KF Dapus No15

See
discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/223431953
Type of violence, age, and gender

differences in the effects of family
violence on childrens behavior
problems: A mega-analysis
Article in Developmental Review March 2006

DOI: 10.1016/j.dr.2005.12.001
CITATIONS READS
141 1,246
5 authors, including:
Michael E Lamb Eva Guterman

University of Cambridge The Open University of Israel
544 PUBLICATIONS 20,987 9 PUBLICATIONS 484 CITATIONS
CITATIONS
SEE PROFILE
SEE PROFILE
Some of the authors of this publication are also working on these related
projects:
Assessing what jury-eligible Scots believe about child sexual abuse View
project
Implementation of a Forensic Interview Protocol for victims, witness and

perpetrador children View project
All content following this page was uploaded by Michael E Lamb on 03 July 2017.
The user has requested enhancement of the downloaded file.

Developmental Review 26 (2006) 89112
www.elsevier.com/locate/dr
Type of violence, age, and gender diVerences

in the eVects of family violence on childrens
behavior problems: A mega-analysis
Kathleen J. Sternberg a, Laila P. Baradaran a, Craig B. Abbott a,
Michael E. Lamb b,, Eva Guterman a
a
Section on Social and Emotional Development, National Institute of Child Health and Human Development,
Suite 8000 6750, Rockledge Drive, Bethesda, MD 20892, USA
b
Department of Social and Developmental Psychology, Faculty of Social and Political Sciences,
University of Cambridge, Free School Lane, Cambridge CB2 3RQ, UK
Received 1 July 2005; revised 15 December 2005

Available online 15 February 2006
Abstract
A mega-analytic study was designed to exploit the power of a large data set combining raw data
from multiple studies (n D 1870) to examine the eVects of type of family violence, age, and gender
on childrens behavior problems assessed using the Child Behavior Checklist (CBCL). Our Wnd-
ings conWrmed that children who experienced multiple forms of family violence were at greater
risk than children who experienced only one form of abuse, and witnesses of inter-parental vio-
lence were at similar risk as victims of violence. Age moderated the eVects of family violence on
externalizing behavior problems, but not on internalizing behavior problems. No main or interac-
tion eVects involving childrens gender were evident. These results underscore the need to take chil-
drens age, type of violence, and type of outcome into account when examining the eVects of family
violence on childrens behavior problems.
Published by Elsevier Inc.
The authors are grateful to the researchers who provided data for inclusion in these analyses: Dante Cicch-
etti, Robert Emery, George Holden, Honore Hughes, Ernest Jouriles, Milling Kinard, Laura McCloskey, Tim
Moore, Maura OKeefe, Kathy Ritchie, Robbie Rossman, Suzanne Salzinger, and Penelope Trickett. We also
thank Jessica Norris for assistance preparing the data for reanalysis and are especially grateful to Jack McAr-
dle for his advice regarding the mega analytic technique. Unfortunately, Kathleen J. Sternberg died before
completion of this report.
*
Corresponding author. Fax: +44 1223 334 550.
E-mail address: mel37@cam.ac.uk (M.E. Lamb).
0273-2297/$ - see front matter. Published by Elsevier Inc.

