A Working Memory Workout: How to Expand the Focus of SerialAttention From One to Four Items in 10 Hours or Less

Paul Verhaeghen, John Cerella, and Chandramallika Basak

Syracuse University

Five individuals participated in an extensive practice study (10 1-hr sessions, 11,000 trials total) on aself-paced identity-judgment

-back task (

ranging from 1 to 5). Within Session 1, response timeincreased abruptly by about 300 ms in passing from



1 to



1, suggesting that the focus of attentioncan accommodate only a single item (H. Garavan, 1998; B. McElree, 2001). Within Session 10, responsetime was dramatically reduced and increased linearly with

for



4, with a slope of about 30 ms. Thedata suggest that working memory consists of a focus of attention governed by a limited-capacity search,expandable through practice, and a content-addressable region outside the focus of attention.

Recent researchers have suggested that working memory opera-tions that involve only a single item have privileged access (Garavan,1998; McElree, 2001; McElree & Dosher, 1989; Oberauer, 2002;Verhaeghen & Basak, in press; Voigt & Hagendorf, 2002). Thesefindings have led to the idea that a

focus of attention

, containing asingle item, may lie at the core of the working memory system. Datafrom an earlier experiment of ours that exemplify this pattern aredepicted in Figure 1 (Verhaeghen & Basak, in press, Experiment 1,college-age sample). The task is a version of the identity-judgment

-back task (after McElree, 2001), in which

was varied from 1 to 5(for a full description of the task, see the Method section). The criticaldata are given by the response times (left panel). As shown in thefigure, the Response Time



trace is close to a step function, witha fast response at



1 and then a jump to slower and statisticallyequal response times for values of



1.How are data such as these to be interpreted? In recent theoriesof working memory, researchers propose that working memory issubdivided into concentric regions that differ in the accessibility of the information stored. Cowan’s (1988, 1995, 1999, 2001) modelhas probably been most influential in this regard. In the model,Cowan proposes a hierarchical two-tier structure for workingmemory, distinguishing a zone of immediate access, labeled thefocus of attention, from a larger, activated portion of long-termmemory in which items are stored in a readily available but notimmediately accessible state. (In accordance with the terminologyintroduced by McElree in, e.g., McElree, 2001, accessibility of anelement in working memory is defined by the time needed toretrieve it; availability is defined by the probability that the ele-ment is retrieved correctly.)The focus of attention is typically considered to be capacitylimited, and to contain a fixed number of items; the activatedportion of long-term memory is not thought to be capacity limited,but is subject to interference and decay. From this, it follows thatone can measure the size of the focus of attention by the numberof items that can be accessed immediately. Operationally, one canmake the measurement for any task that requires retrieval fromworking memory while varying the number of items to be retained.One should expect a jump in response times when the limit of thefocus of attention is reached. If one accepts these premises, theresults depicted in Figure 1 can be interpreted as follows: The itemin the



1 condition has privileged access because, unlike theitems in the



1 conditions, it is held in the focus of attention andhence, no retrieval process is needed prior to the comparisonprocess. For items stored in working memory outside the focus of attention, an additional retrieval step is necessary at a considerableresponse time cost. Moreover, because items stored outside thefocus of attention are stored in content-addressable format (likeitems stored in long-term memory; McElree, 2001), access timesfor these items should be unchanged across values of

. Therefore,in the

-back task, a step function of response time over

isexpected, with the step situated at



2.Although the results shown in Figure 1 may seem straightfor-ward, they are not in line with the expectations from standardtheories of working memory. Early research has pointed at amagical number of seven items (plus or minus two; Miller, 1956).This estimate of the maximum capacity of working memory isprobably too large. It was derived from forward digit span tasks,and it has been shown that these tasks are contaminated with theeffects of rehearsal. Rehearsal allows for the creation of chunksduring encoding and for memorization into long-term memory,thereby inflating the estimate of the fundamental capacity limit(Cowan, 2001). Cowan (for an extensive review, see Cowan,2001) ascribes to the focus of attention a size of not one, not seven,but four (plus or minus one). Evidence for his estimate is derivedfrom a variety of experiments in a large number of researchdomains, including cluster sizes in free recall from episodic mem-ory, the limits of perfect recall from immediate memory, proactiveinterference effects, the limits of cued partial reports, subitizingspans, multiple-object tracking, and the limits of consistentlymapped search. The range of the evidence is staggering: Cowan’sTable 1 lists 41 different “selected key references” (p. 90) in 17different research domains, all pointing to the magical numberfour.

