Psychological Bulletin © 2016 American Psychological Association

2016, Vol. 142, No. 7, 758 –799 0033-2909/16/$12.00 http://dx.doi.org/10.1037/bul0000046

What Limits Working Memory Capacity?

Klaus Oberauer Simon Farrell

University of Zurich University of Western Australia and University of Bristol

Christopher Jarrold Stephan Lewandowsky

University of Bristol University of Bristol and University of Western Australia

We review the evidence for the 3 principal theoretical contenders that vie to explain why and how
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

working memory (WM) capacity is limited. We examine the possibility that capacity limitations arise from
This document is copyrighted by the American Psychological Association or one of its allied publishers.

temporal decay; we examine whether they might reflect a limitation in cognitive resources; and we ask
whether capacity might be limited because of mutual interference of representations in WM. We evaluate
each hypothesis against a common set of findings reflecting the capacity limit: The set-size effect and its
modulation by domain-specificity and heterogeneity of the memory set; the effects of unfilled retention
intervals and of distractor processing in the retention interval; and the pattern of correlates of WM tests.
We conclude that—at least for verbal memoranda—a decay explanation is untenable. A resource-based
view remains tenable but has difficulty accommodating several findings. The interference approach has
its own set of difficulties but accounts best for the set of findings, and therefore, appears to present the
most promising approach for future development.

Keywords: working memory, capacity limits, decay, resources, interference

Working memory (WM) is the system that holds mental repre- information people can remember over short periods of time (in
sentations available for processing. Its limited capacity is a limit- the order of seconds), but there are reasons to believe (discussed
ing factor for the complexity of our thoughts (Halford, Cowan, & below) that the capacity limit also applies to people’s ability to
Andrews, 2007; Oberauer, 2009). Measures of WM capacity have make information in the current environment simultaneously avail-
been identified as major determinants of cognitive development in able for processing.
childhood (Bayliss, Jarrold, Gunn, & Baddeley, 2003) and in old Hypotheses about what limits WM capacity can be organized
age (Park et al., 2002; Salthouse, 1994), as well as of individual into three groups: (a) Some theories assume that representations in
differences in intellectual abilities (Conway, Kane, & Engle, 2003; WM decay over time, unless decay is prevented by some form of
Jarrold & Towse, 2006). Understanding why WM capacity is restoration process such as rehearsal. According to this view, WM
limited is, therefore, an essential step toward understanding why has limited capacity because only a limited amount of information
human cognitive abilities are limited, why individuals differ in can be rehearsed before it fades away into an irrecoverable state
these abilities, and how abilities develop over the life span. (Baddeley, Thomson, & Buchanan, 1975; Schweickert & Boruff,
In this article we use the term WM capacity in a descriptive 1986). (b) Alternatively, WM capacity has been characterized as a
sense, referring to the fact that people can hold only a limited
limited resource that needs to be shared by representations held
amount of mental content available for processing. The capacity
available simultaneously and processes to be carried out at the
limit is usually operationalized as a limit on how much new
same time (Case, Kurland, & Goldberg, 1982; Just & Carpenter,
1992; Ma, Husain, & Bays, 2014). This resource could be contin-
uous or discrete, and the discrete variant is often referred to as a
“slot model” (Cowan, Rouder, Blume, & Saults, 2012). (c) A third
This article was published Online First March 7, 2016.
approach is to explain the limited capacity of WM as arising from
Klaus Oberauer, Department of Psychology–Cognitive Psychology,
University of Zurich; Simon Farrell, School of Psychology, University of interference between representations that do not decay on their
Western Australia and School of Experimental Psychology, University of own and are not resource-limited (Nairne, 1990; Oberauer &
Bristol; Christopher Jarrold, School of Experimental Psychology, Univer- Kliegl, 2006; Saito & Miyake, 2004).
sity of Bristol; Stephan Lewandowsky, School of Psychology, University After over 50 years of research on this topic, experimental
of Western Australia and School of Experimental Psychology, University psychologists have accrued a large and detailed database of rele-
of Bristol. vant studies. Perhaps unsurprisingly, the existing data do not
The research reported in this article was supported by a grant from the
appear to universally support any one of the three accounts of WM
Swiss National Science Foundation (100014_135002) to Klaus Oberauer.
Correspondence concerning this article should be addressed to Klaus
capacity. Given this state of affairs, it is useful to step back and ask
Oberauer, University of Zurich, Department of Psychology–Cognitive Psy- how well each of the three explanatory approaches outlined above
chology, Binzmühlestrasse 14/22, 8050 Zürich, Switzerland. E-mail: accord with the data, and which data are particularly diagnostic.
k.oberauer@psychologie.uzh.ch The aim of the present article is to evaluate critically the explan-

758
 WM CAPACITY 759

atory power of these three hypotheses in light of a common set of models can never rule out the entire set of possible models incor-
findings pertinent to the capacity limit of WM. porating a particular hypothesis.
In light of these considerations our approach in this review is the
Terms of the Competition: Analytical Approach following: We try to identify, for each of the three hypotheses
under investigation, predictions that follow from it in the context
Our review focuses primarily on evaluating each hypothesis on of all existing theories or models that incorporate that hypothesis
its own as an explanation of the capacity limit of WM, for two as the main cause of the WM capacity limit. Table 1 provides an
reasons. First, explanations by a single hypothesized mechanism or overview of the theories we used as context to determine the
process are preferable over multicausal explanations because they predictions of each hypothesis. We selected these theories because
are more parsimonious. Second, analyzing each hypothesis in they explain the WM capacity limit fairly unambiguously accord-
isolation enables us to identify which empirical findings can be ing to only one of the three hypotheses under investigation; this
explained by that hypothesis on its own, and which findings
excludes many theories that draw on a combination of hypotheses,
challenge it. This analytical approach will be informative even for
or that make no unambiguous assumptions as to what causes the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

theories that combine multiple causes of the capacity limit. Toward

WM capacity limit. Where we find that a prediction derived from
This document is copyrighted by the American Psychological Association or one of its allied publishers.

the end of this article we will, therefore, consider the potential for
a hypothesis in the context of all theories in Table 1 is borne out
combining different mechanisms to move toward a complete
by the data we regard the evidence as strongly supporting the
model of WM capacity.
hypothesis. Conversely, where we find that a prediction is not
Evaluating hypotheses in isolation is potentially hazardous be-
supported empirically, we regard that as a challenge to proponents
cause the predictions following from each hypothesis depend on
other assumptions with which they are combined in a theory or of the hypothesis: Although it remains possible that the hypothesis,
model (Newell, 1973). This problem can be circumvented by when put in the context of a new model, meets that challenge, we
evaluating each hypothesis about what underlies the WM capacity argue that the burden of proof then lies with the proponents of that
limit in the context of a fully fleshed-out computational model of hypothesis to present such a model.
WM. In our review we draw on computational models incorporat- In addition, we ask whether the hypothesis, in conjunction with
ing the hypothesis in question where possible. At the same time additional assumptions that are made by some but not all theories
this approach engenders another limitation: Evaluating a hypoth- incorporating that hypothesis, can explain a given finding. Where
esis in the context of a particular theory or model can only that is the case, the finding provides support for the hypothesis, but
determine to what extent the conjunction of all assumptions in the the support is weaker than in cases where the hypothesis predicts
model is able to explain certain findings; it is difficult to attribute the finding, because the explanation depends on additional as-
the empirical success or failure of a model to one hypothesis sumptions that are made only by some theories incorporating the
incorporated in that model. For instance, if one interference model hypothesis. To summarize, our evaluation of each hypothesis with
fails to explain an important phenomenon, proponents of an inter- respect to each finding aims to determine which of four logical
ference explanation of WM capacity can always argue that the relations holds between the hypothesis and the finding: (a) The
interference hypothesis might work better in the context of another finding is predicted by one of the hypotheses, meaning that it
model. As the number of possible models incorporating an as- follows from the hypothesis without any additional assumptions
sumption is potentially infinite, empirically ruling out individual that are not shared by all known theories incorporating the hypoth-

Table 1
Theories Used as Context to Derive Predictions From Hypotheses

Decay Resources Interference

Phonological-loop model (Baddeley et al., Neo-Piagetian general resource model Feature model (Nairne, 1990)
1975; Schweickert & Boruff, 1986) (Case et al., 1982)
Limited-capacity trace-decay theory Multiple-resource model (Alloway et Interference model (Oberauer & Kliegl, 2001,
(Jensen, 1988; Salthouse, 1996) al., 2006; Logie, 2011) 2006)
Primacy model (Page & Norris, 1998) 3CAPS (Just & Carpenter, 1992) SOB (Lewandowsky & Farrell, 2008b) and
SOB-CS (Oberauer, Lewandowsky, et al.,
2012)
Task-switching model (Towse & Hitch, Slot model (Luck & Vogel, 2013; Temporal-clustering-and-sequencing model
1995; Towse, Hitch, & Hutton, 2000) Cowan et al., 2012) (Farrell, 2012)
Computational phonological loop model Resource models of visual WM (Ma
(Burgess & Hitch, 1999, 2006) et al., 2014)
Time-based resource-sharing model
(Barrouillet et al., 2004; Camos et al.,
2009)
Note. Theories in the table were selected because they attribute the working memory (WM) capacity limit unambiguously to decay, limited resources,
or interference, respectively. Some theories of WM were not included because they combine two or three of the hypotheses, or make no clear assumptions
about what causes the capacity limit. We regard the time-based resource-sharing model as a decay model because, unlike resource models, it assumes that
decay is the root cause of the capacity limit of WM, and an attentional resource is assumed to play a role only insofar as it counteracts decay (through
refreshing). Without decay, there would be no role for a resource in that model.
 760 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

esis; (b) the finding can be explained by one of the hypotheses, and their short-term memory (STM) for visual objects declines
meaning that the finding is predicted by the hypothesis together with an increasing number of objects to be remembered (Luck &
with additional assumption that have been proposed in an existing Vogel, 1997; Miller, 1956). This ubiquitous observation has been
theory, or that can reasonably be made; (c) the finding challenges referred to as the effect of memory set size or of memory load. It
one of the hypotheses, meaning that the hypothesis, in the context is a direct reflection of the WM capacity limit: The concept of
of any known theory, predicts the absence of the finding, and (d) limited WM capacity implies that performance gets worse as the
the finding is consistent with the hypothesis, meaning that the amount of information to be held in WM is increased and even-
finding provides no evidence in favor or against the hypothesis. tually surpasses that limit. Therefore, any explanation of WM
capacity must explain the set-size effect.
B: Effects of retention interval and distractor processing.
The Playing Field: Findings for Evaluating
Representations in WM are vulnerable to processing during a
Hypotheses About WM Capacity
retention interval (RI) placed between study and test, which can
We evaluate all three hypotheses against a set of findings that lead to forgetting in the order of seconds. Experimental control
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

we regard as informative for our question, based on the conjunc- over cognitive processes during the RI is often achieved by asking
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tion of two criteria: relevance and diagnosticity. We use the first people to engage in a specific processing task—such as counting,
criterion, relevance, to delimit a set of phenomena that are gener- mental arithmetic, or reading aloud—while maintaining a memory
ally agreed among researchers to be manifestations of the capacity set. We will refer to these processing demands as distractor tasks.
limit of WM. We use the second criterion, diagnosticity, to select Distractor tasks have been placed after presentation of the entire
findings within the set of relevant phenomena that count as evi- memory set, as in the Brown-Peterson paradigm (J. Brown, 1958;
dence in favor or against at least one of the three hypotheses under Jarrold, Tam, Baddeley, & Harvey, 2011; Peterson & Peterson,
investigation. Specifically, we regard as diagnostic any finding 1959), interleaved with presentation of individual items, as in the
that stands in one of three logical relations (out of the four defined complex-span paradigm (Daneman & Carpenter, 1980; Turner &
above, excluding consistency) to at least one of these hypotheses: Engle, 1989), or interleaved with recall of individual items (Le-
The hypothesis predicts the finding, it can explain the finding, or wandowsky, Duncan, & Brown, 2004; Lewandowsky, Geiger, &
it is challenged by the finding. Oberauer, 2008). Distractor processing during the RI typically has
a detrimental effect on memory accuracy. There is general agree-
ment that this detrimental effect reflects the limited capacity of
Relevant Phenomena
WM, because the processing demand is thought to place an addi-
Concerning the first criterion—relevance—we consider three tional load on this capacity, thereby reducing the effective capacity
broad phenomena as manifestations of the WM capacity limit: (a) available for holding the memory set. Therefore, we regard the
The set-size effect on accuracy, (b) the effects on memory of effects of distractor processing during the RI as a phenomenon that
manipulations of the retention interval and the events during that every viable explanation of the WM capacity must account for. In
interval, and (c) the pattern of correlations among tests thought to this context we will also discuss findings on the effect of varying
measure WM capacity and related cognitive tasks. Each of those the duration of an “unfilled” RI, that is, an interval between study
three phenomena, in turn, is characterized by a number of findings and test during which mental activity is not experimentally con-
that specify the precise nature of the phenomenon. Every viable trolled, because some of these findings are diagnostic with regard
theory of WM capacity must explain these three phenomena, to the three hypotheses.
including the detailed findings characterizing them. The informa- C: Individual differences. A viable explanation of WM ca-
tive findings we include in this review are the findings that reflect pacity should also explain, or at least be consistent with, findings
aspects of these three phenomena, and at the same time are concerning individual differences—including age differences—in
diagnostic for the three hypotheses. WM capacity (Conway, Jarrold, Kane, Miyake, & Towse, 2007),
We next briefly introduce each phenomenon, together with our because much of the evidence for a capacity limit applying
reasons for selecting it. Our review will be organized by these broadly to all kinds of complex cognition arises from that
three broad phenomena. In each section, we explain how each of research. In particular, correlational data show that the WM
the three hypotheses accounts for the phenomenon reviewed in it. capacity limit has a high degree of generality across contents
In doing so we will spell out the predictions following from each and testing procedures (Kane et al., 2004; Oberauer, Süß,
hypothesis, the diagnostic findings speaking to these predictions, Schulze, Wilhelm, & Wittmann, 2000; Unsworth, Fukuda,
and the additional assumptions by which each hypothesis needs to Awh, & Vogel, 2014). If different measures of WM capacity
be embellished to explain specific findings. Tables 2 to 4 provide limits were only weakly correlated, the very idea of a singular
an overview of these findings, together with our assessment of WM capacity would be questionable. In addition, correlational
their logical relation to each of the three hypotheses. In what data are informative because—as we will explain below—
follows we will cross-link discussion of each finding in the text different hypotheses about WM capacity make different predic-
with the corresponding entries in Tables 2 to 4 using letters to refer tions about which other variables are correlated with measures
to the three broad phenomena, and numbers to refer to individual of WM capacity.
findings characterizing the phenomenon in question.
A: Set-size effects. As the amount of material a person tries to
On the Choice of Informative Findings
hold in WM increases, memory accuracy decreases. For instance,
people find it increasingly more difficult to remember a list of A comparative evaluation of hypotheses against data in a field
digits or words for immediate serial recall as the list gets longer, as broad as WM capacity is necessarily selective. By making the
 WM CAPACITY 761

reasons for our selection of data explicit we tried to rein in to some In short, our selection of evidence for this review does not
extent the arbitrariness and potential bias involved in prioritizing reflect a judgment of the importance of a set of findings for WM
some pieces of evidence over others. We identified three basic research in general. Rather, it reflects the relevance of findings for
phenomena that are commonly regarded as direct expressions of the specific question we ask: How best to explain the limited
the capacity limit of WM, and we argue that every successful capacity of WM?
explanation of the WM capacity limit must explain these phenom- Although we have endeavored to be explicit about the rea-
ena. A viable explanation of these basic phenomena must be in sons for our selection, and impartial in the choice of findings
agreement with the empirical details known about them, and included, we expect that our choice of informative findings will
therefore, we consider the research characterizing set-size effects, be questioned by some. We hope that this will initiate a debate
effects of distractor processing during the RI, and correlational about which findings should be regarded as benchmarks for
findings in some detail. evaluating the hypotheses under consideration—in other words:
At the same time, we exclude from consideration a vast number What needs to be explained by a viable explanation of the WM
of well-established empirical findings about WM, such as the capacity limit?
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

effects of serial position in memory for lists (Nipher, 1878), the

This document is copyrighted by the American Psychological Association or one of its allied publishers.

effects of presentation rate and presentation modality (Penney, The Contenders: Three Hypotheses About What
1975), the effects of irrelevant sounds on verbal serial-order mem- Limits WM Capacity
ory (Jones & Macken, 1993), or the effects of cueing attention to We start the competition with an introduction of the three
an item within WM (Lepsien & Nobre, 2006). These and many hypotheses under consideration. After that we will evaluate each of
other findings are highly informative about the mechanisms of them in light of diagnostic findings, organized by the three broad
WM, but they do not speak as directly to the capacity limit of WM phenomena outlined above.
as the three phenomena introduced above, because they are not
generally agreed to be direct manifestations of the capacity Decay
limit. For the same reason we excluded the set-size effect on
response times (Lange, Cerella, & Verhaeghen, 2011; Stern- The first hypothesis we investigate is that WM capacity is
berg, 1966): Whereas the increasing time for access to WM limited by the rapid decay of WM representations over time.
contents with increasing set size could reflect the capacity limit Theories assigning an important role to decay invariably assume
of WM, it could equally reflect the longer duration of search that decay can be counteracted by one or several forms of resto-
through a larger set, independent of the capacity limit. Hence, ration. Earlier research focused primarily on subvocal articulation
unlike the set-size effect on accuracy, the set-size effect on as a process for maintaining verbal representations in WM (Bad-
response times is not unambiguously a manifestation of the WM deley et al., 1975). A domain-specific rehearsal mechanism based
capacity limit. on the spatial orientation of attention might also be available for
We also excluded from consideration several findings that be- maintaining spatial information (Awh, Jonides, & Reuter-Lorenz,
long to one of the three broad phenomena we identified above, but 1998; but see Belopolsky & Theeuwes, 2009).
that are not diagnostic. For instance, the finding that distractor More recently, proponents of decay introduced the assumption
processing impairs memory for individual visual features as much that verbal memoranda can be maintained by at least two pro-
as memory for feature bindings (Allen, Baddeley, & Hitch, 2006; cesses; subvocal articulation and attention-based refreshing
(Camos, Lagner, & Barrouillet, 2009).1 Refreshing is conceptual-
Morey & Bieler, 2013) is an instance of the effect of distractor
ized as a domain-general process of strengthening memory traces
processing on memory. Yet, this finding does not help to adjudi-
by directing central attention to them (Barrouillet, Bernardin,
cate between the three hypotheses under consideration, because
Portrat, Vergauwe, & Camos, 2007; Raye, Johnson, Mitchell,
none of them implies that the WM capacity limit should or should
Greene, & Johnson, 2007). Central attention is thought to be
not apply equally to features and to bindings. Likewise, the strong
limited to one process at a time, thereby creating a bottleneck
correlation of WM capacity with fluid intelligence (Conway et al.,
(Pashler, 1994): Central attention can be devoted to refreshing only
2003) is perhaps the one correlational finding about WM that has
during time intervals in which it is not recruited by another
received more attention than any other, but it is not diagnostic,
cognitive process. One implication of this assumption is that
because all three hypotheses explain it in essentially the same way:
refreshing, like articulatory rehearsal, has to proceed sequentially,
Reasoning ability is limited by the amount of task-relevant infor-
strengthening one memory item at a time. On these assumptions
mation that we can hold in WM at the same time, and the details the capacity of WM results from the race between decay and
of this explanation have more to do with our assumptions about restoration: People can maintain as much information as they can
reasoning than with our assumptions about why WM capacity is reliably rehearse or refresh before it decays beyond recovery.
limited.
Finally, we limit the scope of our empirical review to behav- Resources
ioral data from healthy individuals, excluding data from special
populations with certain pathologies or neurological damage, as The concept of a limited resource is often used informally in
well as data from neuroscience. Whereas these data are highly cognitive psychology to describe the fact that the efficiency and
informative about the mechanisms of WM, we found them not
to be diagnostic for adjudicating between the three hypotheses 1
We will use “restoration” as the general term for any hypothetical
about the nature of the WM capacity limit, because the hypoth- process by which decaying memory traces are restored, encompassing
eses do not make differential predictions for these kinds of data. articulatory rehearsal, visual-spatial rehearsal, and attentional refreshing.
 762 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

accuracy of information processing is limited. When used in this assuming that the resource consists of discrete slots because the
way, the term resource does not refer to an explanatory construct latter’s lesser flexibility can only accentuate but not resolve any
but rather summarizes a set of phenomena in need of explanation. challenges.
In contrast to this informal use of the term, there is a more formal,
well-defined resource concept (Anderson, Reder, & Lebiere, 1996;
Interference
Ma et al., 2014; Navon & Gopher, 1979; Tombu & Jolicoeur,
2003). Well-defined resource concepts differ in their details but Interference accounts of the WM capacity limit assume that our
they share a set of assumptions: A resource is a limited quantity ability to hold several representations available at the same time is
that enables a cognitive function (e.g., holding a representation limited by mutual interference between these representations.
available) or process (e.g., retrieving or transforming a represen- Three forms of interference have been identified theoretically; they
tation), such that its efficiency and success probability increases are schematically illustrated in Figure 1.2
monotonically with the resource amount allocated to it. The re- First, interference arises from the confusion between item rep-
source can be allocated flexibly to a broad range of representations resentations. Interference by confusion arises naturally from a
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

and processes, and it can be subdivided into portions allocated in retrieval mechanism called competitive queuing, which is incor-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

parallel to different recipients. Resource sharing implies that pri- porated in many formal models of WM (Hurlstone, Hitch, &
oritizing one cognitive function or process occurs at the expense of Baddeley, 2014; Lewandowsky & Farrell, 2008b). Competitive
others that need the same resource at the same time. It is this queuing describes retrieval from WM as a competition between
well-defined resource concept, rather than the unconstrained in- several retrieval candidates that are activated at retrieval. The more
formal notion of resources, that we consider as a possible expla- a representation is activated, the more likely it is to be selected for
nation of the capacity limit. retrieval. Some models assume that the activation is continuously
The precise predictions of a resource theory depend on the maintained during the retention interval (Page & Norris, 1998),
assumptions the theory makes in two regards: Which cognitive whereas others assume that representations are reactivated at re-
functions or processes need the resource, and how the resource trieval through context cues (Burgess & Hitch, 2006). Context
quantity assigned to a function or process translates into an ob- cues can be representations of the present list context (discrimi-
servable level of performance (i.e., the so-called performance- nating the current memory set from other memory sets in previous
resource function; Norman & Bobrow, 1975). Here we consider trials), ordinal list positions (discriminating items within lists), or
the family of resource theories characterized by the following spatial locations (discriminating items in spatial arrays). Confusion
assumptions: (a) Maintaining a representation in WM requires arises when competing representations are activated as strongly as,
allocating some amount of a resource to it for the duration of or even stronger than, the target representation. This happens when
maintenance, and the success in maintaining a representation is a contextual cues are not sufficiently distinctive from each other to
monotonically increasing function of the resource amount allo- selectively cue the target information (see Figure 1A). As an
cated to it. (b) Carrying out a cognitive operation requires allocat- intuitive analogy for interference by confusion, think of reading a
ing part of the resource to it for the duration of the operation; the printed text: With smaller line spacing the lines are harder to
speed and accuracy of the operation is a monotonically increasing distinguish, and the chance increases that the reader’s eye jumps to
function of its resource share. (c) Maintenance and cognitive the wrong line. Interference by confusion is a feature of most
operations require the same resource, at least within a broad computational models of WM, including those that attribute the
content domain (i.e., verbal, visual, and spatial). capacity limit to decay (e.g., Burgess & Hitch, 1999; Oberauer &
Whereas most resource theories assume that resources can be Lewandowsky, 2011; Page & Norris, 1998).
subdivided into quantities of any size, a more constrained version A second form of interference arises from superposition of
of resource theory—slot theory— has gained popularity in the several distributed representations. Distributed representations can
literature on visual WM (Fukuda, Awh, & Vogel, 2010; Luck & be patterns of activations over a set of units in a neural network, or
Vogel, 2013; Zhang & Luck, 2008). According to slot theories, the patterns of connection weights between units. When several such
resource underlying short-term maintenance of information con- patterns are encoded, they are added together (i.e., superimposed),
sists of a limited number of discrete units or slots that can be and as a consequence, each individual pattern is distorted by the
allocated to individual items or chunks. As a consequence, the others (see Figure 1B). The amount of distortion increases with the
resource is not infinitely divisible—when K slots are available,
WM can at best hold representations of K chunks. If a task requires 2
A fourth form of interference arises when memory items are main-
holding more than K elements in WM, only a subset of K elements tained by persistent activation of their representations, and these represen-
can be represented in WM and no information is available about tations inhibit each other. Lateral inhibition is a common feature of
any additional elements. Here we are not concerned with the competitive (k-winner-takes-all) networks, and it underlies the buffer
debate between proponents of discrete slots and proponents of model developed by Davelaar, Goshen-Gottstein, Ashkenazi, Haarmann,
continuous resources (for a systematic comparison in the visual and Usher (2005) as a component of their model of free recall. In this
model each item is represented by a unit that reactivates itself and inhibits
domain see van den Berg, Awh, & Ma, 2014), and instead treat all other units. As set size increases, the number of active units in the buffer
both positions as members of the family of resource explanations increases up to a point where the sum total inhibition a unit receives from
of WM capacity. Because the assumption of continuously divisible all other units exceeds its self-activation, so that the unit’s activation
resources is more flexible than the discrete-resource notion, in the rapidly drops to zero, and the corresponding item is irreversibly forgotten.
Although technically an interference model, the buffer model behaves
following we focus on the hypothesis of continuous limited re- essentially like a resource model (including the irreversibility assumption
sources. Any challenge arising from the data for the continuous discussed in the context of resource theories below). Therefore, we focus
version of the resource hypothesis also applies to the version in this section on the remaining three forms of interference.
 WM CAPACITY 763
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. Three forms of interference. (A) Interference by confusion: Two items, each represented by a single
unit in a neural network, are associated to two partially overlapping contexts. The figure shows the degree of
activation (darkness of shading) at retrieval, using Context 1 as retrieval cue. Because of context overlap, Item
2 is activated little less than Item 1, such that with the addition of random noise, Item 2 could win the competition
for retrieval. (B) Interference by superposition: Distributed representations of two items—shown as vectors and
as patterns of shading of the units of a neural network—are associated to their respective contexts. The
associations are superimposed in the matrix of connection weights between item and context units. At retrieval,
when Context 1 is used as cue, the retrieved vector (Retr. 1) is a superposition of Items 1 and 2 (Item 2
contributing less because of only partial context overlap). The retrieved vector is a distorted version of the
original Item 1. (C) Interference by feature overwriting. Two distributed item representations are shown, together
with the retrieved vector when Item 1 is recalled. The two right-most features, which are shared by two items,
have been overwritten.

