This article was downloaded by: [University Of Pittsburgh]

On: 16 June 2013, At: 02:44

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Clinical and Experimental

Neuropsychology
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/ncen20

Use of the Wechsler Adult Intelligence Scale Digit

Span subtest for malingering detection: A meta-
analytic review
a a a a
Lindsey J. Jasinski , David T. R. Berry , Anni L. Shandera & Jessica A. Clark
a
University of Kentucky, Lexington, KY, USA
Published online: 03 Mar 2011.

To cite this article: Lindsey J. Jasinski , David T. R. Berry , Anni L. Shandera & Jessica A. Clark (2011): Use of the
Wechsler Adult Intelligence Scale Digit Span subtest for malingering detection: A meta-analytic review, Journal of
Clinical and Experimental Neuropsychology, 33:3, 300-314

To link to this article: http://dx.doi.org/10.1080/13803395.2010.516743

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or
systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution
in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the
contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and
drug doses should be independently verified with primary sources. The publisher shall not be liable for
any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused
arising directly or indirectly in connection with or arising out of the use of this material.
 JOURNAL OF CLINICAL AND EXPERIMENTAL NEUROPSYCHOLOGY
2011, 33 (3), 300–314

Use of the Wechsler Adult Intelligence Scale Digit Span

subtest for malingering detection:
A meta-analytic review

Lindsey J. Jasinski, David T. R. Berry, Anni L. Shandera, and Jessica A. Clark

University of Kentucky, Lexington, KY, USA

Twenty-four studies utilizing the Wechsler Adult Intelligence Scale (WAIS) Digit Span subtest—either the Reliable
Digit Span (RDS) or Age-Corrected Scaled Score (DS-ACSS) variant—for malingering detection were meta-an-
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013

alytically reviewed to evaluate their effectiveness in detecting malingered neurocognitive dysfunction. RDS and
DS-ACSS effectively discriminated between honest responders and dissimulators, with average weighted effect
sizes of 1.34 and 1.08, respectively. No significant differences were found between RDS and DS-ACSS. Similarly,
no differences were found between the Digit Span subtest from the WAIS or Wechsler Memory Scale (WMS).
Strong specificity and moderate sensitivity were observed, and optimal cutting scores are recommended.

Keywords: Digit span; Malingering; Feigning; Wechsler Adult Intelligence Scale–Third Edition; Wechsler Memory
Scale–Third Edition; Wechsler; Meta-analysis.

Neuropsychological assessments provide information assessments. In nonforensic contexts, estimates of feign-

vital to diagnosis, prognosis, and treatment following an
injury or illness; however, test results may be rendered
useless when feigning or suboptimal effort occurs. Thus,
it has been strongly recommended that an examination
of malingering should be included in every neuropsy-
chological test battery (Bush, et al., 2005). According
to the Diagnostic and Statistical Manual of Mental
Disorders–Fourth Edition, Text Revision (DSM–IV–TR;
American Psychiatric Association, 2000), malingering
is the “intentional production of false or exaggerated
physical or psychological symptoms, motivated by exter-
nal incentives” (p. 739). Malingering is thought to be
a multidimensional construct, with distinctions made
between psychiatric, neurocognitive, and somatic feign-
ing (Larrabee, 2007). In general, estimates of the base rate
of neurocognitive feigning have demonstrated substan-
tial rates of malingering, particularly in forensic settings.
For instance, Frederick, Crosby, and Wynkoop (2000)
estimated 44% of criminal defendants had an invalid test-
taking approach, while Mittenberg, Patton, Canyock,
and Condit (2002) showed rates of malingering around
41% in litigants after a traumatic brain injury (TBI).
Thus, it seems that 40% is a reasonable approximation
of the base rates of neurocognitive feigning in forensic
ing are lower, around 7.3%, but are significant enough to
warrant attention (Berry & Schipper, 2007). These esti-
mates point to the need for well-validated measures of
malingering and their routine utilization.
Current malingering assessment practices
Dedicated malingering tests
Tests developed specifically to assess feigning, also known
as dedicated malingering tests, have received signifi-
cant attention in recent years. Several such tests have
been well supported in the literature (Grote & Hook,
2007). The most common type of dedicated malin-
gering test utilizes a forced-choice memory paradigm,
where patients are shown a stimulus to remember and
then are asked to choose the correct response from
several options after a brief delay. In this model,
feigning is suspected in the presence of either worse-
than-chance performance or scores falling below an a
priori cutting score. Examples of well-validated forced-
choice measures for detecting malingered neurocognitive
dysfunction include the Test of Memory Malingering

Address correspondence to David T. R. Berry, Department of Psychology, University of Kentucky, 012-D Kastle Hall, Lexington,
Kentucky, 40504–0044, USA (E-mail: dtrb85@gmail.com).
© 2011 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
http://www.psypress.com/jcen DOI: 10.1080/13803395.2010.516743

