Journal of Consulting and Clinical Psychology Copyright 1992 by the American Psychological Association, Inc.

1992, Vol. 60. No. 2,196-203 0022-006X/92/S3.00

Methodological Issues in the Study of Sexual Abuse Effects

John Briere
Department of Psychiatry
University of Southern California School of Medicine

Despite its relative infancy, child abuse research has provided a substantial literature on the psycho-
logical sequelae of sexual molestation. These findings have been helpful in informing social policy
and guiding mental health practice. Because of the recency of interest in this area, however, as well
as the costs and time investment associated with more rigorous longitudinal research, many of
these studies have used correlational designs and retrospective reports of abuse. The implications
of this methodology are outlined, and remedies are suggested where possible.

Research on the psychological sequelae of childhood sexual pared with that of other subjects with no equivalent indications
abuse has grown rapidly in the last decade. This burgeoning of sexual abuse. Differences found between these groups, as-
literature reflects not only increased scientific interest in sexual suming equivalence on background variables, are attributed to
abuse and its potential effects but also growing public concern the sexual abuse experiences that discriminate them. When
about this form of child maltreatment. Given findings that per- such differences are found in adults, many years after their
haps one third of women and one sixth of men in our culture abuse, it is generally assumed that childhood sexual victimiza-
have experienced sexual contact with someone substantially tion has long-term effects on psychological adjustment.
older by their mid-teens (Finkelhor, 1979; Finkelhor, Hotaling, It is with regard to the generalization from sexual abuse corre-
Lewis, & Smith, 1989; Russell, 1986; Wyatt, 1986), research lates to conclusions about the long-term impacts of abuse that
linking such events to later psychological symptomatology or the sexual abuse literature is most vulnerable to inferential
distress would necessarily have substantial implications for so- error. The following sections of this article outline some of the
cial policy and clinical intervention. problems inherent in current sexual abuse research methodol-
ogy and offer suggestions for ways in which research in this area
Most recent studies do indicate that, in fact, childhood mo- might be improved. Although most of this discussion will focus
lestation is associated with multiple short- and long-term psy- on long-term correlates of childhood sexual abuse, many of the
chological difficulties. These correlations occur in nonclinical points raised here pertain to research on short-term sequelae as
as well as clinical samples and are present in both men and well.
women. Among the problems and symptoms that have been
associated repeatedly with a childhood sexual abuse history are Cross-Sectional Versus Longitudinal Designs
symptoms of posttraumatic stress, low self-esteem and guilt,
anxiety, depression, somatization, dissociation, interpersonal Subjects of sexual abuse research are often simultaneously
dysfunction, eating disorders, sexual problems, substance questioned about (a) abusive events that occurred many years
abuse, and suicidality (Briere, 1989; Browne & Finkelhor, ago and (b) their current level of psychological functioning.
1986). Childhood abuse reports are then designated as independent
variables, whereas subjects' responses on psychological mea-
Methodological Issues sures are considered dependent variables. By virtue of the
correlational and retrospective character of such designs, cause
Partly because of the recency of scientific interest in sexual and effect can become blurred. Although one might assume
abuse, research in this area has made use of relatively simple that the victimization experiences reported by subjects ante-
methodologies and designs. As noted elsewhere (Briere, 1988b), date their current psychological functioning, the reverse is pos-
such research may be considered first wave: primarily focused sible as well: Current distress or symptomatology may impact
on probing the link between childhood molestation and both on respondents' retrospective reports of abuse. As will be de-
proximal and more distal psychosocial difficulties. Typically, scribed later, such difficulties may arise from the effects of time
the psychological functioning of subjects who have been identi- on recollection, as well as the influence of current psychologi-
fied (or identify themselves) as sexual abuse victims is com- cal functioning on the accuracy of recall.
Cross-sectional research can be problematic when time-spe-
cific abuse sequelae are examined. Clinical experience and pre-
I wish to thank an anonymous reviewer for suggestions regarding
liminary research suggest, for example, that abuse-related
this article.
Correspondence concerning this article should be addressed to John symptomatology can wax and wane across the life span (Frie-
Briere, Department of Psychiatry, University of Southern California drich & Reams, 1987). This variability may reflect developmen-
School of Medicine, 1934 Hospital Place, Los Angeles, California tal issues: For example, certain intimacy and sexual problems
90033. may emerge as the child victim grows to be an adolescent and
196
 SPECIAL SECTION: METHODOLOGICAL ISSUES 197

confronts social and biological prodding to form sexual/ro- inherent in accurately matching nonabused to abused subjects
mantic connections with others (Briere, 1989; Gelinas, 1983; in studies using previously identified abused children (e.g., how
Maltz & Holman, 1987). Other sleeper effects first evident in to adequately match on presence of abuse risk factors [Finkel-
adolescence or adulthood may include aggression, substance hor, 1980] in children who were not, in fact, abused). Although
abuse, and proneness to revictimization. Assessment of these much could be written on the methodology of longitudinal
intermittent or stage-specific effects (Kagan's [1971] "hetero- child abuse research, the costs and difficulties inherent in fol-
typic continuity") often suffers from methodologies that rely on lowing subjects over extended periods of time have made such
one-shot adult retrospective reports. Such research requires the studies far less common in the child abuse literature. As a re-
adult subject not only to recall that a given symptom occurred sult, although in no way denigrating the greater value of longi-
at some point in the past but also to approximate when the tudinal methodologies, I have chosen to focus this article on
symptom in question began and ended. issues more common to the conduct of retrospective, correla-
Most important, cross-sectional research is rarely very help- tional abuse research.