number of patterns that are superimposed. Intuitively, interference (Mercer & McKeown, 2010) and tactile (Bancroft, Servos, &
by superposition can be understood in analogy to a printer that Hockley, 2011) STM.
prints two or more words on top of each other, as in a palimpsest: Finally, interference could arise from feature overwriting as
The more words are superimposed on a page, the harder it gets to defined in the feature model of Nairne (1990) and the interference
reconstruct each of them. Interference by superposition arises model of Oberauer and Kliegl (2001, 2006). Like superposition,
naturally in models of WM that use distributed representations the idea of feature overwriting applies to distributed representa-
(G. D. A. Brown, Preece, & Hulme, 2000; Farrell & Le- tions in which each item is coded as a vector of features. Feature
wandowsky, 2002; Matthey, Bays, & Dayan, 2015; Oberauer, overwriting means that when two items share a feature, that feature
Lewandowsky, Farrell, Jarrold, & Greaves, 2012). Direct evidence is overwritten in one of them (Figure 1C). As an analogy, think of
for this form of distortion of WM representations comes from a type-setter with a limited number of types for each letter: When
experiments using stimuli from low-dimensional feature spaces a new text requires a letter that has already been used, the needed
that enable precise control over the features of memoranda. For letters are cut from the older text and pasted into the new text,
instance, Huang and Sekuler (2010) asked participants to repro- leaving gaps that render the older text increasingly illegible. Fea-
duce the spatial frequency of one of two gratings held in WM, and ture overwriting is in some sense the opposite of superposition:
found that the reproduced frequency was biased toward the fre- Interference by superposition leads to distortions of distributed
quency of the other grating (cf. Dubé, Zhou, Kahana, & Sekuler, representations where they differ from each other, whereas feature
2014). Similar biases from distractors have been shown in auditory overwriting leads to distortion of representations where they match
 764 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

(compare Figures 1B and 1C). As a consequence, interference by uted (Cowan, 2005), perhaps with larger weights for more com-
superposition is more severe if the representations interfering with plex chunks if it is assumed that they require a larger resource
each other are dissimilar, whereas interference by feature over- share. The interference hypothesis implies that the degree of mu-
writing is more severe when they are similar. There is some tual interference increases with the number of representations in
evidence for feature overwriting in WM for verbal materials WM, but also depends on the relations of overlap and similarity
(Lange & Oberauer, 2005; Oberauer & Lange, 2008), but not with between them, as we will explain in more detail.
visual materials (Jünger, Kliegl, & Oberauer, 2014). One series of For these reasons we regard evidence on whether memory is a
experiments (Oberauer, Farrell, Jarrold, Pasiecznik, & Greaves, function of the number of elements, their complexity, and/or the
2012) tested the opposing predictions of the two mechanisms of time it takes to restore them as diagnostic for our question. As we
interference and obtained support only for the superposition mech- review in detail below, current findings imply that both the number
anism. of elements in a memory set and their complexity affect perfor-
Evaluation of the interference hypothesis is facilitated by the mance. This pattern (finding A1a in Table 2) has been observed
fact that we can rely on computational models for determining its consistently with both verbal materials (Chen & Cowan, 2005;
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

predictions. The two forms of interference most favored by the Service, 1998) and visual materials (Alvarez & Cavanagh, 2004;
This document is copyrighted by the American Psychological Association or one of its allied publishers.

evidence—interference by confusion of items, and interference Hardman & Cowan, 2015), for different forms of complexity.
from superposition—are implemented in a computational model of Representative data are reproduced in Figure 2. In contrast, there
WM, the SOB-CS model (Oberauer, Lewandowsky, et al., 2012), is no evidence for an effect of time needed for rehearsal or
which allows us to determine what predictions the mechanisms refreshing on memory once other variables—such as the complex-
imply when operating jointly. The combination of interference by ity of the memoranda—are controlled (A1b; Jalbert, Neath, Bireta,
confusion with feature overwriting is implemented in the model of & Surprenant, 2011; Service, 1998).
Oberauer and Kliegl (2001, 2006). Computational models assist in A second piece of evidence shows directly that the set-size
unambiguously deriving predictions from theoretical assumptions, effect arises independently of time: The typical limitation of visual
making the process of evaluating these assumptions in light of data WM to about 2–3 objects has been found even at a retention
more rigorous (Farrell & Lewandowsky, 2010). Therefore, we will interval of zero (Sewell, Lilburn, & Smith, 2014; Tsubomi, Fu-
rely, where possible, on computational interference models to kuda, Watanabe, & Vogel, 2013). This finding (A2 in Table 2) is
unambiguously determine the predictions following from the in- illustrated in Figure 3.
terference hypothesis as specified in this section. Another controversial issue concerning the set-size effect is
In what follows we present the competition between the decay whether materials from different content domains (i.e., the verbal,
hypothesis, the resource hypothesis, and the interference hypoth- visual, or spatial domain) tax the same capacity limit. The set-size
esis across three rounds, one for each of the three broad sets of effect is in part domain specific (A3): Increasing the memory set
findings. Within each round we first provide a brief summary of by adding items from the same domain has been found to impair
the informative findings speaking to the phenomenon discussed in memory more than adding items from a different domain. Dual-set
that round, followed by a discussion of each hypothesis in turn, studies asking participants to remember two sets of materials from
during which we will introduce details on the informative findings different domains (e.g., spatial locations and digits) have consis-
as they become relevant in light of the specific predictions of each tently found a reduced—and sometimes no— effect of the size of
hypothesis. For most findings we present at least one representa- one set on memory for the other, suggesting separate capacity
tive study in a figure that explains the study design and shows the limits for the verbal and the visual-spatial domain (Cocchini,
relevant data. Logie, Della Sala, MacPherson, & Baddeley, 2002; Cowan, Saults,
& Blume, 2014; Fougnie & Marois, 2011; Fougnie, Zughni, God-
win, & Marois, 2015; Oberauer & Kliegl, 2006; Towse &
Round A: The Set-Size Effect
Houston-Price, 2001).
Every test of WM asks people to temporarily hold a set of At the same time there is also robust evidence for a cross-
mental content elements—such as digits, words, sounds, or visual domain set-size effect (A4), implying a domain-general capacity
objects—available for some mental operation. The operation to be limit (Cowan et al., 2014; Oberauer & Kliegl, 2006; Saults &
carried out could consist of reporting the set after a delay, making Cowan, 2007). Set-size effects across domains have been found to
a recognition judgment on elements of the set, or manipulating be more pronounced if the task requires maintenance of bindings
elements in the set. The accuracy of the requested operation between items and their contexts, such as their list positions or
typically declines with increasing size of the set to be held in their locations in space (Depoorter & Vandierendonck, 2009;
WM—also known as the memory load. This set-size effect on Fougnie & Marois, 2011; but see Cowan et al., 2014). Figure 4
accuracy can be regarded the most direct and unambiguous man- shows representative data demonstrating both the cross-domain
ifestation of the capacity limit of WM. set-size effect and the additional domain-specific set-size effect.
A controversial issue tightly linked to the nature of WM capac- The set-size effect is reduced not only for mixed sets from
ity is what scale is most appropriate for measuring WM load. On different content domains, but also with mixed sets of stimuli from
the decay hypothesis, WM load should be measured in terms of the different categories within a domain (see Figure 5). For instance,
time it takes to rehearse or refresh a memory set (Schweickert & lists composed of a set of letters followed by a set of digits are
Boruff, 1986). In contrast, the resource hypothesis and the inter- recalled better than equally long lists consisting entirely of letters,
ference hypothesis assign no role to time per se. According to the and lists composed of digits followed by letters are recalled better
resource hypothesis, WM load should be quantified in terms of the than lists consisting entirely of digits (Sanders & Schroots, 1969).
number of chunks among which the resource needs to be distrib- Likewise, visual arrays of four objects are easier to remember
 WM CAPACITY 765

Table 2
Summary of Informative Findings and Evaluations of Hypotheses in Round A: Findings Characterizing the Set-Size Effect
on Accuracy

Index Finding Decay Resource Interference

A1 Memory depends on the number of elements in a memory set, and on the complexity of ⫺ ⫹ 0
the elements (e.g., number of phonemes in a word, or number of features of a visual
object) (A1a), but not on the duration of reproducing the memory set (A1b)
A2 The set-size effect is also observed with a retention interval of zero ⫺ ⫹⫹ ⫹⫹
A3 The set-size effect is in part domain-specific: Memory sets mixing elements from ⫹ ⫹ ⫹⫹
different content domains are easier to remember than domain-pure sets
A4 Cross-domain set-size effect: Extending a memory set by adding elements from a ⫹ ⫹⫹ 0
different content domain impairs memory
A5 Heterogeneity benefit: Memory is better for heterogeneous sets (consisting of items ⫺ ⫺ ⫹⫹
from different classes) than for homogeneous sets within a domain
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Note. Table entries reflect our judgment of the logical relation between a finding and a hypothesis: The hypothesis predicts (⫹⫹) or can explain (⫹) the
finding, it is consistent with the finding (0) or it is challenged by the finding (⫺); see text for explanation.

when people have to remember the colors of two objects and the Alternatively, refreshing could be assumed to proceed at a rapid
orientations of the other two objects, compared with when they rate, but with each refreshing event only strengthening the re-
need to remember four colors, or four orientations (Olson & Jiang, freshed item by a small amount. With increasing set size, each item
2002), and mixed arrays of shapes and textures are better remem- has to wait longer in between two refreshing events, implying that
bered than pure arrays of one kind of feature (Delvenne & Bruyer, the amount of memory strength lost through decay in between
2004). We will refer to this phenomenon as the benefit of set refreshing events increasingly exceeds the gain in strength through
heterogeneity (A5). Although set heterogeneity within a domain refreshing, leading to a net loss of memory strength over time.
has been investigated less often than the effects of domain com- With these assumptions there is no constant capacity limit— either
binations, the benefit of heterogeneous sets has been observed measured in terms of total refreshing duration or of number of
consistently. In what follows we review how well each of the three items— beyond which any additional WM contents would be
hypotheses accounts for the findings A1 to A5. instantly forgotten. Rather, as the memory set increases, there is an
increasing rate of net loss of memory strength over time, resulting
Decay in an increased rate of forgetting.
The units of measurement of the capacity limit (A1, A2). Even if it does not imply a constant capacity limit on the time
Under a decay account the set-size effect can be explained as an dimension, the decay hypothesis predicts that memory declines as
effect of the time it takes to sequentially restore a memory set the time required for restoration of an item increases. In the
of a given size: Larger sets take longer to rehearse or refresh, following section we show that, on balance, the evidence fails to
increasing the risk of memory contents being lost through decay support that prediction (finding A1b).
before they can be strengthened again. The duration of articu- Is the set-size effect an effect of rehearsal time? The idea of
latory rehearsal can be measured, at least approximately, by the a time limit on WM has initially received support in the verbal
time it takes a person to speak a list of verbal items aloud domain from the word-length effect (Baddeley et al., 1975):
(Mueller, Seymour, Kieras, & Meyer, 2003). On that basis the Lists of words that take longer to say—and therefore arguably
capacity of WM for verbal materials has been estimated to longer to rehearse by subvocal articulation—are harder to re-
correspond to an articulation duration of about 2 s (Schweickert member in order. The correlation between speaking duration
& Boruff, 1986). In contrast, there is no established method for and serial recall accuracy, however, could be because of a third
measuring the duration of refreshing. Vergauwe, Camos, and variable related to both (Lewandowsky & Oberauer, 2008).
Barrouillet (2014) proposed a refreshing rate of 50 ms per item. Two such variables have been identified: First, when the speak-
The capacity of WM for visual materials, which cannot be ing duration and the complexity (i.e., number of syllables) of
maintained through articulation or the allocation of spatial artificial words is varied independently, memory depends on
attention and, therefore, must rely entirely on refreshing, rarely complexity but not duration (Service, 1998). When speaking
exceeds four items (Luck & Vogel, 1997; cf. Cowan, 2001). If duration is varied while holding the number of syllables con-
this capacity limit arises because only about four items can be stant, a word-length effect is found only for a specific set of
refreshed sequentially before they are lost through decay, visual materials but not others (Lovatt, Avons, & Masterson, 2000),
WM representations would have to decay beyond recovery suggesting that the purely time-based word-length effect re-
within 200 ms. This is highly unlikely because it would imply flects a confound between speaking duration and some other
catastrophic forgetting of visual materials whenever central feature of words (see Figure 2B). Second, the number of ortho-
attention is diverted by only a single trial of a choice task, graphic neighbors in the language has recently been identified
which already engages the attentional bottleneck for several 100 as a confounding variable (Jalbert, Neath, Bireta, et al., 2011;
ms. No such catastrophic effects have been observed (Mak- Jalbert, Neath, & Surprenant, 2011). Therefore, the word-length
ovski, Shim, & Jiang, 2006; Ricker & Cowan, 2010). effect appears not to reflect an effect of rehearsal duration, but
 766 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 2. Memory depends on number and complexity of elements in the memory set (finding A1). (A)
Change detection accuracy decreases with increasing number of objects and of features. The left panel shows a
trial with four objects and four features (color, orientation, length, and presence or absence of a black dot);
participants need to decide whether or not there was a change from the memory array to the test array. The right
panel shows data redrawn from Figure 4A in Hardman and Cowan (2015): Accuracy declined with the number
of objects and with the number of features per object. (B) Serial recall depends on complexity of pseudowords
(i.e., the number of phonemes), not on speaking duration per word (Service, 1998). (C) Serial recall as a function
of number of chunks in a list. Chunks could be single words or prelearned word pairs (Chen & Cowan, 2005).
Lenient scoring reflects recall of words regardless of order; strict scoring reflects recall of words in correct list
position. With lenient scoring, accuracy depended nearly exclusively on the number of chunks, regardless of
their complexity (i.e., single vs. two-word chunks); with strict scoring, chunk complexity also affected accuracy.
See the online article for the color version of this figure.

other variables such as word complexity or the density of a memory accuracy. Parmentier, Elford, and Maybery (2005) found
word’s neighborhood in the mental lexicon. that memory for serial order of spatial locations declines with
Attempts to find evidence for a correlation between memory and increasing length of the path connecting subsequent locations, as
rehearsal duration for spatial memoranda have had mixed success:
Smyth and Scholey (1994) manipulated the relation between size
and distance of stimuli in the Corsi block task.3 Displays with 3
In the Corsi-block task, participants see an irregular spatial array of
smaller stimuli separated by a larger distance increased the time “blocks,” which are highlighted in turn, and they try to reproduce the order
for moving between the stimuli at recall but had no effect on of highlighted blocks by pointing at them.
 WM CAPACITY 767
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. The set-size effect for visual arrays is observed even at a retention interval of zero (finding A2),
as shown by Tsubomi et al. (2013): Participants remembered arrays of colored squares of varying set sizes.
After a retention interval of 0 (no delay) or 1 s (delay), memory was tested by a bicolored square in the
location of one array item. Participants decided which of the two colors in the probe matched the original
color in that location in the array. (A) Example trials of the standard memory condition with a 1 s retention
interval (above), and the zero retention-interval condition (below); (B) Data of Experiment 1 of Tsubomi
et al. (2013). Probability of a correct response was calculated from the reported values of Cowan’s K,
an estimate of the number of items available in working memory (WM; Cowan, 2001). Accuracy declined
with set size but was indistinguishable between the two conditions, implying that decay cannot
explain the capacity limit that causes the set-size effect. See the online article for the color version of this
figure.

well as with increasing path complexity (e.g., number of path from the duration of recall. For instance, Dosher and Ma (1998)
crossings). In a review of the relevant literature, Parmentier (2011) investigated serial recall of digits, letters, and single-syllable
came to the conclusion that these effects of path characteristics are words as a function of output duration. They found that proportion
more likely to arise from difficulties during encoding rather than correct was well described by a decreasing function of output
from delays imposed during maintenance. This conclusion meshes duration, regardless of list length and material. These functions,
well with the conclusion that effects of word length on verbal however, differed substantially for spoken recall and recall via
serial recall result from complexity rather than articulation dura- keyboard—the latter took about 50% longer but resulted in equally
tion of the words. good memory performance. Other studies manipulated the pace of
Taken together, neither the word-length effect nor the corre- recall either through instruction or the duration of intervening
sponding movement-length effects in spatial serial recall provide activity, and found no effect on memory (Cowan et al., 2006;
good evidence for a role of rehearsal duration in memory. At the Lewandowsky et al., 2004; Oberauer & Lewandowsky, 2008). In
same time, the data reviewed above do not rule out the more conclusion, the time for overt reproduction of memory items is not
general assumption that the duration of processes during mainte- related to memory performance when confounding variables are
nance and retrieval affects memory. For instance, words from a taken into account (finding A1b). This result questions the assump-
sparse orthographic neighborhood could be harder to retrieve, tion that memory depends on the time for restoring decaying
leading to longer retrieval times and by implication, more decay of traces, inasmuch as restoration involves covert reproduction of the
the remaining list words. In support of this notion, Cowan et al. material, for instance by articulatory rehearsal. There is still room
(1992) observed that recall of all list words was impaired when the for the assumption of a restoration process—such as refreshing—
first three words to be recalled were long compared with when that is not thought to require reproduction of the material.
they were short. However, Lovatt, Avons, and Masterson (2002) A capacity limit without delay. The set-size effect in visual
were able to replicate this effect only with the specific set of words WM cannot be explained as reflecting the race between decay and
used by Cowan, and even then the effect was eliminated when the restoration, because the typical limitation of visual WM to about
analysis was limited to trials in which the first three words were 2–3 objects has been found even at a retention interval of zero
recalled correctly, implying that it is not the recall duration of the (A2). For instance, Tsubomi et al. (2013) presented participants
initial words but recall errors that adversely affected recall of with arrays of a variable number of colored squares for one second,
further words. immediately followed by a single bicolored square in the location
The prediction that recall duration affects memory is further of one of the squares in the memory array. The bicolored square
called into question by findings dissociating memory performance served as a visual mask and as the response probe: Participants had
 768 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 4. Domain-specific and domain-general set-size effects (findings A3 and A4). (A) Initial display and
first updating step of a trial in the experiment of Oberauer and Kliegl (2006). After encoding two or four initial
items, participants worked through eight successive updating steps; each step involved updating of one memory
item. Numerical items (digits) were updated by arithmetic operations; spatial items (locations in the frame) were
updated by mental shifts in the direction of the arrow. (B) Asymptotic accuracy (at sufficiently long presentation
durations for each updating step) for set size 2 (two digits or two locations), set size 2 ⫹ 2 (two digits and two
locations, illustrated in A), and set size 4 (four digits or four locations). Relative to set size 2, accuracy declined
when adding two items from the other content domain (set size 2 ⫹ 2), showing the domain-general set-size
effect. Accuracy declined more when adding two items from the same content domain (set size 4), showing the
domain-specific set-size effect.

to decide which of the two colors of the bicolored square matched Effects of domain and of set heterogeneity (A3–A5). The
the color previously seen in the same location in the memory array. decay hypothesis can offer a straightforward explanation for the
Capacity estimates with this procedure were indistinguishable cross-domain set-size effect (A4 in Table 2) by assuming a
from those with a 1-s retention interval (see Figure 3). domain-general process of restoration, such as attentional refresh-
Proponents of decay could argue that encoding of a visual array ing. Memoranda from different domains—such as words and spa-
into WM is a sequential process, so that at the time of test (i.e., tial locations—must time-share the sequential refreshing mecha-
when the target object is covered by the bicolored mask) some nism. Therefore, adding any additional information that needs to
delay has already elapsed after encoding. This argument faces two be refreshed impairs the chances of surviving decay for all other
problems. First, encoding of colors into WM takes about 50 ms per memoranda, regardless of their content domain (Vergauwe, Bar-
item (Vogel, Woodman, & Luck, 2006). A decay theory would rouillet, & Camos, 2010).
have to assume that representations decay within 150 ms to explain Decay theories can explain the partial domain-specificity of
why WM capacity is limited to about three objects. Second, Sewell set-size effects (A3) by assuming that different content domains
et al. (2014) demonstrated that the stimuli of a visual array are have separate rehearsal processes that can run in parallel. Decay
encoded into WM in parallel: They compared simultaneous and theories agree in assuming that verbal memoranda are maintained
sequential presentation of up to four visual stimuli, displaying each through articulatory rehearsal, and some have argued for an anal-
stimulus in the sequential condition for as long as the entire array ogous spatial rehearsal process based on shifts of spatial attention
in the simultaneous condition. If stimuli were encoded sequen- (Awh et al., 1998). A mixed set of verbal and spatial items could
tially, performance should be worse in the simultaneous condition, be easier to remember than a pure set of either material because
whereas parallel encoding predicts no difference. Sewell and col- parallel articulatory and spatial rehearsal could maintain the verbal
leagues found no difference between presentation conditions, im- and spatial subsets of a mixed set, respectively, without competing
plying that stimuli were encoded in parallel. In the same experi- for time.
ment Sewell and colleagues replicated the observation of a set-size More problematic for decay-rehearsal theories is the benefit of
effect with a negligible retention interval. heterogeneous sets within a domain (A5). For instance, the finding
The finding of a set-size effect without a delay does not rule out that mixed sets of shapes and textures are easier to remember than
an impact of decay over retention intervals longer than one second, pure sets of shapes or of textures (Delvenne & Bruyer, 2004; see
but it shows that the set-size effect—and by implication, the fact Figure 5) would have to be explained by assuming independent,
that WM capacity for visual materials is severely limited— does parallel restoration processes for shapes and for textures. Simi-
not arise primarily from a race between decay and restoration larly, better memory for jointly remembering a list of consonants
during a delay between encoding and test. and a list of digits, compared with remembering two lists of the
 WM CAPACITY 769
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 5. Heterogeneity benefit (finding A5). (A) Example of the mixed (heterogeneous) arrays, combining
shapes and textures, in the change-detection experiment of Delvenne and Bruyer (2004). Participants tried to
remember arrays of two or four stimuli for 0.9 s, and decided whether a single centrally presented probe stimulus
matched one of the stimuli in the array. (B) Accuracy in homogeneous trials (Shapes: two or four shapes;
Textures: two or four textures) and the heterogeneous trials (Mixed: one color and one shape, or two colors and
two shapes). Performance was better for heterogeneous arrays than for both kinds of homogeneous trials,
demonstrating the heterogeneity benefit.

same category (Sanders & Schroots, 1969), would have to be The units of measurement of the capacity limit (A1, A2).
explained by assuming separate, parallel processes of restoration The resource hypothesis entails no commitment concerning what
for digits and for consonants. counts toward the load on WM capacity: If the performance-
One could argue at this point that a decay account of the WM resource function is assumed to be the same for all kinds of
capacity limit could explain the heterogeneity advantage by ap- representations, then the only variable that affects WM perfor-
pealing to additional mechanisms. For example, it could be as- mance is the number of objects or chunks among which the
sumed that, in addition to decay, items in homogeneous sets resource is to be shared. Alternatively, a resource theory can
interfere more with each other than items in heterogeneous sets assume that more complex chunks require more resources to
(see our discussion of interference below). This move would achieve the same level of memory performance. Chen and Cowan
delegate much of the explanation of the set-size effect to (2005) have systematically investigated the contributions of the
interference, raising the question whether decay is still needed number and the complexity of chunks to performance on a verbal
to explain one of the main empirical manifestations of the WM test. They varied complexity by contrasting single-word
capacity limit of WM. chunks to two-word chunks consisting of pairs that participants
Conclusion. The findings characterizing the set-size effect had learned to criterion before WM testing commenced. Chen and
provide little support for the idea that the set-size effect arises from Cowan (2005) found that memory for the occurrence of items on
a race between decay and restoration: The set-size effect is not an a list, regardless of their order, depends on the number of chunks
effect of the time it takes to rehearse the memoranda. The heter- to be remembered (for a replication see Chen & Cowan, 2009). In
ogeneity benefit is difficult to explain by the decay hypothesis. contrast, memory for the serial order of items in a list is better
Perhaps the most decisive evidence against an explanation of the
characterized as a function of the complexity of the chunks (see
set-size effect in terms of decay is the fact that the set-size effect
Figure 2C).
is observed even in the absence of any time interval over which
For visual stimuli, Alvarez and Cavanagh (2004) noted that
decay could express itself.
performance on change-detection tasks correlates with the visual
complexity of the to-be-remembered visual objects. For instance,
Resources change detection is better for arrays of colored squares than for
Assuming that a resource is needed for maintenance in WM, the arrays of Chinese characters when the number of objects is the
resource hypothesis provides a straightforward explanation of the same. Awh, Barton, and Vogel (2007) argued that this finding
set-size effect: As the number of representations held in WM merely reflects the fact that changes in more complex objects are
increases, the resource must be divided among more elements, more subtle and, therefore, a more precise representation is needed
leaving each of them with a smaller share. Resource models have to detect them. However, Brady and Alvarez (2015) showed that
been very successful in quantitatively accounting for the effect of people can remember a greater number of simple objects than
set size on the precision of recall of visual features (Bays, 2014; complex objects even when the changes in the complex-object
Ma et al., 2014; van den Berg et al., 2014). trials are drastic, such as replacing a cube by a Chinese character.
 770 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

This result confirms that the complexity of visual objects affects are represented, that is, what proportion of their feature dimensions
WM performance. Other research operationalizing complexity as they have in common. The other is similarity, defined as the
the number of visual features to be remembered for each object proportion of features two items have in common within the same
found that memory declines with object complexity even when the feature space. For instance, when a memory set consists of a red
required precision is held constant (Cowan, Blume, & Saults, circle and a blue square, the two items vary on two dimensions,
2013; Hardman & Cowan, 2015; Oberauer & Eichenberger, 2013; both of which they share— both items have a color and a shape. At
see Figure 2A for representative results). the same time, the two items share none of their features. In this
At least for visual WM, Cowan and his colleagues have memory set, feature-space overlap is perfect, but similarity is zero.
proposed an explanation of the effects of both the number of In contrast, consider a memory set consisting of a red circle and a
chunks and the number of features per chunk (i.e., one aspect of spoken syllable. These two items share few, if any, feature dimen-
complexity) within a discrete-resource account (Cowan et al., sions, because—leaving aside the possibility of synaesthesia—
2013). Therefore, although the details of how different aspects spoken syllables have no color and no shape, and geometric figures
of complexity affect different aspects of performance are not have no phonological features.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

yet well understood, we argue that the resource account is able The two kinds of relations between representations—similarity
This document is copyrighted by the American Psychological Association or one of its allied publishers.

