J Fam Viol (2008) 23:303–313 DOI 10.1007/s10896-008-9152-0
ORIGINAL ARTICLE
Gender Differences in the Connections Between Violence Experienced as a Child and Perpetration of Intimate Partner Violence in Young Adulthood Xiangming Fang & Phaedra S. Corso
Published online: 8 February 2008 # Springer Science + Business Media, LLC 2008
Abstract This paper uses longitudinal and nationally representative survey data to investigate the direct relationship between three forms of child maltreatment (neglect, physical abuse, and sexual abuse), and future intimate partner violence (IPV) perpetration in the USA. We further examine the indirect effect that child maltreatment has on future IPV perpetration through the presence of youth violence perpetration, and the roles of socioeconomic factors on committing youth violence and IPV. Analyses indicate that gender differences exist for the developmental relationship between child maltreatment and young adult IPV perpetration, and the effects of socioeconomic factors on youth violence and IPV perpetration. For males, the direct effects of being neglected/physically abused as a child on IPV perpetration are not significant. However, the indirect effects of being neglected/physically abused on IPV perpetration through the presence of youth violence perpetration are significant. For females, the direct effects of being neglected/physically abused on IPV perpetration are significant. The indirect effect of being neglected on IPV perpetration is significant, while the indirect effect of childhood physical abuse is not significant. Childhood sexual abuse is not significantly directly associated with IPV perpetration for females; however, for males, it is the
X. Fang (*) National Center for Injury Prevention and Control, Centers for Disease Control and Prevention, 4770 Buford Highway, MS F-64, Atlanta, GA 30341, USA e-mail:
[email protected] P. S. Corso Department of Health Policy and Management, College of Public Health, University of Georgia, Athens, GA 30602, USA
strongest (i.e., largest effect size) direct predictor of IPV perpetration. The indirect effects of childhood sexual abuse on IPV perpetration are not significant for both females and males. Keywords Child maltreatment . Intimate partner violence . Gender differences . Socioeconomic factors Intimate partner violence (IPV) is a national problem affecting millions of adults each year in the USA. Results from the National Violence Against Women (NVAW) survey conducted from November, 1995 to May, 1996 indicate that each year an estimated 8.5 million intimate partner victimizations occur among US population ages 18 and older. This violence results in approximately 2 million injuries requiring medical attention and 1,300 deaths annually (Tjaden and Thoennes 2000). The costs of IPV against women alone exceed an estimated $8.3 billion each year (CDC 2003; Max et al. 2004).1 As recognition of IPV as a serious societal problem increases, more attention has been directed to risk and protective factors for IPV perpetration, especially the link between child maltreatment victimization and future perpetration of IPV. For example, several studies have found that children who have experienced child maltreatment (neglect, physical abuse, or sexual abuse) are more likely as adults to conduct IPV (Black et al. 1999; Ehrensaft et al. 2003; Magdol et al. 1998; Stith et al. 2000); however, there is little agreement regarding the mechanisms that link the two phenomena. Several proposed theories to explain the link between child maltreatment and IPV perpetration include social learning theory (Bandura 1977; Bandura et al. 1962;
1
The number $8.3 billion is measured in 2003 dollars.
304
Feshbach 1980), attachment theory (Egeland et al. 1987, 1988), ecological or multifactor approach (Belsky 1980; Langeland and Dijkstra 1995), and biological or genetic theories of aggressive behavior (Muller et al. 1995). Of these theories, social learning theory is the predominate theory espoused in existing literature. Based on social learning theory, two hypotheses are proposed that provide contrasting ideas about factors that link child physical abuse (one form of child maltreatment) and IPV perpetration. The first hypothesis is that exposure to physical abuse teaches children that violence toward intimates is legitimate and that violence is an effective way of resolving conflicts with partners (O’Leary 1988; Simons et al. 1995; Straus and Smith 1990; Straus et al. 1980). Thus, experiencing physical abuse during childhood may directly increase the probability that children will grow up to behave aggressively toward their intimate partners. Simons et al. (1995) labeled this hypothesis the family roles perspective. Another hypothesis is that the link between physical abuse and IPV perpetration can be mediated by, or be indirectly affected by a child’s antisocial behavior. This hypothesis, labeled by Simons et al., as the antisocial behavior trait perspective, proposes that childhood exposure to physical abuse first increases the risk for IPV perpetration through promoting violent behavior in adolescence (youth violence). This aggressive behavior is then carried through to adulthood and is used with intimate partners (Capaldi and Clark 1998; Herrenkohl et al. 2004; Simons et al. 1995). Although both views build upon classic social learning theory, the first emphasizes the direct effect of physical abuse on adult IPV perpetration while the second focuses on the indirect effect of physical abuse on adult IPV perpetration through its effect on youth violence. Tests of these two hypotheses have yielded conflicting results. Simons et al. (1995) found support for the antisocial behavior trait perspective over the family roles perspective to explain the link between physical abuse and IPV perpetration for both men and women. However, their findings may not be generalizable to the US population due to the specificity of the sample used in the study, which consisted of all white, two-parent families living in small towns. Capaldi and Clark (1998) also found that the effect of unskilled parenting (poor discipline) on male children’s later aggression toward an intimate partner is mediated by the development of antisocial behavior during adolescence. However, because participants were males recruited from schools with a higher than average delinquency rate for a medium-sized metropolitan area, it is likely that the sample that Capaldi and Clark used in their study may not be representative of other populations. Therefore, their findings may have limited generalizability. In contrast to the findings of Simons et al. and Capaldi et al., Herrenkohl et al. (2004) found no significant pathway
J Fam Viol (2008) 23:303–313
