Neuropsychologia 49 (2011) 1476–1486
Contents lists available at ScienceDirect
Neuropsychologia
journa l homepage: www.e lsev ier .com/ loca
Encoding strategy accounts for individual differe
measur
A.C. Link
MRC Cognition
a r t i c l
Article history:
Received 4 Au
Received in re
19 November
Accepted 22 N
Available onlin
Keywords:
Visual short-term memory
Memory capacity
Individual differences
Selective attention
IQ
apaci
ve b
recen
ction
ed u
act d
rator
encoding dominates individual differences in change detection measures of visual short-term memory.
In a first study, we unpredictably varied whether short-term memory was probed using change detec-
tion or whole report, encouraging participants to adopt the same encoding strategy throughout the
tasks. Change detection performance of lower-IQ individuals improved. In a second study, we found
that deficits in top-down attentional selectivity can be alleviated in participants with low change detec-
tion performance by providing helpful grouping information during encoding. Finally, a meta-analysis of
1. Introdu
Individu
a popular t
One reason
(VSTM) cap
that the tw
not identic
Eagle, 2003
uals to diffe
intelligence
Two dis
both, an en
is followed
memory is
∗ Correspon
E-mail add
alejandro.vice
daniel.mitchel
rhodri.cusack@
0028-3932/$ –
doi:10.1016/j.neuroimaging data from 112 participants performing a variety of VSTM tasks showed that performance
correlates with activity in several parietal and frontal regions during the encoding but not the mainte-
nance phase. Taken together, these results support the notion that encoding strategy and not short-term
memory capacity itself largely determines individual differences in visual change detection performance.
© 2010 Elsevier Ltd. All rights reserved.
ction
al differences in visual short-term memory have been
opic in both behavioural and neuroimaging research.
for this is the link between visual short-term memory
acity and intelligence. Numerous studies have shown
o are highly related (e.g. Kyllonen & Christal, 1990) but
al (Ackerman, Beier, & Boyle, 2005; Conway, Kane, &
). However, it is less clear what factors cause individ-
r in their performance in VSTM tasks and measures of
.
tinct paradigms are often used to measure VSTM. In
coding display containing the items to be memorized
by a maintenance delay. The paradigms differ in how
then assessed. In change detection (CD), a probe is
ding author. Tel.: +44 01223 355294.
resses: annika.linke@mrc-cbu.cam.ac.uk (A.C. Linke),
nte-grabovetsky@mrc-cbu.cam.ac.uk (A. Vicente-Grabovetsky),
l@mrc-cbu.cam.ac.uk (D.J. Mitchell),
mrc-cbu.cam.ac.uk (R. Cusack).
presented and participants are asked to decide whether it is the
same or different from the corresponding stimulus during encod-
ing. In report tasks, participants have to report the stimuli they
remember—if the items are letters for instance, by saying them
or typing them in. Report tasks in which only a subset of items
has to be remembered, are referred to as “partial report” (PR),
while tasks in which all items have to be remembered are known
as “whole report” (WR) tasks. In a recent study, we found that
the relationship between VSTM performance and intelligence was
dependent largely on whether individuals were assessed using
CD or WR (Cusack et al., 2009). WR performance increased with
load, peaking at a set-size of 4 and staying constant as set-size
increased further. CD estimates of the number of items in mem-
ory also peaked at set-size 4, but thereafter showed a marked dip
as load increased. Importantly, IQ correlatedwith the size of this dip
but not with VSTM capacity (as measured byK=N× (H− FA), where
N= number of items in set, H= proportion of hits, FA = proportion of
false alarms; Cowan, 2001). This indicates that an additional cog-
nitive process during change detection underlies the relationship
between intelligence and VSTM capacity.
Numerous researchers (e.g. Heitz et al., 2006; Jarrold & Towse,
2008; Kane & Engle, 2002) have suggested that selective atten-
see front matter © 2010 Elsevier Ltd. All rights reserved.
neuropsychologia.2010.11.034es of VSTM
e ∗, A. Vicente-Grabovetsky, D.J. Mitchell, R. Cusack
and Brain Sciences Unit, 15 Chaucer Road, CB2 7EF Cambridge, UK
e i n f o
gust 2010
vised form
2010
ovember 2010
e 3 December 2010
a b s t r a c t
Visual short-term memory (VSTM) c
vidual differences in performance ha
domains in children and adults. We
component necessary for change dete
Veldsman, & Mitchell, 2009). It remain
ing change detection have most imp
present recent findings from our labote /neuropsychologia
nces in change detection
ty is often assessed using change detection tasks, and indi-
een shown to predict cognitive aptitudes across a range of
tly showed that intelligence correlates with an attentional
rather than with memory capacity per se (Cusack, Lehmann,
nclear, however, whether different attentional strategies dur-
uring the encoding or maintenance of information. Here we
y supporting the hypothesis that attentional selection during

A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486 1477
tion is the driving factor that determines VSTM capacity as well
as performance in cognitively demanding tasks such as assess-
ments of IQ. Engle (2002) for instance argues that working memory
is a manifestation of attentional capabilities, which in turn cor-
relate with
participants
encoding p
instead of
the memory
ments of VS
such that d
pants are d
Thefirst
port for thi
encoding st
viated by en
unclear, how
determines
individuals
therefore, d
ies conduct
VSTM capa
brain that a
nitively dem
From pr
networks a
very simila
Mesulam, 1
instance ha
demanding
terior parie
amount of i
2004, 2005
is hard to d
upon simila
Mitchell an
responded
ory tasks. A
between th
capacity (e
Engle, & Kh
In a neuroi
in attention
memory ca
ory perform
that memo
attention d
potential in
gest that ac
the mainten
tion of atten
the other h
that differe
selectiondu
ter at selec
also been e
colleagues
selection ac
ing only th
Vogel, & Oh
globus pall
tions that r
short-term
seem to ari
in these fro
in short-ter
In this paper we present two new behavioural experiments to
probe the link between attentional processes at encoding and per-
formance on change detection and whole report tasks. We then
examine individual differences across a set of neuroimaging exper-
from
ral a
t for
agni
two
ual
ion ta
enco
y of
ted c
erim
havi
Block
he fir
on d
M ca
M w
sed
emem
rese
gher
ce w
tima
nal f
issue
duri
2009
y at
er-IQ
ered
isru
prob
es. It
are p
r, and
.
aim
ntify
er s
ems
ed th
ts. In
diff
y. In
we r
ould
t disc
as re
hen
nal
strat
. Me
betw
ticipa
n-ve
’s Cu
(N= 6intelligence. We therefore suggested that lower-IQ
adopted suboptimal attentional selection during the
hase of the CD task, attempting to encode everything
selecting a manageable subset of items, thus making
representations more fragile. Importantly, WR assess-
TM appear to foster a more selective encoding strategy
ifferences between low and high performing partici-
iminished.
twostudiespresented in thispaperprovide further sup-
s hypothesis by showing that individual differences in
rategy dominate CD measures of VSTM and can be alle-
couraging more selection during encoding. It remained
ever, whether this impact of attentional selection only
which items are encoded or also has an impact on how
maintain this information in short-term memory. We,
rewuponresults fromfivedifferentneuroimaging stud-
ed in our laboratory to investigate the link between
city as measured by CD and activity in regions of the
re commonly activated by short-term memory and cog-
anding tasks.
evious neuroimaging studies, it is clear that the brain
ctivated by attention and short-term memory are
r (Awh & Jonides, 2001; LaBar, Gitelman, Parrish, &
999; Mayer et al., 2007). Duncan and Owen (2000) for
ve shown that some regions are activated by cognitively
tasks irrespective of the precise nature of the task. Pos-
tal cortex is one of the regions that reliably reflects the
nformation held in short-term memory (Todd & Marois,
) but how far this activation is purely memory-specific
isentangle. In attentional and perceptual tasks drawing
r stimuli to those often used in change detection tasks,
d Cusack (2008) showed that posterior parietal cortex
in the same load dependent manner as during mem-
dditionally, numerous studies have found correlations
e ability to control attention and short-term memory
.g. Awh & Jonides, 2001; Bleckley, Durso, Crutchfield,
anna, 2003; Kane, Bleckley, Conway, & Engle, 2001).
maging study, Linden et al. (2003) showed that limits
seemed to be the constraining factor for short-term
pacity, but how and when attention influenced mem-
ance remained unclear. Kane and Engle (2002) argue
ry capacity is determined by demands on executive
uring the maintenance of information in the face of
terference. Similarly, Curtis and D’Esposito (2003) sug-
tivity of dorsolateral prefrontal cortex (DLPFC) during
ance phase of memory tasks reflects top-down direc-
tion to control rehearsal of information in memory. On
and, Vogel, McCollough and Machizawa (2005) suggest
nces in VSTM capacity relate to efficiency of attentional
ringencoding,withhighcapacity individualsbeingbet-
ting the relevant information to remember. This has
ndorsed by McNab and Klingberg (2008) and Edin and
(Edin et al., 2009) whose findings suggest attentional
ts as a gatekeeper to short-term memory by select-
e most relevant items during encoding (also see Awh,
, 2006). They showed that the middle frontal gyrus,
idus and basal ganglia serve attentional control func-
egulate which information is passed on for storage in
memory. Individual differences in VSTM capacity, thus,
se from different abilities to filter relevant information
ntal and subcortical regions before they are maintained
m memory by parietal cortex.
iments
of neu
we tes
and m
in the
individ
detect
during
strateg
presen
2. Exp
2.1. Be
2.1.1.
