Introduction

The 2004 National Institute of Justice report, Violence Against Women: Identifying Risk Factors, concludes that early interventions could go a long way toward preventing intimate partner violence. Recommendations include incorporating violence prevention into a range of existing programs and services for youth and adults (Office of Justice Programs, 2004). There are, however, unanswered questions about risk factors for engagement in partner violence that need to be answered to inform the content and timing of intervention programs for the large number of adolescents and young adults who are at potential risk (Hickman et al., 2004). Finding out whether young adolescents’ aggressive behaviors predict subsequent perpetration and victimization in relationships with intimate partners can inform public health approaches for addressing domestic and other forms of violence (Graffunder et al., 2004).

Using data from a 6-year longitudinal study situated in a distressed urban community, this study examines the relationship between boys’ and girls’ reports of peer-directed aggression during middle school and subsequent involvement as victims and perpetrators of partner physical violence. Physical partner violence accounts for the largest proportion of domestic abuse and its associated costs (National Center for Injury Prevention and Control, 2003). Establishing this empirical base is especially important for economically disadvantaged communities, where multiple types of violence, including fatal and non-fatal domestic violence, contribute to ongoing disparities in social, physical, and mental health (Briere and Jordan, 2004; West, 2004).

Youth aggression and partner violence

During middle school, peer-directed aggressive behaviors peak among both boys and girls and then decline from 9th through 12th grade. In the 2003 Youth Risk Behavior Survey (YRBS), for example, 39% of 9th graders report being in a recent fight, compared to 26% of 12th graders, and weapon carrying dropped slightly from 18% to 16% (Centers for Disease Control and Prevention, 2003). During the high school years, however, young people also become involved in dating and intimate partner relationships, and problems dealing with aggression in these relationships emerge. Partner violence is widespread, with the prevalence increasing rapidly between the ages of 15 and 25 years (O’Leary, 2000). Once initiated, aggressive patterns of interacting and resolving conflict with intimate partners can be very difficult to break (Ehrensaft et al., 2003).

Although controversial in its interpretation, data collected using the widely administered and adapted Conflict Tactics Scale show that young adult males and females in the United States often report similar levels of partner violence victimization and perpetration. Violence often begins in the dating relationships of adolescents and continues through courtships, longer-term relationships, and marriage, when more serious violence may occur (Archer, 2000a,b; Coker et al., 2002; Howard and Wang, 2003). In a national sample of young adults, Morse (1995) found that both males and females engaged in frequent minor assaults, but that young men were more likely to repeatedly beat their partner and to cause injury requiring medical treatment. This pattern continues throughout adulthood. Although women, in general, may strike as often, men hit harder, and women may underestimate the potential risk of being seriously injured or killed (Campbell, 2004; Schafer and Caetano, 1998). Women who engage in perpetration are at increased risk of being severely abused by partners, and often report that their use of force was primarily or exclusively in self-defense (Siegel and Williams, 2003). Clearly, what Archer has termed “aggression between heterosexual partners” is interactive and reciprocal (Frieze, 2000). There are strong associations for both males and females between victimization and perpetration of partner violence (Stith et al., 2004).

There are relatively few longitudinal studies that begin prior to adulthood and examine multiple forms of violence victimization and perpetration by males and females. In a cohort of New Zealand youth followed from birth through age 21, individuals with histories of early antisocial behavior problems were more likely to behave violently toward their partners and to be victims of violence (Woodward et al., 2002). In a 20-year prospective study on intergenerational transmission of partner violence, adolescent conduct disorders were a strong predictor of partner violence perpetration for both genders (Ehrensaft et al., 2003). Herrenkohl and colleagues also report associations between adolescent violence assessed at ages 15 to 18 and subsequent intimate partner violence (Herrenkohl et al., 2003). Behavioral under-control and aggression have been related to concurrent and subsequent relationship aggression among adults assessed at 25 and 29 years of age (Grekin, 2004; Siegel, 2000).

