Subitizing reflects visuo-spatial object individuation capacity.

Abstract

Subitizing is the immediate apprehension of the exact number of items in small sets. Despite more than a 100years of research around this phenomenon, its nature and origin are still unknown. One view posits that it reflects a number estimation process common for small and large sets, which precision decreases as the number of items increases, according to Weber's law. Another view proposes that it reflects a non-numerical mechanism of visual indexing of multiple objects in parallel that is limited in capacity. In a previous research we have gathered evidence against the Weberian estimation hypothesis. Here we provide first direct evidence for the alternative object indexing hypothesis, and show that subitizing reflects a domain general mechanism shared with other tasks that require multiple object individuation.

Full text

Page 1

Brief article
Subitizing reflects visuo-spatial object individuation capacity
Manuela Piazzaa,b,c,⇑, Antonia Fumarolaa,d, Alessandro Chinelloa,b, David Melchera,b
aCenter for Mind/Brain Sciences, University of Trento, Italy
bDepartment of Cognitive Sciences, University of Trento, Italy
cINSERM, U562, Cognitive Neuroimaging Unit, Gif/Yvette, France
dDepartment of Psychology ‘‘Gaetano Kanizsa’’, University of Trieste, Italy
a r t i c l e i n f o
Article history:
Received 19 February 2010
Revised 6 May 2011
Accepted 21 May 2011
Available online xxxx
Keywords:
Subitizing
Visuo-spatial working memory
Visuo-spatial attention
Capacity limits
Individuation
a b s t r a c t
Subitizing is the immediate apprehension of the exact number of items in small sets.
Despite more than a 100 years of research around this phenomenon, its nature and origin
are still unknown. One view posits that it reflects a number estimation process common for
small and large sets, which precision decreases as the number of items increases, according
to Weber’s law. Another view proposes that it reflects a non-numerical mechanism of
visual indexing of multiple objects in parallel that is limited in capacity. In a previous
research we have gathered evidence against the Weberian estimation hypothesis. Here
we provide first direct evidence for the alternative object indexing hypothesis, and show
that subitizing reflects a domain general mechanism shared with other tasks that require
multiple object individuation.
? 2011 Elsevier B.V. All rights reserved.
1. Introduction
The exact nature and origin of subitizing, the immediate
apprehension of the exact number of items in small sets, is
currently debated. One hypothesis posits that it reflects a
numerosity estimation process common for small and large
sets, which precision decreases as the number of items in-
creases, according to Weber’s law (Dehaene & Changeux,
1993; Gallistel & Gelman, 1991). In a previous investiga-
tion, however, we have discarded this account by showing
that enumeration responses (in terms of accuracy, esti-
mates distributions, and reaction times) dramatically differ
for sets of few items compared to sets with a large number
of items with identical ratios (e.g. 1, 2, 3, ... , 8 vs. 10, 20,
30, ... , 80). Moreover, according to the single estimation
process hypothesis, individual variability in subitizing
capacity should correlate with the individual variability in
the precision of large numerosity estimation. Thus, for
example, a small subitizing capacity should indicate a
roughinternalrepresentationofnumericalquantity,which,
in turns, should produce low accuracy in large numerosity
estimation. Contrary to this prediction, however, we have
shown that the two capacities do not correlate across sub-
jects (Revkin, Piazza, Izard, Cohen, & Dehaene, 2008).
An alternative view on subitizing proposes that it
reflects a mechanism of individuating multiple objects in
parallel (Trick & Pylyshyn, 1994) that is not specific to
the domain of number processing. The term ‘‘individua-
tion’’ is here used to emphasize the fact that items are,
through this mechanism, perceived as specific individuals
with a given identity and spatial location. According to this
view, such parallel individuation mechanism would be
common to any tasks requiring multiple objects individua-
tion. One such task is visual working memory (VWM),
where subjects encode multiple objects at a time to subse-
quently compare them to other objects. Like subitizing,
visual working memory also shows capacity limits of
around three to four items (Luck & Vogel, 1997), even if
the exact estimates of such limit are not fixed, but vary
depending on the participants and task parameters (Alva-
rez & Cavanagh, 2004; Bays & Husain, 2008; Melcher,
2001; Melcher & Morrone, 2007).
