Intelligence 32 (2004) 431 – 444

Is working memory fractionated onto different components of

intelligence? A reply to Mackintosh and Bennett (2003)
Roberto Colom*, Pei Chun Shih
Facultad de Psicologı́a, Universidad Autónoma de Madrid, Madrid 28049, Spain

Received 3 February 2004; received in revised form 18 June 2004; accepted 25 June 2004

Abstract

A study was conducted in which 226 participants performed 12 tests, 6 thought to reflect verbal, quantitative,
and spatial working memory (WM), and 6 of crystallized (Gc), fluid (Gf), and spatial (Gv) cognitive abilities.
Confirmatory factor analyses (CFAs) were computed to test the unitary nature of the WM system. Six primary
latent factors were obtained from the verbal, quantitative, and spatial tests. Three second-order latent factors were
then obtained from the primary latent factors, following Mackintosh and Bennett’s procedure [Intelligence 31
(2003) 519–531]. A very high correlation was observed among these second-order latent factors. Given this high
correlation, those second-order latent factors were collapsed into a single second-order latent factor. The restricted
model fitted the data as well as the base model. The result strongly supports the unitary nature of the WM system.
It is proposed that this system can be conceived as a single pool of general purpose resources used to temporarily
preserve a reliable mental representation of information about any task.
D 2004 Elsevier Inc. All rights reserved.

Keywords: Working memory; Working memory general resources; Working memory specific resources; Cognitive abilities;
Intelligence

1. Introduction

Shah and Miyake (2003) designated eight theoretical questions to ask a group of leading researchers
within the working memory (WM) literature: basic mechanisms and representations in WM, the control

* Corresponding author. Tel.: +34 91 497 41 14; fax: +91 497 52 15.
E-mail address: [Link]@[Link] (R. Colom).

0160-2896/$ - see front matter D 2004 Elsevier Inc. All rights reserved.
doi:10.1016/[Link].2004.06.011
432 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

and regulation of WM, the unitary vs. nonunitary nature of WM, the nature of WM limitations, the role
of WM in complex cognitive activities, the relationships of WM to long-term memory and knowledge,
the relationships of WM to attention and consciousness, and the biological implementation of WM.
Although all of these questions are important, two of them were considered especially controversial:

! Is WM a unitary construct, or does it consists of multiple separable subsystems?

! What are the mechanisms that constrain the capacity of WM (a limited supply of activation,
processing speed, decay, inhibition, interference, and skills)?

The high correlation usually found between WM and intelligence supports the view that the latter is
linked with a unitary WM system (Engle, Tuholski, Laughlin, & Conway, 1999; Kane, Hambrick,
Tuholski, Wilhelm, Payne, & Engle, 2004; Duncan et al., 2000; Colom, Flores-Mendoza, & Rebollo,
2003; Colom, Rebollo, Palacios, Juan-Espinosa, & Kyllonen, in press; Colom, Rebollo, Abad, & Shih,
submitted for publication; Colom, Shih, Abad, Rebollo, & Flores-Mendoza, submitted for publication;
Colom, Abad, Rebollo, & Shih, submitted for publication). However, there are some researchers
claiming that there is evidence that WM is not a unitary domain-general system.
Shah and Miyake (1996) support the separability of WM resources. Their studies presumably
demonstrate that spatial and verbal WM are separable and that cognitive activities in these two domains
are fuelled by two different pools of resources. They predict that bif measures of spatial WM and
measures of verbal WM predict performance on spatial tasks and language tasks equally well, it may
suggest the existence of a single pool of general purpose resources that can be used for a variety of
cognitive processes across different domains. In contrast, if spatial WM measures predict spatial
performance only and verbal WM measures predict language performance only, then such domain
specificity may point to the separability of WM resources for spatial thinking and language processingQ
(p. 6).
Their first study found a correlation of .66 between spatial WM and a composite spatial score, and a
correlation of .45 between verbal WM and a verbal SAT score. However, the correlations across content
domains were .07 and .12, respectively. Their second study found similar results. Despite these results,
those researchers concluded that bboth spatial thinking and language comprehension would be
constrained primarily by the size of the pool of the general purpose resourcesQ (p. 21, emphasis added).
Shah and Miyake’s (1996) studies were taken as a starting point by Mackintosh and Bennett (2003) It
is surprising that the later researchers considered seriously two correlational studies that have analyzed
data from 54 (first study) and 30 (second study) participants. It is widely recognized that correlations are
not especially reliable for samples under 100 participants (Detterman, 1989). Moreover, there are several
hardly understandable results in Shah and Miyake’s (1996) studies. First, spatial WM correlates .37 with
the verbal SAT and .61 with the quantitative SAT. Second, spatial short-term memory (STM) correlates
.25 with the verbal SAT and .64 with the quantitative SAT. Third, verbal WM correlates .44 with the
verbal SAT and .13 with the quantitative SAT. Finally, verbal STM correlates .43 with the verbal SAT
and .16 with the quantitative SAT. These results do not support the view that spatial and verbal WM are
fuelled by two different pools of resources.
Nevertheless, Mackintosh and Bennett (2003) advocate the separability of WM resources. They
measured verbal, quantitative, and spatial WM. Verbal WM was measured by the reading span test
(Baddeley, Logie, Nimmo-Smith, & Brereton, 1985), quantitative WM was measured by the mental
counters test (Larson & Sacuzzo, 1989), and spatial WM was measured by the Daneman and Tardif’s
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444 433

