Content uploaded by Michael J Kane
Author content
All content in this area was uploaded by Michael J Kane
Content may be subject to copyright.
Working Memory Capacity and Fluid Intelligence Are Strongly Related
Constructs: Comment on Ackerman, Beier, and Boyle (2005)
Michael J. Kane
University of North Carolina at Greensboro
David Z. Hambrick
Michigan State University
Andrew R. A. Conway
University of Illinois at Chicago
The authors agree with P. L. Ackerman, M. E. Beier, and M. O. Boyle (2005) that working memory
capacity (WMC) is not isomorphic with general fluid intelligence (Gf) or reasoning ability. However, the
WMC and Gf/reasoning constructs are more strongly associated than Ackerman et al. (2005) indicate,
particularly when considering the outcomes of latent-variable studies. The authors’ reanalysis of 14 such
data sets from 10 published studies, representing more than 3,100 young-adult subjects, suggests a strong
correlation between WMC and Gf/reasoning factors (median r ⫽ .72), indicating that the WMC and Gf
constructs share approximately 50% of their variance. This comment also clarifies the authors’ “exec-
utive attention” view of WMC, it demonstrates that WMC has greater discriminant validity than
Ackerman et al. (2005) implied, and it suggests some future directions and challenges for the scientific
study of the convergence of WMC, attention control, and intelligence.
Theorists have recently speculated that individual differences in
working memory capacity (WMC) may explain reasoning ability,
general intelligence (Spearman’s g), or both (e.g., Engle, 2002;
Kyllonen & Christal, 1990; Su¨ß, Oberauer, Wittmann, Wilhelm, &
Schulze, 2002). In response to these claims, Ackerman, Beier, and
Boyle (2005) conducted a critical review and meta-analysis of the
literature and rejected the notion that WMC is isomorphic to g,
general fluid intelligence (Gf), or any other cognitive ability.
Indeed, they remained generally skeptical that WMC research
would significantly advance either the understanding or measure-
ment of human intelligence. Perhaps it is surprising, then, that we
find several points of agreement with Ackerman et al. (2005),
despite our position that WMC and Gf are closely related and that
exploring this relationship will yield important theoretical insights
into the nature of normal cognitive variation (e.g., Conway, Kane,
& Engle, 2003; Engle & Kane, 2004; Engle, Kane, & Tuholski,
1999; Engle, Tuholski, Laughlin, & Conway, 1999; Kane & Engle,
2002).
We concur with Ackerman et al. (2005) on three main conclu-
sions: (a) WMC is not equivalent to g, Gf, or reasoning ability; (b)
WMC is primarily a domain-general construct; (c) WMC is more
closely related to Gf and reasoning than is short-term memory
(STM). The latter two arguments are now well supported by
large-sample, latent-variable studies. WMC tasks correlate
strongly with each other and with a wide range of cognitive
abilities regardless of their surface characteristics, and WMC and
Gf share substantial variance that is independent of STM (for
reviews, see Daneman & Merikle, 1996; Kane et al., 2004). Re-
garding the central issue of a WMC–intelligence isomorphism,
Ackerman et al.’s (2005) meta-analysis suggests that individual
tests of WMC and Gf (or reasoning) share only about 20% of their
variance, on average, and so these constructs are not synonymous
(see also Ackerman, Beier, & Boyle, 2002). We have similarly
argued in recent reviews of latent-variable studies that the corre-
lations between WMC and Gf constructs do not approach 1.0, even
when both are measured broadly, and so speculations about iso-
morphism are falsified (Conway et al., 2003; Kane et al., 2004).
Nonetheless, we also find major points of disagreement with
Ackerman et al. (2005) and discuss them below. Namely, when we
consider the data from latent-variable studies, we find that the
WMC–Gf association is much stronger than they suggested. WMC
also shows greater discriminant validity than Ackerman et al.
(2005) admitted, correlating more strongly with some constructs
than with others, and so WMC represents much more than a “crud
factor” (p. 52) in its association with other cognitive abilities.
Finally, the link between WMC and attention control is potentially
important to a complete understanding of intelligence, but it also
presents considerable challenges for future theoretical and empir-
ical work that we discuss—particularly regarding measurement.
To put our critique in proper context, however, we first present an
overview of our theoretical perspective. We also hope that this
Michael J. Kane, Department of Psychology, University of North Caro-
lina at Greensboro; David Z. Hambrick, Department of Psychology, Mich-
igan State University; Andrew R. A. Conway, Department of Psychology,
University of Illinois at Chicago.
