The relative importance of wife abuse as a risk factor for violence against children

Pergamon

Child Abuse & Neglect, Vol. 24, No. 11, pp. 1383–1398, 2000 Copyright © 2000 Elsevier Science Ltd. Printed in the USA. All rights reserved 0145-2134/00/$–see front matter

PII S0145-2134(00)00194-0

THE RELATIVE IMPORTANCE OF WIFE ABUSE AS A RISK FACTOR FOR VIOLENCE AGAINST CHILDREN EMIKO A. TAJIMA School of Social Work, University of Washington, Seattle, WA, USA

ABSTRACT Objective: To investigate the relative importance of wife abuse as a risk factor for physical child abuse, physical punishment, and verbal child abuse. The study explored the importance of wife abuse relative to blocks of parent, child, and family characteristics and also relative to specific risk factors. Method: This study re-analyzed a sub-sample (N ⫽ 2,733) of data from the 1985 National Family Violence Survey. Hierarchical logistic regressions were conducted, using five different criterion variables measuring physical child abuse, physical punishment, and verbal abuse separately and in combination. Results: Blocks of parent, child, and family characteristics were more important predictors of violence towards children than was wife abuse, though the presence of wife abuse in the home was a consistently significant specific risk factor for all forms of violence against children. Of specific risk factors, a respondent’s history of having been hit as an adolescent was a larger risk factor for physical child abuse than was wife abuse. Wife abuse was an important predictor of physical punishment. Non-violent marital discord was a greater factor in predicting likelihood of verbal child abuse than was wife abuse. Conclusions: Though this study confirms the association between wife abuse and violence towards children, it cautions us not to overlook the contribution of other factors in our attempts to understand the increased risk attributed to wife abuse. © 2000 Elsevier Science Ltd. Key Words—Child abuse, Physical punishment, Verbal abuse, Wife abuse, Risk factors.

INTRODUCTION TO BETTER UNDERSTAND the problem of violence against children, it is important to study the contexts in which it occurs. In recent years, attention has focused on child abuse in the context of wife abuse. Studies have revealed that children are at increased risk of physical abuse and/or punishment when wife abuse occurs in the home (Bowker, Arvitell, & McFerron, 1988; Ross, 1996; Schechter & Edleson, 1994; Stark & Flitcraft, 1988; Straus & Smith, 1990). The co-existence of wife abuse and child abuse is amply documented in the literature and wife abuse is understood to be a clear risk factor for child abuse. What is not found in the literature, however, is much analysis of wife abuse relative to other risk factors for child abuse. In other words, how great a predictor is wife abuse compared to other significant risk factors? The purpose of the present study was therefore to examine the relative importance of wife abuse in understanding violence against children. Because different forms of violence can have distinct etiologies and different correlates,

This paper uses data from the “National Family Violence Resurvey” conducted by Richard J. Gelles and Murray A. Straus, with funding from the National Institute of Mental Health, grant MH40027. Received for publication November 16, 1998; final revision received March 21, 2000; accepted March 30, 2000. Requests for reprints should be sent to Emiko A. Tajima, Ph.D., Assistant Professor, University of Washington School of Social Work, 4101 15th Avenue NE, Seattle, WA 98105-6299. 1383

1384

E. A. Tajima

three separate types of violence against children were explored: physical abuse; physical punishment; and verbal abuse. The following research questions were investigated: What is the relative importance of parent characteristics, child factors, family characteristics, and wife abuse for understanding violence against children? How important is wife abuse relative to other risk factors? What are the greatest specific risk factors for different forms of violence against children? Knowledge derived from this research may lead to greater understanding of physical child abuse, corporal punishment, and verbal abuse of children. A Theoretical Model of Child Abuse Theoretical models of child abuse can be analyzed in terms of their levels of analysis, structure, assumptions about etiology, and their complexity (Azar, 1991). This study used an ecological model of child abuse and considered multiple levels of analysis. As put forth by Bronfenbrenner (1979), the ecological perspective looks at the developing individual, the environment, and their evolving interaction. Bronfenbrenner identified four levels of analysis: microsystems (the immediate setting), mesosystems (relations between settings), exosystems (broader social system settings), and macrosystems (overarching patterns of ideology and/or institutional organization). Belsky (1980) applied this perspective to the problem of child maltreatment, incorporating another level, termed “ontogenic development” to account for what the parent brings to the interaction. The present study analyzed variables related to the following levels of analysis: (1) ontogenic (parent); (2) microsystem, (child characteristics, family factors); and (3) mesosystem (wife abuse) (see Table 1 for list of variables in the analysis). With regard to structure, the model used in this study considered multiple forms of violence against children and employed multiple predictors, viewing abuse as an interaction between the parent and child in the context of the family setting. As Belsky (1980) observed, “Since the parent-child system (the crucible of child maltreatment) is nested within the spousal relationship, what happens between husbands and wives—from an ecological point of view— has implications for what happens between parents and their children” (p. 326). The etiology of child abuse is assumed to be related to multiple, interacting factors, including characteristics of the parent, the child, the family, the larger social context, as well as broader cultural norms and beliefs. As Azar (1991) noted, early theories of child maltreatment were single cause theories, followed by more complex theories that considered multiple factors; however, even these later works simply generated “lists of the components of single factor theories with little attempt to specify contingent relationships between components or prioritize their contribution to causality” (p. 35). By utilizing a more complex model of child abuse, the present study sought to address this weakness. Though causality cannot be shown, this research adds to the literature by identifying risk factors and exploring the relative importance (i.e., size and significance of odds ratios) of each risk factor. Risk Factors for Violence Against Children Wife abuse. Much research on the link between wife abuse and child abuse focuses solely on rates of violence against children in homes with wife abuse, usually looking only at physical child abuse. Clinical studies indicate that between 40 and 70% of men who batter their partners also abuse their children (40%, O’Keefe, 1995; 63%, Giles-Sims, 1985; 70%, Bowker et al., 1988), while rates of abuse committed by battered women reportedly range from 44% to 56% (O’Keefe, 1995; Stacey & Shupe, 1983; Giles-Sims, 1985). A methodological shortcoming of many studies that rely on non-representative samples is their failure to use comparison groups. Consequently, though the above studies showed that child abuse and wife abuse frequently co-occurred, they failed to demonstrate the increased risk posed by wife abuse. Certain clinical sample studies did use comparison groups (e.g., Holden & Ritchie, 1991; Stark

