Verbal fluency deficits in Parkinson’s disease:
A meta-analysis
JULIE D. HENRY and JOHN R. CRAWFORD
School of Psychology, University of Aberdeen, Scotland
(Received July 17, 2003; Revised December 12, 2003; Accepted January 27, 2004)
Abstract
A meta-analysis of 68 studies with a total of 4644 participants was conducted to investigate the sensitivity of tests
of verbal fluency to the presence of Parkinson’s disease (PD) relative to healthy controls. Both phonemic and
semantic fluency were moderately impaired but neither deficit qualified as a differential deficit relative to verbal
intelligence or psychomotor speed. However, PD patients were significantly more impaired on semantic relative to
phonemic fluency (rs 5 .37 vs. .33, respectively), and confrontation naming, a test of semantic memory that
imposes only minimal demands upon cognitive speed and effortful retrieval, was associated with a deficit that was
of a comparable magnitude to the deficits upon each of these types of fluency. Thus, the disorder appears to be
associated with particular problems with semantic memory. Tests that impose heavy demands upon switching may
also be disproportionately affected. Demented and non-demented PD patients differ quantitatively but not
qualitatively in terms of the relative prominence of deficits on tests of phonemic and semantic fluency. However,
patients with dementia of the Alzheimer’s type and demented PD patients can be differentiated from one another by
the relative magnitude of deficits upon these two measures. (JINS, 2004, 10, 608–622.)
Keywords: Parkinson’s, Fluency, Semantic memory, Switching
INTRODUCTION
In addition to motor abnormalities, it has long been recog-
nised that Parkinson’s disease (PD) is also associated with
a number of cognitive deficits, of which it has been sug-
gested that executive dysfunction is particularly prominent
(Della Sala, 1988). Executive functioning is considered to
be responsible for the more complex, or supervisory as-
pects of cognition such as self-directed planning and strat-
egy formation, future-orientated, goal-directed and non-
habitual behavior (Crawford & Henry, in press; Phillips,
1997; Shallice, 1988; Stuss & Benson, 1986). It has been
argued that PD is associated with a number of specific ex-
ecutive deficits, including problems with effortful process-
ing (Weingartner et al., 1984), the use of internal attentional
cues (Brown & Marsden, 1988), cognitive set-shifting (Zec
et al., 1999), and self-directed strategy formation (Taylor
et al., 1986). Moreover, Della Sala (1988) has suggested
that executive dysfunction can account for all of the cogni-
tive deficits associated with non-demented PD.
Consistent with this possibility, it has been found that
patients with PD often perform poorly on tests designed to
capture executive dysfunction, including the Wisconsin Card
Sorting Test (WCST; Gotham et al., 1988; Tsai et al., 1994)
tests of verbal fluency (Auriacombe et al., 1993; Cooper
et al., 1991; Flowers et al., 1995; Matison et al., 1982) and
the Stroop interference test (Hanes et al., 1996a). More-
over, neuropathologically, PD is associated with neuronal
loss in the substantia nigra that leads to dopamine depletion
in the nigro-striatal projection. This, in turn, leads to func-
tional abnormalities in the subcorticofrontal circuits (De
Long & Georgopolis, 1981). Since there is a great deal of
evidence that executive processes rely heavily upon the in-
tact functions of frontal structures (see, e.g., Shallice, 1988;
Stuss & Benson, 1986) the presence of frontal abnormali-
ties would therefore suggest that deficits in this aspect of
cognition should be especially marked.
However, patients with PD may be impaired upon virtu-
ally all measures of cognitive function, and this includes
tests presumed to make only minimal demands on execu-
tive processes. As Miller (1984) and others (Crawford &
Henry, in press; Laws, 1999) have pointed out, a deficit on
an executive measure is not by itself sufficient to infer the
Reprint requests to: Julie D. Henry, School of Psychology, King’s Col-
lege, University of Aberdeen, AB24 3HN. E-mail: j.d.henry@abdn.ac.uk
Journal of the International Neuropsychological Society (2004), 10, 608–622.
Copyright © 2004 INS. Published by Cambridge University Press. Printed in the USA.