Andrew R. A. Conway is now at the Department of Psychology,
Princeton University.
We are grateful to Klaus Oberauer for making the complete correlation
matrix available from Su¨ß et al. (2002).
Correspondence concerning this article should be addressed to Michael
J. Kane, Department of Psychology, University of North Carolina at
Greensboro, P. O. Box 26170, Greensboro, NC 27402-6170. E-mail:
mjkane@uncg.edu
Psychological Bulletin Copyright 2005 by the American Psychological Association
2005, Vol. 131, No. 1, 66 –71 0033-2909/05/$12.00 DOI: 10.1037/0033-2909.131.1.66
66
review clarifies aspects of our position that seem to have promoted
some misunderstanding.
An Executive Attention View of WMC
Our view of WMC derives from correlational, experimental, and
quasi-experimental studies of WMC variation (for more detailed
reviews, see Engle & Kane, 2004; Engle, Kane, & Tuholski,
1999). Much of this research is based on “complex span” tasks,
such as Daneman and Carpenter’s (1980) reading span task, which
present short lists of stimuli for immediate recall. In contrast to
“simple” STM span tasks, the presentation of the memory items is
interpolated with a secondary processing task such as verifying
mathematical equations. The most critical findings for our per-
spective are that individual differences in performance of WMC
tasks predict a variety of higher order cognitive abilities, from
learning to comprehension to reasoning, even when the WMC and
ability tests bear no surface similarity to each other. Moreover,
statistical or experimental control of subjects’ mnemonic strate-
gies, processing skills, or motivation does not eliminate— or even
attenuate—WMC–ability correlations, and so these factors cannot
account for the relationship. STM storage and rehearsal factors
cannot either. WMC and STM tasks yield correlated but separate
factors when factor analyzed, and the WMC factor is more
strongly associated with Gf than is STM.
If the storage, rehearsal, processing, and strategic processes
tapped by WMC tasks do not drive the correlations with ability,
then what does? We have inferred that the residual “secret ingre-
dient” is an attention-control capability that is elicited by WMC
tasks to a greater degree than by STM tasks. To wit, WMC tasks
require subjects to maintain or recover access to stimulus repre-
sentations in the face of (a) proactive interference from prior test
trials and (b) mandatory shifts of conscious focus away from those
memory representations to the secondary processing task. Direct
support for our attention hypothesis comes from findings that
extreme groups of high and low WMC span scorers differ in the
performance of prototypical “attention-control” tasks such as di-
chotic listening, Stroop, antisaccade, and flanker tasks and that
high WMC individuals often can be made to perform like low
WMC individuals by dividing their attention. In short, people with
lower WMC show poorer control over thought and action than do
those with higher WMC by failing more often to prevent or
recover from prepotent responses and by showing slower and less
flexible allocation of visual attention to objects in space.
These findings lead us to the following view: The WMC con-
struct reflects primarily a domain-general attentional capability to
sustain or recover access to (or activation of) representations of
task-relevant stimuli, goals, or response productions, and to control
the influences of interference and competition on goal-directed
thought and behavior. However, one must be mindful that WMC
tasks are complex and multiply determined. They measure some
attention-control capabilities and other capabilities as well (e.g.,
storage, processing skill, mnemonic strategies). Moreover, al-
though WMC tasks are more influenced by attention-control pro-
cesses than by storage and rehearsal, whereas the reverse seems
true for STM tasks, both WMC and STM tasks tap executive and
storage processes to some degree. Finally, the WMC construct and
WMC tasks do not account for, or perfectly measure, the full
extent of possible attention and executive capabilities and pro-
cesses. That is, WMC does not fully explain all manner of atten-
tional or executive variance. Indeed, our labs have discovered
several “attentional” domains that are unrelated to WMC variation
(Kane, Poole, Tuholski, & Engle, 2003). So, when we argue that
WMC is tied to attention control, we mean that WMC’s domain-
general predictive power reflects primarily, but not uniquely, vari-
ance attributable to attention-control capabilities. Most of the
variance that WMC shares with the common abilities underlying
complex cognition, such as reasoning, reflects domain-general
attentional processes of maintenance and control.
