CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 19

Working Memory Capacity as tion (Daneman & Carpenter, 1980).

My lab uses a version of this task in
Executive Attention which subjects read aloud a series
of two to seven sentences, each of
Randall W. Engle1 which is followed by an unrelated
School of Psychology, Georgia Institute of Technology, Atlanta, Georgia word. After the last sentence-word
combination is read, subjects try to
recall the list of unrelated words.
The critical score is the number of
Abstract tionship has been important to at- words recalled.
Performance on measures of tempts to explain cognition for The operation-span task is simi-
working memory (WM) capac- many years. lar in format to the reading-span
ity predicts performance on a It is helpful to consider this task (Turner & Engle, 1989). Sub-
wide range of real-world cog- question in the context of Baddeley jects read aloud a series of opera-
nitive tasks. I review the idea and Hitch’s (1974) formulation of tion-word strings such as “Is 4/2 
that WM capacity (a) is separa- the working memory (WM) system, 3  6? (yes or no) DOG.” They re-
ble from short-term memory, which consists of temporary mem- spond as to whether or not the
(b) is an important component ory stores with associated mecha- equation is correct and then read
of general fluid intelligence, nisms for rehearsing stored infor- the capitalized word aloud. After a
and (c) represents a domain- mation and a mechanism of central set of two to seven such operation-
free limitation in ability to con- or executive attention that regu- word strings, we measure the
trol attention. Studies show lates the contents of the active por- number of words recalled. A third
that individual differences in tion of memory. Performance on task is the counting-span task, in
WM capacity are reflected in measures of WM capacity corre- which subjects see from two to
performance on antisaccade, lates with performance on a variety seven displays of targets and dis-
Stroop, and dichotic-listening of higher-order cognitive tasks in- tractors and count the number of
tasks. WM capacity, or execu- volving, for example, reading com- targets in each display. They then
tive attention, is most impor- prehension, complex learning, and recall the list of digits correspond-
tant under conditions in which reasoning (Daneman & Carpenter, ing to the numbers of targets in the
interference leads to retrieval 1980). What mechanisms are re- displays, in order. The critical score
of response tendencies that sponsible for this correlation? is the number of digits recalled.
conflict with the current task. These tasks must involve some
fundamental aspect of cognition
Keywords because performance on them pre-
working memory capacity; at- MEASURES OF dicts performance on a wide range
tention WM CAPACITY of higher-order cognitive tasks, in-
cluding tasks involving reading
A variety of WM tasks are useful and listening comprehension, lan-
I am an avid baseball fan, and in predicting performance on a guage comprehension, ability to
when I am listening to a game on wide range of cognitive tasks that follow directions, vocabulary
the radio, particularly if the game are more closely related to real- learning, note taking, writing, rea-
involves the Atlanta Braves, my world activities. Further, the tasks soning, bridge playing, and learn-
wife will occasionally tell me some- apparently reflect a common con- ing to write computer programs
thing that she would like for me to struct because they account for (Engle, 2001).
do. However, often, and especially similar variance and all load on the
in the middle of a tense game, I same factor in a factor analysis.
will not even notice that she is talk- Generally, in each of these tasks the
ing to me. Does this ability to block subject receives items to recall and WHY DOES PERFORMANCE
out information have anything to also performs another attention- ON WM TASKS PREDICT
do with working memory? Is there demanding task that is interleaved PERFORMANCE ON
some relationship between the between receiving the items for re- HIGHER-ORDER
ability to control attention and the call. For example, the reading-span COGNITIVE TASKS?
amount of information that can be task was the first task used to
kept temporarily active in mem- study WM capacity and its rela- Originally, psychologists thought
ory? Indeed, the idea of such a rela- tionship with higher-order cogni- performance on WM-capacity

