Child Abuse & Neglect 38 (2014) 1902–1913

Contents lists available at ScienceDirect

Child Abuse & Neglect

Exploring the relationship between child physical abuse and adult dating violence using a causal inference approach in an emerging adult population in South Korea Wesley G. Jennings a,∗ , MiRang Park b , Tara N. Richards c , Elizabeth Tomsich d , Angela Gover e , Ráchael A. Powers a a b c d e

Department of Criminology, University of South Florida, USA Department of Police Administration, Hannam University, Republic of Korea School of Criminal Justice, University of Baltimore, USA Department of Public Affairs and Social Research, Texas A&M International University, USA School of Public Affairs, University of Colorado Denver, USA

a r t i c l e

i n f o

Article history: Received 5 April 2014 Received in revised form 13 August 2014 Accepted 21 August 2014 Available online 15 September 2014 Keywords: Child maltreatment Child abuse Intimate partner violence Dating violence Propensity scores

a b s t r a c t Child maltreatment is one of the most commonly examined risk factors for violence in dating relationships. Often referred to as the intergenerational transmission of violence or cycle of violence, a fair amount of research suggests that experiencing abuse during childhood significantly increases the likelihood of involvement in violent relationships later, but these conclusions are primarily based on correlational research designs. Furthermore, the majority of research linking childhood maltreatment and dating violence has focused on samples of young people from the United States. Considering these limitations, the current study uses a rigorous, propensity score matching approach to estimate the causal effect of experiencing child physical abuse on adult dating violence among a large sample of South Korean emerging adults. Results indicate that the link between child physical abuse and adult dating violence is spurious rather than causal. Study limitations and implications are discussed. © 2014 Elsevier Ltd. All rights reserved.

Introduction Research on intimate partner violence (IPV) persistently documents controlling and abusive relationship behaviors in samples of young people across the globe (Hines & Straus, 2007). For instance, surveys on 38 Asian, North American, South American, European, Middle Eastern, and Australian/New Zealand university sites provided evidence that one quarter of college students reported perpetrating physical IPV in the prior year (Hines & Straus, 2007). Explanations on the origins of physical relationship violence commonly focus on early learning processes, linking childhood maltreatment, or physical abuse, sexual abuse, and/or witnessing violence in childhood with adolescent and adult involvement in violent relationships (Jennings, Park, Tomsich, Gover, & Akers, 2011). The relationship between these constructs is commonly referred to as the cycle of violence. However, broader examination of the literature reveals the multi-determined or ecological nature of violence across the life course, where children exposed to violence concomitantly experience a multitude of overlapping risk factors for violent intimate relationships (Hong, Kim, Yoshihama, & Byoun, 2010; Hong, Lee, Park, & Faller, 2011). Few cycle

∗ Corresponding author. http://dx.doi.org/10.1016/j.chiabu.2014.08.014 0145-2134/© 2014 Elsevier Ltd. All rights reserved.

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1903

of violence studies employ analyses capable of addressing selection bias, and the reliance on correlational designs in extant research may overstate the relationship between physical child abuse and IPV perpetration and victimization (Jennings, Richards, Tomsich, Gover, & Powers, 2013). Although dating violence research has increasingly included samples outside of Western settings, cycle of violence applications in countries such as South Korea remain rare (Kim, Kim, Choi, & Emery, 2014; Yick & Oomen-Early, 2008), hindering an understanding of the cross-cultural applicability of the cycle of violence. Cross-national research on experiencing physical force as punishment during childhood and dating violence situates incidence rates in South Korea above the median (Chan, Straus, Brownridge, & Leung, 2008; Straus, 2010). Chan et al. (2008) reported that nearly one third of South Korean college students reported physical dating violence perpetration (34%) and/or victimization (28%) in the prior year. In Straus’s (2010) review of research on corporal punishment, South Korea ranked 8 out of 32 countries, just below the United States. Specifically, 60% of South Korean respondents did not strongly disagree that they were spanked or hit frequently prior to the age of 12 (Straus, 2010). Attitudes in South Korea regarding physical punishment of children appear favorable, with over 90% of college students in Straus’s (2010) sample reporting that a “hard spanking” is sometimes necessary to discipline a child. Recent cultural and political shifts in South Korea reflect changing norms regarding dating and family. Women comprise an increasing proportion of the workforce (Rogers, Ballantyne, & Draper, 2007) and rising numbers of South Koreans are choosing not to marry or delaying marriage (Jones, 2005; Kreider & Simmons, 2003). Nonetheless, some contend policy regarding dating and family violence lags behind these shifts. Despite the successful passage of two major anti-domestic violence bills in 1997 [These pieces of legislation are the Prevention of Domestic Violence and Victim Protection Act and the Special Act for Punishment of the Crime of Domestic Violence.], the legislation prioritized family preservation, leading criminal justice actors and state-funded counseling centers to encourage survivors to “forgive and forget” (Heo & Rakowski, 2014; Postmus & Hahn, 2007). Modifications to the Acts’ language in the 2000s supporting police intervention and victims’ rights are viewed by some as largely symbolic (Heo & Rakowski, 2014). Similarly, critics of South Korean child abuse policy point out that although South Korean law requires mandatory reporting of child abuse by physicians and teachers, a lack of consequences for failure to fulfill such responsibilities results in minimal reporting by these professionals (Kim & Jeong, 2002). With the launch of the new South Korean government in 2013, independent of the violent crimes classified by criminal justice agencies, new categories of crimes that are serious and need to be rooted out have been introduced, or the so called “four social evils”, which include sexual violence, school violence, unsafe food, and domestic violence. In this vein and relevant to domestic violence, the Republic of Korea’s Prosecution Service announced a three strikes measure compelling the detention of repeat offenders of domestic violence for investigation if the same perpetrator commits physical perpetration three times within three years. These distinctions in policy between the United States and South Korea, alongside the similarities in the rates of experiencing physical force in childhood and IPV as an adult, provide a compelling setting for applying the cycle of violence theory, and therefore contributes to the small but growing literature on dating violence in South Korea (Kim et al., 2014). In addition, the current study overcomes methodological limitations of previous research by applying a quasi-experimental analysis to control for the ecological nature of exposure to violence and determine the causal relationship between experiencing physical child abuse and physical IPV victimization and perpetration among a large sample of South Korean college students. Literature Review The origins of violent behavior are commonly framed within a social learning or intergenerational transmission of violence perspective (Akers & Sellers, 2009; Bandura, 1979). Social learning theory proposes that individuals learn violent behaviors in childhood through observing parents or caretakers instrumentally using violence to manipulate and control others (Akers & Jennings, 2009; Akers & Sellers, 2009; Bandura, 1978). Whether a child solely witnesses violence or also experiences abuse, social learning theory anticipates a greater likelihood of violent perpetration and victimization in adolescence and adulthood relative to children not exposed to violence (Bernard & Bernard, 1983; Mihalic & Elliott, 1997). A multitude of studies identify physical child abuse as a risk factor for involvement in a violent relationship in adolescence or adulthood (Gómez, 2011; Hamby, Finkelhor, & Turner, 2012; Kendra, Bell, & Guimond, 2012; Laporte, Jiang, Pepler, & Chamberland, 2011; Riggs & Kaminski, 2010). For instance, using the U.S. National Longitudinal Study of Adolescent Health data, Gómez (2011) reported that respondents with a history of physical or sexual child abuse had 90% greater odds of IPV perpetration compared with those who did not experience physical or sexual child abuse. Similar cycle of violence effects for dating violence victimization and perpetration emerge among adolescents. Hamby et al. (2012) concluded that the odds of physical dating violence victimization among an American sample of teens increased by 174% with experiences of childhood physical abuse, 97% with experiences of childhood psychological abuse, and 331% with experiences of a parent interfering with a child’s access to their other parent. Witnessing domestic violence additionally acted as a risk factor, increasing the odds of adolescent physical dating violence victimization by 120% in comparison to respondents not exposed to assaults in their family (Hamby et al., 2012). Similarly, Laporte et al. (2011) assessed risk factors for IPV among 12 to 19-year-olds sampled from a Canadian youth protection agency. Both male and female teens who experienced childhood victimization, defined as emotional or physical child abuse or corporal punishment in childhood, were at a greater risk for IPV perpetration (Laporte et al., 2011). However, males displayed the strongest relationship between childhood victimization and IPV perpetration, particularly those