The Strength of the WMC–Gf Correlation
Despite Ackerman et al.’s (2005) suggestion of a “jingle fal-
lacy” (p. 51), researchers have not considered WMC a candidate
mechanism of general intellectual ability simply because our the-
ories use language similar to Spearman’s (1927). Rather, pursuit of
a WMC–Gf association led from the facts that nomothetic models
of cognition accord working memory a central role (e.g., Ander-
son, 1993; Meyer & Kieras, 1999) and that WMC variation pre-
dicts cognitive abilities very broadly. As well, computational mod-
els suggest that g is more easily implemented at the domain-free,
stable, architectural level of a system rather than at the domain-
specific, malleable levels (Ohlsson, 1998), and neuroscience data
suggest that common prefrontal cortex circuitry is involved in
working memory, attention control, and novel reasoning (Duncan,
1995; Kane & Engle, 2002). WMC and control processes are
therefore natural candidates for those underlying general intellec-
tual abilities.
Although WMC is not equivalent to Gf or reasoning ability, we
believe that the association between WMC and these constructs is
stronger than Ackerman et al. (2005) suggested. Our argument
hinges on whether one should draw inferences about hypothetical
constructs from individual tasks or, instead, from latent variables
derived from multiple tasks. We think it is unwise to base argu-
ments about the association between underlying constructs on
modest correlations between tasks. Two tasks that reflect, in part,
highly related constructs might nonetheless correlate weakly be-
cause of other sources of variability (i.e., measurement error). In
contrast to Ackerman et al. (2005), then, we place more stock in
findings from latent-variable studies, which eliminate the measure-
ment error that confounds interpretation of individual tasks and
their correlations.
1
Latent-variable procedures, such as confirma
-
tory factor analysis, statistically isolate the shared variance among
tasks that are thought to reflect a hypothetical construct, such as
WMC or Gf. In doing so, they eliminate error variance that is
unique to any one test and yield a purer measure of the construct.
For example, a Gf latent variable based on syllogistic and analog-
ical reasoning tests is free of deduction- and induction-specific
ability; a WMC latent variable based on tests for words and for
arrows is free of content-specific storage. As Ackerman et al.
(2005) argued with respect to g, construct measurement based on
multiple, varied tests is always more valid and reliable than that
based on any individual, multidetermined test, regardless of the
construct in question—including WMC.
1
Ackerman et al. (2005) did point out that latent-variable analyses of
meta-analytic data, which are what they had to work with, are potentially
problematic and should be interpreted with caution.
67
WORKING MEMORY AND FLUID INTELLIGENCE: COMMENT
In Ackerman et al.’s (2005) analyses, individual WMC and
Gf/reasoning tests correlated around .45, indicating 20% shared
variance (see their Table 1, “Average WM” row).
2
In contrast,
latent-variable studies of young adults, using confirmatory factor
analysis or structural equation modeling, generally produce path
estimates of .60 –.80 between WMC and Gf, indicating approxi-
mately 35%– 65% shared variance (Ackerman et al., 2002; Colom,
Rebollo, Palacios, Juan-Espinosa, & Kyllonen, 2004; Conway,
Cowan, Bunting, Therriault, & Minkoff, 2002; Engle, Tuholski, et
al., 1999; Hambrick, 2003; Kane et al., 2004; Kyllonen & Christal,
1990; Mackintosh & Bennett, 2003; Miyake, Friedman, Rettinger,
Shah, & Hegarty, 2001; Su¨ß et al., 2002). Interpreting the
WMC–Gf associations across these studies is difficult, however,
because the models were structured differently from each other
and many included additional constructs in their models (e.g.,
knowledge, speed, STM) that could affect the WMC–Gf path
estimates.
We therefore reanalyzed the data from published latent-variable
studies that tested at least 100 healthy young adults on multiple
measures of both WMC and Gf and that reported enough infor-
mation for us to conduct our own confirmatory factor analyses on
the data (i.e., they reported standard deviations for each measure
along with all the relevant pairwise correlations, which allowed us
to recover the covariance matrices to be used as input for the
analyses). For each data set, we specified two separate factors
(using maximum likelihood estimation) from the WMC and Gf
tests and allowed them to correlate. Table 1 lists the tasks we factor
analyzed from each study. We included all the immediate-memory
tasks labeled working memory tasks by the original authors and all
the novel reasoning tasks originally labeled as Gf or reasoning
tasks (i.e., we dropped tasks that reflected primarily STM or
crystallized abilities). Table 1 also presents the WMC–Gf corre-
lations, which ranged from .41 to 1.00 with a median value of .72.