Copyright © 2002 American Psychological Society

20 VOLUME 11, NUMBER 1, FEBRUARY 2002

tasks correlated with performance or suppress information. WM ca-

on other cognitive tasks because pacity is about memory only indi- PROACTIVE INTERFERENCE
the processing portion of the WM rectly. Greater WM capacity does
tasks (e.g., skill in reading the sen- mean that more items can be main- Kane and I tested this hypothe-
tences in the reading-span task) tained as active, but this is a result sis with subjects from the upper
was similar to the task being pre- of greater ability to control atten- quartile (high spans) and the lower
dicted (reading comprehension) tion, not a larger memory store. quartile (low spans) on one of the
(Daneman & Carpenter, 1980). Thus, greater WM capacity also WM-capacity tasks. The high- and
However, individual differences means greater ability to use atten- low-WM-capacity subjects com-
on the processing portion of the tion to avoid distraction. pleted three trials on which they
WM tasks (reading, doing arith- Proactive interference refers to saw 10 words for later recall, then
metic, counting, etc.) are unimpor- the difficulty people encounter performed another task for 16 s
tant to the correlation between when a new behavior is associated (this task had the function of pre-
number of items recalled in the with a context associated with venting rehearsal of the 10 words),
WM tasks and performance on other behaviors. For example, al- and then tried to recall the words
higher-order tasks. The correlation though you probably find your car (Kane & Engle, 2000). Both groups
between WM span and reading easily the first time you park in a of subjects recalled about 60% of
comprehension, for example, is new mall, after many shopping the words from the first trial. How-
not reduced in statistical analyses trips, you may have difficulty re- ever, subjects with low spans
that control for reading or arith- calling where you parked your car showed more proactive interfer-
metic expertise (Engle, Cantor, & because of all the previous places ence over succeeding trials than
Carullo, 1992). Moreover, individ- you parked in the mall. Dealing did people with high spans. That
uals who have equal arithmetic with the effects of proactive inter- is, low-WM-capacity subjects
skill still demonstrate differences ference is one of the primary func- showed a greater loss of recall with
in the number of words recalled in tions of WM (Kane & Engle, 2000). each new list. When the subjects
the operation-span task that corre- Without the effects of interference, performed a secondary task at the
late with performance in reading most of the information people same time (i.e., when they were un-
comprehension (Conway & Engle, know and need to function in the der what is referred to as an in-
1996). If domain-specific skills or world could be retrieved from creased cognitive load), the high-
expertise cannot account for why long-term memory sufficiently and low-span groups performed at
performance on WM-capacity quickly and accurately for them to about the same level; in other
tasks correlates with performance perform even complex cognitive words, high-span subjects’ perfor-
on higher-order tasks, then what functions quite well. It is generally mance decreased under cognitive
does? considered a truism that tempo- load, but low-span subjects’ perfor-
The term capacity, as used in rarily retained information that is mance was unaffected by load. The
discussions of short-term memory not rehearsed will be lost in 20 s or finding that divided attention in-
(STM), often conjures up images of so. However, Keppel and Under- creased proactive interference for
a limited number of items or wood (1962) found that retention people with high spans suggests
chunks that can be stored (e.g., 7  for a list of three items was nearly that, under normal conditions, they
2). However, my sense is that WM perfect after such a delay if there used attentional control to combat
capacity is not about individual had been no previous lists pre- the effects of proactive interfer-
differences in how many items can sented for recall. That is, when in- ence. In contrast, the fact that cog-
be stored per se but about differ- terference is relatively absent, there nitive load did not affect proactive
ences in the ability to control atten- is little decline in the recall of infor- interference for low-WM-capacity
tion to maintain information in an mation as delay increases. This subjects suggests that they do not
active, quickly retrievable state. finding is quite informative. Effects normally allocate attention to resist
Thus, WM capacity is just as im- generally attributed to traditional interference.
portant in retention of a single rep- STM are likely to be observed only
resentation, such as the representa- when the effects of interference
tion of a goal or of the status of a force the individual to maintain in-
changing variable, as it is in deter- formation in a verbatim, relatively WM, STM, AND GENERAL
mining how many representations active state. It is also under those FLUID INTELLIGENCE
can be maintained. WM capacity is conditions that individual differ-
not directly about memory—it is ences in WM capacity become im- I believe that measures of WM
about using attention to maintain portant. capacity reflect both memory pro-

Published by Blackwell Publishing Inc.

 CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 21

cesses and executive attention, study provided inferential evidence should not differ in the prosaccade
whereas traditional measures of that performance on WM-capacity condition. However, if individual
STM reflect primarily memory pro- tasks is related to performance on differences in WM capacity corre-
cesses such as grouping, chunking, other cognitive tasks primarily be- spond to differences in executive
and rehearsal. For example, in the cause of individual differences in attention, subjects with low WM
traditional digit-span task, subjects executive attention. However, this spans should be hurt more than
are read or shown a list of digits surmise is based on logic, and, at subjects with high spans by the
and asked to recall them in order. this point, I have not provided em- need to avoid making the conflict-
Recognition of familiar sequences pirical evidence linking individual ing response in the antisaccade
in the list, such as one’s telephone differences in WM capacity and per- task.
number or address, and the ability formance on tasks that psycholo- As expected, the two groups of
to do verbal rehearsal have a large gists would generally agree are at- subjects did not differ in time to
effect on success in this task. tention tasks. I now describe some identify target letters in the prosac-
My colleagues and I used a new findings that support this con- cade condition. But, although both
structural equation modeling analy- tention. groups were slowed in the antisac-
sis2 to test this and the idea that the cade condition, the subjects with
construct measured by WM-capac- low spans were hurt more than the
ity tasks is closely associated with subjects with high spans. In a sec-
general fluid intelligence3 (Engle, ANTISACCADE TASK ond study, subjects performed an
Tuholski, Laughlin, & Conway, extended set of antisaccade trials,
1999). Although the WM and STM In an attempt to directly mea- and again we found that low-span
constructs were highly correlated sure the relationship between WM subjects were substantially slower
with each other in this analysis, a capacity and executive attention, to identify the letters than high-
model with separate constructs for my colleagues and I (Kane, Bleck- span subjects were. Figure 1 shows
WM and STM fit the data better ley, Conway, & Engle, 2001) tested the first saccades following the on-
than a model that combined the two individuals with high and low WM set of the flashing cue: Low-span
into a single construct. That is, two capacity on the antisaccade task. In subjects were more likely to follow
separate psychological mechanisms this task, subjects fixate in the mid- the natural tendency to look in the
were needed to explain the results. dle of a visual display but must re- direction of the cue than were
When the variance common to spond to target information briefly high-span subjects.
WM and STM (presumably due to presented randomly to one side or The antisaccade results would
memory processes common to the the other of the display. Just before not be predicted by any account in
two tasks) was statistically re- the target is presented, an atten- which WM capacity reflects a limi-
moved, WM still correlated with tion-attracting cue occurs on the tation in number of items. The re-
general fluid intelligence. How- side opposite where the target will sults also would not be predicted
ever, the left-over STM variance appear. The cue always predicts by any account in which individual
did not correlate with intelligence. that the target will occur on the op- differences in WM capacity reflect
Recall our logic that WM com- posite side of the display. Optimal knowledge specific to the span
prises both executive attention and performance in the antisaccade task tasks or in which the pertinent at-
memory processes, whereas STM requires that the subject resist the tentional resources are domain
comprises largely memory pro- strong tendency to shift attention specific. The data, however, are
cesses. Thus, removing the com- (as well as eye movements, or sac- consistent with a view that the un-
mon variance would eliminate the cades) to the attention-capturing derlying factor responsible for the
memory processes, and the resid- cue. Most experiments also include relationship between measures of
ual variance in WM would consti- a prosaccade control condition in WM capacity and performance on
tute executive attention. Although which the attention-capturing cue higher-order cognitive tasks is a
measures of WM capacity such as occurs on the same side of the dis- domain-free executive-attention sys-
operation and reading span are play as the subsequent target. Thus, tem. Although individuals possess-
certainly not pure, these results are the natural tendency to look at the ing different WM capacities will
consistent with the idea that the la- cue facilitates performance when it show differences in number of
tent variable resulting from WM occurs in the prosaccade condition items stored in a variety of mem-
tasks reflects a mechanism that one but hurts performance in the anti- ory tasks, this is a result of differ-
might think of as executive atten- saccade condition. ing ability to maintain and inhibit
tion, and that it is strongly related The performance of subjects information, particularly in the
to general fluid intelligence. This with high and low WM spans face of distraction and interference.

Copyright © 2002 American Psychological Society

22 VOLUME 11, NUMBER 1, FEBRUARY 2002

only when the context makes it diffi-

cult to maintain the appropriate task
goal do people with high WM spans
perform better than people with low
WM spans.