Paul Verhaeghen, John Cerella, and Chandramallika Basak, Departmentof Psychology and Center for Health and Behavior, Syracuse University.This research was supported in part by National Institute on Aging GrantAG-16201. Marc Howard and Kara Bopp provided valuable comments ona draft of this article.Correspondence concerning this article should be addressed to PaulVerhaeghen, Department of Psychology, 430 Huntington Hall, SyracuseUniversity, Syracuse, NY 13244-2340. E-mail: pverhaeg@psych.syr.edu

Journal of Experimental Psychology: Copyright 2004 by the American Psychological AssociationLearning, Memory, and Cognition2004, Vol. 30, No. 6, 1322–13370278-7393/04/$12.00 DOI: 10.1037/0278-7393.30.6.1322

1322

We are faced, then, with opposing expectations. On one hand,the handful of recent studies cited in the first paragraph point to afocus of attention that can accommodate no more than a singleitem. On the other hand, there is a large and venerable body of literature supporting the claim that the focus of attention can holdfour items comfortably. We set up the present experiment to try toreconcile the two outcomes. Before we turn to the rationale anddescription of our study, we review the evidence for a narrowfocus in more detail and take up some existing explanations for thediscrepancy between narrow-focus and wide-focus results.

Evidence for a Narrow Focus of Attention in WorkingMemory

Prior to Verhaeghen and Basak’s (in press), at least four studies,using different paradigms, have converged on the conclusion thatthe width of the focus of attention is not four items but one item.The decisive variable is item accessibility, that is, the number of items that can be retrieved without necessitating a cost in responsetime. Step or threshold effects in the Response Time



WorkingMemory active load function, such as seen in Figure 1, are takenas an indication that the focus of attention is restricted to thenumber of items prior to the step—in the case of Figure 1, this isonly a single item.In one of the precedents to the Verhaeghen and Basak (in press)result, Garavan (1998) asked participants to keep a separate run-ning count of triangles and squares that appeared in a randomsequence on a computer screen. Presentation was self-paced, al-lowing for the recording of response times. Response times wereabout 500 ms slower when the stimulus shape was changed fromthe previous trial than when both successive stimuli had the sameshape. Garavan interpreted the results as showing that participantskept separate mental counters for each stimulus shape, and thatswitching between counters in working memory was a time-consuming process. This, in turn, suggests that only one of thecounters was immediately accessible for updating and hence re-sided in the focus of attention, whereas the other counter was heldin temporary storage outside the focus of attention. Further re-search using this paradigm has shown that the cost associated withswitching between counters is largely due to retrieval demands(Voigt & Hagendorf, 2002).In another study, using a speed–accuracy methodology in aprobe-recognition task with serially presented nine-word lists,McElree (1998) found that retrieval speed for the last item wasfaster than retrieval speed for any of the other items presented (seeMcElree & Dosher, 1989; Wickelgren, Corbett, & Dosher, 1980,for earlier demonstrations of this effect). McElree interpreted thisresult as implying that the last item processed resided in the focusof attention, whereas the items presented previously were stored ina less accessible state. In addition, in this study, items werecategorized semantically, with three items for each category; allitems from the category last presented were retrieved faster thanthe other items in the list. McElree (1998) interprets this result asindicative of some amount of clustering in the stored representa-tions in working memory.In a third study, again using a speed–accuracy procedure, McEl-ree (2001) found that access times in an identity-judgment

-back task were faster for the



1 condition than they were in either the



2 or the



3 conditions. His interpretation was that in the



1 conditions, the items presented could not all be accommo-dated in the focus of attention, necessitating slow search processeson a significant portion of the trials. A model was fitted to the data,depicting performance as a mixture of a fast matching process

Figure 1.

Data for Experiment 1 from Verhaeghen and Basak (in press) depicting response time (ms) andaccuracy (percent correct) as a function of

in a group of younger adults in an identity-judgment

-back task.The response time data show a step function in passing from



1 to



1, indicating immediate access to nomore than a single item, taken as evidence for a focus of attention containing only a single item; the accuracydata show a gradual decline. Error bars denote standard errors. Adapted from Figure 3 in “Aging and Switchingof the Focus of Attention in Working Memory: Results From a Modified N-Back Task,” by P. Verhaeghen &C. Basak, in press,

Quarterly Journal of Experimental Psychology

1323

WORKING MEMORY WORKOUT

(operating when the item is residing in the focus of attention) anda slow search process (when it is not). In a follow-up experiment,McElree contrasted a standard 3-back condition with a condition inwhich participants had to respond positively to any item up to andincluding the item 3-back. The latter condition challenged partic-ipants to maintain three rather than one item in an accessible state;speed of performance suffered accordingly, suggesting that onlyone item could be retained in the focus of attention with highprobability.In a fourth study, Oberauer (2002) used a working memory-updating task in which participants had to update a set of single-digit numbers stored in working memory one by one using simplearithmetic operations. Oberauer found that updating occurredfaster when an item had just been updated than it did when otheritems had intervened (Oberauer, 2002). This again suggests thatworking memory had privileged access only to the single item lastupdated.