These correlations, representing 3,168 subjects from 14 different
samples, suggest that WMC and Gf/reasoning constructs share
approximately half of their variance. Moreover, this estimate is
conservative because many of these studies tested undergraduates,
who likely represented restricted ranges of both WMC and Gf. In
addition, some studies administered WMC and Gf tasks in differ-
ent sessions, and so the extent to which WMC and Gf fluctuate
over days also attenuated the correlation. Finally, many of the
studies sampled WMC and Gf tasks with very different band-
widths, using limited types of WMC tasks but a broad variety of Gf
tasks, or vice versa. As Ackerman et al. (2005) noted, bandwidth
asymmetry attenuates correlations between factors (see also Su¨ß et
al., 2002). But even without considering any such attenuation, it is
clear that our estimate of 50% shared variance between WMC and
Gf is considerably greater than the 20% estimate provided by
Ackerman et al. (2005).
Discriminant Validity and the “Crud Factor” Criticism
Psychologists often validate their hypothetical constructs
through a process of triangulation (e.g., Campbell & Fiske, 1959).
For example, if putative tests of Construct A correlate more
strongly with each other than with tests of Construct B, then
Construct A is said to exhibit discriminant validity. Ackerman et
al. (2005) argued that WMC measures “do not show substantial
discriminant validity—meaning that they correlate significantly
and substantially with many different abilities, rather than with one
or two key abilities” (p. 52). We agree that WMC measures
correlate positively with many cognitive variables, but we also
emphasize that discriminant validity is a complex issue with re-
spect to Gf because its marker tests must, by definition, correlate
broadly with many other abilities. If a construct, such as Gf or
WMC, is truly pervasive in complex cognition, then it simply
cannot correlate with only “one or two key abilities.” Instead,
discriminant validity will be indicated if tests of the ostensibly
pervasive construct correlate more strongly with some constructs
than with others.
This is precisely the pattern that WMC shows. With respect to
narrowly defined tasks, we find WMC variation to be unrelated to
(a) delayed memory for supraspan lists in the absence of interfer-
ence (Conway & Engle, 1994; Kane & Engle, 2000), (b) stimulus-
driven visual orienting (Kane, Bleckley, Conway, & Engle, 2001),
(c) identification of masked letters via a novel response mapping
(Kane et al., 2001), (d) subitizing one to three visual items (Tu-
holski, Engle, & Baylis, 2001), and (e) some varieties of task-set
switching and inefficient visual search (Kane et al., 2003).
3
In
contrast, as we noted above, WMC does correlate with attention
and memory tasks provoking substantial interference and conflict,
and so WMC is important to only some basic cognitive capacities.
With respect to more broadly defined abilities, Ackerman et al.’s
(2005) meta-analysis shows that WMC correlates considerably
more strongly with nonverbal reasoning and ECTs than with
2
Note that, by way of contrast, the meta-analytic correlations between
individual STM and Gf/reasoning tasks averaged .30, indicating only 9%
shared variance. This low estimate is consistent with findings from latent-
variable studies (e.g., Conway et al., 2002; Engle, Tuholski, et al., 1999).
However, Ackerman et al.’s (2005) confirmatory factor analyses, depicted
in their Figures 2 and 4, seem to provide an exception to this trend: Their
model for WMC and ability factors yielded a .50 working memory–g
correlation, and a parallel model for STM yielded a .49 STM–g correlation.
The apparent similarity is misleading, however, because the g factors were
defined very differently between the two models. In particular, the g factor
in the WMC analysis (Figure 2) was based primarily on Gf/reasoning
indicators, whereas the g factor in the STM analysis (Figure 4) was based
equally on Gf/reasoning and knowledge-crystallized ability indicators
(note also the different g loadings for the Reasoning factor between the
hierarchical models for the WMC and STM data sets, depicted in their
Figures 1 and 3, respectively). Thus, the strong association here between
STM and g is not particularly surprising because STM factors often have
been observed to correlate with crystallized ability even after controlling
for WMC (e.g., Engle, Tuholski, et al., 1999). Furthermore, WMC and g
demonstrate a stronger relation in the hierarchical factor model depicted in
their Figure 1 (where WMC has a g factor loading of .89) than STM and
g do in the hierarchical model depicted in their Figure 3 (where STM has
a g factor loading of .51). The Ackerman et al. (2005) findings therefore do
not compromise the general view that WMC is more strongly associated
with Gf than is STM.
3
Many of these tasks represent those that Ackerman et al. (2005) would
classify as “elementary cognitive tasks” (ECTs), thus complicating any
interpretation of the WMC–ECT relationship. We suggest that analyses of
ECTs’ attention demands will be critical to understanding their correlations
(see Conway, Kane, & Engle, 1999; but see Ackerman et al., 2002). We
also note that three of the nine tasks classified by Ackerman et al. (2005)
as ECTs were titled “reasoning” tasks by the original authors, and so they
may have had reasonably high Gf loadings. This could explain, in part,
their strong association with WMC.