Other cross-sectional and shorter-term longitudinal research provides an inconsistent picture of the relationship of early aggression to partner violence. Using cross-sectional analyses from the National Study of Adolescent Health, Roberts (2003a) documents that aggressive behaviors are associated with reports of intimate partner abuse. In a study of the co-occurrence of peer violence and dating violence among European and Mexican American adolescents interviewed at 16–20 years of age and again one year later, males who engaged in peer violence perpetration were more likely to be at risk for dating violence and sexual aggression; this finding did not hold for females (Ozer, 2004). In annual surveys with 8th and 9th grade urban youth, having been in physical fights with a peer predicted onset of serious physical dating violence victimization for males, but not females (Foshee et al., 2004). In most research addressing predictors of partner violence, levels of early aggressive behaviors among females typically do not approach those reported by girls in the urban settings in which the current study was conducted. Thus, findings may not be generalizeable to female perpetration in contexts where early adolescent physical aggression is high among both genders.

Other risk factors for partner violence

As Widom and others point out, one of the most consistent findings for both genders is the relationship between childhood victimization, including verified child abuse and neglect, and subsequent involvement in domestic violence (Ehrensaft et al., 2003; Widom, 1989, 1994). This is in keeping with social learning theories that posit child maltreatment leads to an acceptance of violence in adult intimate relationships, and that patterns of aggression learned in the family of origin are exhibited first in relationships with peers in school, then reinforced and repeated in one's own intimate and family relationships (Ponce et al., 2004; Wolfe and Wekerle, 1997). In a sample of college students, those who were physically abused or who witnessed domestic violence in childhood were at greater risk for partner violence during high school, and those who were victimized during high school were at greater risk for victimization as young adults. Young men who assaulted partners had also experienced more family violence than other men (Koss et al., 1987). Here, too, there is contradictory evidence: in the above-cited study of 8th and 9th graders, reports of friends engaging in dating violence, but not inter-parental violence, predicted reports of one's own dating violence (Arriaga and Foshee, 2004). In other research, witnessing family violence has sometimes been associated with dating violence perpetration and adult partner violence for females but not males (Desai et al., 2002; Wolf and Foshee, 2003), although Kinsfogel and Grych (2004) found a relationship for males but not females.

In addition to peer aggression and domestic violence exposures, other risk factors have been associated with partner violence. For example, substance use during adolescence has been related to later partner abuse, although findings for males and females have been inconsistent. In one-year longitudinal analyses, Roberts (2003b) reports prior substance use is a risk for females but not males. By contrast, Foshee et al. (2004) find that substance use predicts victimization among males but not females. Other youth risk behaviors, including initiation of sexual intercourse prior to high school, have received less attention, but are associated with long-term patterns of behavior, including sexual risk taking and multiple partners, that may be related to dating and domestic violence (O’Donnell et al., 2001).

Finally, a range of sociodemographic characteristics have been linked with partner violence, including the chronic strains and stresses of living in poverty. Indeed, structural determinants such as economic disadvantage may account for elevated rates among some racial and ethnic groups, including African Americans (Rennison and Welchens, 2000). Potential protective factors include church affiliation, educational aspirations, and living in a two-parent household (Foshee et al., 2004; Maxwell et al., 2003). Because these factors are also related to youth aggression, it is important to take them into account in investigating the influence of adolescent peer violence on subsequent partner violence victimization and perpetration.

Study hypotheses

We use data from the Reach for Health longitudinal study of urban youth to test the following hypotheses: (1) Higher levels of 8th grade aggression predict subsequent partner violence victimization and perpetration among both males and females; (2) The relationship between 8th grade aggression and subsequent partner violence remains significant, controlling for social and demographic characteristics, including race/ethnicity, as well as potential protective factors such as living in a two-parent family, church attendance, and academic performance; and (3) 8th grade aggression remains an independent predictor of subsequent partner violence for both genders, controlling for potential confounders, including early substance use, sexual initiation, and Exposures to physical aggression in the childhood home.