0010-0277/$ - see front matter ? 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.cognition.2011.05.007
⇑Corresponding author at: INSERM, Cognitive Neuroimaging Unit,
NeuroSpin Center, CEA Saclay, 91191 Gif-sur-Yvette, France. Tel.: +33 (0)1
69 08 19 06.
E-mail address: manuela.piazza@gmail.com (M. Piazza).
Cognition xxx (2011) xxx–xxx
Contents lists available at ScienceDirect
Cognition
journal homepage: www.elsevier.com/locate/COGNIT
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 2

In the developmental literature, this multiple object
tracking mechanism is sometimes defined as based on ‘‘ob-
ject files’’, intended as temporary representations of indi-
vidual objects from a scene (for a review, see (Feigenson,
Dehaene, & Spelke, 2004)). Physiologically, we may think
of this mechanism as an internal map whereby a limited
number of salient objects, as well as their locations can
be highlighted in parallel and subsequently used for
actions such as grasping or eye movements (Xu & Chun,
2009), or for cognitive tasks such as matching them with
other objects or assessing their number (Gottlieb, 2007).
We thus reason that if subitizing relies on such a do-
main-general process of visuo-spatial individuation, which
is not specific to numerical judgements, then the existing
inter-individual variability in subitizing (Revkin et al.,
2008) and VWM capacity (Vogel & Machizawa, 2004)
should tightly correlate, in the absence of correlation be-
tween either of these measures with the precision of large
numerosity estimation (Halberda, Mazzocco, & Feigenson,
2008; Piazza et al., 2010). We further reasoned that if the
individuation process needs to be accessed simultaneously
by the requirements of different tasks, as in a dual task
condition, then we should observe decreased capacity.
According with this idea, even an apparently basic ability
like subitizing should be impaired if its core resource
(the individuation ‘‘map’’) is being used for another task.
To test this hypothesis, we measured enumeration accu-
racy with and without a concurrent VWM task. Finally,
complementary to this prediction, we also reasoned that
if large numerosity estimation abilities do not heavily rely
on the individuation map, then they should not be im-
paired by a concurrent individuation task. Thus, we mea-
sured large numerosity comparison performance with
and without a concurrent VWM task.
2. Methods
2.1. Single task experiment
Sixteen
age = 26.2 years), naïve to the scope of the research, gave
written informed consent. The experiment took place in a
quiet, dimly lit room. Participants sat in front of the com-
puter monitor at a viewing distance of about 50 cm and
with their face fixed on a chinrest. Vocal and manual re-
sponses were recorded by a microphone and the E-prime
response box respectively. Each participant performed
the following three tasks, in randomized order.
healthyparticipants(10males, mean
2.1.1. Dots counting task
Participants were presented with arrays of one to eight
colored dots appearing in a central gray circle subtending
3?, and asked to name aloud their number as quickly and
accurately as possible (for one exemplar stimulus and the
exact trial structure see Fig. 1, panel A). In order to make
sure that participants’ estimation was based on numerosi-
ty and not on other factors, dots were generated so that,
across numerosities, half were of constant dot density
and the other half of constant dot size (Revkin et al.,
2008). Dot colors varied randomly among nine easily
discriminable colors, selected without replacement. Re-
sponses given within 1600 ms were entered by the exper-
imenter with a keyboard. The experiment started with 10
training trials, and comprised 160 trials, organized in five
blocks.
2.1.2. Visual working memory task (hereafter VWM task)
Participants were presented with two arrays (a sample
and a target array, separated by a retention interval) of
one to eight dots of different colors (selected randomly
without replacement), and were asked to perform a vocal
same–different judgment. Apparatus and stimuli were
the same as the dot counting task (see Fig. 2, panel A). In
half of the trials the test array was identical to the sample
array, while in the remaining half the color of one item was
changed. Responses given within 2000 ms were entered by
the experimenter with a keyboard. The experiment started
with 10 training trials, and comprised 160 trials, organized
in five blocks.