68
KANE, HAMBRICK, AND CONWAY
perceptual speed or knowledge. Similarly, Su¨ß et al. (2002) found
that WMC correlated more strongly with reasoning than with
creativity, speed, or even memory factors. It appears to us that
Ackerman et al. (2005) contradicted themselves, for in arguing
elsewhere in their article that WMC and reasoning are not isomor-
phic, they relied on Kyllonen and Christal’s (1990) finding that
WMC correlated more strongly with speed than did reasoning
whereas reasoning correlated more strongly with knowledge than
did WMC. We suggest that if this is the case, then WMC demon-
strates discriminant validity. WMC is not uniformly promiscuous
in its associations with other constructs. Coupled with the fact that
WMC and Gf–reasoning constructs share about half of their vari-
ance, WMC represents much more than a crud factor in the study
of cognitive ability.
WMC, Gf, and Attention Control
Our argument so far is that WMC represents a distinct
cognitive-ability construct that is strongly related to Gf and novel
reasoning. In addition, we believe that executive attention pro-
Table 1
Correlations Between WMC and Gf/Reasoning Factors Derived From Confirmatory Factor Analyses of Data From Latent-Variable
Studies With Young Adults
Study WMC tasks Gf/reasoning tasks r(95% CI)
Kyllonen & Christal (1990)
Study 2: n ⫽ 399 ABC numerical assignment, mental
arithmetic, alphabet recoding
Arithmetic reasoning, AB grammatical reasoning,
verbal analogies, arrow grammatical reasoning,
number sets
.91 (.89, .93)
Study 3: n ⫽ 392 Alphabet recoding, ABC21 Arithmetic reasoning, AB grammatical reasoning,
ABCD arrow, diagramming relations,
following instructions, letter sets, necessary
arithmetic operations, nonsense syllogisms
.79 (.75, .82)
Study 4: n ⫽ 562 Alphabet recoding, mental math Arithmetic reasoning, verbal analogies, number
sets, 123 symbol reduction, three term series,
calendar test
.83 (.80, .85)
Engle, Tuholski, et al. (1999; N
⫽ 133)
Operation span, reading span,
counting span, ABCD, keeping
track, secondary memory/
immediate free recall
Raven, Cattell culture fair .60 (.48, .70)
Miyake et al. (2001; N ⫽ 167) Letter rotation, dot matrix Tower of Hanoi, random generation, paper
folding, space relations, cards, flags
.64 (.54, .72)
Ackerman et al. (2002; N ⫽
135)
ABCD order, alpha span, backward
digit span, computation span,
figural-spatial span, spatial span,
word-sentence span
Ravens, number series, problem solving,
necessary facts, paper folding, spatial analogy,
cube comparison
.66 (.55, .75)
Conway et al. (2002; N ⫽ 120) Operation span, reading span,
counting span
Raven, Cattell culture fair .54 (.40, .66)
Su¨ß et al. (2002; N ⫽ 121
a
)
Reading span, computation span,
alpha span, backward digit span,
math span, verbal span, spatial
working memory, spatial short-
term memory, updating
numerical, updating spatial,
spatial coordination, verbal
coordination
Number sequences, letter sequences,
computational reasoning, verbal analogies,
fact/opinion, senseless inferences, syllogisms,
figural analogies, Charkow, Bongard, figure
assembly, surface development
.86 (.81, .90)
Hambrick (2003; N ⫽ 171) Computation span, reading span Raven, Cattell culture fair, abstraction, letter sets .71 (.63, .78)
Mackintosh & Bennett (2003;
N ⫽ 138
b
)
Mental counters, reading span,
spatial span
Raven, mental rotations 1.00
Colom et al. (2004)
Study 1: n ⫽ 198 Mental counters, sentence
verification, line formation
Raven, surface development .86 (.82, .89)
Study 2: n ⫽ 203 Mental counters, sentence
verification, line formation
Surface development, cards, figure classification .73 (.66, .79)
Study 3: n ⫽ 193 Mental counters, sentence
verification, line formation
Surface development, cards, figure classification .41 (.29, .52)
Kane et al. (2004; N ⫽ 236) Operation span, reading span,
counting span, rotation span,
symmetry span, navigation span
Raven, WASI matrix, BETA III matrix, reading
comprehension, verbal analogies, inferences,
nonsense syllogisms, remote associates, paper
folding, surface development, form board,
space relations, rotated blocks
.67 (.59, .73)
Note. WMC ⫽ working memory capacity; Gf ⫽ general fluid intelligence; 95% CI ⫽ the 95% confidence interval around the correlations; WASI ⫽
Wechsler Abbreviated Scale of Intelligence.
a
N with the complete data set available (personal communication, K. Oberauer, July 7, 2004).
b
N for each pairwise correlation ranged from 117 to 127.