In t
selecti
of VST
of VST
depres
were r
were p
was hi
eviden
this es
additio
An
rupted
et al.,
strateg
in low
consid
couldd
paper,
set-siz
items
weake
display
The
by ide
at high
way it
repeat
contex
in two
strateg
other,
there c
did no
task w
types w
Additio
due to
2.1.1.1
ences
16 par
low no
Cattell
studyour laboratory that have allowed separate modelling
ctivity during encoding and maintenance. Specifically,
a relationship between change detection performance
tude of activity in different regions of interest (ROIs)
task phases. Our results confirm the hypothesis that
differences in VSTM capacity as measured by change
sks are related to different attentional strategies used
ding, with lower-IQ participants adopting a suboptimal
trying to encode all stimuli even when the information
learly exceeds their capacity limits.
ents
oural experiments
ed vs. intermixed probes
st experiment we tested the hypothesis that attentional
uring encoding influences change detection measures
pacity. Cusack et al. (2009) found that CD measures
ere contaminated by some cognitive process that
performance at higher set-sizes, so that fewer items
bered when eight were presented than when just four
nted. Measuring VSTM using WR gave an estimate that
and less variable, particularly for larger set-sizes. No
as found for a relationship between non-verbal IQ and
te of VSTM capacity. Instead IQ was correlated with an
actor that contaminated change detection estimates.
that was not fully resolved is what process was dis-
ngchangedetection.Onepossibility (favoured inCusack
) is that change detection encourages a maladaptive
encoding of trying to remember everything, particularly
individuals. However, another possibility that must be
is that the presentation of the probe in change detection
ptmemory recall. Toexplain thepatternof results in that
e interference would have to be more severe at higher
might be, for example, that when a greater number of
resented, the memory traces created for each item are
hence more easily disrupted by the subsequent probe
of the current experiment was to investigate this issue,
ing the portion of a change detection task that suffers
et-sizes, particularly in those with lower IQ: is it the
are encoded, or the nature of the probe? To do this, we
e comparison between CD and WR, but in two different
one, as in Cusack et al. (2009), we presented CD and WR
erent blocks. This allowed for adaptation of encoding
a second pair of blocks, which were identical to each
andomly intermixed the CD and WR trials. In this case,
be no difference in encoding strategy, as the volunteer
over until the probe phase of each trial what response
quired. Any residual difference between the two trial
they were intermixed had to be due to effects at probe.
effects between blocked and intermixed trials had to be
egic differences, presumably at encoding.
thod. Since Cusack et al. (2009) found substantial differ-
een WR and CD performance in a lower-IQ group only,
nts (13 females, 22–61 years old, M= 53, SD= 12.0) with
rbal intelligence (M= 30.1, SD= 2.7, as estimated using
lture Fair Test), were selected from a previous larger
1, 21–64 years old, M= 47, SD= 12.97, grouped into a

1478 A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486
high and low IQ group by median-split) to directly assess whether
performance could be improved in these participants. Mean age
in this sample was high as participants were recruited during the
summer and not from the typical student population. Each par-
ticipant com
trials were C
blocks, equ
mixed. As in
closely mat
set-sizes of
nance perio
the screen b
sen from th
white) was
with height
rectangle ∼
screen, subj
not be close
appeared a
remembere
next trial d
the CD task
letter at the
(chosen at
two button
anywhere d
50% chance
could be ma
the next tri
button map
the particip
in total (96
completed
trial types w
theexperim
Ethics Com
2.1.1.2. Ana
size was es
To calculate
a double-hi
items were
(KCD =N× (H
of hits, FA
culate the
number of
subtracted
vative assum
letters that
mating the
KWR =C−E
were not in
submitted t
context (blo
to investiga
This was fo
interest.
2.1.1.3. Res
remembere
in the two d
All main
icantly wor
and in CD vs
main effect
but this w
erform
now w
) perf
dditi
2.546
sk an
orta
repl
, perf
r WR
set-s
ff in
D se
).
therm
ce in
s; in
egic
swi
and
mor
ch W
t to W
acro
tly, m
d CD
indiv
enco
s som
renc
it might be some artefact associated more specifically with
ponse production.
. Discussion. These results concur with our previous finding,
owthathow itemsareencodedhasa strong influenceonhow
ople perform in a change detection task. In agreement with
t al. (2005) and McNab and Klingberg (2008), we have pre-
hypothesized that the efficiency of attentional selection,
y whether an individual tries to encode all stimuli at once
cts a few, determines change detection performance and
ed to intelligence. We have explored this by unpredictably
g participants either using WR or CD, but we can also obtain
ging evidence by instructing participants to encode only a
of the presented stimuli. In the next experiment we there-
ed samples t-tests, two-tailed significance values reported.pleted four blocks. In one of the four blocks, all of the
D; in another block, they were all WR; and in two of the
al numbers of WR and CD trials were randomly inter-
Cusack et al. (2009), the WR and CD procedures were
ched. Identical encoding displays were presented, with
2, 3, 4, 6 or 8 letters, followed by a 1200 ms mainte-
d. The encoding displays were presented for 183 ms,
ackground was mid-grey, letters were randomly cho-
e set ABDEFGHJKMNQRTY, and their colour (black or
randomized. Letters were presented in upper case Arial
∼3.8◦ visual angle at a location randomised within a
28◦ by 37◦ visual angle centred at the middle of the
ect to the constraint that the centres of two letters could
r than ∼6.7◦. In the WR task, after the delay period a box
nd participants were asked to type all the letters they
d using a standard keyboard and then press “Enter”. The
id not commence until a response had been made. In
, a probe display was presented that comprised a single
location of one of the letters in the encoding display
random), and participants were asked to press one of
s to indicate whether the probe letter had been present
uring encoding, irrespective of location. There was a
that it had been present. As for the WR task, a response
de at any point after the onset of the probe display and
al began right after the response had been made. The
ping was counterbalanced across participants. Five of
ants, performed longer sessions comprising 960 trials
trials/set size/task) and the remaining 11 participants
480 trials in total (48 trials/set size/task). Blocks and
ithin each block were fully randomised. Approval for
entwasgivenby theCambridgePsychologicalResearch
mittee.
lysis. The number of items remembered at each set-
timated for each procedure, correcting for guessing.
the number of letters remembered in the CD tasks,
gh-threshold model was used, which assumes that KCD
remembered perfectly and the remainder were guessed
− FA) where N= number of items in set, H= proportion
= proportion of false alarms; Cowan, 2001). To cal-
number of letters remembered in the WR trials, the
letters correctly reported was counted and from this
an estimate of the guessing rate. This used the conser-
ption that the participants had learnt the subset of 15
could appear in the stimuli (hence possibly overesti-
guessing rate and slightly underestimating capacity):
×N/(15 −N) where E= number of letters reported that
the display, C= correctly reported letters. The data were
o a three-way repeated measures ANOVA, with factors
cked, intermixed), task (WR, CD) and set-size (5 levels)
te the effect of intermixing probes on performance K.
llowed by post-hoc analysis of the significant effects of
ults. Fig. 1 shows the estimates of the number of items
d (K) as a function of set-size for the two tasks (WR, CD)
ifferent contexts (blocked, intermixed).
effects were significant with people performing signif-
se in blocked vs. mixed trials (F(1, 15) = 13.142, p< .005)
. WR tasks (F(1, 15) = 29.893, p< .001). Furthermore, the
of set-size was significant (F(4, 60) = 26.008, p< .001)
as expected given that K is dependent on the set-
Fig. 1. P
do not k
WR-Mix
only.
size. A
15) = 1
text, ta
Imp
results
blocks
than fo
vidual
drop-o
sizes (C
p= .052
Fur
forman
set size
of strat
for task
worse,
with a
in whi
tle cos
p= .19
Las
WR an
sized;
when
remain
interfe
tively,
the res
2.1.1.4
andsh
well pe
Vogel e
viously
namel
or sele
is relat
probin
conver
subset
1 Pairance in the four different conditions of Study 1. When participants
hether they will be probed using WR or CD (VSTM-CD-Mix, VSTM-
ormance significantly improves in the CD condition (red-dashed line)
onally, the interaction between context and task (F(1,
, p< .005) and the three-way interaction between con-
d set-size were significant (F(4, 60) = 3.432, p< .05).