doi:10.1016/j.dr.2005.12.001
90 K.J. Sternberg et al. / Developmental Review 26 (2006) 89112
Keywords: Child abuse; Spouse abuse; Family violence; Child adjustment; Behavior problems
In the last few decades, researchers have made remarkable progress understanding fam-
ily violence and its eVects on childrens development. Links between exposure to family
violence and a broad range of behavioral, emotional and social disorders have been
explored using a variety of research designs and methods. Despite substantial progress,
however, it remains unclear how diVerent forms of victimization (e.g., being a victim of
physical child abuse, observing spouse abuse, and being both a victim and observer of
physical abuse) aVect behavior and development when the eVects of other factors are con-
trolled. It is unclear, for example, whether witnessing violence between parents has diVer-
ent eVects than being the direct victim of violence, whether being both a victim and a
witness of violence adds to the negative eVects of experiencing only one type of violence, or
whether the eVects of diVerent types of family violence vary depending on the age and gen-
der of the child. The present study was designed to explore these questions using the mega-
analytic method (McArdle & Horn, 2002) to analyze a large data set compiled by many
researchers.
Developmental psychopathology provides a comprehensive theoretical framework for
attempting to understand the dynamic multidimensional processes whereby family vio-
lence aVects childrens behavior and emotional adjustment (Cicchetti & Toth, 1995; Mash
& Dozois, 1996). Abnormal and unusual circumstances, especially violence within the fam-
ily, disrupt the normal course of growth and adaptation because experiences with caregiv-
ers at one developmental period aVect childrens adaptation to subsequent developmental
issues. Children who fail to develop interpersonal trust, receive little aVection from others,
or have authoritarian parents (all common characteristics of abused children), miss impor-
tant socializing experiences and often fail to develop adaptive attachment strategies, which
in turn makes emotional and behavior problems more likely (Cicchetti & Toth, 1995). Fur-
thermore, children learn emotional regulation naturally through the emotional expressions
and explanations given by their caregivers. Children living in families with violent histories
live in a world of emotional turmoil and extremes, making it diYcult for them to under-
stand, label, and regulate their own internal states appropriately. Thus, the emotional and
behavioral problems shown by abused children may reXect their inadequate attempts to
regulate strong emotions (Crittenden & Ainsworth, 1989).
As this framework would predict, children who have witnessed spousal violence or have
been direct victims of physical abuse or both have more behavioral and emotional prob-
lems than children without such experiences. Reviews of research on the psychological
adjustment of physically abused children and child witnesses of spousal abuse suggest that
all types of family violence are associated with diverse psychological problems including
aggression, anxiety, depression, aggressive peer relationships, poor school performance
and a host of other cognitive, social and emotional diYculties, although associations
between abusive acts and speciWc psychological outcomes are sometimes quite weak (Cic-
chetti & Toth, 1995; Edleson, 1999; JaVe, Wolfe, & Wilson, 1990; Kitzmann, Gaylord,
Holt, & Kenny, 2003; Kolko, Blakely, & Engleman, 1996; Margolin & Gordis, 2004; Salz-
inger, Feldman, Ng-Mak, Mojica, & Stockhammer, 2001; Silverman & Gelles, 2001; Stern-
berg et al., 1993; Wolfe, Crooks, Lee, McIntyre-Smith, & JaVe, 2003; Wolfe & McGee,
1994).
K.J. Sternberg et al. / Developmental Review 26 (2006) 89112 91
Two issues regarding the eVects of diVerent forms of violence on childrens behavior
problems are not yet clariWed, however. One concerns the additional risk when children are
both targets of physical violence and witnesses of spousal violence and the second concerns
the diVerential eVects of witnessing as opposed to being a direct victim of violence. Because
childrens vulnerability to developmental problems increases as the number of stressors
increases (Rutter, 1983), and because marital violence and child physical abuse frequently
co-occur (Appel & Holden, 1998; Jouriles & LeCompte, 1991), researchers have hypothe-
sized that children who experience both types of violence would evince more problems
than children who experienced only one form of violence. Hughes and colleagues (Hughes,
Parkinson, & Vargo, 1989) reported such a double whammy eVect, with children who had
experienced both forms of violence having more severe problems than children who had only
witnessed abuse, but other researchers have either failed to Wnd an additive eVect or have
found it only in speciWc subgroups (e. g., Silverman & Gelles, 2001; Sternberg, Lamb,
Guterman, & Abbott, in press). Recent meta-analyses have been inconclusive. Kitzmann
et al. (2003) reported that children who were physically abused and witnessed violence tended
to have poorer outcomes than those who only witnessed violence, but the average eVect sizes
did not diVer signiWcantly. Wolfe et al. (2003) also reported a trend in this direction, but their
sample included only four relevant studies so their conclusions were weak.
As far as the second issue is concerned, Sternberg et al. (1993) found that being a target
of violence had more negative consequences than being a witness of spousal violence. In
their meta analysis of 8 studies in which these groups were compared, Kitzmann and her
colleagues (2003) found that direct victims tended to have poorer outcomes than witnesses,
although the diVerence was not statistically signiWcant.
The developmental psychopathology approach also emphasizes the need to understand
the impact of potential moderators, such as childrens age and gender, but the relevant
research has yielded contradictory results. Several researchers have found that exposure to
family violence has a particularly negative impact on infants and preschoolers (Fantuzzo,
Boruch, Beriama, & Atkins, 1997; Fantuzzo et al., 1991; Hughes, 1988; Stagg, Wills, &
Howell, 1989). Other researchers have reported more prominent behavioral and emotional
problems among school-aged children exposed to family violence, however (Carlson, 1990;
Edleson, 1999; Sternberg et al., 1993). In addition, the symptom patterns tend to diVer,
making comparisons diYcult when researchers do not use the same comprehensive assess-
ment tools. For example, pre-school children from violent families may suVer from sleep
disturbances, somatic complaints, emotional distress, fears of being alone, and bedwetting
(Osofsky, 1995; Osofsky & Scheeringa, 1997; Zeanah & Scheeringa, 1997), whereas school-
aged children from such families have diYculty adapting to school environments and
establishing healthy relationships: they have poorer levels of academic performance, inter-
act with their peers more aggressively, and show more symptoms of depression and anxiety
(Carlson, 1990; JaVe, Hurley, & Wolfe, 1990a). At diVerent stages, children face speciWc
developmental challenges that can be disrupted by exposure to family violence, whose
eVects thus vary depending on the intensity and form of violence, vulnerability at certain
periods of development, and interactions between developmental stage and exposure to
diVerent types of violence (Margolin & Gordis, 2000). According to the organizational
model of development embraced by the developmental psychopathology perspective, early
experiences provide a foundation for subsequent adaptations that may moderate or exac-
erbate the impact of later events (Lamb, Gaensbauer, Malkin, & Schultz, 1985; Sroufe,
1979; Sroufe, Carlson, Levy, & Egeland, 1999). Thus, exposure to family violence during
early childhood, when the capacity for emotional regulation is emerging and childrens
identiWcation with their parents is strongest, is often thought to have more severe and more
enduring negative eVects on future adaptation than later exposure (Fantuzzo et al., 1991,
1997; Hughes, 1988; Stagg et al., 1989). However, although there is some evidence that the
association between family violence and childrens behavioral adjustment may vary as a
function of the childs developmental status at the time of the exposure, the evidence is
scanty and suggestive rather than conclusive (Trickett & McBride-Chang, 1995).
There is some evidence that the eVects of family violence vary depending on the gender
of the children involved, however. Many studies of family violence suggest that boys are
more vulnerable than girls to the impact of stressful life events including family violence
(JaVe, Wolfe, Wilson, & Zak, 1986). By contrast, several investigators have found that girls
who experienced family violence are more likely to manifest internalizing and externalizing
behavior problems than boys are (Cummings, Pepler, & Moore, 1999; Holden & Ritchie,
1991; Sternberg, Lamb, & Dawud-Noursi, 1998) perhaps because girls are more sensitive
than boys to aVective cues and the states of others (Zahn-Waxler, 1993) or because identiW-
cation with a battered mother leads girls to develop a devalued and victim-like identity
(Chodorow, 1991; Davis & Carlson, 1987). Yet other researchers report similar conse-
quences for boys and girls (Carlson, 1990; Fantuzzo et al., 1991; Grych, Jouriles, Swank,
McDonald, & Norwood, 2000; OKeefe, 1994).
Research on factors such as type of violence, age and gender that moderate the eVects of
violence on childrens adjustment is diYcult to conduct, not least because many subjects
are required, especially when the eVects are small and variable. Single studies typically do
not involve samples big enough to fully test multiple moderating variables. For example, of
the 16 studies reviewed by Widom (1989), 11 had 20 or fewer subjects per group. In a more
recent review (Kolko et al., 1996) only 13 of the 29 studies of child witnesses included com-
parison groups, and 8 of the 13 had 25 or fewer subjects per group. Other problems include
diYculties working with samples that are often at high psychosocial risk; reliance on mea-
sures of unknown reliability and validity; dependence on reports of behavior problems by
individuals who may also be traumatized and/or want to protect their children; and diY-
culties directly accessing child victims, some of whom may be too young to be interviewed
and whose parents may need to approve their participation (Osofsky, 2003). Researchers
have thus turned to meta-analysis to synthesize the results of numerous smaller-scale stud-
ies to investigate moderating eVects (Fleiss & Gross, 1991; Victor, 1995). Two recent meta-
analyses have indicated that children who witness domestic violence have more behavioral
and emotional problems than children who have not witnessed domestic violence (Kitz-
mann et al., 2003; Wolfe et al., 2003). Kitzmann et al. (2003) examined 118 studies and
reported a signiWcant association between exposure to family violence and child behavior
problems. Witnesses had signiWcantly worse outcomes than non-witnesses but witnesses
did not diVer signiWcantly from physically abused victims or physically abused witnesses,
suggesting that children exposed to these three types of family violence were similarly
aVected. By contrast, Wolfe et al. (2003) reported (using the results of 41 studies) that being
abused physically, in addition to witnessing domestic violence, minimally, but not signiW-
cantly, increased the eVects of witnessing alone. In both meta-analyses, the attempt to
examine the moderating eVect of childrens age and gender were limited by the small num-
ber of studies in which the eVects of age and gender were considered and by considerable
heterogeneity within each of the groups. Indeed, as indicated earlier, the moderating eVects
of age and gender have proven to be frustratingly diYcult to illuminate (Cummings, 1998),
and no clear pattern has yet emerged. No meta-analyses on the eVects of child physical
abuse have been published.
Meta-analysis may also not be the best technique for exploring possibly subtle eVects such
as these (Eysenck, 1994; Strube & Hartmann, 1991; Wilson & Rachman, 1983). Not only
must meta-analysts assume that constructs, measures, and scales of measurement are compa-
rable; they also risk making the ecological fallacy: imputing group data to individuals
(McArdle & Horn, 2002; Olkin, 1995; Robinson, 1950). An alternative approach to the syn-
thesis of data from multiple studies is mega-analysis (McArdle & Horn, 2002). Mega-analysis
involves the statistical analysis of individual raw-scores from many previous studies. The
primary diVerence between mega-analysis and meta-analysis is that the latter combines the
summary statistics (eVect sizes and signiWcance levels) from individual studies, whereas mega-
analysis combines the raw-scores from those studies, and such meta-analysis with individual
data has been placed at the pinnacle of a hierarchy of evidence (Olkin, 1995, p. 135).
Mega-analysis is limited by the need for all researchers to have used the same dependent mea-
sures and to make their raw data available, but when multiple raw data sets are available,
mega-analysis oVers several advantages over meta-analysis, especially when there are few
studies available for meta-analysis: greater power, greater precision and reliability of the
parameter estimates, greater Xexibility with respect to what hypotheses can be tested, and
greater Xexibility with respect to the analytic techniques that can be applied to the data.
A review of studies focused on child psychopathology raises many questions about the
ways children could respond to physical abuse and or violence between their parents. Stud-
ies focused on childrens coping strategies in socially stressful situations alert us to the
diverse reactions children might have when their emotional and physical security are not
assured by meaningful adult Wgures. Our Wrst goal in this study was thus to conWrm that
experiencing diVerent types of family violence had distinctive eVects on children. We pre-
dicted that children who had witnessed spousal violence or had been the victims of family
violence or both were more likely to have externalizing and internalizing behavior prob-
lems than children from no-violence comparison groups. Furthermore, we expected that
children who were both victims and witnesses of family violence would have the highest
levels of externalizing and internalizing problems because they experienced violence in two
developmentally crucial relationships. Also, we expected that children who were the targets
of physical violence would be at greater risk for externalizing and internalizing problems
than those who only witnessed it between their parents. Another question, which is often
asked but rarely addressed because of limited sample size, concerns gender diVerences in
the way children of diVerent ages respond to maltreatment. In general, boys appear more
vulnerable to diverse psychosocial stresses than girls (see Zaslow & Hayes, 1986, for a
review) and at Wrst glance this appears to be the case for child physical abuse and exposure
to parental violence as well. A more careful review of the literature suggests, however, that
boys and girls responses may be more diverse, especially when age is considered. Boys and
girls face distinctive psychosocial challenges at diVerent stages of development (Emery,
1982; Fantuzzo & Lindquist, 1989; Hershorn & Rosenbaum, 1985) and these may aVect
their ability to cope with exposure to diVerent kinds of family violence. We explored this
question in a mega-analytic study of raw data compiled in multiple studies in which the
Child Behavior Checklist (CBCL) had been used to assess the eVects of family violence on
children. Our combined data set provided adequate power not only for testing the eVects of
diVerent forms of abuse on childrens behavior problems, but also for testing multiple
moderating variables.
Method
Data gathering
We identiWed 24 studies in which researchers had obtained accounts of childrens mal-

treatment histories and used the CBCL (Achenbach, 1991) to measure their behavior prob-
lems. We asked the principal investigators if we could include their data in a large data set
compiled to explore the diVerential eVects of diverse types of family violence on boys and
girls of diVerent ages. Thirteen of the researchers provided data sets that included informa-
tion about the childs age, gender, and behavior problems, as assessed by the mothers using
the CBCL. Two researchers no longer had copies of their data sets, Wve declined to partici-
pate, and two did not respond to our invitation. In total, we obtained data on 3,030 sub-
jects who had participated in one of 15 studies. All of the studies were conducted in North
America.
Participants
The Wnal sample for this mega-analysis study included 1870 subjects for whom we
had detailed information about age, gender, behavior problems, and family violence
history. Subjects ranged in age from 4 to 14 years, with 986 boys and 884 girls (see
Table 1). Overall, the composite sample largely comprised lower to middle class fami-
lies. Slightly more than half (53%) of the children were from two-parent families, 23%
came from single parent families, and this information was missing for 24% of the cases.
Just under one-third (30%) of the mothers had less than a high school education, 27%
had graduated from high school, 17% had technical training, associates degrees, or had
not completed college, 6% had at least bachelors degrees, and the relevant information
was not available for 19% of the sample. Nearly half (47%) were white, 34% were non-
white (predominantly African-American), and the relevant information was not avail-
able for 19%.
Table 1
Number of participants by type of family violence, childs age and gender (N D 1870)
Age No-violence Witness spouse Victim physical Abused-witness Total
(years) comparison abuse abuse
Male Female Male Female Male Female Male Female
4.00 17 18 12 11 3 4 63 45 173
5.00 21 22 14 14 4 3 52 50 180
6.00 34 40 23 29 6 6 67 56 261
7.00 62 61 33 22 8 6 45 38 275
8.00 48 56 16 23 6 6 36 24 215
9.00 52 58 26 18 12 9 31 22 228
10.00 47 48 16 21 27 17 40 19 235
11.00 38 33 12 15 13 10 31 20 172
12.00 31 12 12 17 9 7 11 15 114
13.00 2 3 1 2 0 0 3 1 12
14.00 1 1 0 1 0 0 1 1 5
Total 353 352 165 173 88 68 380 291 1870