69
WORKING MEMORY AND FLUID INTELLIGENCE: COMMENT
cesses mediate this strong WMC–Gf association. This does not
mean that attention control completely explains Gf but only that it
is largely responsible for the shared variance between WMC and
Gf. In making this argument, however, we realize that our evi-
dence is provisional, and we appreciate Ackerman et al. (2005)
pointing out some of its limitations. Our work has demonstrated
significant performance differences between high and low WMC
individuals across a number of different “simple” attention tasks,
but these extreme-groups designs cannot quantify the strength of
the WMC–attention association, nor can they determine whether
WMC-related attention differences correspond to WMC-related Gf
differences. We therefore may be making much theoretical hay out
of a relatively modest correlation that is independent of the
WMC–Gf association. What the field needs now, then, is a latent-
variable approach to the problem, in which many subjects com-
plete many marker tests of WMC, Gf, and attention control. These
studies should report the magnitude of the WMC–attention corre-
lation and examine whether the shared variance between WMC
and attention accounts for substantial Gf variance (and more Gf
variance than is accounted for by residual variance from WMC or
attention constructs).
Our attentional view has empirical support, but it also poses a
significant challenge that was familiar to Spearman (1937, see
especially p. 133), who argued that attention could be as vague a
construct as intelligence. If the examination of a WMC–ability link
is to move beyond a simplistic consideration of tasks (such as
WMC span vs. STM span) and toward a deeper understanding of
underlying constructs, then the attentional/executive contributions
to WMC and Gf must be specified. Quasi-experimental tests for
WMC-related differences in attention tasks are a good place to
begin narrowing the field. An additional tack is to consider the
attentional demands of WMC and Gf tasks themselves. For exam-
ple, Unsworth and Engle (2004) argued that (a) WMC span tasks
require attention control because the processing component dis-
places the memory items from attentional focus, which then must
be retrieved from secondary memory in the face of proactive
interference (see also Miyake, Friedman, & Saito, 2003); (b) if the
focus of attention is limited to 4 ⫾ 1 items (Cowan, 2001), then
STM span tasks presenting lists of 4 or more items also require
some secondary-memory retrieval and its attendant control pro-
cesses. In support of these arguments, Unsworth and Engle factor
analyzed two verbal WMC tasks (with list lengths 2–5) and two
STM span tasks (with list lengths 2–7), which yielded three fac-
tors: STM lengths 2– 4, STM lengths 4 –7, and WMC lengths 2–5.
The latter two factors correlated .67 (vs. .39 between WMC and
STM 2– 4), suggesting that WMC span and long STM lists mea-
sured a similar construct. Moreover, recall on WMC list lengths 2,
3, 4, and 5 all correlated equivalently with Gf, but only perfor-
mance on the long STM lists correlated with Gf as strongly.
Finally, the variance shared by WMC span and long STM lists
accounted for 18% of the variance in Gf, and each accounted for
only 5% unique variance. These findings reinforce the point that
immediate memory tasks cannot be dichotomized as reflecting
either STM or WMC; all these tasks rely on storage and attentional
processes. The challenge is in assessing their respective
contributions.
A related challenge to investigating the WMC–Gf association is
constructing reliable WMC tasks that maximize executive-
attention demands. We think that the tasks should require main-
taining rapid access to information that is at least momentarily
outside the limited, conscious focus of attention. (This would also
seem relevant to successful reasoning.) Dual-task procedures pro-
vide one such methodology— but not the only one. Ackerman et
al. (2005) claimed that “nearly all” WMC tasks are dual tasks and
complained that “far too little information is available that pro-
vides an account of how overall performance should be assessed”
(p. 50). We disagree on both counts. First, many WMC tasks
included in their meta-analysis are not dual tasks (e.g., Kyllonen &
Christal, 1990; Su¨ß et al., 2002), and several nondual-task mea-
sures of immediate memory correlate strongly with Gf, such as
spatial STM tasks (Miyake et al., 2001), coordination and trans-
formation tasks (Su¨ß et al., 2002), and verbal STM tasks with long
lists (Unsworth & Engle, 2004). Second, the problem of a dual-
task assessment of WMC is ameliorated when a latent variable is
derived from both dual and nondual tasks; indeed, the WMC–Gf
correlation does not seem to vary much whether one obtains the
WMC construct from dual tasks or from a mixture of dual tasks
and nondual tasks (e.g., Su¨ß et al., 2002). An important goal for
research in this area, then, will be to clarify the shared processes
among dual- and nondual-task measures of the WMC construct
and to determine whether our idea that they reflect an attention-
control capability holds up.