1904

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

victimized by their fathers. Interestingly, victimization within the family was significantly related to dating violence victimization among females, but this finding did not hold for males (Laporte et al., 2011). Although few studies consider the cycle of violence among South Korean samples, Kim et al. (2014) contribute one exception. Kim et al. (2014) surveyed students attending universities located in the Seoul and Kyung-gi provinces of South Korea about physical child abuse, perpetration of partner violence, and neutralizing beliefs that justify criminal actions. Findings from regression analyses indicated that neutralizing beliefs partially mediated the relationship between physical child abuse and IPV perpetration for male respondents and fully mediated the relationship for female respondents. Notwithstanding the breadth of research documenting an intergenerational transmission of violence effect, less consistency emerges regarding the strength of this finding (Alexander, Moore, & Alexander, 1991; Heyman & Smith Slep, 2002). One meta-analysis of 39 cycle of violence studies found a weak to moderate relationship between physical child abuse and adult intimate partner violence victimization and perpetration (Stith et al., 2000). Methodological and conceptual limitations may account for inconsistencies in cycle of violence findings. Limitations of Prior Research Patterns in cycle of violence research initially appear nonsensical, with some studies documenting effects for female respondents only (Douglas & Straus, 2006), others for both males and females (Gómez, 2011; Laporte et al., 2011), and multiple studies proposing covariates that mediate or condition the effect of physical child abuse on IPV (Kendra et al., 2012; Kim et al., 2014). Yet, one may alternately conclude that violent victimization and its consequences are complex and multidetermined (Saunders, 2003), or ecological in nature (Belsky, 1980). An ecological understanding of violent victimization and perpetration across the life course would look to nested, interdependent individual, familial, and community systems. The notion of interrelated systems is mirrored in cycle of violence research, which commonly controls for influences at various levels. For example, at the individual level research has controlled for factors related to impulsivity, such as age (Millett, Kohl, Jonson-Reid, Drake, & Petra, 2013), sex (Renner & Whitney, 2012), low self-control (Gover, Kaukinen, & Fox, 2008; Rebellon, Straus, & Medeiros, 2008; Wright & Fagan, 2013), substance use (Millett et al., 2013; Renner & Whitney, 2012), delinquency (Cui, Ueno, Gordon, & Fincham, 2013; Millett et al., 2013), and risky sexual behavior (Gover et al., 2008). Family system level variables would measure the strength of ties to family, such as family of origin structure (Cui et al., 2013) and parental support (Gover et al., 2008). Relevant covariates to the community level in a cycle of violence model would appraise ties to community elements either encouraging criminality, such as substance using and/or delinquent peers (Wright & Fagan, 2013), or conformity to the law, such as religious institutions (Gover et al., 2008). Taken together, it is possible that mixed research findings on the connection between physical child abuse and behavioral outcomes result from selection bias, as families characterized by violence are typically multi-problem families (Emery, 2011). As a result, analyses failing to account for a myriad of salient risk factors commonly co-presenting with family violence, such as weak social bonds (Hirschi, 1969), or low self-control (Sellers, 1999), may exaggerate the relationship between exposure to violence and subsequent problematic outcomes (Emery, 2011). Although minimal research controls for the ecological assumption that children exposed to physical abuse may be disproportionately exposed to a multitude of cross-system risk factors, Jennings et al.’s (2013) recent dating violence study among U.S. college students presents an exception. In this regard, Jennings et al. (2013) controlled for selection bias through propensity score matching, an approach innovative to the literature on IPV and the intergenerational transmission of violence, and conducive to greater precision in estimating the causal effect of childhood maltreatment. This methodological design demonstrated that after matching child abuse victims and non-victims on numerous confounding variables, the significant relationships between physical child abuse and dating violence perpetration and victimization was eliminated (Jennings et al., 2013). Specifically, the results suggested that the relationship between early physical child abuse and later dating violence was not causal and in fact, spurious. Jennings et al. (2013) speculated that other factors that disproportionately characterize the lives of child abuse victims, such as low self-control, weak social bonds, and involvement in risky behaviors and/or with risky peers, are relevant to understanding the relationship between child abuse and partner violence perpetration and victimization. This research serves as a case study in South Korea on the degree to which a robust and novel statistical procedure may produce results and theoretical implications entirely distinct from predominant approaches, and informs the methodology of the current study. The Current Study Extant research indicates that substantial proportions of college students in South Korea experience physical abuse in their family of origin and/or violent intimate relationships (Chan et al., 2008; Straus, 2010). However, research on IPV in Asian communities is rare among widely circulated journals on intimate partner violence (Chan & Straus, 2011; Yick & Oomen-Early, 2008). Furthermore, multiple studies have established a relationship between physical child abuse and adult IPV, and yet with the exception of Jennings et al. (2013), few control for the threat of selection bias through propensity score matching. In recognition of these issues, the current study addresses the methodological limitations of previous research by using a propensity score matching approach to examine the causal effect of physical child abuse on dating violence among a sample of emerging adults in South Korea. Despite the absence of random assignment, the quasi-experimental research