Classifying types of family violence
In our study, children with no report of violence in their family are labeled the no-
violence comparison group, children who were direct victims of physical abuse by one or
both parents are labeled victims, children whose parents were violent with one another are
labeled witnesses, children who were both victims of physical abuse and witnesses of phys-
ical abuse between parents are labeled abused-witnesses. We use the term family violence
to refer to any of these types of violence.
We assigned children to one of the three family violence groups or the no-violence com-
parison group based on Child Protective Service (CPS) records or family members reports
on the ConXict Tactics Scale (CTS: Straus, 1979). The CTS is a 19-item questionnaire orig-
inally developed for telephone surveys concerned with tactics such as reasoning, verbal
aggression, and minor and serious violence used by adults to resolve disputes with partners
or with their children.
Thirty-one percent of the cases were assigned to a violence group on the basis of
CPS records. For these cases, we used the original investigators classiWcation of
victims, witnesses, abused-witnesses, or no-violence comparison based upon the
CPS records. The remaining 69% of the cases were classiWed based on the CTS data
provided by the investigators. We reviewed the CTS information and assigned each
subject to a violence category (victims, witnesses, abused-witnesses) when they
reported one or more instances of minor or serious violence in the relevant relation-
ship(s).
One hundred and Wfty-six of these children were victims, 338 were witnesses, 761 were
abused-witnesses; and 705 were no-violence comparison children from families with no
known history of violence (see Table 1).
Assessing childrens externalizing and internalizing problems
The Externalizing and Internalizing problems scales of the CBCL (Achenbach, 1991)
were used to assess parents perceptions of their childrens externalizing and internalizing
problems. These two scales have high internal consistency, three-month test-retest reliabil-
ity and concurrent validity with other standard measures of child behavior problems, and
discriminate well between clinic-referred and non-clinical children (see Achenbach, 1991,
for further details). T scores were used to categorize the current study participants into two
groups using Achenbachs (1991) criteria to distinguish between non-clinical (T 6 63) and
clinical (T 7 64) risk status.
Results
Our analyses examined the eVects of family violence, age and gender on the extent to
which childrens behavior problems were severe enough to put them at clinical risk. To
better understand the association between behavior problems and age, we divided the
sample into three age groups of roughly equivalent size: 4- to 6-, 7- to 9-, and 10- to 14-
year-olds. These age groups generally corresponded to pre- and early-elementary school,
mid-elementary school, and late elementary to junior high school, three periods associ-
ated with important transitions in the childrens lives. The speciWc age ranges were deter-
mined in part by methodological needs to have large enough cell sizes to permit analyses
of the three factors while ensuring that data in each cell came from multiple
investigators.
Table 2 presents the proportion of children, grouped by the type of violence experi-
enced, age, and gender, who met the criteria for being in the clinical range (T 7 64) for
externalizing and internalizing problems. Proportions ranged from .14 to .50 indicating
that, regardless of the violence group, age, or gender, more children fell in the non-clinical
range than in the clinical range. However, Pearson 2 tests of association and analysis of
the cell residuals showed that the type of violence was signiWcantly associated with the risk
of having behavior problems in the clinical range. In general, the number of children in the
clinical range was signiWcantly greater than expected in the abused-witness group and sig-
niWcantly lower than expected in the no-violence comparison group. SigniWcant associa-
tions were not found in all age-gender combinations, however, suggesting an interaction
among the three predictor variables.
Overview of analysis
We used hierarchical binomial multiple logistic regression analysis to examine the

total and unique eVects of type of family violence, age, and gender on outcome and to
determine whether childrens age and gender moderated the association between the
Table 2
Counts and proportions of children in the clinical range for externalizing and internalizing behavior problems by
type of violence, age, and gender
Boys Girls
46 years 79 years 1014 years 46 years 79 years 1014 years
Externalizing count (p)
No violence 24 (.33) 34 (.21) 19 (.16) 25 (.31) 33 (.19) 16 (.16)
Witness 12 (.24) 19 (.25) 16 (.39) 16 (.30) 21 (.33) 24 (.43)
Victim 3 (.23) 9 (.35) 20 (.41) 4 (.31) 6 (.29) 10 (.29)
Abused-witness 77 (.42) 48 (.43) 24 (.28) 54 (.36) 24 (.29) 22 (.39)
Total 116 (.37) 110 (.29) 79 (.27) 99 (.33) 84 (.24) 72 (.30)
2 (3) 7.009 16.25 15.21 .93 6.619 15.23
Internalizing count (p)
No violence 15 (.21) 38 (.23) 31 (.26) 11 (.14) 41 (.23) 18 (.19)
Witness 10 (.20) 22 (.29) 18 (.44) 18 (.33) 25 (.40) 20 (.36)
Victim 2 (.15) 8 (.31) 18 (.37) 2 (.15) 5 (.24) 8 (.24)
Abused-witness 63 (.35) 45 (.40) 43 (.50) 50 (.33) 36 (.43) 26 (.46)
Total 90 (.28) 113 (.30) 110 (.37) 81 (.27) 107 (.31) 72 (.30)
2 (3) 8.09 8.83 13.14 11.92 12.89 14.89
Note. Numbers in parentheses are the proportion of children in the group with clinically elevated T scores
(T 7 64). Proportions are based on the number of clinical and non-clinical children for that cell [p (clinical) + p
(non-clinical) D 1.0]. Pearson 2 tests of association test the relation between risk status (clinical vs. non-clinical)
and form of violence. Italicized cells have counts less than expected. Bolded cells have counts greater than
expected. SigniWcance levels of cell counts are based on normalized standardized residuals.

p < .05.

p < .01.
p < .001.
9
p < .10.
type of violence experienced and clinical risk.1 Separate but parallel analyses were con-
ducted for externalizing and internalizing behavior problems. For ease of presentation,
we report the main results in three sections. In the Wrst section, we report simple logistic
regressions in which the outcome variables were regressed on each predictor variable by
itself. These analyses show the eVects of each predictor as if it was entered Wrst in the
hierarchal analysis, and indicate the total association between each predictor variable
and outcome. In the second section, we report the omnibus results of the hierarchical
logistic regressions. The primary purpose of these analyses was to test whether age and
gender moderated the eVect of the type of violence, and was accomplished by testing
whether the three-way and two-way interactions were reliable predictors over and above
their lower-order components. The hierarchical analyses also identiWed the most parsi-
monious models for evaluating the focused group comparisons generated from our
hypotheses. In the third section, we present the results of the planned group comparisons
and tests of simple main eVects that follow the omnibus tests. In all analyses, the depen-
dent measure was dummy coded (clinical D 1 and non-clinical D 0). The independent
variables were also dummy coded. In the simple and hierarchical analyses, the omitted
reference groups were the no-violence comparison group, 4- to 6-year-olds, and males
for type of violence, age, and gender respectively. In the focused analyses, the reference
group was changed when necessary to test the signiWcance of the odds ratios that
reXected our hypotheses.
Preliminary analyses
We conducted preliminary analyses to test the appropriateness of estimating parame-