from child physical abuse through youth violence to IPV for either males or females. Herrenkohl et al. found that a strong direct effect of physical abuse on IPV was maintained for males, but not for females. However, Herrenkohl et al.’s findings may also have limited generalizability, since participants were recruited from public elementary schools serving predominantly high-crime neighborhoods. Given the limited generalizability of the above findings to the entire US population, there is a need for further research with nationally representative samples to test the link between child maltreatment, youth violence perpetration, and IPV perpetration. Many previous studies examined the mechanisms linking physical abuse and IPV perpetration, such as antisocial behavior trait perspective vs. family roles perspective. Much less has been done to examine the mechanisms linking other forms of child maltreatment (neglect and sexual abuse) and IPV perpetration. However, previous work on testing antisocial behavior trait perspective vs. family roles perspective in explaining the link between physical abuse and IPV perpetration may shed some light on the possible mechanisms linking other forms of child maltreatment (neglect and sexual abuse) and IPV perpetration. For example, the first hypothesis would argue for an indirect effect of neglect or sexual abuse on IPV perpetration through its effect on youth violence perpetration, while the second hypothesis would suggest a direct relationship between neglect (or sexual abuse) and IPV perpetration even after controlling for youth violence perpetration. Another of Herrenkohl et al.’s findings suggests a different element of complexity in explaining the relationship between child maltreatment and IPV perpetration–gender differences. Several other researchers have suggested that the intergenerational transmission of violence may operate differently for males and females (Langhinrichsen-Rohling et al. 1995; Mihalic and Elliott, 1997; Stith et al. 2000). However, studies that have investigated gender differences in the link between child maltreatment and IPV perpetration have produced inconsistent findings. For example, Magdol et al. (1998) found that the link between abusive discipline experienced as a child and IPV perpetration is stronger for females. Doumas et al. (1994) found that abused boys are at higher risk for perpetrating partner violence as adults. Additional limitations of previous studies of the link between child maltreatment and IPV include inadequate controlling for confounding individual socioeconomic, family background and contextual factors. For example, most studies examine the relationship between child maltreatment and IPV, while not simultaneously controlling for the effects of other factors such as parent education, family poverty, and neighborhood disadvantages. Evidence has suggested that IPV, youth violence and child maltreatment are not uniformly distributed across the population
J Fam Viol (2008) 23:303–313
and are more likely to occur in families characterized by social and economic disadvantage, parental separation and divorce, and families living in disadvantaged neighborhoods (Fergusson et al. 1986, 1992; Gelles and Cornell 1990; Mocan and Rees 2005; Sampson et al. 1997; Wikstrom and Loeber 2000). Without controlling for these factors, we do not know whether the link between child maltreatment and IPV arises because of the socioeconomic and family context within which IPV, youth violence, and child maltreatment occur, or arise through a cause and effect relationship in which experiencing child maltreatment increases the risk for later youth violence and IPV perpetration. In this study, we use a longitudinal and nationally representative sample to investigate the direct relationship between three forms of child maltreatment (neglect, physical abuse, and sexual abuse) and IPV, expanding on previous studies which have focused solely on the link between physical abuse and IPV. We further estimate and test the indirect effect that child maltreatment has on future IPV perpetration through the presence of youth violence, and the role that gender plays in the developmental relationship. This study advances the methodology in the field of violence prevention research in testing the significance of indirect effects. Previous research has measured evidence of mediation by showing significant paths to and from an intervening variable and by a reduction in the direct effects of the exogenous variables on the outcome upon the addition of the mediator to the model. However, this research is incomplete because the standard errors of indirect effects were not estimated, therefore evidence of mediation (indirect effect) has not previously been directly tested. By applying a recursive bivariate probit model plus bootstrapping techniques, our methodology has allowed us to estimate the standard errors of indirect effects, and therefore directly test the significance of indirect effects.
Methods Research Hypotheses and Empirical Models Based on previous literature, we investigated three hypotheses for the link between child maltreatment, youth violence perpetration, and IPV perpetration: (H1) Controlling for family background factors, adolescent individual factors, and adolescent community factors, child maltreatment (neglect, physical abuse, or sexual abuse) increases risk for youth violence perpetration; (H2) Controlling for family background factors and young adult individual factors, child maltreatment (neglect, physical abuse, or sexual abuse) increases risk for IPV perpetration indirectly
305
through their effects on youth violence behavior; and (H3) Child maltreatment (neglect, physical abuse, or sexual abuse) has a direct effect on IPV perpetration even after controlling for youth violence behavior and other variables. Following the hypotheses, our empirical model consists of the following two equations: IPVi ¼ α0 þ α1 Negi þ α2 PhyAi þ α3 SexAi 0 0 0 þα4 Di þ α5 Fami þ α6 AdultIndi þ 1 YVi þ "i IPVi ¼ 1 ðperpetration of IPVÞ if IPVi > 0 IPVi ¼ 0 ðnon perpetration of IPVÞ if IPVi 0 ð1Þ YVi ¼ β 0 þ β1 Negi þ β2 PhyAi þ β3 SexAi 0 0 0 0 þβ4 Di þ β5 Fami þ β6 AdolIndi þ β 7 Contexti þ ν i YVi ¼ 1 ðperpetration of youth violenceÞ if YVi > 0 YVi ¼ 0 ðnon perpetration of youth violenceÞ if YVi 0 ð2Þ Where IPVi and YVi indicate an individual’s unobserved propensities to engage in IPV perpetration and youth violence perpetration, respectively, Di denotes a vector of demographic variables, Fami denotes a vector of family background characteristics, AdolIndi denotes a vector of individual characteristics during adolescence, AdultIndi denotes a vector of individual characteristics during young adulthood, and Contexti denotes a vector of community contextual factors during adolescence. Negi is equal to 1 if individual i experienced neglect during childhood and equal to 0 otherwise. Similarly, PhyAi ðSexAi Þ is equal to 1 if individual i experienced physical abuse (sexual abuse) during childhood and equal to 0 otherwise. To allow for possible correlation between the omitted factors in the two equations, we assume ɛi and νi are distributed bivariate normal, and E ½"i ¼ E ½ni ¼ 0; Var½"i ¼ Var½n i ¼ 1, and Cov½"i ; n i ¼ r. With these assumptions, the model given by Eqs. 1 and 2 is a standard, recursive bivariate probit regression model (Greene 2003). Notice that youth violence perpetration is not only a dependent variable in Eq. 2, but also an independent variable in Eq. 1 which introduces all the indirect effects as we mentioned in our hypotheses. The model is estimated by maximum likelihood methods. Direct effects and/or indirect effects of a variable which might appear in Eq. 1 and/or Eq. 2 are calculated based on the formulas appearing in Greene (1998). Bootstrapping techniques were used to estimate the standard errors for all indirect effects.2 All analyses were conducted with STATA SE version 9. 2
Bootstrapping is the current preferred method for nonlinear estimators and has largely replaced the delta method (See Horowitz, 2001; Efron and Tibshirani, 1993).