Wife abuse and violence against children

1385

Table 1. Variables in the Analysis, Descriptive Statistics, Definitions (N ⴝ 2733) Criterion Variables Variable Name (Min/Max)

Percentages/ Means

Physical Child Abuse (0, 1)

4.1%

Physical Punishment (0, 1)

61.6%

Physical Abuse or Punishment (0, 1) Verbal Child Abuse (0, 1)

61.9% 55.2%

Physical Abuse, Punishment or Verbal Abuse (0, 1)

74.9%

Wife Abuse (0, 1)

15.7%

Definition Any one or more of the following in the past year: threw something at the child; kicked, bit or hit with a fist; beat up; burned or scalded; threatened with a knife or gun; or used a knife or gun Any one or more of the following in the past year: pushed, grabbed, or shoved the child; slapped or spanked; or hit or tried to hit with an object Any physical child abuse or physical punishment in the past year Any one or more of the following in the past year: insulted or swore at the child; did or said something to spite the child; or threatened to hit or throw something at the child Any physical abuse, physical punishment or verbal child abuse in the past year Predictor Variables

Parent Factors Age (18–81 yrs) Own Father Hit Mother (0, 1) Education (Husband) (0, 1) Physical Health (0, 1) Alcohol Abuse (0, 1) Drug Abuse (Husband) (0, 1) Hit by Parent as Teen (0, 1) Stress (0–12) Race (4 variables) (0, 1)

Child Factors Gender (0, 1) Child Problems (0, 1) Age (3 variables) (0, 1) Family Factors Number of Children (1–8) Time in Community (0–67 yrs) Non-violent Discord (0, 1)

Income ($0–10,000 to 50,000 or over) Family Type (0, 1)

Husband did any of the following to wife in the past year: threatened to hit or throw something at her; threw something at her; pushed, grabbed or shoved; slapped; kicked, bit or hit with a fist; hit or tried to hit with an object; beat up; choked; threatened with a knife or gun; used a knife or fired a gun; or forced sex or attempted forced sex

35.8 yrs 12.7% 86.1% 72.7% 41.0% 7.3% 54.0% 3.15 Black (6.6%) White (81.3%) Latino (5.6%) Other (6.6%)

Respondent age Respondent’s own father hit mother (ever) Husband is a high school graduate Respondent in very good or excellent physical health Respondent reports moderate to high alcohol use Husband used drugs in past year Respondent ever hit by either parent as adolescent Perceived Stress Index

Female (48.8%) 23.9% Infant (6.8%) Middle (70.2%) Older (23.0%)

Child female or male Any delinquency, aggression or other problems Infant under 1 year old Child 1 through 13 yrs Child 14 yrs or over

1.92 15.02 yrs

Number of children in household Number of years respondent lived in the community

80.9%

$25,700

Respondent did any of the following in the past year: insulted or swore at his/her partner; stomped out of room, house, or yard; did or said something to spite partner; threw, smashed, hit or kicked something Household income (in 1985 dollars)

84.1%

Biological family (i.e., no step-children in household)

& Flitcraft, 1988). Focusing on child rearing behavior and physical punishment, Holden and Ritchie indicated that over 50% of the battered women reported that their partners “spanked the children at least once a week,” compared to only 24% for the non-battered comparison group. Stark

1386

E. A. Tajima

and Flitcraft studied the medical records of 116 mothers of children who had been seen at the hospital and whose cases had been referred to child protective services for suspected child abuse, identifying 45% of the mothers as battered women. Compared to women coming in for surgery or prenatal care, the risk of battering was twice as high in the records of the women whose children were allegedly abused. Straus and Smith (1990) explored child abuse among nationally representative samples, using the 1975 and 1985 National Family Violence Surveys (NFVS). Using discriminant analysis they identified risk factors for physical child abuse, finding that wife abuse was a significant factor. Even “minor” violence (defined by Straus and Smith as slapping, pushing, shoving, or throwing things at one’s partner) was associated with a 150% increase in the rate of child abuse (22.3 versus 8.0). In her re-analysis of the 1985 NFVS data, Ross (1996) found that 22.8% of the male batterers also abused their child at least once, compared to 8.5% for non-batterers. Straus (1994) analyzed the 1985 NFVS data with a focus on corporal punishment of children, finding that mothers who had been hit by their partners had a 71% chance of hitting their adolescent child, whereas women who had not been hit had a 48% probability of doing so. Parent characteristics. Age of the parent may be a risk factor for violence against children (Straus, 1994). Parental education has been also been associated with risk of physical child abuse; surprisingly, more education was related to increased risk in two different studies (Margolin & Larson, 1988; Bowker et al., 1988). Rates of child abuse have been compared between different racial groups, with varied results. Ross (1996) found increased risk of physical child abuse by African American fathers compared to Latinos, and greater risk of child abuse by Latina mothers compared to Whites. Straus and Smith (1990) found African American children to be at greater risk of physical abuse relative to White children. Straus (1994) found White parents to be more likely to use corporal punishment than minorities. Having been abused as a child and having witnessed violence in one’s home of origin are regarded as risk factors for perpetrating child abuse (Merrill, Hervig, & Milner, 1996; Ross, 1996; Straus & Smith, 1990) and corporal punishment (Straus, 1994). Parental physical and mental health has also been connected to risk of child abuse (Cicchetti & Rizley, 1981; Margolin & Larson, 1988). Holden and Ritchie (1991) identified stress as having a negative effect on parenting among battered women. Merrill and colleagues (1996) found alcohol to be a significant risk factor for child abuse among both men and women. Child characteristics. Gender of the child has been associated with risk of physical abuse. Ross (1996) found that boys were at increased risk of abuse in homes in which spousal assault had occurred. Straus (1994) noted that rates of violence also varied with the age of the child. In addition, the type of violence and its etiology may also vary by age of the child (Wauchope & Straus, 1990). Children with certain problems are at increased risk of abuse; children with physical disabilities are particularly vulnerable (Goldson, 1998; Oates, 1996), as are children with behavioral problems such as child aggression (O’Keefe, 1995). It is difficult to interpret findings regarding the relationship between child behavior problems and risk of abuse because it is unclear which came first; that is, the causal direction of the relationship is ambiguous. It should be noted that, though examining child characteristics is important, what the child brings to the setting can not singly precipitate violence towards him/her. As Belsky noted (1980), “. . . characteristics of the child make sense as elicitors of maltreatment only when considered vis a` vis the caregiver’s attributes” (p. 324). Family characteristics. Risk of child abuse may be related to the number of years a family has lived in a neighborhood and the social support networks available (Cicchetti & Rizley, 1981; O’Keefe, 1995; Straus & Smith, 1990). Family income was identified as having an inverse relationship with