ful in discriminating abuse-specific from abuse-concurrent or
abuse-antecedent events. In the relative absence of studies that
Report Biases
examine the psychological functioning of children before and
after sexual abuse has transpired (although see the at-risk stud- As indicated above, retrospective research on sexual abuse
ies of Egeland and colleagues [Egeland & Brunnquel, 1979; Ege- effects relies almost entirely on subjects' reports of past events.
land & Farber, 1984; Egeland, Sroufe, & Erickson, 1983]), re- \fet, recollection of abuse in childhood may be affected by what
searchers have yet to determine (a) the premolestation function- Cicchetti and Rizley (1981) refer to as "the influence of contem-
ing of sexually abused children, and thus the extent to which porary adaptation on recall" (p. 40). For example, the adult
abuse "effects" represent at least partially preexisting risk fac- who, by virtue of her need to avoid painful abuse memories, is
tors or psychological disturbance; (b) the exact role of coexist- amnestic for much or all of her childhood victimization (Briere
ing familial dysfunction and other forms of maltreatment; or (c) & Conte, in press; Herman & Schatzow, 1987) may truthfully
the impacts of social or demographic factors as they moderate report no knowledge of having been abused. If such individuals
or exacerbate what are thought to be simple abuse effects. are studied in sexual abuse research, their automatic inclusion
The most obvious resolution of the difficulties presented by in no abuse comparison groups might easily obscure or con-
retrospective, cross-sectional, and correlational designs is to found between-groups differences—especially if, as reported
avoid this form of data-gathering entirely. The ideal research by Briere and Conte (in press), amnestic abuse victims are more
plan for examining sexual abuse and its effects might involve a symptomatic than their nonamnestic but similarly abused co-
longitudinal study whereby (a) subjects and their families are horts. Although conscious suppression of abuse reports may be
randomly selected and evaluated before sexual abuse, so that partially ameliorated by experimenter assurances of confidenti-
baseline conditions and levels of psychological functioning can ality and attempts to provide an environment supportive of
be ascertained; (b) a detailed assessment of the type and extent disclosure, the problem of repressed memories remains a signif-
of sexual abuse is accomplished when it occurs in that unfortu- icant concern.
nate subsample; and (c) both abused and nonabused subjects are The passage of time may also mitigate against accurate or
repeatedly and regularly studied as they progress through the complete recall of childhood traumas (Menard, 1991). There
life span, with within- and between-groups analyses performed are little data available, in this regard, on the incremental im-
to study the form and development of abuse-specific psycholog- pacts of passing time on subjects' specific recall of discrete life
ical effects. Less ideally, children identified as recently sexually events. Similarly, age-specific socialization may influence sub-
abused could be studied with matched, nonabused children as jects' willingness to report sexual abuse. Individuals who grew
both groups matured (e.g., Cohen & Mannarino, 1988), or sub- up in an earlier era, for example, may be more prone to sup-
jects found to have been abused in the past (e.g., as reported in press or deny abuse experiences by virtue of the greater secrecy
clinic or social casework records) could be followed up in the and stigmatization associated with sexual victimization in pre-
present to determine their psychological status (e.g., McCord, vious decades. As a result, one cannot rule out the possibility
1983). that a main effect of abuse history on a given dependent vari-
Longitudinal analyses of sexual abuse effects can be subject able may more accurately reflect an abuse by time (or age) inter-
to difficulties, however, despite their improvement over the action on said measure. An example of the potential impacts of
one-shot retrospective study. Among these are (a) the method time on recollection may be the cohort effects reported by Rus-
whereby subjects are initially selected (e.g., the potential diffi- sell (1984). Russell found that older women report less molesta-
culties in generalizing from solely at-risk subjects), (b) changes tion than do younger women, leading her to suggest that the
in measurement relevance or sensitivity as a function of subject incidence of sexual abuse may be increasing over the years.
age (i.e., subjects who outgrow childhood-specific psychologi- Although this is not an unreasonable hypothesis, it is also possi-
cal tests), (c) the potential effects of repeated measurement ble that the greater passage of time between childhood and
(panel conditioning) on both abused and comparison subjects interview for older subjects resulted in less complete abuse
(e.g., how to control for—as opposed to merely documenting— memories for these subjects. Additionally, these older subjects
reactivity or monitoring effects if they occur), (d) subject attri- may view sexual abuse as more stigmatizing or embarrassing
tion over time, (e) the potential impacts of treatment in the than do younger subjects, and thus they may underreport mo-
abused group during follow-up (i.e., its impact on the natural lestation, a process that may result in older abuse victims being
course of postabuse symptomatology), and (f) the problems inappropriately placed in nonabused comparison groups.
198 JOHN BRIERE

Despite the importance of accurate disclosures in the con- test-retest reliability of sexual abuse reports and the role of
duct of retrospective studies, researchers in this area have found study-specific variables (e.g., method of interview, wording of
no satisfactory way to ensure the validity of subjects' recollec- abuse inquiry) as related to the frequency and extent of subject
tions of childhood sexual abuse (Briere, 1990; Briere & Zaidi, abuse reports.