Methods

The Reach for Health study

In 1994, the Reach for Health (RFH) study was initiated in Brooklyn, New York, as one of seven collaborating partners in a multi-site research agreement (RSVPP) supported by the Office of Minority Health and NICHD to explore strategies for promoting health and reducing risk in economically disadvantaged minority communities. All youth attending 7th or 8th grade at three participating public middle schools during the 1994–1995 and 1995–1996 school years were eligible for participation. With additional funding from NICHD and CDC, students who completed an 8th grade survey and remained in the city were eligible for a high school follow up. Through high school, information was collected on risks associated with sexual activity, substance use, and interpersonal youth violence. The young adult survey, conducted when participants were about 19–20 years of age (2002–2003), expanded our focus to include intimate partner violence. Recruitment and study procedures are described elsewhere (O’Donnell et al., 1997, 2004).

The RFH sample

Three middle schools provided access to a sample of urban young adolescents. Each is located in an economically disadvantaged area with high rates of teen pregnancy, HIV/STD infection, violence-related injuries, and other sources of morbidity. Over 80% of students were eligible for free lunch programs. At each survey administration, all students were invited to participate. Written parental permission and youth assent were obtained prior to study enrollment. Parental consent was provided for 89% of eligible students; completed baseline surveys were obtained from over 95% of those with parental permission. Tracking information obtained during the high school surveys was used to contact participants as young adults. All surveys have been administered as paper-and-pencil questionnaires. As participants have gotten older, surveys have been administered either in small groups or individually at locations where privacy could be assured, including former middle and high schools, as well as the study office and homes.

Of the 1478 participants who completed an 8th grade survey and were living in New York City when high-school tracking procedures were initiated, 87% completed a high school survey. These analyses include data from 977 respondents who completed an 8th grade survey and were resurveyed as young adults, approximately six years later. This represents 66% of 8th graders eligible for follow up and over 74% of those who participated in high school surveys. (With a new round of surveys to be initiated shortly, we anticipate maintaining contact with over 70% of the original sample through young adulthood.) Retention is similar for both years of enrollment.

As with any longitudinal study, it is important to consider the influence of attrition. The RFH study was somewhat less successful in obtaining parental permissions for males at study recruitment during middle school and, again, at high school. In addition, males have been more difficult to retain in the sample. To account for potential bias due to attrition, we examined predictors of attrition for the sample of males and females combined. Gender is the main predictor of study attrition, although other predictors were significant within gender (e.g., for males, perceived maturity, ethnicity, academic performance, substance use, and age; for females, school, household living arrangements, and age). Important for these analyses, 8th grade aggression predicted attrition for males but not females. We therefore split and weighted the sample within gender. We used logistic regression to create predicted probabilities of responding to the young adult survey derived from each respondent's 8th grade information. These weights, which are the inverse of the predicted probabilities, were divided by the average weights for males and females who completed the young adult survey, thus retaining the obtained sample sizes (McGuigan et al., 1997). Analyses using the weighted samples were conducted for each gender separately and results are provided by gender.

Measures

Aggressive behaviors in 8th grade were assessed by 5 items that asked about behavior in the past three months (90 days): (1) Did you tell someone you were going to beat them up, not including your brothers and sisters or other children you live with? (2) Were you in a physical fight (a fight with hitting, kicking, or pushing)? (3) Did you carry a knife or razor (including a box cutter)? (4) Did you carry a gun? (5) Did you tell someone you were going to cut, stab, or shoot them? Items on guns were prefaced by the statement: “We want to know about “real” guns—the ones that shoot bullets, not BB guns, water guns, or any toy guns.” Items on knives were prefaced with “We want to know about using a knife or razor to hurt somebody, not about accidents where someone gets cut.” Response options included two “no” categories (“never” engaged in the behavior and “no” had not engaged in the behavior during the specified time period) and several “yes” categories (e.g., once, twice, about once a month). Responses were collapsed into yes/no categories and were summed to calculate a score for 8th grade aggressive behavior. The range of scores is 0–5 (females: x=1.10, 95% CI .99–1.20, Cronbach's alpha = .59; males: x=1.62; CI 1.47–1.78, Cronbach's alpha = .66).

Social and demographic measures include age in years at 8th grade and race/ethnicity, dichotomized as African American versus others for regression analyses. Household living arrangement was measured with two questions: How often do you live with your mother/father (or a person who is like a mother/father to you)? Responses were recoded into lives with both mother and father all or most of the time versus other. Frequency of going to church, worship services, mosque, or synagogue was measured by one item with response categories ranging from never to almost every week. Self-reported school performance was measured with the question, “How good are your grades compared to other students in your school?” Responses were worse, about the same, better, one of the best or ungraded/special education; these were recoded into a dichotomous variable as better/best grades versus others.