DOI: 10.10170S1355617704104141
608
https://doi.org/10.1017/S1355617704104141
Downloaded from https://www.cambridge.org/core. UQ Library, on 10 Aug 2017 at 02:54:27, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
presence of a differential executive deficit; instead, it must
be shown that the executive deficit is in excess of the aver-
aged performance deficit across a range of other cognitive
tasks that are not considered to impose heavy executive
demands. Moreover, it has been shown that putative mea-
sures of executive functioning such as the WCST, Stroop
interference task and verbal fluency can be dissociated in
PD (Gurd, 1995; Van Spaendonck et al., 1996), and thus, it
may be that only certain aspects of executive functioning
are differentially impaired.
Verbal Fluency Performance in PD
In an attempt to resolve whether PD is associated with a
differential executive deficit, verbal fluency performance
has been studied extensively. Tests of verbal fluency re-
quire time-restricted generation of multiple response alter-
natives under constrained search conditions, and involve
associative exploration and retrieval of words based on pho-
nemic or semantic criteria (known as phonemic or letter
fluency, and semantic or category fluency, respectively).
Both measures are thought to require efficient organisation
of verbal retrieval and recall, as well as self-monitoring
aspects of cognition (the participant must keep track of
responses already given), effortful self-initiation, and inhi-
bition of responses when appropriate (Crawford & Henry,
in press; Perret, 1974; Phillips, 1997; Ruff et al., 1997).
However, whilst some studies have reported significant
deficits on measures of phonemic fluency in non-demented
PD (Azuma et al., 1997; Flowers et al., 1995) others have
failed to do so (Auriacombe et al., 1993; Caltagirone et al.,
1989; Goldman et al., 1998; Ivory et al., 1999; Levin et al.,
1989; Matison et al., 1982; Miller, 1985). Indeed, at least
one study reported that PD patients performed better than
their respective control group on this task (Taylor et al.,
1986). Preserved semantic fluency has also been reported
(Gabrieli et al., 1996; Troyer et al., 1998) but the most
consistent finding is impaired performance (Auriacombe
et al., 1993; Cooper et al., 1991; Flowers et al., 1995; Ma-
tison et al., 1982). Moreover, a number of studies have
reported significant deficits on semantic, but not phonemic
fluency (Auriacombe et al., 1993; Matison et al., 1982).
A number of different explanations have been proposed
to explain why patients with PD are impaired on tests of
verbal fluency. Flowers et al. (1995) for instance, have ar-
gued that the verbal fluency deficit associated with PD re-
flects a mental bradyphrenia that parallels patients’ motor
bradykinesia. Flowers et al. (1995) found that non-demented
PD patients were impaired on both semantic and phonemic
fluency, but that verbal intelligence alone could not account
for poor performance, and other word-production charac-
teristics did not differ from healthy controls.
However, in a meta-analytic review of the verbal fluency
performance of patients with focal cortical lesions, Henry
and Crawford (2004) found that focal frontal lobe injuries
were associated with equivalent phonemic and semantic
fluency deficits (rs 5 .52 and .54 respectively) suggesting
that phonemic and semantic fluency impose comparable
demands upon executive processes. It may therefore be that
a pattern of comparable impairment upon tests of phonemic
and semantic fluency for patients with PD reflects execu-
tive dysfunction. However, to support this hypothesis, it
would also be necessary to demonstrate that for patients
with PD, as for frontal patients (but not for non-frontal
patients), the deficit in verbal fluency qualifies as a differ-
ential deficit relative to current VIQ and psychomotor speed.
Although Flowers et al. (1995) found that the phonemic
and semantic fluency deficits could not be explained by
level of VIQ, no assessment was made of cognitive speed,
and thus it is not clear whether comparable deficits for the
two measures reflects the presence of a bradyphrenia as
Flowers et al. (1995) suggested, or executive dysfunction.
However, as noted earlier, many studies have found that
semantic fluency is more impaired than phonemic fluency,
and this finding has typically been attributed to the pres-
ence of a specific deficit in semantic memory. Consistent
with this interpretation, whilst Henry and Crawford (2004)
found that focal frontal injuries were associated with equiv-
alent phonemic and semantic fluency deficits, semantic flu-
ency was more impaired following focal temporal damage
(r 5 .61), and the deficit was significantly larger than the
corresponding phonemic fluency deficit (r 5 .44). Since
there is a great deal of evidence that temporal structures are
the neural substrates particularly responsible for semantic
memory, this was presumed to reflect the greater reliance of
semantic fluency upon the integrity of semantic memory.