ntly, subsequent post-hoc analyses1 revealed that our
icated Cusack et al. (2009). When presented in separate
ormance for CD (as measured by K) was much worse
(WR vs. CD, t(15) = 5.771, p< .001 across set-sizes; indi-
izes 3,4,6,8, p< .01). Furthermore, there was a distinct
the number of items remembered in CD at higher set-
t-size 4 vs. 8, t(15) = 2.376, p< .05; 4 vs. 6, t(15) = 2.106,
ore, intermixing CD with WR actually improved per-
theCD task (CD-Mixvs. CD, t(15) = 3.711,p< .005across
dividual set sizes: 3, 6, p< .05; 8, p= .051). In the absence
adjustments, it might be expected that the requirement
tching on a trial-to-trial basis would make performance
so the reverse effect is impressive. This is consistent
e selective strategy on CD trials as a result of a context
R must sometimes be performed. There was only a lit-
R when mixed with CD (WR vs. WR-Mix, t(15) = 1.375,
ss set sizes; individual set-size 8, p< .005).
ixing trials did not abolish all of the difference between
(WR-Mix vs.-CD-Mix, t(15) = 2.846, p< 0.05 across set-
idual set-sizes 3 and 4, p< .05). This suggests that even
ding strategies are matched across WR and CD, there
e effect of the probe method. This might be due to an
e effect of the probe display on memory in CD. Alterna-

A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486 1479
fore used a partial report task and provided bottom-up grouping
cues to manipulate the difficulty of attentional selection.
2.1.2. Manipulating the difficulty of attentional selection in PR
We prev
variable es
hypothesize
selection. In
tional strat
compare pe
task that d
way to cont
to use a par
1984; Dunc
this end, w
items in the
simultaneo
This impos
therefore, b
tionally, we
bottom-up
grouping) o
task instruc
Bottom-
will bias att
grouped to
for individu
tional strat
helpful bot
considerabl
measure is
were, there
PR perform
assessed by
ferent pred
we split pa
task. If atte
of VSTM, an
larger for p
participants
dict that pa
benefit mos
encode.
2.1.2.1. Me
were femal
36.3 (SD= 1
ous experim
task descri
VSTM capac
the entire
by a media
age (t(14) =
females; hig
distribution
During t
variable (1,
After a 120
probed. It ei
colour (cho
12 trials pe
randomized
The colour a
on each tri
verbal shor
verbal reco
remained on the screen until the response was made (first for the
colour and then for the number response) and the next trial started
after a 500 ms ITI. Given the small number of trials, rather than esti-
mating VSTM capacity as the maximum K (Cowan, 2001) across all
es, w
sized
ue ca
k et a
he W
of le
and
descr
ere
caseo
them
appe
iatel
ly w
e tr
displ
ters (
toge
In a
diffe
oher
repo
oupin
gro
ted V
perim
. In e
helpf
g a t
was
ittee
. Res
indiv
exten
ith
en K i
g ne
.811
upin
ianc
by a
d gro
ing
en th
iled
ted p
R tas
ng on
. Att
ed-m
apac
l, un
ials w
elpfu
rep
acity
44.17
enca
uent
w peiously found that whole report tasks provided a less
timate of VSTM capacity than change detection and
d that this was because of differences in attentional
order to test whether individuals differ in the atten-
egy they use during CD but not WR, we needed to
rformance in these two tasks to performance in a third
irectly manipulated attentional selection. A common
rol and measure attentional selection in this context is
tial report (PR) task (e.g. Bundesen, Pedersen, & Larsen,
an et al., 1999; Peers et al., 2005; Sperling, 1960). To
e asked participants to remember only some of the
display (target letters defined by case) while ignoring
usly presented distracters (letters of the opposite case).
ed an explicit requirement for selection and should,
e impacted by individual differences in attention. Addi-
manipulated selection difficulty by providing salient
grouping cues, which were either congruent (helpful
r incongruent (unhelpful grouping) with the top-down
tions (see Table 1, second row).
up grouping of stimuli presented during a VSTM task
entional selection towards encoding the items that are
gether (Woodman, Vecera, & Luck, 2003). In particular
als who do not naturally draw upon a selective atten-
egy to encode information into short-term memory,
tom-up grouping should, thus, improve performance
y. According to our hypothesis, the CD but not the WR
contaminated by variability in selection efficiency. We
fore, interested in how bottom-up grouping affected the
ance of participants with high and low VSTM capacity as
the CD vs. the WR task. Importantly, we could make dif-
ictions for the effect of grouping depending on whether
rticipants based on their performance in the WR or CD
ntional selection dominates CD but not WR measures
y effect of manipulating selection difficulty should be
articipants performing badly in the CD task than for
performing badly in the WR task. Specifically, we pre-
rticipants with a low CD estimate of VSTM capacity will
t from clear guidance as to which items they have to
thod. Sixteen participants were tested, of whom 11
e. Ages ranged between 20 and 58 years with a mean of
5.0). Participants were chosen on the basis of a previ-
ent, which screened people using the change detection
bed below. Two groups of eight participants whose
ity was low and high, respectively were selected from
group (N= 80, separated into high and low capacity
n-split). The two groups did not differ in their mean
1.78, p= .097; low: M= 42.5, SD= 15.4, 23–58 years, 6
h: M= 30.0, SD= 12.5, 20–57 years, 5 females) or gender
.
he change detection task, participants were shown a
2, 3, 4, 6, or 8) number of coloured discs for 150 ms.
0 ms memory interval, one of the original discs was
ther remainedunchangedorwas replacedbyadifferent
sen randomly) in 50% of trials. Participants completed
r condition, resulting in 72 trials in total, with order
and counterbalanced with Change/No Change trials.
nd position of the probed item was selected at random
al. The colour change detection task was flanked by a
t-term memory test on two spoken digits, to minimise
ding of the colours (Todd & Marois, 2004). The probe
set-siz
all set-
both tr
(Cusac
In t
arrays
ration
those
pants w
upper
typing
had dis
immed
random
the sam
arrays
get let
letters
3T3D).
in the
their c
partial
(by gr
ful” (by
estima
the ex
blocks
3T3D (
yieldin
iment
Comm
2.1.2.2
affect
ferent
KWR) w
betwe
a stron
rho= −
up gro
thevar
inated
KWR an
explain
betwe
two-ta
separa
and W
groupi
2.1.2.3
repeat
factor c
(helpfu
(PR) tr
or unh
The
of cap
14) = 1
betwe
subseq
out hoe estimated K at each set-size and then averaged across
to get mean K. Note that this estimate is affected by
pacity and the performance decrease at higher set sizes
l., 2009).
R task, we presented participants with briefly displayed
tters (either white or black) for 183 ms. The configu-
parameters of the stimulus display were identical to
ibed in the experiment above (Section 2.1.1). Partici-
asked to attend selectively to letters of either lower or
nly (“targets”) and to report the letters theyhad seenby
on a keyboard immediately after the stimulus display
ared. Response was self-paced and the next trial started
y after a response had been made. Letters were chosen
ith the constraint that they could not be repeated in
ial. Three different types of arrays were presented: two
aying either 3 target letters (WR, low load; 3T), or 6 tar-
WR, high load; 6T) alone or an array presenting 3 target
ther with 3 distracters of the opposite letter case (PR;
ddition, we modulated the presentation of the letters
rent conditions, such that two groups were formed by
ent motion and colour. The effect of interest was in the
rt trials, where the grouping could be either “helpful”
g the targets and distracters separately) or “unhelp-
uping 2 targets/distracters with 1 distracter/target). We
STM capacity using the KWR measure as described in
ent above (Section 2.1.1). Participants completed five
ach block 18 trials per condition (3T, 6T, 3T3D (none),
ul), 3T3D (unhelpful)) were presented in random order,
otal of 90 trials per condition. Approval for the exper-
given by the Cambridge Psychological Research Ethics
.
ults. We expected that bottom-up grouping should
iduals performing badly in the CD and WR task to dif-
ts. We therefore correlated memory capacity (KCD and
the effect of grouping, as measured by the difference
n the helpful and unhelpful PR conditions. This revealed
gative correlation of KCD with grouping (Spearman:
, p< .001). It is notable that the extent to which bottom-
g affects performance explains more than half (66%) of
e inKCD, suggesting that thismeasureof capacity is dom-
ttentional factors. In contrast, the correlation between
uping was not significant (Spearman: rho= −.300, n.s.),
only a fraction (9%) of the variance. The difference
e correlations was significant (Z= 2.09, p< .05), using a
Fisher’s r-to-z transform. In the following analysis, we
articipants according to their performance in the CD
ks to more thoroughly analyse the effect of bottom-up
their performance.
entional selection in low and high KCD groups. Using a
easures, mixed-effects ANOVA with between-subjects
ity (low, high) and within-subjects factor grouping type
helpful),weanalysedhowperformance inpartial report
as improved or depressed when grouping was helpful
l relative to the no-grouping condition.
eated-measures ANOVA revealed a significant effect
(F(1, 14) = 9.899, p< .01) and grouping type (F(1,
6, p< .001). In addition, there was an interaction
pacity andgrouping type (F(1, 14) = 27.660,p< .001).We
ly analysed the individual conditions in order to find
rformance varied across the two subgroups. First, the

1480
A.C.Linke
etal./N
europsychologia
49 (2011) 1476–1486
Table 1
Overview of all experiments included in this paper. CD: change detection; WR: whole report; PR: partial report.