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1905

design featured in this study permits a rigorous examination of social learning theories’ explanatory power for IPV in a South Korean context while controlling for ecological assumptions of co-occurring salient risk factors. Methods Sample and Procedures The current study utilizes a large convenience sample of undergraduate college students enrolled in various liberal arts and science classes at a large South Korean university between May and June of 2007 (n = 1,399) [The survey instrument and administration procedures used in this research were approved by the host university’s Institutional Review Board.]. Students were asked to participate in the research by completing the Family and Relationship Experiences and Attitudes Among College Students survey (Gover et al., 2008) during class. This survey consists of 167 questions regarding dating violence perpetration and victimization, relationships with parents, exposure to violence during childhood, risk-taking behaviors, and various demographic questions [All of the items/measures used for this particular study were derived and translated from a survey instrument that was originally used in prior U.S.-based, college student, child abuse and dating violence research (Gover et al., 2008). Specifically, the Family and Relationship Experiences and Attitudes Among College Students survey questionnaire was translated from English into Korean by a native Korean speaker with a Ph.D. (the second author of this manuscript) and back translated into English in an effort to identify and resolve any points of confusion with the original translation.]. Participants were provided with packets including the informed consent documents and survey instrument. Upon completion, participants submitted the sealed package to a member of the research team. Respondents were not compensated for their participation. The research yielded a 96% response rate [It is important to note the high nature of this response rate given that no incentives were provided to participants. The response rate reported here, however, is not necessarily unique to investigations of this kind in the literature. In fact, Jennings et al. (2013) employed a similar methodology and analytical approach among a sample of U.S. college students and obtained a similarly high response rate (99%) without offering incentives to participants. Nevertheless, it is still possible that culture influenced the high response rate obtained in the current study (Dalton & Ong, 2005)]. Regarding the sample demographics, fifty-six percent of the sample was female and the mean participant age was 19.92 (SD = 1.12). Nearly all of the sample reported being raised in a two-parent household (93.5%). Thirty-four percent of the participants were freshman, 32% were sophomores, 20% reported being juniors, and 14% were seniors. Physical dating violence perpetration was reported by 14.8% of the sample and physical dating violence victimization was reported by 12.1%. Measures Dependent Variables. Physical dating violence perpetration (˛ = .86) and victimization (˛ = .88) were measured using nine items from the Revised Conflict Tactics Scale (Straus, Hamby, Boney-McCoy, & Sugarman, 1996) that asked respondents how often they engaged in or were victimized by the following behaviors with a dating partner in the previous 12 months: (a) threw something that could hurt; (b) twisted arm or hair; (c) kicked; (d) slapped; (e) pushed or shoved; (f) punched or hit with his or her hand or an object; (g) choked; (h) slammed against the wall; and (i) grabbed. Specific response options included 0 = this has never happened, 1 = once in the past year, 2 = twice in the past year, 3 = three to five times in the past year, and 4 = six or more times in the past year, and 5 = Not in the past year, but it did happen before. Responses were collapsed into two dichotomous variables where 1 represented perpetrating or being victimized by at least one form of violence against/from a dating partner in the past year and 0 indicated that the respondent had not perpetrated and/or been victimized by dating violence in the prior year. Treatment Variable. Physical child abuse was measured with seven items from the Revised Conflict Tactics Scales (Straus et al., 1996). Specifically, respondents were asked whether or not they had experienced the following behaviors from a parent, guardian, or caretaker: (a) throw something that could hurt you; (b) push, grab, or shove you; (c) pull your hair; (d) slap or hit you; (e) hit you with some object (not including spanking); (f) punish you with a belt, board, cord, or other hard object (not including spanking); or (g) hit you so hard that it left bruises or marks (˛ = .80). The childhood maltreatment measure was recoded into a dichotomous variable that indicated whether respondents had experienced at least one of the seven abusive behaviors during their childhood. Those who had experienced at least one of the seven abusive behaviors during their childhood were coded as 1 = treatment, and those who did not report experiencing at least one of the behaviors were coded as 0 = control [The child physical abuse and IPV measures were significantly skewed, which resulted in their dichotomization (particularly for IPV). Second, an additional reason to dichotomize these variables is purely methodological for the purpose of estimating the propensity scores. When using propensity score methods your main independent variable of interest (e.g., child physical abuse in this study) has to be dichotomous as this is the variable that is considered the “treatment”. Or in other words, this variable must be expressed as dichotomy so that cases can be assigned to mutually exclusive categories (treatment group vs. control group) after matching the cases across the host of covariates/confounders. After this process, the treated and control cases can then be compared across the specific outcome/s of interest.].