1989). This problem has led some clinicians (and journal edi-
tors) to express concerns regarding "what is reported and what
in fact happened" to subjects studied in abuse research (Rich, Effects of Abuse Definition
1990, p. 1389). The question is often raised whether some sex-
ual abuse reports reflect, in fact, fantasies, delusions, or inten- As indicated by Peters, Wyatt, and Finkelhor (1986), sexual
tional misrepresentations for secondary gain. Because there are abuse researchers have used different definitions of what, at
almost no empirical data in this area, the possibility of abuse minimum, constitutes childhood sexual abuse. Russell (1984),
confabulation cannot be overlooked by researchers, even given for example, defines sexual abuse as any unwanted sexual expe-
the common clinical impression that such misrepresentation is riences before age 14, or attempted or completed rape by age 17,
rare (Briere, 1989). Among the ways that one might rule out or any attempted or completed sexual contact that occurred
abuse misrepresentation are (a) independently corroborating between relatives before the victim turned 18. Others restrict
abuse reports from other sources, (b) restricting study to abuse this category to actual sexual contact between someone under
cases that have been validated by the child protection or crimi- 15 years of age and another person 5 or more years older (e.g.,
nal justice systems, and (c) decreasing the potential rewards of Briere & Runtz, 1988). Obviously, such variability influences
falsifying one's childhood history. estimates of sexual abuse incidence or prevalence in any given
Independent corroboration of adults' retrospective abuse re- sample. Less appreciated, however, are the effects of different
ports is problematic in two respects. First, because sexual abuse abuse definitions on the identification of abuse-related psycho-
often occurs in secrecy and frequently results in embarrassment logical disturbance. It is not unreasonable to assume, for exam-
or shame (Finkelhor, 1979), many victims report that they did ple, that researchers who restrict themselves to earlier or more
not disclose their molestation to others at the time it occurred. intrusive forms of abuse might report more extreme outcomes
than those using broader definitions (Peters, 1988). Until re-
As a result, a significant proportion of subjects would be unable
searchers settle on a standard definition of what does and does
to provide corroborating information were it required. Second,
not constitute sexual abuse, findings regarding abuse correlates
even if corroboration were possible, the intrusion and possible
must be evaluated in terms of the specific definition being
distress to subjects inherent in investigating the veracity of their
used.
reports (i.e., contacting and questioning relatives, neighbors, or
friends) seemingly precludes its widespread use (Briere, 1990).
Because the majority of sexual abuse cases are never brought Nonequivalent Comparison Groups
to the attention of authorities, data collection limited to sub-
jects whose cases had been validated (founded) by child pro- Most research on the long-term sequelae of sexual abuse uses
tection or criminal justice agencies would significantly reduce designs best considered preexperimental or correlational
the number of subjects available for study. Further, given that (Campbell & Stanley, 1966; Cook & Campbell, 1979). This is
founded cases often differ from others in terms of, for example, because researchers in this area are forced to begin with a sub-
severity of abuse and the social class of the abused child, find- ject variable rather than a manipulated one. Because sexual
ings generated from such unusual samples are likely to have abuse is not a randomly assigned condition, it is impossible to
limited external validity. ensure that abused and nonabused subjects are equivalent in all
Fortunately, concerns about secondary gain for falsely re- other respects and thus that differences found between these
porting sexual abuse are to some extent addressed by the typi- groups are due to abuse, per se. For example, if two groups vary
cal abuse research paradigm, which does not reward or punish not only according to sexual abuse history but also as a function
response to sexual abuse items, per se, and often does not re- of age, social class, or family dysfunction, symptom differences
cord the subject's name or other identifying data. Under such between these groups cannot be linked solely to abuse status.
circumstances, the motive for deliberate falsification of abuse Because sexual abuse is often correlated with other variables,
history is less clear. Researchers should avoid, of course, para- special attention must be paid to the comparison group's status
digms that could reinforce abuse reports, such as offering on these factors. Although matching on relevant variables often
money for participation in research solely to those who report appears to be an attractive option, it is not always clear on which
childhood sexual abuse. variables subjects should be matched. Among those variables
Given the above, it appears that the accuracy of sexual abuse seemingly relevant to child abuse are demographic characteris-
reports cannot be assured, in terms of ruling out either false tics such as age, race, sex, education, and socioeconomic status
positives or false negatives. As a result, retrospective studies in (SES), along with clinical status, various measures of family
this area cannot entirely guarantee the complete validity of functioning, and history of childhood traumas and stressors
their criterion variable. Future studies might include additional other than sexual abuse. Because child abuse is a complex vari-
variables relevant to report bias, such as social desirability, ten- able with complex antecedents, however, even this level of con-
dency toward repression, and attitudes toward abuse disclo- trol is likely to be insufficient. For example, the experimenter
sure, and researchers might especially endeavor to eliminate whose abuse group is sampled from university students and
any conceivable study-specific rewards for reporting sexual whose comparison group (albeit carefully matched on each of
abuse. Furthermore, research may be indicated to probe the the above variables) is drawn from the general population risks
 SPECIAL SECTION: METHODOLOGICAL ISSUES 199

the possibility that unmonitored—but nevertheless important— toms, the potential contributions of injurious events copresent
variables other than abuse still discriminate these groups. with or antecedent to sexual abuse, and the absence of data on
Finally, because matching subjects usually means differen- children's psychological functioning before molestation.