ters from the combined data set by using a modiWed jack-knife procedure in which we
removed the cases from one investigator from the sample and then analyzed the remain-
ing cases. We repeated this procedure for each of the source studies. We determined that
no single data source unduly inXuenced the overall outcome of the logistic regression
and judged that it was appropriate to conduct and report analyses involving the full
data set.
Because some children were classiWed using CPS records and others using the CTS, we
included a variable in all logistic regressions to control for possible diVerences in outcome
in these groups. Preliminary analyses revealed that CTS/CPS had minimal inXuence on
predicting clinically signiWcant internalizing problems, but was a signiWcant predictor of
externalizing behavior problems when other variables were included in the logistic regres-
sion equation. By itself, for example (see Table 6, Step 1), CTS/CPS failed to signiWcantly
predict clinical-level externalizing problems but did so when other variables were included
1
We conducted supplementary analyses in which we analyzed outcome as continuous scores using multiple re-
gression analysis. Externalizing and Internalizing T scores were hierarchically regressed on type of violence, age,
gender, and their interactions, controlling for cts/cps, following steps that paralleled the hierarchical logistic re-
gressions. The pattern of results of the omnibus tests were the same as those of the logistic regression. Type of vi-
olence and type of violence age signiWcantly predicted externalizing problems. Type of violence and age
signiWcantly predicted internalizing problems. Gender and interactions involving gender were not signiWcantly as-
sociated with either outcome. The pattern of results for the group comparisons for type of violence and for age for
externalizing problems were also the same. Two results for type of violence for internalizing problems, however,
were diVerent: victims (57.5 1.68) averaged signiWcantly higher internalizing scores than their no-violence coun-
terparts (54.7 .85, p < .01); abused-witnesses (60.8 .84) did not signiWcantly diVer from witnesses (59.5 1.16).
in the model (see Step 2). When controlling for type of violence, age, and gender, children
classiWed using CPS data had 90% greater odds of high risk externalizing behavior than
children classiWed using the CTS. Preliminary analyses also revealed that the CTS/
CPS age, CTS/CPS gender, CTS/CPS age gender, and CTS/CPS age gender
violence interactions were not signiWcant, indicating no age and gender diVerences
between the CTS and CPS groups.
Simple logistic regression
Simple logistic regressions were conducted to analyze the relation between outcome and
each predictor variable by itself while controlling for CTS/CPS. Type of violence was signiW-
cantly associated with externalizing and internalizing problems (see Table 3). The odds
ratios revealed that children who experienced violence were at 1.95 (victims, p < .01) to 2.91
(abused-witnesses, p < .001) times greater risk for severe externalizing problems, and 1.39
(victims, p D .10) to 2.57 times (abused-witnesses, p < .001) greater risk for severe internaliz-
ing problems than the no-violence comparison children. Additional comparisons between
groups revealed that abused-witnesses were 1.49 (p < .05) and 1.85 (p < .01) times more likely
than victims of externalizing and internalizing problems, respectively, to be in the clinical
range. Abused-witnesses were also at greater risk than witnesses, although not signiWcantly
so (externalizing, 1.27 times greater, p D .09; internalizing 1.21 greater, p D .17). Witnesses
were at greater risk than victims, but again not signiWcantly so (externalizing, 1.23 times
greater, p D .34; internalizing, 1.46 times greater, p D .09). Childrens age also signiWcantly
predicted clinically high levels of externalizing and internalizing problems (see Table 4).
Compared to 4- to 6-year olds, the odds of clinical-level externalizing problems for 7- to 9
and 10- to 14-year-olds were .67 (p < .01) and .69 (p < .01) times smaller, respectively. In con-
trast, the risk for internalizing problems in the clinical range was 1.16 (p D .21) and 1.38
(p < .05) times higher for 7- to 9- and 10- to 14-year-olds, respectively. Childrens gender did
not reliably predict externalizing and internalizing behavior problems (see Table 5).
Table 3
Results of simple logistic regression analysis with type of violence as predictor
Variable B SE Wald 2 df Odds ratio % Odds 95% CI
Externalizing problems
CTS/CPS .64 .13 24.11 1 1.89 +89 1.472.44
Type of violence 60.5 3
Witness .87 .17 27.74 1 2.40 +140 1.733.31
Victim .67 .20 11.66 1 1.95 +95 1.332.86
Abused-witness 1.07 .14 59.04 1 2.91 +191 2.223.82
Internalizing problems
CTS/CPS .25 .13 3.70 1 1.28 +28 1.001.65
Witness .71 .16 19.35 1 2.03 +103 1.482.78
Victim .33 .20 2.66 1 1.39 +39 .942.06
Abused-witness .94 .13 49.52 1 2.57 +157 1.983.34
Note. The No-violence comparison and CTS groups were the reference groups (coded 0).

p < .01.
p < .001.
Table 4
Results of simple logistic regression analysis with childrens age as predictor
CTS/CPS .22 .11 4.06 1 1.25 +25 1.011.55
Age group 12.88 2
79 years .39 .12 10.81 1 .67 33 .53.85
1014 years .37 .13 7.99 1 .69 31 .54.89
CTS/CPS .21 .11 3.48 1 .81 19 .651.01
Age group 6.13 2
79 years .15 .12 1.55 1 1.16 +16 .921.48
1014 years .32 .13 6.13 1 1.38 +38 1.071.78
Note. The 4- to 6-year-olds and CTS group were reference groups (coded 0).

p < .05.
p < .01.
Table 5
Results of simple logistic regression analysis with childrens gender as predictor
CTS/CPS .17 .11 2.38 1 1.18 +18 .961.46
Gender .09 .10 .87 1 .91 9 .741.11
CTS/CPS .17 .11 2.30 1 .85 15 .681.05
Gender .12 .10 1.31 1 .89 11 .731.09
Note. Boys and the CTS group were the reference groups (coded 0).
Hierarchical multiple logistic regression
Hierarchical analyses explored the unique associations between type of violence, age,
gender, and outcome, and tested whether age and gender moderated the eVects of violence.
The analyses were conducted in four steps: at step 1, CTS/CPS was entered to control for
its inXuence on outcome; at step 2, type of violence, age, and gender were entered; at step 3,
the three two-way interaction variables were entered; and at step 4, the three-way interac-
tion variable was entered. For each step we report the Likelihood Ratio 2 which tests the
signiWcance of the block of variables entered at that step, and the Wald 2 which tests the
signiWcance of the unique contribution of each variable controlling for all other variables
in the equation. The Wald 2 tests for violence and age are omnibus tests because these
variables have more than two groups. We defer reporting the results of the group compari-
sons for these variables to the third section of the results. We report two indices of model
Wt, R2L and R2N (Neglekerke), because there is no single agreed-upon index of Wt for logis-
tic regression (Cohen, Cohen, West, & Aiken, 2003). Both of the indices are based on likeli-
hood measures and are interpreted as the proportion of null deviance accounted for by the
set of predictors in the model. Values of both indices range from 0 to 1. R2N is typically
larger than R2L and diVers from the latter by taking into account sample size (Tabachnick
& Fidell, 2001). Logistic R2 measures for good logistic regression models tend to be smaller
than R2 in good models in OLS Regression (Cohen et al., 2003).
Externalizing behavior
Table 6 presents the results of the hierarchical logistic regression for externalizing prob-
lems. Step 2 reveals that only type of violence reliably predicted clinical-level externalizing
problems while controlling for age and gender, and consistently did so in Steps 3 and 4
also. In Step 3 the two-way interactions as a block accounted for a signiWcant proportion
of null deviance, but only the violence age interaction reliably contributed to the model.
Step 4 shows that the three-way interaction did not signiWcantly contribute to the model
over and above the main and two-way interaction components. Although the unique con-
tribution of age was not signiWcant at Step 2, it was signiWcant in the later steps, when the
violence age interaction term was also in the model. Gender and interactions involving
gender, however, were not signiWcant predictors at any step of the analyses. In summary,
then, our results show that type of violence consistently predicted externalizing problems,
even when controlling for other factors, and that childrens age moderated those eVects
whereas gender did not.
Table 6
Hierarchical logistic regression predicting externalizing problems
Variable Model Wt Block df Wald 2
R 2
L R2Na LR (df)
2
Step 1 .001 .002 2.47(1)
CTS/CPS 1 2.49
Step 2 .032 .054 70.48(6)
CTS/CPS 1 24.97
Type of violence 3 52.82
Age 2 5.06
Gender 1 .20
Step 3 .041 .070 21.68(11)
CTS/CPS 1 23.20
Age 2 12.48
Gender 1 .07
Violence age 6 14.26
Violence gender 3 3.79
Age gender 2 2.12
Step 4 .044 .074 6.06(6)
CTS/CPS 1 23.72
Age 2 6.75
Gender 1 .00
Violence age 6 11.82
Age gender 2 .21
Violence age Gender 6 6.04
Note. Reference groups (coded 0): Type of violence, no-violence comparison; Age, 4- to 6-year olds; Gender,
boys; and CTS/CPS, CTS. Variables entered as a block at each step are in bold.
a
Neglekerke R2.

p 6 .05.

p 6 .01.
p 6 .001.
The hierarchical results also indicate that the model including the two-way interac-
tions, Step 3, was the best Wtting yet parsimonious model for predicting high risk exter-
nalizing problems. Using R2N as the Wt index, the model accounted for 5.4% of the null
deviance at Step 2, improved to 7.0% of the null deviance at Step 3, and improved
only slightly to 7.4% of the null deviance at Step 4 (full model). Since gender and its
interactions did not reliably contribute to the model at Step 3, we examined the
consequences of excluding them as recommended by Cohen et al. (2003). Removing
gender, violence gender, and age gender as a block from the Step 3 model did not
signiWcantly alter it ( likelihood ratio 2(6) D 6.21, p D .40). The reduced
model accounted for 6.6% of the null deviance. Based on these results, we concluded
that the best model for evaluating our focused questions with externalizing problems as
the outcome included type of violence, age, and violence age, controlling for CTS/
CPA.
Internalizing behavior
Table 7 presents the results of the hierarchical regression for internalizing problems.
At Step 2, type of violence and age were signiWcant unique predictors of internalizing
problems in the clinical range. Type of violence was a signiWcant predictor in Steps 3 and
4 also, but age was not. In Step 3, the two-way interaction variables did not account for
signiWcant null model deviance, over and above the main eVect variables, and in Step 4,
the three-way interaction did not signiWcantly contribute to the model. Gender and inter-
actions involving gender were not signiWcant predictors at any step of the analyses. In sum-
mary, these results show that type of violence reliably predicted clinically high internalizing
problems but that neither age nor gender moderated its eVects. Age, however, signiWcantly
predicted internalizing problems independent of violence in a model that excluded the
interactions, but gender did not. The Wt indices of the full model (Step 4) changed little
from the reduced model at step 2 indicating that the model that included only the main
eVect variables was the best of the three models for predicting internalizing problems.
Because gender did not reliably contribute to the model at Step 2, we excluded it when
evaluating the focused questions. Even when gender was excluded, the model accounted
for the same level of null deviance (5.2%) as when it was included.
Focused group comparisons
The adjusted odds of externalizing and internalizing behavior problems in the clinical
range for type of violence age, based on the omnibus hierarchical analyses, are presented
in Fig. 1. The adjusted odds were used to compute the odds ratios for the focused tests of
the eVects of type of violence and age. For type of violence, we tested (a) whether children
in the witness, victim, and abused-witness groups were more likely to be at clinical risk
than children in the no-violence group; (b) whether children from the abused-witness
group were more likely to be at clinical risk than children in either the witness or victim
groups; and (c) whether children from the victim group were more likely to be at clinical
risk than children from the witness group. For age group, we tested whether 4- to 6-year-
olds were at greater risk than 7- to 9-year-olds and 10- to 14-year-olds. Because of the sig-
niWcant type of violence age interaction (see Fig. 1A), results for externalizing problems
are reported as simple main eVects.
Table 7
Hierarchical logistic regression predicting internalizing problems
Variable Model Wt Block df Wald 2
R 2
L R2N a
LR (df)
2
Step 1 .001 .002 2.20(1)
CTS/CPS 1 2.18
Step 2 .031 .052 68.54(6)
CTS/CPS 1 2.65
Age 2 14.87
Gender 1 .30
Step 3 .034 .059 8.45(11)
CTS/CPS 1 2.02
Violence 3 11.94
Age 2 2.28
Gender 1 .76
Violence age 6 1.49
Violence Gender 3 3.71
Age Gender 2 2.87
Step 4 .035 .060 2.23(6)
CTS/CPS 1 1.92
Violence 3 8.76
Age 2 .77
Gender 1 1.19
Violence age 6 2.26
Age gender 2 1.46
Violence Age Gender 6 2.21
Note. Reference groups (coded 0): type of violence, no-violence comparison; age, 4- to 6-year olds; gender, boys;
and CTS/CPS, CTS. Variables entered as a block at each step are in bold.
a
Neglekerke R2.