306
J Fam Viol (2008) 23:303–313
Data
two survey questions reflecting physical violence against partners from Wave III of Add Health study:
Data for this study comes from the National Longitudinal Study of Adolescent Health (Add Health), a nationally representative study examining the health-related behaviors and outcomes of adolescents in grades 7 through 12 and their outcomes in young adulthood (Udry 2003). Individual, family, school, and community-level information was collected in two waves between 1994 (Wave I) and 1996 (Wave II). A third wave was conducted among Wave I respondents in 2001 and 2002 (Wave III) to examine the effects of adolescent experiences and factors on subsequent adult outcomes. Details regarding sampling methods and interviewing techniques are described elsewhere (Harris et al. 2003; Udry 2003). Briefly, in-home interviews were conducted for Wave I and Wave III using laptop computers in order to protect confidentiality and prevent interviewer or parental influence. Sample weights were computed for 18,924 respondents at Wave I, which adjust for stratification, oversampling, and other sampling issues to produce a nationally representative sample. Wave III data collection involved 14,322 original Wave I respondents, aged 18 to 26 at the time of Wave III interview, who have a sampling weight at Wave I and could be relocated and reinterviewed between April, 2001 and April, 2002. Additional sample weights were computed at Wave III for use in cross-sectional and longitudinal analyses. For this study, we used data only for respondents who reported being involved in at least one relationship in the 2 years preceding the Wave III survey. This reduced the sample for analyses to 10,320 participants. An additional 968 respondents were excluded because they were missing information outcome variables or covariates, resulting in a final study sample of 9,352 participants. Of the study participants, 5,190 (55.5%) are female. After applying the adjustment of sample weights, females account for 51.9% of the sample. For the participants in the study sample, data describing selfreported youth violence behaviors from Wave I of Add Health study were matched with self-reported IPV perpetration within young adult sexual relationships and retrospective reports of child maltreatment during Wave III of Add Health study.
&
Measures Dependent Variables The measures of IPV perpetration (IPVi) and youth violence perpetration (YVi) used in the analyses are dichotomous measures based on participants’ yes/no responses to multiple items. The measure of IPV perpetration includes
&
Threatened partner with violence, pushed, shoved, or threw something at partner Slapped, hit or kicked partner
Participants are categorized as IPV perpetrators if they reported conducting any of the listed acts of violence during the preceding two years. The measure of youth violence perpetration includes five items reflecting interpersonal violence from Wave I of Add Health study: & & & & &
Took part in a group fight Hurt someone badly enough to need bandages or care from a doctor or nurse Used or threatened to use a weapon to get something from someone Pulled a knife or gun on someone Shot or stabbed someone
Similarly, participants are categorized as youth violence perpetrators if they reported committing any of the listed offenses during the preceding 1 year. Independent Variables The measures of the three types of child maltreatment (neglect, physical abuse, and sexual abuse) are also dichotomized on the basis of respondents’ retrospective reports of child maltreatment prior to when the respondent started sixth grade. The measure of neglect is constructed using the following two questions from Wave III of Add Health: &
How often had your parents or other adult care-givers & &
Left you home alone when an adult should have been with you? Not taken care of your basic needs, such as keeping you clean or providing food or clothing?
Following Leeb et al.’s (2004) work, neglect is coded “1” if the respondent reported that care-givers had left him or her home alone as a child more than five times when an adult should have been with the child, or had not taken care of his or her basic needs at least once, and “0” otherwise. The measure of physical abuse is constructed using the following question from Wave III of Add Health: &
How often had your parents or other adult care-givers slapped, hit, or kicked you?
Again, following Leeb et al.’s work, physical abuse was coded “1” if the respondent reported that care-givers had slapped, hit, or kicked him or her more than 5 times, and
J Fam Viol (2008) 23:303–313
“0” otherwise. The measure of sexual abuse uses the following question from Wave III of Add Health: &
How often had one of your parents or other adult caregivers touched you in a sexual way, forced you to touch him or her in a sexual way, or forced you to have sexual relations?
Sexual abuse is coded “1” if the respondent reported care-givers had touched the child in a sexual way, forced the child to touch him or her in a sexual way, or forced the child to have sexual relations at least once, and a “0” reflects no history of childhood sexual abuse. Our models control for the following demographic variables (Di)—age, age squared,3 gender, and race. Other control variables include an adolescent individual factor (AdolIndi), adolescent community factors (Contexti), family background characteristics (Fami), and individual characteristics during young adulthood (AdultIndi). The adolescent individual factor is religiosity during adolescence. Previous research has found individual religiosity is a protective factor for both IPV and youth violence perpetration (e.g., Cunradi et al. 2002; Resnick et al. 2004). Moreover, individual religiosity is found to be impacted by parental religiosity and parental religiosity is associated with the risk of perpetrating child maltreatment (e.g., Dyselin and Thomsen 2005). Therefore, individual religiosity may be related to both child maltreatment and subsequent youth violence and IPV and is included in this study as a potential confounding variable. Adolescent community factors include an index of community economic disadvantage (CED), which was constructed using the following three county-level items drawn from the 1990 US Census: & & &
Proportion of family with income below poverty Proportion aged 25+ without high school diploma or equivalency Unemployment rate
The overall Cronbach alpha for this scale is 0.79. We standardized each item individually and then took the mean to construct the CED index. Following Levitt (1998), we also include the total number of crimes in the county per 100,000 population to control for the impact of omitted factors that may influence youth violence. Family background characteristics include parent (mother or the main caregiver) education, family poverty, and whether the participant is from two-parent family. Parent education is divided into six categories (coded 1 to 6): “8th grade or less (includes “Never went to school”)”, “More
3
Because of the shape of age–crime curve (see Farrington, 1986; Hindelang, 1981), we also include age squared as a control variable.