Wife abuse and violence against children

1387

risk of child abuse (Bowker et al., 1988; Cicchetti & Rizley, 1981; Straus & Smith, 1990). Family type may also be relevant to understanding child abuse. O’Keefe (1995) identified three family types: single-parent, biological, and step-families. The present study looked at two-parent homes, and compared biological versus step-families. Non-violent marital discord in the home is an important but understudied potential risk factor; most research on the link between wife abuse and child abuse considers only physical violence between the marital couple. Straus and Smith (1990) found marital conflict increased the risk of physical child abuse. However, Hershorn and Rosenbaum (1985) compared abused mothers to those in non-violent but maritally discordant relationships and found little difference in punitive child-rearing methods between the groups. The present study analyzed data from the 1985 National Family Violence Survey (NFVS) to address previously unexamined questions. Prior analyses of the NFVS established wife abuse as a key risk factor for violence against children; the present research built upon these findings by examining wife abuse relative to other risk factors. By examining physical child abuse, physical punishment, and verbal child abuse, this research also broadened the scope of previous analyses. This study aimed to: (1) identify risk factors for physical child abuse, physical punishment, and for verbal child abuse; (2) compare the relative contribution of parent characteristics, child factors, family characteristics, and wife abuse to predicting violence against children; and (3) identify the most important specific risk factors for each type of violence.

METHOD The Data and Sample This study analyzed data from the 1985 NFVS, which was a nationally representative sample of 6,002 households (representative of adult, heterosexual persons with telephone service). Both men and women were interviewed, with one respondent (selected randomly) interviewed per household. Telephone interviews were conducted by trained, female interviewers, with a response rate of 84%; telephone numbers were chosen by random digit dialing, stratified by region and size of place (Straus, 1990). The present study analyzed data on those respondents who were married (or living together) with at least one child under 18. This sub-sample consisted of 2,733 cases; of these, 456 reported wife abuse as defined below. The NFVS used the Conflict Tactics Scale (CTS) to measure spousal assault and child abuse. The CTS is a widely used instrument designed to capture honest responses to sensitive questions about domestic violence and child abuse. To measure violence towards children, respondents were asked how many times in the past year they engaged in each of 19 different behaviors with their child (selected randomly if there were multiple children). To measure spousal assault, respondents indicated whether they had done any of the 19 behaviors to their partner (in the past year), then indicated whether their partner had done any to them. Straus (1990), one of the creators of the CTS, reported the Alpha coefficient of reliability for the CTS as .83 for Husband-to-Wife violence, .62 for Parent-to-Child violence and .77 for Parent-to-Child verbal aggression. Elsewhere, the alpha coefficients were said to range between .41 and .96 (Straus, 1994). The concurrent and construct validity of the CTS is discussed in detail in Straus and Gelles (1990). Though widely used, the CTS has important limitations. Focused on counting the “number of violent acts,” the CTS has been criticized for failing to consider the immediate and broader context of the incident and for disregarding the process and consequence of the violence, especially in relation to spousal assault (Dobash & Dobash, 1992; Schechter, 1982; Straus & Hamby, 1997). Furthermore, the 1-year time period used in the CTS may be too long for accurate recall; however, several items are low frequency behaviors and call for a longer referent period. The 1-year time

1388

E. A. Tajima

span was ultimately selected as an imperfect but necessary compromise (Straus & Gelles, 1990). As the CTS only included measures of violence in the context of interpersonal conflict, it did not measure child sexual abuse or child neglect. Variables in the Analysis (See Table 1) Criterion variables. Physical child abuse was defined by the following CTS items: throwing something at the child; kicking, biting or hitting with a fist; beating up; burning or scalding; threatening with a knife or gun; and using a knife or gun. Physical punishment included pushing, grabbing, or shoving; slapping or spanking; and hitting or trying to hit with an object. Though these physical punishments would be considered abuse by many, they nonetheless have lower potential for injury and/or are deemed acceptable, if not normative in the United States (Straus, 1994). In distinguishing between abuse and physical punishment the present study followed Wauchope and Straus’ (1990) example of considering the “level of severity permitted by law and custom,” as well as the injury potential of the behavior (p. 137). A criterion variable combining “any physical abuse or punishment” was also analyzed. Verbal child abuse was measured by the items insulting or swearing at; doing or saying something to spite the child; and threatening to hit or throw something at the child. A final model explored the combination criterion variable: “any physical abuse, punishment, or verbal abuse.” Predictor Variables Wife abuse. Items used to measure wife abuse included threatening to hit or throw something at; throwing something at; pushing, grabbing or shoving; slapping; kicking, biting or hitting with a fist; hitting or trying to hit with an object; beating up; choking; threatening with a knife or gun; using a knife or firing a gun; and forced sex or attempted forced sex. All items were from the CTS except for the last item, which was a survey question asked of female respondents. Parent characteristics. Parent characteristics included age; whether respondent’s father hit mother; education; physical health; alcohol use; drug use; whether respondent was hit by a parent as a teen; stress; and race (see Table 1). All of these variables have been identified in the literature as risk factors for physical child abuse or punishment. Child characteristics. Child variables included gender; presence of child problems (i.e., aggression, delinquency or other problems); and age of the child. Because of its high correlation with age of respondent, the variable age of the child was recoded. After bivariate analyses, the age of the child was recoded into three dummy variables, “infant under 1 year,” “age 1–13,” and “age 14 or over.” Only 6.8% of the children were infants; 23% were age 14 or over. Family characteristics. Family factors included number of children in household; years in the community; presence of non-violent marital discord; family income; and family type. Data Analysis Analyses involved several stages. First, bivariate relationships among all predictor variables were computed to explore associations; in several instances variables were combined, recoded or removed to prevent multicollinearity. The NFVS included a cross-sectional sample of the US population, oversamples of Black and Latino respondents, and an oversample for individual states. To make the original sample representative of the US, the data included a weight (“weight3” in the codebook) for use in analyses combining the cross-sectional sample and the three oversamples. Because the present study used

Wife abuse and violence against children

1389

only a sub-sample (two-parent households with children under 18 in the home), a new weight was created. The following formula was used: sub-sample weight ⫽ weight3/(mean weight for sub-sample), Where “mean weight for subsample” ⫽ .990198.