to offer a reasonable explanation for the finding that WM and feature-space overlap— have different consequences for the
performance depends both on the number and the complexity of three kinds of interference—interference by confusion, interfer-
elements in the memory set (A1). ence by superposition, and interference by feature overwriting.
The resource hypothesis correctly predicts that the set-size ef- The degree of interference by confusion decreases with decreasing
fect is observed even at a negligible RI (A2: Sewell et al., 2014; similarity and with decreasing feature-space overlap, because
Tsubomi et al., 2013): The competition for resources takes place as items are less likely to be confused the fewer features, and the
soon as a memory set is encoded and does not change while that fewer feature dimensions, they have in common. Likewise, inter-
set needs to be maintained. ference by feature overwriting decreases with decreasing similarity
Effects of domain and of set heterogeneity (A3–A5). As and decreasing feature-space overlap because of the decreasing
long as all elements in a memory set compete for the same proportion of shared features between representations. Interference
resource, the set-size effect should be the same for sets of homog- by superposition, in contrast, increases with decreasing similarity,
enous and for sets of heterogeneous elements. The assumption of because two representations in the same feature space distort each
a general resource predicts the cross-domain set-size effect (A4), other more severely the more their values on each feature dimen-
but it is insufficient to explain domain-specificity of set-size ef- sion differ from each other (see Figure 1). For instance, superim-
fects (A3): The dual-set studies reviewed above show that memory posing a red circle with a red square leads to mutual distortion of
is better for mixed sets of items from different domains than for the shape but not the color of each item, whereas superimposing a
pure sets of equal size (Cocchini et al., 2002; Oberauer & Kliegl, red circle with a blue square leads to distortion on both feature
2006). Resource theories have accounted for this fact by assuming dimensions. Interference by superposition decreases as the degree
separate resources for verbal and for visuospatial materials (Bad- of feature-space overlap decreases, because the distortion caused
deley, 1986; Logie, 1995). by superposition arises from summing (or averaging) feature val-
More problematic for the resource hypothesis is the heteroge- ues within a shared feature dimension. In a neural-network model
neity benefit (A5). To account for better memory for mixed than such as SOB-CS, different feature spaces are implemented as
for pure sets within the verbal or the visual domain (e.g., mixed different sets of units over which representations are distributed
lists of digits and letters, or of colors and orientations), resource (Oberauer, Lewandowsky, et al., 2012). A purely visual represen-
theories would have to assume separate resources for digits and tation of a red circle and a purely phonological representation of a
letters, or for different kinds of visual features. This is logically word do not interfere with each other because their distributed
possible but questions the elegance and parsimony of any resource representations are distributed over different, nonoverlapping sets
theory of WM. of units in a neural network.
Conclusion. The resource hypothesis offers a viable explana- The units of measurement of the capacity limit (A1, A2).
tion for the set-size effect. The only challenge for the resource From the interference perspective any attempt to measure WM
hypothesis arises from the heterogeneity benefit. load by counting or adding up some quantity characterizing indi-
vidual memoranda—such as their number, their complexity, or
their duration of restoration—is futile, because the capacity limit
Interference
arises from the interaction between representations in WM. One
Interference depends on the relations between representations in and the same representation can generate much interference in the
WM. A representation is conceptualized as a set of features, which context of one memory set (e.g., a noun among other nouns), and
can be described as a vector of feature values across several feature very little interference in the context of another (e.g., the same
dimensions (Nairne, 1990), as a point in a feature space defined by noun among a set of colors). That said, everything else being
these feature dimensions (G. D. A. Brown, Neath, & Chater, equal, the interference hypothesis predicts that memory declines as
2007), or as a pattern of activation across a set of units in a neural the number of elements in the set increases, because more repre-
network (Farrell & Lewandowsky, 2002; Lewandowsky & Farrell, sentations in WM imply more mutual interference between them.
2008b). To characterize the relation between two (or more) rep- The interference hypothesis makes no general prediction about the
resentations in WM we need to distinguish two aspects (for a more effect of complexity. Some instances of a complexity effect can
detailed treatment see Oberauer, Lewandowsky, et al., 2012). One easily be accommodated by the interference hypothesis. For in-
is the degree of overlap of the feature spaces in which two items stance, the fact that more complex words— consisting of more
 WM CAPACITY 771

syllables or more phonemes—are harder to remember (Service, evidence that several apparently different dimensions of quantity,
1998) arises naturally in an interference model because more such as space, time, and numerical quantity, share a common
complex words introduce more information into the same (phono- representational medium. Therefore, spoken digits could activate
logical) feature space, thereby increasing interference. Other in- numerical quantities that overlap with the spatial arrangement of
stances of the complexity effect are more challenging for the colors in an array, leading to some interference between digits and
interference hypothesis. For instance, WM for visual objects de- color arrays. The studies reviewed by Walsh (2003) thus provide
clines as more features on different feature dimensions need to be an existence proof for cross-domain representational overlap that
remembered for each object (Hardman & Cowan, 2015; Oberauer is not apparent from a surface analysis of the stimuli. However, for
& Eichenberger, 2013), although there is no reason why adding this kind of explanation of the cross-domain set-size effect to be
information on one feature dimension (e.g., shape or size) should satisfying, independent evidence must be provided for the overlap
interfere with information on another feature dimension (e.g., of representations of the specific stimuli used in each particular
color). On balance, the interference hypothesis is consistent with experiment.
finding A1, but it does not predict or explain it. Conclusion. The interference hypothesis provides a viable
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Interference is instantaneous—as soon as two or more represen- explanation of the set-size effect. It is consistent with the effects of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tations enter WM, they interfere with each other. Interference number and complexity of memoranda, and it predicts the capacity
limits the information that can be held in WM simultaneously, not limit in the absence of a retention interval. The interference hy-
its retention over time. Therefore, the interference hypothesis pothesis predicts that interference is smaller between than within
correctly predicts that a set-size effect is observed even at a domains, and smaller for heterogeneous than homogeneous sets
negligible retention interval (A2). within a domain. At the same time, an interference account has yet
Effects of domain and of set heterogeneity (A3–A5). An to offer a convincing explanation for why even very different
interference account of WM capacity necessarily predicts that memoranda, with apparently minimal feature-space overlap, inter-
memory sets of items from different content domains are easier to fere with each other in WM. Therefore, we can at best say that the
remember than domain-pure sets (A3). Mixed sets of verbal and hypothesis is consistent with a cross-domain set-size effect.
nonverbal items are easier to remember than pure sets because
verbal and nonverbal representations have relatively little feature- Round A: Summary
space overlap, thereby reducing the chance for all three kinds of
Table 2 summarizes the score sheet of round A. The decay
interference. A specific prediction following from interference by
hypothesis was hit hardest by the data: It is challenged by the fact
confusion is that, compared with pure lists, mixed lists lead to
that the set-size effect is not an effect of time (A1, A2), and by the
fewer confusions between list items, in particular between items
heterogeneity benefit within domains (A5). The resource and the
from different categories. This has been found for mixed lists of
interference hypothesis remain stronger contenders, with slightly
verbal, visual, and spatial items (Farrell & Oberauer, 2014).
more points for the interference hypothesis, because it predicts
The interference hypothesis also predicts the heterogeneity ben-
three of the findings, whereas the resource hypothesis predicts only
efit within content domains (A5): Mixed sets of colors and orien-
two, and is challenged by one, the heterogeneity benefit.
tations (Olson & Jiang, 2002) or of shapes and textures (Delvenne
& Bruyer, 2004) are easier to recall than pure sets because of Round B: Retention Interval and
reduced feature-space overlap: These kinds of items are repre-
Distractor Processing
sented in very low-dimensional feature spaces that do not overlap.
Mixed sets within the verbal domain, such as combinations of When trying to temporarily remember new information, concur-
digits and letters (Sanders & Schroots, 1969), arguably do not rently engaging in an unrelated processing task impairs memory
benefit from reduced feature-space overlap, because all verbal performance. This phenomenon has been regarded as a manifes-
materials are encoded primarily through their phonological fea- tation of the WM capacity limit since the early days of WM
tures, so that they share the feature space of phonetic features research (Baddeley & Hitch, 1974; Case et al., 1982; Daneman &
(Baddeley, 1966; Conrad, 1964). However, heterogeneous verbal Carpenter, 1980).
lists benefit from reduced interference by confusion: A confusion The degree to which distractor processing impairs memory
of a digit with a letter is less likely than confusions within each depends on several characteristics of the distractor task. One
class of stimuli. well-replicated finding is that memory performance decreases as
Whereas interference theories provide an explanation for the the cognitive load imposed by a distractor task increases (B1;
effects of domain specificity and of set heterogeneity, they have no Figure 6), where cognitive load is defined as the proportion of the
ready explanation for the cross-domain set-size effect (A4): Add- available processing time during which central cognitive processes
ing items to a memory set decreases memory even when the items are actually engaged by the distractor task (Barrouillet, Bernardin,
have no apparent feature-space overlap with each other, such as & Camos, 2004; Barrouillet et al., 2007; Conrad & Hull, 1966). In
spoken digits and arrays of colors (Morey & Cowan, 2004; Saults practice, cognitive load is usually varied through the pace at which
& Cowan, 2007). Representations with no feature-space overlap a series of processing operations of roughly constant difficulty is
should not interfere with each other. A possible explanation for required. For instance, Conrad and Hull (1966) asked participants
mutual impairment of memory for such very different stimuli is to remember four consonants while reading aloud digits at a pace
that their representations share feature dimensions that are not of 0.4 or 0.8 s per digit. Memory was impaired more when the
apparent from a description of the nominal stimuli. The way same number of digits had to be read at a faster pace.
people represent a stimulus does not necessarily match the way the Independent of pace, the duration of distractor processing dur-
experimenter describes it. For instance, Walsh (2003) summarizes ing the RI has been found to affect memory in some studies
 772 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

Table 3
Summary of Informative Findings and Evaluations of Hypotheses in Round B: Findings on Retention-Interval and
Distractor-Processing Effects

Index Finding Decay Resource Interference

B1 The impairment of memory by processing distractors in the retention interval increases ⫹ ⫺ ⫹

with the cognitive load imposed by the processing task
B2 The duration of distractor processing in the retention interval affects memory if and ⫺ ⫺ ⫹⫹
only if distractors differ from each other
B3 The duration of an unfilled retention interval impairs visual and spatial WM in some ⫹ 0 ⫺
experiments
B4 Domain-specific effect of processing: Processing distractors from the same content ⫹ ⫹ ⫹⫹
domain as the memoranda leads to a larger impairment
B5 Cross-domain impairment of memory by processing: Memory is impaired by processing ⫹ ⫹⫹ 0
of distractors from another domain than the memoranda
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

B6 Heterogeneity benefit: Processing distractors from different classes as the memoranda ⫺ ⫺ ⫹⫹

This document is copyrighted by the American Psychological Association or one of its allied publishers.

(within the same domain) impairs memory less than processing of distractors from
the same class
Note. Table entries reflect our judgment of the logical relation between a finding and a hypothesis: The hypothesis predicts (⫹⫹) or can explain (⫹) the
finding, it is consistent with the finding (0) or it is challenged by the finding (⫺); see text for explanation. WM ⫽ working memory.

(Chechile, 1987; Conrad & Hull, 1966; Peterson & Peterson, whether or not memory declines over a filled RI depends on the
1959), but not in others (Barrouillet et al., 2004; Humphreys et al., variability of distractor processing (see Figure 7). At the same
2010). One relevant moderator is the degree of variability of the time, some studies using visual or spatial memoranda have found
distractor material processed (B2): If the material is highly repet- that extending the duration of the RI impairs memory even in the
itive—such as repeatedly speaking the same word—the duration of absence of a concurrent processing task (B3; Lilienthal, Hale, &
this activity has no effect on memory, at least for verbal memo- Myerson, 2014; Mercer & McKeown, 2014; Ricker & Cowan,
randa. In contrast, when the processed material is variable—such 2010).
as speaking different words—the detrimental effect on memory Like the set-size effect, the impairment of memory by con-
increases with the duration of processing (Lewandowsky, Geiger, current processing is in part domain-specific (B4): Having to
Morrell, & Oberauer, 2010; Lewandowsky et al., 2008; McFarlane process materials from the same domain as the memory content
& Humphreys, 2012). Hence, at least for verbal memoranda, is more detrimental than having to process materials from

Figure 6. Effect of cognitive load by a distractor task (finding B1). (A) Example trials of Experiment 3 of
Barrouillet et al. (2007). Participants remembered lists of letters, and in between made parity judgments or
location judgments on digits. The figure shows the sequence of events between encoding of two list items (red
letters) in a condition with low cognitive load (CL), in which participants have to make four judgments on digits,
and a condition of high CL, in which they have to make eight judgments in the same total time. (B) Memory
span as a function of cognitive load. Cognitive load was estimated as the summed response times of all
judgments in between two letters, divided by the total available time. Memory is impaired more by a distractor
task as the proportion of time spent on the distractor task increases. See the online article for the color version
of this figure.
 WM CAPACITY 773
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 7. The effect of the duration of the retention interval depends on the variability of distractors processed
in that interval (finding B2). (A) Example trials from Experiment 3 of Lewandowsky et al. (2010): Participants
remembered lists of letters, and in between read distractor words aloud. (B) Relative to a baseline without
reading of distractors, letter recall declined when one word was read after each letter. Accuracy did not decline
further when the retention interval was extended by having participants repeat the distractor word four times (no
distractor variability), but it did decline when they read three different words (distractor variability). See the
online article for the color version of this figure.

another domain (Chein, Moore, & Conway, 2011; Davis, Rane, other nonwords than by processing of words (Conlin & Gather-
& Hiscock, 2013; Hale, Myerson, Rhee, Weiss, & Abrams, cole, 2006). We now discuss how the three hypotheses fare in light
1996; Jarrold, Tam, Baddeley, & Harvey, 2010; Jarrold et al., of findings B1 to B6, summarized in Table 3.
2011). In addition to this domain-specific effect of processing
on memory, most studies have also found an—albeit smaller— Decay
impairment of memory by processing material in a different
domain (B5; Chein et al., 2011; Jarrold et al., 2011; Vergauwe The initial motivation for assuming rapid decay of traces in
et al., 2010). For instance, memory for spatial patterns (Darley short-term or WM came from the observation of rapid forget-
& Glass, 1975) and for color-shape conjunctions (Allen et al., ting over an RI filled with a distractor task (J. Brown, 1958).
2006) is impaired by orally counting backward. Memory for The decay hypothesis implies that memory performance de-
spatial locations (Klauer & Stegmaier, 1997) and for colors clines over an increasing RI if restoration processes such as
(Makovski, 2012) is impaired by binary decisions on verbal rehearsal are prevented during that interval by a distractor
stimuli, such as parity judgments on digits. Conversely, mem- activity. If restoration can be accomplished by both articulatory
ory for verbal lists is impaired by nonverbal decisions (Jarrold rehearsal and refreshing (Camos et al., 2009), then distractor
et al., 2011; Vergauwe, Dewaele, Langerock, & Barrouillet, processing preventing articulatory rehearsal (such as articula-
2012). Both the domain-specific and the domain-general effect tory suppression)4 as well as distractor tasks engaging central
of distractor processing have been replicated numerous times. attention (such as tasks requiring response selection or retrieval
Figure 8 (top panel) illustrates these effects. from long-term memory) are predicted to impair memory for
Finally, a heterogeneity benefit for distractor processing has verbal materials.
been observed in all but one of the studies investigating it (B6; Cognitive load (B1). The decay hypothesis, together with
bottom panel of Figure 8): If memoranda and distractors come the assumption of attention-based refreshing, can explain why
from the same domain, distractor processing damages memory less memory declines with increasing cognitive load imposed by a
when memoranda and distractors are drawn from different cate- concurrent distractor task that demands central attention (B1 in
gories than when they are drawn from the same category. For Table 3; Barrouillet et al., 2007). Cognitive load is defined as
instance, memory for lists of digits is impaired more by concurrent the proportion of time of the RI during which central attention
processing of numbers than of words, whereas memory for lists of is engaged by the distractor task. Refreshing is assumed to
words is disrupted more by concurrent processing of words than of compete with the distractor task for the central attentional
numbers (Conlin, Gathercole, & Adams, 2005; Li, 1999; Turner & bottleneck. Therefore, higher cognitive load implies a larger
Engle, 1989; for a partial exception to this pattern see Macken &
Jones, 1995). Similarly, recall of lists of words is disrupted more 4
Articulatory suppression refers to asking participants to continuously
by processing of other words than by processing of nonwords, say aloud a simple series of syllables, such as “ba, ba, ba . . .,” with the
whereas recall of lists of nonwords is disrupted more by processing purpose of preventing articulatory rehearsal.
 774 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

performance can be explained (Oberauer & Lewandowsky,

2011).
Retention interval and the amount of distractor processing
(B2, B3). If decay is to play any role in explaining the capacity
limit of WM, it must lead to measurable forgetting when restora-
tion is prevented. As decay theories assume different restoration
processes for verbal and nonverbal memoranda, and the relevant
evidence differs substantially between these domains, we discuss
them separately.
Verbal memoranda. Two kinds of restoration processes have
been assumed for verbal memoranda, articulatory rehearsal, and
refreshing. Camos et al. (2009) asserted that the protective ef-
fects of rehearsal and of refreshing are additive. If the beneficial
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

effects of rehearsal and refreshing are assumed to be additive,

This document is copyrighted by the American Psychological Association or one of its allied publishers.

this implies that the effects of preventing each of these restoration

processes also must be additive. Therefore, preventing either ar-
ticulatory rehearsal or refreshing should lead to forgetting over
time, and preventing both should lead to more rapid forgetting over
time. Subvocal articulatory rehearsal can be prevented through
articulatory suppression—asking participants to repeat a simple
utterance continuously. Experiments varying the duration of a
retention interval during which participants engaged in articulatory
suppression found no decline of memory with longer retention
intervals (B2; Humphreys et al., 2010; Lewandowsky et al., 2004;
Longoni, Richardson, & Aiello, 1993; Phaf & Wolters, 1993;
Vallar & Baddeley, 1982). Refreshing can be prevented by simple
binary decision tasks that engage central attention (Barrouillet et
al., 2007). Variations of retention intervals filled with binary
decision tasks have not revealed any decline of memory over time
(Barrouillet, Portrat, Vergauwe, Diependaele, & Camos, 2011;
Oberauer & Lewandowsky, 2014). Relaxing the assumption of
additive benefits from rehearsal and refreshing would not help the
decay hypothesis: Even when both forms of restoration are pre-
vented by asking participants to engage in an attentionally de-
manding task and articulatory suppression at the same time, mem-
ory for lists of letters still does not decline over time (Oberauer &
Lewandowsky, 2008, 2013). These findings are incompatible with
a central prediction from the decay hypothesis.
Figure 8. Top: Domain-specific effect (finding B4) and domain-general It has been argued that the decay assumption can be reconciled
effect (finding B5) of distractor processing in the complex-span experiment with the finding that memory does not decline during a RI filled
of Chein et al. (2011): Participants remembered lists of letters (verbal) or with restoration-preventing distractor activity. The argument is
of dot locations in a grid (spatial), combined with lexical decision (verbal) that memory performance depends on the cognitive load imposed
or symmetry judgments (spatial) as processing demand. Relative to a
by distractor processing, and because cognitive load is a proportion
simple-span task with no processing assignment, memory was impaired by
a processing demand in the other domain (domain-general effect), but was
of two time intervals, it can be held constant as the RI is increased.
more impaired by processing in the same domain (domain-specific effect). Therefore, memory performance is predicted to stay constant over
Bottom: Heterogeneity benefit (finding B6) in the complex-span study of variations of the RI as long as cognitive load is held constant
Turner and Engle (1989): Memory for digits was more impaired by (Barrouillet et al., 2011). This argument is, however, not logically
concurrent processing of digits (a condition with homogeneous materials sound (Oberauer & Lewandowsky, 2014). From the observation
used for memory and distractor task) than of words (heterogeneous con- that cognitive load has an effect on memory performance it does
dition, using different materials for the memory and the distractor task). not follow that memory depends only on cognitive load. In fact, the
Conversely, memory for words was more impaired by processing of words decay assumption implies that memory depends on RI in addition
(homogeneous) than of digits (heterogeneous). to cognitive load. Specifically, decay implies that memory must
decline with increasing RI for any constant level of nonminimal
cognitive load, for the following reason: Consider the fate of a
proportion of time during which refreshing is prevented, leav- WM representation during any arbitrary, reasonably short interval
ing memory traces to decay, and a lower proportion of time in the RI. There are two logically possible scenarios of what
during which decay can be counteracted by refreshing. Com- happens to that representation. One possibility is that cognitive
putational modeling has shown that with these assumptions the load is low enough so that articulatory rehearsal and/or refreshing
approximately linear effect of cognitive load on serial recall can fully compensate the adverse effect of decay, so that no net
 WM CAPACITY 775

control for temporal distinctiveness of successive trials. Temporal

distinctiveness refers to the discriminability of memories on the
psychological time dimension. Distinctiveness models of memory,
such as SIMPLE (G. D. A. Brown et al., 2007), assume that
temporal distinctiveness of two events—such as the current trial
and the preceding trial in an experiment— depends on the ratio of
the time intervals that have passed since the two events. If the RI
of a WM task is increased while the intertrial interval is held
constant, the temporal distinctiveness of the current trial relative to
the preceding trial is reduced, leading to more confusion between
trials—that is, more proactive interference.
Effects of temporal distinctiveness can be separated from decay
effects by varying both the RI and the intertrial interval (ITI). By
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

choosing an appropriate ITI for each level of RI, temporal distinc-

This document is copyrighted by the American Psychological Association or one of its allied publishers.

tiveness can be held constant across variations of RI (see Figure

10A). With this design, distinctiveness theories predict no effect of
Figure 9. Schematic time line of memory strength of an item undergoing
decay and refreshing. Strength is increased through refreshing (arrows RI whereas decay theories predict that memory declines with
marked with R) but declines through decay when central attention is increasing RI. Two recent studies testing WM for colors using this
engaged by another process (e.g., encoding or refreshing another item, or design have shown that memory performance varies with temporal
processing a distractor). (A) Low cognitive load, such that refreshing fully distinctiveness, and the effect of RI disappears when distinctive-
compensates decay. Memory strength does not decline over an increasing ness is held constant (Shipstead & Engle, 2013; Souza & Ober-
retention interval. (B) Higher cognitive load, such that refreshing only auer, 2015). An effect of temporal distinctiveness was also ob-
partly compensates decay. Memory strength declines over an increasing served for short-term recognition of complex visual stimuli
retention interval. (Mercer, 2014). A further study (Ricker, Spiegel, & Cowan, 2014)
using arrays of unfamiliar characters or letters (with articulatory
loss of memory strength occurs (Scenario A in Figure 9). Alter- suppression) as memory materials obtained mixed evidence, with
natively, cognitive load is high enough to prevent full compensa- a strong effect of RI and an effect of ITI that was smaller, and
tory restoration, implying a net loss of memory strength (Scenario nonsignificant in two out of four experiments, suggesting that there
B in Figure 9). Under the first scenario, memory accuracy can be is an effect of decay in addition to an effect of proactive interfer-
maintained at a constant level regardless of the duration of the RI, ence. Taken together, the evidence for decay of visual stimuli in
whereas under the second scenario, the net loss of memory WM is mixed, and whether or not decay plays a role might depend
strength during any interval of the RI accumulates as the RI is on the stimuli, the experimental parameters, and the procedure of
increased, implying more forgetting over a longer RI. Now con- testing, in as yet unknown ways (see Figure 10B).
sider two levels of cognitive load, such that memory is worse at the A number of studies have observed that WM for spatial loca-
higher level. The only way this effect of cognitive load can be tions declines over unfilled RIs, but the decline is in most cases
explained within a decay theory is to assume that (at least) at the very shallow, amounting to negligible forgetting after 10 s or more
higher level of cognitive load, decay cannot be fully compensated (B3; Hole, 1996; Jones, Farrand, Stuart, & Morris, 1995; Parmen-
by restoration (as in Scenario B). If both levels of cognitive load tier & Jones, 2000; Phillips & Christie, 1977; Ploner, Gaymard,
allowed full compensation of decay, memory would not differ Rivaud, Agid, & Pierrot-Deseilligny, 1998). Again, these studies
between them. It follows that (at least) at the higher level of have not controlled temporal distinctiveness, so the small effect of
cognitive load, there must be a net loss of memory strength over RI could reflect distinctiveness rather than decay. One recent study
any time interval in the RI. Therefore, at that level of cognitive has demonstrated substantial forgetting of spatial information over
load, a decay theory must predict that memory declines with longer time while holding temporal distinctiveness constant, but only
RIs. The opposite has been observed, disconfirming a prediction when the screen went blank during the RI, thereby removing
from the decay hypothesis (Oberauer & Lewandowsky, 2014). environmental support for a hypothetical visual rehearsal process
Visual and spatial memoranda. Whereas in the verbal do- (Lilienthal et al., 2014). The substantial forgetting during RIs of 1
main the evidence against a role for decay in WM is strong, the versus 4 s in Lilienthal et al. (2014) is difficult to reconcile with the
picture is more ambiguous in the visual-spatial domain (B3). negligible forgetting observed over even longer blank-screen RIs
Turning first to visual information, several experiments on WM for in other studies (e.g., Jones et al., 1995). The available experiments
visual features such as colors, orientations, or shapes have shown differ in many regards that could explain the highly variable
a decline of accuracy over unfilled RIs (Gold, Murray, Sekuler, effects of unfilled RI. One potentially relevant variable is the time
Bennett, & Sekuler, 2005; Mercer & Duffy, 2015; Morey & Bieler, available for consolidation of information in WM (Jolicœur &
2013; Pertzov, Bays, Joseph, & Husain, 2013; Ricker & Cowan, Dell’Acqua, 1998; Nieuwenstein & Wyble, 2014). For instance,
2010; Sakai & Inui, 2002; Zhang & Luck, 2009), whereas others Jones et al. presented a set of dot locations sequentially for 2 s per
have not (Gorgoraptis, Catalao, Bays, & Husain, 2011; Kahana & dot, whereas Lilienthal et al. presented each dot for just 1 s. A
Sekuler, 2002; Magnussen & Greenlee, 1999; Vogel, Woodman, study by Ricker and Cowan (2014) showed that the rate of forget-
& Luck, 2001). ting over unfilled RIs was substantially diminished when more
Interpretation of these findings is further complicated by the fact time was allowed for consolidation of information in WM, for
that the experiments cited in the preceding paragraph did not instance by presenting items sequentially, or allowing more time
 776 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 10. Effects of unfilled retention intervals (RI) and intertrial intervals (ITI) on working memory (WM)
for visual materials (finding B3). (A) Design of the change-detection experiments of Shipstead and Engle (2013),
varying the ITI (from response in the preceding trial to encoding of a new memory array in the current trial) and
the RI (from encoding to onset of the test display in the current trial). Each row shows the time line of one
condition; the conditions in rows 1 and 4 have equal temporal crowdedness (i.e., lack of distinctiveness), defined
as RI/(RI ⫹ ITI). (B) Memory performance, measured as Cowan’s K, in four representative experiments varying
RI and ITI, displayed as a function of RI (left) and of temporal crowdedness (right). Black and grey: Experiments
1 and 3 of Ricker et al. (2014), respectively; red: Experiment 4 of Shipstead and Engle (2013); white: Souza and
Oberauer (2015). Within each study, circles reflect the short, and squares the long ITI condition. In all four
experiments memory declined with longer RIs. In the experiments of Shipstead and Engle (2013), and of Souza
and Oberauer (2015), but not those of Ricker and colleagues (2014), memory was better with longer ISIs, so that
performance depended on temporal distinctiveness: Two conditions with different RIs but equated for temporal
distinctiveness resulted in equal performance (see the two intermediate red and white data points in Panel B).
See the online article for the color version of this figure.