(Tombaugh, 1996), Word Memory Test (Green, Allen, &
Astner, 1997), Letter Memory Test (Inman et al., 1998),
Validity Indicator Profile (Frederick & Crosby, 2000),
Portland Digit Recognition Test (Binder, 2002), and
Victoria Symptom Validity Test (Slick, Hopp, Strauss,
& Thompson, 1997). See Vickery, Berry, Inman, Harris,
and Orey (2001), and Grote and Hook (2007) for more
comprehensive reviews.
While dedicated malingering tests are by far the most
widely used in clinical practice and boast the most empir-
ical support, they suffer from some limitations. One of the
strongest criticisms of dedicated malingering tests is their
vulnerability to coaching, or the provision of information
to examinees prior to an evaluation regarding the pres-
ence and/or nature of malingering indices. For example,
attorneys might discuss specific tests with clients or warn
them not to fail tests where they have to choose answers
from two options. The impact of attorney coaching has
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
been well documented in the malingering literature to
date (Gutheil, 2003; Lees-Haley, 1997; Suhr & Gunstad,
2007; Wetter & Corrigan, 1995). Additionally, increasing
amounts of information available via the Internet make
self-coaching an easy task when preparing for a neu-
ropsychological evaluation (Bauer & McCaffrey, 2006;
Ruiz, Drake, Glass, Marcotte, & van Gorp, 2002). For
example, Allen and Green (2001) documented a steady
decline in failure rates of the Computerized Assessment
of Response Bias (CARB) over a six-year period corre-
sponding closely to increasing Internet availability, and
Berry and Schipper (2007) describe a recent search of the
Psychological Assessment Resources (PAR, Inc.) web-
site that provided extensive information on the names,
acronyms, and purpose of several malingering mea-
sures. This, coupled with the finding that nearly 20% of
Americans surveyed felt that intentional distortion or
misrepresentation within the context of the legal compen-
sation system was acceptable (Public Attitude Monitor,
1992), points to the importance of developing accurate
tests for detecting malingering.
In addition to vulnerability to coaching, another con-
cern with dedicated malingering tests is the administra-
tion time required. For many well-validated procedures,
upwards of 20 minutes additional time is needed for a
single dedicated malingering test. Given Rogers’s (1997)
recommendation that multiple indicators of feigning be
employed, the added testing time could be significant.
Further, dedicated tests of malingering provide data rele-
vant only to effort, but no information on other cognitive
functions, making them time consuming with very lit-
tle relevant diagnostic information obtained (Mathias,
Greve, Bianchini, Houston, & Crouch, 2002).
Embedded indices
In response to the notable limitations of dedicated malin-
gering tests, interest in indices embedded within stan-
dard neuropsychological tests has grown in recent years.
Malingering indices developed within large-scale bat-
teries have included subtests from the Wechsler Adult
Intelligence Scale Revised (WAIS–R; Wechsler, 1981) and
DIGIT SPAN AND MALINGERING DETECTION 301
Third Edition (WAIS–III; Wechsler, 1997a; Digit Span:
Trueblood & Schmidt, 1993; Vocabulary–Digit Span,
WAIS Discriminant Function: Mittenberg, Theroux-
Fichera, Zielinski, & Heilbronner, 1995; Reliable Digit
Span, RDS: Greiffenstein, Baker, & Gola, 1994),
from the Wechsler Memory Scale–Revised (WMS–
R; Wechsler, 1987) and Third Edition (WMS–III;
Wechsler, 1997b; Logical Memory recognition: Killgore
& DellaPietra, 2000; Discriminant Function: Mittenberg,
Patton, & Legler, 2003), and from the Halstead–Reitan
Battery (Discriminant Function: Mittenberg, Rotholc,
Russell, & Heilbronner, 1996). Additionally, several indi-
vidual tests have been examined for their ability to detect
malingering, including the Wisconsin Card Sorting Test
(Bernard, McGrath, & Houston, 1996), Auditory Verbal
Learning Test (Binder, Villaneuva, Howieson, & Moore,
1993), California Verbal Learning Test (Millis, Putnam,
Adams & Ricker, 1995), Finger Tapping Test (Arnold
et al., 2005), and Rey Complex Figure Test (Meyers &
Volbrecht, 1999), among others. See Larrabee (2007) for
a more comprehensive review of the malingering research
on several of these instruments.
Methodological considerations
Research methodology
Methodological rigor has been a topic of considerable
attention with regard to malingering research. While case
studies and differential prevalence designed studies have
been utilized in the past, they are not currently accepted
as standard practice for malingering research. In con-
trast, the most frequently utilized methodologies include
simulation and known-groups design studies. Simulation
studies involve a group of participants asked to feign
symptoms of a disorder while completing a series of
tests, who are compared to a group of “normal” par-
ticipants asked to respond honestly or to perform to
the best of their ability on the same test battery. The
honest group often has no prior psychiatric or neurolog-
ical diagnosis. However, Rogers (2008) points out that
such simulation designs have only “marginal relevance
to malingering because they do not differentiate feigned
from genuine disorders” (p. 413). Rather, the stronger ver-
sion of the simulation design study, enhanced simulation
design, employs a clinically relevant honest comparison
group, thus providing direct information on the differ-
ences between the feigned and authentic disorder. For
example, in traumatic brain injury research, the clinically
relevant group would be participants who had sustained a
brain injury but were not seeking compensation and had
verified honesty (Rogers, 2008).
Simulation studies are often utilized due to their strong
internal validity, control over group assignment, and
convenience. However, concerns are often raised regard-
ing the generalizability of simulation studies. Several
methodological considerations have been recommended
to strengthen the external validity of a study. For
instance, in addition to the malingering group being
clinically relevant or having experienced the condition
 302 JASINSKI ET AL.
of interest (e.g., brain injury), honest and malinger-
ing groups should be matched demographically. Further,
Rogers (2008) recommends that malingering instructions
be clear and comprehensible to the participants, as well
as specific. A warning for participants to feign believably
and relevant proportional incentives, both positive for
following instructions, and negative for failing to do so,
should be included to increase motivation in the malin-
gering group. Instructions should include information
on relevant symptoms of interest, as well as detection
strategies and avoidance techniques. Additionally, allow-
ing preparation time for participants after they are given
symptom information greatly enhances similarity to real-
world evaluations, as does the inclusion of both malinger-
ing and nonmalingering tests within the research battery.
Finally, a postexperimental check for understanding and
cooperation with instructions is necessary to exclude data
from subjects who were not compliant. If utilized, each
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
of these methodological considerations can help maintain
strong internal validity, while improving external validity
(Rogers, 2008).
Known-groups methodology, in contrast, is often uti-
lized for its strong external validity, at the expense of
some internal validity and study control. Known-groups
designs employ a fully clinically relevant malingering
group thought to be feigning based on failure of one
or more well-validated malingering tests. This “known”
malingering group is then compared to a clinically rel-
evant honest group of similar background and injury
severity. The generalizability of such results is very high,
given that both groups have experienced the necessary
antecedent events and motivation to malinger similar
to those seen in clinical practice. However, known-
groups methodology is limited based upon the qual-
ity of the malingering indicators used to determine
group status, and imperfect measures result in imperfect
groups and subsequently lower internal validity (Rogers,
2008).
Research methodology is an important factor to con-
sider when conducting meta-analytic studies, because
differing methodologies greatly impact study outcomes,
resultant operating characteristics, and diagnostic clas-
sification parameters of any malingering detection tool.
Because each has strengths and weaknesses, well-
validated malingering measures should have supportive
evidence from both types of methodologies, per Rogers’s
(1997) recommendations. Further, specific methodolog-
ical characteristics should be coded and explored as
confounds and contributing factors to the findings, as is
the case in the current meta-analytic review.
Statistical methods
Diagnostic validity refers to a “test’s ability to differ-
entiate persons with and without a specified disorder”
(Larrabee & Berry, 2007, p. 14), in this case malinger-
ing. A cutting score is a score at or below which, based
on prior research, an individual is thought to be malin-
gering. Sensitivity and specificity are the most commonly
reported statistics in malingering detection studies and
provide information on the test’s ability to classify malin-
gering. Sensitivity (SN) refers to the probability of a
positive test sign (i.e., failure of malingering test) in indi-
viduals who are malingering. Specificity (SP), in contrast,
refers to the probability of a negative test sign (i.e., pass-
ing malingering test) in individuals who are responding
honestly. The hit rate is the overall efficiency of the test,
taking into account both sensitivity and specificity rates.
It is important to note that distributions from which
these scores are derived usually overlap, and thus there
is no score that perfectly separates malingering and hon-
est groups, resulting in the need to choose the most
effective cutting score based on minimization of relevant
errors (Greve & Bianchini, 2004). While sensitivity and
specificity rates are the most frequently cited statistics in
malingering detection, they do not represent the entire
picture. When derived from malingering studies, these
statistics are typically based on equal sample sizes with
known base rates of malingering approximating 50%.
Positive and negative predictive powers, in contrast, take
into account the relevant base rate and therefore pro-
vide a more pertinent statistic for clinicians. Additionally,
they are more relevant for clinicians in that they pre-
dict the probability that an individual is either honest
or malingering based on test scores. Specifically, positive
predictive power (PPP) refers to the probability of malin-
gering given a positive test score, while negative predictive
power (NPP) is the probability of the absence of malin-
gering, or honest responding, given a negative test score.
However, given that these values vary with the base rate
of each malingering study, they are not explored in the
present meta-analysis, but can be easily calculated using
the relevant sensitivity and specificity values provided
subsequently. Finally, an effect size describes a test’s abil-
ity to differentiate between two groups, with larger effect
sizes indicative of larger average differences between the
groups. Greater average differences, then, translate into
more accurate diagnostic classification statistics. Each
of these statistics are examined in further detail in the
present study.
Goals of present study
The goal of the present study is to provide a com-
prehensive, meta-analytic evaluation of the ability
of the Digit Span subtest in detecting malingered
neurocognitive dysfunction. The Digit Span subtest
is embedded within the revised, third, and fourth
editions of the Wechsler Adult Intelligence Scale
(WAIS–R, WAIS–III, WAIS–IV), as well as the revised,
third, and fourth editions of the Wechsler Memory Scale
(WMS–R, WMS–III, WMS–IV). It requires an exami-
nee to repeat increasing strings of digits in forward and
reverse order, until two consecutive strings of the same
length are missed. A variation, Reliable Digit Span (RDS;
Greiffenstein et al., 1994), has been developed using the
sum of the longest strings of digits forward and back-
ward where both trials were passed. RDS was developed
specifically as a malingering detection tool and has been
extensively studied.