Experiment fMRI? N Stimuli Set-sizes Task Probe (if CD) Encoding duration Maintenance duration
1. Blocked vs. intermixed probes No 16 Letters 2,3,4,6,8 WR+CD Single letter 183ms 1200ms
2. Perceptual grouping No 16 Letters (WR)
Colours (CD)
3,6 (WR and PR)
1,2,3,4,6,8 (CD)
WR, PR, CD Single colour 183ms (WR/PR)
150ms (CD)
0ms (WR/PR) 1200ms (CD)
3. Lateralised retrocueing Yes 15 Red or blue shapes 2, 4 CD Single shape 750ms 1.5–17.4 s (randomly
selected without
replacement from set of
intervals of exponentially
increasing duration)
4. Parts and wholes Yes 17 3 or 6 points 1, 2 triangles CD/2AFC Triangle 1, 6 or 11 s 1, 6 or 11 s
5. Retinotopy 1 Yes 31 Four gratings 2 CD Two gratings 300ms 1, 6 or 11 s
6. Retinotopy 2 Yes 26 Four gratings 2 CD Single grating 300ms 1, 6 or 11 s
7. Item familiarity Yes 23 Abstract shapes 1, 2 CD Single shape 1000ms 1, 6, or 11 s

A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486 1481
Fig. 2. Results group
held in memo hen p
rightmost colu
effect of hel
subgroup o
better perfo
KCD (t(7) = 9
ticipants. In
grouping co
for high KC
greater imp
the use of a
ing strategy
participants
more select
2.1.2.4. Atte
divided par
WR task (KW
for the divis
differ in ag
years, 6 fem
or gender d
from the div
participants
(2009) show
than CD es
tasks encou
expected a
formance in
repeated-m
(F(1, 14) = 6
interaction
2.1.2.5. Tar
on the prop
the PR trials
ity in the tw
main effect
F(1, 14) = 7
type (helpfu
interaction
Helpful gro
number of d
low
rtici
e abi
nd su
gh K
ry ca
intru
0.760
tion
. Dis
ty as
sele
on w
i, low
on
mbe
epor
nship
anc
in thof the partial report task. The y-scales display the effect of helpful and unhelpful
ry (KPR) and number of distracters reported. The leftmost column shows results w
mn shows results for high and low KWR groups.
pful and unhelpful grouping was compared within each
f participants. Not surprisingly, helpful grouping led to
rmance compared to unhelpful grouping for both low
.606, p< .001) and high KCD (t(7) = 7.695, p< .001) par-
terestingly, the increase in performance in the helpful
ndition was much larger for low KCD participants than
D participants (t(14) = 5.861, p< .001; see Fig. 2). This
rovement for low KCD participants after encouraging
more selective strategy suggests that it is the encod-
normally used that differs between low and high KCD
, with high KCD participants automatically adopting a
ive strategy.
ntional selection in low and high KWR groups. We also
ticipants according to their VSTM capacity during the
R) and conducted the same analyses as described above
ion by KCD. Again, the low and high KWR groups did not
e (t(14) = 1.65, p= .121; low: M= 42.1, SD= 15.8, 20–57
ales; high: M= 30.4, SD= 12.5, 22–40 years, 5 females)
istribution. Additionally, this division was independent
ision into low and high KCD groups with 6 low/high KCD
for the
n.s.) pa
and th
tion, a
and hi
memo
tracter
14) = 1
interac
2.1.2.6
capaci
tional
selecti
stimul
mance
the nu
items r
relatio
perform
mancebeing in the opposite KWR group. Since Cusack et al.
ed that WR estimates of capacity were less variable
timates (likely due to the nature of the probe in WR
raging more selection, even for low-K participants), we
smaller difference between the improvement in per-
the PR trials when splitting participants by KWR. The
easures ANOVA showed a main effect of grouping type
0.821, p< .001), but no effect of capacity (F< 1) or their
(F(1, 14) = 3.575, n.s.), thus confirming our prediction.
get selectivity. We also ran the same ANOVA as above
ortion of distracters reported during the probe phase of
. This allowed us to directly investigate target selectiv-
o groups of participants. We again found a significant
of the between-subjects factor CD capacity (low or high,
.627, p< .05) and the within-subjects factor grouping
l or unhelpful, F(1, 14) = 21.030, p< .001), as well as an
between these two factors (F(1, 14) = 13.843, p< .005).
uping decreased distracter intrusions relative to the
istracters reported in the unhelpful grouping condition
support the
mines perfo
used. More
in CD are a
out irreleva
those with
are, hence,
2.1.3. Beha
The two
Cusack et
detection
attentional
ulated the t
report and c
depression
change det
of suboptim
performanc
ment maniing relative to no grouping, in terms of estimated number of targets
articipants are grouped into high and low KCD subgroups, while the
KCD (t(7) = 4.435, p< .005), but not high KCD (t(7) = 1.234,
pants. This again suggests a relationship between KCD
lity to select relevant and suppress irrelevant informa-
pports the notion that the difference between low KCD
CD participants is one of attentional selection and not
pacity per se. In contrast, for WR, an ANOVA on the dis-
sions showed only a main effect of grouping type (F(1,
, p< .005), while the main effect of group (F< 1), and the
(F< 1) were again not significant.
cussion. The data show that participants with low
assessed by change detection suffer from poor atten-
ction during the encoding phase. When attentional
as aided by helpful bottom-up grouping of the relevant
-KCD participants significantly improved in perfor-
the PR task. This effect was present in terms of both
r of items in memory and the number of distracter
ted, a direct measure of mis-selection. Importantly, the
between attentional selection and improvement in
e was only seen when grouping participants by perfor-
e CD task, but not the WR task. Together, these results
idea that what differs between participants and deter-
rmance in CD tasks is the encoding strategy naturally
specifically, individuals with high capacity measures
ble to adopt a selective encoding strategy and filter
nt items using top-down executive resources, whereas
low capacity measures show suboptimal selection and
greatly helped by bottom up grouping cues.
vioural experiments—summary
behavioural experiments presented here, along with
al. (2009), provide converging evidence that change
measurements of VSTM capacity are affected by
processesduringencoding. Thefirst experimentmanip-
ype of response probe presented, by intermixing whole
hange detection probes. This intermixing alleviates the
in performance seen at higher set sizes during pure
ection, suggesting that encoding strategy in the form
al attentional selection is responsible for decreased
e in change detection paradigms. The second experi-
pulated the grouping of items in a partial report task

1482 A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486
Table 2
ROIs included in the meta-analysis.
ROI Full description Peak coordinates (MNI) Source
Superior IPS Superior intraparietal sulcus −21, −70, 42 Xu and Chun (2006)
23, −
Inferior IPS −21,
26, −
“Silver” IPS −23,
23, −
MD-IPS −37,
37, −
MD-IFS −42,
42, 2
LOC −44,
42, −
MFG −42,
48, −
Coordinates ar l2mni
and showed
tively relate
This sugges
attentional
therefore, s
verbal IQ (s
detection ta
that aims a
high set-siz
2.2. Meta-a
2.2.1. Intro
While th
gest that att
of short-ter
strategy us
an impact o
maintenanc
have focuse
vant during
or maintena
is possible
tion during
this informa
be hypothe
individuals
not vary, in
during enco
activation d
ing it possib
encode and
capacity.
We ther
mates of V
encoding an
3–7 in Tabl
regions of in
ory and cog
differed acr
results.