307
than 8th grade but did not graduate from high school”, “High school graduate (or equivalent)”, “Some post-high school training/college”, “College graduate”, and “Postgraduate or professional training”. A dichotomous family poverty indicator was constructed using the household income measure such that a value of one represented annual household income of $20,000 or less, and zero otherwise. Individual characteristics during young adulthood include marital status, school enrollment, employment status, religiosity during young adulthood and own educational attainment. Own educational level is measured by the highest grade or year of regular school that the participants have completed.
Table 1 Descriptive statistics of the sample Variable
Mean (SD) or % (SE) Female
Independent variables IPV perpetration 30.2% (0.6) Youth violence 23.4% (0.6) Child maltreatment Neglected 16.2% (0.5) Physically abused 8.4% (0.4) Sexually abused 4.6% (0.3) Demographical characteristics Age in wave 1 (Age1) 15.4 (1.8) Age in wave 3 (Age3) 21.7 (1.8) White 69.9% (0.6) Black 14.8% (0.5) Asian 3.2% (0.3) Native 0.5% (0.1) Other race 0.7% (0.1) Hispanic 10.9% (0.4) Individual factors during adolescence Religiosity in wave 1 0.014 (0.834) Community factors Econ. disadv. index −0.019 (0.920) County crime rate 5630 (2661) (per 100,000) Family characteristics Parent education 3.6 (1.2) Family poverty 22.1% (0.6) Two-parent family 73.0% (0.6) Individual factors during young adulthood Religiosity in wave 3 0.017 (0.763) Married 23.5% (0.6) School enrollment 38.3% (0.7) Employed 68.2% (0.6) Own education level 13.23 (2.04)
Male
15.9% (0.6) 38.4% (0.8) 22.2% (0.6) 8.4% (0.4) 4.5% (0.3) 15.54 (1.83) 21.93 (1.87) 68.4% (0.7) 14.2% (0.5) 3.3% (0.3) 1.0% (0.2) 0.9% (0.2) 12.2% (0.5) −0.124 (0.828) −0.025 (0.916) 5633 (2608)
3.6 (1.3) 21.9% (0.7) 74.4% (0.7) −0.175 (0.752) 16.5% (0.6) 32.3% (0.7) 74.1% (0.7) 12.94 (1.97)
Mean with standard deviation (SD) is presented for a continuous variable, while rate (in percentage form) with standard error (SE) is presented for a dichotomous variable
308
The means and standard deviations associated with dependent and independent variables are summarized in Table 1. All statistics listed in Table 1 and all the following analyses used sample weights to adjust for stratification and oversampling of underrepresented groups. After the adjustment of sample weights, the study sample has a very similar racial/ethnic distribution to the nation. Table 1 shows a larger percentage of females than males reported engaging in IPV perpetration, while a larger percentage of males than females reported engaging in YV perpetration. These numbers are consistent with evidence suggesting that females perpetrate some forms of violence at rates proportional or higher to males (Archer 2002; Magdol et al. 1998; Moffitt et al. 2001; Morse 1995; Straus and Gelles 1990). In terms of child maltreatment, for males, 22.2% of the sample reported having experienced neglect during their childhood, while 16.2% of females reported neglect. For both males and females, approximately 8.4% were physically abused and 4.5% were sexually abused during their childhood.
Results Bivariate Probit Results We first test the significance of gender effect. Following Greene (2003), we include the interaction terms between the gender dummy and each variable in the IPV and youth violence equations in the pooled regressions, and then test the hypothesis that these interactions are jointly significantly different from zero. The chi-squared statistic with 19 degrees of freedom under the null hypothesis is 31.68, while the critical value at 95% confidence is 30.14. Since the test statistic suggests that the regression models for males and females are systematically different, subsequent regressions are estimated separately for males and females. We start our maximum likelihood estimation of the bivariate probit model by allowing the correlation between the disturbances of Eqs. 1 and 2, ρ, to vary freely. The estimated value of ρ is 0.1986 for females and 0.0862 for males, with the chi-squared statistic equal to 0.2744 and 0.0272 respectively. The critical value at 95% confidence with one degree of freedom is 3.84, so the null hypothesis that ρ equals zero is not rejected, suggesting that the two structural disturbances may well be uncorrelated for both male and female bivariate probit regressions. Knapp and Seaks (1998) show that a likelihood-ratio test of whether the correlation coefficient of the disturbances in Eqs. 1 and 2 is equal to zero can be used as a Hausman endogeneity test. The low values of chi-squared (0.2744 and 0.0272) suggest there is no evidence against the hypothesis that youth violence perpetration is exogenous. Based on these
J Fam Viol (2008) 23:303–313 Table 2 Results from bivariate probit model by gender Variable
Coefficient (Robust SE) Female
Equation for youth violence perpetration Neglected 0.208 (0.084)** Physically abused 0.201 (0.106)* Sexually abused −0.040 (0.155) Parent education −0.103 (0.028)*** Family poverty 0.227 (0.092)** Two-parent family 0.026 (0.067) Religiosity in wave 1 −0.150 (0.038)*** Econ. disadv. index −0.032 (0.031) County crime rate (%) 0.007 (0.014) Equation for IPV perpetration Youth violence 0.220 (0.058)*** Neglected 0.182 (0.088)** Physically abused 0.222 (0.111)** Sexually abused 0.200 (0.153) Parent education −0.022 (0.028) Family poverty −0.048 (0.069) Two-parent family −0.066 (0.073) Religiosity in wave 3 −0.107 (0.039)*** Married 0.019 (0.064) School enrollment −0.019 (0.066) Employed 0.070 (0.058) Own education level −0.053 (0.017)***
Male
0.252 (0.083)*** 0.310 (0.120)** 0.105 (0.181) −0.081 (0.029)*** −0.089 (0.111) −0.092 (0.092) −0.025 (0.040) 0.011 (0.029) 0.027 (0.012)** 0.247 (0.092)*** 0.093 (0.103) 0.060 (0.159) 0.621 (0.204)*** −0.081 (0.036)** 0.071 (0.095) 0.055 (0.104) 0.022 (0.050) 0.154 (0.099) −0.110 (0.096) −0.095 (0.096) −0.035 (0.023)
*p<0.10 **p<0.05 ***p<0.01
results, the models were reestimated with ρ constrained to equal zero. We also considered the possibility that major risk factors (three types of child maltreatment, youth violent behavior, family background characteristics and community factors) might interact with each other. Because regression results reveal that none of the interactions among these factors are significant, our subsequent regression models do not include any interaction terms. Table 2 presents the results from the bivariate probit model with ρ=0 for both males and females. Neglect and physical abuse are at least marginally significantly associated with youth violence for both males and females. However, both neglect and physical abuse are more strongly predictive of youth violence for males than for females. Controlling for other variables, being involved in youth violence is strongly predictive of young adult IPV perpetration for both males and females. For females, both childhood neglect and physical abuse are significantly directly associated with IPV. However, for males, the direct associations are not significant. Sexual abuse is significantly directly associated with IPV perpetration for males, but not significantly associated with IPV perpetration for females.