The Digit Span subtest was chosen as the focus for
the present review for several reasons. Primarily, it is
embedded within the Wechsler Adult Intelligence Scale
(WAIS–R, WAIS–III, WAIS–IV) and Wechsler Memory
Scale (WMS–R, WMS–III, WMS–IV), which are among
the most frequently utilized neuropsychological instru-
ments in clinical practice today (Camera, Nathan, &
Puente, 2000; Rabin, Barr, & Burton, 2005). As such,
retrospective malingering analyses can be conducted if a
clinician administered either of these instruments and is
in need of information regarding malingering status. Such
retrospective analyses can also be particularly useful in
attaining research goals. Further, it is virtually identical
across several revisions of the WAIS and WMS, thereby
increasing the number of studies available for the present
meta-analysis, subsequently improving the reliability of
the findings. Finally, the Digit Span subtest is easily and
quickly administered either individually or as part of a
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
battery, so is particularly user friendly in a clinical setting.
Thus, the focus of the present study is on a meta-analytic
review of the Digit Span subtest and its ability to detect
malingering.
METHOD
A priori inclusion criteria
Malingering studies utilizing the Digit Span subtest of
the revised or third editions of the WAIS and WMS were
surveyed through March 2009. Studies investigating the
usefulness of either the total age-corrected scaled score
(DS-ACSS) or Reliable Digit Span (RDS) were included.
To be considered for inclusion, manuscripts were required
to come from a peer-reviewed journal, peer-reviewed arti-
cle in press, or English-language dissertation. Conference
presentations and papers were not included in the present
study, as they do not typically undergo as thorough a
review process. In cases where dissertation results were
later published, only the published paper was included for
evaluation.
Publications meeting these criteria were examined and
were included if they utilized known-groups or enhanced
simulation methodology (e.g., with a clinically relevant
honest comparison group). Thus, studies using student
or community volunteers instructed to malinger were
included provided they were compared to a group with
a history of traumatic brain injury or other neurologi-
cal disorder. Studies were excluded if comparisons were
made only with normal controls, as they do not pro-
vide information on a test’s ability to discriminate feigned
from true injury (Rogers, 1997). In the present study, clin-
ical samples primarily included traumatic brain injury
and chronic pain patients, although several mixed neuro-
logical groups were also included. Chronic pain patients
were included because they often feign neurocognitive
symptoms in addition to somatic complaints (Iverson &
McCracken, 1997). Groups feigning psychiatric symp-
toms or mental retardation were not included in the
present review to reduce the heterogeneity among groups
and the possibility of comparing “apples to oranges,” a
DIGIT SPAN AND MALINGERING DETECTION 303
common criticism of meta-analysis (Lipsey & Wilson,
2001). Thus, the primary focus of this paper was on
feigned neurocognitive symptoms. The use of the Digit
Span test in detecting other types of feigning should be
explored in future studies.
Data collection
Keyword-guided searches of the PsycINFO and
MedLine databases were undertaken to identify stud-
ies of any version of the WAIS or WMS used to detect
malingering. Combinations of the keywords “malinger∗ ,”
“feign∗ ,” “dissimulat∗ ,” “fak∗ ,” and “pattern analys∗ ”
were crossed with “WAIS,” “WAIS-R,” WAIS-III,”
“WMS,” “WMS-R,” and “WMS-III,” both in acronym
and written form, as well as “Wechsler.” The asterisks
are used in search databases to allow for variable suffixes
(such as -ing, -ed, -s, -ion). Additionally, several of the
prominent pattern analysis techniques used in malinger-
ing literature were searched for with either the WAIS or
the WMS, including “performance curve,” “magnitude
of error,” “violation of learning principle,” and “atypical
pattern analysis.” Reference sections of review papers,
book chapters, test manuals, and empirical studies were
scrutinized to identify additional manuscripts. Lastly,
email contact with highly published individuals using the
WAIS or WMS for malingering detection was attempted
in order to solicit any papers in press or under review.
As a result of these keyword-guided searches, 74 papers
were identified, 43 of them utilizing a version of the
WAIS, 24 reviewing a version of the WMS, and 4 stud-
ies using both the WAIS and the WMS. It was necessary
to exclude studies on the following basis: lack of a clini-
cally relevant control group (n = 10), lack of any feigning
group (n = 5), the use of a correlational or differential
prevalence design (n = 6), or because the Digit Span
subtest was used to determine malingering status for a
known-groups design (n = 2). Further, 21 studies were
excluded because they examined subtests of the WAIS or
WMS other than Digit Span for malingering detection.
Finally, 6 papers were also excluded on the basis of being
a conference presentation. A total of 24 papers (WAIS:
n = 23; WMS: n = 1) were considered appropriate based
on the a priori inclusion criteria and were coded. These
studies are listed in the Appendix.
Selection of data for review
In some cases, multiple contrasts were available from
a single study. Here, only the independent contrasts
were selected because using effect sizes from overlapping
groups increases statistical dependence and inflates the
mean effect sizes for the study (Lipsey & Wilson, 2001).
One source of nonindependence involved comparing a
single feigning group with multiple honest groups. For
example, a few studies included multiple honest groups
of mild and severe TBI patients, all compared to a sin-
gle malingering group. In this case, the contrast using a
group of mild TBI patients was chosen because it was
 304 JASINSKI ET AL.
thought to be most clinically relevant, given that mild
TBI patients are most likely to present for neuropsy-
chological testing and have the greatest motivation to
exaggerate symptoms, compared to those with a severe
TBI where compensation could theoretically be based
upon physical findings and notable functional decline.
Another source of nonindependence of contrasts was the
use of several malingering groups compared to a single
honest group. In this situation, the more stringent classi-
fication criteria for the malingering group were used (e.g.,
choosing definite over possible malingered neurocogni-
tive dysfunction, MND, based on the Slick, Sherman, &
Iverson, 1999, criteria).
A final source of nonindependent contrasts arose from
the overlapping nature of the Digit Span age-corrected
scaled score (DS-ACSS) and Reliable Digit Span (RDS)
variants, where the same trial scores are used to com-
pute each scale. Four of the studies included provided
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
enough information to calculate effect sizes for both vari-
ants. In order to create independent contrasts, two studies
were randomly assigned to contribute to the DS-ACSS
analyses, while the other two were assigned to the RDS
analyses. While this method slightly reduced the number
of studies available for each of the analyses, the benefits of
independent effects outweighed the desire for additional
studies.
Data extraction
Demographic variables such as age, gender, and mean
education level were extracted from each of the studies
and are presented in the Appendix. Additionally, the type
of group was coded according to the following system:
traumatic brain injury, mixed neurological, and chronic
pain. The type of participants in the malingering group
was also coded, including patients, students, or com-
munity volunteers. Means and standard deviations for
RDS and DS-ACSS were extracted when available, as
were sensitivity and specificity, hit rate, and methodolog-
ical characteristics of each study, specific to the research
design type. Methodological characteristics deemed rele-
vant in this study were used to rate the methodological
strength of each study and are based on recommenda-
tions from prior literature reviews (Berry, Baer, Rinaldo,
& Wetter, 2002; Berry & Schipper, 2007). A total method-
ological strength score for each contrast was calculated
and was divided by the number of possible points for
each design type (e.g., Sim, enhanced simulation; KG,
known groups). This was done because 16 variables were
coded for simulation design studies, while known-groups
studies had only 7. Thus, a proportion score was used
to directly compare the designs. Higher scores represent
methodologically stronger studies.
In order to ascertain coding accuracy, 42% (10) of pub-
lications were cross-coded by a second rater. Reliabilities
were found to be adequate for demographic variables
(r = .97), methodological characteristics (r = .92), means
and standard deviations (r = 1.0), and diagnostic accu-
racy statistics (sensitivity r = 1.0, specificity r = 1.0).
Discrepancies were discussed until 100% consensus was
reached.
RESULTS
A total of 24 studies were selected and were coded accord-
ing to the inclusion criteria. Of these, two studies did not
provide enough information (i.e., standard deviations) to
calculate an effect size, while an additional two studies
provided enough information to compute two indepen-
dent effect sizes. In other words, each of these two studies
included at least two feigning and two honest groups
for independent comparison. Thus, the analysis of effect
sizes included 24 independent contrasts. Further, of the
24 studies examined, only 19 studies provided informa-
tion on the relevant diagnostic accuracy statistics. See
Appendix for a summary of each study, including effect
sizes and diagnostic accuracy statistics.
Calculation of effect sizes was completed utilizing the
Hedges’s d statistic from the meta-analytic software pro-
gram DSTAT (Johnson, 1989). Further analyses were
conducted using the Statistical Package for the Social
Sciences (SPSS) software. The Hedges’s d value repre-
sents the distance between two sample means in terms of
pooled standard deviation units. Although the alterna-
tive effect size Cohen’s d is often used, it has been shown
to be biased with small sample sizes, such as those often
utilized in neuropsychology research (Hedges & Olkin,
1985). Hedges’s d is a standardized metric thought to be
an unbiased estimate of effect size by correcting for sam-
ple size with the inverse variance, thus giving more weight
to studies with larger sample sizes (Lipsey & Wilson,
2001). Thus, in the text below, all d values are Hedges’s d.
The next step in data analysis was screening the effect
sizes for outliers, or unusually high or low values that
might skew subsequent analyses. An important consid-
eration when conducting outliers analysis is the bal-
ance between retaining all possible comparisons in order
to investigate any methodologically interesting determi-
nants of variability in the dataset, and excluding extreme
values unlikely to be representative of observations within
the population of interest. In the present study, ade-
quate variability was seen in the effect sizes, so the desire
for a representative set of findings from the popula-
tion was a priority, and outliers were excluded. Further,
such extreme effect sizes have disproportionate influence
on meta-analytical statistics (e.g., means, variances) and
may distort them (Lipsey & Wilson, 2001). Thus, a step-
and-fences procedure (Tukey, 1977) was completed for
the set of 24 independent contrasts. This procedure uses
the median d value to determine the likelihood of a
value belonging to the same distribution. Values outside
of the inner fence, located approximately 2.67 standard
deviations away from the median d value, are consid-
ered outliers. Application of this process resulted in two
outliers being identified from the set of independent con-
trasts (d values: 2.97 and 3.46), both of which contributed
to a positively skewed distribution. It is noteworthy that
one outlier was from a study with a very small sam-
ple size (n = 7 for the feigning group) while the other