tially discarding other subjects, this procedure can reduce the Although clinical experience and the characteristics of some
generalizability of obtained between-groups differences, espe- postabuse sequelae (e.g., flashbacks to specific abusive events in
cially if the matching variables include risk factors for abuse. childhood) suggest that many of the difficulties associated with
Individuals who match an abused sample on negative family a child abuse history are in fact abuse effects, it is erroneous to
environment and economic distress, for example, may best rep- make such conclusions solely on the basis of the available re-
resent an atypical population: people who were at risk to be search. The role of science in such endeavors is to as precisely as
abused but, for whatever reasons, were not victimized. This possible determine (a) the exact relationships between child-
problem also may occur when different groups of abused sub- hood sexual abuse and various types of psychological dysfunc-
jects (e.g., sexually abused, physically abused, and psychologi- tion; (b) which "effects" are likely to be epiphenomena of sexual
cally neglected) are matched and compared because the resul- abuse (i.e., arising from third variables, including other types of
tant groups may no longer represent the (unmatched) popula- maltreatment); and (c) how preabuse functioning, family-of-ori-
tions from which they were drawn. gin dynamics, and other variables moderate these relation-
Rather than matching to approximate equivalence, it is ships.
usually best to draw representative abused and nonabused sub- Because correlational designs are, in and of themselves, in-
jects from the same population, whether it be university stu- sufficient to imply causation, the abuse researcher must use
dents, clients from a mental health center, or a random sample strategies that lessen the number of possible alternative explana-
of women from a specific geographic area. If differences on tions for their findings. This is, of course, a common goal of
relevant variables emerge, the results should be qualified ac- research design. Kazdin (1980) notes that
cordingly in the discussion section. Ultimately, although the
researcher may wish to match on any number of subject dimen- Experimental results by their very nature are ambiguous because
sions, Miller's (1987) conclusion remains true: "Matching can- any particular finding may depend upon one of the many unique
features of the setting and experimental arrangements where the
not be used to make two groups equivalent when the groups are finding was obtained. Because ambiguity can never be elimi-
not in fact equivalent" (p. 50). nated, minimizing ambiguity is the primary task of research de-
sign, (p. 4)
Other Forms of Abuse
Ambiguity regarding abuse sequelae is probably best ad-
Since the beginning of research in this area, there has been a dressed by procedures that rule out competing explanations for
tendency for investigators to examine sexual abuse in a relative symptomatology. Alternative hypotheses, for example, are that
vacuum. As noted more generally by Rosenberg (1987), abuse abuse effects are at least in part the results of dysfunctional
researchers have tended to overlook other forms of maltreat- family dynamics that support sexual abuse and produce psycho-
ment that a child might experience in a given family, despite the logical disturbance (Fromuth, 1986) and that concomitant phys-
fact that physical, psychological, and sexual child abuse fre- ical or psychological abuse may account for some of the diffi-
quently occur together. Recent research by Briere and Runtz culties otherwise attributed to sexual abuse (Briere & Runtz,
(1990), for example, points to the confusion that may accrue 1990). Attention to these possibilities suggests what are likely to
from examining one form of abuse alone. In this study various be minimal criteria for modern abuse sequelae research: that
forms of child maltreatment were found to covary, and each the design control for extraneous variables by both (a) sampling
type of abuse was correlated with a number of later psychoso- from a single population (e.g., inner city adolescents or patients
cial difficulties. When all other forms of abuse were controlled from a dissociative disorders clinic) and (b) formally including
by canonical correlation analysis, however, each type of abuse relevant factors as additional independent variables or predic-
was found to have a considerably smaller number of unique tors so that their interactions with sexual abuse can be tested
psychological correlates. directly.
The copresence of various child abuse types and sequelae Alternatively, variables relevant to competing etiologic sce-
suggest that, at minimum, researchers should examine not only narios may be examined through structural equation modeling
the main effect of sexual abuse on psychological functioning (see, for example, Wyatt & Newcomb, 1990). This approach
but also its interaction with physical and psychological mal- allows one to examine specific hypotheses (models) regarding
treatment factors. Other designs might consider all three forms the causal links among such variables and permits consider-
of abuse simultaneously, using multivariate procedures such as ation of phenomena that may mediate between cause and ef-
multiple regression or canonical correlation. fect. Such analyses are only valid, however, if the sample size is
adequate, the variables in question have been well operational-
Conclusions Regarding Causality ized and reliably measured, and the proposed model is based
on adequate theory (MacDonald, 1977). Even this more sophis-
As noted at several points in this article, reliance on retrospec- ticated analytic approach, however, cannot prove a specific
tive, nonlongitudinal designs generally precludes definitive in- causal hypothesis; it can only provide information on how well
ferences regarding abuse effects. Among the impediments to a given model fits the data at hand. Because of the error that
such conclusions are the potentially confounding effects of si- almost inevitably creeps into the estimation of linear models,
multaneously determining previous abuse and current symp- however, this goodness-of-fit criterion may be misleading in
200 JOHN BRIERE

some instances (Anderson & Shanteau, 1977). Ultimately, the mental health outpatients, university students, professionals,
various weaknesses inherent in correlational designs demand and members of the general population. Although most studies
that abuse researchers avoid unequivocally attributing causality have restricted themselves to female subjects, recent investiga-
to findings not derived from rigorous longitudinal method- tions have increasingly included male subjects (e.g., Briere,
ology. Evans, Runtz, & Wall, 1988; Urquiza & Crowley, 1986). Simi-
larly, although most long-term sequelae studies have been of
Measurement Issues Caucasian subjects, more attempts are being made to sample
from other racial and ethnic groups (e.g., Wyatt, 1986). Socioeco-
Research on abuse sequelae has been compromised, to some nomic status often varies with subject type; for example, univer-
extent, by the use of questionable measurement systems. Fre- sity students are typically from the middle class, whereas some
quently, investigators use either home-spun measures of un- patient groups and general population samples have lower SES.