(1987) spatial span task. They also measured fluid intelligence, crystallised intelligence, and spatial
ability, by the Progressive Matrices test (Raven, 1962), the Mill Hill Vocabulary Scale, and the
Vandenberg and Kruse’s (1978) mental rotation test, respectively.
Their results suggest that the content of both ability and WM tests influences their degree of
association. They performed confirmatory factor analyses (CFAs) to test a model with one ability and
one WM test in each latent factor (Gc=Mill Hill and verbal WM; Gf=Progressive Matrices and mental
counters; Gv=mental rotation and spatial WM). They found correlations among the latent factors of .67
(Gc–Gf), .73 (Gf–Gv), and .28 (Gc–Gv), thus arriving at the conclusion that bour ability and WM tests
are measuring three partially discriminable sets of cognitive operationsQ (p. 526).
They go beyond this conclusion saying that the finding that verbal WM is relatively independent of
spatial WM is consistent with the distinction between the classical distinction between the spatial
sketchpad and the phonological loop (Baddeley & Hitch, 1974). Verbal and spatial WM tasks place
different processing demands on the participants (Baddeley & Logie, 2003).
However, Mackintosh and Bennett (2003) acknowledged that (1) their study is not very large, and (2)
they have only one measure of ability and one of WM within each of the three possible content domains.
Concerning the second point, they write: bmultiple measures are clearly needed in order to identify and
study the relationship among the three factors suggested by our best-fitting modelQ (p. 529).
The present study is designed to surpass the later limitation acknowledged by Mackintosh and
Bennett (2003). Verbal, quantitative, and spatial WM were measured by two tasks each. Crystallised,
fluid, and spatial cognitive abilities were also measured by two standardized tests each. WM and
cognitive ability primary latent factors were obtained by CFA. Those primary factors were then
collapsed into higher order factors representing Gc, Gf, and Gv. However, contrary to Mackintosh and
Bennett (2003), we endorse the unitary nature of the WM system. If we are right, then a restricted model
in which Gc, Gf, and Gv are collapsed into a single second-order latent factor will not lose goodness of
fit with respect to the base model. If this demonstrates to be the case, then Mackintosh and Bennett
(2003) nonunitary view should be considered questionable.

2. Method

2.1. Participants

Two hundred twenty-six participants took part in the study. The sample comprised Psychology
undergraduates. The students participated to fulfill a course requirement. The mean age was 20.42
(S.D.=3.25).

2.2. Administered tests

The six WM tests considered in the present study require the switching of control processes forth and
back between the representation of the items and the processing component. Verbal WM was measured
by the ABCD and Alphabet tests. Both tests were modelled after the CAM Battery (Kyllonen & Christal,
1990). The ABCD task was employed by Ackerman, Beier, and Boyle (2002) as well as by Engle et al.
(1999) as a measure of verbal WM. Quantitative WM was measured by the mental counters and
computation span tests. The counter test was modelled after Larson and Sacuzzo (1989) and it was
434 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

employed by Mackintosh and Bennett (2003) as a measure of quantitative WM. The computation span
test was modelled after Ackerman et al. (2002).Finally, spatial WM was measured by the dot matrix and
letter rotation tests. Both tests were modelled after the Miyake, Friedman, Rettinger, Shah, and Hegarty
(2001) study. Both spatial tests involve visuospatial storage (dot locations and spatial orientations,
respectively) with a concurrent visuospatial processing (verification of spatial matrix equations or mental
rotation). A more detailed description follows.