came from the only study utilizing inmates as part of the
feigning group. Removal of these outliers brought greater
normality to the distribution.
Following outlier removal, 22 independent contrasts
using either the DS-ACSS (n = 12) or RDS (n = 10)
variant remained for analysis. The distribution of the
sample was roughly normal. The d values for the
independent contrasts ranged from 0.92 to 2.46, with
a weighted mean effect size of 1.25 (95% confidence
interval, CI = 1.13–1.37) and a median of 1.23. Overall,
these mean values indicate that the two variants of the
Digit Span test separate malingering and honest groups
by just over one pooled standard deviation, a significant
effect size that is considered large (Cohen, 1988). Thus,
honest participants had higher scores on the DS-ACSS
and RDS subtests than did those thought to be malinger-
ing, indicating that malingerers suppressed their scores
beyond what would be observed in a clinical population
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
responding honestly.
Representativeness of study findings
While these effect sizes are substantial, several factors
should be considered before having confidence in the
results of a meta-analysis, including the possibility that
the findings are not characteristic of all studies con-
ducted in an area. Specifically, meta-analysis has often
been challenged on the basis of the “file drawer problem,”
whereby only statistically significant results are accepted
for publication in peer-reviewed journals, leaving open
the possibility of small effects or negative findings going
unpublished. The result would be effect sizes that overes-
timate the actual effect in the population. In response to
such criticism, Rosenthal (1991) developed the fail-safe N
statistic, using it to estimate the number of studies with
a null effect that would have to exist in researcher’s “file
drawers” in order to reduce the observed mean effect size
from the sample studies to a nonsignificant level. Using
this procedure, the fail-safe N based on the set of 22
independent contrasts in this study resulted in an esti-
mated 244 unpublished null findings necessary to render
the overall effect nonsignificant (p > .05). Based on this
analysis, it appears highly improbable that this number of
studies exist, thereby increasing confidence in the findings
of this study as representative of the population at large.
An additional consideration is the extent to which
methodological rigor influenced the overall effects
derived from the studies. Intuitively, one would assume
that studies with increased methodological rigor would
result in less variable, and thus more reliable, effect
sizes. In order to ascertain the relationship between
methodological quality and the effect sizes, a correla-
tion coefficient was calculated using the methodological
strength score assigned to each study and resulted in
r = −.24 (p = .28). This indicates that, overall, method-
ological quality was not significantly associated with the
magnitude of effect size.
Thus, based on the sizable mean effect sizes, large value
of the fail-safe N, and lack of a significant relationship
between methodological rigor and effect size, the results
DIGIT SPAN AND MALINGERING DETECTION 305
of this study can be interpreted as representative of the
population of studies in the area of malingering and
the Digit Span test, including both DS-ACSS and RDS.
This subtest is able to differentiate honest and malinger-
ing groups by more than one pooled standard deviation
on average. However, what is not yet known is extent
of variability in the effect sizes and the diagnostic accu-
racy of DS-ACSS and RDS. These issues will be explored
subsequently.
Variability of effect sizes
When summarizing data across studies, a further consid-
eration is the homogeneity of the effect sizes derived, with
significant heterogeneity suggesting that a single mean
effect size should not be used to represent an overall
effect. In other words, the effect sizes may differ from
the population mean by more than just sampling error,
rather by systematic differences that can be further exam-
ined. The Q statistic is commonly used in meta-analysis,
with a significant result indicating heterogeneity. It rep-
resents an asymptotic chi-square distribution with k −
1 degrees of freedom (Hedges & Olkin, 1985). Results
from the 22 independent contrasts indicate that a sig-
nificant amount of variability in the effect sizes was not
observed, Q(21) = 30.58, p < .08, although the results
approached significance. Hence, the overall mean effect
size does appear to adequately represent the overall abil-
ity of the Digit Span test to detect malingering.
Within- and between-measure differences
Although the overall effect size showed no statistically
significant heterogeneity, theoretically derived modera-
tors were explored to examine their potential impact on
the data, as the idea of predetermined focused contrasts is
potentially unrelated to overall variability and is therefore
useful to examine (Rosenthal, 1991). Further, since the
Q statistic noted above approached significance, modera-
tors may be useful in explaining the observed variability,
despite lack of formal statistical significance. This explo-
ration included the variant utilized (RDS or DS-ACSS)
and the test battery from which Digit Span scores were
derived (versions of WAIS or WMS), in addition to other
methodological moderators discussed below.
A major source of hypothesized variability is between-
variant differences, given that the Digit Span test can
be used in the form of DS-ACSS or RDS. To examine
this, separate effect sizes were calculated for each vari-
ant after outlier extraction, and the results are presented
in Table 1. Overall, both RDS and DS-ACSS showed
robust effect sizes, with RDS having a slightly larger
effect size (d = 1.34, 95% CI = 1.18 − 1.50) than DS-
ACSS (d = 1.08, 95% CI = 1.01 − 1.50). Further analysis
of these effect sizes showed within-measure homogene-
ity tests to be significant for RDS (Q = 15.89, p < .05),
indicating that the overall effect size for RDS is not able
to describe the dataset accurately, and further moderator
analyses could help explain the variability seen within the
 306 JASINSKI ET AL.
TABLE 1
Effect size data by subscale
d 95% CI k
RDS 1.34 1.18 – 1.50 10
DS-ACSS 1.08 1.01 – 1.50 12
p value >.05
Note. Data presented are from independent contrasts, outlier-
parsed dataset. RDS = reliable digit span, DS-ACSS = digit
span age-corrected scaled score, d = weighted mean corrected
Hedges’s g effect size, 95% CI = 95% confidence interval around
the weighted mean effect size, k = number of studies included in
analysis.
RDS data. Homogeneity tests were not significant, how-
ever, for the DS-ACSS data (Q = 12.25, p > .05).
In addition to examining the homogeneity of each
variant, the QB statistic was used to test for significant
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
differences between the ability of the RDS and DS-ACSS
to separate honest and feigning groups. The QB statistic
is analogous to an F-test and uses the Scheffé method
to protect against Type I error. In essence, this analog to
analysis of variance (ANOVA) partitions the overall Q
statistic for the study into that part that can be explained
by sampling error within the two groups (Qw ; for RDS
and DS-ACSS) and the portion that can be explained by
the independent variable (QB ; Lipsey & Wilson, 2001).
Essentially, this is the within- and between-groups vari-
ance used in ANOVA and here applied to the differences
between the RDS and DS-ACSS. Using this method, no
significant differences were found in the ability of the
RDS or DS-ACSS to separate honest and feigning groups
(QB = 2.44, p > .05). In general, it appears that both RDS
and DS-ACSS show strength in malingering detection,
and either could be used with confidence.
Another potential moderator was the test battery in
which Digit Span was administered. Table 2 shows the
results from this analysis, and as can be seen, nearly all
versions of the WAIS and WMS resulted in strong effect
sizes, with the lowest seen in the WMS–R. However,
it is important to note that only one study was avail-
able for both the WMS–III and WMS–R; thus, these
TABLE 2
Effect size data by test version
d 95% CI k on the effect sizes was conducted to see whether the real-
WAIS–III 1.28 1.06 – 1.50 7
WAIS–R 1.26 1.10 – 1.42 10
WMS–III 1.27 0.74 – 1.80 1 effect size outcome relative to a simulation or laboratory
WMS–R 1.02 0.55 – 1.49 1 design (n = 7). Results of these and other methodological
Mixed 1.21 0.92 – 1.50 3 moderator analyses appear in Table 3. Again using the QB
p value >.05
Note. Data presented are from independent contrasts, outlier-
parsed dataset. WAIS–III = Wechsler Adult Intelligence Scale–
Third Edition, WAIS–R = Wechsler Adult Intelligence Scale–
Revised edition, WMS–III = Wechsler Memory Scale–Third
Edition, WMS–R = Wechsler Memory Scale–Revised edition,
Mixed = combination of WAIS and WMS Digit Span subtests,
d = weighted mean corrected Hedges’s g effect size, 95% CI =
95% confidence interval around the weighted mean effect size,
k = number of studies included in analysis.
data should be interpreted with caution. When examin-
ing tests of homogeneity across each test version, results
were found to be homogenous for the WAIS–III (Q =
6.11, p > .05) and mixed WAIS/WMS studies (Q = 4.95,
p > .05). In contrast, significant heterogeneity was found
for the WAIS–R (Q = 18.48, p < .05), again suggesting
that the overall effect size is not an accurate descriptor,
and further moderator analyses would be helpful. It is
important to note that the other scales may not have
shown significant variability in part because of small sam-
ple size. Homogeneity testing was not completed for the
WMS–III or WMS–R given that only one effect size was
available from these studies. Again, the QB statistic was
used, and no significant differences were found in the
ability of the WAIS test versions or WMS test versions
in differentiating malingering and honest groups (QB =
1.03, p > .05). This finding of similar effect sizes across
test versions is crucial given the new revisions of both
tests that have been recently developed. The results of
this study suggest that the Digit Span subtest could be
used effectively as a malingering detection tool regardless
of the test version administered and likely can be used in
newer or future versions of the test. Further, retrospec-
tive data analysis that relies on older test versions can be
conducted with confidence.
Methodological moderators
Overall, very little variability was found in the effect sizes
across both the type of variant and test version. However,
as mentioned previously, additional theoretically derived
methodological moderators may be an interesting and
important area of consideration. Further, the hetero-
geneity found in the RDS effect sizes as well as those
derived from the WAIS–R warrant further analyses to
explain variability that extends beyond sampling error.
A major hypothesized source of additional variability
may be the methodological design with which the study
was conducted, given differences in reliability and validity
found between these methods (Rogers, 1997). Further, the
exploration of simulation versus known-groups design
has been a topic of interest in much literature, and exam-
ination of these variables may shed light onto the quality
and applicability of each design.
As such, a focused contrast of the effect of study design
world setting, and hence increased external validity, of
those in known-groups studies (n = 15) moderated the
statistic, results showed no significant differences between
study design type (QB = 2.17, p > .05). Interestingly,
this suggests that results from both simulation (d = 1.27,
95% CI = 1.05–1.23) and known-groups studies (d =
1.29, 95% CI = 1.22–1.36) provided similar effect sizes,
and that both can be considered valuable sources of
information on malingering and the Digit Span test.
Turning to other methodological moderators, those
hypothesized in advance from enhanced simulation stud-
ies included the presence of incentive offered for effort,