2.1.3. Dots comparison task
Participants were presented with two dots arrays (black,
on a gray circular background, presented laterally of a cen-
tral white fixation cross) and judged, without exact count-
ing, which one contained more dots by pressing the
response box button on the side of the larger array. The ar-
rays remained on the screen until subjects gave their re-
sponse. Dots number varied from 10 to 44, such that the
numerical ratio between the two arrays spanned five val-
ues: 1.06, 1.14, 1.23, 1.33, or 1.6. The arrays were generated
to be equated on half the trials in dot size and in the other
half in occupied area (Piazza, Izard, Pinel, Le Bihan, &
Dehaene, 2004). The experiment started with 10 training
trials and comprised 140 trials, organized in seven blocks.
2.2. Dual task experiments
Two new groups of subjects performed two separate
experiments, one investigating the pattern of interference
between VWM and counting, and the other investigating
the pattern of interference between VWM and large num-
ber estimation.
2.2.1. Dots counting and VWM
Seventeen healthy adult subjects (seven males, mean
age = 22.6 years) were tested. In the same trial, they per-
formed two tasks, a counting and a working memory task,
in a typical dual task condition. In order to obtain a baseline
measure,theyalso performedthecountingtask alonewhile
ignoringtheworkingmemorystimulionthescreen,andthe
VWM task alone while ignoring the enumeration stimuli, in
separate blocks. Participants were first presented with a
memory set of either two or four colored circles displayed
near fixation. The circles were 1? in diameter and were in
one of eight colors (black, white, red, green, blue, yellow,
purple or brown), selected randomly without replacement.
The memory set was then replaced with the counting set,
consisting in arrays of one to eight Gabor stimuli (oriented
contrast gratings windowed by a Gaussian function) dis-
played against a mean gray background and subsequently
masked with 24 randomly oriented Gabor stimuli. Each
2
M. Piazza et al./Cognition xxx (2011) xxx–xxx
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 3

Gaborstimulussubtended2?invisualangleandwaslocated
within a radius of 8? from the center of the screen. After the
mask,thememorytestsetwasshown.Attheendofthetrial
participants were asked to first report the number of Gabor
stimulibytypinginthenumberofitemsusingthekeyboard
(primary task) and then report whether the twocolored cir-
cle displays were identical or different (secondary task),
againviakeypress(seeFig.4,panelA).On90%oftrials,there
were 1, 2, 4, 6 or 8 stimuli presented. The other 10% of trials,
3,5,7or9stimuliwerepresentedtoensurethatparticipants
didnotalwaysguessanevennumberwhenunsure(thefirst
trial was set to always show 3, 5, 7 or 9 stimuli, for this rea-
son). The counting and the VWM stimuli never spatially
overlapped, as the area within a 3? radius from fixation
was reserved to the VWM stimuli. The experiment started
with 20 training trials and comprised 330 experimental tri-
als, organized in three blocks.
2.2.2. Dots comparison and VWM
Fifteen healthy adult subjects (four males, mean
age = 28 years), were tested. In the same trial, participants
performed both a dot comparison and a working memory
task, in a typical dual task condition. In order to obtain a
baseline measure, they also performed the dot comparison
task alone, while ignoring the working memory stimuli on
the screen, in a separate block. The condition order (dual
task first or control task first) was randomized across par-
ticipants. Trials started with the presentation of a memory
set consisting in one to eight colored dots which were dis-
played within a central gray circle (subtending 4.8?) for
500 ms. Dots were identical to the ones used in the single
WM task (see description above). The memory set was
then cleared from the screen and replaced after 500 ms
with the comparison sets, consisting of two arrays of black
dots, appearing in two lateral white circles at an offset of
5.7? from fixation bilaterally. Dots were identical to the
ones used in the single dots comparison task (see descrip-
tion above). The comparison sets were presented for
1000 ms, and followed by 500 ms blank. Finally, the
memory test set appeared in the central circle, for
250 ms, and was followed by 250 ms blank, marking the
end of the trial. Participants were asked to first judge
which lateral array contained more dots by pressing the
mouse keys on the same side as the larger array (primary
500
600
700
800
900
1000
0
0
0,1
0.1
0,2
0.2
0,3
0.3
0,4
0.4
0,5
1234
Set size
5678
Response Time (ms)
Error rate
Accuracy
Speed
200 ms
1500 ITI
How many?