Hierarchical logistic regression was used to identify the contribution of blocks of variables and the contribution of specific predictor variables in explaining different forms of violence against children. To examine the relative importance of wife abuse, 21 predictor variables were entered in four separate blocks. Predictor variables included 11 parent characteristics, four child factors, five family factors, and wife abuse. This method yielded coefficients (adjusted odds ratios) for each predictor variable, controlling for all other variables in the model; it also produced a pseudo R2 statistic for each block of variables. The pseudo R2 represents an estimate of the proportion of the variance in the criterion variable that is explained by the block. In this way, the contribution of each block can be identified and compared to that of other blocks of risk factors. Hierarchical regression is a method in which blocks of variables can be entered in a theoreticallybased ordering. The five models included parent characteristics as the first block, child characteristics second, family characteristics third, and wife abuse as the final block. Many of the variables were correlated to some extent and shared a certain percentage of the explained variance in the criterion variables. (The highest zero-order correlation among predictor variables was .47, however all other zero-order correlations were below .32.) When a block was entered, it was allowed to explain as much of the variance in the criterion measure as it could, then remaining blocks were in turn allowed to explain as much of the remaining variance as they could. Thus, the sequencing of the blocks had some influence on how much variance they appeared to explain. In particular, the block that was entered first may seem to explain more variance simply because it was able to account for all of the shared explained variance. Because these models were being used to examine the importance of wife abuse as a risk factor relative to other sets of factors, this study selected the more conservative route of entering the wife abuse block in the last position. To examine the impact of the sequencing, regressions were also run with blocks entered in different order. When wife abuse was entered first, it explained about twice the amount of variance as it explained when entered last. It should be noted that large samples such as the NFVS may yield statistically significant findings even for very small relationships. Therefore this study reported the adjusted odds ratio (Exp[B]) to indicate the magnitude of the relationship in addition to its significance level. RESULTS Physical Child Abuse Physical child abuse occurred in 4.1% (n ⫽ 112) of the cases in this sample. Looking at the relative contribution of different blocks of risk factors, Table 2 shows that, controlling for all other predictors, wife abuse explained less than 1% (.5%) of the variance in this criterion variable. Parent characteristics, as a block, explained more of the variation in physical abuse than any other block (6.4%). Child factors accounted for 2.1% of the variance and family characteristics explained 2.0%. The total explained variance for the model was only 11.0%. This indicates that important predictors of physical abuse were not included in the model. However, it can also be explained by the fact that this was a large, heterogeneous sample. Risk factors may be important predictors among certain subgroups, but such relationships could be obscured in analyses using the full sample. Analysis of the coefficients for the specific predictor variables in the final model revealed five significant relationships. Respondents who had been hit by a parent as an adolescent were more than twice as likely to abuse their child, a finding consistent with Ross’ (1996) study. Surprisingly,

1390

E. A. Tajima Table 2. Risk Factors for Physical Child Abusea (Hierarchical Logistic Regression) (N ⴝ 2388) Model Pseudo R2 ⴝ .110 (Nagelkerke) Block

Variable

Parent Characteristics

Respondent’s age Resp.’s father hit mother Husband high school grad Very good/excell. health Mod.–high alcohol use Husband uses drugs Resp. hit as adolescent Stress index Black respondentb Latino respondentb Other race respondentb Gender (female ⫽ 1, male ⫽ 0) Any child problems Infant ⬍1 yrc Child 14 yrs or overc Number of children Years in community Marital discord Income Biological family (non-step) Any wife abuse

Child Characteristics

Family Characteristics

Wife Abuse a b c

Block Pseudo R2

Adjusted Odds Ratio Exp(B)

Sig

.064

1.01 0.93 0.71 0.65 0.83 0.79 2.27 1.09 1.29 1.73 1.59 0.58 1.56 0.48 1.21 1.21 1.00 2.23 1.06 1.96 1.69

.386 .794 .264 .067 .403 .570 .001 .047 .544 .180 .235 .014 .050 .264 .497 .063 .601 .060 .502 .049 .040

.021

.020

.005

No child abuse n ⫽ 2617 (95.9%); Yes child abuse n ⫽ 112 (4.1%). Excluded category is “White respondent.” Excluded category is “Child age 1 through 13 years.”

biological families (i.e., not a step-family) were twice as likely to report child abuse. This result is contrary to what O’Keefe (1995) had theorized; it is possible that step-families were simply less likely to disclose child abuse, possibly because of ongoing custody situations. The presence of wife abuse increased the likelihood of child abuse by about 70%. Thus, even though wife abuse explained little of the total variance relative to blocks of parent, child, and family characteristics, when comparing homes with and without wife abuse, it was clear that the presence of wife abuse significantly increased the odds of child abuse. Higher scores on the stress index were associated with increased odds of abuse. If the child was female, she was 42% less likely to be abused. None of the other variables were significantly related to the criterion variable. No differences were found between Whites and non-Whites. Physical Punishment A majority of respondents (61.6%, n ⫽ 1,674) in this sample reported some physical punishment. As shown in Table 3, the total explained variance for this model was 30.3%, with child characteristics accounting for the most variation (19.9%). Parent characteristics explained 8.4%, and family characteristics explained 1.1%. As a block, when controlling for all other predictors, wife abuse accounted for only .9% of the variation in physical punishment. Of the specific predictor variables in the model, however, wife abuse was the most important positive risk factor, increasing the risk of physical punishment by over two and a half times. The presence of child problems (i.e., aggression, delinquency or other problems), doubled the risk of physical punishment. Non-violent marital discord in the home increased the odds by 86%. If the parent had been hit as a teen, or if the family was biological, the likelihood of physical punishment increased by about 50%. Several variables were related to reductions in the risk of physical punishment. Female children

Wife abuse and violence against children

1391

Table 3. Risk Factors for Physical Punishmenta (Hierarchical Logistic Regression) (N ⴝ 2382) Model Pseudo R2 ⴝ .303 Block