TABLE 3
Results of categorical methodological effect size moderator analyses
Manipulation Condition
Study design KG
Sim
Financial incentive Yes
No
Warning believable Yes
No
KG obj. criteria Yes
No
Note. Analyses conducted with outlier-parsed dataset. k = number of studies. KG = known-groups
design, Sim = simulation design.
∗∗∗ p significant at <.001.
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
preparation time, warning participants to feign believ-
ably, and the use of coaching, while those from known-
groups studies included having malingering status based
on objective criteria. Although each of these modera-
tors contributed to the methodological strength score
discussed previously, each was hypothesized as particu-
larly important and warrants individual exploration as a
possible contributing moderator. Interestingly, the effect
of preparation time and coaching on malingering indices
or symptoms could not be examined due to the lack of
any enhanced simulation design studies utilizing those
methodological characteristics.
The next methodological moderator that was exam-
ined was the presence of financial incentive in enhanced
simulation designed studies. Because the incentive was
not consistently awarded to all participants (e.g., use of a
drawing) and due to the very small number of studies uti-
lizing simulation design with financial incentive (n = 3),
this variable was coded as either being present or not.
Results of this analysis are presented in Table 3 and indi-
cate that the presence of financial incentive resulted in
larger effect sizes than when no such incentive was used
(QB = 19.58, p < .001). Given that the RDS variant and
WAIS–R battery showed significant heterogeneity, a spe-
cific contrast examining the moderating effect of incen-
tive was undertaken for those groups. For the WAIS–R,
the presence of financial incentive resulted in larger effect
sizes when using an incentive (QB = 9.00, p < .01). For
RDS, the analysis could not be completed due to hav-
ing only one study using the RDS and financial incentive
and one study without incentive. Intuitively, one would
assume that increased effort via the use of an incentive
would result in less blatant malingering and thus smaller
effect sizes. However, in this case incentives resulted in
more obvious feigning.
Turning to the moderating effect of warning partici-
pants to feign believably, results in Table 3 showed that
the presence of a warning significantly moderated effect
size outcome (QB = 19.71, p < .001). The use of a warn-
ing significantly reduced the effect size. The results should
be interpreted with caution given the very small sample of
those not provided a warning (n = 2); however, this may
indicate that the use of a warning results in less blatant
DIGIT SPAN AND MALINGERING DETECTION 307
k Mean d-value QB
14 1.29 1.04
8 1.17
3 1.34 19.58∗∗∗
5 1.08
6 1.16 19.71∗∗∗
2 1.41
11 1.22 15.91∗∗∗
3 1.51
malingering. Again turning to the RDS and WAIS–R,
results for the WAIS–R were similar, with smaller effect
sizes and less blatant malingering observed in studies uti-
lizing a warning (QB = 8.07, p < .01). Analysis for the
RDS could not be completed because all RDS studies
used a warning.
Finally, turning to the known-groups design studies,
a hypothesized moderator was whether the malingering
classification was based on objective criteria (e.g., well-
validated dedicated malingering tests). Results in Table 3
show that the presence of objective criteria significantly
moderated the effect size outcome (QB = 15.91, p < .001).
As might be expected, those studies employing objective
criteria for establishing the known malingering groups
had smaller effect sizes. However, this finding may be an
artifact of having very few studies that did not employ
objective criteria and thus should be interpreted with cau-
tion. When examining the RDS and WAIS–R, the use
of objective criteria significantly moderated the effect for
both groups (RDS, QB = 4.49, p < .05; WAIS–R, QB =
16.82, p < .001), with smaller effect sizes seen in studies
utilizing this methodological quality.
Diagnostic accuracy
In addition to examining the ability of the Digit Span
subtest in separating groups based on the effect size,
another important consideration for clinical practice is
the diagnostic accuracy of a test. As such, the sensitivity
(proportion of feigners correctly identified by a test) and
specificity (the proportion of honest individuals correctly
identified as such by a test) were examined across studies.
As noted previously, a few studies provided information
relevant for effect size coding, but sensitivity and speci-
ficity values were not available, and vice versa. Table 4
provides the mean diagnostic accuracy statistics for the 19
independent contrasts from 18 studies (1 study used sep-
arate malingering and honest groups for cross-validation
and thus allowed for 2 independent contrasts). For the
overall sample of contrasts, mean sensitivity values were
moderate (M = 61.4%, SD = 20.5) while mean specificity
 308 JASINSKI ET AL.
TABLE 4
Operating characteristics for overall sample and by subscale
Cut score M k Sensitivity (%)
Total sample 6.9 19
RDS 7.1 9 63.3 (17.6)
DS-ACSS 6.8 10
p-value
Note. Analyses conducted on independent contrasts, outlier-parsed dataset. k = number of studies.
RDS = reliable digit span. DS-ACSS = digit span age-corrected scaled score. Standard deviations
in parentheses.
values were moderately high (M = 86.3%, SD = 10.9).
This resulted in an overall hit rate of 76.3% (SD = 7.0).
A closer examination of the diagnostic accuracy sep-
arately for RDS and DS-ACSS showed a mean sensi-
tivity of 63.3% (SD = 17.6) for RDS (N = 9), while
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
specificity was 86.1% (SD = 11.0). The overall hit rate
for RDS was high average (M = 75.7%, SD = 6.6)
at a mean cutting score of 7.1 (SD = 0.3). Similarly,
the mean sensitivity for the DS-ACSS (N = 10) was
adequate (M = 59.7%, SD = 23.7), with high aver-
age specificity (M = 86.5%, SD = 11.5) and hit rate
(M = 76.8%, SD = 7.6) at a mean cutting score of
6.8 (SD = 1.9). Overall, there was no significant dif-
ference between the measures on sensitivity (t = 0.376,
p = .712), specificity (t = −0.075, p = .941), or hit
rate (t = −0.345, p = .734). These findings again
suggest that either variant can be used effectively in
classifying malingering. However, it is noteworthy that
the cutting score for RDS was much less variable as
indicated by a small standard deviation, suggesting
more consistency across studies in recommended cutting
scores.
DISCUSSION
This purpose of the present study was to provide a
comprehensive meta-analytic review of the ability of the
Digit Span subtest of the Wechsler Adult Intelligence
Scale (WAIS) and Wechsler Memory Scale (WMS) in
detecting malingering, particularly in forensic settings
where simply assuming honesty is not acceptable. More
specifically, the ability of the Reliable Digit Span (RDS)
variant (Greiffenstein et al., 1994) and the Digit Span
age-corrected scaled score variant (DS-ACSS) in sepa-
rating and accurately classifying malingering and hon-
est groups was examined across all studies in the area
through March 2009. The WAIS and WMS are two
of the most frequently utilized neuropsychological tests,
and the examination of embedded malingering indices
within these measures has the potential to allow clin-
icians to verify honesty in patients without adding
unnecessary testing time to a battery. Further, retro-
spective analyses of effort can be conducted when no
malingering tests were given, and embedded indices on
standard neuropsychological tests have been hypothe-
sized to be less susceptible to coaching (Mathias et al.,
2002).
Specificity (%) Hit rate (%)
61.4 (20.5) 86.3 (10.9) 76.3 (7.0)
86.1 (11.0) 75.7 (6.6)
59.7 (23.7) 86.5 (11.5) 76.8 (7.6)
.712 .941 .734
Summary of results: Effect size analysis
Overall, results of the present study indicate that the
Digit Span test is sensitive to malingering and on aver-
age differentiates malingering and honest groups by
more than one pooled standard deviation (d = 1.25,
95% CI = 1.13–1.37). Similarly, both the RDS and
DS-ACSS variants produced large effect sizes (RDS:
d = 1.34, 95% CI = 1.18–1.50; DS-ACSS: d = 1.08,
95% CI = 1.01–1.50). These effect sizes are similar to
the means reported in a recent meta-analysis of sev-
eral dedicated malingering tests, suggesting that the Digit
Span may be a viable alternative to those more time-
consuming tests (Vickery et al., 2001). A file drawer anal-
ysis suggested that 244 null studies would need to exist
to render the effect insignificant, an unlikely prospect.
Homogeneity testing revealed consistency among the
effect sizes and variability consistent with what is
expected given sampling error. Results such as these can
be considered positive, with the observed consistency in
the effect sizes resulting in increased confidence in the
use of the scale, particularly in forensic settings. These
results aid in the Digit Span test meeting Daubert stan-
dards of admissibility in court (Daubert v. Merrell Dow
Pharmaceuticals, 1993).
Further, the effect sizes derived from studies on the
Digit Span subtest were found to be similar in magnitude
across the variants (RDS vs. DS-ACSS) or test battery
(WAIS vs. WMS) utilized. Overall, although the effect
sizes for the RDS and WAIS–R showed significant vari-
ability, there were no significant differences between the
two variants or between any of the test versions in their
ability to separate malingering and honest groups. Effect
sizes from both variants and all test versions were large.
These results are promising for clinicians who may choose
to use either the WAIS or the WMS, depending on the
referral question, indicating that the Digit Span subtest
from either battery could be used interchangeably to ade-
quately detect malingering. Further, should a legal issue
arise after an evaluation has been completed, a statement
regarding the presence or absence of malingering can be
made using the Digit Span results in conjunction with
other malingering indices, given the high likelihood that
either a WAIS or a WMS was administered during the
original testing. For researchers, retrospective data analy-
ses are possible, and results from various test versions can
be combined to produce larger sample sizes and hence