p 6 .05.

p 6 .01.
p 6 .001.
Externalizing behavior
Table 8 presents the simple eVects for comparisons of each of the three violence groups
with the no-violence group. For 4- to 6-year-olds, only the abused-witness children were at
signiWcantly greater risk (1.87 times greater) of reaching the clinical cutoV than the no-vio-
lence children. In the two older age groups, however, children in each of the violence
groups were at signiWcantly greater risk than children in the no-violence group. The 7- to 9-
year-old witnesses, victims, and abused-witnesses were at 2.30 times, 2.27 times, and 3.25
times greater risk than their no-violence peers, respectively. For 10- to 14-year-olds, wit-
nesses, victims, and abused-witnesses were at 4.90 times, 2.66 times, and 3.14 times greater
risk, respectively, than no-violence children.
Comparisons between abused-witnesses and witnesses and victims revealed that
abused-witnesses had the highest odds of severe externalizing problems as 4- to 6-year-
olds and 7- to 9-year-olds, but there was only one signiWcant group diVerence evident in
all the age groups. Four- to six-year-old abused-witnesses were at 1.65 (p < .05) times
A 1.0
0.9
0.8
0.7
Adjusted Odds
0.6
0.5
0.4
0.3
0.2
0.1
0.0
4 to 6 7 to 9 10 to 14
Age Group (Years)
B 1.0
0.9
0.8
0.7
Adjusted Odds
0.6
0.5
0.4
0.3
0.2
0.1
0.0
4 to 6 7 to 9 10 to 14
Age Group (Years)
No-Violence Witness Victim Abused-Witness
Fig. 1. Adjusted odds of (A) externalizing and (B) internalizing behavior problems in the clinical range by type of
violence and age.
greater risk than witnesses of the same age, indicating that, in general, experiencing
both forms of violence did not make clinically high externalizing problems signiWcantly
more likely than did experiencing only one form of violence. Finally, victims and wit-
nesses did not diVer signiWcantly at any age, indicating that children experiencing either
form of violence were similarly likely to have externalizing problems in the clinical
range.
Analyses of the simple eVects for age revealed a signiWcant age eVect only for children in
the no-violence group. The odds of clinically high externalizing problems for 4- to 6- year-
old no-violence children were 1.87 (p < .01) times higher than for the 7- to 9-year- olds, and
2.43 (p < .001) times higher than for the 10- to 14-year-olds.
Internalizing behavior
Although the odds of clinical risk for the witness, victim, and abused-witness
groups increased with age, group diVerences (odds ratios) were consistent across age
Table 8
Results of simple eVects logistic regression predicting externalizing problems from type of violence at each grade
level
46 years (N D 614)
Witness .13 .29 .19 1 1.14 +14 .642.03
Victim .14 .48 .09 1 1.15 +15 .452.98
Abused-witness .63 .22 8.11 1 1.87 +87 1.222.88
79 years (N D 718)
Witness .83 .25 11.49 1 2.30 +130 1.423.71
Victim .82 .35 5.56 1 2.27 +127 1.154.48
Abused-witness 1.18 .22 29.79 1 3.25 +225 2.134.97
1014 years (N D 538)
Witness 1.59 .29 30.60 1 4.90 +390 2.798.60
Victim .98 .30 10.87 1 2.66 +166 1.494.75
Abused-witness 1.15 .27 18.65 1 3.14 +214 1.875.28
Note. The no-violence comparison group was the reference group (coded 0). Tabled values are adjusted by con-
trolling for the eVects of CTS/CPS.

p < .05.

p < .01.
p < .001.
(see Fig. 1B). Comparisons of the three family violence groups to the no-violence
group (see Table 9) revealed that witnesses and abused-witnesses were at signiWcantly
greater risk of clinical-level internalizing problems than the non-violence group (wit-
ness: 2.07 times greater; abused-witness: 2.87 times greater). Although victims had a
32% greater risk of clinical-level problems than no-violence children, the diVerence
was not signiWcant.
Comparisons between abused-witnesses and witnesses or victims revealed that children
in the Wrst group were at 2.17 (p < .001) times greater risk than victims, and at 1.38 (p < .05)
times greater risk than witnesses, of internalizing problems in the clinical range. These
results indicate that, for internalizing problems, experiencing two forms of violence had a
greater negative eVect than experiencing only one form of violence. Lastly, comparison of
witnesses to victims showed that the former were at 1.57 (p < .05) times greater risk than the
latter of being in the clinical range for internalizing problems.
Table 9
Results of logistic regression analysis predicting internalizing problems from type of violence and childrens age
Witness .73 .16 20.38 1 2.07 +107 1.512.85
Victim .28 .20 1.84 1 1.32 +32 .881.97
Abused-witness 1.05 .14 57.73 1 2.87 +187 2.183.76
Age group 15.01 2
79 years .38 .13 8.99 1 1.47 +47 1.141.88
1014 years .50 .14 13.60 1 1.65 +65 1.272.16
Note. The no-violence comparison group and 4- to 6-year-olds were reference groups (coded 0). Tabled values are
adjusted by controlling for the eVects of CTS/CPS.