2.2.1. ABCD
Two categories were used, with five words in each category. The categories were trees (árboles) and
food (comida). Three study frames were displayed on the computer screen for 3 s each. The first frame
indicated the order of two members from the same category (El cedro precede al roble); the second
frame indicated the order of two members from the other category (El ajo no precede a la sal); the third
frame indicated the order of the categories (Los árboles no preceden a la comida). After the third study
screen, an eight-choice answer screen was displayed from which participants selected the correct order of
the words. Participants were allowed 10 s to select a response. The use and ordering of category
members were balanced across items, as were the variations of order (precede, no precede). There were
14 trials and two practice trials. The number of hits was obtained as the participant’s score.

2.2.2. Alphabet
This task required participants to apply successor and predecessor operations to a string with a given
number of letters. A typical trial was

(Screen 1) BTG
(Screen 2) +2
(Screen 3) S

for which the correct response would be: D V I.

The string of letters is presented for 3 s, the operation to apply is presented for 1500 ms, and the
participant has unlimited time to enter a response. The number of letters increased from three to seven
(4 trials at each level and a total of 20 trials). For two trials within a given block, the participant must
add one or two positions, while for the other two trials, the participant must subtract one or two
positions. The number of additions and subtractions are randomized within a given block of trials. The
number of hits is obtained as the participant’s score.

2.2.3. Counters
Three boxes representing counters appeared on the computer screen. At the beginning of each trial,
the value of the three counters was set to 0, 0, 0. A flash appeared above or below one counter for 500
ms. If the flash appeared above the box, the participant must add one (+1) to that counter, but if the flash
appeared below the box, the participant must subtract one to that counter ( 1). The participant’s task is
to keep a running track of the value of the three counters. At the end of each trial, participants reported
the cumulative total of all three counters. The participant had unlimited time to enter a response. The task
comprised 10 trials with five counter changes and 10 trials with seven counter changes (and three
practice trials with three counter changes). Maximum and minimum values used were +3 and 3. The
number of hits was obtained as the participant’s score.
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444 435

2.2.4. Computation span

This task included a verification task and a recall task. Participants were allowed 6 s to verify the
accuracy of a math equation and were instructed to remember the displayed solution, irrespective of its
accuracy. After the final equation of the trial was displayed, participants were prompted to remember in
the correct serial order each of the presented solutions from the equations. Each math equation included
two operations using digits from 1 to 10, and the provided and actual solutions were always single-digit
numbers. Set size ranged from three to seven equation/solutions (15 trials total). The participant score
was obtained after the number of hits in the verification and remembering tasks.

2.2.5. Dot matrix

The requirement is to verify a matrix equation while simultaneously remembering a dot location in a
55 grid. In the matrix equation display, a simple addition or subtraction equation was presented on the
computer screen. Participants were given a maximum of 4.5 s to verify each equation. Immediately after
their response, the computer displayed a dot grid for 1.5 s. After a sequence of between two and five
equation-grid pairs, participants recalled, in any order, which grid spaces contained dots clicking with
the mouse on an empty grid. There were three practice trials with two equations and two dots each, after
which sets progressively increased in size from two to five equations and dots for a total of 12 sets, 3 of
each size. The number of hits in the verification and remembering tasks were obtained as the
participant’s score.

2.2.6. Letter rotation

Each trial consisted of the sequential presentation of a set of the same capital letter (F, J, P, or R) each
of which appeared on a computer screen either normal or mirror imaged and rotated in one of seven
possible orientations (multiples of 458). The participant’s task was to decide whether each letter was
normal or mirror imaged as quickly and as accurately as possible and remember its spatial orientation
(where the top of each letter was pointing). Participants were given a maximum of 3 s to respond normal
or mirror, and then the subsequent letter is displayed. After each set, the participant saw a grid and
marked the places corresponding to the positions of the tops of the presented letters in the correct serial
order. There were three practice trials with two letters each, after which sets progressively increased in
size from two to five letters for a total of 12 sets, 3 of each size. The number of hits in the verification
and remembering tasks were obtained as the participant’s score.
Finally, crystallised intelligence (Gc) was measured by the vocabulary subtest from the Primary
Mental Abilities Battery (PMA-V; Thurstone, 1938) and the verbal reasoning subtest from the
Differential Abilities Battery (DAT-VR; Bennett, Seashore, & Wesman, 1974). Fluid intelligence (Gf)
was measured by the reasoning subtest from the PMA (R; Thurstone, 1938) and the Culture Fair
Intelligence Test (Subtests 1 and 3; TEA, 1997). Spatial ability was measured by the mental rotation test
from the PMA (S; Thurstone, 1938) and the surface development test (Yela, 1969).