known reliability and validity or generic measures that may be This heterogeneity is a positive development in terms of our
insensitive to abuse-specific symptomatology. As a result, find- understanding of abuse in different social contexts. It is more
ings in this area can be difficult to interpret. problematic, however, when researchers seek to generalize
To the extent that adequate reliability and validity is estab- from a given sample to the universe of sexual abuse victims.
lished, the use of study-specific measures is not inherently prob- Not only is generalization hampered by the varying clinical
lematic. Unfortunately, a number of studies have foregone psy- status, occupations, races, and SES of these samples but the
chometric evaluation of their instruments, leaving the con- type and extent of sexual abuse reported by subjects often
sumer to trust that the measure is stable and that it taps the differs significantly from group to group. For example, clinical
construct intended by the researchers. Without such data it is subjects typically report more frequent molestation by more
not clear whether negative findings reflect an absence of be- perpetrators, a longer abuse duration, a greater likelihood of
tween-groups differences or the impacts of unreliability on va- intercourse, and more symptomatology than do nonclinical
lidity. Positive findings, on the other hand, are interpretable subjects (Elliott & Briere, 1991). As a result, findings derived
only to the extent that the measure has construct validity. For from clinical groups may not generalize well to general popula-
these reasons, study-specific instruments should be developed tion samples or to other individuals with less severe abuse histo-
according to accepted psychometric principles (e.g., Anastasi, ries.
1988; DeVellis, 1991) and accompanied by data on reliability Given differences in subject characteristics and abuse sever-
and validity. ity, blanket statements about sexual abuse or its mental health
A second problem for abuse researchers is the generality of sequelae should be carefully avoided when they are based on a
many measures of psychological dysfunction. Because most single study. In fact, short of large, general population studies of
available instruments were developed without reference to sexual abuse and its correlates (i.e., Wyatt & Newcomb, 1990),
abuse or trauma, they may be less sensitive to abuse-specific conclusions drawn from any given child abuse study should be
symptoms (Elliott & Briere, 1991). An example of the promise limited to careful inferences regarding individuals with similar
of abuse specificity is the findings of Briere and Runtz (1990), demographics, social status, and abuse histories.
who found that although a standard measure of low self-esteem An important caveat to the above-noted concerns is that of
was unrelated to child abuse history, a newly created measure study replication. Although the results of a single investigation
incorporating self-denigrating statements often made by may have limited generalizability, findings that are stable
former abuse victims was significantly associated with child- across multiple studies (and therefore multiple samples) have
hood maltreatment. Similarly, Bagley's (in press) community considerably greater applicability to the universe of abuse vic-
study of 345 Canadian women revealed that the Trauma Symp- tims. For example, the association between childhood sexual
tom Checklist (Briere & Runtz, 1989)—a scale developed to abuse and depression found in a number of clinical and non-
specifically tap abuse-related symptomatology—was more ef- clinical studies (Briere & Runtz, 1991; Browne & Finkelhor,
fective than traditional measures such as the Middlesex Hospi- 1986) suggests that this symptom is a common sequel of sexual
tal Questionnaire, the Center for Epidemiological Studies in abuse, to some extent irrespective of subject type. Consistent
Depression (CESD) scale, or the Coopersmith self-esteem inven- replication of findings at least partially addresses those threats
tory in identifying adults who were sexually abused as children. to internal and external validity present in many sexual abuse
The importance of abuse-relevant measures resides not in studies. Meta-analysis, defined as "the statistical analysis of a
their potential ability to identify abuse victims, however, as large collection o f . . . individual studies for the purpose of
much as in the development of an accurate database regarding integrating the findings" (Glass, 1976, p. 6), may be especially
abuse-specific symptom patterns. By identifying precisely how helpful in summarizing the findings of replicated abuse studies
former abuse victims differ from nonvictims, for example, such and, in some instances, uncovering relationships between
research can help clinicians to more accurately diagnose post- abuse and outcome variables not readily apparent in any given
abuse disturbance, as opposed to perhaps missing victims' dis- single study (Rosenthal, 1991).
tress on generic measures of psychological functioning.
Statistical Issues
Constraints on Generalization In addition to the methodological concerns outlined above,
Researchers of sexual abuse sequelae have sampled from a there are several statistical issues that may arise in sexual abuse
wide variety of subject groups, including psychiatric inpatients, research.