250 ms
100 ms
Time
AB 0.5
Fig. 1. Dots counting task. Example of stimuli, trial structure and timing (panel A). Performance (RTs and error rate) as a function of set size (panel B). Error
bars are SEM.
700 ms
Same or different?
1000 ms
Time
200 ms
100 ms
2000 ms
1500 ITI
0
1
2
3
4
5
6
1234
Set size
5678
Number of objects encoded (K)
AB
Fig. 2. Visual working memory task. Example of stimuli, trial structure and timing (panel A). Performance (in terms of number of objects encoded) as a
function of set size (panel B). Error bars are SEM.
M. Piazza et al./Cognition xxx (2011) xxx–xxx
3
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 4

task), and then report whether the two colored circles sets
displayed centrally were identical or different (secondary
task), via keyboard key press. The experiment started with
10 training trials and comprised 200 trials, organized in
five blocks.
3. Results
3.1. Single task experiments
Inallthreetasksweobservedtheexpectedpatternsofre-
sults.Inthedotscountingtask,meancorrectresponsetimes
(RTs)anderrorsincreasedwithsetsize(F(7, 105) = 155.279,
p = 0.001, and F(7, 105) = 55.95, p = 0.000, respectively), but
only starting from numerosity 3 onwards (see Fig. 1, panel
B).
We estimated, for each subject, the subitizing capacity
(hereafter S) by fitting the full RTs curve with a sigmoid
function of numerosity and taking the inflexion point of
that curve (Revkin et al., 2008). Data fit was very good
(R2= 0.89, SD = 0.09) and yielded a mean estimate across
subjects of 4.47 (range = 3.81–5, SD = 0.36; Shapiro–Wilk
test (p = 0.57) confirmed that data was normally distrib-
uted). While such sigmoid fit appeared to be the most
appropriate approach to the present data (see Fig. 1, panel
B), the numerical output used to estimate subitizing (e.g.
the flex of the sigmoid) grossly overestimates the actual
subitizing range (Revkin et al., 2008). Such overestimation
should not be considered as a problem in this study be-
cause the scope of the fitting is to capture the inter-
individual differences, and this method does this very well.
However, in order to thoroughly check the consistency and
reliability of our estimated inter-individual differences, we
also fitted a bilinear function to the data and took the
intersection between the two lines as another estimate of
subitizing (Green & Bavelier, 2003). Given the drop in RTs
for the large numerosity, probably due to guessing end-ef-
fects, we excluded the last data point, and fitted RTs for
numerosities 1–7. This model also fit the data well
(R2= 0.94, SD = 0.03), and gave an estimated subitizing
range of 2.2 (range = 1.39–2.78, SD = 0.5, data not normally
distributed, Shapiro–Wilk p = 0.005). Importantly, as ex-
pected, the ranges estimated with these two methods
highly correlated across subjects (R = 0.83, p = 0.000).
In the VWM task, accuracy declined with increased set
size (set size 1:99%; 2:98%; 3:95%; 4:95%; 5:85%; 6:81%;
7:73%; 8:78) (F(7, 105) = 32.93, p = 0.000) especially start-
ing from sets with more than four objects.
The number of objects encoded at each set size (N), esti-
mated with Cowan’s K formula (Cowan, 2001), increased
up to set size 4, and levelled off thereafter (see Fig. 2, panel
B). For each subject, we took the average K across set sizes
as an estimate of their capacity. Across subjects, the aver-
age estimated K was 3.01 (range = 1.6–4.1, SD = 0.62, data
normally distributed, Shapiro–Wilk test p = 0.69).
In the dots comparison task, performance was modu-
lated by the ratio between the sets, according to Weber’s
law. Accuracy was used to estimate the internal Weber
fraction (hereafter W), a measure of the precision of the
numerical estimation, for each participant, using a method
previously described (Dehaene, 2007; Piazza et al., 2004).