Two additional 8th grade risk behaviors were included: early substance use and sexual initiation. Early substance use was scored as a composite measure of number of substances (cigarettes, alcohol, marijuana) ever used, with a range of 0 to 3 (females: x=1.12; CI=1.03–1.22), Cronbach's alpha = .69; males: x=1.11, CI=.99–1.22; Cronbach's alpha = .67). The question, “Have you ever had sexual intercourse? This is sometimes called ‘going all the way’” was used to assess early sexual initiation, and coded as 0 for no and 1 for yes. Given the age of students, school administrators and our Institutional Review Boards indicated that questions distinguishing among oral, vaginal and anal sex should not be included in the middle school survey.

Exposures to childhood domestic violence were assessed at the young adult survey using measures adapted from the Adverse Childhood Experiences Study (Hillis et al., 2001). To assess parental physical aggression, we asked: “Sometimes, parents or other adults hurt children. Before you turned 18, how often did a parent, step-parent or adult living in your home do the following things to you: (1) push, grab, slap or throw something at you; (2) hit you so hard that you had marks or were injured; (3) threatened you with a knife or gun or used a knife or gun to hurt you.” Participants who answered yes to any question received a score of 1 and those who answered never to all three were coded 0. To assess witnessing parental partner violence, we included separate items to assess mothers’ and fathers’ as perpetrators. Three categories of violence were asked about: (1) push, grab, slap, or throw something [at partner]; (2) kick, butt, hit; and (3) threaten with or use a knife or gun. Respondents who reported witnessing a parent perpetrating one or more types of violence received a score of 1; all others were scored 0.

Measures of victimization and perpetration of partner violence were adapted from the Conflict Tactics Scales (Straus, 1979). For non-weapon related victimization, respondents were asked if anyone that they had dated and/or had sex with had done the following: “(1) hit, punched, or slapped you; (2) thrown something at you or hit you with an object; (3) pushed, grabbed, or shoved you; (4) pulled your hair, scratched, or bit you; or (5) kicked or choked you.” Four additional items assessed weapon-related victimization: “(6) threatened you with a knife; (7) threatened you with a gun; (8) used a knife against you; and (9) used a gun against you. Respondents indicated whether this happened once, more than once, or not at all.” The same set of items assessed perpetration, with the lead: “Have you ever done the following things to someone you were dating and/or having sex with.” For the two measures of partner violence, participants who endorsed any item received a score of 1; all others received a score of 0.

In addition to these dichotomous measures, we also considered two ways to assess severity, which is difficult without additional information about the context of specific acts of partner violence. For example, being threatened with a knife can be more or less serious that being hit, punched or slapped, depending on the circumstance; similarly, repeated experiences of a single type of violence may be more or less serious than exposures to different types. First, we used a higher cutpoint for the two measures (0 or 1 versus 2 or higher). Second, we distinguished between weapon-related and non-weapon violence victimization and perpetration.

In both 8th grade and young adult surveys, the amount of missing data on any single item was low. In 8th grade, missing data ranged from about 1% to 5% on the individual aggression items, and was less than 3% on partner violence items during young adulthood, when mean substitution was used to replace missing values.

Analysis plan

Descriptive information is provided for weighted samples of males and females. Logistic regressions of partner violence victimization and perpetration on 8th grade aggressive behavior were performed separately for each gender. Four models were tested. In model 1, we entered 8th grade aggression as the sole predictor of victimization or perpetration; in model 2, socio-demographic characteristics were added; in model 3, early substance use and sexual initiation were entered, and the final model included exposures to physical aggression in the childhood home. All regressions were run adjusting for year enrolled in the study and middle school attended, neither of which were significant and are not included in reported analyses. Analyses were conducted using robust variance estimates with STATA version 6.