 SPECIAL SECTION: METHODOLOGICAL ISSUES 201

Insufficient statistical power. Studies with inadequate sam- provide statistical equivalence between nonequivalent groups,
ple sizes, unreliable or insensitive measures, or poorly chosen the use of semipartial analysis to establish the relative impor-
statistical analyses decrease the investigator's ability to detect tance or etiologic significance of two or more variables (Pedha-
relationships when they are, in fact, present in the population at zur, 1982; Stevens, 1986), and the use of any partialing tech-
hand (Cohen, 1988; DeVellis, 1991). Reduced statistical power nique when certain underlying assumptions or implicit require-
is often a problem in studies of abuse sequelae, resulting in an ments are not met (Briere, 1988a). Regarding the latter, partial
underestimation of the differences between abused and non- or semipartial analysis may be misleading if the following are
abused subjects. For this reason, it is suggested that (a) sample true:
and cell sizes in any given study minimally attain the magni- 1. There is substantial multicolinearity between control and
tude suggested by the appropriate power analysis (Cohen, abuse variables. Pedhazur (1982) notes that in the extreme case
1988); (b) dependent measures be reasonably reliable and rele- of this problem, "partialling out from one predictor another
vant to abuse sequelae; and (c) statistical tests be appropriate for predictor from which it is highly correlated will generally result
the data, in terms of their sensitivity, underlying assumptions, in a small, even meaningless semi-partial correlation" (p. 167).
and use of the highest level of measurement possible. Further- 2. The control variable is unreliable. As noted by Cohen and
more, researchers who fail to reject the null hypothesis should Cohen (1983), measurement error in the control variable "may
consider reporting the results of relevant power analyses, so that decrease or increase or even change the sign of, a partial rela-
the potential contribution of methodological issues (e.g., low jV) tionship" (p. 407).
to nonsignificant results can be evaluated. 3. The causal or directional relationship between the control
Univariate versus multivariate analyses. Assuming that rea- variable(s) and the abuse variable is unknown. Various writers
sonable sample sizes can be attained, the researcher confronted (e.g., Gordon, 1968; Pedhazur, 1982) have noted that it is inap-
with multiple, related measures of psychological functioning propriate to control for X, while examining the role of X2 on Y
may find multivariate statistical approaches to be advan- if there is a possibility that X2 caused X, (e.g., that sexual abuse
tageous over univariate ones. Such procedures, which include produces family dysfunction as well as symptoms), or that Xj
multivariate analysis of variance, canonical correlation, dis- and X2 interact synergistically, or that X! and X 2 are different
criminant analysis, and multiple regression, can be helpful in measures of the same construct. The former point may be espe-
several ways: They control the experimentwise error rate rela- cially relevant to abuse effects research, where ongoing intrafa-
tive to an unprotected series of univariate analyses; they take milial abuse undoubtedly contributes to a negative family envi-
into account the relationship between dependent variables ronment and where both may have synergistic impacts on later
rather than erroneously treating related dependent variables as psychological adjustment.
independent events; and, in some instances, they may be more At minimum, the algebra of the semipartial correlation is
able to detect between-groups differences than univariate tests. likely to produce a conservative test when the control variable is
Although multivariate analysis can be well suited to sexual not causally antecedent to the abuse. In order to be significant,
abuse research, it can be misused as well. For example, multi- abuse must correlate with symptoms after all variance shared
variate tests are sometimes used in the context of insufficient with the control variables(s) has been removed. Such analyses
sample sizes or inadequate subjects-to-variables ratios, thereby tend to give more of the credit to control variables than to abuse
potentially capitalizing on error variance and producing mis- variables, thereby encouraging an underestimation of the po-
leading results (Tabachnick & Fidell, 1989). Multivariate analy- tential impacts of sexual abuse.
sis also can be inappropriate when encompassing a collection Because covariate, semipartial, or other control variable anal-
of unrelated measures (e.g., subject age, score on a depression yses are inappropriate solutions to nonequivalent groups, and
inventory, and responses on an analogue parenting task), as given the complexity of childhood sexual abuse and its associa-
opposed to a set of variables whose interrelation is likely to be tion with other variables, partialing procedures should be used
both significant and of interest, such as the subscales of the with caution in such research. Among other constraints, par-
Minnesota Multiphasic Personality Inventory (Hathaway & tialized results should be interpreted as such (i.e., solely as "the
McKinley, 1982) or scores on a neuropsychological test battery effect of a variable(s) after having controlled for another vari-
(Applebaum & McCall, 1983). able®" (Pedhazur, 1982, p. 178).
Statistical control issues. Because sexual abuse may be asso-
ciated with familial dysfunction, other forms of maltreatment, Conclusion
or other variables thought to be risk factors for molestation,
researchers have attempted to control for these factors either by This article has outlined a number of ways in which the meth-
matching (as noted previously) or through statistical methods. odology of sexual abuse research might be improved. It should
Of the latter, the most common are the partialing procedures, not be inferred from this critique, however, that the extant litera-
statistical techniques that allow exploration of the relationship ture in this area is fatally flawed. The many studies available on
between child abuse and adult psychological functioning after abuse sequelae have been of considerable assistance to clini-
the variance shared by child abuse and other variables has been cians, legislators, and social policy planners—individuals who
removed (controlled for or partialed out).1
Although these methodologies have many valid applications, 1
Although the present discussion concerns the use of semipartial
there are a number of conditions when partialing procedures correlations, similar issues may be raised regarding the use of related
can lead to excessively conservative or even erroneous conclu- procedures such as analysis of covariance, partial correlation analysis,
sions (Briere, 1988a). These include the use of covariates to or stepwise multiple regression analysis.