for encoding a simultaneous array. This finding converges with the become immune to decay, and different materials might differ in
observation of Sakai and Inui (2002) that the rate of forgetting of the time it takes to consolidate them.
visual features became more shallow as the presentation duration Few studies have investigated forgetting of visual or spatial
was increased from 120 to 1,200 ms. One interpretation of this memoranda as a function of the duration of an RI filled with
result is that representations in WM need to be consolidated to distracting activity (B2). On the decay hypothesis, filling the RI
 WM CAPACITY 777

with distractor processing should impair restoration and, therefore, arising from resource competition, because processing and retrieval
accelerate the decline of memory over time. Ricker and Cowan never compete for resources.
(2010) found change-detection accuracy to decline over an RI A resource account of how processing during the RI impairs
filled with mental arithmetic. The processing task impaired mem- maintenance must make an additional assumption: Once the re-
ory compared with a condition with unfilled RI, but did not lead to source share of a representation in WM falls below a threshold,
faster forgetting over time. Christie and Phillips (1979) asked that representation is irreversibly forgotten, so that even when part
participants to reproduce patterns of randomly filled grids after of the resource is freed later, it cannot be reallocated to that
variable RIs during which they counted backward in steps of three. representation. This is the assumption underlying the 3CAPS
Distractor processing reduced memory compared to a condition model (Just & Carpenter, 1992). In what follows we will discuss
with unfilled RI, but the duration of the RI had no effect. This the resource hypothesis augmented with the irreversibility assump-
result mirrors the findings of Oberauer and Lewandowsky (2013, tion above. Departing from the order of findings in Table 3, we
2014) using verbal memoranda combined with nonverbal distrac- postpone discussion of the effect of cognitive load (B1) because it
tor, which also found that memory was unaffected by the duration is understood better in the context of more general considerations
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

for which the distractor task had to be carried out. Neither of these about the role of the intensity and duration of processing during a
This document is copyrighted by the American Psychological Association or one of its allied publishers.

findings matches the prediction from the decay hypothesis. retention interval.
Cross-domain and domain-specific effects of distractor pro- Retention interval and the amount of distractor processing:
cessing (B4 –B6). The decay hypothesis has no problem explain- Intensity and duration (B2, B3). A processing task that de-
ing the finding that memory is impaired by processing of distrac- mands more of the shared resource should impair memory to a
tors in a different domain (B5). Cross-domain dual-task costs are larger degree. In this context it is important to consider two
to be expected if a domain-general attentional mechanism contrib- dimensions of the resource demand of a processing task, its inten-
utes to memory restoration (Vergauwe et al., 2010). The decay sity and its duration. According to the irreversibility assumption
hypothesis can also explain domain-specific effects of processing introduced above, the amount of forgetting caused by a concurrent
on memory by assuming domain-specific rehearsal processes, such processing task should depend on the intensity of that task’s
as articulatory rehearsal for verbal memoranda, and spatial shifts resource demand, not on its duration: A processing task that
of attention for spatial memoranda (B4). The heterogeneity bene- demands more of the resource share at any point in time leads to
fit, in contrast, is challenging for the decay hypothesis (B6): There more serious resource cuts for the memory items, putting them at
is no reason why, for instance, memory for words should be higher risk of being irreversibly forgotten at that moment (compare
impaired more by processing of words than of digits, whereas Scenarios A and C in Figure 11). As long as a processing task
memory for digits is impaired more by processing of digits than of demands a constant share of the resource, its duration should not
words: Processing of words and of digits should equally disrupt matter: Cutting the resource share of memory representations
articulatory rehearsal. either pushes it below the retrieval threshold right away, leading to
Conclusion. Whereas the decay assumption offers viable expla- instant forgetting, or does not push it below the threshold, allowing
nations for some findings—in particular the effect of cognitive load, indefinite maintenance (compare Scenarios A and B in Figure 11).
and the observation of both domain-general and domain-specific At first glance the prediction that processing duration does not
effects of distractor processing—it is challenged by others. The most matter appears attractive because it matches a large set of findings
problematic result is the lack of forgetting—at least of verbal infor- showing that, when the intensity of a concurrent processing de-
mation— over time, even when articulatory rehearsal, attention-based mand is held constant, its duration has no impact on memory (B2).
refreshing, or both are prevented by a concurrent processing task. Speaking an irrelevant word or syllable aloud impairs memory for
verbal lists, but it does not matter for how long the same utterance
is repeated (Humphreys et al., 2010; Lewandowsky et al., 2010;
Resources
Lewandowsky et al., 2008; Longoni et al., 1993; Oberauer &
The resource assumption has often been invoked to explain why Lewandowsky, 2008; Phaf & Wolters, 1993; Vallar & Badde-
WM maintenance suffers from a processing task carried out during ley, 1982). Likewise, making simple binary decisions impairs
the RI: The processing task is assumed to take away part of the serial recall of verbal lists, but the number of such decisions to
resource needed for maintenance, leaving less to be distributed among be carried out at a constant rate has little impact on memory
the memory items. This explanation, though intuitively appealing, is (Oberauer & Lewandowsky, 2008, 2014). Memory for spatial
less straightforward than it appears. Assume that a memory set is patterns is impaired by concurrent backward counting, but the
encoded by dividing 100% of a resource among its items. In the duration of the backward counting has no effect (Christie &
subsequent RI 50% of the resource is demanded by a processing task. Phillips, 1979). As we noted above, these effects are problem-
As a consequence, the resource share assigned to each item needs to atic for the decay assumption, but they can be accommodated
be cut in half. Once the processing task is finished, it no longer by the resource hypothesis.
requires any part of the resource, so the resource can be given back to There is, however, an equally solid body of evidence showing
the memory items. When memory is tested after the processing task that under certain conditions the duration of concurrent processing
is completed—as is usually the case in dual-task paradigms of WM— has a substantial effect on memory. This is the case whenever the
then the resource share allocated to each representation in WM at the material processed varies over time (B2; see Figure 7): When
time of test is not diminished by the fact that a processing task had to people have to repeat the same distractor word several times, their
be completed in the RI. Therefore, models in which performance memory performance is indistinguishable from that when required
depends on the resource allocation at retrieval (e.g., Lovett, Reder, & to say the word only once, but when people have to say several
Lebiere, 1999) do not explain the effect of concurrent processing as different words in between presentation of each memory item, list
 778 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

This version of the resource model comes down to a resource-

modulated decay model: The chance of irreversibly forgetting any
memory item increases over time, and the rate of forgetting de-
pends on the mean resource share of that item during the time
interval in question. This version of resource theory runs into the
same difficulties as the decay hypothesis: It cannot explain why
processing duration does not matter when the material processed
has low variability (e.g., a series of binary decisions on highly
similar stimuli).
Cognitive load (B1). The resource hypothesis also offers no
obvious way to explain the cognitive-load effect (B1). When
cognitive load is maximal, the entire available time for a process-
ing task is required for processing, implying that any resource
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

amount needed for the processing task is continuously engaged

This document is copyrighted by the American Psychological Association or one of its allied publishers.

by it. Cognitive load can be reduced by reducing the pace of

processing, thereby adding free time in between individual
processing steps— during these intervals the resource is pre-
sumably not needed for processing. Yet, there is no way in
which this intermittently free resource could benefit memory:
The free resource could be allocated to memory representations
Figure 11. Schematic timeline of the resource share of a memory item for a short time but will soon be claimed back by concurrent
when a distractor task temporarily draws away part of the resource from it. processing demands, leaving the memory representation as
The continuous line shows the resource share of the memory representation resource-depleted as before.
over time (from left to right); the thick broken arrow covers the duration of
Perhaps the cognitive-load effect arises from a resource limit
the processing task, and the dotted line is the retrieval threshold, such that
any representation falling below the threshold is irrevocably forgotten
because higher cognitive load increases the intensity of pro-
(symbolized by the evaporation cloud). (A) A short period of concurrent cessing, such that the processing task recruits a larger propor-
processing demand of low intensity in a model without random fluctuations tion of the resource. This is conceivable because cognitive load
of resource assignment. The memory item’s resource share remains above has often been manipulated through a variation of time pressure
threshold, and after the distractor task is completed, the full resource (e.g., Barrouillet et al., 2007), and a resource-based system
amount can be restored to the item. (B) Like A but with a longer period of should respond to time pressure by speeding up processing
distractor processing. The distractor task does no more harm to memory steps through allocating a larger resource share to them (Tombu
than in A. (C) A short period of high-intensity processing demand: The
& Jolicoeur, 2003). Two findings speak against that possibility,
memory representation is instantly forgotten as the processing demand
pushes its resource share below threshold. (D) An extended period of though. First, when time pressure for distractor processing in a
processing with low average resource demand but random fluctuation of complex-span task is increased, people do not increase the
resource assignment (a sample of two time courses is shown as the efficiency of distractor processing (i.e., produce equally accu-
continuous and the broken line). As the processing period is extended, rate responses at higher speed) but rather trade accuracy for
there are more chances of the item’s resource share to fall below threshold, speed (Oberauer & Lewandowsky, 2013). Other studies on time
thereby being irrevocably forgotten. pressure found that time pressure even decreased processing
efficiency (Dambacher & Hübner, 2015). Second, the effect of
recall is worse than when they have to say only a single word cognitive load is also observed in the absence of time pressure:
(Lewandowsky et al., 2010; Lewandowsky et al., 2008; McFarlane
In the experiments of Oberauer, Lewandowsky et al. (2012)
& Humphreys, 2012). Similarly, carrying out four arithmetic op-
participants were free to complete each processing step when
erations impairs memory more than two operations at the same rate
they were ready, and cognitive load was manipulated by vary-
(Gavens & Barrouillet, 2004).
ing the free time between a response and the next stimulus.
An effect of the duration of processing could be accommodated
Memory was again better at lower cognitive load. The
in a resource model by assuming that the allocation of resource
quantities, or the threshold, fluctuates randomly over time (Sce- cognitive-load effect is to a large part a beneficial effect of free
nario D in Figure 11). Assume that an item in WM has its resource time in between distractor processing, and a resource account
share curtailed by a concurrent processing task, but its mean has no way to explain that effect.
resource share is still slightly above threshold. This item could It might be tempting to explain the cognitive-load effect by
survive in WM indefinitely if its resource share remained constant. assuming that the resource is needed for refreshing items, or to
This is not the case, however, if the resource share fluctuates over otherwise protect them from decay. The longer a distractor task
time, and the item is irrevocably forgotten if its resource share at captures some of the resource, the longer memory representations
any point in time falls below a threshold. The chance that the are left to decay. This is essentially the explanation of the
resource share falls below the threshold at least once during a time cognitive-load effect given in the time-based resource-sharing
interval increases with the duration of the interval. Therefore, a (TBRS) model (Barrouillet et al., 2004). This approach implies
longer duration of a resource-demanding processing task should that decay, not a resource limit, is the primary cause of the capacity
lead to more forgetting. limit. The resource limit comes into play only as limiting the
 WM CAPACITY 779

restoration process that mitigates decay. Therefore, we discussed serial recall attribute this recency effect at least in part to response
this account above in the Decay section of Round B. suppression: Once a list item is recalled, it is removed from WM
Cross-domain and domain-specific effects of distractor pro- so that it does not interfere with recall of subsequent items. As
cessing (B4 –B6). The resource-based explanation of dual-task recall nears the end of the list, there are only few items left in WM,
costs in WM implies that a concurrent processing task should reducing interference for the last list items. In line with this
impair memory if processing and maintenance compete for the explanation, the recency effect is larger if all list items up to the
same resource. If a domain-general resource is assumed, then last have been recalled— even though in the wrong order— com-
processing requirements with very little similarity or overlap pared with trials on which people failed to recall some prerecency
with the memory contents should disrupt maintenance. There is items (Farrell & Lewandowsky, 2012). A second line of evidence
substantial evidence supporting this prediction (B5). At the comes from research on WM updating: When a precue indicates a
same time, processing tasks using material from the same broad specific list item as the one to be replaced on the next updating
content domain (verbal vs. visual or spatial) have often been step, people can remove that item from WM before seeing the
found to impair memory more than processing tasks from a replacement stimulus (Ecker, Oberauer, & Lewandowsky, 2014;
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

different domain (B4). These findings can be jointly explained Ecker, Lewandowsky, & Oberauer, 2014). Finally, research from
This document is copyrighted by the American Psychological Association or one of its allied publishers.

by assuming that WM draws on a general resource together with visual WM suggests that when one item, or a subset of items, is
domain-specific resources (Baddeley, 1986; Logie, 2011). cued during the RI as relevant, other items can be removed from
More problematic for the resource hypothesis are findings WM (Souza, Rerko, & Oberauer, 2014; Williams, Hong, Kang,
showing smaller dual-task costs when memoranda and processing Carlisle, & Woodman, 2013).
materials come from different categories within the same content The strong theoretical reasons for assuming that outdated represen-
domain (B6). Explaining this heterogeneity benefit by assuming tations can be selectively removed from WM, together with the
separate resources for different categories within a domain, such as empirical evidence supporting this assumption, suggest an explana-
digits, words, and nonwords, would open the door to a boundless tion for the cognitive-load effect: Lower cognitive load implies more
inflation of resources, rendering the resource theory untestable. free time in between processing of distractors, and that time can be
Conclusion. The hypothesis of a domain-general resource, used to remove distractors, thereby reducing interference. Oberauer,
embellished with appropriate assumptions, provides an attractive Lewandowsky et al. (2012) have implemented this idea in one inter-
explanation for why WM maintenance is often found to be im- ference model of WM, SOB-CS. In SOB-CS, every distractor is
paired by an unrelated processing task even when it has no obvious bound to the context in which it is encountered, and when processing
overlap with the memory contents. The resource hypothesis strug- is complete that distractor is unbound from its context. The mecha-
gles, however, with explaining why the effect of processing on nism by which unbinding takes place is identical to that which
maintenance depends on whether memoranda and processed ma- accomplishes response suppression during recall. Oberauer, Le-
terial come from the same class of stimuli within a domain; why wandowsky et al. (2012) showed that with the inclusion of this
the duration of processing has an impact on memory if and only if unbinding process, SOB-CS produces the linear effect of cognitive
the processed material varies over time; and why it is beneficial for load on memory performance.
memory if a concurrent processing episode is interspersed with Retention interval and the amount of distractor processing
longer intervals of free time. (B2, B3). The interference hypothesis makes a specific predic-
tion for the effect of distractor processing in the RI: The degree to
which memory is impaired should not depend on the duration of a
Interference
distractor-filled RI, but on the number of different representations
Interference theories can account for the adverse effect of dis- engaged during distractor processing: With every new distractor, a
tractor processing on memory by assuming that the representations new representation enters WM and adds to the interference suf-
engaged in processing enter WM and, therefore, interfere with fered by the memoranda. For instance, if the distractor task con-
representations of memory items (Saito & Miyake, 2004). sists of reading aloud words, the amount of interference should
Cognitive load (B1). The effect of cognitive load (B1) poses depend on the number of different words read. This prediction has
a problem for interference theories, because it is not immediately been confirmed (B2): When participants have to speak the same
obvious how low cognitive load—that is, more free time in be- word or syllable repeatedly during maintenance of a verbal list,
tween individual operations on a distractor task—should be ben- forgetting does not depend on how often they repeat the utterance.
eficial for memory. One suggested solution is that the free time is In contrast, if they have to say aloud a series of different words,
used to “remove” representations of previously processed distrac- memory is impaired more the more words need to be spoken
tors from WM, by unbinding them from their encoding context, (Lewandowsky et al., 2010; Lewandowsky et al., 2008; McFarlane
thereby reducing interference with the memoranda (Oberauer, & Humphreys, 2012).
Lewandowsky, et al., 2012). Every theory of WM must assume As already noted, interference is instantaneous, and therefore,
some process by which WM is cleared of no-longer relevant the interference hypothesis does not predict forgetting over an
representations. If this does not happen on its own through decay, unfilled RI. A decline of memory with increasing unfilled delays
it has to be accomplished by some other process, such as unbinding could be explained only through temporal distinctiveness: If the RI
or removal. is increased while the ITI is held constant, temporal distinctiveness
Independent evidence for the selective removal of no-longer of the trials is decreased, making it harder to distinguish the current
relevant information from WM comes from three sources. One is memory set from that of preceding trials, thereby increasing the
the recency effect in immediate serial recall: The last few list items risk of proactive interference (Shipstead & Engle, 2013). It follows
are usually recalled better than the preceding ones. Most models of that the interference hypothesis is challenged by findings of grad-
 780 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

ual memory loss over an unfilled RI when temporal distinctiveness task (Oberauer, Lewandowsky, et al., 2012). Such an explanation
is controlled through a concomitant variation of the intertrial remains preliminary until the representations actually involved in
interval (B3; Lilienthal et al., 2014; Ricker et al., 2014). One a given distractor processing task are determined independently of
potential explanation for these findings within an interference their effect on memory. Therefore, we argue that the interference
framework is that participants generated representations during the hypothesis is consistent with finding B5, but does not yet offer a
RI spontaneously through mind wandering, which often involves satisfactory explanation of it.
visual images (Teasdale, Proctor, Lloyd, & Baddeley, 1993) that Conclusion. To summarize, two mechanisms of interfer-
could interfere with visual memoranda, or through erratic eye ence—interference by confusion and by superposition—jointly
movements that are known to interfere with spatial WM (Pearson provide an accurate account of the detailed pattern of dual-task
& Sahraie, 2003). In the absence of independent evidence of such costs between maintenance and concurrent processing. Yet, for a
self-generated representations, however, such an explanation is complete account of effects of unfilled retention intervals and of
post hoc and therefore unsatisfying. dual-task costs across different domains, an interference model has
Cross-domain and domain-specific effects of distractor pro- to make as yet untested assumptions about the recruitment of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

cessing (B4 –B6). Interference from distractor processing should representations that do not correspond directly to information
This document is copyrighted by the American Psychological Association or one of its allied publishers.

depend on the similarity and the feature-space overlap between given in the environment.
memoranda and distractors. The predicted pattern of these effects
has been explored through simulations with SOB-CS (Oberauer,
Round B: Summary
Lewandowsky, et al., 2012), and can be summarized as follows.
First, if the distractors come from the same category as the Round B favored the interference hypothesis, which correctly pre-
memory items (e.g., both are sets of words), so that they cannot dicted three findings (see Table 3): The fact that the duration of
easily be distinguished by a category difference, distractors tend to distractor processing depends on the variability of distractors (B2), the
be confused with items, leading to an above-chance rate of intru- finding that impairment of memory by processing is reduced when
sions of distractors in recall. Distractor intrusions become more distractors come from a different domain than the memoranda (B4),
prevalent when the similarity between items and distractors within and the fact that it is also reduced when they come from a different
a class of stimuli (e.g., words) is increased. At the same time, class of stimuli (B6). The resource hypothesis predicts only one
distractors more similar to memory items create less interference finding, the cross-domain impairment of memory by processing (B5),
by superposition, reducing the prevalence of other kinds of errors and the decay hypothesis predicts none. Conversely, both the
(i.e., transpositions, other extralist intrusions). Both of these pre- decay and the resource hypothesis are challenged by two findings (B2,
dicted effects have been confirmed experimentally (Oberauer, B6), and the resource hypothesis faces the additional problem of being
Farrell, et al., 2012): When distractors were made similar to the difficult to reconcile with the cognitive-load effect (B1). The inter-
immediately preceding memory items, people were more likely to ference hypothesis is challenged by only one finding: the loss of
recall the correct item, but when they did make an error, they were memory over unfilled retention intervals for some visual and spatial
more likely to confuse the item with the following (similar) dis- memoranda (B3).
tractor, compared with a condition where distractors were dissim-
ilar to all items.
Round C: Individual Differences
Second, when distractors come from a different category than the
memoranda within the same content domain (e.g., letters and digits), Correlations of measures of WM capacity with other variables
interference by confusion is minimal, so that the detrimental effect of can be used to test hypotheses about what causes the capacity
processing on memory is less severe than when distractors come from limit: Whereas a positive correlation between WM capacity and a
the same category. This prediction is borne out by the heterogeneity putative cause—for instance, processing speed— does not imply
benefit (B6; Conlin & Gathercole, 2006; Conlin et al., 2005; Li, 1999; causation, the absence of such a correlation seriously challenges
Turner & Engle, 1989; but see Macken & Jones, 1995). the hypothetical causal link (Underwood, 1975). Conversely, cor-
Third, when the distractors come from a different domain than the relational data can also serve to explore the scope of the WM
memoranda (e.g., verbal vs. spatial), interference is reduced compared capacity limit, asking which cognitive functions and processes are
to distractors from the same domain because contents from different limited to what extent by that capacity limit. The following five
domains have less feature-space overlap, reducing interference by findings from individual-differences research, summarized in Ta-
superposition (as well as interference by feature overwriting). This ble 4, qualify as diagnostic because they speak either to potential
prediction has also been confirmed (B4; Chein et al., 2011; Davis et causes or to the scope of WM capacity, or both.
al., 2013; Hale et al., 1996; Jarrold et al., 2011). First, there is the hierarchical factorial structure of WM
Several studies have found impairment of maintenance by pro- capacity tests, which has been consistently obtained across
cessing of materials in a different domain, compared with a no- studies that used a broad set of WM tests (Table 4, C1): WM
processing control condition (B5; e.g., Chein et al., 2011; Jarrold capacity is a notably general source of variance between indi-
et al., 2011). An interference account can explain these findings by viduals, as shown by the fact that a large variety of tasks used
assuming that distractor processing engages not only representa- to measure it load strongly on a common factor (Kane et al.,
tions of the stimuli to be processed but also of the responses, the 2004; Wilhelm, Hildebrandt, & Oberauer, 2013). Yet, on a
task set, and perhaps of executive control settings. Even if there is lower level of generality, separate factors for verbal-numerical
no feature-space overlap between the memoranda and the distrac- and for visual-spatial WM tasks can be distinguished (Alloway,
tor stimuli, there is arguably feature-space overlap between the Gathercole, & Pickering, 2006; Kane et al., 2004; Oberauer et
memoranda and other representations involved in the processing al., 2000; Shah & Miyake, 1996). Figure 12 shows results from
 WM CAPACITY 781

Table 4
Summary of Informative Findings and Evaluations of Hypotheses in Round C: Individual Differences

Index Finding Decay Resource Interference

C1 The factorial structure of WM tasks includes a general factor of WM capacity (C1a) ⫹ ⫹⫹ ⫹

together with domain-specific factors for verbal-numerical and for visual-spatial
WM (C1b)
C2 WM capacity correlates with processing speed, in particular with the drift-rate ⫹⫹ ⫹ 0
parameter of the diffusion model of response-time distributions from speeded choice
tasks
C3 WM capacity correlates with measures of long-term memory ⫹ ⫹ ⫹
C4 WM capacity correlates with resistance to distraction in attention tasks ⫺ ⫹ ⫹
C5 Some valid measures of WM capacity involve no memory requirement ⫺ ⫹⫹ ⫹⫹
Note. Table entries reflect our judgment of the logical relation between a finding and a hypothesis: The hypothesis predicts (⫹⫹) or can explain (⫹) the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

finding, it is consistent with the finding (0) or it is challenged by the finding (⫺); see text for explanation. WM ⫽ working memory.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

a representative study illustrating the generality and the cess in overcoming distraction in simple attentional paradigms
domain-specificity of individual differences in WM capacity. (C4), such as the antisaccade task (Chuderski, 2014; Shipstead,
Second, WM capacity is correlated with speed on simple tasks, Lindsey, Marshall, & Engle, 2014), the Stroop task (Kane &
in particular with the efficiency of information processing in Engle, 2003; Meier & Kane, 2013), the flanker task (Heitz &
speeded choice tasks (C2; Ratcliff, Thapar, & McKoon, 2010; Engle, 2007; but see Keye, Wilhelm, Oberauer, & van Ravenz-
Schmiedek, Oberauer, Wilhelm, Süß, & Wittmann, 2007; see waaij, 2009), and the prevalence of self-reported task-unrelated
Figure 13). Third, WM capacity has been found to be highly thoughts (McVay & Kane, 2009, 2012). Examples of frequently
correlated with measures of episodic long-term memory (C3; used paradigms for measuring controlled attention are given in
Unsworth, 2010; Unsworth, Brewer, & Spillers, 2009). Figure 14.
Our remaining two diagnostic findings pertain to the relation The fifth diagnostic finding concerns simultaneous attention to
between WM capacity and attention. These last two findings, C4 multiple elements and their relations: Tests of WM capacity based
and C5, further underscore that the scope of WM extends beyond on short-term recall, such as complex-span tasks, correlate highly
tests of immediate memory. The fourth finding is that measures of with performance on relational-integration tests (C5; Oberauer,
WM capacity are positively correlated with indicators of the suc- Süß, Wilhelm, & Wittmann, 2003). In these tasks, people monitor

Figure 12. Structural equation model of simple span tasks (STM) and complex span tasks (WMC) with verbal
(V) and spatial (S) memoranda, reproduced from The generality of working-memory capacity: A latent-variable
approach to verbal and visuo-spatial memory span and reasoning. Kane, M. J., Hambrick, D. Z., Tuholski, S. W.,
Wilhelm, O., Payne, T. W., & Engle, R. W. (2004). Journal of Experimental Psychology: General, 133 (p. 203).
Copyright 2004 by the American Psychological Association. Squares show manifest variables (i.e., measured
scores), and circles show latent variables (i.e., factors). Factors representing working memory (WM) capacity in
different domains (verbal vs. spatial) are distinct, but highly correlated, reflecting a substantial proportion of
general variance shared among them (finding C1).
 782 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 13. Components of response time in the diffusion model (Ratcliff, 1978). The sensory stage involves
stimulus processing and categorization. The central processing stage involves making a decision to select one of
two responses (e.g., whether the stimulus is a consonant or a vowel). The third stage involves motor execution
(e.g., pressing a button). The central stage is modeled as the accumulation of evidence by a diffusion process that
drifts toward one of two boundaries (dotted horizontal lines), representing the two response options. A decision
is made once a boundary is reached. The diffusion process on each trial is noisy (black line); its efficiency is
reflected by its average drift rate (slope of the red line). Estimates of the drift rate were found to correlate highly
with working memory capacity (finding C2; Ratcliff et al., 2010; Schmiedek et al., 2007). See the online article
for the color version of this figure.