2.3. Procedure

Testing took place in two sessions, administered collectively (in groups of no more than 20
participants), for a total of 2 h approximately. The order of task administrations was fixed for all the
participants to minimize any measurement error due to an order by participant interaction. The WM tasks
were administered in Session 1, while the cognitive ability tests were administered in Session 2.
436 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

3. Results

The descriptive statistics are shown in Table 1. Tasks’ correlations and reliability indices (Cronbach’s a)
are also presented in Table 1.
Skew and kurtosis values are within the normal range. The measures meet standard criteria for
univariate normality, with skew values less than 3 and kurtosis values less than 4 (Kline, 1998).
Reliability indices show appropriate values, between a minimum of .73 and a maximum of .93.
The latent-variable approach has several advantages. Because latent variables are what is shared by
multiple tasks used to tap some construct, specific task requirements have less influence on the estimates
of the construct relations. Moreover, because the measurement error for each task is not part of the latent
variable, the latter provides a reliable measure of the construct.
To maximize the advantages of the latent-variable approach and to surpass the acknowledged
limitation of the Mackintosh and Bennett’s (2003) study, six measures of WM and six measures of
cognitive ability are considered in the present study. Verbal, quantitative, and spatial tests are
administered. The base model (Model 1) included six primary latent factors defined by two measures
each. The hierarchical base model (Model 2) comprised three second-order latent factors defined by two
primary latent factors each.
The CFAs were performed with the AMOS program (Arbuckle, 1999). AMOS uses maximum-
likelihood estimation to derive the specific parameters based on the covariance matrix.
The starting point is a model in which six primary latent factors were defined by WM and cognitive
ability measures (Model 1). Verbal WM was defined by alphabet and ABCD, quantitative WM was
defined by mental counters and computation span, and spatial WM was defined by dot matrix and letter
rotation. Verbal ability was defined by PMA-V and DAT-VR, fluid intelligence was defined by PMA-R

Table 1
Descriptive statistics and correlation matrix for the dependent measures
Measures 1 2 3 4 5 6 7 8 9 10 11 12
(1) Alphabet .333 .272 .349 .294 .303 .366 .285 .315 .269 .311 .325
(2) ABCD .220 .282 .238 .246 .297 .231 .255 .218 .252 .263
(3) PMA-V .577 .259 .267 .322 .251 .277 .236 .274 .286
(4) DAT-VR .332 .342 .413 .322 .355 .303 .351 .367
(5) Counters .355 .379 .295 .348 .298 .345 .360
(6) Computation .390 .304 .359 .307 .356 .371
(7) PMA-R .462 .434 .370 .429 .448
(8) Cattell .338 .288 .334 .349
(9) Dot matrix .380 .370 .386
(10) Letter rotation .316 .330
(11) PMA-E .436
(12) Surface development
Mean 7.65 8.29 32.62 26.97 9.38 17.15 21.14 13.40 76.50 46.07 27.20 24.51
S.D. 3.35 4.34 6.69 7.20 3.82 5.36 4.41 2.16 8.69 9.06 11.56 12.82
Skew .46 .35 .02 .16 .26 .25 .36 .50 .15 .06 .26 .61
Kurtosis 0.02 1.24 0.41 0.55 0.61 0.39 0.01 0.44 1.26 0.43 0.56 0.15
Reliability .76 .87 .85 .82 .80 .82 .80 .75 .78 .76 .73 .93
Reliability indices are also shown.
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444 437