Data fit was very good (average R2= 0.90, SD = 0.12), and
yielded a mean estimate of 0.20 (range = 0.09–0.32,
SD = 0.067,normallydistributed,
p = 0.85), in excellent agreement with previous reports
(Piazza et al., 2010).
We conducted correlation analysis between the esti-
mated measures: subitizing capacity (S), VWM capacity
(K) and the numerosity estimation precision (w). If subitiz-
ing relies on a single process of visuo-spatial individuation,
used for multiple tagging of objects in parallel in various
tasks, then subitizing range and VWM capacity should
tightly correlate, in the absence of correlation between
these measures and the precision of numerosity compari-
son (w). Results clearly confirmed these predictions. First,
we replicated the absence of correlation between subitiz-
ing capacity and numerosity comparison precision first ob-
served by Revkin et al. (2008) (R = 0.13, p = 0.63, and
R = 0.12, p = 0.64 for S calculated with the sigmoid and
the bilinear fit respectively). Second, we confirmed that
numerosity comparison precision was also unrelated to
VWM capacity (R = 0.07, p = 0.80). Finally, and crucially,
we observed a significant linear correlation between subi-
tizing capacity and VWM capacity (R = 0.73, p = 0.001; 95%
confidence interval, 0.37–0.9, and Kendall’s Tau correlation
coefficient = .370, p = 0.04, for the sigmoid and the bilinear
fit for deriving S, respectively) (see Fig. 3).
We statistically confirmed that the correlation between
S and VWM is significantly higher that the correlation be-
tween S and W, by performing a Z1bar⁄test (Steiger, 1980)
(Z = 2.228, p = 0.02).
Shapiro–Wilktest
3.2. Dual task experiments
3.2.1. Dots counting and VWM
Without any added VWM load, counting performance
was near perfect for up to four items, with a drop in pro-
portion correct when six or eight items were presented.
Thus, the subitizing range was around four items, consis-
tent with previous reports. The addition of the VWM load
1,5
1.5
2,5
2.5
3,5
3.5
4,5
4.5
3,5
3.5 4 4.5 5
4 4,55
VWM capacity
Subitizing capacity
Fig. 3. Correlation between visual working memory and subitizing
capacity across subjects.
4
M. Piazza et al./Cognition xxx (2011) xxx–xxx
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 5

strongly modulated counting performance (ANOVA main
effect of VWM load (F(2, 32) = 21.78, p < 0.001)) (see
Fig. 2, panel B). The effect of VWM increased for larger
numbers of items (interaction between VWM load and
set size (F(8, 32) = 2.35, p = 0.037). The main effect of
VWM load was significant for a memory load of two items
(F(1, 16) = 13.01, p = 0.002) and for the larger memory load
of four items (F(1, 16) = 47.54, p < 0.001).
In order to more directly test the hypothesis that subi-
tizing capacity can be modulated by concurrent working
memory load, we estimated, for each subject, the subitiz-
ing capacity (S) by fitting the accuracy curve with a
sigmoid function of numerosity, closer to the accuracy data
available in this experiment then a bilinear function fit (see
Fig. 4, panel B). Data fit (R2= 0.89, SD = 0.16) yielded a
average estimate of S across subjects and the three load
conditions of 4.7 (SD = 0.97). Subitizing capacity correlated
across subjects in the no load vs. load conditions (R = 0.70,
p = 0.002), suggesting that our measure is a reliable one.
Despite such inter-subject consistency, however, in confir-
mation to our hypothesis we observed that the estimated S
was higher in the no load condition and decreased propor-
tionally with VWM load (ANOVA main effect of VWM load
F(2,32) = 15.72, p < 0.001, pairwise planned comparisons
all p < 0.001).
Consistent with the hypothesis that VWM and enu-
meration would interfere with each other based on a
shared capacity limit, performance on the VWM task
was best when the number of items to enumerate was
low (especially when there was only one item) and much
worse when there were many items to enumerate. There
was a main effect of the number of items to enumerate
on VWM performance (F(4, 13) = 56.31, p < 0.001). The
presence of an interaction between the VWM set size
and the number of items to enumerate (F(4, 13) = 8.35,
p = 0.001) may be explained by the particularly good per-
formance when there were few items and a small set size
(performance was near perfect for the VWM set size of
two items when there was only one item to enumerate).