p < .01.
p < .001.
Age group comparisons indicated that 7- to 9-year-olds were 1.47 (p < .01) times more
likely and 10- to 14-year-olds were 1.65 (p < .001) times more likely than the 4- to 6-year-
olds to have clinically high internalizing behavior problems.
Discussion
There is consensus that children in physically violent families are more likely to have a
variety of psychological, social, and academic problems than other children, but it has not
been clear whether abused-witnesses have poorer outcomes than children who experience
only one form of violence, whether children who are targets of violence have poorer out-
comes than children who witness spousal violence, or how childrens age and gender mod-
erate the adverse impact of diVerent forms of family violence. We sought to address these
issues by analyzing a combined data set with suYcient power to test the eVects of diVerent
forms of violence on behavior problems as well as the moderating eVects of age and gender.
Our results showed that the type of violence, childrens age, and type of outcome all must
be taken into account.
Our Wrst goal was to conWrm that diVerent types of family violence have distinctive
eVects. We predicted that children who had witnessed family violence, been the victims of
family violence, or both were more likely to have clinically signiWcant externalizing and
internalizing behavior problems than children who did not experience violence. Our results
were mixed, however, and the eVects varied by age and outcome. Seven- to 14-year-olds
who experienced any form of violence indeed showed greater risk of clinically-high exter-
nalizing problems than no-violence children, as predicted, but for 4- to 6-year-olds only
abused-witnesses were at signiWcantly greater risk than no-violence children. A diVerent
picture emerged for internalizing problems, furthermore. In all age groups, witnesses and
abused-witnesses, but not victims, were at greater risk for severe internalizing problems
than their no-violence counterparts. Victims, however, averaged signiWcantly higher inter-
nalizing problem scores than children in the no-violence comparison group.
The potentially additive eVects of being both a victim of child abuse and a witness of
spouse abuse (Hughes et al., 1989) were also investigated. In general, vulnerability to devel-
opmental problems increases as the number of stresses increases (Rutter, 1983), and the
predicted double whammy eVect was clearly evident in analyses of internalizing
problems: Abused-witnesses were consistently at greater risk than victims and witnesses
regardless of age. Support was mixed with respect to externalizing behavior, however.
Abused-witnesses were at greater risk among 4- to 9-year-olds but in the oldest age group
(ages 1014), children in the witness group were at greatest risk.
We predicted that physically abused children would be at greater risk for severe exter-
nalizing and internalizing problems than children who only witnessed violence between
their parents. Our data did not support our prediction. Our Wndings are congruent with
those of other researchers who have reported that, on a variety of measures, children who
observe spouse abuse do not diVer consistently from physically abused children (Kitzmann
et al., 2003).
Childrens speciWc cognitive, emotional, and behavioral responses to family violence
may help explain diVerences in the pattern of adjustment among children living with family
violence. Grych and Finchams (1993) cognitive-contextual model and Davies and Cum-
mingss (1994) emotional security hypothesis share the assumption (although they diVer
with respect to their emphasis) that children react to the meaning of family violence, and
that this meaning is shaped by childrens appraisals and emotional response to the conXict
(Cummings, 1998). These appraisals and emotional responses have been conceptualized
both as mediators and moderators of the association between abuse and child well being
(El-Sheikh & Harger, 2001). Thus, we also examined the ways age and gender moderated
the impact of family violence on behavior problems, predicting that family violence during
early childhood would have more negative eVects on childrens behavior and emotional
problems than later violence and would have more reliable eVects on boys than on girls.
Consistent with the organizational model of development, we expected that exposure to
domestic violence during early childhood, when the capacity for emotional regulation is
emerging and childrens identiWcation with their parents is strongest, would have stronger
negative eVects on adaptation than later exposure because early experiences provide a
foundation that may moderate or exacerbate the impact of later events (Lamb et al., 1985;
Sroufe, 1979; Sroufe et al., 1999). Our Wndings were not consistent with this prediction. The
negative eVects of family violence on internalizing behavior were consistent regardless of
age, and the negative eVects on externalizing behavior problems were greater for 7- to 14-
year-olds. This latter Wnding is consistent with Conte and Schuermans (1987) speculation
that physical harm at later ages may have particularly severe consequences because older
children can reXect consciously on the meaning of the abuse and perhaps feel self-blame or
anger. Widom (1998) indeed reported that children who had been abused just prior to ado-
lescence were more likely to be arrested as juveniles than those that had been abused very
early in their lives. Although our Wndings are consistent with these, however, they do run
counter to the predictions of the organizational model and thus highlight the need for lon-
gitudinal studies designed to explore some of the unexpected age diVerences in the eVects of
diVerent types of family violence.
Although childrens age did not moderate the negative eVects of violence on internaliz-
ing behavior problems, it did have a signiWcant main eVect: older children were at greater
clinical risk than younger children. Moreover, since the level of risk for each form of vio-
lence increased with age, the odds ratios comparing the diVerent forms of violence
remained consistent across age (see Fig. 1B). Thus, at all ages children were at greater risk
when they experienced multiple forms of family violence. Abused-witnesses were 187%
more likely to have clinically signiWcant internalizing problems than no-violence compari-
son children, 117% more likely to have such problems than victims and 38% more likely to
have them than witnesses across ages. Witnesses, too, were 107% more likely to have clini-
cally signiWcant internalizing problems than their no-violence counterparts across ages.
The eVects of the diVerent forms of violence on externalizing behavior problems varied
by age, however. Among children in the no violence group, externalizing behavior prob-
lems became less likely with age while those for the three violence groups were more stable.
As a result, 7- to 14-year-old children who were witnesses or victims or abused-witnesses
were more likely to have severe externalizing behavior problems than their peers in the no-
violence comparison group. Children aged 46 who experienced only one form of violence
were similar to their no-violence comparison group counterparts with respect to externaliz-
ing problems and there were no signiWcant diVerences between witnesses and victims of
family violence. Interestingly, however, witnesses tended to be at greater risk than victims
among the 10- to 14-year-olds.
Although the risk of severe externalizing problems as reported by mothers of children in
the three violence groups remained stable across age, the risk of internalizing problems
increased. In contrast, mothers of children in the no-violence comparison group reported
decreased externalizing problems with age and increased internalizing problems with age.
Perhaps mothers evaluate their childrens behavior problems diVerently at diVerent stages
of development, seeing their youngest children (aged 46) as more susceptible to externaliz-
ing problems in the face of family violence. Interestingly, mothers rated children who only
witnessed spousal violence as having more internalizing problems in all age groups but as
having more externalizing problems only in the older age group (10- to 14-year-olds).
The fact that mothers reported more consistent eVects when they themselves were vic-
tims of violence may reXect the fact that mothers were informants regarding both the chil-
drens behavior problems and their abuse status. Alternately, mothers who are direct
victims of abuse may project their own frustrations or guilt onto their children, and dis-
tressed mothers are known to evaluate their childrens behavior more harshly than objec-
tive observers do (Brody & Forehand, 1986; Hughes, 1988; Hughes & Barad, 1983). As a
result, a better understanding of childrens adjustment can be wrought when researchers
obtain information from multiple informants (Achenbach, McConaughy, & Howell, 1987).
Although self-reports are also biased and thus should not be the sole source of informa-
tion, they provide insight into childrens adjustment and interpretation when used in con-
junction with others reports.
Contrary to some reported Wndings (Cummings et al., 1999; Holden & Ritchie, 1991;
JaVe et al., 1986; Sternberg et al., 1993), gender did not inXuence the eVects of family vio-
lence on childrens behavior problems either when it was the sole predictor or in associa-
tion with age and type of violence. These results are consistent with many reports showing
no gender diVerences in the eVects of family violence (Christopoulos et al., 1987; Davis &
Carlson, 1987; Holden & Ritchie, 1991; Hughes et al., 1989; Moore et al., 1990), and with
the results of two recent meta-analyses (Kitzmann et al., 2003; Wolfe et al., 2003). Of
course, it is possible that the mechanisms linking risk to well being and adjustment are
diVerent for boys and girls and this may explain the somewhat inconsistent patterns of
results.
It is important to note that, regardless of the type of abuse, age, or gender, more chil-
dren had scores in the non-clinical range than in the clinical range. Even in the abused-wit-
ness group, where the risks were consistently highest, only 2850% of the children in any
cell had clinically signiWcant behavior problems. Other researchers have likewise observed
that the majority of children who witness family violence or who are victims of physical
abuse do not show adverse eVects (Edleson, 1999; Margolin & Gordis, 2000), suggesting an
impressive level of resilience in response to poor rearing circumstances, although it is possi-
ble that some of the processes that help children adapt successfully to abusive families may
presage future problems (Graham-Bermann, 2002). Such developmental patterns obvi-
ously need to be explored in longitudinal studies. By the same token, clinically signiWcant
behavior problems were evident in 1435% of the children not exposed to family violence,
underscoring the fact that family violence is only one of many factors that aVects childrens
behavioral problems adversely.
Mega-analysis and meta-analysis are both eVective means for combining the results of
numerous studies in hopes of gaining a clearer, more precise assessment of the eVects of