more reliable outcomes. Further, with the recent release
of the newest versions of the WAIS–IV (Wechsler, 2008)
and WMS–IV (Wechsler, 2009), Digit Span research and
application to malingering detection is supported and
based on these results should generalize to the nearly
identical forms of the subtest in these new test versions.
Moreover, the impact of study methodology on the
effect sizes was explored, again with promising results.
For instance, homogeneity was found among effect sizes
derived from both enhanced simulation and known-
groups studies, implying that both methodologies are
justifiable approaches to determining a test’s ability to
classify malingering. This is consistent with Rogers’s
(1997) recommendation that both designs are vital to test
validation, although it is in contrast with the theoreti-
cal notion that known-groups designs are typically the
stronger methodology. Moreover, it is in contrast with
Rogers and Cavanaugh’s (1983) criticism of simulation
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
studies, specifically that they involve generalizing results
from individuals asked to comply with instructions to
fake to patients who fake when asked to comply. The
findings here are promising for research using enhanced
simulation studies conducted with the recommended
methodological controls (Rogers, 2008), indicating they
are ecologically valid and may provide a reliable estimate
of a test’s ability to detect malingering in clinical settings.
Additional moderators were hypothesized and
explored, as they can provide some insight into method-
ological characteristics that may be important despite
overall consistency in the derived effect sizes. Further,
significant heterogeneity was found for the subset of RDS
effect sizes as well as those derived from the WAIS–R,
suggesting that moderator variables could aid in explain-
ing such variability. Hypothesized moderators included
the use of incentives for effort, the use of a warning to
feign believably in enhanced simulation studies, and the
use of objective criteria in classifying feigning groups
within the known-groups methodology. Each resulted
in significant moderation of the overall effect sizes and
for RDS and WAIS–R effect sizes where they could be
analyzed, indicating that such methodological controls
should be routinely applied in malingering studies. The
use of coaching on symptoms or feigning indices, as well
as the effect of preparation time, could not be analyzed
due to lack of relevant studies.
With regard to the use of incentives, the present results
are limited in that the use of incentives was coded as a
dichotomy, due to inconsistency in how the incentives
were awarded and the small number of studies included
in these analyses (n = 3). In looking at these studies, the
incentive amount ranged from $20 per study participant
to a $100 drawing for a single participant. The results of
the moderator analyses showed that effect size increased
for those offered an incentive, and this was true when eye-
balling the amount of incentive. The study providing the
largest incentive (Iverson & Franzen, 1996) also showed
the largest effect size (d = 2.46). However, intuition sug-
gests that a larger incentive is associated with a more
ecologically valid study and less blatant malingering, thus
reducing the differences between the feigning and hon-
est groups. Prior research in this area has been mixed,
DIGIT SPAN AND MALINGERING DETECTION 309
with Orey, Cragar, and Berry (2000) finding no effect
of incentive, while Rogers and Cruise (1998) reported a
main effect of incentive on an inventory for psychiatric
malingering. It may be the case that incentive has a differ-
ential effect based on the type of feigning (neurocognitive
vs. psychiatric) or based on the type of test adminis-
tered (e.g., self-report vs. oral administration); however,
this has yet to be examined in the literature. Given the
very small sample size, the current results should be
interpreted with extreme caution, but raise an interesting
point for further exploration.
Another methodological characteristic thought to have
an impact on the ecological validity of a study is the
use of a warning to feign believably. While the practice
of providing a warning regarding malingering instru-
ments is not consistent in clinical practice (Slick, Tan,
Strauss, & Hultsch, 2004), Rogers (1997) strongly rec-
ommends using a warning in simulation research. Again,
little research has focused on this specific aspect of simu-
lation methodology, but intuition suggests that inclusion
of a warning might result in smaller group differences, or
more credible feigning. This was the case in the present
study, although again the results should be interpreted
with extreme caution given the majority of studies uti-
lized such a warning.
Finally, the use of objective, valid measures for classify-
ing the malingerers in a known-groups study significantly
moderated the effect sizes in the present meta-analysis.
Specifically, those studies employing well-validated dedi-
cated tests of malingering as well as using the Slick et al.
(1999) criteria to aid in the determination of group status
showed smaller effect sizes, while those not utilizing such
criteria had larger and more variable differences. The use
of objective criteria is now considered standard practice
in known-groups studies, as it increases the internal valid-
ity of the study. These results are not surprising, but are
in need of further exploration with a larger number of
studies.
Summary of results: Diagnostic accuracy
statistics
Overall, the Digit Span test showed strong specificity
(M = 86.3%), with moderate sensitivity (M = 61.4%)
and overall hit rate (M = 76.3%). More specifically, the
RDS and DS-ACSS variants also showed strong speci-
ficity (M = 86.1%; M = 86.5%, respectively) with good
sensitivity (M = 63.3%; M = 59.7%, respectively). There
were no significant differences in the diagnostic accuracy
of the RDS and DS-ACSS, indicating that clinicians may
choose the one best suited to their needs. Although RDS
showed slightly better consistency across recommended
cutting scores, the DS-ACSS is calculated automatically
when scoring the WAIS and is conceptually easier to
understand, making it the likely choice by clinicians.
Further, DS-ACSS utilizes a standardized score and is
not calculated as a simple sum of digits forward and
backward as RDS is, so may be more difficult for the
sophisticated malingerer to gauge the amount of feigning
required to produce a given standard score.
 310 JASINSKI ET AL.
Limitations
The primary shortcoming of the present review is the
small number of studies available for analysis. More
specifically, the limited number of studies across ver-
sions of the WAIS and WMS batteries precluded in-depth
exploration among these tests. Thus, the results of those
analyses should be considered tentative and an area in
need of further research, particularly in light of the
new test versions being released. Further, the effects of
coaching and preparation time are potentially important
moderators that could not be explored here.
Additionally, although the Digit Span subtest may
hold promise as a screening instrument, the present study
was not able to explore this issue thoroughly based on
a limited range of cutting scores presented to maxi-
mize sensitivity, a requirement for using either RDS or
DS-ACSS as a screening tool for malingering. Future
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
research should focus on publishing the full range of
cutting scores and associated sensitivity and specificity
values, to allow clinicians and researchers to further
examine this potentially time-saving diagnostic approach.
In summary, the present review provided evidence to
support the use of the Digit Span subtest, including
RDS and DS-ACSS, in adequately detecting malingering.
The Digit Span subtest performs similarly to many well-
validated dedicated malingering tests. The advantages of
embedded indices, in addition to the large effect sizes
and strong diagnostic accuracy statistics, make the Digit
Span test an essential tool for clinicians and researchers
alike.
Original manuscript received 11 May 2010
Revised manuscript accepted 28 July 2010
REFERENCES
Note: References marked with an asterisk designate studies incl-
uded in the meta-analysis.
Allen, L. M., & Green, P. (2001). Declining CARB failure rates
over 6 years of testing: What’s wrong with this picture?
Archives of Clinical Neuropsychology, 16, 846.
American Psychiatric Association. (2000). Diagnostic and sta-
tistical manual of mental disorders (4th ed., text rev.).
Washington, DC: Author.
Arnold, G., Boone, K. B., Lu, P., Dean, A., Wen, J., Nitch,
S., et al. (2005). Sensitivity and specificity of finger-tapping
test scores for the detection of suspect effort. The Clinical
Neuropsychologist, 19, 105–120.
∗ Axelrod, B. N., Fichtenberg, N. L, & Millis, S. R. (2006).
Detecting incomplete effort with Digit Span from the
Wechsler Adult Intelligence Scale–Third Edition. The
Clinical Neuropsychologist, 20, 513–523.
∗ Babikian, T., Boone, K. B., Lu, P., & Arnold, G. (2006).
Sensitivity and specificity of various digit span scores in the
detection of suspect effort. The Clinical Neuropsychologist,
20, 145–159.
Bauer, L., & McCaffrey, R. J. (2006). Coverage of the Test of
Memory Malingering, Victoria Symptom Validity Test, and
Word Memory Test on the Internet: Is test security threat-
ened? Archives of Clinical Neuropsychology, 21, 121–126.
Bernard, L. C., McGrath, M. J., & Houston, W. (1993).
Discriminating between simulated malingering and closed
head injury on the Wechsler Memory Scale–Revised.
Archives of Clinical Neuropsychology, 8, 539–551.
Bernard, L. C., McGrath, M. J., & Houston, W. (1996). The dif-
ferential effects of simulating malingering closed head injury,
and other CNS pathology on the Wisconsin Card Sorting
Test: Support for the “pattern of performance hypothesis”.
Archives of Clinical Neuropsychology, 11, 231–245.
Berry, D. T. R., Baer, R. A., Rinaldo, J. C., & Wetter, M. W.
(2002). Assessment of malingering. In J. N. Butcher (Ed.),
Clinical personality assessment: Practical approaches. New
York, NY: Oxford University Press.
Berry, D. T. R., & Schipper, L. J. (2007). Detection of feigned
psychiatric symptoms during forensic neuropsychological
evaluations. In G. J. Larrabee (Ed.), Assessment of malin-
gered neuropsychological deficits (pp. 226–263). New York,
NY: Oxford University Press.
Binder, L. M. (2002). The Portland Digit Recognition Test:
A review of validation and clinical use. Journal of Forensic
Neuropsychology, 2, 27–41.
Binder, L. M., Villanueva, M. R., Howieson, D., & Moore, R.
T. (1993). The Rey AVLT recognition memory task measures
motivational impairment after mild head trauma. Archives of
Clinical Neuropsychology, 8, 137–147.
Bush, S. S., Ruff, R. M., Troster, A. I., Barth, J. T., Koffler, S. P.,
Pliskin, N. H., et al. (2005). Symptom validity assessment:
Practical issues and medical necessity. NAN position paper.
Archives of Clinical Neuropsychology, 20, 419–426.
Camara, W., Nathan, J. S., & Puente, A. E. (2000).
Psychological test usage: Implications in professional
psychology. Psychological Testing and Assessment, 31,
141–154.
∗ Chafetz, M. D. (2008). Malingering on the social security dis-
ability consultative exam: Predictors and base rates. The
Clinical Neuropsychologist, 22, 529–546.
Cohen, J. (1988). Statistical power analysis for the behavioral sci-
ences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.
Daubert v. Merrell Dow Pharmaceuticals, 113 S. Ct. 2876 (1993).
∗ Duncan, S. A., & Ausborn, D. L. (2002). The use of Reliable
Digits to detect malingering in a criminal forensic pretrial
population. Assessment, 9, 56–61.
∗ Etherton, J. L., Bianchini, K. J., Ciota, M. A., Heinly, M. T.,
& Greve, K. W. (2006). Pain, malingering and the WAIS–III
Working Memory Index. The Spine Journal, 6, 61–71.
∗ Etherton, J. L., Bianchini, K. J., Greve, K. W., & Heinly, M.
T. (2005). Sensitivity and specificity of Reliable Digit Span in
malingered pain-related disability. Assessment, 12, 130–136.
Frederick, R. I., & Crosby, R. D. (2000). Development and val-
idation of the Validity Indicator Profile. Law and Human
Behavior, 24, 59–82.
Frederick, R. I., Crosby, R. D., & Wynkoop, T. F. (2000).
Performance curve classification of invalid responding
on the Validity Indicator Profile. Archives of Clinical
Neuropsychology, 15, 281–300.
Green, P., Allen, L. M., & Astner, I. K. (1997). The Word
Memory Test: A user’s guide to the oral and computer-
administered forms. Durham, NC: CogniSyst.
∗ Greiffenstein, M., Baker, W. J., & Gola, T. (1994). Validation
of malingered amnesia measures with a large clinical sample.
Psychological Assessment, 6, 218–224.
∗ Greiffenstein, M., Gola, T., & Baker, W. J. (1995). MMPI-2
validity scales versus domain specific measures in detec-
tion of factitious traumatic brain injury. The Clinical
Neuropsychologist, 3, 230–240.
Greve, K. W., & Bianchini, K. J. (2004). Setting empirical
cut-offs on psychometric indicators of negative response
bias: A methodological commentary with recommendations.
Archives of Clinical Neuropsychology, 19, 533–541.
Grote, C. L., & Hook, J. N. (2007). Forced-choice recognition
tests of malingering. In G. J. Larrabee (Ed.), Assessment of
malingered neuropsychological deficits (pp. 44–79). New York,
NY: Oxford University Press.
Gutheil, T. G. (2003). Reflections on coaching by attorneys.
Journal of American Academy of Psychiatry and Law, 31, 6–9.
∗ Heaton, R. K., Smith, H. H., Lehman, R. A. W., & Vogt, A.
T. (1978). Prospects for faking believable deficits on neu-
ropsychological testing. Journal of Consulting and Clinical
Psychology, 46, 892–900.