1906

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

Propensity Score Covariates. To account for selection processes (Pearl, 2009; Rosenbaum & Rubin, 1983; Shadish, Cook, & Campbell, 2002), we included 29 covariates as potential confounders. These covariates were used to create the conditional probability or propensity score used to generate a similarly-situated, matched sample of child physical abuse victims and child physical abuse non-victims. In this regard, covariates found in prior research to influence college students’ risk for dating violence victimization and perpetration were included in the analysis as confounders (Gover, 2004). Social learning theory suggests that witnessing violence in childhood is associated with a higher risk of future perpetration of and victimization from violence. Father-to-mother violence was measured by asking participants if they had witnessed their father perpetrate physical violence against their mother (i.e., father-to-mother violence). Likewise, mother-to-father violence was measured by asking respondents if they had witnessed their mother perpetrate physical violence against their father during their childhood (i.e., mother-to-father violence). Maternal and paternal support measures include four-item scales specific to each parent (items were modified from Hirschi, 1969). Scale items included “How often do you trust your mom/dad?”; “How often do you feel you can talk to her/him about your problems?”; “How often do you think she/he is genuinely interested in you?”; and “How often do you feel that she/he supports you?” (1 = never, 2 = sometimes, 3 = half of the time, 4 = usually, 5 = always). The Cronbach’s alpha values for the two measures were ˛ = .86 and ˛ = .90 for maternal and paternal support, respectively. Religiosity was assessed using a one-item self-report question. Responses were coded as 1 = not religious, 2 = moderately religious, and 3 = very religious. Religious service attendance was measured by asking participants how often in the past year they had attended a religious service (1 = never, 2 = seldom, 3 = monthly, 4 = weekly). Substance use was measured by asking participants whether in the past year they had (a) used alcohol, (b) used cigarettes, (c) used marijuana, (d) used hard drugs, (e) used a fake id, or (f) drank alcohol and drove a vehicle. Responses were coded dichotomously (0 = no; 1 = yes). Peer substance use was measured using these same questions that asked participants whether in the past year their friends had (a) used alcohol, (b) used cigarettes, (c) used marijuana, (d) used hard drugs, (e) used a fake id, or (f) drank and drove a vehicle. Responses were coded dichotomously (0 = no; 1 = yes). Self-control was measured using a 23 item additive scale from Grasmick, Tittle, Bursick, and Arneklev (1993). Sample scale items included the following: “I often act on the spur of the moment without stopping to think”; “I don’t devote much thought and effort to preparing for the future”; “I often do whatever brings me pleasure here and now, even at the cost of some distant goal”; and “I’m more concerned with what happens to me in the short run than in the long run.” Likert response items were used with a range of 1 to 4, with higher numbers indicating lower self-control (˛ = .80). Risky sexual behavior was measured using a three-item scale. Scale items included “How old were you when you had sexual intercourse?” (0 = I have never had sexual intercourse to 6 = 18 years or older); “During your life how many people have you had sexual intercourse with?” (0 = I have never had sexual intercourse to 6 = 6 or more partners); and “During the last 3 months, how many people have you had sexual intercourse with?” (0 = I have never had sexual intercourse to 6 = 6 or more partners; ˛ = .75). Additional demographic variables included respondents’ sex, age, college class rank (freshman, sophomore, junior, senior), growing up in a household with their natural parents, living off campus, and being in an exclusive dating relationship. Analytic Strategy An experimental research design using randomly assigned treatment and control groups is the gold standard for evaluating the treatment effect of a given condition. Having said this, it is obviously not feasible or ethical to randomly assign youth to households where they either will or will not experience child physical abuse and then follow them up years later in young adulthood to assess potentially relevant outcomes such as dating violence perpetration and/or victimization. Therefore, in an effort to utilize more rigorous methods than have generally not been used in the past when assessing these relationships, we employ propensity score matching (PSM) techniques to approximate a quasi-experimental research design where we can statistically assess the potential causal linkages between experiencing child maltreatment early on in the life course and later dating violence perpetration and/or victimization. PSM is becoming more commonplace as an analytical strategy in criminology and related disciplines (for examples, see Loughran et al., 2009; Sampson, Laub, & Wimer, 2006), as this method permits the estimation of propensity (or balancing) scores in an effort to statistically remove observable and systematic differences or imbalances that may be apparent between the treatment and control groups prior to evaluating the outcome/s of interest (Rosenbaum & Rubin, 1983). The propensity scores derived from the available covariates are estimated via a logit model. As applied in the current study, the “treatment” group represents those individuals that did report experiencing childhood maltreatment compared with the “control” group who are represented by individuals who did not report experiencing childhood maltreatment but are, statistically speaking, a similarly-situated group to the treatment group in terms of their covariate risk profile. In this regard, the analysis that follows proceeds in a series of stages. First, PSM is applied using an R program (Ho, Imai, King, & Stuart, 2007; Ho, Imai, King, & Stuart, 2011) relying on statistical convention (Austin, 2009; Rosenbaum, 2002) and utilizing a 1:1 nearest neighbor, propensity score matching algorithm with a strict .05 caliper to create a matched set of treated and control cases that are statistically balanced across the range of relevant covariates previously discussed. The second stage of the analysis involves the estimation of a series of mean-difference comparisons between the individuals who reported experiencing child maltreatment and those who did not, along with an examination of the nature and magnitude of these differences reported as standardized mean differences (e.g., Cohen’s d; Cohen, 1988). These group comparisons are

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1907

Table 1 Bivariate comparisons of covariates before and after matching for child physical abuse victims and child physical abuse non-victims. Variables

Before matching Child physical abuse victims (n = 702)

Male Age Freshman Sophomore Junior Senior Natural parent household Lives off-campus Exclusive dating relationship Witnessed father hit mother Witnessed mother hit father Religious service attendance Religiosity Alcohol use Cigarette use Marijuana use Hard drug use Drink and drive Fake ID use Peer alcohol use Peer cigarette use Peer marijuana use Peer hard drug use Peer drink and drive Peer fake ID use Maternal support Paternal support Risky sexual behavior Low self-control

.61 19.90 .35 .32 .19 .13 .93 .95 .32 .31 .13 1.82 1.60 .97 .37 .01 .01 .07 .10 .98 .58 .01 .02 .13 .19 16.20 14.43 2.19 77.30

Child physical abuse non-victims (n = 550) .51 19.96 .33 .32 .22 .14 .94 .95 .32 .12 .03 1.87 1.55 .97 .30 .01 .01 .06 .07 .98 .53 .01 .01 .10 .11 17.12 16.01 1.90 75.99

After matching Standardized mean difference .21 −.06 .06 .00 −.08 −.01 −.06 .03 .00 .41 .31 −.05 .07 .00 .14 .00 .00 .02 .10 .10 .11 .02 .05 .06 .19 −.26 −.36 .09 .17

Child physical abuse victims (n = 456) .56 19.98 .32 .33 .21 .14 .95 .95 .31 .16 .05 1.85 1.58 .97 .31 .01 .01 .06 .07 .99 .54 .01 .01 .12 .13 16.95 15.61 1.83 76.61

Child physical abuse non-victims (n = 456) .55 19.96 .34 .30 .22 .14 .95 .95 .31 .14 .04 1.86 1.58 .97 .33 .01 .01 .06 .08 .99 .54 .01 .01 .12 .13 16.99 15.63 2.05 76.55

Standardized mean difference .01 0.3 −.06 .07 −.03 .00 .00 .00 .00 .05 .03 −.01 .00 .00 −.04 .00 .00 .00 −.03 .00 .00 .00 .00 .00 .00 .02 −.01 −.06 −.01

Note: Significant mean differences (p < .05) in italics.

evaluated both prior to and following the propensity score estimation. Finally, the potential linkages between experiencing early childhood physical abuse and later dating violence perpetration and victimization are assessed through a comparison of the odds of reporting either outcome both prior to and following the propensity score estimation. Results Table 1 presents the mean difference comparisons, and Fig. 1 provides a graphical illustration of a histogram with overlaid kernel density estimates of the standardized mean differences between the child physical abuse victims both prior to and following the propensity score estimation across the range of covariates. What is readily apparent is that the application of PSM considerably reduced the imbalances that previously existed prior to matching, and that the individuals who reported experiencing child physical abuse prior to when propensity score matching methods were applied evinced significantly more risk across the range of covariates relative to those individuals who did not report experiencing child physical abuse early on in life. Specifically, nine of the covariates were significantly different prior to matching with the child physical abuse victims having a greater proportion of females (M = 0.61 vs. M = 0.51, p < .05), a higher prevalence of witnessing father-to-mother (M = 0.31 vs. M = 0.12, p < .05) and mother-to-father violence (M = 0.13 vs. M = 0.03, p < .05), a higher proportion of cigarette (M = 0.37 vs. M = 0.30, p < .05) and peer cigarette users (M = 0.58 vs. M = 0.53, p < .05), a higher proportion of peers who use fake IDs (M = 0.19 vs. M = 0.11, p < .05), demonstrated significantly less maternal (M = 16.20 vs. M = 17.12, p < .05) and paternal support (M = 14.43 vs. M = 16.01, p < .05), and displayed significantly less self-control (M = 77.30 vs. M = 75.99, p < .05) than their non-child physical abuse victim counterparts. Turning toward the magnitude of these statistically significant differences, the results suggested that the differences that were observed prior to matching ranged from small to moderate in terms of an effect size (d = .14 to d = .41 in absolute value). In contrast, the mean difference comparisons post-matching indicated that all of the nine previously observed significant mean differences with small to moderate magnitude were rendered non-significant with negligible (d = .01 to .07 in absolute value) to null (d = .00) effect sizes. Thus, these results provide relatively robust evidence that the application of PSM was efficient in removing any systematic and observable differences that were apparent prior to matching and generated highly comparable samples based on risk with the only difference being that one group experienced the ‘treatment’ (e.g., child physical abuse) and the other group did not experience the treatment.