2.2.2. Meth
Five div
were includ
change dete
Forced-Cho
the experim
antly
char
nanc
uish
. Exp
corre
tory
ed c
ven
(CPR
h stu
Basi
/ww
The
imat
rt of i
ed fr
the
lishe
enco
with
e w
wer
ith r
ions)
wer
heen
the c
cross
(8 fe
xper
d ha
AfterInferior intraparietal sulcus
Intraparietal sulcus (Silver)
Intraparietal sulcus (Multi Demands Network)
Inferior frontal sulcus (Multi Demands Network)
Lateral occiptal cortex
Middle frontal gyrus
e given in MNI152 space, converted from Talairach space where necessary using ta
that the magnitude of the effect of grouping is nega-
d to capacity in change detection but not whole report.
ts that change detection estimates are related to the
strategy used by the participant at encoding. The data,
upport our hypothesis that individuals with lower non-
tudy 1) or with measures of low capacity in change
sks (study 2) normally adopt an attentional strategy
t encoding all items in a display without prioritizing at
es, when not all items can be remembered.
nalysis of imaging experiments
duction
e behavioural results discussed above (Section 2.1) sug-
entional strategiesduringencoding influenceestimates
m memory capacity, it remains unclear whether the
ed while encoding information into memory also has
n how the items are retained in memory during the
e phase of a change detection task. Different studies
d separately on whether attentional selection is rele-
encoding (Awh et al., 2006; McNab & Klingberg, 2008)
nce (Curtis & D’Esposito, 2003; Kane & Engle, 2002). It
that individuals using more efficient attentional selec-
encoding also show differences in how they maintain
tion in short-term memory. On the other hand, it could
sized that while encoding strategy might differ between
,howtheencoded itemsaremaintained inmemorydoes
dicating that attentional selection only has an impact
ding. Neuroimaging allows us to directly differentiate
uring the different phases of short-term memory, mak-
le to investigate whether people differ in the way they
maintain information and how this is related to their
efore conducted a meta-analysis to assess whether esti-
Import
design
mainte
disting
2.2.2.1
mal or
no his
inform
was gi
mittee
for eac
Visual
(http:/
1997).
approx
the sta
recruit
In
unpub
a first
sented
retrocu
probes
data w
condit
period
found t
due to
aged a
tested
In E
dots an
shape.STM capacity were related to activation during the
d maintenance phase of five VSTM tasks (see studies
e 1) that we have recently conducted. We focused on
terest (ROIs) commonly activated in short-term mem-
nitively demanding tasks (Table 2). The stimuli used
oss the experiments allowing some generalization of
ods
erse VSTM studies with a total of 112 participants
ed in the meta-analysis. Participants had to perform a
ction task in four of these studies and a 2-Alternative-
ice task in the fifth. Table 1 provides an overview of
ental designs and stimuli used in the different studies.
task to inde
data with re
tions. Beta v
the encodin
conditions.
years old, M
Experim
during sho
four sectors
instructed
had to pay a
Table 1 for
changed du
the final co56, 46
−89, 24 Xu and Chun (2006)
84, 28
−80, 38 Silver, Ress, and Heeger (2005)
80, 38
−56, 41 Duncan and Owen (2000)
56, 41
23, 28 Duncan and Owen (2000)
3, 28
−71, 5 Xu and Chun (2006)
69, 0
−10, 52 McNab and Klingberg (2008)
8, 42
(http://imaging.mrc-cbu.cam.ac.uk/downloads/MNI2tal/tal2mni.m).
, the experiments shared many common experimental
acteristics. They all included a clearly defined encoding,
e and probe phase that could be modelled separately to
activity due to these different stages of the VSTM task.
eriments. All subjects in all experiments reported nor-
cted-to-normal visual acuity, normal colour vision, and
of psychological or neurological impairment. All gave
onsent and were paid for taking part. Ethical approval
by the Cambridge Psychology Research Ethics Com-
EC) or the Local Research Ethics Committee (LREC)
dy. In all experiments, stimuli were presented using
c .NET except for Experiment 7, which used Matlab
w.themathworks.com) and the Psychtoolbox (Brainard,
experiments were carried out over a time period of
ely 26 months (with on average 6.37 months between
ndividual experiments, SD= 2.87) andparticipantswere
om a large volunteer panel.
first neuroimaging study (Experiment 3; Mitchell,
dPhDthesis), stimuliwereabstract silhouettes andafter
ding and maintenance period participants were pre-
a retrocue intended to reduce the initial load. This
as followed by another maintenance period before the
e presented. A General Linear Model was fit to the
egressors for the each of the task phases (across load
. Only beta values corresponding to the first encoding
e included in the meta-analysis in order not to con-
codingprocesseswith reactions andalteredprocessing
ue. Betas and estimates of memory capacity were aver-
the two maintenance phases. Fifteen participants were
males, 19–43 years old, M= 28.5, SD= 8.3).
iment 4, participants were presented with three or six
d to mentally manipulate them to form the outline of a
a variable maintenance period, they performed a 2AFC
ntify this shape. A General Linear Model was fit to the
gressors for the each of the task phases and load condi-
alues as well as the estimate of memory capacity during
g and maintenance period were averaged across load
Seventeen participants were tested (10 females, 19–40
= 22.1, SD= 5.6).
ents 5 and 6 investigated retinotopic representations
rt-term memory. Participants had to attend to two of
. The sectors were colour-coded and participants were
before the start of the experiment which colour they
ttention to. Each of the sectors contained gratings (see
an example of the stimuli) whose pattern continuously
ring the encoding phase. Participants had to maintain
nfiguration in short-term memory for a variable time

A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486 1483
period. The two experiments only differed in how the sectors were
oriented and how the probe was presented (two vs. one sectors,
respectively). Seventeen participants were tested in Experiment 5
(8 females, 22–40 years old, M= 25.6, SD= 4.5). Fourteen of the par-
ticipants were scanned a second time with scanning sessions at
least two weeks apart and were treated as separate participants in
the analysis2. Twenty-six participants were tested in Experiment 6
(16 females
The last
ityonchang
either one
maintenanc
frequently a
meta-analy
ber of high f
the encodin
were tested
2.2.2.2. Ima
Siemens Tim
MPRAGE se
image, at 1
echo-plana
similar para
matrix 64 ×
in all but t
TR = 2.150 s
Data
Departmen
http://www
dure. The
timing, co-r
algorithm)
Institute (M
smoothed w
kernel and
pass, 128 s
a general l
to each ex
haemodyna
these expe
and either t
the range o
encoding w
of them lon
of sequent
Passingham
Sakai and P
us to separ
epochs. All
out residua
In orde
Dependent
performanc
analysis usi
ROIs were c
discussed a
these regio
tasks (four
“Silver” IPS
lateral occi
overview).
2 Univariate
of the particip
OIs sh
ies. Al
llowin
re dra
. IPS
rom X
al. (20
from t
capac
(MFG
sho
ere
y rep
ar (Brett, Anton, Valabregue, & Poline, 2002).
each study, task phase and ROI, mean beta values for the
t GLM regressors were extracted from the original SPM
is. Only the encoding and maintenance period were taken
count, averaging across set sizes and conditions within the
ual experiments. Memory capacity was estimated for the
rticipants using Cowan’s (2001) formula (K=N× (H− F) and
(2 × PC − 1) for the 2AFC task; where N= number of items in
oding display, H= proportion of hits, FA = proportion of false
, PC = percent correct). Mean signal strength (mean beta val-
r the relevant regressors) and K values were standardised
each study to account for stimulus differences and were
ted using Pearson’s correlation.
Results
ividual univariate analyses showed that the regions involved
encoding and maintenance (vs. baseline) of these different
ubstantially overlap (see Fig. 4). Importantly, many of the
s commonly involved in short-term memory showed signif-
ctivity during the encoding and maintenance phase of all five
.
oss the 112 participants, with inter-study differences
ed by standardisation of the BOLD and K estimates, mem-
acitypositively correlatedwithencodingactivity in superior
.300, p< .0053), MD-IPS (r= .286, p< .005), Silver IPS (r= .218,
ificance values reported here are uncorrected. Except for Silver IPS and MFG
survive Bonferroni correction for multiple comparisons at ˛= 0.05., 19–35 years old, M= 26.2, SD= 4.6).
study (Experiment 7) investigated the effect of familiar-
edetectionperformance. Participantshad to remember
or two abstract grey-scale shapes during a variable
e period. Two of the 26 shapes were presented very
nd were, thus, very familiar to the participants. For the
sis, beta values were averaged across conditions (num-
requency items in the encoding display) and set-size for
g and maintenance phase. Twenty-three participants
(15 females, 18–38 years old, M= 25.7, SD= 5.4).
ging methods and analysis. All data were acquired on a
Trio 3 Tesla scanner using a 12 channel head coil. An
quence acquired a whole brain T1-weighted structural
mm3 resolution. The functional data was acquired with
r imaging (EPI) sequences. All the EPI sequences shared
meters (TR = 2 s, TE = 30 ms, flip angle = 78◦, 32 slices of
64 with a 25% gap, voxel size 3 mm× 3 mm × 3.75 mm)
he two retinotopy experiments (5 and 6 in Table 1;
, voxel size 2.42 mm × 2.42 mm × 3 mm).
were pre-processed using SPM5 (Wellcome
t of Imaging Neuroscience, London, UK;
.fil.ion.ucl.ac.uk/spm) following a standard proce-
EPI data were corrected for head motion and slice
egistered to the structural (using a mutual information
and affine normalised to the Montreal Neurological
NI) template brain. The resulting images were spatially
ith a 10–14 mm full-width half-maximum Gaussian
temporally filtered to reduce signal drift using a high-
cut-off. Each voxel in these EPI images was fitted with
inear model composed of a set of regressors specific
periment, which were convolved with the canonical
mic response function (HRF). Importantly, in all of
riments we varied the duration of the maintenance
he inter-trial-interval (ITI) or encoding epochs (within
f 1–17.4 s; except for Experiment 7 in which ITI and
ere constant). Jittering with variable intervals (some
g) has been shown repeatedly to achieve separation
ial events in event-related designs (e.g. Rowe and
(2001), using a delay period jitter of 8.5–17.5 s or
assingham (2003), using a jitter of 4–12 s) and allowed
ately model the encoding, maintenance and response
experiments additionally included regressors to model
l movement artefacts.
r to assess whether Blood-Oxygenation-Level-
(BOLD) activity was dependent on an individual’s
e in VSTM tasks we conducted a region of interest
ng the same ROIs for the five different studies. Seven
hosen based on indications from the previous studies
bove (also see Cusack, Mitchell, & Duncan, 2010) that
ns are reliably activated during short-term memory
intraparietal sulcus regions: superior IPS, inferior IPS,
, MD-IPS; an inferior frontal sulcus (MD-IFS) and a
pital cortex (LOC) region; see Fig. 3 and Table 2 for an
Additionally we included one region (Middle Frontal
analysis had shown that it didnotmakeadifferencewhether sessions
ants scanned twice were combined or not.