Thus, Raskin et al. (1992a), for instance, found that non-
demented PD patients did not differ from healthy controls
on a test of phonemic fluency, but were significantly im-
paired on a test of semantic fluency in which specific cues
were provided. Raskin et al. (1992a) suggested that PD
patients possess intact storage systems, but that there may
be a specific deficit in the retrieval of semantic informa-
tion. Auriacombe et al. (1993) also found that semantic but
not phonemic fluency was significantly impaired in PD.
However, this was attributed to a more specific retrieval
deficit in accessing the verbal labels, or phonological shapes,
associated with category exemplars; i.e., not a problem with
the retrieval of semantic information per se, but in lexical
retrieval.
It has also been suggested that PD is associated with a
specific deficit in the executive control mechanisms respon-
sible for the consecutive inhibition and disinhibition of al-
gorithms. This derives from the fact that some studies have
reported that measures of verbal fluency in which partici-
pants must alternate or shift between naming exemplars
that belong to more than one different category, or accord-
ing to more than one type of phonemic criteria, are also
impaired. However, whilst performance on the WCST is
often disrupted (Gotham et al., 1988; Lees & Smith, 1983;
Tsai et al., 1994), deficits on tests of alternating fluency
may not be disproportionate to those associated with the
single fluency condition (Cooper et al., 1991; Gurd, 1995).
Downes et al. (1993) argue that an important factor when
Fluency in PD 609
https://doi.org/10.1017/S1355617704104141
Downloaded from https://www.cambridge.org/core. UQ Library, on 10 Aug 2017 at 02:54:27, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
assessing alternating fluency performance is whether intra-
or extra-dimensional shifting is required, as it was found
that whilst PD patients’ ability to alternate between probes
of the same domain (i.e., semantic–semantic, or phonemic–
phonemic) was intact, generation of exemplars from
different fluency domains (i.e., phonemic–semantic) was
selectively compromised.
Thus, different researchers have advanced different inter-
pretations of the nature of the cognitive impairment associ-
ated with PD, and that is presumed to underlie deficits on
tests of verbal fluency, and this at least partially reflects
discrepancies between studies in terms of the relative mag-
nitude of deficits on different cognitive measures. It has
been suggested that such discrepancies may be attributable
to substantive differences between studies in terms of the
PD patients sampled, and in particular, PD is associated
with a high incidence of dementia.
Verbal Fluency Performance
in Demented PD
It remains unclear whether dementia in PD is analogous to
dementia of the Alzheimer’s type (DAT), or reflects a
clinical0neuropsychological syndrome that Albert (1978)
has named subcortical dementia. It has been claimed that
cortical dementias such as DAT are typified by a pattern of
worse semantic relative to phonemic fluency performance,
and subcortical dementias by the opposite deficit profile.
However, whilst research involving patients with DAT, Hun-
tington’s disease and progressive supranuclear palsy has
found evidence consistent with this distinction (Hodges et al.,
1990; Rosser & Hodges, 1994), the cortical–subcortical dis-
sociation has not been consistently upheld. Suhr and Jones
(1998) for instance, found the pattern of semantic and pho-
nemic fluency deficits to be comparable for patients with
Alzheimer’s, Huntington’s and Parkinson’s dementias. Ques-
tions also remain with respect to the relationship between
cognitive deficits in demented and non-demented PD.Azuma
et al. (1997) found that as mental state decreases, PD is
associated with a reduced ability to use sub-category struc-
ture to facilitate retrieval, suggesting that there is a quali-
tative difference, and that semantic fluency should be
disproportionately impaired relative to phonemic fluency
as the dementia progresses. However, this hypothesis has
not yet been rigorously tested.
Aims
To the present authors’ knowledge, the current paper is the
first to apply meta-analytic techniques to compare perfor-
mance upon tests of phonemic and semantic fluency in PD.
One of the most important advantages of this methodology
is that corrections can be implemented for sampling error,
and thus it will be possible to assess whether discrepancies
between studies reflect the influence of substantive factors
such as dementia status, or artifactual variance. In addition,
using meta-analysis an effect’s generalisability can be sub-
jected to a level of scrutiny not possible in a single study,
and with a level of objectivity and methodological consis-
tency that is difficult to achieve in non-quantitative reviews
(Stanley, 2001).
The first aim was to derive effect size estimates for pho-
nemic and semantic fluency for patients with PD relative to
healthy controls. Comparison of the relative magnitude of
each will help to resolve the inconsistencies noted in the
literature, and permit an assessment of whether the verbal
fluency deficit associated with PD predominantly reflects
executive dysfunction, or problems with semantic memory
(Henry & Crawford, 2004).