and the Culture Fair Intelligence Test (Subtests 1 and 3), and spatial ability was defined by PMA-S and
surface development.
The fit for Model 1 was as follows: v 2(39)=70.28, CMIN/df=1.80, Pb.01, RMSEA=.060 (confidence
interval of .036 to .082). Those fit indices can be considered acceptable (Ackerman et al., 2002; Byrne,
1998). The correlations among the latent factors are all quite large (see Fig. 1). It is important to note that
there is a correlation of .82 between verbal WM and quantitative WM, a correlation of .88 between
verbal WM and spatial WM, and a correlation of 97 between quantitative WM and spatial WM. These
results clearly failed to support the separability of verbal, quantitative, and spatial WM advocated by
Mackintosh and Bennett (2003).
Following Mackintosh and Bennett (2003), the six primary latent factors were collapsed in three
second-order latent factors: Gc was defined by verbal WM and verbal ability, Gf was defined by
quantitative WM and fluid intelligence, and Gv was defined by spatial WM and spatial ability (Model 2).
Perhaps the consideration of quantitative WM as a marker of Gf can be seen as surprising. However, we
here also follow Mackintosh and Bennett, because they considered the mental counters test as a marker
of Gf.
The fit for Model 2 was as follows: v 2(45)=92.02, CMIN/df=2.04, Pb.01, RMSEA=.068 (confidence
interval of .048 to .088). Those fit indices can be considered acceptable. It is noteworthy that, contrary to
Mackintosh and Bennett’s (2003) findings, the correlations among Gc, Gf, and Gv are very large (all
above .90; Fig. 2).
The high correlation among Gc, Gf, and Gv suggests that a single latent factor underlies the primary
latent factors. Taking this possibility into consideration, a restricted model (Model 3) was tested: Gc, Gf,
and Gv were collapsed into a single factor. The standardized regression weights for Model 3 are shown
in Fig. 3.
The fit for Model 3 was: v 2(48)=93.67, CMIN/df=1.95, Pb.01, RMSEA=.065 (confidence interval of
.045 to .084). Therefore, the restricted model (Model 3) fitted the data just as well as the hierarchical
base model (Model 2). The result supports the unitary nature of the WM system, rejecting the

Fig. 1. Model 1. Six correlated latent factors representing WM and cognitive abilities across content domains.
438 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

Fig. 2. Model 2. Hierarchical model for Gc, Gf, and Gv.

fractionation of WM maps onto different components of intelligence endorsed by Mackintosh and

Bennett (2003).

4. General discussion

The findings reported in the present study strongly support the statement that verbal, quantitative, and
spatial WM are not mainly fuelled by different pools of resources. Instead, WM can be considered as a
single pool of general purpose resources. Presumably, this pool is used for several diverse cognitive
processes across different content domains.
Shah and Miyake (1996) supposedly found that measures of spatial WM predict spatial performance
only, while measures of verbal WM predict language performance only. Despite the lack of reliability of
Shah and Miyake (1996) results, we have observed that measures of spatial WM and measures of verbal
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444 439

Fig. 3. Model 3. Restricted model in which the six primary latent factors are collapsed into a single higher order factor.

WM predict performance on spatial and language tasks (Fig. 1). There is a strong positive manifold
between verbal, quantitative, and spatial measures.
Stauffer, Ree, and Carretta’s (1996) findings are consistent with the later statement. They studied
the relationship between g (obtained from the ASVAB) and a general cognitive factor (obtained from
25 computer-based cognitive-components measures). They found a correlation of .95, concluding
that bthe amount of g in the common variance among the cognitive-components tests is greater than
in traditional paper-and-pencil tests, indicating that the amount of reliable variance attributable to
specific abilities is smaller in cognitive-components tests than in traditional paper-and-pencil testsQ
(pp. 200–201).
Mackintosh and Bennett (2003) reported a study in which WM is considered from a nonunitary view.
Although WM may be partly general, it is also domain specific bwith three of these domains
corresponding to Gc, Gv, and GfQ (p. 519). They support the conclusion that verbal and spatial WM tasks
place different processing demands. They endorse Miyake and Shah’s (2003) statement: bWM is not
completely unitary, and any comprehensive theory of WM should be able to account for domain-specific
effectsQ (p. 448).
440 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