DICHOTIC-LISTENING TASK

Yet another attention task in

which WM capacity has been
shown to be important is the ven-
erable dichotic-listening task,
which measures a person’s ability
to repeat aloud words presented to
one ear while ignoring information
presented to the other ear. Con-
way, Cowan, and Bunting (2001)
Fig. 1. Proportion of initial eye movements to the misleading cue in the antisaccade had subjects shadow words in one
condition as a function of working memory span (high vs. low) and block of trials. ear while ignoring words spoken
Adapted from “A Controlled-Attention View of WM Capacity,” by M.J. Kane, M.K.
to the other ear. At some point,
Bleckley, A.R.A. Conway, & R.W. Engle, 2001, Journal of Experimental Psychology: Gen-
eral, 130, p. 176. Copyright 2001 by the American Psychological Association. Adapted each subject’s first name was spo-
with permission of the author. ken as a word in the ignored mes-
sage, and the question was whether,
at the end of the study, the subject
would report hearing his or her
als on which the ink color and the
STROOP TASK name. High-WM subjects should
word name were congruent was 0%,
be better than low-WM subjects at
50%, or 75%. Goal maintenance
ignoring distracting information,
The Stroop task, like the antisac- should be easiest in the 0% condi-
so they should be less likely to re-
cade task, also requires mainte- tion because no trials present match-
port hearing their name under
nance of a single crucial goal in ing ink color and words. Goal main-
these conditions. And although
WM. In this task, subjects are tenance should be hardest in the
only 20% of high-span subjects re-
shown color words and are re- 75%-congruent condition because
ported hearing their name, 65% of
quired to name the color of the ink the word name can be used to re-
low-span subjects reported hearing
in which each word is printed. The spond correctly on the vast majority
their name. Again, the conclusion
word and ink color can be congru- of trials. We found that performance
is that people with low WM spans
ent or incongruent. For example, on incongruent trials differed sub-
are less capable than people with
the word “red” can be printed in stantially for high- and low-WM
high WM spans of doing the men-
blue ink (incongruent) or in red ink subjects and also depended on the
tal work necessary to block dis-
(congruent). If the word and ink proportion of congruent trials. Al-
tracting information.
color are incongruent, there is a though both groups made more er-
strong predisposition to make a re- rors on incongruent trials as the pro-
sponse inappropriate to the task, portion of congruent trials increased
that is, to say the word, rather than (see Fig. 2), the percentage of errors
to say the color of the ink in which did not differ for high- and low-WM CONCLUSIONS
the word is printed. Performance subjects in the 0%- or 50%-congru-
on the Stroop task should rely on ent conditions. When 75% of the tri- WM-capacity tasks measure a
executive attention to maintain the als were congruent, however, peo- construct fundamentally important
goal of naming the color of the let- ple with low spans made almost to higher-order cognition. That
ters even when the word elicits a twice as many errors as people with construct is distinguishable from
stronger response tendency to say high spans. Thus, differences be- STM and is at least related to,
the word. tween people with high and low maybe isomorphic to, general fluid
Kane and I (2001) conducted a spans are not overall differences in intelligence and executive atten-
study in which the percentage of tri- inhibition in all situations; rather, tion. One crucial function of the

Published by Blackwell Publishing Inc.

 CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 23

served, then the model is rejected and

alternative models are then tested.
SEM is better than standard correlation
and regression procedures for two
main reasons. First, latent variables are
constructed from multiple tests that
purportedly measure the same con-
struct. Thus, a construct is defined in
terms of common variance alone, free
of measurement error. Second, causal
relations among latent variables can be
tested by fitting a model to observed
data.
3. General intelligence, or g, can be
thought to consist of two types of intel-
ligence: general crystallized intelli-
gence, which is culturally derived
knowledge, and general fluid intelli-
gence, which is the ability to reason
and to solve novel problems.