However, as noted, the presence of a deficit on a test of
phonemic or semantic fluency does not by itself provide
evidence of executive or semantic memory dysfunction, re-
spectively. Thus, the second aim was to estimate effect sizes
for other cognitive measures in order to assess to what ex-
tent fluency deficits in PD qualify as differential deficits.
Premorbid intelligence as estimated by the NationalAdult
Reading Test (NART; Nelson, 1982) and the reading sub-
test of the Wide Range Achievement Test (WRAT; Jastak &
Wilkinson, 1984) was included to address the possibility
that if a phonemic fluency deficit is present, it reflects the
fact that PD patients have not been successfully matched to
their controls for premorbid ability. However, of particular
importance was to address the possibility that phonemic
and semantic fluency deficits simply reflect a current gen-
eral impairment in verbal abilities (see Miller, 1984). Thus,
the pattern of deficits across fluency versus verbal intelli-
gence as measured by the WAIS (Wechsler, 1955; 1981)
Verbal and Vocabulary scales (VIQ) will be compared.
We will also assess whether deficits on tests of phonemic
and semantic fluency are in excess of deficits on the WAIS
Digit Symbol test (Wechsler, 1955; 1981), a widely used
measure of psychomotor speed (Salthouse, 1992). This will
address the possibility that deficits on tests of verbal flu-
ency simply reflect generalised slowing rather than execu-
tive dysfunction. Performance on tests of phonemic and
semantic fluency will also be compared with the Boston
Naming Test (BNT; Kaplan et al., 1983), a measure of se-
mantic memory that imposes only minimal demands upon
effortful retrieval and cognitive speed.
The third aim is to compare the magnitude of phonemic
and semantic fluency deficits with performance on tests of
alternating fluency. Effect sizes will also be calculated for
the number of categories completed and perseverative er-
rors upon the Wisconsin Card Sorting Test (WCST CC and
WCST PE respectively; Heaton, 1981) as this measure also
imposes demands upon cognitive set-shifting (Miyake et al.,
2000).
A fourth issue relates to whether the relative magnitude
of the deficits for phonemic and semantic fluency are com-
parable for demented and non-demented sub-groups, and
for demented PD patients relative to patients with DAT. The
magnitude of these deficits will therefore be quantified for
each of these sub-groups. Data for patients with DAT will
610 J.D. Henry and J.R. Crawford
https://doi.org/10.1017/S1355617704104141
Downloaded from https://www.cambridge.org/core. UQ Library, on 10 Aug 2017 at 02:54:27, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
be taken from an independent meta-analysis (Henry et al.,
in press). PD patients were only permitted to contribute to
the ‘demented’ or ‘non-demented’ analyses where the de-
mentia status of the patients was specifically indicated in
the study.
METHODS
Sample of Studies
Acomputer-based search involving the Web of Science, Psych
Lit CD-ROM, and Science Direct databases was under-
taken, using the following terms as search parameters;
letter fluency, FA S , semantic fluency, category fluency, con-
trolled oral word association, COWA(T), word fluency ver-
bal fluency, oral fluency, phonemic fluency, executive test,
and frontal test. In addition, a manual search of most
issues of the journals Journal of the International Neuro-
psychological Society, Brain, Neuropsychology, Clinical
Neuropsychologist, Neuropsychologia, Neuropsychiatry,
Neuropsychology and Behavioural Neurology, Journal of
Neuropsychiatry and Clinical Neurosciences, and the Jour-
nal of Clinical and Experimental Neuropsychology was con-
ducted. These journals were selected as they were considered
to be the most relevant to the current area of research (i.e.,
neuropsychological deficits in Parkinson’s disease). The fact
that most but not all issues were searched manually unfor-
tunately reflects the very real problem that the libraries ac-
cessible to us had only incomplete collections of certain
journals, and in particular very early copies of certain jour-
nals were often not available.Asystematic method of search
through these journals was adopted, with every page checked
for references to measures of verbal fluency. The search
was completed in October 2002.
The inclusion criteria were (1) the patient group had to
consist entirely of adults with PD; (2) the study had to
include a healthy control group free from neurological or
psychiatric disease; and (3) a measure of phonemic, seman-
tic, intra- or extra-alternating fluency. Effect size estimates
for premorbid IQ, current VIQ, Digit Symbol, BNT, WCST
CC and WCST PE were derived from studies that also re-
ported verbal fluency results. For inclusion, the study must
also have presented precise statistics convertible to effect
size r (i.e., the M and SD for the patient and control group
separately, or precise statistical test results, F, t, or Z). Since
an effect size expresses a directional relationship, only sta-
tistical test results based on 1 degree of freedom could be
used to derive effect sizes (Rosenthal, 1994). Imprecise
statistical test results were also not included (i.e., where it
was simply stated that p , .05 or p , .01, etc.; Le Bras
et al., 1999; Oyebode et al., 1986). Finally, studies had to
have been published in English in a journal.
