Professional Documents
Culture Documents
Background: Influential studies have cast doubt on the validity of retrospective reports by adults of
their own adverse experiences in childhood. Accordingly, many researchers view retrospective reports
with scepticism. Method: A computer-based search, supplemented by hand searches, was used to
identify studies reported between 1980 and 2001 in which there was a quantified assessment of the
validity of retrospective recall of sexual abuse, physical abuse, physical/emotional neglect or family
discord, using samples of at least 40. Validity was assessed by means of comparisons with contemporaneous, prospectively obtained, court or clinic or research records; by agreement between retrospective reports of two siblings; and by the examination of possible bias with respect to differences
between retrospective and prospective reports in their correlates and consequences. Medium- to longterm reliability of retrospective recall was determined from studies in which the testretest period
extended over at least 6 months. Results: Retrospective reports in adulthood of major adverse
experiences in childhood, even when these are of a kind that allow reasonable operationalisation, involve a substantial rate of false negatives, and substantial measurement error. On the other hand,
although less easily quantified, false positive reports are probably rare. Several studies have shown
some bias in retrospective reports. However, such bias is not sufficiently great to invalidate retrospective
case-control studies of major adversities of an easily defined kind. Nevertheless, the findings suggest
that little weight can be placed on the retrospective reports of details of early experiences or on reports of
experiences that rely heavily on judgement or interpretation. Conclusion: Retrospective studies have a
worthwhile place in research, but further research is needed to examine possible biases in reporting. Keywords: Retrospective recall, prospective measures, validity, reliability, sexual abuse, physical
abuse, neglect, family discord, reporting bias.
The great majority of studies of the long-term psychopathological consequence of childhood experiences have used some type of case-control design in
which data on experiences have been obtained
through interviews in adult life that have relied on
retrospective recall. Accordingly, it is necessary to
know the extent to which such recall is likely to be
valid and, in particular, the extent to which it is free
from biases that might be relevant to the causal
hypothesis being tested. In this paper we address
that question by considering the findings from studies since 1980 with data that provided some form of
corroboration through contemporaneous research
measures or official records or through comparisons
with other informants reporting on the same experiences over the same time period. The background to
the issue is that early studies were largely negative in
their conclusions, with findings suggesting that retrospective reports were not to be relied on (e.g., Yarrow, Campbell, & Burton, 1970). In addition, in
recent years there has been an increasing call for
longitudinal studies on the assumption that they are
generally preferable to cross-sectional enquiries
using retrospective recall (Rutter, 1988; Wierson &
Forehand, 1994). Before presenting the empirical
findings on retrospective recall, it is necessary to
consider how far that assumption is correct. In
addition, we note some of the key methodological
issues that apply to the use of corroborative data.
261
tunity of separating person effects on the environment from environmental effects on the person (Bell,
1968; Bell & Chapman, 1986); and 5) the opportunity to assess attrition bias in a rigorous fashion.
The main limitations of longitudinal data comprise
five rather separate problems. First, some hidden
experiences are not likely to be reported contemporaneously in childhood. Most obviously this applies to sexual abuse within the family. Second,
when studying sequelae in adult life it is inevitable
that there has to be reliance on outdated measures
for the assessment of childhood experiences. That
may be a major problem when the concepts of what
constitute key experiences have changed over time.
Thus, few studies 20 or 30 years ago assessed differences among siblings in their experiences or
relationships in the family, and few used measures
of attachment that would be regarded as satisfactory
today. Third, although some long-term longitudinal
studies have been remarkably successful in maintaining the cooperation of samples (see, e.g., Cairns,
Elder, & Costello, 1996; Fergusson, Horwood, &
Nagin, 2000a; Moffitt, Caspi, Rutter, & Silva, 2001),
attrition is often a problem, if only because individuals move abroad or to a distant area in the
same county and cannot be traced. Fourth, longitudinal data have to rely on retrospective recall for
measures of experiences since the last interview,
which will often involve reporting over a period of
several years. Finally, longitudinal data are very expensive to collect, the costs being very much greater
than for cross-sectional studies using samples of the
same size.
Measurement issues
The assessment of retrospective recall is necessarily
reliant on the quality of the measures of experiences
at both time points and on the comparability of the
measurement at the time in childhood and as retrospectively reported in adult life. Many studies fall
down on this criterion because the measures in
childhood were rather vague and global. Thus, Yarrow et al. (1970) relied on nursery school records
deriving from unstandardised parental reports.