To supplement these analyses and consider the severity of violence, we repeated the regressions using the higher cut point on the outcome measures, thus comparing multiple vs. no/few types of violence. Finally, among those who reported any lifetime partner violence, we examined whether youth who were aggressive in middle school were at greater risk of weapon-related partner violence (threats/use).

Results

The majority of participants self-identified as Black/African American, and most others were Hispanic/Latino (see Table 1). Not shown, at 8th grade, most girls (85.8%) and boys (81.6%) reported living with their mothers; about 10% reported being in mother-absent households. About a quarter (25.7%) of girls and a third (34.4%) of boys lived in a two-parent household. Church attendance was somewhat higher among females: 37.0% of girls compared to 26.3% of males attended church more than once a month. A higher proportion of girls also reported better grades than their peers. Early substance use was similar for males and females, with 59% reporting cigarette, alcohol, or marijuana use; males were more than twice as likely to report sexual initiation by 8th grade.

Table 1 Sample characteristics and percentages of females and males reporting aggressive behaviors and other risks at 8th grade (weighted)
Full size table

Nearly 40% of both genders reported that a parent or other adult in their household had used physical aggression against them. Smaller proportions reported witnessing a parent using physical aggression, including weapon use, against an intimate partner (22.7% females; 19.3% males). Among females who witnessed parental domestic violence (n=125), 44% reported both their fathers and mothers had been perpetrators, 36% had witnessed only their fathers as perpetrators, and 20% had witnessed only their mothers as perpetrators. Among males (n=83), these figures, respectively, are 57.8%, 21.7% and 20.5% (all not shown).

Table 1 also includes 8th grade reports of recent aggressive behavior; for example, 32.1% of girls reported being in physical fight in the last three months, and 20.6% had carried a knife, razor, or box cutter. Among boys, parallel figures are 42.1% and 36.3%. A count of aggressive behaviors shows that 42.1% of females reported no aggression and 4.3% reported four or more types. For boys, 29.3% reported none and 11.9% reported four or more.

Table 2 provides information on proportions of males and females who by young adulthood have been involved in each type of partner violence victimization and perpetration. In addition to information on the table, to examine the correlation between victimization and perpetration, we created two scales that count types of aggressive behaviors. Among females, the Pearson correlation between victimization and perpetration is .58; for males, the correlation is .68. Because the distribution is highly skewed (with the majority reporting no aggressive behaviors), these scales were recoded into dichotomous variables for regression analyses: 35% of females reported partner victimization, and a similar proportion reported perpetration; more males reported victimization (35%) than perpetration (22%). Cross-tabulations of these dichotomous outcomes indicate that 23.8% of females and 17.6% of males reported both victimization and perpetration, 11.3% of females and 4.0% of males reported only perpetration, 11.6% of females and 17.6% of males reported only victimization, and 53.3% of females and 60.9% of males reported neither. Also not shown, among females who reported no 8th grade aggression, 29% reported partner victimization and 25% reported perpetration; among those who reported three or more types of 8th grade aggression, these figures exceed 40%. Among males, differences are greater: among those reporting no 8th grade aggression, 26.0% reported victimization and only 8.8% reported perpetration; among those reporting higher middle school aggression, these figures are 51.7% and 37.0%, respectively.

Table 2 Percentages of females and males reporting lifetime intimate partner physical violence victimization and perpetration (weighted)
Full size table

Table 3 provides results for females of the logistic regressions of partner victimization and perpetration. The first set of models pertains to victimization. Entered alone, 8th grade aggression (model 1) is positively and significantly associated with the outcome of partner victimization (OR 1.28, S.E. 0.10, p≤.001), and remains significant (OR 1.25, S.E. .10, p≤.01) in model 2, which controls for social and demographic characteristics. Model 3 adds other 8th grade risk behaviors; substance use is a significant risk for victimization (OR 1.43; S.E. .14, p≤.001), but sexual initiation is not; 8th grade aggression is no longer significant. The final model adds physical aggression in the childhood home; both types (witnessing parents’ using aggression toward each other and being the victim of parental aggression) are positively and significantly related to victimization (p≤.001). Substance use remains significant (p≤.01), and 8th grade aggression is not.