Hedges, L., & Olkin, I. (1985). Statistical methods for meta-
analysis. New York, NY: Academic Press.
∗ Heinly, M. T., Greve, K. W., Bianchini, K. J., Love, J. M.,
& Brennan, A. (2005). WAIS Digit Span-based indicators
of malingered neurocognitive dysfunction. Assessment, 12,
429–444.
∗ Inman, T. H., & Berry, D. T. R. (2002). Cross-validation
of indicators of malingering: A comparison of nine neu-
ropsychological tests, four tests of malingering, and behav-
ioral observations. Archives of Clinical Neuropsychology, 17,
1–23.
Inman, T. H., Vickery, C. D., Berry, D. T. R., Lamb, D.,
Edwards, C., & Smith, G. T. (1998). Development and ini-
tial validation of a new procedure for evaluating adequacy
of effort given during neuropsychological testing: The Letter
Memory Test. Psychological Assessment, 10, 128–139.
∗ Iverson, G. L., & Franzen, M. D. (1994). The Recognition
Memory Test, Digit Span, and Knox Cube Test as markers
of malingered memory impairment. Assessment, 1, 323–334.
∗ Iverson, G. L., & Franzen, M. D. (1996). Using multiple objec-
tive memory procedures to detect simulated malingering.
Journal of Clinical and Experimental Neuropsychology, 18,
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
38–51.
Iverson, G. L., & McCracken, L. M. (1997). “Postconcussive”
symptoms in persons with chronic pain. Brain Injury, 11,
783–790.
Johnson, B. T. (1989). DSTAT: Software for the meta-analytic
review of research literatures. Hillside, NJ: Lawrence Erlbaum
Associates.
Killgore, W. D., & DellaPietra, L. (2000). Using the WMS–
III to detect malingering: Empirical validation of the rarely
missed index (RMI). Journal of Clinical and Experimental
Neuropsychology, 22, 761–771.
∗ Langeluddecke, P. M., & Lucas, S. K. (2003). Quantitative
measures of memory malingering on the Wechsler Memory
Scale–Third edition in mild head injury litigants. Archives of
Clinical Neuropsychology, 18, 181–197.
∗ Larrabee, G. J. (2003). Detection of malingering using atypical
performance patterns on standard neuropsychological tests.
Clinical Neuropsychologist, 17, 410–425.
Larrabee, G. J. (2007). Assessment of malingered neuropsycholog-
ical deficits. New York, NY: Oxford University Press.
Larrabee, G. J., & Berry, D. T. R. (2007). Diagnostic clas-
sification statistics and diagnostic validity of malingering
assessment. In G. J. Larrabee (Ed.), Assessment of malingered
neurocognitive deficits (pp. 14–26). New York, NY: Oxford
University Press.
Lees-Haley, P. R. (1997). Attorneys influence expert evidence in
forensic psychology and neuropsychology cases. Assessment,
4, 321–324.
Lipsey, M. W., & Wilson, B. D. (2001). Practical meta-analysis.
Thousand Oaks, CA: Sage Publications.
∗ Martin, R. C., Hayes, J. S., & Gouvier, W. D. (1996).
Differential vulnerability between postconcussion self-report
and objective malingering tests in identifying simulated
mild head injury. Journal of Clinical and Experimental
Neuropsychology, 18, 265–275.
∗ Mathias, C. W., Greve, K. W., Bianchini, K. B., Houston, R. J.,
& Crouch, J. A. (2002). Detecting malingered neurocognitive
dysfunction using the Reliable Digit Span in traumatic brain
injury. Assessment, 9, 301–308.
Meyers, J. E., & Volbrecht, M. (1999). Detection of malinger-
ers using the Rey Complex Figure and Recognition Trial.
Applied Neuropsychology, 6, 201–207.
Millis, S. R., Putnam, S. H., Adams, K. M., & Ricker, J. H.
(1995). The California Verbal Learning Test in the detec-
tion of incomplete effort in neuropsychological evaluation.
Psychological Assessment, 7, 463–471.
∗ Mittenberg, W., Azrin, R., Millsaps, C., & Heilbronner, R.
(1993). Identification of malingered head injury on the
Wechsler Memory Scale–Revised. Psychological Assessment,
5, 34–40.
Mittenberg, W., Patton, C., Canyock, E. M., & Condit, D. C.
(2002). Base rates of malingering and symptom exaggeration.
DIGIT SPAN AND MALINGERING DETECTION 311
Journal of Clinical and Experimental Neuropsychology, 24,
1094–1102.
Mittenberg, W., Patton, C., & Legler, W. (October, 2003).
Identification of malingered head injury on the Wechsler
Memory Scale–Third Edition. Paper presented at the annual
conference of the National Academy of Neuropsychology,
Dallas, TX.
Mittenberg, W., Rotholc, A., Russell, E., & Heilbronner,
R. (1996). Identification of malingered head injury
on the Halstead–Reitan Battery. Archives of Clinical
Neuropsychology, 11, 271–281.
∗ Mittenberg, W., Theroux, S., Aguila-Puentes, G., Bianchini, K.,
Greve, K., & Rayls, K. (2001). Identification of malingered
head injury on the Wechsler Adult Intelligence Scale–Third
Edition. The Clinical Neuropsychologist, 15, 440–445.
∗ Mittenberg, W., Theroux-Fichera, S., Zielinski, R. E., &
Heilbronner, R. L. (1995). Identification of malingered
head injury on the Wechsler Adult Intelligence Scale–
Revised. Professional Psychology: Research and Practice, 26,
491–498.
Orey, C., Cragar, D. E., & Berry, D. T. R. (2000). The effects
of two motivational manipulations on the neuropsycholog-
ical performance of mildly head-injured college students.
Archives of Clinical Neuropsychology, 15, 335–348.
Public attitude monitor [Survey]. (1992). Oak Brook, IL:
Insurance Research Council.
Rabin, L. A., Barr, W. B., & Burton, L. A. (2005). Assessment
practices of clinical neuropsychologists in the United States
and Canada: A survey of INS, NAN, and APA Division 40
members. Archives of Clinical Neuropsychology, 20, 33–65.
Rogers, R. (1997). Clinical assessment of malingering and decep-
tion (2nd ed.). New York, NY: Guilford Press.
Rogers, R. (2008). Researching response styles. In R. Rogers
(Ed.), Clinical assessment of malingering and deception (3rd
ed., pp. 411–434). New York: Guilford Press.
Rogers, R., & Cavanaugh, J. L. (1983). “Nothing but the truth”
. . . a reexamination of malingering. Journal of Psychiatry and
Law, 11, 443–459.
Rogers, R., & Cruise, K. R. (1998). Assessment of malingering
with simulation designs: Threats to external validity. Law and
Human Behavior, 22, 273–285.
Rosenthal, R. (1991). Meta-analytic procedures for social
research (Rev. ed.). Newbury Park, CA: Sage.
Ruiz, M. A., Drake, E. B., Glass, A., Marcotte, D., & van
Gorp, W. G. (2002). Trying to beat the system: Misuse of the
Internet to assist in avoiding the detection of psychological
symptom dissimulation. Professional Psychology: Research
and Practice, 33, 294–299.
Slick, D. J., Hopp, G., Strauss, E., & Thompson, G.
(1997). Victoria Symptom Validity Test: Professional manual.
Odessa, FL: Psychological Assessment Resources.
Slick, D. J., Sherman, E. M. S., & Iverson, G. L. (1999).
Diagnostic criteria for malingered neurocognitive dysfunc-
tion: Proposed standards for clinical practice and research.
Clinical Neuropsychologist, 13, 545–561.
Slick, D. J., Tan, J. E., Strauss, E. H., & Hultsch, D. F.
(2004). Detecting malingering: A survey of experts’ practices.
Archives of Clinical Neuropsychology, 19, 465–473.
∗ Strauss, E., Slick, D. J., Levy-Bencheton, J., Hunter,
M., MacDonald, S. W. S., & Hultsch, D. F. (2002).
Intraindividual variability as an indicator of malingering
in head injury. Archives of Clinical Neuropsychology, 17,
423–444.
Suhr, J. A., & Gunstad, J. (2007). Coaching and malingering:
A review. In G. J. Larrabee (Ed.), Assessment of malingered
neuropsychological deficits (pp. 287–311). New York, NY:
Oxford University Press.
∗ Suhr, J., Tranel, D., Wefel, J., & Barrash, J. (1997). Memory
performance after head injury: Contributions of malinger-
ing, litigation status, psychological factors, and medication
use. Journal of Clinical and Experimental Neuropsychology,
4, 500–514.
Tombaugh, T. N. (1996). Test of Memory Malingering
(TOMM). New York, NY: Multi-Health Systems.
 312 JASINSKI ET AL.
∗ Trueblood, W. (1994). Qualitative and quantitative character-
istics of malingered and other invalid WAIS–R and clin-
ical memory data. Journal of Clinical and Experimental
Neuropsychology, 4, 597–607.
∗ Trueblood, W., & Schmidt, M. (1993). Malingering and other
validity considerations in the neuropsychological evaluation
of mild head injury. Journal of Clinical and Experimental
Neuropsychology, 15, 578–590.
Tukey, J. (1977). Exploratory data analysis. Reading, MA:
Addison-Wesley.
Vickery, C. D., Berry, D. T. R., Inman, T. H., Harris, M. J., &
Orey, S. A. (2001). Detection of inadequate effort on neu-
ropsychological testing: A meta-analytic review of selected
procedures. Archives of Clinical Neuropsychology, 16, 45–73.
Wechsler, D. (1981). Wechsler Adult Intelligence Scale–Revised
manual. New York, NY: The Psychological Corporation.
Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013
Wechsler, D. (1987). Wechsler Memory Scale–Revised manual.
San Antonio, TX: The Psychological Corporation.
Wechsler, D. (1997a). WAIS–III: Administration and scoring
manual. San Antonio, TX: The Psychological Corporation.
Wechsler, D. (1997b). Wechsler Memory Scale III:
Administration and scoring manual. San Antonio, TX:
The Psychological Corporation.
Wechsler, D. (2008). WAIS–IV: Administration and scoring
manual. San Antonio, TX: The Psychological Corpor
ation.
Wechsler, D. (2009). WMS–IV: Administration and scoring man-
ual. San Antonio, TX: The Psychological Corporation.
Wetter, M. W., & Corrigan, S. K. (1995). Providing informa-
tion to clients about psychological tests: A survey of attor-
neys’ and law students’ attitudes. Professional Psychology:
Research and Practice, 26, 474–477.
 Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013