This overall trend can be seen by plotting percent correct
in the VWM task as a function of the total number of
items presented during the trial, including both the
memory set and the enumeration stimuli (see Fig. 4, pa-
nel C). The flex in the curve between three and four
items suggests that here was a trade-off in performance
consistent with a limited, shared resource for the two
tasks.
3.2.2. Dots comparison and VWM
One subject responded quasi randomly in the dots com-
parison task, in both the no load and the load task, such that
his psychometric curves were not fittable, and we thus ex-
cluded the data from further analysis. The significant corre-
lation of the weber fraction across subjects in the no load
0
0,2
0,4
0.4
0,6
0.6
0,8
0.8
1
12468
Proportion of errors
Set size
no load
load 2
load 4
2
2
Total number of items in both tasks
3
3
4
4
56810 12
10
0
15
20
25
30
56810 12
1
0.75
0.5
0.25
1
0.2
0
.
?
.
.
.
.
.
.
.
300 ms
500 ms
500 ms
200 ms
500 ms
100 ms
250 ms
Task 1. How many Gabors?
Task 2. Same or different?
1500 ITI
Time
Proportion of errors
AB
C
Fig. 4. Dual-task experiments. Example of stimuli, trial structure and timing (panel A). The Gabor stimuli were presented at 60% contrast, in order to
maintain good visibility but avoid afterimages. Pilot testing revealed that up to nine stimuli could be accurately counted given unlimited time. Performance
in the dots counting task as a function of set size (panel B). Separate lines show performance without a VWM load (black), with a concurrent memory load of
two items (dark gray) or of four items (light gray). Performance in the VWM task as a function of total number of items presented during the memory set
and enumeration set (panel C). Note that data for two and four items includes that found in the control condition in which the participants ignored the
enumeration task and concentrated only on the VWM task. Error bars are SEM.
M. Piazza et al./Cognition xxx (2011) xxx–xxx
5
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 6

vs. load conditions (R = 0.82, p = 0.001), suggests that our
measure is a reliable one. Without any added VWM load,
subjects were quit precise in the dots comparison task,
yielding to anestimated
(SD = 0.09). For the condition with the additional VWM
load, the weber fraction was equal to 0.26, (SD = 0.14),
slightly higher than the no load condition (t(13) = ?2.371,
p = 0.034). Crucially, however, the weber fraction was not
modulated by VWM load (R = 0.187, p = 0.657). In order to
fit the psychometric functions at the individual subject le-
vel, we collapsed trials corresponding to low working
memory load (1–4) vs. high WM load (5–8), in the dual task
and the control task condition separately. Results showed
no effect of working memory load (F(1, 13) = 0.323,
p = 0.679), nor an interaction between task condition (sin-
gle vs. dual) and working memory load (F(1, 13) = 0.059,
p = 0.811).
Performance in the VWM task itself (the secondary
task) remained extremely high, confirming that subjects
were not ignoring the secondary task (average Cowan’s
K = 3.5, range = 0.6–6, SD = 1.41), despite the concurrent
numerosity comparison task.
weber fractionof0.21
4. Discussion
In conflict with the idea that small numerosities are
processed by a Weberian mechanism for extracting
numerical information common for large and small num-
erosity, we found that individual differences in subitizing
capacity do not correlate with individual difference in large
number estimation precision. Thus, the two mechanisms
seem to be of a different nature. This result is in line with
recent work in favor of the notion of a dissociation be-
tween large and small numerosity processing in terms of
attentional resources, showing that while subitizing is
influenced by a concurrent attentional task, numerosity
estimation is not (Burr, Turi, & Anobile, 2010).
Indeed, in agreement with the idea that small numeros-
ities are processed by an object individuation mechanism
dedicated to multiple objects processing, we observed that
individual differences in subitizing capacity tightly corre-
lated with the individual differences in VWM capacity.