a continuously changing array of visual stimuli to detect any part because of the general slowing of information processing in
instance in which a subset of the stimuli have a certain relation to old age, which in turn slows rehearsal of WM contents, leading to
each other (e.g., four dots forming a square, or two airplanes being a larger net loss through decay. Here we extend this argument to
on a collision course; see Figure 15 for examples). We regard this the two forms of restoration proposed in contemporary decay
finding as diagnostic because it demonstrates that WM capacity is theories: Individual and age-related variability in the speed of
not merely a limit on how much information we can remember domain-general attentional refreshing could explain the general
over a short period of time, but also on how much information in factor of WM, whereas variability in articulatory rehearsal and
the environment we can simultaneously attend to and integrate. spatial rehearsal could explain the domain-specific factors of ver-
We next examine how each of the three theoretical contenders bal and visual-spatial WM, respectively.
handles findings C1 to C5. Correlations with processing speed and articulation rate
(C2). One prediction following from the above assumptions is
Decay that independent measures of the efficiency of restoration pro-
cesses should correlate with measures of WM capacity. Evidence
Factorial structure of WM (C1). How does the decay hy-
speaking to this prediction is available from two sources. The first
pothesis fare in light of correlational findings concerning WM
capacity? The strong general factor reflecting the common vari- is the correlation between measures of WM capacity and indicators
ance of WM tests across different domains and paradigms (C1a) of the speed of attention-based refreshing (C2). Refreshing is
could be explained as reflecting individual differences in the assumed to be limited by the central attentional bottleneck (Bar-
general decay rate, or in the efficiency of attention-based refresh- rouillet et al., 2007). The speed of central processes in simple
ing. The domain-specific factors (C1b) could be attributed to the decision tasks, which require the central bottleneck, is reflected in
efficiency of domain-specific forms of rehearsal such as articula- the drift rate of the diffusion model of choice RTs (Sigman &
tory rehearsal for verbal materials, and rehearsal of spatial infor- Dehaene, 2005). The drift rate in turn is highly correlated with
mation through deployment of spatial attention. The explanation of WM capacity (Schmiedek et al., 2007). Moreover, Lee and
variability in WM capacity by variability in the speed of restora- Chabris (2013) demonstrated a direct relationship between the
tion flows directly from Jensen’s “limited-capacity trace-decay processing speed of the central bottleneck and fluid intelligence.
theory” (Jensen, 1988). Jensen assumed that individual differences These findings lend credibility to the idea that WM capacity
in WM capacity arise from differences in the speed of rehearsal. reflects to a substantial degree the efficiency of attention-based
Analogous arguments have been applied to developmental differ- refreshing.
ences: Kail (1992) has proposed that as children grow older, their The second line of evidence pertains to the efficiency of artic-
general processing speed increases, which enables them to re- ulatory rehearsal. Researchers have measured how long people
hearse faster, leading to better WM capacity (see also Gaillard, take to articulate verbal materials aloud as an indicator of their
Barrouillet, Jarrold, & Camos, 2011). Salthouse (1996) has pro- rehearsal speed, and used this measure as a predictor of perfor-
posed that the steep decline of WM capacity in old age is to a large mance on verbal serial recall. Earlier work found a positive cor-
 WM CAPACITY 783
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 14. Example trials of tasks for measuring controlled attention (finding C4). (A) Antisaccade task:
Participants must direct their gaze in the opposite direction of a flashing light to identify a stimulus presented
briefly and then masked. Controlled attention is measured by identification accuracy. (B) Stroop task: Partici-
pants must name the print color as quickly as possible. The first stimulus shows a congruent trial on which the
word matches the print color; the second an incongruent trial on which word and color mismatch. Controlled
attention is measured by the size of the congruency effect. (C) Flanker task. Participants make a speeded
classification on the central stimulus (pressing the left key for H and the right key for S), trying to ignore the
flanking stimuli, which can be congruent (first and second trial) or incongruent (third trial). Controlled attention
is measured as the size of the congruency effect. (D) Task-switch paradigm: Participants make speeded
classification on digits according to one of two task rules, indicated by a task cue preceding each trial. The task
switch cost is the difference between performance on trials requiring a task switch relative to the preceding trial
and performance on task-repetition trials. The congruency cost is the performance difference between trials in
which both tasks would require the same response and trials in which they would require different responses. See
the online article for the color version of this figure.

relation between articulation rate and serial recall performance provide no evidence that the speed of articulatory rehearsal has a
(e.g., Cowan et al., 1998; Kail, 1997). When controlling for the direct causal link to people’s performance on verbal WM tasks.
availability of verbal representations in long-term memory, such as Correlations with long-term memory (C3). We are not
the speed of lexical access (Tehan, Fogarty, & Ryan, 2004; Tehan aware of any attempt to apply a decay theory to explain the
& Lalor, 2000) or vocabulary (Ferguson & Bowey, 2005); how- correlation between WM and long-term memory (C3), but we
ever, measures of rehearsal speed did not account for significant envision two ways in which such an explanation could be worked
variance in serial recall. These findings suggest that individual out. One approach is to explain individual differences in WM
differences in lexical knowledge are a common cause of speed of capacity as arising from differences in the speed and effective-
lexical access, speed of articulation, and verbal serial recall. They ness of rehearsal or refreshing. These restoration processes can
 784 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 15. Example trials of two relational-integration tasks (Oberauer et al., 2003). (A) Finding-squares task:
From each display to the next, two dots change location at random. Participants must detect when four dots in
a display form a square. (B) Verbal monitoring task: From each display to the next one word is exchanged by
a new word. Participants must detect when three words in a row, in a column, or across a diagonal rhyme with
each other. These tasks are valid indicators of working memory (WM) capacity—they load highly on a WM
capacity factor—although they do not require retention of information across a retention interval (finding C5).
See the online article for the color version of this figure.

be argued to not only protect representations in WM from decay tion could build on the hypothesis that representations of task goals
but to also help establishing long-term memory traces. Whereas or task sets implementing the instructions decay over time (Alt-
articulatory maintenance rehearsal has only a limited effect on mann & Gray, 2002). Individual differences in many indicators of
long-term memory (Greene, 1987), refreshing has been shown attentional control can be attributed to failures of goal maintenance
to improve long-term retention (Raye et al., 2007), and provid- (Kane et al., 2007; Kane & Engle, 2003), which in turn could be
ing more time for refreshing during a WM task results in better attributed to decay.
recall of the memoranda in a delayed test (Camos & Portrat, There is scant evidence, however, that representations of task
2015; Loaiza & McCabe, 2012b) Therefore, the efficiency of goals or task sets in WM decay over time. Altmann and Gray
refreshing could be a source of common variance of WM and (2002) based their hypothesis of task-set decay on the observation
long-term memory. In line with this hypothesis, Loaiza and of a gradual increase of response times over successive repetitions
McCabe (2013) have argued that age differences in episodic of the same task in a task-switch paradigm. Subsequent work,
long-term memory can in part be explained by age differences however, showed that this gradual slowing arises not from decay,
in the efficiency of refreshing. but from people’s growing expectation of a task switch: When
The second approach starts from the assumption that indi- participants know the number of task repetitions before the next
vidual differences in decay rate (perhaps in conjunction with task switch, they anticipate the switch and slow down in prepara-
differences in restoration processes) determine how much in- tion for it; in contrast, when the number of task repetitions is
formation can be maintained in WM simultaneously, which in unpredictable, no such slowing is observed (Monsell, Sumner, &
turn determines the size of structures or chunks that can be Waters, 2003). Additional evidence against decay of task sets
formed and encoded into long-term memory. More complex comes from another finding from the task-switch paradigm: When
elaborations and larger chunks arguably improve memory over switching between three tasks, participants are slower to switch
the long term, and this could explain why people with higher back to a task that they have carried out two trials ago than to a
WM capacity measures tend to do better on tests of long-term task that they last carried out longer ago (Mayr & Keele, 2000; for
memory as well. To conclude, although the decay hypothesis a review see Koch, Gade, Schuch, & Philipp, 2010). This is the
does not directly predict the correlation between WM capacity opposite of what would be predicted from task-set decay. More-
and episodic long-term memory, it has no difficulty explaining over, Horoufchin, Philipp, and Koch (2011) have shown that the
it. effects of varying the time between successive tasks, which have
Correlations with measures of attention (C4, C5). In con- been attributed to task-set decay in earlier work, are better ex-
trast, decay-rehearsal theories do have difficulties explaining the plained by temporal distinctiveness than by trace decay. In sum-
correlation of WM capacity with indicators of attentional control mary, the evidence consistently goes against the assumption that
(C4), such as Stroop interference or performance in the antisaccade task representations decay, leaving little room for a decay-based
task (Kane, Conway, Hambrick, & Engle, 2007). To the best of our explanation for the correlation between WM capacity and perfor-
knowledge, no attempt has been made to explain the correlation mance in attention-control tasks.
between WM capacity and measures of attention or cognitive We close this section by considering a further prediction from a
control in terms of decay and restoration. One potential explana- decay account for individual differences: A valid test of WM
 WM CAPACITY 785

capacity must require maintenance over a nonnegligible RI during processes such as response selection (Sigman & Dehaene, 2006),
which individual differences in decay rate and in the efficiency of which are constrained by a domain-general capacity limit. This
restoration processes could influence performance. This prediction capacity limit has been modelled as a resource limit (Navon &
has not been borne out empirically (C5). WM capacity can be Miller, 2002; Tombu & Jolicoeur, 2003). Hence it would not be
measured by a monitoring paradigm that requires no maintenance far-fetched to identify that resource with the resource underlying
of information over any RI because all necessary information is WM capacity.
continuously visible (Oberauer et al., 2003): People watch a We are not aware of a proposal for explaining the correlation
changing set of stimuli and are asked to detect when a target between WM and episodic long-term memory (C3) within a re-
constellation occurs among any subset of stimuli, such as a square source theory. One approach could be formulated in analogy to a
among a subset of dots, or a row or column of rhyming words in decay-based account sketched above: Individuals with more WM
a matrix (see Figure 15). This paradigm is among the most valid maintenance resources can hold larger sets of items available
indicators of WM capacity, judged by its loadings on a general simultaneously, enabling them to form larger chunks and more
WM capacity factor, and among the best predictors of fluid intel- elaborate structures to be encoded into long-term memory. An-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ligence (Buehner, Krumm, & Pick, 2005; Buehner, Krumm, other approach could start from the assumption that retrieval from
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Ziegler, & Pluecken, 2006; Chuderski, 2014; Chuderski, Taraday, long-term memory depends on the same resource as maintenance
Ne˛cka, & Smoleń, 2012; Oberauer, Süß, Wilhelm, & Wittmann, in, or retrieval from, WM. This notion could be justified with the
2008). Individual differences in a task without an RI cannot be fact that retrieval from long-term memory is susceptible to dual-
explained by differences in decay rate or efficiency of restoration task interference (Carrier & Pashler, 1995; Rohrer & Pashler,
processes; therefore, these findings render it highly unlikely that 2003), and it would provide a natural link between the resource
those variables contribute substantially to explaining individual hypothesis and recent theoretical developments by Unsworth and
differences in general WM capacity. Engle (2007). We conclude that the correlation between WM and
Conclusion. The decay hypothesis provides a satisfactory ex- long-term memory measures does not pose a fundamental chal-
planation for the factorial structure of WM capacity, and its cor- lenge for resource theories.
relation with processing speed and episodic long-term memory. It Correlations with measures of attention (C4, C5). The re-
is challenged, however, by the correlation of WM capacity with source assumption can also explain why WM capacity measures
performance on tasks that place little, if any, demand on the are correlated to several measures of attention (C4). Attention is
maintenance of information over time, such as attention-control often characterized as a limited resource, and if that resource
tasks and perceptual monitoring tasks. overlaps with or is identical to the resource underlying WM
capacity, their positive correlation follows as a necessary predic-
tion. At the same time, assuming a general resource that fuels not
Resources
only maintenance in WM but also various attentional functions
The notion of resources has often been invoked to explain the risks diluting the resource concept to a point where it is little more
pattern of correlations of WM tests with each other and with other than a redescription of the correlational findings. For such a
variables: When performance in two tasks is positively correlated, concept to become testable it would be necessary to specify what
researchers routinely assume that they draw in part on the same the resource does in each of the attentional paradigms in which it
resource. Factor analytic findings are interpreted by assuming that is deemed relevant, that is, to characterize its performance-
each factor stands for a resource. Often these interpretations are resource functions for those attentional paradigms. Combined with
merely redescriptions of the findings, because identifying each such specifications, the resource hypothesis would probably not
factor with a resource does not explain why the correlational predict that WM capacity correlates with every measure of atten-
patterns underlying the factor structure are the way they are—any tional function to the same degree, but would rather predict cor-
other factor structure could equally be interpreted in terms of relations specifically with variables sensitive to the shared re-
resources. Resource accounts of individual differences gain ex- source.
planatory value if a resource theory places constraints on the For instance, it could be argued that the resource underlying
resources assumed to exist, so that predictions for the factor WM capacity is needed to maintain a strong representation of a
structure can be made. task goal to avoid goal neglect. Goal neglect refers to the failure to
Factorial structure of WM (C1). As discussed in the preced- implement a goal despite knowing and being committed to that
ing two rounds, the findings of both domain-general and domain- goal. For instance, participants in a Stroop experiment (Figure 14
specific set-size effects, and effects of distractor processing on B) occasionally read the color word instead of reporting its print
memory, require the assumption of a domain-general resource color, despite knowing that they were supposed to do the latter,
together with domain-specific resources for verbal and for visual- and individuals with lower WM capacity commit this kind of error
spatial materials. This set of assumptions matches well with the more frequently (Kane & Engle, 2003). A resource explanation of
WM model of Baddeley (Alloway et al., 2006; Baddeley, 2001, this finding could assume that the WM resource is needed for
2012), and it directly predicts the factor structure of WM capacity maintaining the relevant goal (e.g., naming the print color), and
measures (C1). when the resource runs low, the goal risks losing the competition
Correlations with processing speed and long-term memory against a conflicting habit (e.g., reading the color word). This
(C2, C3). The resource account also offers an explanation for the explanation implies the prediction that WM capacity correlates
correlation of WM capacity and processing efficiency on simple with performance on attentional tasks involving a conflict between
speeded tasks, as reflected in the drift rate of the diffusion model the relevant goal and a strong competing goal or habit, because
(C2). The drift rate has been shown to reflect the speed of central these paradigms require strong goal maintenance to prevent goal
 786 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

neglect. The Stroop task is an instance of an attentional paradigm bined with the assumption that the WM resource is required for
with high goal conflict. Another paradigm inducing goal conflict is goal maintenance, provides a successful explanation of the pattern
the antisaccade paradigm (Figure 14 A). In this paradigm, a visual of correlations of WM capacity with indicators of attentional
cue is flashed on one side of the screen, and participants must control.
make a saccade (i.e., an eye movement) to a target appearing on Finally, if the resource limiting WM capacity is conceptualized
the other side, thereby overcoming the habit of moving the eyes as an attentional resource, it must be expected to also limit the
toward a sudden-onset stimulus. capacity for simultaneously attending to multiple objects in the
In contrast, WM capacity should be predicted to correlate less environment. Such a resource account predicts that measures of
with attentional paradigms in which goal maintenance is less WM capacity correlate with performance on monitoring tasks and
important. For instance, in the flanker paradigm (Figure 14 C) other tasks for measuring relational integration that involve no
participants make speeded responses to a centrally presented stim- retention interval (C5).
ulus flanked by distractor stimuli that can be incongruent with the Conclusion. The resource hypothesis—with the assumption
target stimulus (i.e., they are linked to another response than the of domain-general and domain-specific resources—predicts the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

target), and therefore, must be ignored. A high attention-control factorial structure of WM capacity. It also provides an explanation
This document is copyrighted by the American Psychological Association or one of its allied publishers.

score in the flanker task (i.e., a small cost of incongruent compared for the correlation of WM capacity with processing speed and
to congruent or neutral flankers) does not require minimizing the episodic memory. The resource account, combined with the as-
influence of a competing goal or habit, but minimizing the influ- sumption that goal maintenance requires the WM resource, can
ence of distracting stimuli. The same can be said for visual search offer a detailed and largely accurate account of the pattern of
paradigms, in which efficient search requires attentional filtering correlations of WM capacity with measures reflecting aspects of
of the distractors. attentional control. Finally, the resource hypothesis also correctly
The task-switch paradigm (Figure 14D; Rogers & Monsell, predicts that a measure of WM capacity does not necessarily
1995) also entails strong goal conflict because when participants involve a memory demand.
switch back and forth between two tasks, the currently irrelevant
task still has a strong tendency to intrude in response selection.
Interference
Strong goal maintenance is therefore needed to carry out the
currently relevant task and avoid distraction from the irrelevant An interference account of WM capacity does not point to an
task. However, strong goal maintenance does not help, and perhaps obvious source of individual differences that generalizes across a
even stands in the way of, rapid, seamless switching between two broad range of paradigms and content domains. There are a num-
tasks. Therefore, individuals with good goal maintenance would ber of parameters in interference models that could vary across
not be expected to have smaller task-switch costs than individuals individuals and explain individual and developmental differences
with poor goal maintenance (Herd et al., 2014). Rather, individuals in WM capacity and their factorial structure (C1), and recent work
with good goal maintenance could be predicted to have smaller has explored some of these possibilities.
task-congruency costs, that is, smaller performance costs on trials Factorial structure of WM (C1). One general source of
with conflict between the currently relevant and the currently individual differences could be the ability to control the contents of
irrelevant task. WM by preventing access of irrelevant material (“filtering”) and
To summarize, a resource account linking WM capacity to goal by removing WM contents that are no longer relevant (Oberauer,
maintenance predicts that WM capacity is correlated with success- Lewandowsky, et al., 2012). Evidence for a role of filtering and
ful attentional control on paradigms with high goal conflict, such removal in explaining individual differences in WM capacity is
as the Stroop task and the antisaccade task, but not on paradigms mixed at best. Some findings suggest that individual differences
with low goal conflict, such as the flanker task and visual search. and age differences in WM capacity are related to the efficiency of
For the task-switch paradigm this account entails the prediction filtering out irrelevant stimuli (Jost, Bryck, Vogel, & Mayr, 2011;
that WM capacity is correlated with the congruency effect, but not Vogel, McCollough, & Machizawa, 2005), whereas others speak
with the task-switch cost. against such an association (Cowan, Morey, AuBuchon, Zwilling,
Extant findings provide support for this set of predictions: & Gilchrist, 2010; Mall, Morey, Wolff, & Lehnert, 2014). There is
Indicators of attention from goal-conflict paradigms have been preliminary evidence that the ability to remove information from
found to correlate with WM capacity (e.g., the Stroop effect, Kane WM declines with adult age (Cansino, Guzzon, Martinelli,
& Engle, 2003; performance in the antisaccade task, Unsworth, Barollo, & Casco, 2011). However, one individual-differences
Schrock, & Engle, 2004). In contrast, indicators of attention from study with a memory-updating paradigm found no correlation
paradigms without goal conflict have often been found to have between measures of WM capacity and the efficiency of removal
only negligible correlations with WM capacity (e.g., the flanker of outdated information from WM (Ecker, Lewandowsky, et al.,
task; Keye et al., 2009; Wilhelm et al., 2013; and most paradigms 2014).
of visual search; Kane, Poole, Tuholski, & Engle, 2006; Sobel, The distinctiveness of representations in long-term memory
Gerrie, Poole, & Kane, 2007). Task switch costs are virtually could be a source of domain-specific individual differences. Dis-
uncorrelated with WM capacity (Oberauer, Süß, Wilhelm, & tinctive long-term memory representations play an important role
Sander, 2007). The congruency effect in the task switching para- for retrieval from WM. There is broad agreement among WM
digm has, unfortunately, so far received little attention in researchers that retrieval of an item from WM often returns a
individual-differences research (for a recent exception see Stahl et distorted representation of the original item, which needs to be
al., 2014), so the prediction that it correlates with WM capacity disambiguated through a process often referred to as redintegra-
remains untested. We conclude that the resource hypothesis, com- tion (Hulme, Roodenrys, Brown, & Mercer, 1995; Lewandowsky,
 WM CAPACITY 787

1999; Schweickert, 1993). Redintegration relies on comparing the testing WM) as well as for immediate or delayed recall of longer
distorted representation of an item retrieved from WM to long- lists (as used for testing long-term memory), variations in this
term memory representations of known items in the set of recall parameter also contributed to the common variance of indicators of
candidates. Theories differ in what causes the distortion of mem- WM and of episodic long-term memory in the model. Hence, at
ory traces—in interference models, distortion arises from interfer- least one instantiation of an interference model provides an expla-
ence by superposition. Individuals with more distinctive long-term nation for the correlation between WM capacity and long-term
knowledge can be expected to redintegrate more successfully. The memory (C3).
distinctiveness of long-term knowledge arguably reflects at least in Correlations with measures of attention (C4, C5). How
part the person’s level of expertise in a content domain, so that could an interference account explain the relation between WM
distinctiveness might vary independently in different domains. capacity and resistance to distraction in attentional paradigms
Therefore, individual differences in distinctiveness of long-term (C4)? So far no such explanation has been worked out, so we can
memory representations could explain the domain-specific source only offer a speculative sketch. Performance in attention-control
of variance in WM capacity. paradigms such as the Stroop, the flanker, or the antisaccade tasks
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

A simulation study with the SOB-CS model implemented indi- relies on task sets implementing the instructions. Task sets are
This document is copyrighted by the American Psychological Association or one of its allied publishers.

vidual differences in the removal of irrelevant information as a procedural representations in WM that link conditions (e.g., target
domain-general source of variation, together with differences in stimuli) to actions (e.g., pressing a button). These representations
memory distinctiveness as a domain-specific source (Oberauer, are in principle vulnerable to interference in the same way as other
Lewandowsky, et al., 2012). With these assumptions, the model (declarative) representations in WM. Interference can arise from
was able to reproduce the factorial structure of simple and complex competing task sets. For instance, in the antisaccade task the
span tasks (C1; Kane et al., 2004). habitual task set for moving the eyes toward a flashing light in the
Correlations with processing speed (C2). The interference environment could interfere with the instructed task set for moving
hypothesis does not lend itself to a straightforward explanation of the eyes in the opposite direction, away from the flashing cue. In
why WM capacity is correlated with processing speed. One pos- the task-switch paradigm, proactive interference arises from the
sibility is that interference between representations in procedural currently not relevant task set that has been carried out just seconds
WM influence processing speed (Oberauer, 2009). Procedural ago (Allport, Styles, & Hsieh, 1994). People with high WM
WM holds the current task set—the relevant stimulus and response capacity might be good at protecting the current task set from
categories and the mappings between them. The distinctiveness interference by competing procedural representations, such as re-
of stimulus and response representations, and the robustness of cently used task sets or habits (i.e., strong stimulus–response
bindings between them, can be expected to determine the effi- associations in long-term memory) by either filtering them (i.e.,
ciency of response selection, which translates into the drift rate preventing them from intruding into procedural WM) or by re-
of the diffusion model (Schmiedek et al., 2007). This could moving them from procedural WM (Oberauer, Souza, Druey, &
explain why WM capacity is correlated specifically with the Gade, 2013).
drift rate (C2). As an explanation along these lines has not been This set of assumptions is similar to the idea discussed above of
worked out yet, a conservative assessment is that the interfer- a WM resource responsible for goal maintenance, and it engenders
ence hypothesis is consistent with finding C2, but it does not yet a similar set of predictions: Individuals who are good at establish-
offer an explanation for it. ing robust task sets in procedural WM and protecting them against
Correlations with long-term memory (C3). Differences be- interference should be more successful in overcoming conflict
tween people in their susceptibility to interference could also arise from representations of competing stimulus-response mappings.
from differences in the distinctiveness of context representations Therefore, high WM capacity should be correlated with lower
(see Figure 1). For instance, individuals with more distinct context Stroop interference and better performance in the antisaccade task.
representations, such as list positions, are expected to perform High-capacity individuals should also be better at avoiding mind
better in remembering lists in order, because they are less likely to wandering (McVay & Kane, 2009) by filtering or removing task-
confuse items from different positions, and suffer less interference irrelevant representations from (declarative and procedural) WM.
from superposition of item-context bindings. Differences in con- In the task-switch paradigm individuals with high WM capacity
textual distinctiveness have been shown to contribute to age dif- should show smaller costs of task incongruency. The predictions
ferences in serial recall at the beginning (McCormack, Brown, for task-switch costs depend on how WM capacity is assumed to
Vousden, & Henson, 2000) and at the end of the life span (Maylor, be related to the two processes of controlling the contents of WM,
Vousden, & Brown, 1999). On a more global level, more distinc- filtering and removal. The ability to protect the current task set
tive contexts also serve to distinguish the current memory set from against interference by preventing other representations from en-
those of previous trials, reducing proactive interference—this as- tering procedural WM (i.e., filtering) should, if anything, hinder
sumption could explain why WM capacity is correlated with the the rapid reconfiguration of the task set when a switch to another
susceptibility to proactive interference (Kane & Engle, 2000). task is required. In contrast, the ability to remove representations
In one interference-based computational model of serial and free from WM when they are no longer needed should facilitate task
recall (Farrell, 2012), variability in the distinctiveness of context switching. If people with high WM are good at both filtering and
representations serves as a key source of individual differences of removal, the opposing effect of these processes on task-switch
WM capacity. Simulations with this model provide a detailed costs should result in at best a small correlation of task-switch
account of differences in recall behavior between individuals with costs with WM capacity.
high and with low WM capacity. Because contextual distinctive- In the flanker paradigm, conflict from the flankers arises
ness is relevant for immediate memory of short lists (as used for through the same stimulus-response bindings that mediate the
 788 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

correct response. Therefore, individuals who are able to establish WM capacity with monitoring of multiple visual stimuli (C5). The
strong stimulus-response bindings in procedural WM should be interference hypothesis, by contrast, predicts only the latter and it
more efficient in translating both the target stimulus and the offers a more speculative explanation for the correlation of WM
flankers into representations of the responses mapped to them— capacity with speed measures (C2) than the resource hypothesis.
and when these responses are in conflict with each other, perfor-
mance will suffer no less than for a person with a weaker task set.
Discussion
Therefore, the size of the flanker effect is not predicted to correlate
with WM capacity. In visual-search tasks, no conflicting action We have evaluated three hypotheses about why the capacity of
tendency needs to be overcome, so there is no reason to predict a WM is limited by matching predictions from each hypothesis
correlation of search efficiency with WM capacity. against a set of relevant and diagnostic findings. The assumption
To summarize, the interference hypothesis, when applied to that representations in WM are lost because of rapid decay has
attention-control paradigms along the lines sketched above, can appeal because it is simple and matches our personal experience of
explain the pattern of correlations of WM capacity with indicators rapid forgetting of new information (Jonides et al., 2008). Much of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

from the attention-control tasks that we already reviewed in con- the evidence we have reviewed above, however, speaks against
This document is copyrighted by the American Psychological Association or one of its allied publishers.