APPENDIX
Demographic and methodological characteristics, effect sizes, and classification accuracy of studies included in independent contrasts analyses

Age Ed.
(years) (years) Method Test Characteristics

Study Comp type Grp N Subj. M SD M SD Coa Prep Inc Bel. MS Subtest M SD CS SN SP HR d

Axelrod (2006)a KG M 36 TBI 43.1 11.9 12.5 1.6 n/a n/a n/a n/a .86 DS 6.3 2.1 ≤7 .75 .72 1.05
H 29 TBI 38.2 17.8 12.8 2.1 8.5 2.1 .69
Babikian (2006)a KG M 66 MixedN 42.5 12.3 13.0 2.4 n/a n/a n/a n/a .86 RDS 6.7 2.4 ≤7 .62 .69 0.99
H 56 MixedN 48.4 16.4 13.3 2.4 8.9 2.0 .77
Chafetz (2008)a KG M 16 Dis. 24.1 8.5 10.2 1.5 n/a n/a n/a n/a .43 RDS 4.2 2.4 NA NA NA NA 2.18
H 21 Dis. 30.2 8.9 9.8 1.6 8.0 1.0
KG M 7 Dis. 25.4 11.0 7.1 3.2 n/a n/a n/a n/a .43 RDS 2.9 1.9 NA NA NA NA 3.46c
H 12 Dis. 40.9 12.7 9.8 1.4 7.7 1.0
Duncan (2002)a KG M 54 Crim. 34.7 10.1 8.9 2.2 n/a n/a n/a n/a .57 RDS 5.8 3.4 ≤7 .68 .71 1.19
H 124 Crim. 38.1 9.5 10.5 2.7 8.9 2.1 .72
Etherton (2006)a KG M 32 Pain 43.0 8.5 11.5 3.0 n/a n/a n/a n/a .86 DS 6.4 3.0 ≤7 .75 .75 1.12
H 36 TBI 32.2 14.1 NA NA 9.8 3.1 .75
Sim M 20 Stud. NA NA NA NA N N N Y .38 DS 5.9 2.7 ≤7 .55 .71 1.43
H 49 Pain 43.5 11.9 12.2 2.2 10.0 2.9 .78
Etherton (2005)a KG M 35 Pain 43.3 11.6 NA NA n/a n/a n/a n/a 1.0 RDS 7.2 3.0 ≤7 .60 .79 1.29
H 53 Pain 43.0 8.4 NA NA 10.5 2.3 .92
Greiffenstein (1994)a KG M 43 TBI 39.3 10.0 11.1 2.2 n/a n/a n/a n/a .57 RDS 6.7 1.2 ≤7 .68 .77 1.90
H 30 PPCS 37.6 13.9 11.7 0.8 8.9 1.1 .89
Greiffenstein (1995)a KG M 68 TBI 38.3 11.7 12.2 2.9 n/a n/a n/a n/a .57 RDS 6.6 1.8 ≤7 .89 .80 1.49
H 53 TBI 24.6 12.5 12.5 2.6 9.2 1.7 .68
Heaton (1978)a Sim M 16 Stud/CV 24.4 7.5 12.9 2.4 N N Y N .50 DS 7.0 2.9 NA NA NA NA 0.92
H 16 TBI 26.7 6.5 11.9 1.6 9.5 2.5
Heinly (2005)a KG M 18 TBI NA NA NA NA n/a n/a n/a n/a .86 DS 6.1 2.8 ≤5 .36 .75 1.24
H 45 TBI NA NA NA NA 10.5 3.8 .92
Inman (2002)a Sim M 21 TBI 18.7 0.9 12.5 1.0 N N Y Y .75 RDS 9.2 2.3 <8 .27 .65 1.02
H 24 TBI 18.7 1.7 12.3 0.6 11.6 2.4 1.0
Iverson (1994)a Sim M 40 Stud/Inm 28 6.9 14 2.0 N N N Y .50 DS 2.7 1.7 <5 .90 .90 2.97c
H 20 TBI 28 7.5 13 1.9 8.3 2.3 .90

313
 Downloaded by [University Of Pittsburgh] at 02:44 16 June 2013

314
Iverson (1996)a Sim M 20 Psych. 20.0 1.7 13.7 0.7 N N Y Y .50 DS 2.6 1.4 <4 .78 .91 2.46
H 20 MemImp 34.0 13.2 13.6 2.3 8.2 2.8 1.0
Langeluddecke (2003)b KG M 25 TBI 33.9 12.0 10.8 2.1 n/a n/a n/a n/a 1.0 DS 11.6 3.6 ≤8 .16 .76 1.27
H 50 TBI 38.4 11.3 11.1 3.3 16.3 3.9 .98
Larrabee (2003)a KG M 24 TBI 39.3 11.8 12.5 2.3 n/a n/a n/a n/a .71 RDS 7.4 1.7 ≤7 .50 .73 1.27
H 12 TBI 34.8 16.8 12.6 2.6 10.0 2.2 .94
Martin (1996)a Sim M 51 Stud./TBI 22.1 5.5 14.0 1.4 Y N N N .44 DS 11.3 6.3 ≤ 11 .35 .68 1.15
H 70 TBI 21.6 5.5 14.6 1.3 16.4 3.4 1.0
Mathias (2002)a KG M 24 TBI 38.9 12.3 12.5 2.8 n/a n/a n/a n/a 1.0 RDS 6.8 1.5 ≤7 .67 .82 1.83
H 30 TBI 33.6 13.5 12.5 2.4 10.1 2.0 .93
Mittenberg (1993)a Sim M 39 CV 32.8 12.4 15.5 2.3 N N N Y .56 DS 11.2 3.9 NA NA NA NA 1.02
H 39 TBI 32.9 11.9 13.9 3.0 15.3 4.1
Mittenberg (1995)a Sim M 67 CV 33.7 12.2 15.7 1.9 N N N Y .56 DS 6.9 3.1 NA NA NA NA 1.02
H 67 TBI 33.2 11.3 14.0 3.0 9.7 2.3
Mittenberg (2001)a KG M 36 TBI 42.9 16.0 11.0 3.4 n/a n/a n/a n/a .43 DS 6.1 2.2 NA NA NA NA 1.21
H 36 TBI 30.3 12.9 11.9 1.5 9.5 3.3
Strauss (2002)a Sim M 11 TBI NA NA NA NA N N N Y .56 RDS 5.9 4.1 <7 .79 .85 1.27
H 16 TBI NA NA NA NA 10.3 2.9 .90
Suhr (1997)a KG M 31 TBI 37.6 8.7 12.7 2.2 n/a n/a n/a n/a 1.0 DS 6.3 2.4 NA NA NA NA 1.02
H 20 TBI 35.4 9.5 13.1 2.9 9.0 3.0
Trueblood (1994)a KG M 12 TBI 34.9 NA 11.6 NA n/a n/a n/a n/a .86 DS 4.8 NA <7 .75 .75 NA
H 12 TBI 35.0 NA 11.3 NA 8.3 NA .75
Trueblood (1993)a KG M 8 TBI 37.4 NA 11.6 NA n/a n/a n/a n/a .86 DS 5.9 NA ≤7 .62 .75 NA
H 8 TBI 37.4 NA 12.0 NA 8.9 NA .88

Note. For studies, only the first author is listed. n/a = not applicable; NA = information not available. Comp type = comparison type (KG = known groups; Sim =
enhanced simulation). Grp = group (M = malingering; H = honest). N = number of subjects. Subj. = subject sample (Crim. = criminal pretrial defendants, CV =
community volunteers, Dis. = disability claimants, Inm = Inmates, MemImp = memory impaired, MixedN = mixed neurological, Pain = chronic pain, PPCS = persistent
postconcussive syndrome, Psych. = psychiatric patients, Stud. = students, TBI = traumatic brain injury). Age (M = mean, SD = standard deviation). Ed. = education
level. Method (Coa = coaching provided on either symptoms or avoiding detection, Prep = preparation time given, Inc = incentive given, Bel. = warning to be believable;
N = not present, Y = included in methodology). MS = relative method score. Test scale (RDS = reliable digit span, DS = digit span age-corrected scaled score; M = mean
test score, SD = standard deviation, CS = cutting score, SN = sensitivity, SP = specificity, HR = hit rate, d = corrected Hedges’s g effect size).
a Study using a version of the Wechsler Adult Intelligence Scale (WAIS).
b Study using a version of the Wechsler Memory Scale (WMS).
c Study excluded as an outlier.