Fig. 2. Percentage of errors on incongruent trials in the Stroop task as a function of

working memory span (high vs. low) and percentage of trials that were congruent
(Kane & Engle, 2001). Cond  condition. References
Baddeley, A.D., & Hitch, G. (1974). Working mem-
ory. In G.A. Bower (Ed.), The psychology of
WM system is keeping information nism underlying fluid intelligence, learning and motivation (Vol. 8, pp. 47–89). New
York: Academic Press.
quickly retrievable when the task an important question is how dif- Conway, A.R.A., Cowan, N., & Bunting, M.F.
context provides interfering infor- ferences in executive attention are (2001). The cocktail party phenomenon revis-
ited: The importance of WM capacity. Psy-
mation that would lead to an inap- manifest in what is viewed as intel- chonomic Bulletin & Review, 8, 331–335.
propriate response. ligent behavior. Conway, A.R.A., & Engle, R.W. (1996). Individual
These conclusions raise some in- differences in WM capacity: More evidence for
a general capacity theory. Memory, 4, 577–590.
triguing questions for future re- Recommended Reading Daneman, M., & Carpenter, P.A. (1980). Indi-
search. Do individual differences vidual differences in WM and reading. Journal
in WM capacity–executive atten- Engle, R.W. (2001). (See References) of Verbal Learning and Verbal Behavior , 19 ,
Engle, R.W., & Oransky, N. (1999). 450–466.
tion reflect a central domain-free Engle, R.W. (2001). What is working memory ca-
Multi-store versus dynamic mod-
attention mechanism or domain- els of temporary storage in mem-
pacity? In H.L. Roediger, J.S. Nairne, I. Neath,
& A.M. Suprenant (Eds.), The nature of remem-
specific components? How do ory. In R.J. Sternberg (Ed.), The bering: Essays in honor of Robert G. Crowder (pp.
these differences map onto the nature of cognition (pp. 514–555). 297–314). Washington, DC: American Psycho-
Cambridge, MA: MIT Press. logical Association Press.
main brain structures associated Engle, R.W., Cantor, J., & Carullo, J.J. (1992). Indi-
with executive attention? The pre- vidual differences in WM and comprehension:
A test of four hypotheses. Journal of Experimen-
frontal cortex appears to be impor- Acknowledgments—This work was sup- tal Psychology: Learning, Memory, and Cognition,
tant in maintaining information in ported by Grants F49620-93-1-0336 and 18, 972–992.
an active state, and the anterior cin- F49620-97-1041 from the Air Force Office Engle, R.W., Tuholski, S.W., Laughlin, J.E., & Con-
of Scientific Research. way, A.R.A. (1999). Working memory, short-
gulate is an important structure in term memory and general fluid intelligence: A
detecting or dealing with conflict latent variable approach. Journal of Experimen-
tal Psychology: General, 128, 309–331.
of the type that would occur in the Kane, M.J., Bleckley, M.K., Conway, A.R.A., & En-
Notes
antisaccade and Stroop tasks. gle, R.W. (2001). A controlled-attention view
of WM capacity. Journal of Experimental Psy-
Other brain structures (e.g., the lo- 1. Address correspondence to Ran- chology: General, 130, 169–183.
cus coeruleus) appear to be impor- dall W. Engle, School of Psychology, Kane, M.J., & Engle, R.W. (2000). WM capacity,
tant to adjusting the level of mental Georgia Institute of Technology, 274 proactive interference, and divided attention:
5th St. NW, Atlanta, GA 30332-0170; Limits on long-term memory retrieval. Journal
effort expended on a task. It re- of Experimental Psychology: Learning, Memory,
e-mail: randall.engle@psych.gatech.edu.
mains to be seen whether the dif- 2. Structural equation modeling
and Cognition, 26, 336–358.
Kane, M.J., & Engle, R.W. (2001). Individual differ-
ferences in executive attention I (SEM) is a method of statistical analysis ences in executive attention and the Stroop. Manu-
have described here reflect differ- by which correlations among a large script submitted for publication.
ences in a unified central system or number of tests are analyzed. A model Keppel, G., & Underwood, B.J. (1962). Proactive
is first defined, and if the model is able inhibition in short-term retention of single
are specific to particular compo- items. Journal of Verbal Learning and Verbal Be-
to reproduce the correlations that were
nents of the brain. Also, because observed, the model is thought to be
havior, 1, 153–161.
Turner, M.L., & Engle, R.W. (1989). Is working
WM capacity (executive attention) valid. If the model is unable to repro- memory capacity task dependent? Journal of
appears to be an important mecha- duce the correlations that were ob- Memory and Language, 28, 127–154.

Copyright © 2002 American Psychological Society