However, Kintsch, Healy, Hegarty, Pennington, and Salthouse (2003) insist in that there is a need to
define a set of criteria that should be met before we postulate the existence of a subcomponent in WM:
bin interest of parsimony, it is desirable that additional subcomponents of WM are postulated only when
there is evidence for dissociations between subsystems in the basis of different methodologiesQ (p. 420,
emphasis added).
Shah and Miyake (1996) recognized that the single resource model is able to account for their
results (p. 21): spatial thinking and language comprehension would be constrained primarily by the
size of the pool of the general purpose resources. Oberauer, Süg, Schulze, Wilhelm, and Wittmann
(2000) and Süg, Oberauer, Wittman, Wilhelm, and Schulze (2002) reported a strong association
between verbal and spatial WM. Mackintosh and Bennett (2003) acknowledged that bthe weight of the
evidence is consistent with there being an equally general component of WM that is strongly
associated with gQ (p. 529).
We must say that the weight of the evidence is more than consistent: WM is strongly associated
with g. First, Ackerman et al. (2002) administered 36 ability tests representing verbal, numerical,
spatial, and processing speed abilities; the Progressive Matrices Test; and seven WM tests. The
CFA model depicted in their Fig. 5 shows a .70 correlation between WM and g. Second, Süg et
al. (2002) showed that g correlated .58 with WM. Third, Colom et al. (in press) found (across
three separate studies) an average correlation of .96 between g and WM. Finally, the re-analysis
reported by Colom et al. (in press) from theAckerman et al. (2002) dataset indicated that WM was
the latent factor best predicted by a higher order factor representing g (the structural coefficient
was a large .94).
For illustrative purposes, we have tested a model with the dataset considered in the present study. That
model postulated verbal, quantitative, and spatial WM as primary latent factors, as well as a second-
order WM latent factor. On the other hand, there were primary latent factors for Gc, Gf, and Gv, as well
as a second-order latent factor ( g) underlying those primary factors (Fig. 4). The fit for this model was
quite good: v 2(47)=82.78, CMIN/df=1.76, Pb.01, RMSEA=.058 (confidence interval of .037 to .078).
The computed correlation between WM and g was a large .86.
Given the current state of affairs, why to invest time and effort trying to understand secondary
issues? It is clear that the single pool of general purpose resources is much more important than specific
pools of resources. Although we all must acknowledge that WM is not absolutely unitary, it is time to
learn from the past: as within the cognitive abilities literature, WM can be considered from a
hierarchical view, where the single pool of general purpose resources sits atop the hierarchy (Carroll,
1993, 2003).
Although researchers could agree on that hierarchical view, to say that there are general and
specific resources is to say really nothing. What is needed is a quantification of the importance of
the domain-general and domain-specific resources. Miyake and Shah (2003) expressed this view:
brather than debating whether WM is unitary or not, the key issue may be to specify the sources of
domain-specific effects found in both experimental studies and correlational studies. If one postulates
a domain-general factor, another key issue is to specify the extent to which the domain-general and
domain-specific factors, respectively, contribute to constraining WM performanceQ (p. 449, emphasis
added).
Engle, Kane, and Tuholski, (2003) proposed a hierarchical view based on both domain-general and
domain-specific components. From this view, a general factor can be equated with the domain-general
component of span tasks, whereas specific factors can be equated with domain-specific components.
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444
Fig. 4. Model 4. g and WM defined as higher order correlated factors from the corresponding primary latent factors.

441
442 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

Taking this view into consideration, Colom, Abad, Rebollo, & Shih, (submitted for publication)
quantified the relative importance of these general and specific WM components after an extensive
reanalysis of five datasets. Their results demonstrated that the general component explains about four
times the variance explained by the specific components.
Using the same procedure as Colom et al. (submitted for publication), we have computed the
percentage of variance explained by the general and specific components with the dataset considered in
the present study. The results were entirely consistent with those found by Colom et al. (submitted for
publication) approximately 30% of the common variance is explained by the general component,
whereas between 4.5% and 6% of the common variance is explained by the specific components.
We have proposed elsewhere (Colom et al., submitted for publication; Colom, et al., submitted for
publication) that these general resources of the WM system are required to temporary store the
information of any cognitive task. WM tasks are memory span tasks characterized by the temporary
storage of the information. People differ in their capacity to temporarily store that information. Those
individual differences can be attributed to a general pool of cognitive resources to maintain a reliable
mental representation of information about any task, irrespective of its content.
g is thought to represent the average efficiency of the total set of cognitive functions (Carroll, 1991;
Duncan et al., 2000; Jensen, 1998). That efficiency could depend on a single pool of general purpose
resources provided by a presumed brain’s central processing unit. It is reasonable to state that there is
one general limitation underlying information maintenance within the WM system. Those storage-plus
processing tasks strain that general limitation to a high degree, mainly because the processing
requirement leaves a small amount of resources to the active (and reliable) maintenance of the relevant
information.
Kane and Engle (2002) state that there are too many strong correlations among diverse memory span
tasks and diverse higher order tasks ( g demanding) to deny that a general mechanism is involved. We
strongly endorse this statement. The ability to maintain a reliable memory representation in an active
state could be crucial to general success across complex tasks. It is well documented that increases in
complexity both on memory span tasks and on intelligence tests selectively recruit more activation from
the same brain area, namely, the dorsolateral prefrontal cortex (Duncan et al., 2000; Kane & Engle,
2002). This fact is consistent with our view of the single pool of general purpose resources: memory
span tasks and intelligence tests exhaust the same single general purpose resources.
In summary, verbal, quantitative, and spatial WM are not fuelled by different pools of resources.
Instead, WM can be considered primarily as a single pool of general purpose resources used for several
diverse cognitive processes across different content domains. WM is mainly a unitary system, although
there is a nonnegligible contribution of domain-specific resources. Researchers should first focus their
attention to the understanding of the nature of the single pool of general purpose resources, mainly
because it severely limits the capacity of the WM system. We suggest that the same entity limits WM. If
this entity is mental energy (Just & Carpenter, 1992; Kahneman, 1973), activation (Lovett, Reder, &
Lebiere, 2003), or still another unknown factor, is a question that remains to be answered.