1908

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

Fig. 1. Distribution of standardized mean differences before and after matching for child physical abuse victims (“treated”) and child physical abuse non-victims (“controls”) with overlaid kernel density estimate.

Following the statistical comparisons between the individuals who experienced child physical abuse prior to and postmatching, the final stage of the analysis involved an examination of the odds ratio for experiencing the dating violence perpetration and victimization outcomes for the unmatched and matched samples of child physical abuse and child physical abuse non-victims. Fig. 2 offers an illustration of the prevalence of dating violence perpetration prior to and post-matching. As can be seen, the odds of a child physical abuse victim reporting being a dating violence perpetrator were 2.11 times greater than the odds of a child physical abuse non-victim reporting being a dating violence perpetrator prior to matching {Before Matching (OR = 2.11; 95% CI = 1.50–2.96; p < .001)}. However, following the application of PSM, this statistically significant

Fig. 2. Dating violence perpetration differences before and after matching for child physical abuse victims (“treated”) and child physical abuse nonvictims (“controls”). Note: Physical dating violence perpetrator: Before Matching (OR = 2.11; 95% CI = 1.50–2.96; p < .001)/After Matching (OR = 1.48; 95% CI = 0.97–2.23; p > .05). Abbreviations: OR, odds ratio; CI, 95% confidence interval.

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1909

Fig. 3. Dating violence victimization differences before and after matching for child physical abuse victims (“treated”) and child physical abuse non-victims (“controls”). Note: Physical dating violence victim: Before Matching (OR = 2.17; 95% CI = 1.49–3.15; p < .001)/After Matching (OR = 1.44; 95% CI = 0.92–2.26; p > .05). Abbreviations: OR, odds ratio; CI, 95% confidence interval.

difference was rendered insignificant {After Matching (OR = 1.48; 95% CI = 0.97–2.23; p > .05)}. Virtually identical results were obtained when considering the odds of being a dating violence victim as well. Specifically, child physical abuse victims had greater than two times the odds of experiencing dating violence victimization prior to matching compared to child physical abuse non-victims, but, after matching, this difference became insignificant {Before Matching (OR = 2.17; 95% CI = 1.49–3.15; p < .001)/After Matching (OR = 1.44; 95% CI = 0.92–2.26; p > .05)} (Fig. 3). [As a point of further clarification, our original sample size was 702 child physical abuse victims and 550 child physical abuse non-victims (total sample size = 1,252). After applying nearest neighbor 1:1 propensity score matching (keeping in mind that there were only 550 child physical abuse non-victims to begin with so that only 550 child physical abuse victims could be matched under the most optimal scenario, e.g., total optimal sample size of 1,100), our final sample size of matched child physical abuse treatment cases to child physical abuse control cases was (456 + 456 = 912). Therefore, our initial sample of potentially “matchable” cases of 1,100 was reduced to 912 after the extreme outliers were left unmatched and removed in an effort to increase the precision of the propensity score estimates. As such, we still maintained considerable statistical power to identify statistically significant effects (Raudenbush & Xiaofeng, 2000).] [At the request of one of the reviewers who noted that witnessing violence in the home is often considered as a measure of child abuse in the U.S., we re-estimated all of the analysis using witnessing interparental violence in the home (e.g., mother hit father and/or father hit mother) as the “treatment” condition versus having experienced child physical abuse. And, in these particular analyses, experiencing child physical abuse was considered as a covariate instead. The results indicated that there were 314 cases that reported witnessing interparental violence (258 of which were successfully matched with 258 cases that did not report witnessing interparental violence across the host of covariates including having experienced child physical abuse). Furthermore, the findings demonstrated that the cases that witnessed interparental violence in the home were significantly more likely to have also experienced child physical abuse prior to matching (77.4% versus 48.9%, respectively; OR = 3.57, 95% CI = 2.66–4.79), although after matching this difference diminished to non-significance (49.7% versus 50.7%, respectively; OR = 0.96, 95% CI = 0.65–1.42) as did the statistically significant differences observed across the other covariates prior to matching. In addition, the statistically significant differences observed between the treated cases (e.g., those who witnessed interparental violence) and the control cases (e.g., those who did not witness interparental violence) across the outcomes of dating violence perpetration {Before Matching (OR = 1.78; 95% CI = 1.28–2.48; p < .01)/After Matching (OR = 1.25; 95% CI = 0.81–1.95; p > .05)} and dating violence victimization {Before Matching (OR = 1.66; 95% CI = 1.15–2.39; p < .01)/After Matching (OR = 1.28; 95% CI = 0.79–2.09; p > .05)} were also statistically different prior to matching, but not post-matching. Therefore, it appears that the results are robust and consistent regardless of which child abuse measure is used (e.g., either experiencing child physical abuse or witnessing interparental violence in the home).] Discussion Similar to research in the U.S., studies of domestic violence in South Korea have found that those who experienced child physical abuse or witnessed IPV are more likely to perpetrate and experience domestic violence (Hong et al., 2010). Social learning theory (Bandura, 1978) is typically used to explain this relationship, where children learn through behavioral conditioning. Likewise, Straus (1991) suggests that the use of physical punishment results in children associating love and moral rightness with violence. Taken together, theory and research suggest that experiencing or witnessing violence in the home imparts on the child that violence is justified as a means of conflict resolution. This literature is situated in the larger literature that suggests that there are many risk factors for domestic violence victimization (Stith et al., 2000; Stith, Smith, Penn, Ward, & Tritt, 2004). However, although the results of meta-analyses