202 JOHN BRIERE

could not reasonably wait for the results of rigorous longitu- Cicchetti, D, & Rizley, R. (1981). Developmental perspectives on the
dinal research before intervening in this major social problem. etiology, intergenerational transmission, and sequelae of child mal-
In fact, the first wave of abuse research has largely succeeded in treatment. In R. Rizley & D. Cicchetti (Eds.), Developmental per-
terms of increasing social, clinical, and scientific awareness of spectives on child maltreatment. San Francisco, CA: Jossey-Bass.
sexual abuse and its potential impacts. Cohen, J. (1988). Statistical power analysis for the behavioral sciences
Given this success, it is time for the second wave: the develop- (2d ed.). New York: Academic Press.
ment of more tightly controlled and methodologically sophisti- Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation
analysis for the behavioral sciences (2d ed.). Hillsdale, NJ: Erlbaum.
cated studies that seek to disentangle the antecedents, corre- Cohen, J. A., & Mannarino, A. P. (1988). Psychological symptoms in
lates, and impacts of sexual abuse. Although such studies are sexually abused girls. Child Abuse & Neglect, 12, 433-442.
more likely to be longitudinal in nature, the cross-sectional Cook, T. D, & Campbell, D. T. (1979). Quasi-Experimentation. Boston:
study is far from dead. Instead, future work in this area is likely Houghton Mifflin.
to be characterized by greater attention to design sensitivity, DeVellis, R. F. (1991). Scale development: Theory and applications.
greater control over extraneous variables, and careful infer- Newbury Park, CA: Sage.
ences about causality. Egeland, B., & Brunnquel, D. (1979). An at-risk approach to the study
of child abuse: Some preliminary findings. Journal of American
Academy of Child Psychiatry, 18, 219-235.
References Egeland, B., & Farber, E. A. (1984). Infant-mother attachment: Factors
related to its development and change over time. Child Development,
Anastasi, A. (1988). Psychological testing (6th ed.). New \brk: Macmil-
55,753-771.
lan.
Egeland, B., Sroufe, L. A., & Erickson, M. F. (1983). Developmental
Anderson, N. H., & Shanteau, J. (1977). Weak inference with linear
models. Psychological Bulletin, 84,1155-1170. consequences of different patterns of maltreatment. Child Abuse &
Neglect, 7, 459-469.
Applebaum, M. I, & McCall, R. B. (1983). Design and analysis in
developmental psychology. In P. H. Mussen (Ed.), Handbook of child Elliott, D. M., & Briere, J. (1991). Studying the long-term effects of
psychology: Vol. 1. History, theory, and methods (4th ed., pp. 415- sexual abuse: The Trauma Symptom Checklist (TSC) scales. In
476). New York: Wiley. A. W Burgess (Ed.), Rape and sexual assault (Vol. 3). New \brk:
Bagley, C. (in press). The prevalence and mental health sequels of child Garland.
sexual abuse in a community sample of women aged 18 to 27. Cana- Finkelhor, D. (1979). Sexually victimized children. New York: Free
Press.
dian Journal of Mental Health.
Briere, J. (1988a). Controlling for family variables in abuse effects re- Finkelhor, D. (1980). Risk factors in the sexual victimization of chil-
search: A critique of the "partialling" approach. Journal of Interper- dren. Child Abuse & Neglect, 4, 265-273.
sonal Violence, 3, 80-89. Finkelhor, D, Hotaling, G., Lewis, I. A., & Smith, C. (1989). Sexual
Briere, J. (1988b). The long-term clinical correlates of childhood sexual abuse and its relationship to later sexual satisfaction, marital status,
victimization. Annals of the NewYork Academy of Sciences, 528,327- religion, and attitudes. Journal of Interpersonal Violence, 4, 279-299.
335. Friedrich, W N., & Reams, R. A. (1987). The course of psychological
Briere, J. (1989). Therapy for adults molested as children: Beyond sur- symptoms in sexually abused young children. Psychotherapy:
vival. New \brk: Springer. Theory, Research, and Practice, 24,160-170.
Briere, J. (1990). Accuracy of adults' reports of abuse in childhood: Dr. Fromuth, M. E. (1986). The relationship of childhood sexual abuse
Briere replies (invited letter). American Journal of Psychiatry, 147, with later psychological and sexual adjustment in a sample of college
1389-1390. women. Child Abuse & Neglect, 10, 5-15.
Briere, J., & Conte, J. (in press). Self-reported amnesia for abuse in Gelinas, D. (1983). The persisting negative effects of incest. Psychiatry,
adults molested as children. Journal of Traumatic Stress. 46,312-332.
Briere, I, Evans, D, Runtz, M., & Wall, T. (1988). Symptomatology in Glass, G. (1976). Primary, secondary, and meta-analysis of research.
men who were molested as children: A comparison study. American Educational Researcher, 5, 3-8.
Journal of Orthopsychiatry, 58, 457-461. Gordon, R. A. (1968). Issues in multiple regression. American Journal
Briere, J., & Runtz, M. (1991). Sexual abuse trauma in adults: A review of Sociology, 73, 592-616.
and integration. In J. Briere (Ed.), Treating victims of child sexual Hathaway, S. R., & McKinley, J. C. (1982). Manual for the Minnesota
abuse (pp. 3-13). San Francisco, CA: Jossey-Bass. Multiphasic Personality Inventory. Minneapolis: University of Min-
Briere, J., & Runtz, M. (1988). Symptomatology associated with child- nesota Press.
hood sexual victimization in a non-clinical adult sample. Child Herman, J. L., & Schatzow, E. (1987). Recovery and verification of
Abuse & Neglect, 12, 51-59. memories of childhood sexual trauma. Psychoanalytic Psychology,
Briere, J., & Runtz, M. (1989). The Trauma Symptom Checklist (TSC- 4,1-14.