These results suggest that subitizing and visuo-spatial
working memory share some key components, in particu-
lar those components that exhibit a capacity limit. Inter-
estingly, independent work on the neural basis of
capacity limits point to posterior parietal cortex as the lo-
cus of capacity limits, accounting for the inter-individual
variability in performance in both subitizing (Piazza,
Giacomini, Le Bihan, & Dehaene, 2003) and visuo-spatial
working memory (Todd & Marois, 2004; Vogel & Machiza-
wa, 2004). According to the object individuation hypothe-
sis, at the origin of capacity limitations common to
subitizing and visuo-spatial memory is the architecture
of our sensory-motor system that generates maps of a lim-
ited number of salient objects and their locations in paral-
lel. These maps are instrumental for keeping track of
multiple items in order to guide cognition and action
(Drew & Vogel, 2008; Gottlieb, 2007; Melcher, 2001;
Melcher & Colby, 2008). Eventually, these maps support
perceptual and numerical judgments of the sort explored
in the present study.
By using a dual task paradigm, we provide further evi-
dence that subitizing and visuo-spatial working memory
share common resources. Indeed, we showed that the
maintenance of two or four items in VWM interfered with
enumeration by reducing the subitizing range. Likewise,
the enumeration task interfered with the VWM task in a
predictable manner, consistent with a common capacity
limit. Complementary to this observation, and in agree-
ment with our correlational results, we also show that
loading VWM does not result in a decrease in precision
of large numerosity estimation. This confirms the relative
independence between object indexing processes (subitiz-
ing VWM) and numerosity estimation processes, and even
suggests the possibility that these mechanisms rely on sep-
arable neural systems (see Piazza (2010) for a review of
current imaging data).
The results of this study are also in line with previous
research showing that subitizing is not fully automatic or
‘‘pre-attentive’’, in that its capacity can be reduced by con-
current visuo-attentive tasks (Egeth, Leonard, & Palomares,
2008; Olivers & Watson, 2008; Vetter, Butterworth, & Bah-
rami, 2008; Xu & Liu, 2008). Beyond the debate related to
the natureof subitizing,
the working memory literature in that they suggest that
the ability to store items in memory is closely related to
the ability to quickly select individual. This individuation/
selection step seems to form a first bottleneck which
serves as an upper limit on the ability to track multiple
items or maintain them in memory. Thus, the central
capacity limitation on the number of items we can eventu-
ally recall might be crucially determined by the number of
items we can initially encode (Drew & Vogel, 2008).
theseresultsalso inform
References
Alvarez, G. A., & Cavanagh, P. (2004). The capacity of visual short-term
memory is set both by visual information load and by number of
objects. Psychological Science, 15(2), 106–111.
Bays, P. M., & Husain, M. (2008). Dynamic shifts of limited working
memory resources in human vision. Science, 321(5890), 851–854.
Burr, D., Turi, M., & Anobile, G. (2010). Subitizing but not estimation of
numerosity requires attentional resources. Journal of Vision, 10(6).
Cowan, N. (2001). The magical number 4 in short-term memory: A
reconsideration of mental storage capacity. Behavioral and Brain
Sciences, 24(1), 87–114 (discussion 114–185).
Dehaene, S. (2007). Symbols and quantities in parietal cortex: Elements of a
mathematical theory of number representation and manipulation.
Attention and performance (Vol. XXII). Press HU, pp. 527–574.
Dehaene, S., & Changeux, J. P. (1993). Development of elementary
numerical abilities:Aneuronal
Neuroscience, 5, 390–407.
Drew, T., & Vogel, E. K. (2008). Neural measures of individual differences
in selecting and tracking multiple moving objects. Journal of
Neuroscience, 28(16), 4183–4191.
Egeth, H. E., Leonard, C. J., & Palomares, M. (2008). The role of attention in
subitizing: Is the magical number 1? Visual Cognition, 16(4), 463–473.
Feigenson, L., Dehaene, S., & Spelke, E. (2004). Core systems of number.
Trends in Cognitive Sciences, 8(7), 307–314.
Gallistel, C. R., & Gelman, R. (1991). The preverbal counting process. In W.