Variable

Parent Characteristics

Respondent’s age Resp.’s father hit mother Husband high school grad Very good/excell. health Mod.–high alcohol use Husband uses drugs Resp. hit as adolescent Stress index Black respondentb Latino respondentb Other race respondentb Gender (female ⫽ 1, male ⫽ 0) Any child problems Infant ⬍1 yrc Child 14 yrs or overc Number of children Years in community Marital discord Income Biological family (non-step) Any wife abuse

Child Characteristics

Family Characteristics

Wife Abuse a b c

Block Pseudo R2

Adjusted Odds Ratio Exp(B)

Sig

.084

0.95 1.06 1.17 0.82 0.95 0.70 1.49 1.04 0.76 0.41 0.83 0.71 2.21 0.04 0.13 1.05 1.00 1.86 0.98 1.51 2.60

.000 .732 .354 .127 .660 .101 .000 .090 .213 .000 .386 .001 .000 .000 .000 .367 .683 .000 .545 .003 .000

.199

.011

.009

No punishment n ⫽ 1045 (38.4%); Yes punishment n ⫽ 1674 (61.6%). Excluded category is “White respondent.” Excluded category is “Child age 1 through 13 years.”

were 29% less likely to be physically punished. Consistent with Escovar and Escovar’s (1985) findings, Latinos were 59% less likely to report using physical punishment than White respondents. Children aged 14 years or over and infants under 1 year of age were 87–96% less likely to experience physical punishment than children between those ages. As reported by Straus (1994), older respondents were at decreased risk. Also consistent with Straus’ (1994) study, income was not significantly related to physical punishment. Physical Child Abuse or Physical Punishment The previous models looked at physical abuse and punishment separately; however, most parents who physically abused their children also physically punished them. Though it is useful to study the risk factors for these different behaviors separately, it is also limiting to analyze risk factors for the less severe types of violence without considering whether the parent also engaged in abuse. More than half of all respondents reported either physical abuse or punishment (61.9%, n ⫽ 1,682). The final model explained 29.9% of the variance in this criterion variable (Table 4). The proportion of total variance explained by the four blocks is very similar to that found in the physical punishment model. Risk factors too, were nearly identical to those for physical punishment. This is not surprising: physical punishment occurred far more frequently than abuse, and nearly all respondents who abused their child also physically punished them, therefore the two models include mostly the same cases. Wife abuse increased the likelihood of physical abuse or punishment by more than 150%. Child behavioral problems more than doubled the risk of abuse or punishment. Marital discord, being hit as a teen, and being a biological family all increased the odds by about 50 – 80%. Female children, Latinos (compared to Whites), and older parents had decreased risk of abuse or punishment. Infants and older children were considerably less likely to be abused or punished than children between those ages (about 85–95% less likely).

1392

E. A. Tajima

Table 4. Risk Factors for Physical Child Abuse or Physical Punishmenta (Hierarchical Logistic Regression) (N ⴝ 2382) Model Pseudo R2 ⴝ .299 Block

Variable

Parent Characteristics

Respondent’s age Resp.’s father hit mother Husband high school grad Very good/excell. health Mod.–high alcohol use Husband uses drugs Resp. hit as adolescent Stress index Black respondentb Latino respondentb Other race respondentb Gender (female ⫽ 1, male ⫽ 0) Any child problems Infant ⬍1 yrc Child 14 yrs or overc Number of children Years in community Marital discord Income Biological family (non-step) Any wife abuse

Child Characteristics

Family Characteristics

Wife Abuse a b c

Block Pseudo R2

Adjusted Odds Ratio Exp(B)

Sig

.084

0.95 1.04 1.19 0.81 0.93 0.68 1.52 1.05 0.77 0.43 0.82 0.72 2.18 0.04 0.13 1.04 1.00 1.84 0.97 1.53 2.56

.000 .826 .292 .103 .517 .085 .000 .054 .224 .000 .342 .002 .000 .000 .000 .489 .713 .000 .457 .003 .000

.195

.011

.009

No child abuse or punishment n ⫽ 1036 (38.1%); Yes child abuse or punishment n ⫽ 1682 (61.9%). Excluded category is “White respondent.” Excluded category is “Child age 1 through 13 years.”

Verbal Abuse Verbal abuse of the child was reported in 55.2% (n ⫽ 1,494) of the cases. As shown in Table 5, the predictor variables in this model explained 18.3% of the variance in verbal abuse. Looking at the separate blocks, parent characteristics and child characteristics each explained 6.8% of the variation in verbal abuse, with family characteristics accounting for 4.5%. In this model, controlling for all other predictors, wife abuse accounted for a mere .2% of the variance in verbal child abuse. Considering the relative importance of different risk factors for verbal abuse, it is evident from these data that wife abuse is considerably less important than blocks of parent, child, or family characteristics. Of the risk factors positively related to verbal abuse, the most important specific predictor variable was non-violent marital discord, which increased the risk more than three-fold. Children with behavioral problems were 1.86 times as likely to be verbally abused. The presence of wife abuse in the home increased the risk of verbal abuse by 42%. Also significant was alcohol abuse and having been hit as a teen; these increased the odds of using verbal abuse by 34% and 38% respectively. Greater risk was also associated with greater stress, and higher number of children in the household. As was the case with both physical abuse and punishment, girls were less likely to be verbally abused, about 22% less likely. Infants under 1 year of age were far less likely to be verbally abused than children ages 1 through 13—about 88% less likely. Children 14 and over were no more or less likely to be verbally abused than children between 1 and 13 years of age. Physical Child Abuse, Physical Punishment, or Verbal Abuse The final model explored risk factors for any type of violence towards children (Table 6). Nearly three fourths (74.9%, n ⫽ 2,027) of the respondents reported at least one of these behaviors in the

Wife abuse and violence against children

1393

Table 5. Risk Factors for Verbal Child Abusea (Hierarchical Logistic Regression) (N ⴝ 2376) Model Pseudo R2 ⴝ .183 Block

Variable

Parent Characteristics

Respondent’s age Resp.’s father hit mother Husband high school grad Very good/excell. health Mod.–high alcohol use Husband uses drugs Resp. hit as adolescent Stress index Black respondentb Latino respondentb Other race respondentb Gender (female ⫽ 1, male ⫽ 0) Any child problems Infant ⬍1 yrc Child 14 yrs or overc Number of children Years in community Marital discord Income Biological family (non-step) Any wife abuse

Child Characteristics

Family Characteristics

Wife Abuse a b c

Block Pseudo R2

Adjusted Odds Ratio Exp(B)