violence on children. This mega-analysis and the two recent meta-analyses on domestic
violence (Kitzmann et al., 2003; Wolfe et al., 2003) address some common questions, but
other questions were unique to this mega-analysis. For example, in addition to comparing
witnesses to other groups, we compared abused-witnesses, victims, and children not
exposed to violence. Although the latter three groups have been compared in smaller
studies, they have not been considered in meta- or mega-analyses. Further, all three reviews
tested for the moderating eVects of age and gender, but the current study was able to
extend the meta-analyses by testing the main eVects of age, gender and the age by gender
interaction. Meta-analyses were limited to testing moderating eVects (do group diVerences
vary across age or gender?) because the unit of analysisthe eVect sizeis a standardized
group diVerence and not an estimate for a single group. By contrast, we were able to esti-
mate the likelihood of severe problems for boys and girls in three age groups and make
direct comparisons between groups.
Our results were consistent with several of those obtained in the meta-analyses, but
there were some important diVerences. Unlike the meta-analyses, we were able to fully
test and Wnd evidence for a double whammy eVect on internalizing behavior problems.
Kitzmann et al. (2003) only compared abused-witnesses to witnesses, and thus were
unable to ask whether abused-witnesses had poorer outcomes than victims. They found
a non-signiWcant trend for abused-witnesses to have poorer outcomes than witnesses and
it is possible that the diVerence was not signiWcant because they could include only 18
eVect size estimates in their analyses. Wolfe and his colleagues (Wolfe et al., 2003)
included only four eVect sizes for the same comparison, and did not compare witnesses
and victims while Kitzman et al., included 8 eVect sizes and found a non-signiWcant ten-
dency for victims to have poorer outcomes than witnesses. Although our combined data
came from only 15 studies, our use of individual subject data instead of summary statis-
tics permitted more powerful tests. Despite the fact that witnesses and victims did not
diVer with respect to externalizing problems in our study, witnesses were more likely to
have internalizing problems than victims. Finally, neither of the meta analyses found
that age moderated the eVects of violence, but we found both a main and moderating
eVect for age.
Meta-analysis provides a quantitative summary of Wndings in terms of an overall eVect
size and, more importantly, allows the identiWcation of factors that moderate average eVect
sizes. Glass (1976) deWned meta-analysis as the statistical analysis of the results of individ-
ual studies. Mega-analysis, however, is more akin to a primary investigation in which the
researcher collects and analyzes individual subject data from multiple research sites (J.J.
McArdle, personal communication, March 2005). The goal of our mega-analysis was to
combine data from a large group of studies to seek answers to very speciWc questions
which have been elusive in the past because of small sample sizes and other methodological
issues.
The strength of mega-analysis notwithstanding, the study had several limitations.
First, we examined only the mothers evaluations of their childrens behavioral prob-
lems. Mothers, fathers and teachers observe and interact with children in diVerent con-
texts and therefore should not be expected to report the same behavior problems
(Achenbach et al., 1987; Jouriles, Mehta, McDonald, & Francis, 1997; Youngstrom,
Loeber, & Stouthamer-Loeber, 2000). As a result, information about the childrens
adjustment should ideally be obtained independently from mothers, fathers, teachers,
and children themselves. As mentioned above, furthermore, the highest levels of inter-
nalizing problems at all ages, as well as the highest levels of externalizing problems in
the oldest age group, were reported on the part of children in the abused-witness and
witness groupsgroups in which mothers were themselves victims of family violence.
Sternberg et al. (1993) also found that women who had been abused by their husbands
reported the highest level of behavior problems whereas women who had not
themselves been abused reported similar levels of behavior problems whether or not the
children had been abused. By contrast, children who had themselves been abused (those
in the child abuse and abused-witness groups) reported the highest level of behavior
problems, whereas those exposed to no abuse or spouse abuse only had similarly low
levels of reported behavior problems. Likewise, Kitzmann et al. (2003) observed that in
studies comparing witnesses and nonwitnesses, eVect sizes based on mothers reports
were signiWcantly larger than those based on childrens reports. These results raise con-
cerns that mothers ratings may be inXated because of their own distress, especially
when evaluating their childrens internalizing behavior problems. The exclusive reliance
on parents reports of maltreatment is also potentially problematic (Dodge, Bates, &
Petit, 1990). As shown by Sternberg et al. (in press), researchers who rely solely on
mothers or teachers evaluations are likely to Wnd far more children in need of interven-
tion than those who rely only on childrens reports, whereas decisions about the need
for intervention based solely on childrens reports are likely to overlook some children
and families in need of assistance. Whatever the reasons for the diVerences among infor-
mants, in sum, we are more likely to understand the eVects of family violence if we
obtain information from multiple sources.
Second, like other measures of family violence, the CTS has both strengths and limita-
tions, but many researchers use only this tool to determine whether or not violence has
occurred. Sole reliance on the CTS may promote inaccuracies, especially when researchers
are reliant on one informant. Family violence is a multidimensional, complex problem. It
warrants the use of multiple measures, in addition to measures exploring the speciWc con-
text, meanings, and motives of respondents. In the present mega-analysis, however, we
included data from several studies in which the CTS was the sole index of family violence.
These limitations notwithstanding, our results provided strong support for the follow-
ing key Wndings: (1) to evaluate the eVects of family violence on childrens behavior and
emotional problems we need to take childrens age, type of violence, and type of outcome
into account. (2) Victims of physical child abuse are not at greater risk of behavioral and
emotional problems than children who witness spouse abuse, but children who are both
victims and witnesses of family violence are at the highest risk of externalizing and inter-
nalizing problems, and (3) there are no gender diVerences in the eVects of family violence
on children.
Taken together, the results underscore the importance of multidimensional, interac-
tive approaches to the study of child development (Rutter & Sroufe, 2000). After three
decades of research on family violence, no single risk factor has been identiWed, no sin-
gle pattern of response to maltreatment has been observed, and much of the variability
in outcome remains unexplained. Many children are adversely aVected yet even more
appear to be resilient. Several fundamental questions remain unanswered: Does early
exposure to family violence trigger more profound developmental disruption than later
exposure? Does family violence have delayed eVects? Do delayed eVects occur regard-
less of whether abuse continues? How does the frequency and severity of family vio-
lence aVect childrens outcomes? Our study has illustrated the advantages, such as
increased power and analytic Xexibility, of mega-analysis. Our understanding of family
violence would beneWt from further application of this technique, because so few
researchers have access to samples that are suYciently large and heterogeneous to per-
mit examination of all the factors that moderate the eVect of family violence on chil-
drens adjustment.
References
Achenbach, T. M. (1991). Manual for child behavior checklist proWle. Burlington, VT: University Associates in
Psychiatry.
Achenbach, T. M., McConaughy, S., & Howell, C. T. (1987). Child/adolescent behavioral and emotional problems:
Implications of cross-informant correlations for situational speciWcity. Psychological Bulletin, 2, 213232.
Appel, A. E., & Holden, G. W. (1998). The co-occurrence of spouse and physical child abuse: A review and
appraisal. Journal of Family Psychology, 12, 578599.
Brody, G. H., & Forehand, R. (1986). Maternal perception of child maltreatment as a function of the combined
inXuence of child behavior and maternal depression. Journal of Consulting and clinical Psychology, 54,
237240.
Carlson, B. E. (1990). Adolescent observers of marital violence. Journal of Family Violence, 5, 285299.
Christopoulos, C., Cohn, D. A., Shaw, D. S., Joyce, S., Sullivan-Hanson, J., Kraft, S. P., et al. (1987). Children of
abused women: I. Adjustment at time of shelter residence. Journal of Marriage and the Family, 49, 611619.
Chodorow, N. J. (1991). Feminism and psychoanalytic theory. New Haven, CT: Yale University Press.
Cicchetti, D., & Toth, S. L. (1995). Child maltreatment and attachment organization: Implications for interven-
tion. In S. Goldberg & R. Muir (Eds.), Attachment theory: Social, developmental, and clinical perspectives (pp.
279308). Hillsdale, NJ: Analytic Press.
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the
behavioral sciences (3rd ed.). Mahwah, NJ: Erlbaum.
Conte, J. R., & Schuerman, J. R. (1987). Factors associated with an increased impact of child sexual abuse. Child
Abuse and Neglected, 11, 210211.
Crittenden, P. M., & Ainsworth, M. (1989). Attachment and child abuse. In D. Cicchetti & V. Carlson (Eds.), Child
maltreatment: Theory and research on the causes and consequences of child abuse and neglected (pp. 432463).
New York: Cambridge University Press.
Cummings, E. M. (1998). Stress and coping approaches and research: The impact of marital conXict on children.
Journal of Aggression, Maltreatment and Trauma, 2, 3150.
Cummings, J. G., Pepler, D. J., & Moore, T. E. (1999). Behavior problems in children exposed to wife abuse: Gen-
der diVerences. Journal of Family Violence, 14, 133156.
Davis, L. V., & Carlson, B. E. (1987). Observation of spouse abuse: What happens to the children. Journal of Inter-
personal Violence, 2, 278291.
Davies, P. T., & Cummings, E. M. (1994). Marital conXict and child adjustment: An emotional security hypothe-
sis. Psychological Bulletin, 116(3), 387411.
Dodge, K. A., Bates, J. E., & Petit, G. S. (1990). Mechanisms in the cycle of violence. Science, 250, 16781683.
Edleson, J. L. (1999). Introduction to special issue. Child Maltreatment, 4, 9192.
El-Sheikh, M., & Harger, J. W. (2001). Exposure to interparental conXict and childrens adjustment and physical
health: The moderating role of vagal tone. Child Development, 72(6), 16171636.
Emery, R. E. (1982). Interparental conXict and children of discord and divorce. Psychological Bulletin, 92,
310330.
Eysenck, H. J. (1994). Systematic reviews: Meta-analysis and its problems. British Medical Journal, 309, 789792.
Fantuzzo, J., Boruch, R., Beriama, A., & Atkins, M. (1997). Domestic violence and children: Prevalence and risk in