Fig. 3. R
ous stud
in the fo
sulcus) a
cus), sup
drawn f
Silver et
Activity
memory
Gyrus
during
ROIs w
activit
marsb
For
relevan
analys
into ac
individ
112 pa
K=N×
the enc
alarms
ues fo
within
correla
2.2.3.
Ind
during
tasks s
region
icant a
studies
Acr
remov
ory cap
IPS (r=
3 Sign
all ROIsown to be commonly activated by short-term memory tasks in previ-
l ROIsweredefinedas10 mmspheres aroundpeakactivation reported
g studies: MD regions (IFS: inferior frontal sulcus; IPS: intraparietal
wn from Duncan and Owen (2000). inf. IPS (inferior intraparietal sul-
(superior intraparietal sulcus) and LOC (lateral occipital cortex) are
u and Chun (2006). Silver-IPS (intraparietal sulcus) is drawn from
05) and the MFG region is drawn from McNab and Klingberg (2008).
hese ROIs was extracted and correlated with estimates of short-term
ity.
)) that has been implicated in controlling attention
rt-term memory tasks (McNab & Klingberg, 2008). All
defined as spheres of 10 mm radius around the peak
orted in the respective studies (see Table 2), using

1484 A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486
Fig. 4. Region
analysis durin
respective vox
study showed
t-maps (FDR c
p< .05), MD
dle Frontal
not show si
Fig. 5). In or
during main
actual diffe
the signific
an adaptati
Rubin, 1992
ent this wou
maintenanc
this task ph
that showed
but not ma
p< .005; Silv
Z= 4.10, p< .001; MFG: Z= 4.24, p< .001) indicating that the absence
of correlations with performance during the maintenance period
was not due to noise.
These results suggest that differences in BOLD activation dur-
ing the encoding of items into short-term memory are associated
with how well individuals perform in the task. This is in agree-
ment with the results of the behavioural studies discussed earlier
and points towards a strong influence of encoding strategy on esti-
mates of short-term memory capacity. Additionally, the finding
that activity in superior IPS but not inferior IPS correlates with
performance shows a parallel with Xu and Chun’s (2009) differen-
tiation between “object individuation” in inferior IPS and “object
identification” in superior IPS. According to their findings, “object
individuation” is fixed to about four items and is independent
of encoding demands and complexity while “object identifica-
tion” is thought to be flexible and dependent on the demands of
the task. This difference in flexibility is also supported by recent
data examining the functional connectivity of these regions across
a broad range of tasks. While the inferior IPS shows consistent
strong connectivity with perceptual regions (the LOC), the supe-
rior IPS sometimes connects strongly with perceptual regions and
sometimes with frontal regions (Cusack & Owen, 2008). It seems
able to assume that attentional selection determines how
f the
ther
tha
em
from
s.
cussi
ividu
t to p
d us
entio
rong
Fig. 5. Correla
activity (y-axis
of the ROIs dus activated by the five neuroimaging studies included in the meta-
g encoding and maintenance. Red indicates that activity in the
els was found in all five studies while blue indicates that only one
activity in that region. The figure was created by adding the binary
orrected, p< 0.05) for the five different studies.
-IFS (r= .286, p< .005), LOC (r= .279, p< .005) and Mid-
Gyrus (r= .197, p< .05) while maintenance activity did
gnificant correlations with K in any of these regions (see
der to test whether the lack of significant correlations
tenance was due to a higher degree of noise or reflected
rences in slope, we compared the difference between
reason
manyo
are, fur
finding
term m
items
region
3. Dis
Ind
subjec
assesse
the att
task stant encoding and the maintenance correlations using
on of the Williams–Hotelling test (Meng, Rosenthal, &
). Specifically, if the slopes were not significantly differ-
ld imply that the lack of significant correlations during
e is simply due to higher variability in the data during
ase. Slopes were significantly different for all of the ROIs
significant correlations with capacity during encoding
intenance (sup. IPS: Z= 1.93, p< .05; MD-IPS: Z= 2.73,
er IPS: Z= 3.47, p< .001; MD-IFS: Z= 2.68, p< .005; LOC:
upon Cusac
intelligence
with an ad
has been a
tion and sho
Awh et al.,
trol takes e
(Awh et al.,
Engle, 2002
tion plots for all ROIs (standardized values depicted) during encoding (first row) and mai
) and estimates of short-term memory capacity (x-axis) are marked by an asterisk (*p< 0
ring the maintenance phase of the tasks but are shown here for comparison.itemsreach the “object identification” stage. The results
more, in accordance with McNab and Klingberg’s (2008)
t activity in the middle frontal gyrus predicts short-
ory capacity and acts as a control to filter irrelevant
being encoded and passed on to be stored in parietal
on
al differences in short-termmemorycapacityhavebeen
sychological research for decades and have often been
ing change detection tasks. Here we have shown that
nal strategy used during the encoding phase of the
ly influences change detection performance. This builds
k et al.’s (2009) finding that individual differences in
donot correlatewithmemory capacity itself, but rather
ditional cognitive process during change detection. It
ssumed for a number of years that attentional selec-
rt-term memory are strongly related mechanisms (e.g.
2006). It was unclear, however, when attentional con-
ffect in a short-term memory task, during encoding
2006; McNab & Klingberg, 2008), maintenance (Kane &
; Curtis & D’Esposito, 2003) or both. We have addressed
ntenance (second row). Significant correlations between mean BOLD
.05, **p< 0.005, ***p< 0.001). Correlations were not significant for any

A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486 1485
this question in two behavioural experiments and by conducting
a meta-analysis of five imaging studies recently carried out in our
laboratory.
Our results indicate that individuals differ in the strategy
they adopt
with low p
but instead
sented info
consequent
information
memory (A
viduals diff
irrelevant i
ory capacity
Zwilling an
with childr
that attenti
the older c
older childr
cantly drop
that young
irrelevant in
four items. T
uals showin
is possible t
adaptive st
that they le
A questi
particular a
items to rep
CD the part
effective str
of trials wh
be known a
age broade
predictions
leads to pa
ory with a l
straightforw
in which m
number of i
predict wor
greater num
A relate
choosing th
does the na
matic selec
a sign of p
individuals
during diffi
ing particip
the need fo
forming ba
helpful bot
urally use a
do so, in tu
possible tha
uals lies in t
that they ar
An issue
time availab
Chen, & Jian
apply effec
more rapid
activity du
(Vogel & M
ongoing encoding of the information, especially since encoding
time as well as the delay period in these experiments is short
(100 ms and 900 ms, respectively in Vogel & Machizawa, 2004).