Sig

.068

0.99 1.13 1.11 1.02 1.34 1.26 1.38 1.13 1.15 0.81 0.86 0.78 1.86 0.12 0.89 1.22 1.01 3.26 1.07 1.11 1.42

.361 .415 .475 .879 .002 .242 .001 .000 .458 .339 .414 .007 .000 .000 .356 .000 .084 .000 .054 .393 .013

.068

.045

.002

No verbal abuse n ⫽ 1211 (44.8%); Yes verbal abuse n ⫽ 1494 (55.2%). Excluded category is “White respondent.” Excluded category is “Child age 1 through 13 years.”

past year. Together, predictor variables explained 24.4% of the variance in violence towards children. As a block, parent characteristics explained 6.2% of the variation. Child factors played the largest role, explaining 15.0%, while family characteristics explained 2.6%. Wife abuse accounted for .6% of the variance in this criterion variable, controlling for all other predictor variables. Of the specific variables that were positively related, marital discord was the greatest risk factor for some form of violence towards the child, increasing the odds nearly three times. Wife abuse had the next largest coefficient, more than doubling the likelihood of violence towards children. Having a child with behavioral problems also doubled the risk of violence. Respondents who had been hit by their parents as teens were 53% more likely to report some violence. Respondents in non-step families, and those who reported greater stress were at increased risk of violence towards their child. It is difficult to interpret the results from this final model because this combined criterion variable encompassed such varied behaviors. Nonetheless, analyzing risk factors for any type of violence towards children may inform those interested in understanding violence in all its different forms. As with all previous models, female children were at less risk of experiencing violence. Infants under 1 year of age and teens 14 and over were also at decreased risk compared to those aged 1 through 13. Latinos were about half as likely to be violent towards their child, relative to Whites. Older parents were also at decreased risk of violence, controlling for age of the child. In this model only, drug use by the husband was a significant factor; surprisingly though, this decreased the risk of violence towards children by 47%. LIMITATIONS OF THE DATA AND STUDY The data from the NFVS offered the opportunity to study connections between wife abuse and child abuse among a large, representative sample of families in the US. Nonetheless, these data

1394

E. A. Tajima Table 6. Risk Factors for Physical Child Abuse, Physical Punishment of Verbal Abusea (Hierarchical Logistic Regression) (N ⴝ 2379) Model Pseudo R2 ⴝ .244 Block

Variable

Parent Characteristics

Respondent’s age Resp.’s father hit mother Husband high school grad Very good/excell. health Mod.–high alcohol use Husband uses drugs Resp. hit as adolescent Stress index Black respondentb Latino respondentb Other race respondentb Gender (female ⫽ 1, male ⫽ 0) Any child problems Infant ⬍1 yrc Child 14 yrs or overc Number of children Years in community Marital discord Income Biological family (non-step) Any wife abuse

Child Characteristics

Family Characteristics

Wife Abuse a b c

Block Pseudo R2

Adjusted Odds Ratio Exp(B)

Sig

.062

0.95 1.00 1.13 0.97 1.21 0.53 1.53 1.10 0.95 0.45 1.01 0.76 2.22 0.03 0.28 1.11 1.01 2.94 1.02 1.25 2.32

.000 .985 .505 .811 .112 .013 .000 .000 .838 .001 .969 .014 .000 .000 .000 .074 .112 .000 .690 .140 .000

.150

.026

.006

No child abuse, punishment or verbal n ⫽ 679 (25.1%); Yes child abuse, pun., or verbal n ⫽ 2027 (74.9%). Excluded category is “White respondent.” Excluded category is “Child age 1 through 13 years.”

have several limitations. As is often the case with secondary analysis, ideal measures of important variables may not be available. In the present study, this limitation was most acute in relation to “exosystem” and “macrosystem” variables. The ecological model of child abuse calls for measures of elements such as neighborhood support systems, housing conditions, or social norms, but these were not available. In addition to the limitations of the CTS as a measure of interpersonal violence, there are problems inherent in data that depend upon adult recall. Widom and Shepard (1996) compared documented histories of childhood physical abuse and retrospective self-reports, finding that a substantial number of individuals failed to report childhood abuse. Henry, Moffitt, Caspi, Langley, and Silva’s (1994) research led them to caution against relying upon retrospective reports, especially for psychosocial variables such as childhood family conflict. The present study included measures of abuse as an adolescent and whether or not there had been wife abuse in the respondent’s home of origin. The NFVS data were derived from a survey of self-reported violence in the home, and may have been affected by social desirability bias (Widom & Shepard, 1996). This bias may result in underreporting of violence in general; even more troublesome, social desirability bias may interact with other factors, potentially yielding measures that represent willingness to disclose abuse rather than actual risk of abuse. Although violence may have been underreported, the relationships between predictors and violence may still be accurately estimated. Sample composition further limited the study. The NFVS oversampled African Americans and Latinos but not other minority groups. Consequently, analysis by race was limited to comparisons between African Americans, Whites, Latinos, and those of “other” race. Though single parents were surveyed, they were not asked questions about domestic violence and therefore could not be included in the present study. This was particularly unfortunate, as many battered women are single