nection with goal maintenance in the Resource section: WM decay having a primary role in limiting WM capacity. For verbal
capacity is correlated with the success of overcoming conflict in memoranda the evidence is against decay playing a role in deter-
the Stroop and the antisaccade task, and more generally with the mining retention over the short term; for visual and spatial mem-
ability to prevent intrusions from task-unrelated representations oranda decay might play a role, but is unlikely to determine the
into WM. WM capacity is only negligibly correlated with the capacity limit, because the rate of forgetting that could be attrib-
flanker effect, with task-switch costs, and the efficiency of visual uted to decay is too slow to explain the severe capacity limit
search. Therefore, the interference hypothesis, together with the observed at RIs of just one or two seconds, or even in the absence
assumption that individual differences in WM capacity arise in of any RI (Oberauer et al., 2003; Tsubomi et al., 2013).
part from differences in the effectiveness of filtering and removal, An explanation of WM capacity in terms of resources has
can explain the relations of WM capacity to indicators of atten- considerable strengths but also serious limitations. The main
tional control (C4), although many details of that explanation need strength of this approach is that it explains why memory for some
to be worked out. WM content is often found to be impaired by the concurrent
Finally, on the interference hypothesis we should expect that maintenance or processing of material that appears to have little in
individual differences in WM capacity affect performance on any common with that content. There are two main limitations: The
task that requires access to multiple distinct representations at the resource account cannot explain why memory is impaired more by
same time, whether these are representations of past events (i.e., simultaneous maintenance or processing of material from the same
memory representations) or of stimuli in the environment. There- category than of materials from different categories within a do-
fore, the interference hypothesis provides a natural explanation for main, and resource models cannot offer a coherent explanation for
the fact that monitoring tasks—requiring simultaneous access to how distractor processing impairs memory. In particular, a re-
multiple elements to determine their relations—are as valid mea- source account cannot explain why a longer duration of distractor
sures of WM capacity as tasks measuring STM (C5). processing impairs memory if and only if the distractors differ
Conclusion. Taken together, interference accounts can ex- from each other, and it cannot explain why decreasing cognitive
plain what is known about the correlational structure of WM load by adding free time in between distractors improves memory.
capacity indicators. This explanatory potential has been demon- The interference hypothesis offers a viable account of most of
strated by a simulation with SOB-CS reproducing the factorial the findings in Tables 2 to 4. However, we identified two limita-
structure of a broad range of memory span tests (Oberauer, Le- tions: First, interference does not offer a natural explanation for the
wandowsky, et al., 2012). This explanatory success, however, does observations of time-based forgetting over unfilled RIs when tem-
not arise from the interference hypothesis on its own, but in poral distinctiveness is controlled. Second, interference provides
conjunction with additional assumptions about the sources of no straightforward explanation for why maintenance of a memory
individual differences. Therefore, interference theories do not pre- set is impaired by simultaneous maintenance or processing of other
dict a specific factor structure, and the source of individual differ- materials that have no apparent feature-space overlap with the
ences in interference models of WM is yet to be determined. memory set. These challenges do not appear to be insurmount-
Distinctiveness of representations, together with the effectiveness able—we rather see them as a call for more in-depth analysis of the
of processes that control the contents of WM, are likely to play a representations actually recruited when maintaining or processing
central role in an interference-based explanation of individual the materials in question. In conclusion, we argue that interference
differences. is a promising approach to explaining the capacity limit of WM,
although more theoretical and empirical work needs to be done to
fully realize its potential.
Round C: Summary
Table 4 presents the score sheet for round C. The decay hypoth-
No Family Wise Knock-Out Blows
esis struggled to explain why WM capacity is correlated with
measures of attention that are not prone to decay (C4, C5). The One difficulty in evaluating the three hypotheses is that each of
resource and the interference hypothesis both fared well, with a them actually represents an entire family of possible explanations,
better score for the resource hypothesis because it predicts two consisting of a potentially innumerable set of variants. The decay
findings, the factor structure of WM (C1) and the correlation of hypothesis is invariably accompanied by the assumption of one or
 WM CAPACITY 789

several restoration mechanisms, and the predictions of any decay hana, Zhou, Geller, & Sekuler, 2007). The similarity of high-
theory depend substantially on the details of how restoration is dimensional stimuli such as letters or words can be assessed
thought to work (for a glimpse at the multiplicity of possible through similarity ratings or acoustic confusion measurements,
approaches see Chapter 2 of Lewandowsky & Farrell, 2011). The which can be submitted to multidimensional scaling to model the
resource hypothesis can be fleshed out in many different ways feature space (Farrell, 2006; Lewandowsky & Farrell, 2008a).
concerning the number and scope of resources and the Another approach might be to assess the similarity of patterns of
performance-resource functions for translating resource quantities neural activity during maintenance of different kinds of WM
into expected performance. The interference hypothesis reflects a contents (Kriegeskorte & Kievit, 2013; Kriegeskorte, Mur, &
family of different mechanisms of interference and their combi- Bandettini, 2008). Cross-dimensional congruency effects such as
nations. Therefore, all three hypotheses are highly flexible in what the SNARC5 effect (Nuerk, Wood, & Willmes, 2005) could also
they predict. We have tried to nevertheless pin down predictions be used to detect overlaps of feature spaces of stimuli from
that follow from the basic hypothesis in question irrespective of different domains, such as numbers and spatial locations (cf.
the details, but we cannot logically rule out that versions of each Walsh, 2003).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

hypothesis can be created that escape the challenges we have

This document is copyrighted by the American Psychological Association or one of its allied publishers.

noted. Combined Explanations

One troublesome aspect of the flexibility of all three hypotheses
is that they raise the temptation of circular explanations. In the So far we have focused our investigation on how well decay,
context of decay theories, when forgetting over time is observed, resource limits, or interference can explain the WM capacity limit
researchers conclude that restoration processes were not possible, on their own. This enabled us to identify the strengths and weak-
or insufficient to compensate decay, whereas when no forgetting nesses of each hypothesis in isolation, and provided an evaluation
over time is observed, it is concluded that some form of rehearsal of the most parsimonious explanations of WM capacity. In light of
or refreshing must have prevented decay. This reasoning is circular the fact that all of these explanations face some challenges, we
as long as there is no independent measure of rehearsal or refresh- next ask whether combinations of two or all three of the above
ing, or of the opportunity for engaging in such a restoration hypotheses could provide a more powerful explanation. Some
process. An independent assay of articulatory rehearsal can be theories of WM build on such combinations—models based on
obtained by asking people to rehearse overtly (Tan & Ward, 2008). ACT-R, for instance, combine a resource limit with decay and
Attention-based refreshing is more difficult to measure, and to date interference by confusion (Lovett et al., 1999), and Cowan’s
there is no independent evidence that people engage in refreshing embedded-process theory combines a central, domain-general re-
during WM tasks at all—rather, the occurrence of refreshing is source limit, the focus of attention, with the ability to hold infor-
inferred from the performance data it is meant to explain. Never- mation in the activated part of long-term memory, where they are
theless, at least the opportunity for refreshing—if not the process susceptible to interference and decay (Cowan, 2005).
of refreshing itself— can be assessed by measuring for how long a We argue that any combination that includes a role for decay in
distractor task engages the attentional bottleneck and setting that limiting WM capacity faces difficulties in explaining three find-
time in relation to the time available for the distractor task (Ober- ings: First, there is no forgetting for verbal memory lists over
auer & Lewandowsky, 2013). delays— of 10 s and more— during which both articulatory re-
Resource theories risk becoming circular when the existence of hearsal and attention-based refreshing are engaged by a concurrent
shared resources is inferred from the observation of mutual im- processing demand (Oberauer & Lewandowsky, 2008, 2013). Sec-
pairment of two concurrent tasks, whereas the existence of sepa- ond, the measured capacity for visual stimuli is the same imme-
rate resources is inferred from the (relative) lack of dual-task costs. diately after encoding— before any decay could have hap-
There is no obvious way of measuring the resource demand of a pened—as it is after a 1 s delay (Tsubomi et al., 2013). Third, some
task or process independently of dual-task costs. This is why the of the most valid tasks for measuring WM capacity involve no
resource concept by itself is virtually untestable, as has been noted retention interval (Chuderski, 2014; Oberauer et al., 2003). These
long ago (Navon, 1984). A testable resource theory of WM needs findings leave little room for a contribution of decay to an expla-
to specify which resources exist, what each resource is needed for, nation of the capacity limit.
what its performance-resource function is, and how multiple re- In contrast, a combination of interference with a domain-general
sources operate together (i.e., whether their contributions to a resource limit appears viable. We note that the limitations of
process are combined additively or interactively). Whereas single- resource accounts and interference accounts are complementary:
resource theories meeting these requirements have been proposed The resource hypothesis is challenged by findings that are ex-
(Anderson et al., 1996; Ma et al., 2014), there is no equally plained well by interference, most notably the effects of set het-
well-defined multiple-resource theory of WM to date. erogeneity on the degree to which maintenance is impaired by
Interference theories are at risk of circular explanations when other memory contents or by concurrent processing, and the find-
researchers infer the degree of similarity or feature-space overlap ing that the duration of distractor processing matters if and only if
between two kinds of representations from the observed degree of distractors vary over time. Conversely, the assumption of a general
mutual impairment of tasks recruiting these representations. To resource provides a natural account for the mutual disruption of
escape circularity, interference theorists need to find ways to representations in WM when there is no apparent feature-space
assess similarity and feature-space overlap independently of their
consequences for memory performance. One way to achieve this is 5
Spatial-Numerical Association of Response Codes: People respond
to use stimuli varying in very low-dimensional, well-defined fea- faster with a right key press when making judgments on larger numbers,
ture spaces such as color, orientation, or spatial frequency (Ka- and faster with a left key press when judging smaller numbers.
 790 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

overlap between them. A combination of a general resource with heterogeneous memory sets is more than just an instance of the
interference fits well with theoretical frameworks that explain benefit of dissimilarity within a memory set (Conrad & Hull,
performance on WM tasks as being supported by at least two 1964; Poirier, Saint-Aubin, Musselwhite, Mohanadas, & Ma-
mechanisms: A domain-general core system limited to hold one or hammed, 2007).
a few chunks, referred to as primary memory (Unsworth & Engle, Theoretical prospects. We identified two promising avenues
2007) or the focus of attention (Cowan, 2005; McElree, 2006), for understanding the capacity limit of WM, a purely interference-
supplemented by mechanisms for maintaining and retrieving in- based model, or a model combining interference with a limited
formation in long-term memory over short periods of time. Be- resource. Here we highlight a few challenges that theorists will
cause retrieval from long-term memory is generally assumed to be have to meet to develop these approaches further.
limited by interference, it would be reasonable to assume that A first question for an interference theory of WM capacity is
interference constrains the contribution of the supplementary whether—and if so, how—interference in WM differs from
mechanisms. In contrast, the core system might be a limited interference in long-term memory. Interference limits our abil-
resource. In light of the fact that interference alone explains most ity to remember events and facts over the long term, but
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of the findings indicative of the WM capacity limit, the scope of long-term memory is not constrained by a severe capacity limit
This document is copyrighted by the American Psychological Association or one of its allied publishers.

the core mechanism might be very limited. A thorough investiga- of the kind that characterizes WM. From the perspective of
tion of the mutual disruption of maintenance of verbal and visual- unitary memory models such as SIMPLE (G. D. A. Brown et
spatial memory sets led Cowan et al. (2014) to the conclusion that al., 2007) or the temporal-context model (Howard & Kahana,
the domain-general core mechanism holds just one item. There- 2002; Sederberg, Howard, & Kahana, 2008), there is no qual-
fore, the core component of the WM system might be a focus of itative difference between WM and long-term memory: The
attention holding, in most circumstances, a single item or chunk contents of WM are simply those memory contents that are best
(McElree, 2006; Oberauer & Hein, 2012). accessible, given the currently available retrieval cues. From
this perspective, the capacity limit of WM is merely a reflection
of the general limit on our ability to retrieve information from
Outlook
memory. One proposal for demarcating a special role for WM
Where to from here? In this final section we briefly sketch within a unitary framework is that the contents of WM can be
possible avenues for advancing our understanding of the capacity accessed directly from the currently active context, whereas
limit of WM through further empirical and theoretical work. retrieval from episodic long-term memory requires first retriev-
Empirical desiderata. Whereas most of the findings in Ta- ing their context, which then can be used as cue to retrieve the
bles 2 to 4 are clear-cut phenomena with solid empirical support, content associated to it (Farrell, 2012).
our review identified three areas in need of further empirical Whereas unitary models emphasize the continuity of WM with
consolidation. First, it has emerged that the set-size effect on long-term memory, they tend to neglect the close link of WM to
accuracy is an effect not only of the number of elements or chunks, attention. As we have noted throughout this review, the limited
but also their complexity (A1), but our knowledge of the effects of capacity of WM applies not only to memory for recent events but
complexity remains patchy. In our review we summarized several also to apprehension of information in the present perceptual
findings under the umbrella term of “complexity effects,” but it is environment, for instance when monitoring the relations between
far from clear that, for instance, the number of phonemes in a multiple stimuli (Oberauer et al., 2003), or when reporting visual
word, the number of words in a chunk, and the number of features features of objects that have been masked only a few milliseconds
of a visual object all reflect the same kind of complexity. Com- before (Sewell et al., 2014; Tsubomi et al., 2013). One task for
plexity is a complex term, encompassing a variety of ways in further developing interference models of WM is to apply them to
which characteristics of memoranda can be varied, and we have interference between representations of multiple objects attended
only just begun to chart this territory empirically. to simultaneously.
Second, the role of time in forgetting of visual and spatial We also need to work out how interference affects the rep-
memoranda (B3) is in need of further clarification: Under which resentation of task sets in procedural WM to understand why
conditions does memory decline over an unfilled retention interval, WM capacity is correlated with the efficiency of response
or a filled retention interval? When such a decline is observed, is selection in simple speeded choice tasks (Schmiedek et al.,
it because of decay or related to reduced temporal distinctiveness? 2007), and with measures of controlled attention (Kane et al.,
The mixed evidence on these questions reflects the large variety of 2007). This effort could build on modeling work that aims to
materials and procedures used for addressing them, and it will take understand WM and executive control within a unitary frame-
a systematic effort to tease apart the variables that determine under work (Chatham et al., 2011; Herd et al., 2014; Oberauer et al.,
which conditions temporal factors affect memory for visual and 2013). Extending the notion of interference to capacity limits on
spatial information in WM. attention might lead to an understanding of the mechanisms of
A third phenomenon on which more research is desirable is interference that departs substantially from that in models of
the heterogeneity benefit within content domains (A5 and B6). memory.
Whereas the available evidence consistently shows heterogene- One possible difference between interference in memory and
ity benefits, a systematic exploration of its origins is missing. In interference in attention could be that memory relies on informa-
light of our analysis, which revealed that these findings are tion coded in synaptic connection weights, whereas attention op-
highly diagnostic for adjudicating between the interference erates on information coded by ongoing neural activity. In unitary
hypothesis and its competitors, filling this gap seems important. memory models information is maintained in connection weights.
One open question, for instance, is whether the benefit of In contrast, stimuli currently attended to are coded by patterns of
 WM CAPACITY 791

neural activity, and this is also true for at least some stimuli held separable? Child Development, 77, 1698 –1716. http://dx.doi.org/10
in WM (Emrich, Rigall, Larocque, & Postle, 2013)—although .1111/j.1467-8624.2006.00968.x
apparently only those currently attended to (Lewis-Peacock, Drys- Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set:
dale, Oberauer, & Postle, 2012). If representations of several items Exploring the dynamic control of tasks. In C. Umiltá & M. Moscovitch
are simultaneously represented in WM through persistent neural (Eds.), Attention & performance (Vol. XV, pp. 421– 452). Cambridge,
firing patterns, do they interfere with each other, and can this MA: MIT Press.
Altmann, E. M., & Gray, W. D. (2002). Forgetting to remember: The
interference be characterized by mechanisms analogous to those
functional relationship of decay and interference. Psychological Science,
governing associative memory models?
13, 27–33. http://dx.doi.org/10.1111/1467-9280.00405
The second promising approach for explaining WM capacity is Alvarez, G. A., & Cavanagh, P. (2004). The capacity of visual short-term
a combination of interference with a resource limit. The challenge memory is set both by visual information load and by number of objects.
for this approach is to integrate these two hypotheses into a precise Psychological Science, 15, 106 –111. http://dx.doi.org/10.1111/j.0963-
mechanistic model. One path toward an integration would to be 7214.2004.01502006.x
start from an interference model and add a resource-limited central Anderson, J. R., Reder, L. M., & Lebiere, C. (1996). Working memory:
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

component to it that maintains one, or a small number, of repre- Activation limitations on retrieval. Cognitive Psychology, 30, 221–256.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

sentations (e.g., the free-recall model of Davelaar et al., 2005). http://dx.doi.org/10.1006/cogp.1996.0007

Such a model will have to specify how the capacity-limited com- Awh, E., Barton, B., & Vogel, E. K. (2007). Visual working memory
ponent cooperates with the interference-limited component in gen- represents a fixed number of items regardless of complexity. Psycho-
erating behavior on various paradigms for studying WM. Deter- logical Science, 18, 622– 628. http://dx.doi.org/10.1111/j.1467-9280
mining how the assumed mechanisms of two components operate .2007.01949.x
together engenders a level of complexity that is best handled by Awh, E., Jonides, J., & Reuter-Lorenz, P. A. (1998). Rehearsal in spatial
working memory. Journal of Experimental Psychology: Human Percep-
computational modeling.
tion and Performance, 24, 780 –790. http://dx.doi.org/10.1037/0096-
This challenge illustrates a general point (cf. Farrell & Le-
1523.24.3.780
wandowsky, 2010; Hintzman, 1991): A computational imple-
Baddeley, A. D. (1966). Short-term memory for word sequences as a
mentation of one’s assumptions about how the WM system function of acoustic, semantic and formal similarity. The Quarterly
works—as a set of equations or a simulation program— helps to Journal of Experimental Psychology, 18, 362–365. http://dx.doi.org/10
uncover inconsistencies of assumptions and unanticipated be- .1080/14640746608400055
havior resulting from the interaction of several mechanisms. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press.
Most important, computational modeling enables researchers to Baddeley, A. D. (2001). Is working memory still working? American
unambiguously determine the predictions that follow from a Psychologist, 56, 851– 864. http://dx.doi.org/10.1037/0003-066X.56.11
hypothesis—for instance, about the cause of the WM capacity .851
limit—in conjunction with a set of additional assumptions. To Baddeley, A. (2012). Working memory: Theories, models, and controver-
the extent that future theorizing about WM and its capacity sies. Annual Review of Psychology, 63, 1–29. http://dx.doi.org/10.1146/
limit is based on computational modeling, a future review will annurev-psych-120710-100422
be better placed to systematically map out which findings are Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower
core predictions of which models, and what the shortcomings of (Ed.), Recent advances in learning and motivation (Vol. VIII, pp.
47–90). New York, NY: Academic Press.
particular models tell us about the processes that underlie
Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and
benchmark phenomena.
the structure of short term memory. Journal of Verbal Learning and
Verbal Behavior, 14, 575–589. http://dx.doi.org/10.1016/S0022-
Conclusion 5371(75)80045-4
Bancroft, T. D., Servos, P., & Hockley, W. E. (2011). Mechanisms of
To conclude, we argue that two theoretical approaches hold interference in vibrotactile working memory. PLoS ONE, 6. http://dx
the best promise for an adequate explanation of the WM ca- .doi.org/10.1371/journal.pone.0022518
pacity limit. One is an explanation based only on interference. Barrouillet, P., Bernardin, S., & Camos, V. (2004). Time constraints and
Researchers following this route should make it a priority to resource sharing in adults’ working memory spans. Journal of Experi-
develop a detailed explanation of interference between very mental Psychology: General, 133, 83–100. http://dx.doi.org/10.1037/
different contents in WM. The other approach is to combine the 0096-3445.133.1.83
Barrouillet, P., Bernardin, S., Portrat, S., Vergauwe, E., & Camos, V.
interference hypothesis with a domain-general core mechanism
(2007). Time and cognitive load in working memory. Journal of Exper-
of very limited scope. Work along this line needs to flesh out in
imental Psychology: Learning, Memory, and Cognition, 33, 570 –585.
more detail how the resource limit is to be combined with the http://dx.doi.org/10.1037/0278-7393.33.3.570
mechanisms of interference. Barrouillet, P., Portrat, S., Vergauwe, E., Diependaele, K., & Camos, V.
(2011). Further evidence for temporal decay in working memory: Reply
to Lewandowsky and Oberauer (2009). Journal of Experimental Psy-
References chology: Learning, Memory, and Cognition, 37, 1302–1317. http://dx
Allen, R. J., Baddeley, A. D., & Hitch, G. J. (2006). Is the binding of visual .doi.org/10.1037/a0022933
features in working memory resource-demanding? Journal of Experi- Bayliss, D. M., Jarrold, C., Gunn, D. M., & Baddeley, A. D. (2003). The
mental Psychology: General, 135, 298 –313. http://dx.doi.org/10.1037/ complexities of complex span: Explaining individual differences in
0096-3445.135.2.298 working memory in children and adults. Journal of Experimental Psy-
Alloway, T. P., Gathercole, S. E., & Pickering, S. J. (2006). Verbal and chology: General, 132, 71–92. http://dx.doi.org/10.1037/0096-3445.132
visuospatial short-term and working memory in children: Are they .1.71
 792 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

Bays, P. M. (2014). Noise in neural populations accounts for errors in Chen, Z., & Cowan, N. (2005). Chunk limits and length limits in imme-
working memory. The Journal of Neuroscience, 34, 3632–3645. http:// diate recall: A reconciliation. Journal of Experimental Psychology:
dx.doi.org/10.1523/JNEUROSCI.3204-13.2014 Learning, Memory, and Cognition, 31, 1235–1249. http://dx.doi.org/10
Belopolsky, A. V., & Theeuwes, J. (2009). No functional role of attention- .1037/0278-7393.31.6.1235
based rehearsal in maintenance of spatial working memory representa- Chen, Z., & Cowan, N. (2009). Core verbal working-memory capacity:
tions. Acta Psychologica, 132, 124 –135. http://dx.doi.org/10.1016/j The limit in words retained without covert articulation. The Quarterly
.actpsy.2009.01.002 Journal of Experimental Psychology: Human Experimental Psychology,
Brady, T. F., & Alvarez, G. A. (2015). No evidence for a fixed object limit 62, 1420 –1429. http://dx.doi.org/10.1080/17470210802453977
in working memory: Spatial ensemble representations inflate estimates Christie, D. F. M., & Phillips, W. A. (1979). Simple drawing and pattern
of working memory capacity for complex objects. Journal of Experi- completion techniques for studying visualization and long-term visual
mental Psychology: Learning, Memory, and Cognition, 41, 921–929. knowledge. Memory & Cognition, 7, 360 –367. http://dx.doi.org/10
http://dx.doi.org/10.1037/xlm0000075 .3758/BF03196940
Brown, G. D. A., Neath, I., & Chater, N. (2007). A temporal ratio model Chuderski, A. (2014). The relational integration task explains fluid reason-
of memory. Psychological Review, 114, 539 –576. http://dx.doi.org/10 ing above and beyond other working memory tasks. Memory & Cogni-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

.1037/0033-295X.114.3.539 tion, 42, 448 – 463. http://dx.doi.org/10.3758/s13421-013-0366-x

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Brown, G. D. A., Preece, T., & Hulme, C. (2000). Oscillator-based mem- Chuderski, A., Taraday, M., Ne˛cka, E., & Smoleń, T. (2012). Storage
ory for serial order. Psychological Review, 107, 127–181. http://dx.doi capacity explains fluid intelligence but executive control does not.
.org/10.1037/0033-295X.107.1.127 Intelligence, 40, 278 –295. http://dx.doi.org/10.1016/j.intell.2012.02.010
Brown, J. (1958). Some tests of the decay theory of immediate memory. Cocchini, G., Logie, R. H., Della Sala, S., MacPherson, S. E., & Baddeley,
The Quarterly Journal of Experimental Psychology, 10, 12–21. http:// A. D. (2002). Concurrent performance of two memory tasks: Evidence
dx.doi.org/10.1080/17470215808416249 for domain-specific working memory systems. Memory & Cognition,
Buehner, M., Krumm, S., & Pick, M. (2005). Reasoning ⫽ working 30, 1086 –1095. http://dx.doi.org/10.3758/BF03194326
memory ⫽ attention. Intelligence, 33, 251–272. http://dx.doi.org/10 Conlin, J. A., & Gathercole, S. E. (2006). Lexicality and interference in
.1016/j.intell.2005.01.002 working memory in children and in adults. Journal of Memory and
Buehner, M., Krumm, S., Ziegler, M., & Pluecken, T. (2006). Cognitive Language, 55, 363–380. http://dx.doi.org/10.1016/j.jml.2006.05.003
abilities and their interplay. Reasoning, crystallized intelligence, work- Conlin, J. A., Gathercole, S. E., & Adams, J. W. (2005). Stimulus simi-
ing memory components, and sustained attention. Journal of Individual larity decrements in children’s working memory span. The Quarterly
Differences, 27, 57–72. http://dx.doi.org/10.1027/1614-0001.27.2.57 Journal of Experimental Psychology, 58, 1434 –1446. http://dx.doi.org/
Burgess, N., & Hitch, G. J. (1999). Memory for serial order: A network 10.1080/02724980443000683
model of the phonological loop and its timing. Psychological Review, Conrad, R. (1964). Acoustic confusions in immediate memory. British
106, 551–581. http://dx.doi.org/10.1037/0033-295X.106.3.551 Journal of Psychology, 55, 75– 84. http://dx.doi.org/10.1111/j.2044-
Burgess, N., & Hitch, G. J. (2006). A revised model of short-term memory 8295.1964.tb00899.x
and long-term learning of verbal sequences. Journal of Memory and Conrad, R., & Hull, A. J. (1964). Information, acoustic confusion and
Language, 55, 627– 652. http://dx.doi.org/10.1016/j.jml.2006.08.005 memory span. British Journal of Psychology, 55, 429 – 432. http://dx.doi
Camos, V., Lagner, P., & Barrouillet, P. (2009). Two maintenance mech- .org/10.1111/j.2044-8295.1964.tb00928.x
anisms of verbal information in working memory. Journal of Memory Conrad, R., & Hull, A. J. (1966). The role of the interpolated task in
and Language, 61, 457– 469. http://dx.doi.org/10.1016/j.jml.2009.06 short-term retention. The Quarterly Journal of Experimental Psychol-
.002 ogy, 18, 266 –269. http://dx.doi.org/10.1080/14640746608400039
Camos, V., & Portrat, S. (2015). The impact of cognitive load on delayed Conway, A. R. A., Jarrold, C., Kane, M. J., Miyake, A., & Towse, J. N.
recall. Psychonomic Bulletin & Review, 22, 1029 –1034. http://dx.doi (Eds.). (2007). Variation in working memory. New York, NY: Oxford
.org/10.3758/s13423-014-0772-5 University Press.
Cansino, S., Guzzon, D., Martinelli, M., Barollo, M., & Casco, C. (2011). Conway, A. R. A., Kane, M. J., & Engle, R. W. (2003). Working memory
Effects of aging on interference control in selective attention and work- capacity and its relation to general intelligence. Trends in Cognitive
ing memory. Memory & Cognition, 39, 1409 –1422. http://dx.doi.org/10 Sciences, 7, 547–552. http://dx.doi.org/10.1016/j.tics.2003.10.005
.3758/s13421-011-0109-9 Cowan, N. (2001). The magical number 4 in short-term memory: A
Carrier, L. M., & Pashler, H. (1995). Attentional limits in memory re- reconsideration of mental storage capacity. Behavioral and Brain Sci-
trieval. Journal of Experimental Psychology: Learning, Memory, and ences, 24, 87–114. http://dx.doi.org/10.1017/S0140525X01003922
Cognition, 21, 1339 –1348. http://dx.doi.org/10.1037/0278-7393.21.5 Cowan, N. (2005). Working memory capacity. New York, NY: Psychology
.1339 Press. http://dx.doi.org/10.4324/9780203342398
Case, R., Kurland, M., & Goldberg, J. (1982). Operational efficiency and Cowan, N., Blume, C. L., & Saults, J. S. (2013). Attention to attributes and
the growth of short-term memory span. Journal of Experimental Child objects in working memory. Journal of Experimental Psychology:
Psychology, 33, 386 – 404. http://dx.doi.org/10.1016/0022- Learning, Memory, and Cognition, 39, 731–747. http://dx.doi.org/10
0965(82)90054-6 .1037/a0029687
Chatham, C. H., Herd, S. A., Brant, A. M., Hazy, T. E., Miyake, A., Cowan, N., Day, L., Saults, J. S., Keller, T. A., Johnson, T. D., & Flores,
O’Reilly, R., & Friedman, N. P. (2011). From an executive network to L. (1992). The role of verbal output time in the effects of word length on
executive control: A computational model of the n-back task. Journal of immediate memory. Journal of Memory and Language, 31, 1–17. http://
Cognitive Neuroscience, 23, 3598 –3619. http://dx.doi.org/10.1162/ dx.doi.org/10.1016/0749-596X(92)90002-F
jocn_a_00047 Cowan, N., Elliott, E. M., Saults, J. S., Nugent, L. D., Bomb, P., &
Chechile, R. A. (1987). Trace susceptibility theory. Journal of Experimen- Hismjatullina, A. (2006). Rethinking speed theories of cognitive devel-
tal Psychology: General, 116, 203–222. http://dx.doi.org/10.1037/0096- opment. Increasing the rate of recall without affecting accuracy. Psy-
3445.116.3.203 chological Science, 17, 67–73. http://dx.doi.org/10.1111/j.1467-9280
Chein, J. M., Moore, A. B., & Conway, A. R. A. (2011). Domain-general .2005.01666.x
mechanisms of complex working memory span. NeuroImage, 54, 550 – Cowan, N., Morey, C. C., AuBuchon, A. M., Zwilling, C. E., & Gilchrist,
559. http://dx.doi.org/10.1016/j.neuroimage.2010.07.067 A. L. (2010). Seven-year-olds allocate attention like adults unless work-
 WM CAPACITY 793

ing memory is overloaded. Developmental Science, 13, 120 –133. http:// Farrell, S. (2012). Temporal clustering and sequencing in short-term mem-
dx.doi.org/10.1111/j.1467-7687.2009.00864.x ory and episodic memory. Psychological Review, 119, 223–271. http://
Cowan, N., Rouder, J. N., Blume, C. L., & Saults, J. S. (2012). Models of dx.doi.org/10.1037/a0027371
verbal working memory capacity: What does it take to make them work? Farrell, S., & Lewandowsky, S. (2002). An endogenous distributed model
Psychological Review, 119, 480 – 499. http://dx.doi.org/10.1037/ of ordering in serial recall. Psychonomic Bulletin & Review, 9, 59 –79.
a0027791 http://dx.doi.org/10.3758/BF03196257
Cowan, N., Saults, J. S., & Blume, C. L. (2014). Central and peripheral Farrell, S., & Lewandowsky, S. (2010). Computational models as aids to
components of working memory storage. Journal of Experimental Psy- better reasoning in psychology. Current Directions in Psychological
chology: General, 143, 1806 –1836. http://dx.doi.org/10.1037/a0036814 Science, 19, 329 –335. http://dx.doi.org/10.1177/0963721410386677
Cowan, N., Wood, N. L., Wood, P. K., Keller, T. A., Nugent, L. D., & Farrell, S., & Lewandowsky, S. (2012). Response suppression contributes
Keller, C. V. (1998). Two separate verbal processing rates contributing to recency in serial recall. Memory & Cognition, 40, 1070 –1080. http://
to short-term memory span. Journal of Experimental Psychology: Gen- dx.doi.org/10.3758/s13421-012-0212-6
Farrell, S., & Oberauer, K. (2014). Working memory for cross-domain
eral, 127, 141–160. http://dx.doi.org/10.1037/0096-3445.127.2.141
sequences. The Quarterly Journal of Experimental Psychology, 67,
Dambacher, M., & Hübner, R. (2015). Time pressure affects the efficiency
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