J Fam Viol (2008) 23:303–313
309
Almost no family background characteristics included in this study are directly and significantly associated with IPV perpetration for either males or females, except that for males, parent education is directly and significantly associated with IPV perpetration. For the relationship between family background characteristics and youth violence perpetration, low parent education and family poverty significantly increase the risk of perpetrating youth violence for females, while for males, only low parent education significantly increases the likelihood of youth violence perpetration. For individual factors, religiosity is a significant protective factor for both IPV and youth violence perpetration for females, but it is not for males. Community economic disadvantage is not significantly associated with youth violence perpetration for either males or females. Although living in a high crime neighborhood significantly increases the risk of perpetrating youth violence for males, this effect is not significant for females. In the IPV equation, high own educational attainment is a significant protective factor for young adult IPV perpetration for females, but not for males. Other young adulthood individual socioeconomic factors, such as marital status, school enrollment, and employment status, are not significantly associated with IPV perpetration for both males and females. Direct and Indirect Effects The estimated marginal effects of the main variables in the IPV and youth violence equations are presented in Table 3.
Controlling for other variables, perpetration of youth violence is strongly and significantly associated with young adult IPV perpetration for both males and females. Being involved in violent behavior as a youth increases the likelihood of IPV perpetration by 7.78% for females and 5.55% for males, respectively. For females, being neglected as a child not only has a positive and significant direct effect (6.46%) on adult IPV perpetration but also a positive and significant indirect effect (0.57%) on adult IPV perpetration through its effect on youth violence behavior. For males, the direct effect (2.09%) is not significant but the indirect effect (0.57%) is significant. In terms of childhood physical abuse, being physically abused as a child has a positive and significant direct effect (7.96%) on young adult IPV perpetration for females, while the indirect effect (0.57%) is not significant. In contrast, the indirect effect (0.71%) is marginally significant for males, while the direct effect (1.34%) is not significant. Being neglected as a child is associated with a 6.37% increase in the probability of committing youth violence for females, and a 9.63% increase in the likelihood of committing youth violence for males. Being physically abused as a child increases the likelihood of youth violence perpetration by 6.22% for females, and 12.03% for males. Due to the nonsignificant associations between sexual abuse and youth violence, the indirect effects of sexual abuse on young adult IPV perpetration through the development of violent behavior as a youth are not significant for either females or males. For females, the direct effect of sexual abuse on young adult IPV perpetra-
Table 3 Estimated direct and indirect effects by gender (standard errors in parentheses) Variables
Youth violence Neglected
Youth violence Females
Males
Females
Direct (total)
Direct (total)
Direct
–
– 0.0637** (0.0271)
Physically abused Sexually abused
0.0622* (0.0345) −0.0114 (0.0437)
Parent education
−0.0301*** (0.0081) 0.0692** (0.0296) 0.0074 (0.0194) –
Family poverty Two-parent family Own education level County crime rate (%)
IPV perpetration
0.0020 (0.0042)
0.0778*** (0.0211) 0.0963*** (0.0324)
0.0646** (0.0320)
0.1203** (0.0475) 0.0401 (0.0697)
0.0796** (0.0414) 0.0720 (0.0572)
−0.0304*** (0.0110) −0.0332 (0.0406) −0.0346 (0.0349) – 0.0101** (0.0047)
−0.0075 (0.0095) −0.0165 (0.0234) −0.0228 (0.0255) −0.0182*** (0.0057) –
Males Indirect
Direct
−
0.0555*** (0.0213) 0.0209 (0.0237)
0.0057** (0.0027) 0.0057 (0.0035) −0.0006 (0.0035) −0.0024** (0.0010) 0.0053* (0.0029) 0.0004 (0.0019) – 0.0002 (0.0003)
Bootstrap estimates of the standard errors for indirect effects came from 1,000 bootstrap samples *p<0.10 **p<0.05 ***p<0.01
0.0134 (0.0362) 0.1763*** (0.0707) −0.0176** (0.0079) 0.0158 (0.0216) 0.0118 (0.0219) −0.0076 (0.0049) –
Indirect − 0.0057** (0.0029) 0.0071* (0.0039) 0.0029 (0.0072) −0.0017* (0.0009) −0.0019 (0.0022) −0.0018 (0.0021) – 0.0006 (0.0004)
310
tion is also not significant. For males, however, childhood sexual abuse directly and significantly increases the likelihood of young adult IPV perpetration by 17.63% even though the indirect effect is not significant. Several familial background characteristics have significant effects on IPV and youth violence perpetration. For both males and females, the higher the educational level of the child’s parent, the lower the likelihood of conducting violence as a youth. For females, the indirect effect of parent education on IPV perpetration is negative and significant. For males this indirect effect is marginally significant. The direct effect of parent education on IPV perpetration is negative and significant for males, while for females, the direct effect is negative but not significant. Living in a poor family significantly increases the likelihood of youth violence perpetration by 6.92% for females, but has a nonsignificant effect for males. The indirect effect of family poverty on adult IPV perpetration is positive and marginally significant for females (0.53%), while it is not significant for males. The direct effect of family poverty on IPV perpetration is not significant for either females or males. Living with both parents is not significantly associated with youth violence or IPV perpetration, for either females or males. Living in a high crime neighborhood is not significantly associated with youth violence perpetration for females, but has a significant effect on youth violence perpetration for males. A one percentage point decrease in the county crime rate reduces the likelihood of youth violence perpetration by 1.01% for males. The indirect effect of living in a high crime neighborhood on young adult IPV perpetration is not significant for either females or males.