Wife abuse and violence against children

1395

mothers, abused by estranged partners, and many clients of child protective service agencies are also single parents (Schechter & Edleson, 1994). Furthermore, the generalizability of the present study is limited in certain ways. This research used nationally representative data, therefore findings are generalizable to US households, which is valuable—it is important to examine risk factors for violence towards children in the general population. However, patterns of violence in the general population may differ from those found among “clinical” populations. Therefore, when using a nationally representative sample, implications for clinical practice or agency policy should be drawn with caution. The field needs both research that is broadly generalizable and work based on clinical populations. It would be useful to replicate this study in “clinical” settings such as shelters, counseling centers, hospitals, or legal advocacy programs. Because the data were gathered in 1985, information on the incidence and prevalence of child abuse and wife abuse may not be generalizable to the present time. For example, Straus and Gelles (1986) noted changes in rates of child abuse and wife abuse between 1975 and 1985 that may have been related to historical period. Though not an incidence study, the present study focused on patterns of risk factors which may also be affected by historical changes such as increased services for victims of domestic violence, or greater awareness of connections between various forms of violence. Despite these limitations, the present study may serve as an analytic model for studying risk factors for child abuse; the patterns and risk factors identified in the present study might be re-examined as new data become available to compare changes in these relationships over time. SUMMARY AND DISCUSSION This study of risk factors for different forms of violence towards children suggests that sets of parent, child, and family characteristics are more important predictors of violence towards children than is the presence of wife abuse in the home. Compared to other blocks of factors, wife abuse explained relatively little of the variation in the criterion variables. Parent characteristics accounted for the greatest percentage of the explained variance in physical child abuse, though the total explained variance for that model was only 11%. Child characteristics accounted for about two-thirds of the variance in the physical punishment model, which had a total explained variance of about 30%. Both parent and child factors accounted for the same proportion of explained variance for the verbal abuse model. Child characteristics were key predictors in the fifth model, which looked at any violence towards children. Family factors consistently accounted for a greater percentage of the explained variance than wife abuse. These results raise two questions: Why did the models explain so little of the variance, especially for child abuse? Why does wife abuse account for so little variance? Part of the answer to these questions may lie with the sample. Sample heterogeneity may conceal relationships that would be stronger if the sample were analyzed by subgroups. The fact that it is a non-clinical sample is also important. Though our knowledge of predictors of child abuse are often derived from clinical populations, it may be that in the general population, wife abuse plays a less important role than other sets of risk factors. Future replication with a clinical sample would be informative. Part of the answer may lie with the variables in the model. This analysis attempted to include key predictors, however the rates of explained variance indicate that important risk factors were omitted. Future research might include measures of attitudes and cognitions related to violence, norms about hitting children, and measures of parenting education. Including neighborhood level or social structure variables (e.g., unemployment rate, crime rate) might also increase the explained variance. Finally, part of the answer may relate to the etiological complexity of violence towards children. No single set of risk factors explained more than 6.4% of the variance in child physical abuse. The particularly low explained variance for physical abuse may also be

1396

E. A. Tajima

related to its highly skewed distribution (only 4% reported abuse), meaning that there was little overall variance to explain. Though domestic violence has been highlighted as a key risk factor for child abuse by researchers and practitioners, it is only one piece of the puzzle. However, even though it may explain little of the variance in violence towards children relative to sets of parent and child characteristics, wife abuse is a risk factor that can be identified and addressed by policy-makers and service providers. As an individual predictor, wife abuse was a consistently significant risk factor for all forms of violence towards children, increasing the likelihood of physical child abuse, physical punishment, and verbal child abuse. However, it was not always the most important specific predictor. Though these findings confirm the association between wife abuse and violence against children, they also caution us not to overlook the contribution of other elements in our attempts to understand the increased risk attributed to wife abuse. For example, the respondent’s childhood history of abuse was a greater risk factor for physical child abuse than was the presence of wife abuse in the home. Therefore, in looking to the existing parental relationship to better explain violence against children, we must not ignore the intergenerational transmission of violence, as parents who were hit as teens are, in turn, at greater risk of becoming violent towards their own children. This is consistent with Merrill and colleagues (1996) conclusion that “early social learning and the intergenerational transmission of abuse perspectives may be more important in explaining child physical abuse than intimate partner physical violence” (p. 1059). Concordant with Straus’ (1994) findings, respondents who had been hit as adolescents were more likely to physically punish their own child, showing that the intergenerational transmission of violence hypothesis relates also to corporal punishment. Wife abuse was the largest single positive risk factor for physical punishment, greater than having a child with behavioral problems, though all specific child characteristics were significant. In the literature the focus is on wife abuse as a risk factor specifically for physical child abuse. However, this work highlights that wife abuse may have an even stronger connection to physical punishment. Further research is needed to explain this connection. This study also identified non-violent marital discord as a key risk factor for physical punishment. Though most research focuses on physical battering of the mother, these data suggest that we also examine the process by which non-violent marital discord increases the likelihood of violence against children. This research begins exploration of the little-studied area of verbal child abuse. Non-violent marital discord was the most important specific risk factor for verbal child abuse. This might be explained by the fact that neither of these behaviors involve physically confrontational violence; perhaps parents who resolve marital conflict in non-physical ways are also more likely to resolve parent-child conflict in a non-physical manner. Non-violent marital discord was a greater risk factor for verbal child abuse than was wife abuse. Interestingly, parents who had been hit as adolescents were more likely to verbally abuse their child, pointing again to possible intergenerational transmission. Thus, to better understand verbal violence towards children, we must explore parental conflict resolution tactics, in addition to the process by which this form of violence may be transmitted from one generation to another. This study found that drug use by the husband reduced the risk of violence, for the final combination variable, “any physical abuse, punishment, or verbal abuse.” This is curious because parental drug abuse has been discussed in the literature as a factor increasing the risk of abuse (O’Keefe, 1995). Indeed, in bivariate analysis conducted for the present study, drug use by the husband was positively correlated with all forms of violence towards children. However, drug use is also correlated with important risk factors for violence against children, such as wife abuse, alcohol abuse, having been hit by one’s own parents, and having wife abuse in one’s home of origin. It is possible that any positive association between husband drug use and violence towards the child may be explained by these correlates of drug use. Once these variables were taken into account, the relationship between drug use by the husband and any violence towards the child was

Wife abuse and violence against children

1397

negative. Perhaps certain drugs serve as sedatives, decreasing the risk of violence on the part of the user. Another interpretation is that drug using husbands are at home less, and therefore have decreased “time at risk.” An alternative possibility is that wives of drug abusers are, for some reason, less likely to be violent towards the child. Finally, perhaps respondents who admitted to drug use were less likely to disclose violence towards their children. It should also be noted that the data were gathered in 1985, before the impact of widespread crack cocaine usage was felt. Further investigation of this topic would help to clarify the connection between drug use and violence against children. These findings support theoretical models of child abuse that consider multiple levels of analysis. The present study included blocks of variables related to ontogenic (parent), microsystem (family and child factors), and mesosystem (wife abuse) levels. Each of these blocks of factors helped to explain the variation in the criterion variables, with some factors playing a greater role for certain forms of violence. Future research should seek to incorporate more variables related to the exosystem and macrosystem levels, such as neighborhood support systems or social norms, as was done by Magdol and colleagues (1997), who included multiple measures of “social ties” in their study of partner violence. Longitudinal research might help identify risk factors for later child abuse and help conceptualize how they contribute to risk. Toward this end, future research could model the longitudinal work of Henry, Caspi, Moffitt, and Silva (1996), who studied predictors of criminal convictions, finding that certain childhood family factors contributed to a “generalized” risk of conviction, while childhood temperament specifically predicted violent offenses. This study found Latinos less likely to be violent towards their children compared to Whites, but found no differences between African Americans and Whites, or between “other” race and Whites. These findings emphasize the variation that exists between minority groups and caution us not to limit analysis to the categories of White versus “minority.” Future research might also compare risk factors for different types of violence towards children between and within different racial and ethnic subgroups. This research has also shown the value of conducting separate analyses for different types of violence towards children. Though there was some overlap, risk factors were different when examining physical abuse, physical punishment, and verbal child abuse. These results suggest possibly distinct etiologies for the different forms of violence towards children, though at the same time acknowledges elements common to all. Acknowledgements—The author would like to thank Julia H. Littell, Jeffrey L. Edleson, and Mary R. Gillmore for their helpful comments.