Futureexperiments could investigatewhether encoding time inter-
ith th
intr
ctivi
divid
form
(Vog
ble t
e sam
l, 20
at w
wer
less
acro
gher
the
Mar
gh it
than
ve s
ilitie
cours
ring
le, ex
onal
d ne
ctivit
ual d
Olese
ingha
pite
oura
robed
aged
ment
ion t
prob
en sh
at w
rform
obe
cued
(Jian
gan,
Woo
s littl
ff we
t set
emb
l bra
one
anc
rall,
how
t-ter
ced
elec
repo
anc
ting
res o
ct of
ancduring the encoding phase of change detection tasks,
erforming participants not selecting a subset of items
trying to encode everything, even when the pre-
rmation clearly exceeds capacity, and performance
ly suffers. This is in accordance with suggestions that
is filtered for relevancy before entering short-term
wh et al., 2006; McNab & Klingberg, 2008) and that indi-
er in their attentional ability to separate relevant from
tems. Drawing upon the finding that short-term mem-
increases during childhood, Cowan, Morey, AuBuchon,
d Gilchrist (2009) conducted a change detection study
en (7–8 and 12–13 years old) and adults and showed
onal efficiency of seven-year-olds was equal to that of
hildren and adults for low set-sizes. Compared to the
en and adults, however, attentional efficiency signifi-
ped for higher set-sizes. Cowan et al. (2009) concluded
children show a similar attentional ability to filter out
formation as adults as long as set-size does not exceed
his result parallels our observation of lower-IQ individ-
ga significantdrop inperformanceathigher set-sizes. It
hat young children, like lower-IQ individuals, use a mal-
rategy to encode information into short-term memory
arn to adjust when growing older.
on that remains is which aspect of the tasks encourages
ttentional strategies. In WR, individuals choose which
ort, but in CD, a single item is probed at random. If in
icipant encodes only a subset of the sample (the most
ategy at high set-sizes) then there will be a proportion
en one of the other items is probed, and nothing will
bout it. If this is found discouraging, it might encour-
r but shallower encoding. This would lead to different
by different VSTM models. If a broader attentional set
rtial encoding of many items, then a model of mem-
imited number of slots (e.g., Zhang & Luck, 2008) would
ardly predict a loss of information. In contrast, a model
emory resources can be divided between an arbitrary
tems (e.g., Bays & Husain, 2008), would not necessarily
se performance as a result of shallower encoding of a
ber of items.
d question is how much control individuals have in
eir attentional strategy: Is strategy consciously chosen;
ture of the task encourage the unconscious and auto-
tion of a particular strategy; or is poorer performance
oorer strategic control? Kane et al. (2007) found that
with low memory capacity reported mind-wandering
cult tasks considerably more often than better perform-
ants. By using a partial report paradigm that imposed
r selection, we could show that those individuals per-
dly in CD significantly improved in the presence of
tom-up cues to selection. Individuals who do not nat-
selective encoding strategy can thus be encouraged to
rn improving their performance considerably. It is also
t the difference between lower- and higher-IQ individ-
he ability to switch attentional strategies when noticing
e not performing well in a task.
not examined in the current study is the effect of the
le for encoding (cf. Bays, Catalao, & Husain, 2009; Eng,
g, 2006). It might be, for example, that it takes time to
tive selection, and that in some people this process is
than in others. ERP studies of VSTM show sustained
ring the maintenance period that relates to capacity
achizawa, 2004). It is possible that this partly reflects
acts w
It is
brain a
that in
vant in
varies
more a
into th
Lengye
tion th
with lo
due to
similar
and hi
This is
Todd &
althou
rather
that ha
possib
Of
ity du
examp
attenti
tion ha
then a
individ
2009;
& Pass
Des
behavi
was p
encour
experi
detect
of the
has be
such th
ory pe
the pr
single
probes
& O’Re
sizes (
ing ha
drop o
effect a
of rem
parieta
2007),
perform
Ove
study s
of shor
influen
tional s
whole
perform
sugges
measu
ful effe
performe effects we have described.
iguing that individual differences were associated with
ty during encoding, and not maintenance. It might be
uals’ memory capacity is similar, but that how well rele-
ation is selected to be encoded into short-term memory
el et al., 2005). Alternatively, some individuals may be
o “compress” – encode items efficiently – and so fit more
e underlying memory capacity (cf. Orbán, Fiser, Aslin, &
08). Thishypothesismakes thecounter-intuitivepredic-
hen capacity limits have not been reached, individuals
capacitywould showmoreactivityduringmaintenance,
efficient compression. At higher load, activity would be
ss participants. Hence, the difference between lower
load would be greater in higher-capacity individuals.
pattern reported in the literature (Linden et al., 2003;
ois, 2005; Vogel & Machizawa, 2004; Vogel et al., 2005),
is usually attributed to greater activity for high load
less activity for low load. We are not aware of studies
earched for the measure that would discriminate these
s.
e, there could also be circumstances in which activ-
maintenance would track memory performance. For
periments 4–7 did not impose strong requirements for
selection during the maintenance period; if informa-
eded to be maintained in the face of distracting stimuli,
y during this task phase may have also correlated with
ifferences in performance (Clapp, Rubens, & Gazzaley,
n, Macoveanu, Tegner, & Klingberg, 2007; Sakai, Rowe,
m, 2002).
the predominant effects of encoding strategy, our first
l study also showed that the way in which memory
had a residual effect even when participants were
to adopt the same encoding strategy throughout the
. It is, therefore, reasonable to assume that change
asks are vulnerable to interference by characteristics
e (Makovski, Sussma, & Jiang, 2008). For instance, it
own that configural information is encoded in VSTM,
hen the context is disrupted during the probe, mem-
ance decreases. Presenting all remembered items in
display improves performance relative to showing a
item or disrupting the characteristics of the un-cued
g, Olson, & Chun, 2000; Vidal, Gauchou, Tallon-Baudry,
2005). Such effects have greater impact at higher set
dman et al., 2003), suggesting that bottom-up group-
e effect below capacity. This is in accordance with the
saw in our change detection task, which only came into
sizes above capacity. Interestingly, bottom-up grouping
ered items differentially alters brain activity in distinct
in areas responsible for VSTM maintenance (Xu & Chun,
of these being the superior IPS region correlating with
e in our meta-analysis.
our current results replicated and extended our earlier
ing that individual differences in alternative measures
m memory – change detection and whole-report – are
by distinct cognitive components, in particular, atten-
tion during encoding. Intermixing change detection and
rt probes during a VSTM task improved participants
e on change detection without affecting whole report,
that the attentional set used during encoding affects
f change detection. In a partial report task, the help-
bottom-up grouping during encoding selectively aided
e of participants with a low change detection estimate

1486 A.C. Linke et al. / Neuropsychologia 49 (2011) 1476–1486
of VSTM capacity, indicating that change detection measures of
VSTM are dominated by individual differences in attentional selec-
tivity during encoding. Performance in whole report tasks, on the
other hand, is much less affected by encoding strategy. Further-
more, individualdifferences inVSTMcapacity correlatedwithBOLD
signal during encoding but not maintenance. Taken together, these
results sup
assessed by
differences
nent that co
capacity its
Acknowled
We wou
ments and f
running Exp
References
Ackerman, P. L
The same
Awh, E., & Joni
ing memo
Awh, E., Vogel
memory. N
Bays, P. M., Ca
memory is
Bays, P. M., &
resources
Bleckley, M. K
Individual
cation. Psy
Brainard, D. H
Brett, M., Anto
usinganSP
mapping o
Neuroimag
Bundesen, C.,
tion from
of Experim
Cusack, R., Mit
tion switc
Neuroscien
Clapp, W. C., R
disruption
Conway, A. R.
its relation
Cowan, N. (20
tion of me
Cowan, N., Mo
Seven-yea
loaded. De
Curtis, C. E.,&D
working m
Cusack, R., Leh
not visual
Bulletin &
Cusack, R., &
but not fro
method. In
Duncan, J., &
lobe recru
475–483.
Duncan, J., Bu
(1999). Sys
Psychology
Edin, F., Klingb
anism for
National A
Eng,H.Y., Chen
visual stim
Engle, R. W. (2
tions in Psy
Heitz, R. P., R
Engle, R. W
intelligenc
136.
Jarrold, C., & Towse, J. N. (2008). Individual differences in working memory. Neuro-
science, 139, 39–50.
Jiang, Y.,Olson, I. R., &Chun,M.M. (2000).Organizationof visual short-termmemory.
Journal of Experimental Psychology: Learning,Memory, andCognition,26, 683–702.
Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex in working-memory
capacity, executive attention, and general fluid intelligence: An individual-
differences perspective. Psychonomic Bulletin & Review, 9, 637–671.
. J., Bl
ntion
and
ral, 13
. J., Br
7). Fo
orking
621.
, P. C.
ory c
. S., Gi
lap of
pariso
D. E. J
l. (200
ation
–153
i, T.,
emo
hology
. S., B
7). Co
oimag
F., & K
orking
., Rose
nts. Ps
, D. J.
ed num
ge.
, D. J.,
etal co
ex, 18,
P. J., M
orkin
–105
., Fise
ks by
nited
., Lud
5). At
–148
B., &
vity in
ory. N
, & Pas
s. Na
., Row
46 cr
. A.,
ntion i
, G. (1
Monog
., & M
erior p
J., & M
rence
Neuro
., Gau
ion in
256.
. K., &
s in v
. K., M
vidual
503.
an, G.
al wor
Chun,
mem
Chun
of the
6–18
Chun,
gnitiv
., & L
king mport the hypothesis that variations in performance as
change detection tasks are primarily determined by
in encoding strategy and it is this attentional compo-
rrelates with measures of IQ, not short-term memory
elf.
gment
ld like to thank Michele Veldsman for her helpful com-
or acquiring data for Experiment 1 and Ciara Greene for
eriment 7.
