Studies were obtained by conducting computerized database searches of PsycLIT
from 1974 to 1995, Sociofile from 1974 to 1995, PsycInfo from 1967 to 1995, Dissertation
Abstracts International up to 1995, and ERIC from 1966 to 1995. Key terms entered for
these databases were adjustment or effect or effects, college or undergraduate or
undergraduates, and sex abuse or sexual abuse or child and adult and sexual. Studies
that we already knew were also included. Reference lists of all obtained studies were read
to locate additional studies.
To be included, studies must either have used samples exclusively of college
students, or, if noncollege subjects were also included, then results of measures of
college students had to be reported separately. For inclusion in analyses of psychological
correlates of CSA, studies had to (a) include a control group that contained no students
with CSA experiences; (b) use a distinct CSA group, rather than a general
"abused" group that could include participants without a history of CSA; (c)
report on at least one of the 18 symptoms described below; and (d) provide sufficient data
to compute one or more effect sizes. Studies not including reports of psychological
correlates were included if they contained data on reactions to CSA, either
retrospectively recalled or current reflections; these data had to be classifiable into
mutually exclusive negative, neutral, or positive categories. Studies were also included
if they contained data on self-reported effects of CSA.
As in other meta-analyses (e.g., Jumper, 1995 ; Oliver & Hyde, 1993 ), a
single study could report data for more than one sample. Fromuth and Burkhart (1989)
examined two male student samples - one from the Midwest and another from the Southeast-and
reported separate statistics for these two samples. These samples were thus treated as
distinct. Further, male and female samples within a single study were treated as distinct
when results were reported separately for them (cf. Rind &
Tromovitch, 1997 ); this was done to examine gender differences. Many studies reported
more than one result, using different measures, for the same psychological correlate
(e.g., a depression result from the Beck Depression Inventory and another from the Symptom
Checklist). In these cases, effect sizes ( r s) were computed for each result and
were then averaged using Fisher Z transformations to obtain a single mean effect
size. This practice has been used in other meta-analyses (e.g., Erel &
Burman, 1995 )
and has been recommended by Rosenthal (1984) . The mean effect size thus computed for a
given sample for a particular psychological correlate constituted a
"symptom-level" effect size. Finally, numerous studies reported results for more
than one type of psychological correlate from a single sample (e.g., anxiety and
depression). As in other meta-analyses (e.g., Neumann et al., 1996 ), we treated multiple
different correlates in two ways. First, we computed for each sample with multiple
different psychological correlates a "sample-level" effect size by averaging the
symptom-level effect sizes from that sample using Fisher Z transformations. We
later conducted a meta-analysis on these sample-level effect sizes. Second, we analyzed
different psychological correlates (i.e., symptoms) separately in a series of
symptom-level meta-analyses.
Applying the above criteria produced 59 usable studies (see the Appendix ), consisting of 36 published studies, 21 unpublished
dissertations, and 2 unpublished master's theses. These studies yielded 70 independent
samples for estimating prevalence rates, 54 independent samples for computing 54
sample-level and 214 symptom-level effect sizes, 21 independent samples that provided
retrospectively recalled reaction data, 10 independent samples that provided data on
current reflections, and 11 independent samples that provided data on self-reported
effects. Prevalence rates were based on 35,703 participants (13,704 men and 21,999 women).
Effect size data for psychological correlates were based on 15,824 participants (3,254 men
from 18 samples and 12,570 women from 40 samples) - actual numbers of participants are
somewhat higher than these because one study, not included in the above totals ( Haugaard
& Emery, 1989 ), failed to provide exact sample sizes for men and women. Reaction and
self-reported effects data were based on 3,136 participants (783 men from 13 samples and
2,353 women from 14 samples) -actual numbers of participants are somewhat higher because
one study, not included in the above totals ( Schultz & Jones, 1983 ), failed to
report exact sample sizes for men and women.