Acknowledgements

The research referred to in this article was supported by a grant funded by the Spanish bMinisterio de
Ciencia y Tecnologı́aQ (Grant No. BSO2002-01455).
 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444 443

References

Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2002). Individual differences in working memory within a nomological network
of cognitive and perceptual speed abilities. Journal of Experimental Psychology. General, 131(4), 567 – 589.
Arbuckle, J. L. (1999). SPSS: AMOS 4.0. SmallWaters.
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), The Psychology of Learning and Motivation,
vol. 8. (pp. 47 – 89). New York7 Academic Press.
Baddeley, A. D., & Logie, R. H. (2003). Working memory: The multiple component model. In A. Miyake, & P. Shah (Eds.),
Models of working memory: Mechanisms of active maintenance and executive control, (2nd ed., pp. 28–61). New York7
Cambridge University Press.
Baddeley, A., Logie, R., Nimmo-Smith, I., & Brereton, N. (1985). Components of fluent reading. Journal of Memory and
Language, 23, 119 – 131.
Bennett, G. K., Seashore, H. G., & Wesman, A. G. (1974). Differential aptitude test. Madrid, Spain7 TEA, S.A.
Byrne, B. M. (1998). Structural equation modelling with LISREL, PRELIS, and SIMPLIS: Basic concepts, applications, and
programming. Mahwah, NJ7 Erlbaum.
Carroll, J. B. (1991). No demonstration that g is not unitary, but there is more to the story: Comment on Kranzler and Jensen.
Intelligence, 15, 423 – 436.
Carroll, J. B. (1993). Human cognitive abilities. Cambridge, UK7 Cambridge University Pres.
Carroll, J. B. (2003). The higher-stratum structure of cognitive abilities: Current evidence supports g and about ten broad
factors. In H. Nyborg (Ed.), The Scientific Study of General Intelligence: Tribute to Arthur R. Jensen (pp. 5 – 21).
Cambridge, UK7 Elservier Science/Pergamon Press.
Colom, R., Abad, F. J., Rebollo, I., & Shih, P. C. (2004). Memory span and General Intelligence: A Latent-Variable Approach.
Paper Submitted for Publication.
Colom, R., Flores-Mendoza, C., & Rebollo, I. (2003). Working memory and intelligence. Personality and Individual
Differences, 34, 33 – 39.
Colom, R., Rebollo, I., Abad, F. J., & Shih, P. (2004). Complex span tasks, simple span tasks, and cognitive abilities: A re-
analysis of key studies. Paper Submitted for Publication.
Colom, R., Rebollo, I., Palacios, A., Juan-Espinosa, M., & Kyollen, P. (2004). Working memory is (almost) perfectly predicted
by g. Intelligence, 32, 277 – 296.
Colom, R., Shih, P., Abad, F. J., Rebollo, I., & Flores-Mendoza, C. (2004). Short-term memory and working memory are
conditioned by the same underlying capacity limitations: A latent-variable analysis. Unpublished Manuscript.
Daneman, M., & Tardif, T. (1987). Working memory and reading skill re-examined. In M. Coltheart (Ed.), Attention and
Performance, vol. XII. (pp. 491 – 508). Hove, Sussex7 Erlbaum.
Detterman, D. K. (1989). The future of intelligence research. Intelligence, 13, 199 – 203.
Duncan, J., Seitz, R. J., Kolodny, J., Bor, D., Herzog, H., Ahmed, A., Newell, F. N., & Emslie, H. (2000). A neural basis for
general intelligence. Science, 289, 457 – 460.
Engle, R. W., Kane, M. J., & Tuholski, S. W. (2003). Individual differences in working memory capacity and what they tell us
about controlled attention, general fluid intelligence, and functions of the prefrontal cortex. In A. Miyake, & P. Shah (Eds.),
Models of working memory, (2nd ed., pp. 102–134). New York7 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(3), 309 – 331.
Jensen, A. (1998). The g factor. Westport7 Praeger.
Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory.
Psychological Review, 99, 122 – 149.
Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ7 Prentice-Hall.
Kane, M., & Engle, R. (2002). The role of prefrontal cortex in WMC, executive attention, and general fluid intelligence: An
individual-differences perspective. Psychonomic Bulletin and Review, 9(4), 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 visuo-spatial memory span and reasoning. Journal of
Experimental Psychology. General, 133, 189 – 217.
444 R. Colom, P. Chun Shih / Intelligence 32 (2004) 431–444