33– 44. http://dx.doi.org/10.1080/17470218.2013.795177

of perceptual processing in decisions under conflict. Psychological Re-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Ferguson, A. N., & Bowey, J. A. (2005). Global processing speed as a

search, 79, 83–94.
mediator of developmental changes in children’s auditory memory span.
Daneman, M., & Carpenter, P. A. (1980). Individual differences in working
Journal of Experimental Child Psychology, 91, 89 –112. http://dx.doi
memory and reading. Journal of Verbal Learning and Verbal Behavior,
.org/10.1016/j.jecp.2004.12.006
19, 450 – 466. http://dx.doi.org/10.1016/S0022-5371(80)90312-6
Fougnie, D., & Marois, R. (2011). What limits working memory capacity?
Darley, C. F., & Glass, A. L. (1975). Effects of rehearsal and serial list Evidence for modality-specific sources to the simultaneous storage of
position on recall. Journal of Experimental Psychology: Human Learn- visual and auditory arrays. Journal of Experimental Psychology: Learn-
ing and Memory, 1, 453– 458. http://dx.doi.org/10.1037/0278-7393.1.4 ing, Memory, and Cognition, 37, 1329 –1341. http://dx.doi.org/10.1037/
.453 a0024834
Davelaar, E. J., Goshen-Gottstein, Y., Ashkenazi, A., Haarmann, H. J., & Fougnie, D., Zughni, S., Godwin, D., & Marois, R. (2015). Working
Usher, M. (2005). The demise of short-term memory revisited: Empir- memory storage is intrinsically domain specific. Journal of Experimen-
ical and computational investigations of recency effects. Psychological tal Psychology: General, 144, 30 – 47. http://dx.doi.org/10.1037/
Review, 112, 3– 42. http://dx.doi.org/10.1037/0033-295X.112.1.3 a0038211
Davis, L. C., Rane, S., & Hiscock, M. (2013). Serial recall of visuospatial Fukuda, K., Awh, E., & Vogel, E. K. (2010). Discrete capacity limits in
and verbal information with and without material-specific interference: visual working memory. Current Opinion in Neurobiology, 20, 177–
Implications for contemporary models of working memory. Memory, 21, 182. http://dx.doi.org/10.1016/j.conb.2010.03.005
778 –797. http://dx.doi.org/10.1080/09658211.2012.756037 Gaillard, V., Barrouillet, P., Jarrold, C., & Camos, V. (2011). Develop-
Delvenne, J.-B., & Bruyer, R. (2004). Does visual short-term memory store mental differences in working memory: Where do they come from?
bound features? Visual Cognition, 11, 1–27. http://dx.doi.org/10.1080/ Journal of Experimental Child Psychology, 110, 469 – 479. http://dx.doi
13506280344000167 .org/10.1016/j.jecp.2011.05.004
Depoorter, A., & Vandierendonck, A. (2009). Evidence for modality- Gavens, N., & Barrouillet, P. (2004). Delays of retention, processing
independent order coding in working memory. The Quarterly Journal of efficiency, and attentional resources in working memory span develop-
Experimental Psychology: Human Experimental Psychology, 62, 531– ment. Journal of Memory and Language, 51, 644 – 657. http://dx.doi.org/
549. http://dx.doi.org/10.1080/17470210801995002 10.1016/j.jml.2004.06.009
Dosher, B. A., & Ma, J. J. (1998). Output loss or rehearsal loop? Output- Gold, J. M., Murray, R. F., Sekuler, A. B., Bennett, P. J., & Sekuler, R.
time versus pronunciation-time limits in immediate recall for forgetting- (2005). Visual memory decay is deterministic. Psychological Science,
matched materials. Journal of Experimental Psychology: Learning, 16, 769 –774. http://dx.doi.org/10.1111/j.1467-9280.2005.01612.x
Memory, and Cognition, 24, 316 –335. http://dx.doi.org/10.1037/0278- Gorgoraptis, N., Catalao, R. F. G., Bays, P. M., & Husain, M. (2011).
7393.24.2.316 Dynamic updating of working memory resources for visual objects. The
Journal of Neuroscience, 31, 8502– 8511. http://dx.doi.org/10.1523/
Dubé, C., Zhou, F., Kahana, M. J., & Sekuler, R. (2014). Similarity-based
JNEUROSCI.0208-11.2011
distortion of visual short-term memory is due to perceptual averaging.
Greene, R. L. (1987). Effects of maintenance rehearsal on human memory.
Vision Research, 96, 8 –16. http://dx.doi.org/10.1016/j.visres.2013.12
Psychological Bulletin, 102, 403– 413. http://dx.doi.org/10.1037/0033-
.016
2909.102.3.403
Ecker, U. K. H., Lewandowsky, S., & Oberauer, K. (2014). Removal of
Hale, S., Myerson, J., Rhee, S. H., Weiss, C. S., & Abrams, R. A. (1996).
information from working memory: A specific updating process. Jour-
Selective interference with the maintenance of location information in
nal of Memory and Language, 74, 77–90. http://dx.doi.org/10.1016/j.jml
working memory. Neuropsychology, 10, 228 –240. http://dx.doi.org/10
.2013.09.003 .1037/0894-4105.10.2.228
Ecker, U. K. H., Oberauer, K., & Lewandowsky, S. (2014). Working Halford, G. S., Cowan, N., & Andrews, G. (2007). Separating cognitive
memory updating involves item-specific removal. Journal of Memory capacity from knowledge: A new hypothesis. Trends in Cognitive Sci-
and Language, 74, 1–15. ences, 11, 236 –242. http://dx.doi.org/10.1016/j.tics.2007.04.001
Emrich, S. M., Riggall, A. C., Larocque, J. J., & Postle, B. R. (2013). Hardman, K. O., & Cowan, N. (2015). Remembering complex objects in
Distributed patterns of activity in sensory cortex reflect the precision of visual working memory: Do capacity limits restrict objects or features?
multiple items maintained in visual short-term memory. The Journal of Journal of Experimental Psychology: Learning, Memory, and Cogni-
Neuroscience, 33, 6516 – 6523. http://dx.doi.org/10.1523/JNEUROSCI tion, 41, 325–347. http://dx.doi.org/10.1037/xlm0000031
.5732-12.2013 Heitz, R. P., & Engle, R. W. (2007). Focusing the spotlight: Individual
Farrell, S. (2006). Mixed-list phonological similarity effects in delayed differences in visual attention control. Journal of Experimental Psychol-
serial recall. Journal of Memory and Language, 55, 587– 600. http://dx ogy: General, 136, 217–240. http://dx.doi.org/10.1037/0096-3445.136.2
.doi.org/10.1016/j.jml.2006.06.002 .217
 794 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

Herd, S. A., O’Reilly, R. C., Hazy, T. E., Chatham, C. H., Brant, A. M., & Jones, D. M., & Macken, W. J. (1993). Irrelevant tones produce an
Friedman, N. P. (2014). A neural network model of individual differ- irrelevant speech effect: Implications for phonological coding in work-
ences in task switching abilities. Neuropsychologia, 62, 375–389. http:// ing memory. Journal of Experimental Psychology: Learning, Memory,
dx.doi.org/10.1016/j.neuropsychologia.2014.04.014 and Cognition, 19, 369 –381. http://dx.doi.org/10.1037/0278-7393.19.2
Hintzman, D. L. (1991). Why are formal models useful in psychology? In .369
W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Jonides, J., Lewis, R. L., Nee, D. E., Lustig, C. A., Berman, M. G., &
Essays on human memory in honor of Bennet B. Murdock (pp. 39 –56). Moore, K. S. (2008). The mind and brain of short-term memory. Annual
Hillsdale, NJ: Erlbaum. Review of Psychology, 59, 193–224. http://dx.doi.org/10.1146/annurev
Hole, G. J. (1996). Decay and interference effects in visuospatial short- .psych.59.103006.093615
term memory. Perception, 25, 53– 64. http://dx.doi.org/10.1068/ Jost, K., Bryck, R. L., Vogel, E. K., & Mayr, U. (2011). Are old adults just
p250053 like low working memory young adults? Filtering efficiency and age
Horoufchin, H., Philipp, A. M., & Koch, I. (2011). The dissipating task- differences in visual working memory. Cerebral Cortex, 21, 1147–1154.
repetition benefit in cued task switching: Task-set decay or temporal http://dx.doi.org/10.1093/cercor/bhq185
distinctiveness? Journal of Experimental Psychology: Human Percep- Jünger, E., Kliegl, R., & Oberauer, K. (2014). No evidence for feature
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

tion and Performance, 37, 455– 472. http://dx.doi.org/10.1037/a0020557 overwriting in visual working memory. Memory, 22, 374 –389. http://
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Howard, M. W., & Kahana, M. J. (2002). A distributed representation of dx.doi.org/10.1080/09658211.2013.794243

temporal context. Journal of Mathematical Psychology, 46, 269 –299. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehen-
http://dx.doi.org/10.1006/jmps.2001.1388 sion: Individual differences in working memory. Psychological Review,
Huang, J., & Sekuler, R. (2010). Distortions in recall from visual memory: 99, 122–149. http://dx.doi.org/10.1037/0033-295X.99.1.122
Two classes of attractors at work. Journal of Vision, 10, 1–27. http://dx Kahana, M. J., & Sekuler, R. (2002). Recognizing spatial patterns: A noisy
.doi.org/10.1167/10.2.24 exemplar approach. Vision Research, 42, 2177–2192. http://dx.doi.org/
Hulme, C., Roodenrys, S., Brown, G., & Mercer, R. (1995). The role of 10.1016/S0042-6989(02)00118-9
long-term memory mechanisms in memory span. British Journal of Kahana, M. J., Zhou, F., Geller, A. S., & Sekuler, R. (2007). Lure
Psychology, 86, 527–536. http://dx.doi.org/10.1111/j.2044-8295.1995 similarity affects visual episodic recognition: Detailed tests of a noisy
.tb02570.x exemplar model. Memory & Cognition, 35, 1222–1232. http://dx.doi
Humphreys, M. S., Maguire, A. M., McFarlane, K. A., Burt, J. S., Bolland, .org/10.3758/BF03193596
S. W., Murray, K. L., & Dunn, R. (2010). Using maintenance rehearsal Kail, R. (1992). Processing speed, speech rate, and memory. Developmen-
to explore recognition memory. Journal of Experimental Psychology: tal Psychology, 28, 899 –904. http://dx.doi.org/10.1037/0012-1649.28.5
Learning, Memory, and Cognition, 36, 147–159. http://dx.doi.org/10 .899
.1037/a0017687 Kail, R. (1997). Phonological skill and articulation time independently
Hurlstone, M. J., Hitch, G. J., & Baddeley, A. D. (2014). Memory for serial contribute to the development of memory span. Journal of Experimental
order across domains: An overview of the literature and directions for Child Psychology, 67, 57– 68. http://dx.doi.org/10.1006/jecp.1997.2393
future research. Psychological Bulletin, 140, 339 –373. Kane, M. J., Conway, A. R. A., Hambrick, D. Z., & Engle, R. W. (2007).
Jalbert, A., Neath, I., Bireta, T. J., & Surprenant, A. M. (2011). When does Variation in working memory capacity as variation in executive atten-
length cause the word length effect? Journal of Experimental Psychol- tion and control. In A. R. A. Conway, C. Jarrold, M. J. Kane, A. Miyake,
ogy: Learning, Memory, and Cognition, 37, 338 –353. http://dx.doi.org/ & J. N. Towse (Eds.), Variation in working memory (pp. 21– 48). New
10.1037/a0021804 York, NY: Oxford University Press.
Jalbert, A., Neath, I., & Surprenant, A. M. (2011). Does length or neigh- Kane, M. J., & Engle, R. W. (2000). Working-memory capacity, proactive
borhood size cause the word length effect? Memory & Cognition, 39, interference, and divided attention: Limits on long-term memory re-
1198 –1210. http://dx.doi.org/10.3758/s13421-011-0094-z trieval. Journal of Experimental Psychology: Learning, Memory, and
Jarrold, C., Tam, H., Baddeley, A. D., & Harvey, C. E. (2010). The nature Cognition, 26, 336 –358. http://dx.doi.org/10.1037/0278-7393.26.2.336
and position of processing determines why forgetting occurs in working Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the
memory tasks. Psychonomic Bulletin & Review, 17, 772–777. http://dx control of attention: The contributions of goal neglect, response com-
.doi.org/10.3758/PBR.17.6.772 petition, and task set to Stroop interference. Journal of Experimental
Jarrold, C., Tam, H., Baddeley, A. D., & Harvey, C. E. (2011). How does Psychology: General, 132, 47–70. http://dx.doi.org/10.1037/0096-3445
processing affect storage in working memory tasks? Evidence for both .132.1.47
domain-general and domain-specific effects. Journal of Experimental Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W.,
Psychology: Learning, Memory, and Cognition, 37, 688 –705. http://dx & Engle, R. W. (2004). The generality of working memory capacity: A
.doi.org/10.1037/a0022527 latent-variable approach to verbal and visuospatial memory span and
Jarrold, C., & Towse, J. N. (2006). Individual differences in working reasoning. Journal of Experimental Psychology: General, 133, 189 –
memory. Neuroscience, 139, 39 –50. http://dx.doi.org/10.1016/j 217. http://dx.doi.org/10.1037/0096-3445.133.2.189
.neuroscience.2005.07.002 Kane, M. J., Poole, B. J., Tuholski, S. W., & Engle, R. W. (2006). Working
Jensen, A. R. (1988). Speed of information processing and population memory capacity and the top-down control of visual search: Exploring
differences. In S. H. Irvine & J. W. Berry (Eds.), Human abilities in the boundaries of “executive attention”. Journal of Experimental Psy-
cultural context (pp. 105–145). Cambridge, United Kingdom: Cam- chology: Learning, Memory, and Cognition, 32, 749 –777. http://dx.doi
bridge University Press. .org/10.1037/0278-7393.32.4.749
Jolicœur, P., & Dell’Acqua, R. (1998). The demonstration of short-term Keye, D., Wilhelm, O., Oberauer, K., & van Ravenzwaaij, D. (2009).
consolidation. Cognitive Psychology, 36, 138 –202. http://dx.doi.org/10 Individual differences in conflict-monitoring: Testing means and cova-
.1006/cogp.1998.0684 riance hypothesis about the Simon and the Eriksen Flanker task. Psy-
Jones, D., Farrand, P., Stuart, G., & Morris, N. (1995). Functional equiv- chological Research, 73, 762–776. http://dx.doi.org/10.1007/s00426-
alence of verbal and spatial information in serial short-term memory. 008-0188-9
Journal of Experimental Psychology: Learning, Memory, and Cogni- Klauer, K. C., & Stegmaier, R. (1997). Interference in immediate spatial
tion, 21, 1008 –1018. http://dx.doi.org/10.1037/0278-7393.21.4.1008 memory: Shifts of spatial attention or central-executive involvement?
 WM CAPACITY 795

The Quarterly Journal of Experimental Psychology, 50, 79 –99. http:// & Cognition, 42, 1118 –1129. http://dx.doi.org/10.3758/s13421-014-
dx.doi.org/10.1080/027249897392233 0421-2
Koch, I., Gade, M., Schuch, S., & Philipp, A. M. (2010). The role of Loaiza, V. M., & McCabe, D. P. (2013). The influence of aging on
inhibition in task switching: A review. Psychonomic Bulletin & Review, attentional refreshing and articulatory rehearsal during working memory
17, 1–14. http://dx.doi.org/10.3758/PBR.17.1.1 on later episodic memory performance. Aging, Neuropsychology, and
Kriegeskorte, N., & Kievit, R. A. (2013). Representational geometry: Cognition: A Journal on Normal and Dysfunctional Development, 20,
Integrating cognition, computation, and the brain. Trends in Cognitive 471– 493.
Sciences, 17, 401– 412. http://dx.doi.org/10.1016/j.tics.2013.06.007 Loaiza, V. M., & McCabe, D. P. (2012b). Temporal-contextual processing
Kriegeskorte, N., Mur, M., & Bandettini, P. (2008). Representational in working memory: Evidence from delayed cued recall and delayed free
similarity analysis - connecting the branches of systems neuroscience. recall tests. Memory & Cognition, 40, 191–203. http://dx.doi.org/10
Frontiers in Systems Neuroscience, 2, 4. http://dx.doi.org/10.3389/neuro .3758/s13421-011-0148-2
.06.004.2008 Logie, R. H. (1995). Visuo-spatial working memory. Hove: Erlbaum.
Lange, E. B., Cerella, J., & Verhaeghen, P. (2011). Ease of access to list Logie, R. H. (2011). The functional organization and capacity limits of
items in short-term memory depends on the order of the recognition working memory. Current Directions in Psychological Science, 20,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

probes. Journal of Experimental Psychology: Learning, Memory, and 240 –245. http://dx.doi.org/10.1177/0963721411415340
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Cognition, 37, 608 – 620. http://dx.doi.org/10.1037/a0022220 Longoni, A. M., Richardson, J. T. E., & Aiello, A. (1993). Articulatory
Lange, E. B., & Oberauer, K. (2005). Overwriting of phonemic features in rehearsal and phonological storage in working memory. Memory &
serial recall. Memory, 13, 333–339. http://dx.doi.org/10.1080/ Cognition, 21, 11–22. http://dx.doi.org/10.3758/BF03211160
09658210344000378 Lovatt, P., Avons, S. E., & Masterson, J. (2000). The word-length effect
Lee, J. J., & Chabris, C. F. (2013). General cognitive ability and the and disyllabic words. The Quarterly Journal of Experimental Psychol-
psychological refractory period: Individual differences in the mind’s ogy, 53, 1–22. http://dx.doi.org/10.1080/713755877
bottleneck. Psychological Science, 24, 1226 –1233. http://dx.doi.org/10 Lovatt, P., Avons, S. E., & Masterson, J. (2002). Output decay in imme-
.1177/0956797612471540 diate serial recall: Speech time revisited. Journal of Memory and Lan-
Lepsien, J., & Nobre, A. C. (2006). Cognitive control of attention in the guage, 46, 227–243. http://dx.doi.org/10.1006/jmla.2001.2806
human brain: Insights from orienting attention to mental representations. Lovett, M. C., Reder, L. M., & Lebiere, C. (1999). Modeling working
Brain Research, 1105, 20 –31. http://dx.doi.org/10.1016/j.brainres.2006 memory in a unified architecture: An ACT-R perspective. In A. Miyake
.03.033 & P. Shah (Eds.), Models of working memory. Mechanisms of active
Lewandowsky, S. (1999). Redintegration and response suppression in maintenance and executive control (pp. 135–182). Cambridge: Cam-
serial recall: A dynamic network model. International Journal of Psy- bridge University Press. http://dx.doi.org/10.1017/CBO9781139174909
chology (Special Issue: Short-Term/Working Memory), 34, 434 – 446. .008
Lewandowsky, S., Duncan, M., & Brown, G. D. A. (2004). Time does not Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory
cause forgetting in short-term serial recall. Psychonomic Bulletin & for features and conjunctions. Nature, 390, 279 –281. http://dx.doi.org/
Review, 11, 771–790. http://dx.doi.org/10.3758/BF03196705 10.1038/36846
Lewandowsky, S., & Farrell, S. (2008a). Phonological similarity in serial Luck, S. J., & Vogel, E. K. (2013). Visual working memory capacity: From
recall: Constraints on theories of memory. Journal of Memory and psychophysics and neurobiology to individual differences. Trends in
Language, 58, 429 – 448. http://dx.doi.org/10.1016/j.jml.2007.01.005 Cognitive Sciences, 17, 391– 400. http://dx.doi.org/10.1016/j.tics.2013
Lewandowsky, S., & Farrell, S. (2008b). Short-term memory: New data .06.006
and a model. In B. H. Ross (Ed.), The psychology of learning and Ma, W. J., Husain, M., & Bays, P. M. (2014). Changing concepts of
motivation (Vol. 49, pp. 1– 48). London, United Kingdom: Elsevier. working memory. Nature Neuroscience, 17, 347–356. http://dx.doi.org/
http://dx.doi.org/10.1016/S0079-7421(08)00001-7 10.1038/nn.3655
Lewandowsky, S., & Farrell, S. (2011). Computational modeling in cog- Macken, W. J., & Jones, D. M. (1995). Functional characteristics of the
nition: Principles and practice. Thousand Oaks, CA: Sage. http://dx.doi inner voice and the inner ear: Single or double agency? Journal of
.org/10.4135/9781483349428 Experimental Psychology: Learning, Memory, and Cognition, 21, 436 –
Lewandowsky, S., Geiger, S. M., Morrell, D. B., & Oberauer, K. (2010). 448. http://dx.doi.org/10.1037/0278-7393.21.2.436
Turning simple span into complex span: Time for decay or interference Magnussen, S., & Greenlee, M. W. (1999). The psychophysics of percep-
from distractors? Journal of Experimental Psychology: Learning, Mem- tual memory. Psychological Research, 62, 81–92. http://dx.doi.org/10
ory, and Cognition, 36, 958 –978. http://dx.doi.org/10.1037/a0019764 .1007/s004260050043
Lewandowsky, S., Geiger, S. M., & Oberauer, K. (2008). Interference- Makovski, T. (2012). Are multiple visual short-term memory storages
based forgetting in verbal short-term memory. Journal of Memory and necessary to explain the retro-cue effect? Psychonomic Bulletin & Re-
Language, 59, 200 –222. http://dx.doi.org/10.1016/j.jml.2008.04.004 view, 19, 470 – 476. http://dx.doi.org/10.3758/s13423-012-0235-9
Lewandowsky, S., & Oberauer, K. (2008). The word-length effect provides Makovski, T., Shim, W. M., & Jiang, Y. V. (2006). Interference from filled
no evidence for decay in short-term memory. Psychonomic Bulletin & delays on visual change detection. Journal of Vision, 6, 1459 –1470.
Review, 15, 875– 888. http://dx.doi.org/10.3758/PBR.15.5.875 http://dx.doi.org/10.1167/6.12.11
Lewis-Peacock, J. A., Drysdale, A. T., Oberauer, K., & Postle, B. R. Mall, J. T., Morey, C. C., Wolff, M. J., & Lehnert, F. (2014). Visual
(2012). Neural evidence for a distinction between short-term memory selective attention is equally functional for individuals with low and high
and the focus of attention. Journal of Cognitive Neuroscience, 24, working memory capacity: Evidence from accuracy and eye movements.
61–79. http://dx.doi.org/10.1162/jocn_a_00140 Attention, Perception, & Psychophysics, 76, 1998 –2014. http://dx.doi
Li, K. Z. H. (1999). Selection from working memory: On the relationship .org/10.3758/s13414-013-0610-2
between processing and storage components. Aging, Neuropsychology, Matthey, L., Bays, P. M., & Dayan, P. (2015). A probabilistic palimpsest
and Cognition: A Journal on Normal and Dysfunctional Development, 6, model of visual short-term memory. PLoS Computational Biology, 11.
99 –116. http://dx.doi.org/10.1076/anec.6.2.99.784 http://dx.doi.org/10.1371/journal.pcbi.1004003
Lilienthal, L., Hale, S., & Myerson, J. (2014). The effects of environmental Maylor, E. A., Vousden, J. I., & Brown, G. D. A. (1999). Adult age
support and secondary tasks on visuospatial working memory. Memory differences in short-term memory for serial order: Data and a model.
 796 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

Psychology and Aging, 14, 572–594. http://dx.doi.org/10.1037/0882- Navon, D. (1984). Resources - a theoretical soupstone? Psychological
7974.14.4.572 Review, 91, 216 –234. http://dx.doi.org/10.1037/0033-295X.91.2.216
Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action: Navon, D., & Gopher, D. (1979). On the economy of the human-processing
The role of backward inhibition. Journal of Experimental Psychology: system. Psychological Review, 86, 214 –255. http://dx.doi.org/10.1037/
General, 129, 4 –26. http://dx.doi.org/10.1037/0096-3445.129.1.4 0033-295X.86.3.214
McCormack, T., Brown, G. D. A., Vousden, J. I., & Henson, R. N. A. Navon, D., & Miller, J. (2002). Queuing or sharing? A critical evaluation
(2000). Children’s serial recall errors: Implications for theories of short- of the single-bottleneck notion. Cognitive Psychology, 44, 193–251.
term memory development. Journal of Experimental Child Psychology, http://dx.doi.org/10.1006/cogp.2001.0767
76, 222–252. http://dx.doi.org/10.1006/jecp.1999.2550 Newell, A. (1973). You can’t play twenty questions with nature and win.
McElree, B. (2006). Accessing recent events. In B. H. Ross (Ed.), The In W. G. Chase (Ed.), Visual information processing (pp. 283–308).
psychology of learning and motivation (Vol. 46, pp. 155–200). San New York, NY: Academic Press.
Diego, CA: Academic Press. Nieuwenstein, M., & Wyble, B. (2014). Beyond a mask and against the
McFarlane, K. A., & Humphreys, M. S. (2012). Maintenance rehearsal: bottleneck: Retroactive dual-task interference during working memory
The key to the role attention plays in storage and forgetting. Journal of consolidation of a masked visual target. Journal of Experimental Psy-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Experimental Psychology: Learning, Memory, and Cognition, 38, 1001– chology: General, 143, 1409 –1427. http://dx.doi.org/10.1037/a0035257
This document is copyrighted by the American Psychological Association or one of its allied publishers.