Statistical Analysis
Meta-analysis is a rigorous, quantitative alternative to the
traditional review process, as it involves statistical integra-
tion of results. The basis of this methodology is the effect
size, a standardised statistic that quantifies the magnitude
of an effect. Two basic types of metric exist that can be used
to quantify effect size, known as the r- and the d-families.
Although mathematically equivalent, they are associated
with different interpretations of what the effect size repre-
sents. Whilst exemplars of the r family characterise the
degree of correlation between two variables, e.g. the point-
biserial correlation between group membership (i.e., pres-
ence or absence of PD), and the variable of interest (i.e.,
performance on the cognitive measure of interest), d family
members exemplify this relationship in terms of the stan-
dardized difference between these two variables calibrated
in terms of the standard deviation. As a consequence of its
greater generality of interpretation, consistency of meaning
and more salient practical meaning, r is the more useful
effect size estimate (see Rosenthal & DiMatteo, 2001), and
thus, in the present study this effect size was employed.
It should be noted that because the correlation coefficient
is associated with a slight bias, Fisher (1928) derived a
transformation of r that Snedecor and Cochran (1989) have
recommended should be employed during statistical analy-
ses in preference to r. However, this transformed estimate
is itself associated with a bias, and in a Monte Carlo analy-
sis, Field (2001) reported that for random effects meta-
analytic models, transformed effect-size estimates produced
substantial upward biases of a larger magnitude than the
corresponding downward biases associated with untrans-
formed correlation coefficients. Thus, in the present study,
untransformed correlation coefficients have been employed
for statistical analyses.
For each construct, effects were pooled to derive an es-
timate of the mean, with each effect weighted for sample
size to correct for sampling error. To do so, the random
effects meta-analytic model was selected in preference to
the more commonly employed fixed effects model as it
yields more generalisable parameter estimates. This is be-
cause, in the fixed effects model, the mean is presumed to
reflect a common underlying effect parameter that gives
rise to the sample observations. However, in the random
effects model the mean represents a hyperparameter, as it
allows for substantive differences beyond sampling error
that differentiate the effects contributing to each respective
mean (Raudenbush, 1994).
Statistically, the crucial difference between these meth-
odologies is in the calculation of standard errors and confi-
dence intervals, which for the random effects model are
typically larger. The National Research Council (1992) ar-
gues that the fixed effects model should be the exception
rather than the rule, as it may lead to inappropriately strong
conclusions. Thus, although more technically demanding,
it was considered important to use the random effects model
in the present work.
To estimate the degree of heterogeneity of the effects
contributing to each mean, the homogeneity statistic Q and
the random effects variance ~s
u
2
) were estimated, as well as
the SD of random effects, and the 95% confidence intervals
Fluency in PD 611
https://doi.org/10.1017/S1355617704104141
Downloaded from https://www.cambridge.org/core. UQ Library, on 10 Aug 2017 at 02:54:27, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
(CI) within which random effects can be expected to fall. Q
quantifies within-group heterogeneity (i.e. the degree to
which the studies contributing to each respective mean can
be regarded as homogenous). If the Q statistic associated
with a mean effect is significant, this suggests that there are
substantive differences between the studies contributing to
that particular mean. In contrast, a non-significant estimate
of Q suggests that once sampling error has been removed,
no substantive differences between the studies contributing
to the respective mean in question remain (i.e. the null hy-
pothesis of homogeneity of effects cannot be rejected).
It was also important to test whether the difference in the
magnitude of mean effects between, for instance, phonemic
versus semantic fluency, was statistically significant. How-
ever, there is no agreed method for statistically comparing
mean effects using the random effects meta-analytic model.
A particular difficulty is whether the degrees of freedom
(df ) in such analyses should be based on N (the number of
participants) or K (the number of studies). In the present
work, a relatively large number of studies were included,
and therefore, t tests were computed using the more conser-
vative K as the df.