Table 3 Odds ratios and robust standard errors for logistic regressions of intimate partner physical violence victimization and perpetration on middle school aggression for females, controlling for background characteristics, other 8th grade risk behaviors, and childhood domestic violence exposures (weighted)
Full size table

The second set of columns pertains to female partner violence perpetration. Again, 8th grade aggression (model 1) is positively associated with the outcome (OR 1.43, S.E. .11, p≤.001), and remains significant in model 2 (OR 1.44, S.E. .11, p≤.001). In contrast to victimization, early aggression continues to predict subsequent perpetration (OR 1.31, S.E. .11, p≤.001), controlling for other 8th grade risk behaviors (model 3); substance use is also a risk (p≤.001) but, again, sexual initiation is not. Earlier aggression remains significant in the final model (OR 1.30; S.E. .12, p≤.01), even when exposures to physical aggression in the childhood home are included (model 4). Those who reported parental physical aggression or witnessed aggression between parents and their partners were about two to three times as likely to report perpetration.

Table 4 presents results for males. Earlier aggression is positively associated with partner violence victimization when entered alone (OR 1.32, S.E. .10, p≤.001), controlling for social and demographic characteristics (OR 1.30, S.E. .10, p≤.001), early substance use and sexual initiation (OR 1.25, S.E. .11, p≤.01), and finally, exposures to domestic violence during childhood (OR 1.24, S.E. .11, p≤.05). It also remains significant in all four models with the outcome of perpetration: model 1 (OR 1.43, S.E. .12, p≤.001; model 2: OR 1.39, S.E. .12, p≤.001; model 3: OR 1.22, S.E. .12, p≤.05; model 4: OR 1.22, S.E. .12, p≤.05). Exposures to childhood domestic aggression are positively and significantly associated with both outcomes (p≤.001). In the final model, early sexual initiation is a significant risk (p≤.05).

Table 4 Odds ratios and robust standard errors for logistic regressions of intimate partner physical violence victimization and perpetration on middle school aggression for males, controlling for background characteristics, other 8th grade risk behaviors, and childhood domestic violence exposures (weighted)
Full size table

Finally, not shown, we considered whether early aggression is related to more serious partner violence. To do so, we first repeated the logistic regression analyses using a higher cut point on the victimization and perpetration measures (i.e., none or only one type of violence versus two or more). Employing the full model, the odds ratios for 8th grade aggression are consistent, remaining significant for perpetration among males and females and for victimization among males. We also examined whether early aggression elevated reports of weapon-related violence among the subset of participants who reported any lifetime partner violence. Among both males and females, less than 2% reported weapon-related violence but no non-weapon related violence. Results of logistic regressions indicate that, among females who reported partner violence, early aggression was positively associated with weapon-related victimization and perpetration (p≤.05). Early aggression did not distinguish weapon-related violence among males.

Discussion

Our study has been successful in following a large, community-based sample of urban youth from early adolescence, during which time both males and females engaged in high levels of aggressive behaviors, through young adulthood, by which time sizeable proportions report being the victim of or perpetrating intimate partner violence. Identifying a link between the aggressive behaviors of young adolescents and their subsequent involvement in partner violence presents an important opportunity for intervening to disrupt the pathway to adult partnership violence (Wekerle, 1999). The importance of intervening early is underscored by the fact that those who experience partner violence once have greater risk of repeat occurrences, and that the stability of violence within relationships is high, even among teenagers (O’Leary and Smith Slep, 2003).

The prevalence of partner violence perpetration by both genders is clearly high. The prevalences reported are consistent with findings from community samples examining dating and couple violence in other economically distressed populations (Archer, 2000a,b; Straus, 1999; White and Chen, 2003). They remain, however, surprising to those who assume that females are more likely to be victims and less likely to be perpetrators than males. Moreover, it is striking that, in our study, females are more likely than males to report most forms of both non-weapon and weapon-related perpetration, while males and females report similar levels of victimization. These figures certainly raise concerns about potential emotional and physical harm to both men and women. They also call attention to the fact that female perpetrators may be especially at risk since female aggression, whether or not committed in self-defense, may be met by more injury-causing, and even lethal, violence by males. Addressing aggression early and breaking the cycle are important for both genders (Yates, 2003).