1910

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

examining the relationship between child abuse and later domestic violence victimization is significant, the effect is generally weak to moderate (Stith et al., 2000). Furthermore, the results of these studies are correlational, rather than causal. In other words, it is unknown whether experiencing child abuse leads to a greater likelihood of experiencing adult dating violence or rather those who experience child physical abuse also are more likely to experience a range of risk factors that increase their likelihood of experiencing victimization. Indeed, the only study to date that has isolated the causal impact of child physical abuse on IPV suggested that the relationship may be spurious (Jennings et al., 2013). To that end, this study utilized a quasi-experimental methodology to examine whether child physical abuse in South Korea is a causal factor in adult dating violence victimization and perpetration. As expected, there were significant differences between those who experienced child physical abuse and those who did not. Many of these differences are in line with previous research. For example, those who experience child physical abuse are also more likely to witness domestic violence in the household (Hong et al., 2011), have insecure attachments to parents (Morton & Browne, 1998), and be involved in adolescent delinquency (Smith & Thornberry, 1995). Therefore, it appears that child physical abuse co-occurs with other known risk factors. Furthermore, this study replicated earlier findings with regard to the relationship between childhood abuse and domestic victimization and perpetration as child physical abuse victims were significantly more likely to report dating violence. However, once these groups were matched on relevant characteristics, the effect of child maltreatment became insignificant. In other words, once adults who experienced child physical abuse were matched with a sample of adults who did not experience child physical abuse, both groups were equally likely to experience adult dating violence. The results of this study suggest that the causal relationship between experiencing physical abuse as a child and later dating violence may be spurious. While experiencing childhood physical abuse is a factor associated with adult dating violence, it likely exists in tandem with other problems within the family that together contribute to the underlying mechanism that supports the cycle of violence. For example, risk factors for childhood maltreatment found in previous research using South Korean samples include poor parent–child relationship quality, family poverty, mother’s alcohol use, and parents’ endorsements of corporal punishment and Confucian-type family structures (Hong et al., 2011). Many of these risk factors, including the Confucian-type family structures have also been cited as risks factors for domestic violence in South Korea (Hong et al., 2010). In a multi-nation study, South Korea emerged as having a higher than average approval for corporal punishment (Straus, 2009). Although the origins are different, similar to the U.S., justification for violence is rooted in historical cultural and religious values. Confucianism influences the attitudes and beliefs regarding the use of violence against children and women in the home. Male dominance and the belief that violence should remain a private matter contribute to the use of intimate partner violence among South Koreans. For example, in a sample of Korean American families, females in patriarchal relationships experienced a greater likelihood of violence (Kim & Sung, 2000). Likewise, the importance placed on obedience in the household supports the use of more severe forms of physical punishment (e.g., use of objects such as belts) (Hong et al., 2011). This study focused on the relationship between experiencing physical child abuse and adult dating violence, but taken together, these previous studies suggest that the use of and attitudes surrounding corporal punishment may also contribute to the likelihood of domestic violence. Future research should explore the relationship between these cultural and religious values and domestic violence in South Korea. Given that both domestic violence and child maltreatment may stem from the same underlying cultural traditions, it is not surprising that children are at a greater risk of being abused if there is domestic violence in the home (Hong et al., 2011). Furthermore, research suggests that these forms of violence are related to the help-seeking behaviors of women in abusive relationships. Kim and Lee (2011) investigated help seeking behaviors among a clinical sample of battered women in metropolitan and provincial South Korea. Surprisingly, although one quarter called the police, one quarter pursued legal or shelter assistance, and over half sought medical treatment, only 15% contacted sources of informal support, such as family or neighbors (Kim & Lee, 2011), which further reflects the importance placed on intact male-headed households. Although a number of factors influenced women’s choice to seek help, the authors pointed to the importance of child safety on mother’s decision-making. Child abuse from the partner was associated with a greater likelihood of turning to formal (e.g., police) and informal (e.g., friends) help-seeking sources. The authors suggested that South Korea should provide integrated services to address both IPV and child abuse. Examining the link between cultural values and domestic violence in South Korea may be particularly pertinent as current research suggests that public opinion regarding child maltreatment and domestic violence may be changing to reflect less support for the use of violence in the home (Hong et al., 2010; Hong et al., 2011). This suggests that similar to the U.S. in the 1990s, South Korea may be experiencing a cultural shift in how they define domestic violence and the culpability of offenders. However, policy reform is still in its infancy. Although there have been notable efforts undertaken by advocacy groups in South Korea since the 1970s and the passage of the Prevention of Domestic Violence and Victim Protection Act (Protection Act) in 1997 (Postmus & Hahn, 2007), some criticize these efforts as ineffective as male dominance is supported through tradition and the implementation of these laws. In addition, informed by the ecological approach to understanding violence, risk factors for domestic violence include not only individual-level determinants, but also familial, community, and sociocultural factors. Therefore any potential policy initiatives and interventions should be approached from a multi-systemic standpoint. Also, despite legal reforms, Hong et al. (2010) suggest that police officer training in South Korea is still inadequate and recommend inter-agency collaboration with professions such as social workers to more effectively handle cases. Furthermore, currently, the Prevention of Domestic Violence and Victim Protection Act (Protection Act) only applies to married