33): Early data on a new scale. Journal of Interpersonal Violence, 4, Kagan, J. (1971). Change and continuity in infancy. New \brk: Wiley.
151-163. Kazdin, A. E. (1980). Research design in clinical psychology. New York:
Briere, J., & Runtz, M. (1990). Differential adult symptomatology asso- Harper & Row.
ciated with three types of child abuse histories. Child Abuse & Ne- MacDonald, K. I. (1977). Path analysis. In C. A. O'Muircheartaigh &
glect, 14, 357-364. C. Payne (Eds.), The analysis of survey data (Vol. 2). New York: Wiley.
Briere, J., & Zaidi, L. Y. (1989). Sexual abuse histories and sequelae in Maltz, W, & Holman, B. (1987). Incest and sexuality: A guide to under-
female psychiatric emergency room patients. American Journal of standing and healing. Lexington, MA: Lexington Books.
Psychiatry, 146, 1602-1606. McCord, J. (1983). A forty year perspective on effects of child abuse
Browne, A., & Finkelhor, D. (1986). Impact of child sexual abuse: A and neglect. Child Abuse & Neglect, 7, 265-270.
review of the research. Psychological Bulletin, 99, 66-77. Menard, S. (1991). Longitudinal research. Sage University Paper Series
Campbell, D. T., & Stanley, J. C. (1966). Experimental and quasi-experi- on Quantitative Applications in the Social Sciences, 07-076. New-
mental designs for research. Chicago: Rand McNally. bury Park, CA: Sage.
 SPECIAL SECTION: METHODOLOGICAL ISSUES 203

Miller, S. A. (1987). Developmental research methods. Englewood Russell, D. E. H. (1986). The secret trauma: Incest in the lives of girls and
Cliffs, NJ: Prentice-Hall. women. New York: Basic Books.
Pedhazur, E. J. (1982). Multiple regression in behavioral research (2nd Stevens, J. (1986). Applied multivariate statistics for the social sciences.
ed.). New York: Holt, Rinehart & Winston. Hillsdale, NJ: Erlbaum.
Peters, S. D. (1988). Child sexual abuse and later psychological prob- Tabachnick, B. G., & Fidell, L. S. (1989). Using multivariate statistics.
lems. In G. E. Wyatt & G. J. Powell (Eds.), Lasting effects of child New York: Harper & Row.
sexual abuse (pp. 101-117). Newbury Park, CA: Sage. Urquiza, A. J., & Crowley, C. (1986, May). Sex differences in the survi-
Peters, S. D., Wyatt, G. E., & Finkelhor, D. (1986). Prevalence. In D. vors of childhood sexual abuse. Paper presented at the Fourth Na-
Finkelhor, S. Araji, L. Baron, A. Browne, S. D. Peters, & G. E. Wyatt, tional Conference on the Sexual Victimization of Children, New
A sourcebook on child sexual abuse (pp. 15-59). Newbury Park, CA: Orleans, LA.
Sage. Wyatt, G. E. (1986). The sexual abuse of Afro-American and White
Rich, C. L. (1990). Accuracy of adults' reports of abuse in childhood American women in childhood. Child Abuse & Neglect, 9,507-519.
(letter). American Journal of Psychiatry, 147,1388. Wyatt, G. E., & Newcomb, M. (1990). Internal and external mediators
Rosenberg, M. S. (1987). New directions for research on the psychologi- of women's sexual abuse in childhood. Journal of Consulting and
Clinical Psychology, 58, 758-767.
cal maltreatment of children. American Psychologist, 42,166-171.
Rosenthal, R. (1991). Mela-analytic procedures for social research.
Newbury Park, CA: Sage. Received August 5,1991
Russell, D. E. H. (1984). Sexual exploitation: Rape, child sexual abuse, Revision received October 3,1991
and workplace harassment. Newbury Park, CA: Sage. Accepted October 14,1991 •

MEMBERS OF UNDERREPRESENTED GROUPS:

REVIEWERS FOR JOURNAL MANUSCRIPTS WANTED
If you are interested in reviewing manuscripts for APA journals, the APA Publications and Commu-
nications Board would like to invite your participation. Manuscript reviewers are vital to the pub-
lication process. As a reviewer, you would gain valuable experience in publishing. The P&C Board
is particularly interested in encouraging members of underrepresented groups to participate more in
this process.
If you are interested in reviewing manuscripts, please write to Leslie Cameron at the address below.
Please note the following important points:
• To be selected as a reviewer, you must have published articles in peer-reviewed journals. The
experience of publishing provides a reviewer with the basis for preparing a thorough, objective
review.
» To select the appropriate reviewers for each manuscript, the editor needs detailed information.
Please include with your letter your vita. In your letter, please identify which APA journal you
are interested in and describe your area of expertise. Be as specific as possible. For example,
"socialpsychology" is not sufficient—you would need to specify "social cognition" or "attitude
change" as well.
• Reviewing a manuscript takes time. If you are selected to review a manuscript, be prepared to
invest the necessary time to evaluate the manuscript thoroughly.
Write to Leslie Cameron, Journals Office, American Psychological Association, 750 First Street,
NE. Washington, DC 20002-4242.