As a final methodological note, we consider Ackerman et al.’s
(2005) argument that WMC span tasks are problematic because
they are typically scored only for the storage component of the task
and thus people who differ in accuracy on the processing compo-
nent by 5% or more are indistinguishable. In fact, it is extremely
rare for individuals to differ this much on the processing compo-
nent of a WMC span task because performance is generally at
ceiling. For example, during the Spring of 2003, 420 students at
the University of Illinois at Chicago completed two verbal WMC
tasks, and the mean numbers of processing errors, out of 42 total
trials, were only 1.2 (SD ⫽ 1.5) and 1.3 (SD ⫽ 1.5), respectively.
One might be tempted to argue from this that the time to perform
the processing component may be more important, or more sen-
sitive to trade-off, than accuracy. However, statistically or exper-
imentally controlling time on the processing component does not
diminish the correlation between WMC span and complex cogni-
tion, and it sometimes increases it (Conway & Engle, 1996; Engle,
Cantor, & Carullo, 1992; Friedman & Miyake, 2004). Thus, the
correlation between WMC and ability is not impacted by the fact
that many WMC tests are dual tasks, and so we think that the norm
of scoring only the storage component of WMC span tasks is well
justified. Future experimental and correlational research will de-
termine whether our attentional view of WMC and Gf is equally
well justified.
References
Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2002). Individual differ-
ences in working memory within a nomological network of cognitive
and perceptual speed abilities. Journal of Experimental Psychology:
General, 131, 567–589.
Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working memory
and intelligence: The same or different constructs? Psychological Bul-
letin, 131, 30 –60.
Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant
70
KANE, HAMBRICK, AND CONWAY
validation by the multitrait–multimethod matrix. Psychological Bulletin,
56, 81–105.
Colom, R., Rebollo, I., Palacios, A., Juan-Espinosa, M., & Kyllonen, P. C.
(2004). Working memory is (almost) perfectly predicted by g. Intelli-
gence, 32, 277–296.
Conway, A. R. A., Cowan, N., Bunting, M. F., Therriault, D., & Minkoff,
S. (2002). A latent variable analysis of working memory capacity,
short-term memory capacity, processing speed, and general fluid intel-
ligence. Intelligence, 30, 163–183.
Conway, A. R. A., & Engle, R. W. (1994). Working memory and retrieval:
A resource-dependent inhibition model. Journal of Experimental Psy-
chology: General, 123, 354 –373.
Conway, A. R. A., & Engle, R. W. (1996). Individual differences in
working memory capacity: More evidence for a general capacity theory.
Memory, 4, 577–590.
Conway, A. R. A., Kane, M. J., & Engle, R. W. (1999). Is Spearman’s G
determined by speed or working memory capacity? [Review of the book
The G factor: The science of mental ability]. Psycoloquy, 10(74). Re-
trieved from http://psycprints.ecs.soton.ac.uk/archive/00000709/
Conway, A. R. A., Kane, M. J., & Engle, R. W. (2003). Working memory
capacity and its relation to general intelligence. Trends in Cognitive
Sciences, 7, 547–552.
Cowan, N. (2001). The magical number 4 in short-term memory: A
reconsideration of mental storage capacity. Behavioral and Brain Sci-
ences, 24, 87–185.
Daneman, M., & Carpenter, P. A. (1980). Individual differences in working
memory and reading. Journal of Verbal Learning and Verbal Behavior,
19, 450 –466.
Daneman, M., & Merikle, P. M. (1996). Working memory and language
comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3,
422– 433.
Duncan, J. (1995). Attention, intelligence, and the frontal lobes. In M.
Gazzaniga (Ed.), The cognitive neurosciences (pp. 721–733). Cam-
bridge, MA: MIT Press.
Engle, R. W. (2002). Working memory capacity as executive attention.
Current Directions in Psychological Science, 11, 19 –23.
Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in
working memory and comprehension: A test of four hypotheses. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 18,
972–992.