., Beier,M. E., &Boyle,M.O. (2005).Workingmemoryand intelligence:
or different constructs? Psychology Bulletin, 131, 30–60.
des, J. (2001). Overlapping mechanisms of attention and spatial work-
ry. Trends in Cognitive Sciences, 5, 119–126.
, E. K., & Oh, S.-H. (2006). Interactions between attention and working
euroscience, 139, 201–208.
talao, R. F. G., & Husain, M. (2009). The precision of visual working
set by allocation of a shared resource. Journal of Vision, 9, 1–11.
Husain, M. (2008). Dynamic shifts of limited working memory
in human vision. Science, 321, 851–854.
., Durso, F. T., Crutchfield, J. M., Engle, R. W., & Khanna, M. M. (2003).
differences in working memory capacity predict visual attention allo-
chonomic Bulletin & Review, 10, 884–889.
. (1997). The psychophysics toolbox. Spatial Vision, 10, 433–436.
n, J.-L., Valabregue, R., & Poline, J.-B. (2002). Region of interest analysis
Mtoolbox. InPresentedat the8th international conferenceon functional
f the human brain, June 2–6 Sendai, Japan,. Available on CD-ROM in
e, 16
Pedersen, L. F., & Larsen, A. (1984). Measuring efficiency of selec-
briefly exposed visual displays: A model for partial report. Journal
ental Psychology. Human Perception and Performance, 10, 329–339.
chell, D. J., & Duncan, J. (2010). Discrete object representation, atten-
hing, and task difficulty in the parietal lobe. Journal of Cognitive
ce, 22, 32–47.
ubens, M. T., & Gazzaley, A. (2009). Mechanisms of working memory
by external interference. Cerebral Cortex, 20, 859–872.
A., Kane, M. J., & Eagle, R. W. (2003). Working memory capacity and
to general intelligence. Trends in Cognitive Sciences, 7, 547–552.
01). The magical number four in short-term memory: A reconsidera-
ntal storage capacity. The Behavioral and Brain Sciences, 24, 87–114.
rey, C. C., AuBuchon, A. M., Zwilling, C. E., & Gilchrist, A. L. (2009).
r-olds allocate attention like adults unless working memory is over-
velopmental Science, 13, 120–133.
’Esposito,M. (2003). Persistentactivity in theprefrontal cortexduring
emory. Trends in Cognitive Sciences, 7, 415–423.
mann,M., Veldsman,M., &Mitchell, D. J. (2009). Encoding strategyand
working memory capacity correlates with intelligence. Psychonomic
Review, 16, 641–647.
Owen, A. M. (2008). Distinct networks of connectivity for parietal
ntal regions identified with a novel alternative to the “resting state”
Proceedings of Cognitive Neurosciene Society San Francisco.
Owen, A. M. (2000). Common regions of the human frontal
ited by diverse cognitive demands. Trends in Neurosciences, 26,
ndesen, C., Chavda, S., Olson, A., Humphreys, G. W., & Shibuya, H.
tematic analysis of deficits in visual attention. Journal of Experimental
, 128, 450–478.
erg, T., Johansson, P., McNab, F., Tegner, J., & Compte, A. (2009). Mech-
top-down control of working memory capacity. Proceedings of the
cademy of Sciences of the United States of America, 106, 6802–6807.
,D.,& Jiang,Y. (2006).Visualworkingmemory for simpleandcomplex
uli. Psychonomic Bulletin & Review, 12, 1127–1133.
002). Working memory capacity as executive attention. Current Direc-
chological Science, 11, 19–23.
edick, T. S., Hambrick, D. Z., Kane, M. J., Conway, A. R. A., &
. (2006). Working memory, executive function, and general fluid
e are not the same. The Behavioral and Brain Sciences, 29, 135–
Kane, M
atte
span
Gene
Kane, M
(200
of w
614–
Kyllonen
mem
LaBar, K
over
com
Linden,
et a
soci
1518
Makovsk
ing m
Psyc
Mayer, J
(200
Neur
McNab,
to w
Meng, X
ficie
Mitchell
limit
brid
Mitchell
pari
Cort
Olesen,
to w
1047
Orbán, G
chun
the U
Peers, P
(200
1469
Rowe, J.
Acti
mem
Sakai, K.
ation
Sakai, K
area
Silver, M
atte
Sperling
ogy
Todd, J. J
post
Todd, J.
diffe
ioral
Vidal, J. R
mat
244–
Vogel, E
ence
Vogel, E
indi
500–
Woodm
visu
Xu, Y., &
term
Xu, Y., &
ings
1876
Xu, Y., &
in Co
Zhang, W
woreckley, M. K., Conway, A. R. A., & Engle, R. W. (2001). A controlled-
view of working memory capacity: Individual differences in memory
the control of visual orienting. Journal of Experimental Psychology:
0, 169–183.
own, L. H., McVay, J. C., Silvia, P. J., Myin- Germeys, I., & Kwapil, T. R.
r whom the mind wanders, and when: An experience-sampling study
memory and executive control in daily life. Psychological Science, 18,
, & Christal, R. E. (1990). Reasoning ability is (little more than) working
apacity?!. Intelligence, 14, 389–433.
telman, D. R., Parrish, T. B., & Mesulam, M. M. (1999). Neuroanatomic
working memory and spatial attention networks: A functional MRI
n within subjects. Neuroimage, 10, 695–704.
., Bittner, R. A., Muckli, L., Waltz, J. A., Kriegeskorte, N., Goebel, R.,
3). Cortical capacity constraints for visual working memory: Dis-
of fMRI load effects in a fronto-parietal network. Neuroimage, 20,
0.
Sussma, R., & Jiang, Y. V. (2008). Orienting attention in visual work-
ry reduces interference from memory probes. Journal of Experimental
: Learning, Memory, and Cognition, 34, 369–380.
ittner, R. A., Nikolic, D., Bledowski, C., Goeble, R., & Linden, D. E. J.
mmon neural substrates for visual working memory and attention.
e, 36, 441–453.
lingberg, T. (2008). Prefrontal cortex and basal ganglia control access
memory. Nature Neuroscience, 11, 103–107.
nthal, R., & Rubin, D. (1992). Comparing correlated correlation coef-
ychology Bulletin, 111, 172–175.
(2009). The role of human posterior parietal cortex in representing a
ber of objects. Unpublished PhD thesis. University of Cambridge: Cam-
& Cusack, R. (2008). Flexible, capacity-limited activity of posterior
rtex in perceptual as well as visual short-term memory tasks. Cerebral
1788–1798.
acoveanu, J., Tegner, J., & Klingberg, T. (2007). Brain activity related
g memory and distraction in children and adults. Cerebral Cortex, 17,
4.
r, J., Aslin, R. N., & Lengyel, M. (2008). Bayesian learning of visual
human observers. Proceedings of the National Academy of Sciences of
States of America, 105, 2745–2750.
wig, C., Rorden, C., Cusack, R., Bonfiglioli, C., Bundesen, C., et al.
tentional functions of parietal and frontal cortex. Cerebral Cortex, 15,
4.
Passingham, R. E. (2001). Working memory for location and time:
prefrontal area 46 relates to selection rather than maintenance in
euroimage, 14, 77–86.
singham, R. E. (2003). Prefrontal interactions reflect future task oper-
ture Neuroscience, 6, 75–81.
e, J. B., & Passingham, R. E. (2002). Active maintenance in prefrontal
eates distractor-resistant memory. Nature Neuroscience, 5, 479–484.
Ress, D., & Heeger, D. J. (2005). Topographic maps of visual spatial
n human parietal cortex. Journal of Neurophysiology, 94, 1358–1371.
960). The information available in brief visual presentations. Psychol-
raphs, 74, 1–29.
arois, R. (2004). Capacity limit of visual short-term memory in human
arietal cortex. Nature, 428, 751–754.
arois, R. (2005). Posterior parietal cortex activity predicts individual
s in visual short-term memory capacity. Cognitive, Affective & Behav-
science, 5, 144–155.
chou, H. L., Tallon-Baudry, C., & O’Regan, J. K. (2005). Relational infor-
visual short-term memory: The structural gist. Journal of Vision, 5,
Machizawa, M. G. (2004). Neural activity predicts individual differ-
isual working memory capacity. Nature, 428, 748–751.
cCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal
differences in controlling access to working memory. Nature, 438,
F., Vecera, S. P., & Luck, S. J. (2003). Perceptual organization influences
king memory. Psychonomic Bulletin & Review, 10, 80–87.
M.M. (2006).Dissociableneuralmechanisms supportingvisual short-
ory for objects. Nature, 440, 91–95.
, M. M. (2007). Visual grouping in human parietal cortex. Proceed-
National Academy of Sciences of the United States of America, 104,
771.
M. M. (2009). Selecting and perceiving multiple visual objects. Trends
e Sciences, 13, 167–174.
uck, S. J. (2008). Discrete fixed-resolution representations in visual
emory. Nature, 453, 233–235.