REFERENCES Azar, S. (1991). Models of child abuse: A metatheoretical analysis. Criminal Justice and Behavior, 18, 30 – 46. Belsky, J. (1980). Child maltreatment: An ecological integration. American Psychologist, 35, 320 –335. Bowker, L. H., Arvitell, M., & McFerron, J. R. (1988). On the relationship between wife beating and child abuse. In K. Yllo & M. Bograd (Eds.), Feminist perspectives on wife abuse (pp. 158 –174). Newbury Park, CA: Sage. Bronfenbrenner, U. (1979). The ecology of human development. Cambridge, MA: Harvard University Press. Cicchetti, D., & Rizley, R. (1981). Developmental perspectives on the etiology, intergenerational transmission and sequelae of child maltreatment. New Directions for Child Development, 11, 31–56. Dobash, R. E., & Dobash, R. P. (1992). Women, violence and social change. London: Routledge. Escovar, L. A., & Escovar, P. L. (1985). Retrospective perception of parental child-rearing practices in three culturally different college groups. International Journal of Intercultural Relations, 9, 31– 49. Giles-Sims, S. J. (1985). A longitudinal study of battered children of battered wives. Family Relations, 34, 205–210. Goldson, E. (1998). Children with disabilities and child maltreatment. Child Abuse & Neglect, 22, 663– 667. Henry, B., Caspi, A., Moffitt, T. E., & Silva, P. A. (1996). Temperamental and familial predictors of violent and nonviolent criminal convictions: Age 3 to age 18. Developmental Psychology, 32, 614 – 623.

1398

E. A. Tajima

Henry, B., Moffitt, T. E., Caspi, A., Langley, J., & Silva, P. A. (1994). On the “remembrance of things past:” A longitudinal evaluation of the retrospective method. Psychological Assessment, 6, 92–101. Hershorn, M., & Rosenbaum, A. (1985). Children of marital violence: A closer look at the unintended victims. American Journal of Orthopsychiatry, 55, 260 –265. Holden, G. W., & Ritchie, K. L. (1991). Linking extreme marital discord, child rearing, and child behavior problems: Evidence from battered women. Child Development, 62, 311–327. Magdol, L., Moffitt, T. E., Caspi, A., Newman, D. L., Fagan, J., & Silva, P. A. (1997). Gender differences in partner violence in a birth cohort of 21-year-olds: Bridging the gap between clinical and epidemiological approaches. Journal of Consulting and Clinical Psychology, 65, 68 –78. Margolin, L., & Larson, O. W., III. (1988). Assessing mothers’ and fathers’ violence towards children as a function of their involuntary participation in family work. Journal of Family Violence, 3, 209 –223. Merrill, L. L., Hervig, L. K., & Milner, J. S. (1996). Childhood parenting experiences, intimate partner conflict resolution, and adult risk for child physical abuse. Child Abuse & Neglect, 20, 1049 –1065. Oates, K. (1996). The spectrum of child abuse: Assessment, treatment, and prevention. New York: Brunner/Mazel, Inc. O’Keefe, M. (1995). Predictors of child abuse in maritally violent families. Journal of Interpersonal Violence, 10, 3–25. Ross, S. (1996). Risk of physical abuse to children of spouse abusing parents. Child Abuse & Neglect, 20, 589 –598. Schechter, S., & Edleson, J. (1994, June). In the best interest of women and children: A call for collaboration between child welfare and domestic violence constituencies. Briefing paper presented at the conference on Domestic Violence and Child Welfare: Integrating Policy and Practice for Families, Racine, WI. Schechter, S. (1982). Women and male violence: The visions and struggles of the battered women’s movement. Boston: South End Press. Stacey, W., & Shupe, A. (1983). The family secret: Domestic violence in America. Boston: Beacon Press. Stark, E., & Flitcraft, A. H. (1988). Women and children at risk: A feminist perspective on child abuse. International Journal of Health Services, 18, 7–118. Straus, M. A. (1990). Ordinary violence, child abuse, and wife beating: What do they have in common? In M. A. Straus & R. J. Gelles (Eds.), Physical violence in American families: Risk factors and adaptations to violence in 8,145 families (pp. 403– 424). New Brunswick, NJ: Transaction Publishers. Straus, M. A. (1994). Beating the devil out of them: Corporal punishment by parents and its effects on children. New York: Lexington Books. 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: Transaction Publishers. Straus, M. A., & Gelles, R. J. (1986). Societal change and change in family violence from 1975 to 1985 as revealed by two national surveys. Journal of Marriage and the Family, 48, 465– 479. Straus, M. A., & Hamby, S. L. (1997). Measuring physical and psychological maltreatment of children with the Conflict Tactics Scales. In G. Kaufman Kantor & J. L. Jasinski (Eds.), Out of the darkness: Contemporary research perspectives on family violence (pp. 119 –135). Thousand Oaks, 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: Risk factors and adaptations to violence in 8,145 families (pp. 245–261). New Brunswick, NJ: Transaction Publishers. Wauchope, B. A., & Straus, M. A. (1990). Physical punishment and physical abuse of American children: Incidence rates by age, gender, and occupational class. In M. A. Straus & R. J. Gelles (Eds.), Physical violence in American families: Risk factors and adaptations to violence in 8,145 families (pp. 133–143). New Brunswick, NJ: Transaction Publishers. Widom, C. S., & Shepard, R. L. (1996). Accuracy of adult recollections of childhood victimization: Part 1. Childhood physical abuse. Psychological Assessment, 8, 412– 421.