Discussion We used data from a nationally representative longitudinal study (Add Health study) to examine the developmental relationships between three forms of child maltreatment victimization (neglect, physical abuse, and sexual abuse) and adult IPV perpetration, and the gender differences in the developmental relationships. Our study finds strong support for the gender difference in the link between child maltreatment and IPV perpetration. Overall, childhood neglect/physical abuse is a stronger predictor of young adult IPV perpetration for females than males, while childhood neglect/physical abuse is a stronger predictor of youth violence perpetration for males than females. Childhood sexual abuse is a significant predictor of IPV perpetration for males, but not a significant predictor of IPV perpetration for females. For males, the indirect effect of being physically abused on later young adult IPV perpetration is significant while
J Fam Viol (2008) 23:303–313
the direct effect is not significant. These results, which are consistent with Simons et al.’s (1995) and Capaldi and Clark’s (1998) findings, support the theory of antisocial behavior trait perspective over family roles perspective to explain the developmental relationship between being physically abused as a child and young adult IPV perpetration for males. The link between neglect and IPV perpetration is similar to that between physical abuse and IPV perpetration. The significant indirect effect (but nonsignificant direct effect) of neglect on IPV perpetration suggests that the theory of antisocial behavior trait perspective may also be applied in explaining the link between neglect and IPV perpetration, i.e., like their physically abused counterparts, male neglected children may foster an antisocial orientation that, in turn, increases the risk for both youth violence and IPV perpetration. The indirect effect of childhood sexual abuse on IPV perpetration is not significant for males, but it is the strongest direct predictor for IPV perpetration. Obviously, the results contradict the antisocial behavior trait framework. If the theory of family role perspective is applied to explain these findings, it suggests that exposure to sexual abuse teaches male children that violence toward an intimate is legitimate and therefore increases the risk that children will grow up to behave aggressively toward their intimate partners. However, further etiologic research is needed to examine the appropriateness of using this theory in explaining the link between sexual abuse and IPV perpetration for males, or whether some other theories may better explain the findings that compared to male nonmaltreated children, male sexually abused children may grow up to behave more aggressively toward intimates but not the public in general. For females, the direct effect of being physically abused on later IPV perpetration is significant, while the indirect effect is not. These results suggest that for females, the theory of family role perspective may better explain the link between childhood physical abuse and young adult IPV perpetration than the theory of antisocial behavior trait perspective. In contrast, both the indirect and direct effects of being neglected as a child on young adult IPV perpetration are significant. These results suggest that both the family role and antisocial behavior trait perspectives may explain the link between childhood neglect and young adult IPV perpetration for females to some extent. However, the magnitude of direct effect is much greater than that of indirect effect. Thus, the theory of family role hypothesis might be more effective than the theory of antisocial behavior trait hypothesis in explaining the developmental relationship between childhood neglect/ physical abuse and young adult IPV perpetration for females. However, as we mentioned before, since the theory of family role perspective was originally proposed
J Fam Viol (2008) 23:303–313
in explaining the link between physical abuse and IPV perpetration, more research is needed to examine the appropriateness of applying this theory in explaining the link between neglect and IPV perpetration. In contrast to our results with the male sample, both direct and indirect effects of childhood sexual abuse on young adult IPV perpetration are not statistically significant for females. These results suggest that gender makes a difference when considering the mechanisms linking physical abuse/ neglect and adult IPV perpetration for males and females. Thus, the theory of antisocial behavior trait perspective may be more effective than the theory of family role perspective in explaining the developmental relationship between physical abuse/neglect and adult IPV perpetration for males; while the theory of family role perspective may be more effective than the theory of antisocial behavior trait perspective in explaining the developmental relationship between physical abuse/neglect and adult IPV perpetration for females. One explanation for the gender differences is that, in the face of maltreatment, males may tend to develop externalizing problems, whereas females may tend to develop internalizing problems (e.g., depression, rejection sensitivity; e.g., Crijnen et al. 1997; Kratzer and Hodgins 1997). If so, males’ maltreatment would result in general aggression against partners and peers. Females’ maltreatment may only result in violence toward an intimate partner, because intimate relationships may exacerbate the internalizing problems (e.g., if a women is rejected by her partner) and thus lead to violence. Note that either of these models could be applied to adolescent males or females, but overall, externalizing may be more likely among adolescent males, while internalizing may be more likely among adolescent females. Limitations of the Study This study includes a number of limitations, mostly due to the nature of the secondary data. First, this study relies on retrospective reports to measure child maltreatment. The use of retrospective measures of child maltreatment has been questioned by a number of researchers (see Widom and Morris 1997; Widom et al. 1999). The most frequently cited disadvantage of retrospective measures is recall bias which may limit the validity and reliability of these reports. Second, all of our violence measures are based on the participants’ self reports. The participants may not accurately remember details of their lives, or they may choose not to tell the truth. In turn, this raises the issue of the extent to which measurement errors in the self-reported variables may have influenced the study results. In addition, the Add Health study includes only one question about childhood experience of sexual abuse by the participant’s parents or other adult care-
311
givers. Sexual abuse by strangers, therefore, is not included in the measure of sexual abuse used in this study. The third limitation is that we only investigate IPV perpetration in sexual relationships. We excluded nonsexual romantic relationships because, in contrast to data on sexual relationships, the Add Health data does not have information on the ending dates of non-sexual relationships. Therefore, we do not know if non-sexual romantic relationships occurred in the preceding two years. Excluding nonsexual romantic relationships may have some effects on our findings; however, we expect effects to be minimal since 86% of romantic relationships, as reported in the Add Health study, are sexual relationships. The fourth limitation is the potential selection bias. This study focuses on the young adults who reported sexual relationships in the preceding two years. We excluded those who did not report sexual relationships partly because we do not know if participants do not report sexual relationships because there was no sexual relationship in the preceding 2 years or because of participant non-disclosure. We examined the distribution of child maltreatment and the perpetration of youth violence between those who reported sexual relationships and those who did not, and found that the percent of those reporting child maltreatment and youth violence were approximately equal (no statistically significant difference). This might alleviate our worries about the selection bias. Finally, we were not able to account for the potential endogeneity of child maltreatment. We have a very limited number of individual and family variables included in the model. As a result, the connections that emerge between child maltreatment and IPV perpetration/youth violence perpetration may be misspecified due to the possibility that some omitted variables are sufficiently correlated with both child maltreatment and adult IPV perpetration/youth violence perpetration. This is a limitation of most work in this area. To properly address this shortcoming would require more knowledge developed in testing the endogeneity of common independent variables in bivariate probit model settings, and in studying the effects of individual, family background and community factors on child maltreatment, youth violence and IPV perpetration.