Wve major U.S. cities. Journal of the American Academy of Child and Adolescent Psychiatry, 36, 116122.
Fantuzzo, J., DePaola, L. M., Lambert, L., Martino, T., Anderson, G., & Sutton, S. (1991). EVects of interparental
violence on the psychological adjustment and competencies of young children. Journal of Consulting and Clin-
ical Psychology, 59, 258265.
Fantuzzo, J., & Lindquist, C. (1989). Violence in the home: The eVects of observing conjugal violence on children.
Journal of Family Violence, 4, 7790.
Fleiss, J. L., & Gross, A. J. (1991). Meta-analysis in epidemiology, with special reference to studies of the associa-
tion between exposure to environmental tobacco smoke and lung cancer: A critique. Journal of Clinical Epide-
miology, 44, 127139.
Glass, G. V. (1976). Primary, secondary, and meta-analysis. Educational Researcher, 5, 38.
Graham-Bermann, S. A. (2002). Child abuse in the context of domestic violence. In J. E. B. Myers, L. Berliner, J.
Briere, C. T. Hendrix, C. Jenny, & T. A. Reid (Eds.), The APSAC handbook on child maltreatmentsecond ed.,
pp. 119129). Thousand Oaks, CA: Sage Publications.
Grych, J. H., & Fincham, F. D. (1993). Childrens appraisals of marital conXict: Initial investigations of the cogni-
tive-contextual framework. Child Development, 64(1), 215230.
Grych, J. H., Jouriles, E. N., Swank, P. R., McDonald, R., & Norwood, W. D. (2000). Patterns of adjustment
among children of battered women. Journal of Consulting and Clinical Psychology, 68, 8494.
Hershorn, M., & Rosenbaum, A. (1985). Children of marital violence: A closer look at the unintended victims.
American Journal of Orthopsychiatry, 55(2), 260266.
Holden, G. W., & Ritchie, K. L. (1991). Linking extreme marital discord, child rearing, and child behavior prob-
lems: Evidence for battered women. Child Development, 13, 5168.
Hughes, H. M., & Barad, S. J. (1983). Psychological functioning of children in battered womens shatter: A model
preventative program. Family Relations, 31, 495502.
Hughes, H. M., Parkinson, D., & Vargo, M. (1989). Witnessing spouse abuse and experiencing physical abuse: A
double whammy. Journal of Family Violence, 4, 197209.
Hughes, H. M. (1988). Psychological and behavioral correlates of family violence in child witnesses and victims.
American Journal of Orthopsychiatry, 58, 7790.
JaVe, P. G., Hurley, D. J., & Wolfe, D. (1990a). Childrens observations of violence: I. Critical issues in child develop-
ment and intervention planning. Canadian Journal of Psychiatry Revue Canadienne de Psychiatrie, 35, 466470.
JaVe, P. G., Wolfe, D. A., & Wilson, S. K. (1990). Children of battered women. Thousand Oaks, CA: Sage Publications.
JaVe, P., Wolfe, D., Wilson, S., & Zak, L. (1986). Family violence and child adjustments: A comparative analysis of
girls and boys behavioral symptoms. American Journal of Psychiatry, 143, 7477.
Jouriles, E. N., & LeCompte, S. H. (1991). Husbands aggression towards wives and fathers aggression toward
children: Moderating eVects of child gender. Journal of Consulting and Clinical Psychology, 59, 190192.
Jouriles, E. N., Mehta, P., McDonald, R., & Francis, D. J. (1997). Psychometric properties of family members
reports of parental physical aggression toward clinic-referred children. Journal of Consulting and Clinical Psy-
chology, 65, 309318.
Kitzmann, K. M., Gaylord, N. K., Holt, A. R., & Kenny, E. D. (2003). Child witnesses to domestic violence: A
meta-analytic review. Journal of Consulting and Clinical Psychology, 71, 339352.
Kolko, J., Blakely, E., & Engleman, D. (1996). Children who witness domestic violence: A review of the empirical
literature. Journal of Interpersonal Violence, 11, 193283.
Lamb, M. E., Gaensbauer, T. J., Malkin, C. M., & Schultz, L. A. (1985). The eVects of child maltreatment on secu-
rity of infant-adult attachment. Infant Behavior and Development, 8, 3545.
Margolin, G., & Gordis, E. B. (2000). The eVects of family and community violence on children. Annual Review of
Psychology, 51, 445479.
Margolin, G., & Gordis, E. B. (2004). Childrens exposure to violence in the family and community. Current Direc-
tions in Psychological Science, 13, 152155.
Mash, E. J., & Dozois, J. A. (1996). Child psychopathology: A developmental-systems perspective. In E. J. Mash &
R. A. Barkley (Eds.), Child psychopathology (pp. 360). New York: Guilford.
McArdle, J. J., & Horn, J. L. (2002). The beneWts and limitations of mega-analysis with illustrations for the WAIS.
Paper presented at the international meeting of CODATA, Montreal, QE.
Moore, T., Pepler, D., Weinberg, B., Hammond, L., Waddell, J., & Weiser, L. (1990). Research on children from
violent families. Canadas Mental Health, 38, 1923.
Olkin, I. (1995). Statistical and theoretical considerations in meta-analysis. Journal of Clinical Epidemiology, 48, 133146.
OKeefe, M. (1994). Linking marital violence, motherchild/fatherchild aggression, and child behavior problems.
Journal of Family Violence, 9, 6378.
Osofsky, J. D. (2003). Prevalence of childrens exposure to domestic violence and child maltreatment: Implica-
tions for prevention and intervention. Clinical Child and Family Psychology Review, 6, 161170.
Osofsky, J. D. (1995). The eVect of exposure to violence on young children. American Psychologist, 50, 782788.
Osofsky, J. D., & Scheeringa, M. S. (1997). Community and domestic violence exposure: EVects on development
and psychopathology. In D. Cicchetti & S. L. Toth (Eds.), Developmental perspectives on trauma: Theory,
research, and intervention (pp. 155180). Rochester, NY: University of Rochester Press.
Robinson, W. S. (1950). Ecological correlations and the behavior of individuals. American Sociological Review,
15, 351357.
Rutter, M. (1983). Stress, coping, and development: Some issues and some questions. In N. Garmezy & M. Rutter
(Eds.), Stress, coping, and development in children (pp. 141). New York: McGraw-Hill.
Rutter, M., & Sroufe, L. A. (2000). Developmental psychopathology: Concepts and challenges. Development and
Psychopathology, 12(3), 265296.
Salzinger, S., Feldman, R. S., Ng-Mak, D. S., Mojica, E., & Stockhammer, T. F. (2001). The eVect of physical abuse
on childrens social and aVective status: A model of cognitive and behavioral processes explaining the associa-
tion. Development and Psychopathology, 13, 805825.
Silverman, A., & Gelles, R. J. (2001). The double whammy revised: The impact of exposure to domestic violence
and being a victim of parent to child violence. Indian Journal of Social Work, 62, 305327.
Sroufe, L. A. (1979). The coherence of individual development: Early care, attachment, and subsequent develop-
mental issues. American Psychologist, 34, 834841.
Sroufe, L. A., Carlson, E. A., Levy, A. K., & Egeland, B. (1999). Implications of attachment theory for develop-
mental psychopathology. Development and Psychopathology, 11, 113.
Stagg, V., Wills, G. D., & Howell, M. (1989). Psychopathology in early child witnesses of family violence. Topics in
Early Childhood Special Education, 9, 7387.
Sternberg, K. J., Lamb, M. E., & Dawud-Noursi, S. (1998). Understanding domestic violence and its eVects: Mak-
ing sense of divergent reports and perspectives. In G. W. Holden, R. GeVner, & E. W. Jouriles (Eds.), Children
exposed to family violence (pp. 121156). Washington, DC: American Psychological Association.
Sternberg, K. J., Lamb, M. E., Greenbaum, C., Cicchetti, D., Dawud-Noursi, S., Cortes, R. M., et al. (1993). EVects
of domestic violence on behavior problems and depression. Developmental Psychology, 29, 4452.
Sternberg, K.J., Lamb M.E., Guterman, E., & Abbott, C.B. (in press). EVects of early and later domestic violence
on childrens behavior problems and depression: A longitudinal, multi-informant perspective. Child Abuse and
Neglect.
Straus, M. A. (1979). Measuring intrafamily conXict and violence: The conXict tactics scales. Journal of Marriage
and the Family, 41, 7588.
Strube, M. J., & Hartmann, D. P. (1991). A critical appraisal of meta-analysis. British Journal of Clinical Psychol-
ogy, 21, 129139.
Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics (4th ed.). Needham Heights, MA: Allyn &
Bacon.
Trickett, P. K., & McBride-Chang, C. (1995). The developmental impact of diVerent forms of child abuse and
neglect. Developmental Review, 15, 311337.
Victor, J. (1995). The challenge of meta-analysis: Discussion. Indications and contra-indications for meta-
analysis. Journal of Clinical Epidemiology, 48, 58.
Widom, C. S. (1989). Does violence beget violence? A critical examination of the literature. Psychological Bulletin,
106, 328.
Widom, C. S. (1998). Child victims: Searching for opportunities to break the cycle of violence. Applied and Preven-
tive Psychology, 7, 225234.
Wilson, G. T., & Rachman, S. J. (1983). Meta-analysis and the evaluation of psychotherapy outcome: Limitations
and liabilities. Journal of Consulting and Clinical Psychology, 51, 5464.
Wolfe, D. A., Crooks, C. V., Lee, V., McIntyre-Smith, A., & JaVe, P. G. (2003). The eVects of childrens exposure to
domestic violence: A meta-analysis and critique. Clinical Child and Family Psychology Review, 6, 171187.
Wolfe, D. A., & McGee, R. (1994). Dimensions of child maltreatment and their relationship to adolescent adjust-
ment. Developmental Psychopathology, 6, 165181.
Youngstrom, E., Loeber, R., & Stouthamer-Loeber, M. (2000). Patterns and correlates of agreement between par-
ent, teacher, and male adolescent ratings of externalizing and internalizing problems. Journal of Consulting
and Clinical Psychology, 68, 10381050.
Zahn-Waxler, C. (1993). Warriors and worriers: Gender and psychopathology. Development and Psychopathol-
ogy, 5, 7989.
Zaslow, M., & Hayes, C. (1986). Sex diVerences in childrens responses to psychosocial stress: Toward a cross-cul-
tural context analysis. In M. E. Lamb, A. L. Brown, & B. RogoV (Eds.), Advances in developmental psychology
(Vol. 4, pp. 285338). Hillsdale, NJ: Erlbaum.
Zeanah, C. H., & Scheeringa, M. S. (1997). The experience and eVects of violence in infancy. In J. D. Osofsky (Ed.),
Children in a violent society (pp. 97123). New York, NY: Guilford Press.
View publication stats

KF Dapus No15

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

KF Dapus No15

Caricato da

Copyright:

Formati disponibili

See

discussions, stats, and author profiles for this publication at:

Type of violence, age, and gender

Article in Developmental Review March 2006

Michael E Lamb Eva Guterman

Implementation of a Forensic Interview Protocol for victims, witness and

The user has requested enhancement of the downloaded file.

Type of violence, age, and gender diVerences

Received 1 July 2005; revised 15 December 2005

0273-2297/$ - see front matter. Published by Elsevier Inc.

We identiWed 24 studies in which researchers had obtained accounts of childrens mal-

Total 353 352 165 173 88 68 380 291 1870

Classifying types of family violence

Assessing childrens externalizing and internalizing problems

We used hierarchical binomial multiple logistic regression analysis to examine the

We conducted preliminary analyses to test the appropriateness of estimating parame-

Simple logistic regression

Hierarchical multiple logistic regression

Focused group comparisons

No-Violence Witness Victim Abused-Witness

View publication stats

Potrebbero piacerti anche