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1911

relationships. In this regard, South Korea does not have a specific law for domestic violence in unmarried dating relationships. Therefore, even though domestic violence is enforced strictly under this Act, unmarried dating violence perpetrators and victims are not protected by this law. Recognizing the theoretical and methodological contributions of the current study, there are several notable limitations. Although one limitation to this research is its sample of convenience, curbing theoretical conclusions about the application of social learning theory in a South Korean context, Straus (2009) argues that unrepresentative convenience samples can result in valid cross-national comparisons. Titled the “national context effect,” Straus (2009) indicates that so long as convenient samples are comparable, valid theory testing is possible. The national context effect suggests that comparing the current study’s theoretical application to that in other nations in convenience-based college samples is not unsound. However, it is important to note that this study was limited in scope to one country; this study should be replicated in other countries, both Western and non-Western. Furthermore, future research should explore this relationship using non-college student samples. In particular, research has recently begun to understand the extent to which adolescent relationships feature dating violence (Black et al., 2011). Considering that adolescents that experience child abuse may still be living with their abusive parent(s), it is possible that the effects of child abuse may be more proximally related to dating violence among adolescents. Second, it is important to note that domestic violence is underreported in official and unofficial data sources. Furthermore, domestic violence that features coercive control (i.e., intimate partner terrorism) or emotional abuse may be particularly underrepresented in research (Johnson & Ferraro, 2000). As such, it is possible that this study has captured less severe forms of domestic violence. Victims of child maltreatment may be more likely to become involved in relationships that feature intimate partner terrorism. Furthermore, this study focused on physical child abuse. Future research should explore other forms of child maltreatment including psychological, sexual, and neglect. In addition, it is important that we briefly note some limitations of propensity score methods more generally as well. For example, we relied on a nearest neighbor 1:1 matching method, which is the most commonly used and suggested propensity score matching method (Ho et al., 2011; Rosenbaum, 2002). Having said this, this method will result in data loss as the extreme outliers on the propensity score distribution will likely not “find a match” (e.g., these cases will be dropped resulting in reduced effective sample size). Although, the cost of this data loss is more than offset by the benefit of increasing the precision in the estimates and ultimately arriving at the most comparable matched groups of treated and control cases (Ho et al., 2011; Rosenbaum, 2002). Pearl (2009) also notes that the potential for hidden variable bias is a limitation of the propensity score matching method (as it is similarly a limitation for any type of traditional regression model) in that the propensity score matching process only accounts for the variables that are observed/observable. However, equally important, is the fact that propensity score matching does not operate under the same constraints that traditional regression models do with regard to sample size and multicollinearity concerns as, generally speaking, the more covariates the better, and the more covariates the greater the precision in the obtained estimates. Thus, while we are not arguing that propensity score methods are a panacea for any and all research questions, they do (at least here in the current study) demonstrate the utility of how applying this particular analytic strategy in order to increase the precision of estimates can lead to disparate, and at times, opposite conclusions relative to those that have been previously reported in the literature (e.g., the linkage between child physical abuse and IPV being spurious). Ultimately, as is true with any statistical analysis, caution should be taken when interpreting the results within the constraints of the known strengths and limitations of the method employed in order to not over- or understate the findings from any one study. Nevertheless, we believe that this study has taken an important step in further unpacking the causal, or in fact spurious, relationship between child maltreatment and later adult dating violence perpetration and victimization among a cultural group (South Koreans) that has largely been ignored in this literature. Taken together, the intergenerational transmission of violence theory suggests that child maltreatment is linked to later dating violence through mechanisms of social learning. However, the majority of the research that examines the cycle of violence has been correlational and focused in the U.S. The results of this study suggest that the relationship between child physical abuse and dating violence may be spurious. Using a college student sample of South Koreans, this study used a quasiexperimental design to isolate the causal impact of child physical abuse and found that once children from families that featured child physical abuse are matched with children from families that did not feature such abuse, the groups’ likelihood of perpetrating and experiencing later dating violence became statistically equivalent. In recognition of these important findings, we encourage future studies to build upon the cross-cultural and international focus of this research in broadening the applicability and in testing the relevance of theoretical and empirical discussions regarding child maltreatment and dating violence going forward. References Akers, R. L., & Jennings, W. G. (2009). The social learning theory of crime and deviance. In Handbook on crime and deviance. New York: Springer. Akers, R. L., & Sellers, C. S. (2009). Criminological theories: Introduction, evaluation and application (5th ed.). New York, NY: Oxford University Press. Alexander, P. C., Moore, S., & Alexander, E. R. (1991). What is transmitted in the intergenerational transmission of violence? Journal of Marriage and Family, 53, 657–668. Austin, P. C. (2009). Some methods of propensity-score matching had superior performance to others: Results of an empirical investigation and Monte Carlo simulations. Biomedical Journal, 51, 171–184. Bandura, A. (1978). Social learning theory of aggression. Journal of Communication, 28(3), 12–29. Bandura, A. (1979). Social learning theory. Englewood Cliffs, NJ: Prentice Hall. Belsky, J. (1980). Child maltreatment: An ecological integration. American Psychologist, 35(4), 320–335.