Conclusions This study finds that, in general, children who experienced child maltreatment are more likely to commit youth violence and young adult IPV perpetration. Therefore, there are additional health benefits for prevention programs and policies aimed at reducing child maltreatment than previously reported. Findings also indicate gender differences exist for the developmental relationship between childhood
312
neglect/physical abuse and IPV perpetration: the link between childhood neglect/physical abuse and IPV perpetration is stronger for females than males, whereas the link between childhood neglect/physical abuse and youth violence perpetration is stronger for males than females. Childhood sexual abuse is not significantly associated with young adult IPV perpetration for females; however, for males, it is the strongest predictor of young adult IPV perpetration. Gender differences also exist for the roles of socioeconomic factors play on perpetrating youth violence and young adult IPV. Our results suggest it may be important to account for gender differences when designing the optimal time and setting for violence prevention programs. Most IPV prevention programs have targeted middle- or high-school aged students and were set in a school setting (Whitaker et al. 2006). Our results suggest that IPV prevention programs should begin even earlier. Children who have been maltreated, especially girls who are victims of physical abuse/ neglect and boys who are victims of sexual abuse, are good candidates for IPV prevention. Our findings also indicate perpetration of youth violence is significantly associated with adult IPV perpetration for both males and females, and therefore suggest preventing youth violence may be key to preventing IPV. Acknowledgements This research uses data from Add Health, a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with cooperative funding from 17 other agencies. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 (addhealth@unc. edu). The authors would like to thank the following individuals for their helpful comments on this manuscript: Daniel Whitaker, James Mercy, Qian An, Jennifer Wyatt, Rebecca Leeb, and John R. Lutzker. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.
References Archer, J. (2002). Sex differences in physically aggressive acts between heterosexual partners: A meta-analytical review. Aggression and Violence Behavior, 7, 313–351. Bandura, A. (1977). Social learning theory. Englewood Cliffs, NJ: Prentice Hall. Bandura, A., Ross, R., & Ross, S. (1962). Transmission of aggression through imitation of aggressive models. Journal of Abnormal and Social Psychology, 63, 575–582. Belsky, J. (1980). Child maltreatment: An ecological integration. American Psychologist, 35, 320–335. Black, D. A., Schumacher, J. A., Smith, A. M., & Heyman, R. E. (1999). Partner, child abuse risk factor literature review:
J Fam Viol (2008) 23:303–313 National Network on Family Resiliency, National Network for Health. Retrieved from http://www.nnh.org/risk. Capaldi, D. M., & Clark, S. (1998). Prospective family predictors of aggression toward female partners for at-risk young men. Developmental Psychology, 24, 1175–1188. Centers for Disease Control and Prevention (CDC) (2003). Costs of intimate partner violence against women in the United States. Atlanta, GA: CDC, National Center for Injury Prevention and Control. Retrieved from http://www.cdc.gov/ncipc/pub-res/ipv_ cost/ipv.htm. Crijnen, A. A. M., Achenbach, T. M., & Verhulst, F. C. (1997). Comparisons of problems reported by parents of children in 12 cultures: Total problems, externalizing, and internalizing. Journal of the American Academy of Child & Adolescent Psychiatry, 36, 1269–1277. Cunradi, C. B., Caetano, R., & Schafer, J. (2002). Religious affiliation, denominational homogamy, and intimate partner violence among U.S. couples. Journal for the Scientific Study of Religion, 41(1), 139–151. Doumas, D., Margolin, G., & John, R. S. (1994). The intergenerational transmission of aggression across three generations. Journal of Family Violence, 9, 157–175. Dyselin, C. W., & Thomsen, C. J. (2005). Religiosity and risk of perpetrating child physical abuse: An empirical investigation. Journal of Psychology and Theology, 33(4), 291–298. Efron, B. J., & Tibshirani, R. (1993). An introduction to the bootstrap. New York: Chapman and Hall. Egeland, B., Jacobvitz, D., & Papatola, K. (1987). Intergenerational continuity of abuse. In R. J. Gelles, & J. B. Lancaster (Eds.) Child abuse and neglect: Biosocial dimensions. New York: Aldine. Egeland, B., Jacobvitz, D., & Sroufe, L. A. (1988). Breaking the cycle of abuse. Child Development, 59, 1080–1088. Ehrensaft, M. K., Cohen, P., Brown, J., Smailes, E., Chen, H., & Johnson, J. G. (2003). Intergenerational transmission of partner violence: a 20-year prospective study. Journal of Consulting and Clinical Psychology, 71, 741–753. Farrington, D. P. (1986). Age and crime. In M. Tonry, & N. Morris (Eds.) Crime and justice: An annual review of research (Vol. 7). Chicago: University of Chicago Press. Fergusson, D. M., Horwood, L. J., & Lynskey, M. T. (1992). Family change, parental discord and early offending. Journal of Child Psychology and Psychiatry, 33, 1059–1075. Fergusson, D. M., Horwood, L. J., Kershaw, K. L., & Shannon, F. T. (1986). Factors associated with reports of wife assault. Journal of Marriage and the Family, 48, 407–412. Feshbach, S. (1980). Child abuse and the dynamics of human aggression and violence. In G. Gerbner, C. J. Ross, & E. Zigler (Eds.) Child abuse: An agenda for action. New York: Oxford University Press. Gelles, R. J., & Cornell, C. P. (1990). Intimate violence in families (2nd ed.). Newbury Park, CA: Sage. Greene, W. H. (1998). Gender economics courses in liberal arts colleges: Further results. Journal of Economic Education, 29, 291–300. Greene, W. H. (2003). Econometric analysis (5th ed.). Upper Saddle River, NJ: Prentice Hall. Harris, K. M., Florey, F., Tabor, J., Bearman, P. S., Jones, J., Udry, J. R. (2003). The National Longitudinal Study of Adolescent Health: Research Design. Retrieved from http://www.cpc.unc. edu/projects/addhealth/design. Herrenkohl, T. I., Mason, W. A., Kosterman, R., Lengua, L. J., Hawkins, J. D., & Abbott, R. D. (2004). Pathways from physical childhood abuse to partner violence in young adulthood. Violence and Victims, 19, 123–136.