Since dementia status will moderate the magnitude of
deficits across individual studies, for each statistical com-
parison, only studies that assessed both variables of interest
were included. For example, although in total 80 PD groups
were tested on phonemic fluency, and 66 PD groups on
semantic fluency, since only 50 groups were assessed on
both phonemic and semantic fluency, when conducting in-
ferential statistics to compare phonemic and semantic flu-
ency, only data from these 50 groups were permitted to
contribute to the analyses. This ensured that the partici-
pants being compared upon the two measures were equated
for dementia severity (i.e., it is exactly the same partici-
pants being compared upon each of these measures).
It should also be noted that because the same participants
were compared upon each measure, paired t tests were em-
ployed for all statistical comparisons. Mean effects were
also calculated for each of the non-fluency variables iden-
tified (premorbid IQ, current VIQ, Digit Symbol, BNT,
WCST CC and WCST PE) and compared with the corre-
sponding effects for phonemic and semantic fluency. Again,
to ensure that dementia severity was controlled for, only
studies that assessed both the fluency and non-fluency vari-
able of interest were included in each comparison.
Finally, the null hypothesis that the mean effect size is
zero was tested with the statistic Z; if the value of Z exceeds
1.96, this indicates that the mean effect differs significantly
from zero at the .05 level. To interpret how important a
particular effect was in practical terms, Cohen’s (1977)
guidelines were adopted. These suggest that a correlation of
.1 should be regarded as representing a small effect, .3 as
medium, and .5 as large. In addition, squares of the effect
size multiplied by 100 were also presented as these latter
quantities represent the percentage of the variance ac-
counted for (PVAF) by group membership (i.e., the pres-
ence of PD versus being a member of the healthy adult
population) on a measure of interest. It should be noted that
for inferential statistics comparisons were made using the
PVAF by group membership upon each of the measures of
interest because the difference between effect sizes is non-
linear as r increases and thus PVAF is the more appropriate
index when comparing variables.
RESULTS
Participant Characteristics
Sixty-eight studies published between 1983 and 2002 met
the inclusion criteria specified, and in total, data from 2644
PD patients and 2000 controls contributed to these analy-
ses. References for the 68 studies included in this meta-
analysis are provided in theAppendix. Patients and controls
did not differ significantly in terms of age (M 5 65.01,
SD 5 6.97 vs. M 5 63.16, SD 5 8.13 respectively) or edu-
cation (M 5 12.90, SD 5 2.14 vs. M 5 12.80, SD 5 2.23,
respectively). However, a significantly higher proportion
of the patient group were male (63.27% vs. 47.85% male,
respectively, p , .001). For patients with PD, the mean
Hoehn and Yahr (1967) score, an index of disease severity
that categorises level of disability according to stages be-
tween 1 (mild disability)and5(complete invalidism), was
2.32 (SD 5 0.52). The mean duration of illness was 5.66
(SD 5 2.83) years.
Effect Sizes for Patients With PD
Relative to Healthy Controls
Table 1 presents estimates of the mean effects for phonemic
and semantic fluency, their variability, and practical impor-
tance in terms of the PVAF for studies that include both of
these measures. In addition, mean effects are presented for
premorbid IQ, current VIQ, Digit Symbol, BNT, WCST CC
and WCST PE, calculated using only those studies that in-
cluded the particular non-fluency measure of interest in
addition to phonemic or semantic fluency. As noted previ-
ously, this methodology ensures that exactly the same par-
ticipants are contributing to the mean effects for the two
variables of interest. This is particularly important given
that, as expected, the magnitude of the deficits for phone-
mic and semantic fluency in terms of the PVAF were both
substantially and significantly related to dementia severity
as measured by the Mini Mental State Examination (MMSE;
Folstein et al., 1975); phonemic: r 52.77, K5 41, p ,.001,
semantic: r 52.67, K 5 29, p , .001.
Thus, it can be seen in Table 1 that for each non-fluency
measure, for instance premorbid IQ, two mean effects have
been calculated; one for studies that also assess phonemic
fluency (r 5 .14; K 5 19), and one for studies that also
assess semantic fluency (r 5 .08, K 5 17). Each fluency
mean effect was also re-calculated for these comparisons.
For comparisons with premorbid IQ, current VIQ, Digit
Symbol, BNT, WCST CC and WCST PE the mean effects
612 J.D. Henry and J.R. Crawford
https://doi.org/10.1017/S1355617704104141
Downloaded from https://www.cambridge.org/core. UQ Library, on 10 Aug 2017 at 02:54:27, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