Four main ways to check the validity of retrospective reports have been used in this paper: i) Direct comparisons of retrospective reports with prior
official records (e.g., from schools, hospitals or
courts); ii) similar comparisons with prior parental
reports; iii) agreement between the retrospective reports of two siblings; and iv) indirect comparison in
terms of the differences between retrospective and
prospective reports in relation to risk correlates and
outcome.
A particular problem with respect to comparability
arises when the measures at the childhood time
period derive from the parent and those obtained
retrospectively in adulthood from the subject (i.e.,
the person who was a child at the first time period).
262
Aims of review
The purpose of this paper was to review the empirical
research findings on the validity of retrospective recall with respect to four main domains of important
risk experiences in childhood: 1) sexual abuse; 2)
physical abuse or severe physical punishment; 3)
physical or emotional neglect; and 4) chronic family
conflict or disharmony. Various other domains of
potential interest (e.g., verbal abuse or poor parental
supervision) could not be reviewed because we found
no studies that examined their validity.
Methods
To identify the relevant literature, a computer-based
search in the databases medline and PsycINFO
using the full-text search strategy retrospect* and
[reliability, validity] and [adverse child* exper*,
child* abu*, neglect] or [parent* bonding, parent*
rearing, parent* relationship, parent* attachment]
was conducted for the years 1980 to 2001. It produced 18 articles. Two were reviews, 9 reported
studies with systematic evaluation of retrospective
recall in adulthood of adverse experiences in childhood, and 7 were irrelevant to the subject. Hand
searching led to the identification of a further 30
reports. Attention was confined to studies using
quantitative measures on samples of at least 40
participants and in which corroboration was available through contemporaneous prospectively obtained court or clinic or research records, or through
comparisons between siblings for their retrospective
recall of the same experiences. This resulted in a
reduction to 8 studies in which there were validating
data for retrospective recall (see Table 1). In addition,
attention was paid to a further 6 studies with findings on long-term reliability of retrospective recall
(see Table 2). A list of all identified studies is available from the authors.
Results
Contemporary official records and retrospective
recall
Williams (1994) conducted a study on 129 women
with clinically documented severe childhood sexual
abuse (CSA). The majority (86%) were AfricanAmerican. About 17 years after having been treated
for the abuse, the women were re-interviewed about
CSA without being told that a comparison was going
(Williams, 1994;
Banyard & Williams, 1996)
Reports of 52 patients/siblings,
39 controls/siblings compared
Study description
Note: CSA: childhood sexual abuse; CPA: childhood physical abuse; NEG: neglect.
Main references
Study
Family conflict
Childparent
attachment
CSA
CSA
CPA
Ever in correctional
institution
Early home
environment
Parental negativity
Mixed category of
abuse and neglect
NEG
CPA
CSA
Childhood adversities
.21
.40
correlation
.14
.42
).01
.72
.14
.74
).01
.77
).09
Kappa
both abused
one abused
both abused
one abused
both neglected
one neglected
physical violence
interfamilial
intercourse
4251% of the
questions showed
significant agreement
if
if
if
if
if
if
.62
.64
.17
.60
.38.52
.20.44
.76.84
.90.92
.11.20
.40.50
.41
Sensitivity
Estimates of validity
yes vs. no
CSA in females
CSA in males
CPA
if outcome good
if outcome poor
if outcome good
if outcome poor
263
264
Main references
Study description
Childhood adversities
Representative sample
of 2,750 adolescents
983 young adults
244 Community
residents
Family relations,
early home
Parental Bonding
Instrument
Correlation
.84
.80
.45
.47
1.00
.82
.66
.80.90
whites
blacks
three yrs interval
265
266
Only half the reports of sexual abuse were confirmed by the other sister but the corroboration rates
were substantially higher for physical abuse (71%)
and for neglect (82%) giving rise to kappa values of
.58, .66 and .80 respectively. What was striking,
however, was that corroboration was almost entirely
dependent on whether or not both sisters had
experienced the same adversity. When they had (as
was the case in the great majority of pairs) the kappa
values ranged from .72 to .77 but when they had not
(as was the case for 20 pairs with respect to both
physical and sexual abuse and 8 pairs with respect
to neglect), agreement was essentially zero, with
kappa values ranging from ).09 to .14. Bifulco et al.