1912

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

Bernard, M. L., & Bernard, J. L. (1983). Violent intimacy: The family as a model for love relationships. Family Relations, 32, 283–286. Black, M. C., Basile, K. C., Breiding, M. J., Smith, S. G., Walters, M. L., Merrick, M. T., Chen, J., & Stevens, M. R. (2011). The National Intimate Partner and Sexual Violence Survey (NISVS): 2010 summary report. Atlanta, GA: National Center for Injury Prevention and Control, Centers for Disease Control and Prevention. Chan, K. L., & Straus, M. A. (2011). Prevalence and correlates of physical assault on dating partners. Family & Intimate Partner Violence Quarterly, 3(3), 197–216. Chan, K. L., Straus, M. A., Brownridge, A. T., & Leung, W. C. (2008). Prevalence of dating partner violence and suicidal ideation among male and female university students worldwide. Journal of Midwifery & Women’s Health, 53, 529–537. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates. Cui, M., Ueno, K., Gordon, M., & Fincham, F. D. (2013). The continuation of intimate partner violence from adolescence to young adulthood. Journal of Marriage and Family, 75(2), 300–313. Dalton, R. J., & Ong, N.-N. T. (2005). Authority orientations and democratic attitudes: A test of the “Asian Values” hypothesis. Japanese Journal of Political Science, 6, 1–21. Douglas, E. M., & Straus, M. A. (2006). Assault and injury of dating partners by university students in 19 countries and its relation to corporal punishment experienced as a child. European Journal of Criminology, 3, 293–318. Emery, C. R. (2011). Controlling for selection effects in the relationship between child behavior problems and exposure to intimate partner violence. Journal of Interpersonal Violence, 26(8), 1541–1558. Gómez, A. M. (2011). Testing the cycle of violence hypothesis: Child abuse and adolescent dating violence as predictors of intimate partner violence in young adulthood. Youth & Society, 43(1), 171–192. Gover, A. R. (2004). Risky lifestyles and dating violence: A theoretical test of violent victimization. Journal of Criminal Justice, 32, 171–180. Gover, A. R., Kaukinen, C., & Fox, K. A. (2008). The relationship between violence in the family of origin and dating violence among college students. Journal of Interpersonal Violence, 23, 1667–1693. Grasmick, H. G., Tittle, C. R., Bursick, R. J., & Arneklev, B. J. (1993). Testing the core empirical implications of Gottfredson and Hirschi’s general theory of crime. Journal of Research in Crime and Delinquency, 30, 5–29. Hamby, S., Finkelhor, D., & Turner, H. (2012). Teen dating violence: Co-occurrence with other victimizations in the National Survey of Children’s Exposure to Violence (NatSCEV). Psychology of Violence, 2(2), 111–124. Heyman, R. E., & Smith Slep, A. M. (2002). Do child abuse and interparental violence lead to adulthood family violence? Journal of Marriage and Family, 64(4), 864–870. Heo, M. S., & Rakowski, C. A. (2014). Challenges and opportunities for a human rights frame in South Korea context and strategizing in the anti-domestic violence movement. Violence Against Women,. Advance online publication. doi: 1077801214536287. Hines, D. A., & Straus, M. A. (2007). Binge drinking and violence against dating partners: The mediating effect of antisocial traits and behaviors in a multi-national perspective. Aggressive Behavior, 33, 441–457. Hirschi, T. (1969). Causes of delinquency. Berkeley: University of California Press. Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Analysis, 15, 199–236. Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2011). Matchit: Nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 8, 1–28. Hong, J. S., Kim, S. M., Yoshihama, M., & Byoun, S. J. (2010). Wife battering in South Korea: An ecological systems analysis. Children and Youth Services Review, 32(12), 1623–1630. Hong, J. S., Lee, N. Y., Park, H. J., & Faller, K. C. (2011). Child maltreatment in South Korea: An ecological systems analysis. Children and Youth Services Review, 33(7), 1058–1066. Jennings, W. G., Park, M., Tomsich, E. A., Gover, A. R., & Akers, R. L. (2011). Assessing the overlap in dating violence perpetration and victimization among South Korean college students: The influence of social learning and self-control. American Journal of Criminal Justice, 36, 188–206. Jennings, W. G., Richards, T. R., Tomsich, E. A., Gover, A. R., & Powers, R. A. (2013). A critical examination of the causal link between child abuse and adult dating violence perpetration and victimization from a propensity score matching approach. Women and Criminal Justice, 23, 167–184. Johnson, M. P., & Ferraro, K. J. (2000). Research on domestic violence in the 1990: Making distinctions. Journal of Marriage and the Family, 62(4), 948–963. Jones, G. W. (2005). The “flight from marriage” in Southeast and East Asia. Journal of Comparative Family Studies, 36(1), 93–119. Kendra, R., Bell, K. M., & Guimond, J. M. (2012). The impact of child abuse history, PTSD symptoms, and anger arousal on dating violence perpetration among college women. Journal of Family Violence, 27(3), 165–175. Kim, C. S., & Jeong, Y. S. (2002). Survey of child abuse and awareness of teachers about child abuse in Bucheon city. Sahoe Bokji Review, 7, 145–166. Kim, J. Y., Kim, H. J., Choi, J. W., & Emery, C. (2014). Family violence and dating violence in Korea. Journal of Family Violence, 29, 23–33. Kim, J. Y., & Lee, J. H. (2011). Factors influencing help-seeking behavior among battered Korean women in intimate relationships. Journal of Interpersonal Violence, 26(15), 2991–3012. Kim, J. Y., & Sung, K. (2000). Conjugal violence in Korean American families: A residue of the cultural tradition. Journal of Family Violence, 15(4), 331–345. Kreider, R. M., & Simmons, T. (2003). Marital status: 2000. Census 2000 brief (BOC publication No. C2KBR-30). Washington, DC: U.S. Census Bureau. Retrieved from http://www.census.gov/prod/2003pubs/c2kbr-30.pdf Laporte, L., Jiang, D., Pepler, D. J., & Chamberland, C. (2011). The relationship between adolescents’ experience of family violence and dating violence. Youth & Society, 43(1), 3–27. Loughran, T. A., Mulvey, E. P., Schubert, C. A., Fagan, J., Piquero, A. R., & Losoya, S. H. (2009). Estimating a dose–response relationship between length of stay and future recidivism in serious juvenile offenders. Criminology, 47, 699–740. Mihalic, S. W., & Elliott, D. (1997). A social learning theory model of marital violence. Journal of Family Violence, 12(1), 21–47. Millett, L. S., Kohl, P. L., Jonson-Reid, M., Drake, B., & Petra, M. (2013). Child maltreatment victimization and subsequent perpetration of young adult intimate partner violence: An exploration of mediating factors. Child Maltreatment, 18(2), 71–84. Morton, N., & Browne, K. D. (1998). Theory and observation of attachment and its relation to child maltreatment: A review. Child Abuse & Neglect, 22(11), 1093–1104. Pearl, J. (2009). Causality: Models, reasoning, and inference (2nd ed.). New York: Cambridge University Press. Postmus, J. L., & Hahn, S. A. (2007). Comparing the policy response to violence against women in the USA and South Korea. International Social Work, 50(6), 770–782. Raudenbush, S. W., & Xiaofeng, L. (2000). Statistical power and optimal design for multisite randomized trials. Psychological Methods, 5, 199–213. Rebellon, C. J., Straus, M. A., & Medeiros, R. A. (2008). Self-control in global perspective: An empirical assessment of Gottfredson and Hirschi’s general theory within and across 32 national settings. European Journal of Criminology, 5(3), 331–362. Renner, L. M., & Whitney, S. D. (2012). Risk factors for unidirectional and bidirectional intimate partner violence among young adults. Child Abuse & Neglect, 36(1), 40–52. Riggs, S. A., & Kaminski, P. (2010). Childhood emotional abuse, adult attachment, and depression as predictors of relational adjustment and psychological aggression. Journal of Aggression, Maltreatment & Trauma, 19(1), 75–104. Rogers, W., Ballantyne, A., & Draper, H. (2007). Is sex-selective abortion morally justified and should it be prohibited? Bioethics, 21(9), 520–524. Rosenbaum, P. R. (2002). Observational studies (2nd ed.). New York: Springer-Verlag. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observation studies for casual effects. Biometrika, 70, 41–55.

W.G. Jennings et al. / Child Abuse & Neglect 38 (2014) 1902–1913

1913

Sampson, R. J., Laub, R. J., & Wimer, C. (2006). Does marriage reduce crime? A counterfactual approach to within-individual causal effects. Criminology, 44, 465–508. Saunders, B. E. (2003). Understanding children exposed to violence toward an integration of overlapping fields. Journal of Interpersonal Violence, 18(4), 356–376. Sellers, C. S. (1999). Self-control and intimate violence: An examination of the scope and specification of the general theory of crime. Criminology, 37(2), 375–404. Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Boston: Houghton Mifflin. Smith, C., & Thornberry, T. P. (1995). The relationship between childhood maltreatment and adolescent involvement in delinquency. Criminology, 33(4), 451–481. Stith, S. M., Rosen, K. H., Middleton, K. A., Busch, A. L., Lundeberg, K., & Carlton, R. P. (2000). The intergenerational transmission of spouse abuse: A meta-analysis. Journal of Marriage and Family, 62, 640–654. Stith, S. M., Smith, D. B., Penn, C. E., Ward, D. B., & Tritt, D. (2004). Intimate partner physical abuse perpetration and victimization risk factors: A meta-analytic review. Aggression and Violent Behavior, 10(1), 65–98. Straus, M. A. (1991). Discipline and deviance: Physical punishment of children and violence and other crime in adulthood. Social Problems, 38, 133–154. Straus, M. A. (2009). The national context effect: An empirical test of the validity of cross national research using unrepresentative samples. Cross-Cultural Research, 43(3), 183–205. Straus, M. A. (2010). Prevalence, societal causes, and trends in corporate punishment by parents in world perspective. Law & Contemporary Problems, 73(1), 1–30. Straus, M. A., Hamby, S. L., Boney-McCoy, S., & Sugarman, D. B. (1996). The revised Conflict Tactics Scales (CTS2): Development and preliminary psychometric data. Journal of Family Issues, 17, 283–316. Wright, E. M., & Fagan, A. (2013). The cycle of violence in context: Exploring the moderating roles of neighborhood disadvantage and cultural norms. Criminology, 51(2), 217–249. Yick, A. G., & Oomen-Early, R. (2008). A 16-year examination of domestic violence among Asians and Asian Americans in the empirical knowledge base: A content analysis. Journal of Interpersonal Violence, 23, 1075–1094.