, this very small subset (n = 56) represents only a quarter of the perpetrators only group (56/200 = .28) and less than 10% of all perpetrators when including the mutual violence group (56/746 = .08). Among those individuals who lack a family history of violence, perpetrating IPV produces an increase in depressive symptoms consistent with those reporting mutual violence. Thus, our results demonstrate a deleterious relationship between IPV and depressive symptoms for most of the sample irrespective of whether the respondent reports victimization, perpetration, or mutual violence.

Discussion

Using longitudinal data we examine whether changes in intimate partner violence exposure correspond to changes in individual level of depressive symptoms across the period from adolescence to young adulthood. Relying on interviews conducted with a contemporary, diverse sample, we find that consistent with our hypotheses, IPV victimization and perpetration are associated with increases in depressive symptoms. Furthermore, these results were present for young men, as well as for young women. However, the relationship between IPV and depressive symptoms does not appear to be stronger for female respondents relative to their male counterparts, even though overall young women score higher on self-reports of depressive symptoms. These findings are important in documenting that young men are not immune to negative psychological outcomes associated with IPV victimization or perpetration, even though, as prior research demonstrates, young women are more vulnerable to physical injuries (Whitaker, Haileyesus, Swahn and Saltzman 2007). It is possible that other indicators of well-being could capture gendered relationships that are not apparent using a straightforward index of depressive symptoms.

Our work is consistent with prior research emphasizing that victimization has negative implications for emotional well-being. We expand on these findings by showing that perpetration is also related to depressive symptoms. This finding is potentially important as victimization has a more intuitive, straightforward relationship to declines in mental health. We argue that perpetration may be significantly related to depressive symptoms because it is: a) a marker of involvement in an intimate relationship characterized by extensive conflict and other negative dynamics; b) viewed by the individual as a failure experience (inability to handle conflict constructively); and c) seen by the partner and others as inappropriate and harmful, even if the actions have not risen to the level of official intervention.

Our longitudinal data reveal the fluid nature of IPV. The vast majority of respondents do not remain in one category of perpetration, victimization, or mutual violence across interview waves. For example, 21% of the individuals who appear as perpetrator or victim only at wave 3 report mutual victimization with the same partner at wave 4 (not shown). Further, mutual violence represents the modal response pattern in this non-clinical sample. Nevertheless, analyses provide some support for the notion that subgroups that include perpetration (perpetrator only or mutual) reflect between-individual differences in levels of depression while groups that include any victimization (mutual or victim-only) are more sensitive to within-individual changes.

We also examine the moderating influence of early victimization by parents and peers. Although moderation effects are not apparent in analyses examining overall variations relying on the basic perpetration and victimization scales, an effect is identified in analyses that take into account the specific form of IPV (whether mutual, victimization only or perpetration only). Findings reveal that among youths reporting parental physical violence during adolescence, IPV perpetration (absent victimization) appears to be less psychologically distressing. In contrast, youths without this early familial history (76% of the sample) were vulnerable to increases in depressive symptoms following IPV perpetration. Given that the composition of the “perpetrator only” group is predominantly female, we are unable to assess gender differences by IPV type (victim only, perpetrator only, mutual violence). Nonetheless, our findings have implications for future theorizing about gender, IPV, and outcomes associated with IPV. Prior work theorizes that IPV perpetration may be particularly distressing for adolescent girls and women as it represents a departure from feminine norms of behavior (Caetano and Cunradi 2003). Yet it will be important in future work to further explore contextual variations in the lives of adolescent girls and women, and how IPV experiences (victimization and perpetration), influence outcomes of psychological well-being.

Finally, our analyses yield little support for the notion that IPV experiences demonstrate a cumulative effect on depressive symptoms. The accumulation of IPV does not appear to offer additional negative contributions to the relationship between IPV and depressive symptoms beyond those demonstrated by the current or most recent relationship. Additionally, prior IPV exposure does not amplify the relationship between IPV exposure on depressive symptoms. Consistent with research examining other stressors (e.g., Avison and Turner, 1988), these results suggest a recency effect, meaning that it is the most recent IPV exposure that is particularly salient when thinking about current levels of depressive symptoms.

While the current study represents an improvement over prior work by using a longitudinal design, and a time-varying measure of IPV, some additional limitations remain. The current data do not assess the severity of intimate partner violence and more importantly, sustained injuries. As noted by several scholars, while female respondents may engage in relationship violence as frequently or more frequently than their male counterparts, they are more likely to be injured (Archer 2000). It is possible that sustaining physical injury might operate as a moderator of the IPV and depressive symptoms relationship. Additionally, the current study only assesses a single dimension of psychological well-being, namely depressive symptoms. Differences by age, gender or race/ethnicity may exist between intimate partner violence and other dimensions of psychological well-being such as anxiety, PTSD and major depressive disorder. Furthermore, our focus on adolescents and young adulthood may not fully capture the negative effects associated with cumulative IPV. It is possible that examinations of these trajectories in middle- and late-adulthood would reveal distinct gendered patterns.

Few respondents report continual involvement in intimate partner violence across relationships. A more common pattern is for violence to be present in one or two relationships. Nevertheless, intimate partner violence appears to have a significant negative impact on the psychological well-being of young adults even after accounting for other risk factors and individuals’ own proclivity towards depressive symptoms. Psychological distress, including depressive symptoms, may undermine self-confidence and self-efficacy (Bandura 1989), thus compromising young people’s ability to comfortably navigate the transition to adult identities and roles. Consequently, the costs of intimate partner violence may be long term, and have additional implications for individuals’ choices associated with family formation and stability, as well as economic and educational attainment. Nevertheless, the findings also provide indications of the malleability of one's mental health suggesting some individuals may demonstrate resiliency in recovering from IPV.