For each study, the following information was coded:
(a) all statistics, if provided, on psychological correlates of CSA, including
means, standard deviations, t tests, F ratios, correlations, chi squares,
degrees of freedom, and sample sizes;
(b) types of psychological correlates reported;
(c) all statistics regarding relations between moderator variables (e.g., force,
penetration, frequency of CSA) and psychological correlates;
(d) sex of participants;
(e) definition of CSA, including ages that defined a "child" and an older
person, whether peer experiences were included, whether CSA experiences were limited to
contact sex or also included noncontact sexual experiences, and whether CSA experiences
were limited to unwanted sex or also included willing sexual experiences;
(f) all reaction data, if provided, including both retrospectively recalled
reactions to and current reflections on the CSA experiences;
(g) all self-reported effects data, if provided, including responses to how these
experiences affected participants overall and how they affected their sex lives;
(h) types of family environment measures used; and
(i) all statistics on family environment measures, including their
[Page 28]
relations with CSA and with psychological correlates.
Together, the three basic sets of statistics (differences between CSA and
control participants in adjustment, differences between CSA and control participants in
family environment, and the relationship between family environment and adjustment) were
used to address the question of whether significant relationships between CSA and
adjustment were spurious, attributable to the confounding variable of family environment.
Finally, the results of all analyses using statistical control were coded (e.g., examining
the relationship between CSA and adjustment, holding family environment factors constant).
These data were used to directly examine whether any significant relations between CSA and
psychological adjustment were spurious.
Coding of the studies resulted in 18 categories of psychological correlates of
CSA; several additional correlates were infrequently reported and were therefore not
considered in the meta-analyses. These 18 correlates, along with the measures used to
assess them in the various studies, were as follows:
1. Alcohol problems - based on the Michigan Alcoholism Screening Test (MAST;
Brady, Foulks, Childress, & Pertschuk, 1982 ), the alcohol subscale of the Millon
Clinical Multiaxial Inventory (MCMI; Millon, 1982 ), and investigator-authored items.
2. Anxiety - based on the Anxiety subscale of the Symptom Checklist (SCL-90-R;
Derogatis, Lipman, & Covi, 1973 ), the Hopkins Symptom Checklist (HSCL;
Derogatis, Lipman, Rickels, Ulenhuth, & Covi, 1974 ), the Brief Symptom Inventory (BSI; Derogatis
& Spencer, 1982 ), the Trauma Symptom Checklist (TSC-33 and TSC-40; Briere &
Runtz, 1989 ), the MMPI form R ( Hathaway & McKinley, 1967 ), the MCMI, the Institute
of Personality and Ability Testing Anxiety Scale Questionnaire (IPAT; Krug,
Scheier, & Cattell, 1976 ), the State-Trait Anxiety Inventory (STAI;
Spielberger, Gorsuch, & Lushene, 1970 ), and investigator-authored items.
3. Depression - based on the Depression subscales of the SCL-90-R, the
HSCL, the
BSI, the TSC-33 and 40, the MMPI form R, the Hugo Short Form of the MMPI (HSF; Hugo, 1971
), and the MCMI; depression-related items from the Clinical Analysis Questionnaire
(CAQ; Cattell, 1973 ); the Beck Depression Inventory (BDI; Beck, Ward,
Mendelson, Mock, & Erbaugh, 1961 ); and investigator-authored items.
4. Dissociation - based on the Dissociative Experiences Scale (DES; Bernstein
& Putnam, 1986 ), Briere's Dissociation Scale ( Briere & Runtz, 1988b ), and the
dissociation subscale from the TSC-33 and 40. This symptom indicates experiences such as
depersonalization, memory loss, and not feeling like oneself.
5. Eating disorders - based on the Bulimia Test (BULIT; Smith &
Thelen, 1984
), the Bulimia Diagnostic Instrument ( Nevo, 1985 ), the Eating Attitudes Test (EAT-26;
Garner, Olmsted, Bohr, & Garfinkel, 1982 ), the Eating Disorder Inventory (EDI;
Garner, Olmsted, & Polivy, 1983 ), and investigator-authored items.
6. Hostility - based on the Hostility subscale of the SCL-90-R and the BSI. This
symptom reflects thoughts, feelings, or actions that are characteristic of anger.
7. Interpersonal sensitivity - based on the Interpersonal Sensitivity subscale of
the SCL-90-R, HSCL, and BSI. This symptom reflects feelings of uneasiness and marked
discomfort when interacting with others, as well as feelings of personal inadequacy and
inferiority, especially compared with others.