Kintsch, W., Healy, A. F., Hegarty, M., Pennington, B. F., & Salthouse, T. A. (2003). Models of working memory: Eight
questions and some general issues. In A. Miyake, & P. Shah (Eds.), Models of working memory: Mechanisms of active
maintenance and executive control, (2nd ed., pp. 412–441). New York7 Cambridge University Press.
Kline, P. (1998). Principles and practice of structural equation modellling. New York7 Guilford Press.
Kyllonen, P., & Christal, R. (1990). Reasoning ability is (little more than) working memory capacity?! Intelligence, 14,
389 – 433.
Larson, G. E., & Sacuzzo, D. P. (1989). Cognitive correlated of general intelligence: Toward a process theory of g. Intelligence,
13, 5 – 31.
Lovett, M., Reder, L. M., & Lebiere, C. (2003). Modeling working memory in a unified architecture: An ACT-R perspective. In
A. Miyake, & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control, (2nd
ed., pp. 135–182). New York7 Cambridge University Press.
Mackintosh, N. J., & Bennett, E. S. (2003). The fractionation of working memory maps onto different components of
intelligence. Intelligence, 31, 519 – 531.
Miyake, A., & Shah, P. (2003). Toward Unified Theories of Working Memory. In P. Shah, & A. Miyake (Eds.), Models of
working memory, (2nd ed., pp. 442– 481). Cambridge, UK7 Cambridge University 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 Experimental Psychology: General,
130(4), 621 – 640.
Oberauer, K., Süg, H., Schulze, R., Wilhelm, O., & Wittmann, W. (2000). Working memory capacity—Facets of a cognitive
ability construct. Personality and Individual Differences, 29(6), 1017 – 1046.
Raven, J. C. (1962). Extended Guide to Using the Mill Hill Vocabulary Scale with the Progressive Matrices Scales. London7
H.K. Lewis.
Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial thinking and language processing: An
individual differences approach. Journal of Experimental Psychology. General, 125, 4 – 27.
Shah, P., & Miyake, A. (2003). Models of working memory: An introduction. In P. Shah, & A. Miyake (Eds.), Models of
working memory, (2nd ed., pp. 1– 27). Cambridge, UK7 Cambridge University Press.
Stauffer, J., Ree, M., & Carretta, T. (1996). Cognitive-components tests are not much more than g: An extension of Kyllonen’s
analyses. The Journal of General Psychology, 123(3), 193 – 205.
Süg, H., Oberauer, K., Wittman, W., Wilhelm, O., & Schulze, R. (2002). Working memory capacity explains reasoning
ability—and a little bit more. Intelligence, 30, 261 – 288.
TEA (1997). Manual del test de factor g de Cattell [Handbook of the Cattell’s Culture-Fair Intelligence Test]. Madrid: TEA.
Thurstone, L. (1938). Primary mental abilities. Psychometric Monographs, 1.
Vandenberg, S. G., & Kruse, A. R. (1978). Mental rotations: Group tests of three-dimensional spatial visualization. Perceptual
and Motor Skills, 47, 599 – 604.
Yela, M. (1969). Desarrollo de superficies [Surface development]. Madrid, Spain7 TEA.