Kessen, A. Ortony, & F. I. M. Craik (Eds.), Thoughts memories and
emotions: Essays in honor of George Mandler (pp. 65–81). Hillsdale, NJ:
Lawrence Erlbaum Associates.
Gottlieb, J. (2007). From thought to action: The parietal cortex as a bridge
between perception, action, and cognition. Neuron, 53(1), 9–16.
Green, C. S., & Bavelier, D. (2003). Action video game modifies visual
selective attention. Nature, 423(6939), 534–537.
model.
JournalofCognitive
6
M. Piazza et al./Cognition xxx (2011) xxx–xxx
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007

Page 7

Halberda, J., Mazzocco, M. M., & Feigenson, L. (2008). Individual
differences in non-verbal number acuity correlate with maths
achievement. Nature, 455(7213), 665–668.
Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for
features and conjunctions. Nature, 390(6657), 279–281.
Melcher, D. (2001). Persistence of visual memory for scenes. Nature,
412(6845), 401.
Melcher, D., & Colby, C. L. (2008). Trans-saccadic perception. Trends in
Cognitive Sciences, 12(12), 466–473.
Melcher, D., & Morrone, M. C. (2007). Trans-saccadic memory: Building a
stable world from glance to glance. In R. P. G. Van Gompel, M. H.
Fischer, W. S. Murray, & R. L. Hill (Eds.), Eye movement research: A
window on mind and brain (pp. 213–236). Oxford: Elsevier.
Olivers, C., & Watson, D. (2008). Subitizing requires attention. Visual
Cognition, 16, 439–462.
Piazza, M. (2010). Neurocognitive start-up tools for symbolic number
representations. Trends in Cognitive Sciences, 14(12), 542–551.
Piazza, M., Facoetti, A., Trussardi, A. N., Berteletti, I., Conte, S., Lucangeli,
D., et al. (2010). Developmental trajectory of number acuity reveals a
severe impairment in developmental dyscalculia. Cognition, 116(1),
33–41.
Piazza, M., Giacomini, E., Le Bihan, D., & Dehaene, S. (2003). Single-trial
classification of parallel pre-attentive and serial attentive processes
using functional magnetic resonance imaging. Proceedings of the Royal
Society of London. Series B: Biological Sciences, 270(1521), 1237–1245.
Piazza, M., Izard, V., Pinel, P., Le Bihan, D., & Dehaene, S. (2004). Tuning
curves for approximate numerosity in the human intraparietal sulcus.
Neuron, 44(3), 547–555.
Revkin, S. K., Piazza, M., Izard, V., Cohen, L., & Dehaene, S. (2008). Does
subitizing reflect numerical estimation? Psychological Science, 19(6),
607–614.
Steiger, J. (1980). Tests for comparing elements of a correlation matrix.
Psychological Bulletin, 87, 245–251.
Todd, J. J., & Marois, R. (2004). Capacity limit of visual short-term memory
in human posterior parietal cortex. Nature, 428(6984), 751–754.
Trick, L. M., & Pylyshyn, Z. W. (1994). Why are small and large numbers
enumerated differently? A limited capacity preattentive stage in
vision. Psychological Review, 100, 80–102.
Vetter, P., Butterworth, B., & Bahrami, B. (2008). Modulating attentional
load affects numerosity estimation: Evidence against a pre-attentive
subitizing mechanism. PLoS ONE, 3(9), e3269.
Vogel, E., & Machizawa, M. (2004). Neural activity predicts individual
differences in visual working memory capacity. Nature, 428(6984),
748–751.
Xu, X., & Liu, C. (2008). Can subitizing survive the attentional blink? An
ERP study. Neuroscience Letters, 440(2), 140–144.
Xu, Y., & Chun, M. M. (2009). Selecting and perceiving multiple visual
objects. Trends in Cognitive Sciences, 13(4), 167–174.
M. Piazza et al./Cognition xxx (2011) xxx–xxx
7
Please cite this article in press as: Piazza, M., et al. Subitizing reflects visuo-spatial object individuation capacity. Cognition (2011),
doi:10.1016/j.cognition.2011.05.007