(1997) reported that corroboration did not seem to be
a function of confiding between the sisters, but the
almost total lack of corroboration in the case of pairs
with discordant experiences clearly casts serious
doubt on the reality of the validation. As the authors
noted, it could be argued that the sisters were unable
to differentiate between their own and the others
experiences. Whether or not that constitutes the
explanation, the findings would certainly seem to
raise questions about the utility of this research
strategy to test the validity of retrospective recall.
Robins et al. (1985) also used a sibling comparison design to test the validity of retrospective
recall. The degree of agreement in recall of pairs of
adults aged 3050 years in which one was a psychiatric patient was compared with pairs of controls
neither of whom were psychiatric patients. The
agreement within pairs of both sorts was said to be
well above chance levels for a range of variables
reflecting different aspects of family discord and did
not differ between patients and controls. Robins
et al. reported a significant agreement for 40% to
50% of 150 questions but actual levels of agreement
were not reported.
267
more factors that showed an association to the official records than to self-reports in this study.
268
Infantile amnesia
It is well known that most adults have few, if any,
memories of discrete events occurring during the
first two years of life. This conclusion has been
supported by systematic studies of both ordinary
events (such as the birth of a sibling or admission to
hospital see Usher & Neisser, 1993) or very unusual ones such as urethral catheterisation (Goodman, Quas, Batterman-Fauce, Riddlesberger, &
Kuhn, 1994). The evidence suggests that it is not
that the experiences are not registered at the time
(because even quite young children can remember
events that occurred some months ago), but rather
that recall in adulthood is unusual. This may be
because concepts in adult life are so different from
those in childhood (Howe & Courage, 1993; Lewis,
1995). Alternatively, following the emergence of
conversational language, repeated retrieval of memories through conversations about the past may
preserve them over time (Nelson, 1993). Or perhaps
repeated retrieval through non-language, as well as
language, stimuli (because both increase greatly
post-infancy) may increase the accessibility of
memories (Barr & Hayne, 2000). Whatever the explanation, it is clear that adults cannot be expected
to have an accurate recall of events that took place in
infancy.
269
Discussion
It is all too clear that there are major methodological
problems involved in the assessment of the validity
of retrospective recall of seriously adverse experiences in childhood. On the face of it, the best method
would seem to be the comparison of contemporaneous and retrospective accounts as obtained in epidemiological/longitudinal studies of non-clinical
populations. The main problem is that the contemporaneous accounts usually involve a different
informant (generally the parent) from that used at
follow-up (generally the subjects themselves), and
that the accounts in childhood do not always cover
the complete age period. The follow-up into adult life
of groups whose abuse or neglect or family conflict
has been documented by the courts or by clinic
attendance has the additional serious limitation that
only a minority of cases of serious adversity get
documented in this way. Accordingly, retrospective
reports of abuse/neglect by individuals in non-clinical, non-court control groups are more likely to be
valid accounts of non-referred adversity than false
positives. Also, it cannot be assumed that the retro-
270
Correspondence to
r PsyDr Jochen Hardt, Klinik und Poliklinik fu
chosomatische Medizin und Psychotherapie, Universitatsklinikum Mainz, Untere Zahlbacherstr. 8,
D-55131 Mainz, Germany; Email: hardt@mail.
uni-mainz.de
References
Achenbach, T.M., McConaughy, S.H., & Howell, C.T.
(1987). Child/adolescent behavioral and emotional
problems: Implications of cross-informant correlations for situational specificity. Psychological Bulletin,
101, 212232.
Amaya-Jackson, L., Socolar, R.R.S., Hunter, W., Runyan, D.K., & Colinders, R. (2000). Directly questioning children and adolescents about maltreatment.
Interpersonal Violence, 15, 725759.
Banyard, V.L., & Williams, L.M. (1996). Characteristics
of child sexual abuse as correlates of womens
adjustment: A prospective study. Journal of Marriage
and the Family, 58, 853865.
Barr, R., & Hayne, H. (2000). Age-related changes in
imitation: Implications for memory development. In
C. Rovee-Collier, L. Lipsitt, & H. Haynee (Eds.),
271
272
273