8. Locus of control - based on the Locus of Control (LOC) scales by Nowicki and
Duke (1974) , Coleman et al. (1966) , and Rotter (1966) . This scale measures the extent
to which one feels in control of one's life.
9. Obsessive-compulsive symptomatology - based on the Obsessive-Compulsive
subscales of the SCL-90-R, HSCL, and BSI. This symptom is concerned with unremitting and
irresistible thoughts, impulses, and actions that are ego alien or unwanted.
10. Paranoia - based on the Paranoia subscales of the SCL-90-R,
HSCL, BSI, MCMI,
MMPI form R, HSF, and CAQ. This symptom reflects a disordered mode of thinking, consisting
of thoughts involving, for example, projection, hostility, suspiciousness, grandiosity,
and delusions.
11. Phobia - based on the Phobic Anxiety subscales of the SCL-90-R and BSI. This
symptom reflects a persistent fear response of an irrational and disproportionate nature
to a specific person, place, object, or situation.
12. Psychotic symptoms - based on the Psychoticism subscales of the BSI, SCL-90-R,
MCMI, MMPI (form R and HSF, Sc scale), CAQ, and Tennessee Self-Concept Scale (TSCS;
Fitts,
1964 ). For these measures, high scores indicate attributes such as mental confusion and
delusions (i.e., first-rank symptoms of schizophrenia such as hallucinations and
thought-broadcasting).
13. Self-esteem - based on the TSCS, Rosenberg Self-Esteem Scale ( Rosenberg, 1965
), Self-Ideal Discrepancy subscale of the Family Perception Grid ( Kelly, 1955 ), the Self
subscales of the McPearl Belief Scale ( McCann & Pearlman, 1990 ), subscales from the
Erwin Identity Scale ( Erwin & Delworth, 1980 ), and the Coopersmith Self-Esteem
Inventory ( Coopersmith, 1967 ).
14. Sexual adjustment - based on Finkelhor's Sexual Self-Esteem Scale (
Finkelhor,
1981 ), Reed's (1988) Romantic and Sexual Self-Esteem Survey, the Derogatis Sexual
Functioning Inventory (DSFI; Derogatis & Melisaratos, 1979 ), the Psychosexual
Functioning Questionnaire ( Schover, Friedman, Weiler, Heinman, &
LoPiccolo, 1982 ),
the Sexual Arousability Inventory ( Hoon, Hoon, & Wincze, 1976 ), subscales from the
TSC-33 and 40 and the Erwin Identity Scale, and investigator-authored items.
15. Social adjustment - based on the Social Support Questionnaire (
Sarason,
Levine, Basham, & Sarason, 1983 ); the Interpersonal Relationship Scale (
Schlein, Guerney, & Stover, 1971 ); the Inventory of Interpersonal Problems ( Horowitz,
Rosenberg, Baer, & Ureno, 1988 ); the Texas Social and Behavioral Inventory (
Helmreich & Stapp, 1974 ); the Social Adjustment Scale (SAS; Weissman &
Bothwell,
1976 ); Rathus' Assertiveness Schedule ( Rathus, 1973 ); Rotter's Interpersonal Trust
Scale ( Rotter, 1967 ); the Intimacy Attitude Scale ( Treadwell, 1981 ); the Intimacy
Behavior Scale ( Treadwell, 1981 ); subscales from the TSCS, McPearl Belief Scale, the
College Self-Expression Scale ( Galassi, DeLo, Galassi, & Bastien, 1974 ), the Student
Development Task and Lifestyle Inventory ( Winston, Miller, & Prince, 1987 ), and the
Miller Social Intimacy Scale (MSIS; Miller & Lefcourt, 1982 ); and
investigator-authored items.
16. Somatization - based on MacMillan's Health Opinion Survey (
MacMillan, 1957 );
subscales from the HSCL, TSC-33 and 40, BSI, SCL-90-R, MCMI, MMPI form R, HSF, and
CAQ;
and investigator-authored questions. This symptom reflects bodily related distress such as
headaches and pain; it also includes gastrointestinal, respiratory, and cardiovascular
complaints and complaints of sleeping problems.