Our study provides evidence that perpetration of violence during middle school is related to perpetrating partner violence by young adulthood. This finding holds for males and females, even after controlling for other early risk behaviors, including substance use and sexual initiation, and exposure to physical aggression in the childhood home. In addition, middle school aggression among boys predicts subsequent partner victimization, again controlling for potential confounders. Establishing this link between the peer-directed aggression, typical among many middle schoolers, and subsequent violence with intimate partners is important for developing evidence-based prevention and advocating for early intervention. Results highlight the need for violence prevention efforts that help young adolescent males and females learn non-violent strategies for resolving conflicts in the intimate relationships they will form over the next period of their lives. Our data suggest the importance of beginning such programs early, especially in communities where youth are at high risk of peer violence as well as subsequent violence in their intimate relationships. Attention must also be paid to differences in patterns of risks that may lead to partner violence. For example, we find that while early substance use places females at risk for partner violence, early sexual initiation may be a better marker for males. This suggests the need for gender-specific targets for early intervention.

Despite limitations in our measures of childhood exposure to domestic violence, consistent with other studies, we find that witnessing parents fighting or being the target of parental aggression places males and females at risk for both victimization and perpetration. In our study we do not distinguish between what might be relatively low levels of aggression (e.g., a parent pushing a child) and more serious child abuse and domestic violence, nor can we determine whether these childhood exposures pre- or post-date middle school aggression. In addition, we rely upon retrospective accounts of childhood exposures to household aggression. These reports are not independently verified, and might be biased by faulty memory or influenced by subsequent events, including the respondents’ experiences as a perpetrator or victim of partner violence. However, we have identified that middle school aggression contributes independently to partner violence by young adulthood; thus, early aggression provides a target for intervention that schools and other programs can address, often without the complications involved in assessing parents’ behavior at home.

It is also notable that none of the demographic or social characteristics—race/ethnicity, 8th grade age, household structure, church attendance, self-reported school performance—is related to partner violence. In several cases, the lack of association may stem from the use of single-item measures (for example, for school performance and church attendence); thus, fuller measures of school attachment and religiosity may yield different results.

While this research sheds light on the development of partner violence, the sample was limited to a population that is disproportionately poor, urban, and African American; thus, results cannot be generalized to other groups, especially those in which aggressive behaviors by females are less normative, both in early adolescent relationships and later in young adulthood. It is possible that the prevalence of partner violence, and particularly perpetration, is lower among males due to attrition bias. Although useful in reducing bias, weighting procedures may not fully account for behavior not observed or assessed. Despite efforts to retain participants in the sample and weight our results, we note that more violent males were likely to drop out. Not surprisingly, among those who remained in the study, more males than females report having been arrested, convicted of a felony, and spending time in jail. Our results may therefore underestimate the prevalence of partner violence, and particularly perpetration, among males, as well as the magnitude of the relationships between early aggression and subsequent partner violence.

Another limitation includes the exclusive focus on physical aggression. We have not examined psychological abuse, which often precedes physical abuse, or sexual aggression (Archer, 2000a,b; O’Leary, 1999). We also do not examine the meaning of partner violence in the context of young adults’ relationships. At the time of the survey, 18% of females were currently living with a partner, 82% had a main partner within the last year, and about half reported three or more sex partners during their lifetime. Among males, 11% said they were living with a partner, 67% reported a main partner within the last year, and three-quarters reported three or more lifetime sex partners. A better understanding of partner violence in the context of sexual and intimate relationships of young adults is needed to inform the content of interventions aimed at reducing couple violence (West and Rose, 2000). In addition, we note that our outcome measures of partner violence assess lifetime experiences accumulated by early adulthood. Given that dating violence can begin in early adolescence and escalate through high school, more research examining the development of partner violence from early adolescence forward is needed.

Despite these limitations, our research is one of the few longitudinal studies that follow young people from middle school to young adulthood and assess multiple forms of violence and early risk and protective factors. Our findings highlight the importance of incorporating strategies for teaching youth how to resolve conflicts without resorting to violence and maintain healthy relationships, especially in communities where levels of violence inside and outside the home are high.