Theoretical Resolutions

There have been two attempts to reconcile the conflicting find-ings concerning the size of the focus of attention at a theoreticallevel. One proposal focused on the requirements of the task (theserial-attention explanation; Garavan, 1998), and the other focusedon structural characteristics of working memory (the tripartite-architecture explanation; Oberauer, 2002).Garavan’s (1998) explanation draws on the specifics of the tasksinvolved. In all of the narrow-focus procedures, attention is di-rected serially to different elements that are either stored in work-ing memory or are being encoded into working memory. The serialrequirement probably necessitates a controlled switch of attentionto successive items. For example, in the Garavan running-counttask, a switch in stimulus shape necessitates access to the storedcount for each shape and a controlled update of that count. In theMcElree (1998) probe-recall task, attention is directed to eachsuccessive item, which is then encoded explicitly into workingmemory. In the McElree (2001)

-back task, the item shown on thescreen has to be explicitly compared with an item stored inworking memory; the stored representation then has to be eitherstrengthened or updated. In the Oberauer (2002) task, attentionneeds to be directed to the stored representation of the appropriateitem, which then needs to be consciously updated. Thus, whatdistinguishes narrow-focus outcomes from wide-focus outcomesmay be the requirement to shift attention serially within either thestored representations or the to-be-encoded items in the narrow-focus contexts. Oberauer uses the term object switching for thisattentional process, and Garavan uses the term attention switching;we (see also Verhaeghen & Basak, in press, and Voigt & Hagen-dorf, 2002) prefer to use the term focus switching—the switchingof the focus of attention from one item to another. It is this focusswitching that precipitates the characteristic jump in response timewhen an active working memory load of two (or more) rather thanof one must be maintained.Oberauer (2002) has advanced a different kind of explanation.He argues that the conflicting results are not due to the processesinvolved but that, rather, the two outcomes derive from differentmemory structures. Oberauer argues for the existence of a concen-tric tripartite store with functionally distinct regions. In effect, thetripartite store combines the architecture proposed by Cowan (e.g.,1995) with the architecture proposed by McElree (2001) andGaravan (1998). The three regions are characterized by theirincreasing accessibility to cognitive processes. The first and out-ermost tier, the activated part of long-term memory, serves mainlyto store information over brief periods; the processes operating inthis region concern encoding and maintenance. The second tier,lying within the activated region, is a capacity-limited region of direct access corresponding to Cowan’s focus of attention, wherea limited number of items are stored that are likely to be selectedfor subsequent processing. The third and innermost tier, which lieswithin the region of direct access, is the focus of attention, corre-sponding to the focus of attention proposed by McElree andGaravan and containing only a single chunk of information,namely, the chunk actually selected as the object of the nextcognitive operation.Oberauer’s (2002) theory was designed to accommodate datafrom a variant of the memory-updating task. In his version of thetask, two sets of single-digit numbers were presented, and set sizewas either one or three. One set of numbers was marked forsubsequent updating (using mental arithmetic operations); theother set was marked for storage only. First and foremost, Ober-auer observed the focus-switching effect indicative of a narrowfocus of attention: When the current item matched the last itemupdated, response times were faster than when a different itemneeded to be accessed. Second, Oberauer found that speed of updating was dependent on the size of the to-be-updated set andnot on the size of the to-be-stored set. His interpretation was thatthe two sets were stored in distinct regions of working memory.The storage-only set was placed in the activated part of long-termmemory. The to-be-updated set needed to be held in the moreaccessible region of direct access, ready to be retrieved into thefocus of attention. Because items compete for activation in theregion of direct access, retrieval times in that store were dependenton the number of items stored.

The Size of the Focus of Attention: A StructuralLimitation or an Allocation Artifact?

The limitations imposed on the focus of attention by both theserial-attention explanation and the tripartite-architecture explana-tion are hard and fast. Both theories treat a focus of one as builtinto the working memory machinery. Oberauer’s (2002) position isthat the focus of attention “holds at any time the one chunk that isactually selected as the object of the next cognitive operation” (p.412). McElree (2001) states that “focal attention is able to maintainone temporally extended event only” (p. 820). Garavan (1998)claims that “this subset [i.e., the attentional focus as a subset of short-term memory] contains just one item” (p. 275). The singleexception is that this event may contain multiple items if they canbe coded into a chunk that forms a unitary processing epoch(Garavan, 1998), for instance a set of categorically related words(McElree, 1998). The limitations imposed on the region of directaccess, on the other hand, have been taken to be more fluid and todepend on constraints such as overwriting or crosstalk (Oberauer,2002).It seems to us that the position that the focus of attention islimited to a single item is excessively rigid even under circum-stances that require serial switching. One of the basic suppositionsof attentional theory has been that attentional capacity can beallocated flexibly across the perceptual–cognitive field, and thatthis allocation follows the needs of the participant, the demands of

1324

VERHAEGHEN, CERELLA, AND BASAK