17. Suicidal ideation and behavior - based on the Reasons for Living Inventory (
Linehan, Goodstein, Nielsen, & Chiles, 1983 ), the Suicide Behaviors Questionnaire (
Linehan & Nielsen, 1981 ), and investigator-authored items.
18. Wide adjustment - based on the General Well-Being Schedule ( McDowell &
Newell, 1987 ); total or global scores from the HSCL, TSC-33 and 40, SCL-90-R, and BSI;
subscales of the Comrey Personality Scales ( Comrey, 1970 ) and the TSCS;
investigator-created variables derived from combining scales of standard measures; and
investigator-authored items. This factor is a general measure of psychological adjustment
or symptomatology and, when derived by combining items or measures, is analogous to
Jumper's (1995) "psychological symptomatology" and Neumann et al.'s (1996)
"general symptomatology."
[Page 29]
The effect size used in this review was r , the Pearson correlation
coefficient. For CSA-psychological adjustment relations, positive r s indicated
poorer adjustment for CSA participants compared to control participants. For CSA-family
environment relations, positive r s indicated poorer family functioning for CSA
subjects. For family environment-adjustment relations, positive r s indicated that
poorer family functioning was associated with poorer adjustment. Pearson r s were
also computed to assess the magnitude of the relation between various moderating variables
(e.g., force) and outcome measures (i.e., psychological adjustment and self-reported
reactions). Positive r s indicated that higher levels of moderators were associated
with higher levels of symptoms or more negative reactions to the CSA. Finally, Pearson r
s were computed to assess the size of the differences in reactions and self-reported
effects between men and women who had CSA experiences. In this case, positive r s
indicated that men reported fewer negative reactions or effects than women, or conversely,
that they reported more positive reactions or effects than women.
Formulas for calculating r were taken from Rosenthal (1984,
1995) . A number of studies reported results separately for different types of CSA
participants (e.g., Collings, 1995 ; Roland, Zelhart, & Dubes, 1989 ; Sedney &
Brooks, 1984 ). To make the effect sizes in these cases comparable to those in the
majority of studies that compared participants with all types of CSA experiences with
controls, we combined all CSA subgroups in a given study into a single CSA group and then
compared this group with its control group (cf. Neumann et al., 1996 ). [*2]
[*2] Combination of CSA subgroups was achieved by computing a weighted mean, and by
computing the "true" variance of all CSA participants. The "true"
variance is the value that would have resulted from computing the variance of the scores
of all CSA participants irrespective of their subgrouping. This value was obtained by (a)
adding the sum of the squares of the CSA subgroups to get the within sum of squares for
these subgroups, (b) calculating the between-means sum of squares for the CSA subgroups,
(c) adding the within and between sum of squares to get the sum of squares total for the
subgroups, and (d) dividing the sum of squares total by the number of CSA scores minus 1.
Using the derived mean and variance, the CSA group was then compared with the control
group. This procedure produced results that were comparable to those of most other studies
that used one overall CSA group and was thus chosen over contrasting the means of the CSA
subgroups with the control mean.
Sample-level and symptom-level effect sizes across studies were
compared and combined meta-analytically using formulas taken from Rosenthal (1984) and
Shadish and Haddock (1994) . Combining effect sizes involved transforming r s into
Fisher Z s and then weighting the Fisher Z s by the degrees of freedom ( df
= N - 3) associated with their samples. The mean weighted Fisher Z was
transformed back to a mean weighted effect size, referred to as the unbiased effect
size estimate (r u). This metric was used to estimate the effect
size in the population and is considered to be unbiased because it weighs more heavily
larger samples whose effect sizes are generally considered to be more precise population
estimates ( Rosenthal, 1984 ; Shadish & Haddock, 1994 ). Statistical significance of
the effect size estimates was determined by computing their 95% confidence intervals; an
interval not including zero indicated an effect size estimate was significant ( Shadish
& Haddock, 1994 ).
To establish interrater reliability for coding, Bruce Rind
and Philip Tromovitch independently coded studies for psychological correlates, reactions,
self-reported effects, family environment-CSA relations, family environment-adjustment
relations, and results of statistical control. Interjudge agreement for these codings
ranged from 85% to 100%; all disagreements were resolved by discussion.
|