Engle, R. W., & Kane, M. J. (2004). Executive attention, working memory
capacity, and a two-factor theory of cognitive control. In B. Ross (Ed.),
The psychology of learning and motivation (pp. 145–199). New York:
Academic Press.
Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999). Individual differ-
ences in working memory capacity and what they tell us about controlled
attention, general fluid intelligence and functions of the prefrontal cor-
tex. In A. Miyake & P. Shah (Eds.), Models of working memory:
Mechanisms of active maintenance and executive control (pp. 102–134).
New York: Cambridge University Press.
Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. A. (1999).
Working memory, short-term memory, and general fluid intelligence: A
latent variable approach. Journal of Experimental Psychology: General,
128, 309 –331.
Friedman, N. P., & Miyake, A. (2004). The reading span test and its
predictive power for reading comprehension ability. Journal of Memory
and Language, 51, 136 –158.
Hambrick, D. Z. (2003). Why are some people more knowledgeable than
others? A longitudinal study of language acquisition. Memory & Cog-
nition, 31, 902–917.
Kane, M. J., Bleckley, M. K., Conway, A. R. A., & Engle, R. W. (2001).
A controlled-attention view of working-memory capacity. Journal of
Experimental Psychology: General, 130, 169 –183.
Kane, M. J., & Engle, R. W. (2000). Working memory capacity, proactive
interference, and divided attention: Limits on long-term memory re-
trieval. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 26, 333–358.
Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex in
working-memory capacity, executive attention, and general fluid intel-
ligence: An individual-differences perspective. Psychonomic Bulletin &
Review, 9, 637– 671.
Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W.,
& Engle, R. W. (2004). The generality of working memory capacity: A
latent-variable approach to verbal and visuospatial memory span and
reasoning. Journal of Experimental Psychology: General, 133, 189 –
217.
Kane, M. J., Poole, B. J., Tuholski, S. W., & Engle, R. W. (2003,
November). Working memory capacity and executive control in search
and switching. Paper presented at the annual meeting of the Psy-
chonomic Society, Vancouver, British Columbia, Canada.
Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more
than) working-memory capacity?! Intelligence, 14, 389 – 433.
Mackintosh, N. J., & Bennett, E. S. (2003). The fractionation of working
memory maps onto different components of intelligence. Intelligence,
31, 519 –531.
Meyer, D. E., & Kieras, D. E. (1999). Pre´cis to a practical unified theory
of cognition and action: Some lessons learned from EPIC computational
models of human multiple-task performance. In D. Gopher & A. Koriat
(Eds.), Attention and performance XVII: Cognitive regulation of perfor-
mance: Interaction of theory and application (pp. 17– 88). Cambridge,
MA: MIT Press.
Miyake, A., Friedman, N. P., Rettinger, D. A., Shah, P., & Hegarty, M.
(2001). How are visuospatial working memory, executive functioning,
and spatial abilities related? A latent-variable analysis. Journal of Ex-
perimental Psychology: General, 130, 621– 640.
Miyake, A., Friedman, N. P., & Saito, S. (2003, July). Individual differ-
ences in working memory span tasks: The role of attention in encoding
and retrieval. Paper presented at the Variation in Working Memory
Symposium, Chicago, IL.
Ohlsson, S. (1998). Spearman’s g ⫽ Anderson’s ACT?: Reflections on the
locus of generality in human cognition. Journal of the Learning Sci-
ences, 7, 135–145.
Spearman, C. (1927). The abilities of man: Their nature and measurement.
London: Macmillan.
Spearman, C. (1937). Psychology down the ages. London: Macmillan.
Su¨ß, H.-M., Oberauer, K., Wittmann, W. W., Wilhelm, O., & Schulze, R.
(2002). Working-memory capacity explains reasoning ability—and a
little bit more. Intelligence, 30, 261–288.
Tuholski, S. W., Engle, R. W., & Baylis, G. C. (2001). Individual differ-
ences in working memory capacity and enumeration. Memory & Cog-
nition, 29, 484 – 492.
Unsworth, N., & Engle, R. W. (2004). Verbal working and short-term
memory spans and their relation to fluid abilities: Evidence from list-
length effects. Manuscript submitted for publication.
Received July 9, 2004
Revision received July 21, 2004
Accepted August 3, 2004 䡲
71
WORKING MEMORY AND FLUID INTELLIGENCE: COMMENT
A preview of this full-text is provided by American Psychological Association.
Content available from Psychological Bulletin
This content is subject to copyright. Terms and conditions apply.