J Fam Viol (2008) 23:303–313 Hindelang, M. J. (1981). Variations in sex-race-age-specific incidence rates of offending. American Sociological Review, 46, 461–474. Horowitz, J. L. (2001). The Bootstrap. In J. J. Heckman, & E. Leamer (Eds.) Handbook of Econometrics (Vol. 5) (pp. 3159–3228). Amsterdam: North-Holland. Knapp, L. G., & Seaks, T. G. (1998). A Hausman test for a dummy variable in probit. Applied Economics Letters, 5, 321–323. Kratzer, L., & Hodgins, S. (1997). Adult outcomes of child conduct problems: A cohort study. Journal of Abnormal Child Psychology, 25, 65–81. Langeland, W., & Dijkstra, S. (1995). Breaking the intergenerational transmission of child abuse: Beyond the mother-child relationship. Child Abuse Review, 4, 4–13. Langhinrichsen-Rohling, J., Neidig, P., & Thorn, G. (1995). Violent marriages: Gender differences in levels of current violence and past abuse. Journal of Family Violence, 10, 159–176. Leeb, R. T., Mueller, M., Swahn, M. H., Feldman, M., & Melanson, C. (2004). Sugar & spice, snips & snails: Gender effects on risk factors for delinquency & violence in youth. Presented at 2004 Victimization of Children & Youth: An International Research Conference, Portsmouth, New Hampshire. Levitt, S. D. (1998). Juvenile crime and punishment. Journal of Political Economy, 106, 1156–1185. Magdol, L., Moffitt, T. E., Caspi, A., & Silva, P. A. (1998). Developmental antecedents of partner abuse: A prospective-longitudinal study. Journal of Abnormal Psychology, 107, 375–389. Max, W., Rice, D. P., Finkelstein, E., Bardwell, R. A., & Leadbetter, S. (2004). The economic toll of intimate partner violence against women in the United States. Violence and Victims, 19, 259–272. Mihalic, S. W., & Elliott, D. (1997). A social learning theory model of marital violence. Journal of Family Violence, 12, 21–47. Mocan, M., & Rees, D. (2005). Economic conditions, deterrence and juvenile crime: Evidence from Micro data. American Law and Economics Review, 7, 319–349. Moffitt, T. E., Caspi, A., Rutter, M., & Silva, P. A. (2001). Sex differences in physical violence and sex similarities in partner abuse. In T. E. Moffitt, A. Caspi, M. Rutter, & P. A. Silva (Eds.) Sex differences in antisocial behavior: Conduct order, delinquency, and violence in the Dunedin Longitudinal Study. New York: Cambridge University Press. Morse, B. J. (1995). Beyond the conflict tactics scale: Assessing gender differences in partner violence. Violence and Victims, 10, 251–272. Muller, R. T., Hunter, J. E., & Stollak, G. (1995). The intergenerational transmission of corporal punishment: A comparison of social learning and temperament models. Child Abuse and Neglect, 19, 1323–1335. O’Leary, K. D. (1988). Physical aggression between spouses: A social learning theory perspective. In V. B. Van Hasselt, R. L.
313 Morrison, A. S. Bellack, & M. Hersen (Eds.) Handbook of family violence. New York: Plenum. Resnick, M. D., Ireland, M., & Borowsky, I. (2004). Youth violence perpetration: what protects? What predicts? Findings from the National Longitudinal Study of Adolescent Health. Journal of Adolescent Health, 35(424), e1−e10. Sampson, R. J., Raudenbush, S., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918–924. Simons, R. L., Wu, C., Johnson, C., & Conger, R. D. (1995). A test of various perspectives on the intergenerational transmission of domestic violence. Criminology, 33, 141–172. 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 the Family, 62, 640–654. Straus, M. A., & Gelles, R. J. (1990). Physical violence in American Families: Risk factors and adaptations to violence in 8,145 families. New Brunswick, NJ: Translation Publishers. Straus, M. A., Gelles, R. J., & Steinmetz, S. K. (1980). Behind closed doors: Violence in the American family. Beverly Hills, CA: Sage. Straus, M. A., & Smith, C. (1990). Family patterns and child abuse. In M. A. Straus, & R. J. Gelles (Eds.) Physical violence in American families. New Brunswick, NJ: Transaction Books. Tjaden, P., & Thoennes, N. (2000). Extent, nature, and consequences of intimate partner violence: findings from the National Violence Against Women Survey. Washington, DC: Department of Justice (US). Retrieved from URL: http://www.ojp.usdoj.gov/nij/ pubs-sum/181867.htm. Udry, J. R. (Ed.) (2003). The National Longitudinal Study of Adolescent Health (Add Health), Waves I and II, 1994–1996; Wave III, 2001–2002. Chapel Hill, NC: Carolina Population Center, University of North Carolina at Chapel Hill. Whitaker, D. J., Morrison, S., Lindquist, C., Hawkins, S. R., O’Neil, J. A., Nesius, A. M., Mathew, A., & Reese, L. (2006). A critical review of interventions for the primary prevention of perpetration of partner violence. Aggression and Violence Behavior, 11, 151–166. Widom, C. S., & Morris, S. (1997). Accuracy of adult recollections of childhood victimization, Part 2: Childhood sexual abuse. Psychological Assessment, 8, 412–421. Widom, C. S., Weiler, B. L., & Cottler, L. B. (1999). Childhood victimization and drug use: A comparison of prospective and retrospective findings. Journal of Consulting and Clinical Psychology, 67, 867–880. Wikstrom, P. H., & Loeber, R. (2000). Do disadvantaged neighborhoods cause well-adjusted children to become adolescent delinquents? A study of male juvenile serious offending, individual risk and protective factors, and neighborhood context. Criminology, 38, 1109–1142.