1018. http://dx.doi.org/10.1037/a0026783 Nipher, F. E. (1878). On the distribution of errors in numbers written from
McVay, J. C., & Kane, M. J. (2009). Conducting the train of thought: memory. Transactions of the Academy of Science of St. Louis, 3, CCX-
Working memory capacity, goal neglect, and mind wandering in an CCXI.
executive-control task. Journal of Experimental Psychology: Learning, Norman, D. A., & Bobrow, D. G. (1975). On data-limited and resource-
Memory, and Cognition, 35, 196 –204. http://dx.doi.org/10.1037/ limited processes. Cognitive Psychology, 7, 44 – 64. http://dx.doi.org/10
a0014104 .1016/0010-0285(75)90004-3
McVay, J. C., & Kane, M. J. (2012). Why does working memory capacity Nuerk, H. C., Wood, G., & Willmes, K. (2005). The universal SNARC
predict variation in reading comprehension? On the influence of mind effect: The association between number magnitude and space is amodal.
wandering and executive attention. Journal of Experimental Psychology: Experimental Psychology, 52, 187–194. http://dx.doi.org/10.1027/1618-
3169.52.3.187
General, 141, 302–320. http://dx.doi.org/10.1037/a0025250
Oberauer, K. (2009). Design for a working memory. Psychology of Learn-
Meier, M. E., & Kane, M. J. (2013). Working memory capacity and Stroop
ing and Motivation: Advances in Research and Theory, 51, 45–100.
interference: Global versus local indices of executive control. Journal of
http://dx.doi.org/10.1016/S0079-7421(09)51002-X
Experimental Psychology: Learning, Memory, and Cognition, 39, 748 –
Oberauer, K., & Eichenberger, S. (2013). Visual working memory declines
759. http://dx.doi.org/10.1037/a0029200
when more features must be remembered for each object. Memory &
Mercer, T. (2014). The loss of short-term visual representations over time:
Cognition, 41, 1212–1227. http://dx.doi.org/10.3758/s13421-013-
Decay or temporal distinctiveness? Journal of Experimental Psychol-
0333-6
ogy: Human Perception and Performance, 40, 2281–2288. http://dx.doi
Oberauer, K., Farrell, S., Jarrold, C., Pasiecznik, K., & Greaves, M. (2012).
.org/10.1037/a0038141
Interference between maintenance and processing in working memory:
Mercer, T., & Duffy, P. (2015). The loss of residual visual memories over
The effect of item-distractor similarity in complex span. Journal of
the passage of time. The Quarterly Journal of Experimental Psychology:
Experimental Psychology: Learning, Memory, and Cognition, 38, 665–
Human Experimental Psychology, 68, 242–248. http://dx.doi.org/10
685. http://dx.doi.org/10.1037/a0026337
.1080/17470218.2014.975256
Oberauer, K., & Hein, L. (2012). Attention to information in working
Mercer, T., & McKeown, D. (2010). Updating and feature overwriting in
memory. Current Directions in Psychological Science, 21, 164 –169.
short-term memory for timbre. Attention, Perception, & Psychophysics, http://dx.doi.org/10.1177/0963721412444727
72, 2289 –2303. http://dx.doi.org/10.3758/BF03196702 Oberauer, K., & Kliegl, R. (2001). Beyond resources: Formal models of
Mercer, T., & McKeown, D. (2014). Decay uncovered in nonverbal short- complexity effects and age differences in working memory. European
term memory. Psychonomic Bulletin & Review, 21, 128 –135. http://dx Journal of Cognitive Psychology, 13, 187–215. http://dx.doi.org/10
.doi.org/10.3758/s13423-013-0472-6 .1080/09541440042000278
Miller, G. A. (1956). The magical number seven plus or minus two: Some Oberauer, K., & Kliegl, R. (2006). A formal model of capacity limits in
limits on our capacity for processing information. Psychological Review, working memory. Journal of Memory and Language, 55, 601– 626.
63, 81–97. http://dx.doi.org/10.1037/h0043158 http://dx.doi.org/10.1016/j.jml.2006.08.009
Monsell, S., Sumner, P., & Waters, H. (2003). Task-set reconfiguration Oberauer, K., & Lange, E. B. (2008). Interference in verbal working
with predictable and unpredictable task switches. Memory & Cognition, memory: Distinguishing similarity-based confusion, feature overwriting,
31, 327–342. http://dx.doi.org/10.3758/BF03194391 and feature migration. Journal of Memory and Language, 58, 730 –745.
Morey, C. C., & Bieler, M. (2013). Visual short-term memory always http://dx.doi.org/10.1016/j.jml.2007.09.006
requires general attention. Psychonomic Bulletin & Review, 20, 163– Oberauer, K., & Lewandowsky, S. (2008). Forgetting in immediate serial
170. http://dx.doi.org/10.3758/s13423-012-0313-z recall: Decay, temporal distinctiveness, or interference? Psychological
Morey, C. C., & Cowan, N. (2004). When visual and verbal memories Review, 115, 544 –576. http://dx.doi.org/10.1037/0033-295X.115.3.544
compete: Evidence of cross-domain limits in working memory. Psycho- Oberauer, K., & Lewandowsky, S. (2011). Modeling working memory: A
nomic Bulletin & Review, 11, 296 –301. http://dx.doi.org/10.3758/ computational implementation of the Time-Based Resource-Sharing the-
BF03196573 ory. Psychonomic Bulletin & Review, 18, 10 – 45. http://dx.doi.org/10
Mueller, S. T., Seymour, T. L., Kieras, D. E., & Meyer, D. E. (2003). .3758/s13423-010-0020-6
Theoretical implications of articulatory duration, phonological similar- Oberauer, K., & Lewandowsky, S. (2013). Evidence against decay in
ity, and phonological complexity in verbal working memory. Journal of verbal working memory. Journal of Experimental Psychology: General,
Experimental Psychology: Learning, Memory, and Cognition, 29, 1353– 142, 380 – 411. http://dx.doi.org/10.1037/a0029588
1380. http://dx.doi.org/10.1037/0278-7393.29.6.1353 Oberauer, K., & Lewandowsky, S. (2014). Further evidence against decay
Nairne, J. S. (1990). A feature model of immediate memory. Memory & in working memory. Journal of Memory and Language, 73, 15–30.
Cognition, 18, 251–269. http://dx.doi.org/10.3758/BF03213879 http://dx.doi.org/10.1016/j.jml.2014.02.003
 WM CAPACITY 797

Oberauer, K., Lewandowsky, S., Farrell, S., Jarrold, C., & Greaves, M. Phaf, R. H., & Wolters, G. (1993). Attentional shifts in maintenance
(2012). Modeling working memory: An interference model of complex rehearsal. The American Journal of Psychology, 106, 353–382. http://
span. Psychonomic Bulletin & Review, 19, 779 – 819. http://dx.doi.org/ dx.doi.org/10.2307/1423182
10.3758/s13423-012-0272-4 Phillips, W. A., & Christie, D. F. M. (1977). Components of visual
Oberauer, K., Souza, A. S., Druey, M. D., & Gade, M. (2013). Analogous memory. The Quarterly Journal of Experimental Psychology, 29, 117–
mechanisms of selection and updating in declarative and procedural 133. http://dx.doi.org/10.1080/00335557743000080
working memory: Experiments and a computational model. Cognitive Ploner, C. J., Gaymard, B., Rivaud, S., Agid, Y., & Pierrot-Deseilligny, C.
Psychology, 66, 157–211. http://dx.doi.org/10.1016/j.cogpsych.2012.11 (1998). Temporal limits of spatial working memory in humans. Euro-
.001 pean Journal of Neuroscience, 10, 794 –797. http://dx.doi.org/10.1046/
Oberauer, K., Süß, H.-M., Schulze, R., Wilhelm, O., & Wittmann, W. W. j.1460-9568.1998.00101.x
(2000). Working memory capacity - facets of a cognitive ability con- Poirier, M., Saint-Aubin, J., Musselwhite, K., Mohanadas, T., & Ma-
struct. Personality and Individual Differences, 29, 1017–1045. http://dx hammed, G. (2007). Visual similarity effects on short-term memory for
.doi.org/10.1016/S0191-8869(99)00251-2 order: The case of verbally labeled pictorial stimuli. Memory & Cogni-
Oberauer, K., Süß, H.-M. , Wilhelm, O., & Sander, N. (2007). Individual tion, 35, 711–723. http://dx.doi.org/10.3758/BF03193309
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

differences in working memory capacity and reasoning ability. In A. R. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

A. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N. Towse (Eds.), 85, 59 –108. http://dx.doi.org/10.1037/0033-295X.85.2.59
Variation in working memory (pp. 49 –75). New York, NY: Oxford Ratcliff, R., Thapar, A., & McKoon, G. (2010). Individual differences,
University Press. aging, and IQ in two-choice tasks. Cognitive Psychology, 60, 127–157.
Oberauer, K., Süß, H.-M., Wilhelm, O., & Wittmann, W. W. (2003). The http://dx.doi.org/10.1016/j.cogpsych.2009.09.001
multiple faces of working memory - storage, processing, supervision, Raye, C. L., Johnson, M. K., Mitchell, K. J., Greene, E. J., & Johnson,
and coordination. Intelligence, 31, 167–193. http://dx.doi.org/10.1016/ M. R. (2007). Refreshing: A minimal executive function. Cortex, 43,
S0160-2896(02)00115-0 135–145. http://dx.doi.org/10.1016/S0010-9452(08)70451-9
Oberauer, K., Süß, H.-M., Wilhelm, O., & Wittmann, W. W. (2008). Ricker, T. J., & Cowan, N. (2010). Loss of visual working memory within
Which working memory functions predict intelligence? Intelligence, 36, seconds: The combined use of refreshable and non-refreshable features.
Journal of Experimental Psychology: Learning, Memory, and Cogni-
641– 652. http://dx.doi.org/10.1016/j.intell.2008.01.007
tion, 36, 1355–1368. http://dx.doi.org/10.1037/a0020356
Olson, I. R., & Jiang, Y. (2002). Is visual short-term memory object based?
Ricker, T. J., & Cowan, N. (2014). Differences between presentation
Rejection of the “strong-object” hypothesis. Perception & Psychophys-
methods in working memory procedures: A matter of working memory
ics, 64, 1055–1067. http://dx.doi.org/10.3758/BF03194756
consolidation. Journal of Experimental Psychology: Learning, Memory,
Page, M. P. A., & Norris, D. (1998). The primacy model: A new model of
and Cognition, 40, 417– 428. http://dx.doi.org/10.1037/a0034301
immediate serial recall. Psychological Review, 105, 761–781. http://dx
Ricker, T. J., Spiegel, L. R., & Cowan, N. (2014). Time-based loss in visual
.doi.org/10.1037/0033-295X.105.4.761-781
short-term memory is from trace decay, not temporal distinctiveness.
Park, D. C., Lautenschlager, G., Hedden, T., Davidson, N. S., Smith, A. D.,
Journal of Experimental Psychology: Learning, Memory, and Cogni-
& Smith, P. K. (2002). Models of visuospatial and verbal memory across
tion, 40, 1510 –1523. http://dx.doi.org/10.1037/xlm0000018
the adult life span. Psychology and Aging, 17, 299 –320. http://dx.doi
Rogers, R. D., & Monsell, S. (1995). Costs of a predictable switch between
.org/10.1037/0882-7974.17.2.299
simple cognitive tasks. Journal of Experimental Psychology: General,
Parmentier, F. B. R. (2011). Exploring the determinants of memory for
124, 207–231. http://dx.doi.org/10.1037/0096-3445.124.2.207
spatial sequences. In A. Vandierendonck & A. Szmalec (Eds.), Spatial
Rohrer, D., & Pashler, H. E. (2003). Concurrent task effects on memory
working memory (pp. 68 – 86). Hove: Psychology Press.
retrieval. Psychonomic Bulletin & Review, 10, 96 –103. http://dx.doi.org/
Parmentier, F. B. R., Elford, G., & Maybery, M. (2005). Transitional 10.3758/BF03196472
information in spatial serial memory: Path characteristics affect recall Saito, S., & Miyake, A. (2004). On the nature of forgetting and the
performance. Journal of Experimental Psychology: Learning, Memory, processing-storage relationship in reading span performance. Journal of
and Cognition, 31, 412– 427. http://dx.doi.org/10.1037/0278-7393.31.3 Memory and Language, 50, 425– 443. http://dx.doi.org/10.1016/j.jml
.412 .2003.12.003
Parmentier, F. B. R., & Jones, D. M. (2000). Functional characteristics of Sakai, K., & Inui, T. (2002). A feature-segmentation model of short-term
auditory temporal-spatial short-term memory: Evidence from serial or- visual memory. Perception, 31, 579 –589. http://dx.doi.org/10.1068/
der errors. Journal of Experimental Psychology: Learning, Memory, and p3320
Cognition, 26, 222–238. http://dx.doi.org/10.1037/0278-7393.26.1.222 Salthouse, T. A. (1994). How many causes are there of aging-related
Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. decrements in cognitive functioning? Developmental Review, 14, 413–
Psychological Bulletin, 116, 220 –244. http://dx.doi.org/10.1037/0033- 437. http://dx.doi.org/10.1006/drev.1994.1016
2909.116.2.220 Salthouse, T. A. (1996). The processing-speed theory of adult age differ-
Pearson, D., & Sahraie, A. (2003). Oculomotor control and the mainte- ences in cognition. Psychological Review, 103, 403– 428. http://dx.doi
nance of spatially and temporally distributed events in visuo-spatial .org/10.1037/0033-295X.103.3.403
working memory. The Quarterly Journal of Experimental Psychology, Sanders, A. F., & Schroots, J. J. F. (1969). Cognitive categories and
56, 1089 –1111. http://dx.doi.org/10.1080/02724980343000044 memory span III. Effects of similarity on recall. The Quarterly Journal
Penney, C. G. (1975). Modality effects in short-term verbal memory. of Experimental Psychology, 21, 21–28. http://dx.doi.org/10.1080/
Psychological Bulletin, 82, 68 – 84. http://dx.doi.org/10.1037/h0076166 14640746908400190
Pertzov, Y., Bays, P. M., Joseph, S., & Husain, M. (2013). Rapid forgetting Saults, J. S., & Cowan, N. (2007). A central capacity limit to the simul-
prevented by retrospective attention cues. Journal of Experimental Psy- taneous storage of visual and auditory arrays in working memory.
chology: Human Perception and Performance, 39, 1224 –1231. http:// Journal of Experimental Psychology: General, 136, 663– 684. http://dx
dx.doi.org/10.1037/a0030947 .doi.org/10.1037/0096-3445.136.4.663
Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individ- Schmiedek, F., Oberauer, K., Wilhelm, O., Süß, H. M., & Wittmann,
ual verbal items. Journal of Experimental Psychology, 58, 193–198. W. W. (2007). Individual differences in components of reaction time
http://dx.doi.org/10.1037/h0049234 distributions and their relations to working memory and intelligence.
 798 OBERAUER, FARRELL, JARROLD, AND LEWANDOWSKY

Journal of Experimental Psychology: General, 136, 414 – 429. http://dx load and presentation rate. European Journal of Cognitive Psychology,
.doi.org/10.1037/0096-3445.136.3.414 5, 417– 433. http://dx.doi.org/10.1080/09541449308520128
Schweickert, R. (1993). A multinomial processing tree model for degra- Tehan, G., Fogarty, G., & Ryan, K. (2004). The contribution to immediate
dation and redintegration in immediate recall. Memory & Cognition, 21, serial recall of rehearsal, search speed, access to lexical memory, and
168 –175. http://dx.doi.org/10.3758/BF03202729 phonological coding: An investigation at the construct level. Memory &
Schweickert, R., & Boruff, B. (1986). Short-term memory capacity: Magic Cognition, 32, 711–721. http://dx.doi.org/10.3758/BF03195861
number or magic spell? Journal of Experimental Psychology: Learning, Tehan, G., & Lalor, D. M. (2000). Individual differences in memory span:
Memory, and Cognition, 12, 419 – 425. http://dx.doi.org/10.1037/0278- The contribution of rehearsal, access to lexical memory, and output
7393.12.3.419 speed. The Quarterly Journal of Experimental Psychology, 53, 1012–
Sederberg, P. B., Howard, M. W., & Kahana, M. J. (2008). A context- 1038. http://dx.doi.org/10.1080/713755933
based theory of recency and contiguity in free recall. Psychological Tombu, M., & Jolicoeur, P. (2003). A central capacity sharing model of
Review, 115, 893–912. http://dx.doi.org/10.1037/a0013396 dual-task performance. Journal of Experimental Psychology: Human
Service, E. (1998). The effect of word length on immediate serial recall Perception and Performance, 29, 3–18. http://dx.doi.org/10.1037/0096-
depends on phonological complexity, not articulatory duration. The 1523.29.1.3
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Quarterly Journal of Experimental Psychology, 51A, 283–304. http://dx Towse, J. N., & Hitch, G. J. (1995). Is there a relationship between task
This document is copyrighted by the American Psychological Association or one of its allied publishers.

.doi.org/10.1080/713755759 demand and storage space in tests of working memory capacity? The
Sewell, D. K., Lilburn, S. D., & Smith, P. L. (2014). An information Quarterly Journal of Experimental Psychology, 48, 108 –124. http://dx
capacity limitation of visual short-term memory. Journal of Experimen- .doi.org/10.1080/14640749508401379
tal Psychology: Human Perception and Performance, 40, 2214 –2242. Towse, J. N., Hitch, G. J., & Hutton, U. (2000). On the interpretation of
http://dx.doi.org/10.1037/a0037744 working memory span in adults. Memory & Cognition, 28, 341–348.
Shah, P., & Miyake, A. (1996). The separability of working memory http://dx.doi.org/10.3758/BF03198549
resources for spatial thinking and language processing: An individual Towse, J. N., & Houston-Price, C. M. T. (2001). Combining representa-
differences approach. Journal of Experimental Psychology: General, tions in working memory: A brief report. British Journal of Develop-
125, 4 –27. http://dx.doi.org/10.1037/0096-3445.125.1.4 mental Psychology, 19, 319 –324. http://dx.doi.org/10.1348/
Shipstead, Z., & Engle, R. E. (2013). Interference within the focus of
026151001166100
attention: Working memory tasks reflect more than temporary mainte-
Tsubomi, H., Fukuda, K., Watanabe, K., & Vogel, E. K. (2013). Neural
nance. Journal of Experimental Psychology: Learning, Memory, and
limits to representing objects still within view. The Journal of Neuro-
Cognition, 39, 277–289. http://dx.doi.org/10.1037/a0028467
science, 33, 8257– 8263. http://dx.doi.org/10.1523/JNEUROSCI.5348-
Shipstead, Z., Lindsey, D. R. B., Marshall, R. L., & Engle, R. E. (2014).
12.2013
The mechanisms of working memory capacity: Primary memory, sec-
Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task
ondary memory, and attention control. Journal of Memory and Lan-
dependent? Journal of Memory and Language, 28, 127–154. http://dx
guage, 72, 116 –141. http://dx.doi.org/10.1016/j.jml.2014.01.004
.doi.org/10.1016/0749-596X(89)90040-5
Sigman, M., & Dehaene, S. (2005). Parsing a cognitive task: A character-
Underwood, B. J. (1975). Individual differences as a crucible in theory
ization of the mind’s bottleneck. PLoS Biology, 3, e37. http://dx.doi.org/
construction. American Psychologist, 30, 128 –134. http://dx.doi.org/10
10.1371/journal.pbio.0030037
.1037/h0076759
Sigman, M., & Dehaene, S. (2006). Dynamics of the central bottleneck:
Unsworth, N. (2010). On the division of working memory and long-term
Dual-task and task uncertainty. PLoS Biology, 4, e220.
memory and their relation to intelligence: A latent variable approach.
Smyth, M. M., & Scholey, K. A. (1994). Characteristics of spatial memory
Acta Psychologica, 134, 16 –28. http://dx.doi.org/10.1016/j.actpsy.2009
span: Is there an analogy to the word length effect, based on movement
time? The Quarterly Journal of Experimental Psychology, 47, 91–117. .11.010
http://dx.doi.org/10.1080/14640749408401145 Unsworth, N., Brewer, G. A., & Spillers, G. J. (2009). There’s more to the
Sobel, K. V., Gerrie, M. P., Poole, B. J., & Kane, M. J. (2007). Individual working memory capacity-fluid intelligence relationship than just sec-
differences in working memory capacity and visual search: The roles of ondary memory. Psychonomic Bulletin & Review, 16, 931–937. http://
top-down and bottom-up processing. Psychonomic Bulletin & Review, dx.doi.org/10.3758/PBR.16.5.931
14, 840 – 845. http://dx.doi.org/10.3758/BF03194109 Unsworth, N., & Engle, R. W. (2007). The nature of individual differences
Souza, A. S., & Oberauer, K. (2015). Time-based forgetting in visual in working memory capacity: Active maintenance in primary memory
working memory reflects temporal distinctiveness, not decay. Psycho- and controlled search from secondary memory. Psychological Review,
nomic Bulletin & Review, 22, 156 –162. http://dx.doi.org/10.3758/ 114, 104 –132. http://dx.doi.org/10.1037/0033-295X.114.1.104
s13423-014-0652-z Unsworth, N., Fukuda, K., Awh, E., & Vogel, E. K. (2014). Working
Souza, A. S., Rerko, L., & Oberauer, K. (2014). Unloading and reloading memory and fluid intelligence: Capacity, attention control, and second-
working memory: Attending to one item frees capacity. Journal of ary memory retrieval. Cognitive Psychology, 71, 1–26. http://dx.doi.org/
Experimental Psychology: Human Perception and Performance, 40, 10.1016/j.cogpsych.2014.01.003
1237–1256. http://dx.doi.org/10.1037/a0036331 Unsworth, N., Schrock, J. C., & Engle, R. W. (2004). Working memory
Stahl, C., Voss, A., Schmitz, F., Nuszbaum, M., Tüscher, O., Lieb, K., & capacity and the antisaccade task: Individual differences in voluntary
Klauer, K. C. (2014). Behavioral components of impulsivity. Journal of saccade control. Journal of Experimental Psychology: Learning, Mem-
Experimental Psychology: General, 143, 850 – 886. http://dx.doi.org/10 ory, and Cognition, 30, 1302–1321. http://dx.doi.org/10.1037/0278-
.1037/a0033981 7393.30.6.1302
Sternberg, S. (1966). High-speed scanning in human memory. Science, Vallar, G., & Baddeley, A. D. (1982). Short-term forgetting and the
153, 652– 654. http://dx.doi.org/10.1126/science.153.3736.652 articulatory loop. The Quarterly Journal of Experimental Psychology
Tan, L., & Ward, G. (2008). Rehearsal in immediate serial recall. Psycho- Section A: Human Experimental Psychology, 34, 53– 60. http://dx.doi
nomic Bulletin & Review, 15, 535–542. http://dx.doi.org/10.3758/PBR .org/10.1080/14640748208400857
.15.3.535 van den Berg, R., Awh, E., & Ma, W. J. (2014). Factorial comparison of
Teasdale, J. D., Proctor, L., Lloyd, C. A., & Baddeley, A. D. (1993). working memory models. Psychological Review, 121, 124 –149. http://
Working memory and stimulus-independent thought: Effects of memory dx.doi.org/10.1037/a0035234
 WM CAPACITY 799

Vergauwe, E., Barrouillet, P., & Camos, V. (2010). Do mental processes chology: Human Perception and Performance, 32, 1436 –1451. http://
share a domain-general resource? Psychological Science, 21, 384 –390. dx.doi.org/10.1037/0096-1523.32.6.1436
http://dx.doi.org/10.1177/0956797610361340 Walsh, V. (2003). A theory of magnitude: Common cortical metrics of
Vergauwe, E., Camos, V., & Barrouillet, P. (2014). The impact of storage time, space and quantity. Trends in Cognitive Sciences, 7, 483– 488.
on processing: How is information maintained in working memory? http://dx.doi.org/10.1016/j.tics.2003.09.002
Journal of Experimental Psychology: Learning, Memory, and Cogni- Wilhelm, O., Hildebrandt, A., & Oberauer, K. (2013). What is working
tion, 40, 1072–1095. http://dx.doi.org/10.1037/a0035779 memory capacity, and how can we measure it? Frontiers in Psychology,
Vergauwe, E., Dewaele, N., Langerock, N., & Barrouillet, P. (2012). 4, 433. http://dx.doi.org/10.3389/fpsyg.2013.00433
Evidence for a central pool of general resources in working memory. Williams, M., Hong, S. W., Kang, M.-S., Carlisle, N. B., & Woodman,
Journal of Cognitive Psychology, 24, 359 –366. http://dx.doi.org/10 G. F. (2013). The benefit of forgetting. Psychonomic Bulletin & Review,
.1080/20445911.2011.640625 20, 348 –355. http://dx.doi.org/10.3758/s13423-012-0354-3
Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural Zhang, W., & Luck, S. J. (2008). Discrete fixed-resolution representations in visual
measures reveal individual differences in controlling access to work- working memory. Nature, 453, 233–235. http://dx.doi.org/10.1038/nature06860
ing memory. Nature, 438, 500 –503. http://dx.doi.org/10.1038/ Zhang, W., & Luck, S. J. (2009). Sudden death and gradual decay in visual
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

nature04171 working memory. Psychological Science, 20, 423– 428. http://dx.doi

Vogel, E. K., Woodman, G. F., & Luck, S. J. (2001). Storage of features, .org/10.1111/j.1467-9280.2009.02322.x
This document is copyrighted by the American Psychological Association or one of its allied publishers.

conjunctions and objects in visual working memory. Journal of Exper-

imental Psychology: Human Perception and Performance, 27, 92–114.
http://dx.doi.org/10.1037/0096-1523.27.1.92 Received November 24, 2014
Vogel, E. K., Woodman, G. F., & Luck, S. J. (2006). The time course of Revision received December 14, 2015
consolidation in visual working memory. Journal of Experimental Psy- Accepted December 15, 2015 䡲