Vision Research 85 (2013) Contents lists available at SciVerse ScienceDirect. Vision Research. journal homepage:

Similar documents
Tracking the Development of Automaticity in Memory Search with Human Electrophysiology

Visual Short-term Memory Capacity for Simple and Complex Objects

Does Contralateral Delay Activity Reflect Working Memory Storage or the Current Focus of Spatial Attention within Visual Working Memory?

Manuscript under review for Psychological Science. Direct Electrophysiological Measurement of Attentional Templates in Visual Working Memory

Does contralateral delay activity reflect working memory storage or the current focus of spatial attention within visual working memory?

Selective storage and maintenance of an object s features in visual working memory

Does scene context always facilitate retrieval of visual object representations?

Report. Spatial Attention Can Be Allocated Rapidly and in Parallel to New Visual Objects. Martin Eimer 1, * and Anna Grubert 1 1

Templates for Rejection: Configuring Attention to Ignore Task-Irrelevant Features

Running Head: VISUAL WORKING MEMORY LOAD AND ATTENTIONAL CONTROL

Electrophysiological Substrates of Auditory Temporal Assimilation Between Two Neighboring Time Intervals

Visual Working Memory Capacity for Color Is Independent of Representation Resolution

Attentional Blink Paradigm

On perfect working-memory performance with large numbers of items

Selective bias in temporal bisection task by number exposition

The impact of item clustering on visual search: It all depends on the nature of the visual search

Limitations of Object-Based Feature Encoding in Visual Short-Term Memory

Memory-Based Attention Capture when Multiple Items Are Maintained in Visual Working. Memory. Andrew Hollingworth and Valerie M.

Active suppression after involuntary capture of attention

Independence of Visual Awareness from the Scope of Attention: an Electrophysiological Study

Competition in visual working memory for control of search

Electrophysiological Indices of Target and Distractor Processing in Visual Search

Early posterior ERP components do not reflect the control of attentional shifts toward expected peripheral events

The impact of numeration on visual attention during a psychophysical task; An ERP study

Figure 1. Source localization results for the No Go N2 component. (a) Dipole modeling

(Visual) Attention. October 3, PSY Visual Attention 1

An EEG/ERP study of efficient versus inefficient visual search

Delineating the Neural Signatures of Tracking Spatial Position and Working Memory during Attentive Tracking

Journal of Experimental Psychology: Learning, Memory, and Cognition

EEG Analysis on Brain.fm (Focus)

Discrete Resource Allocation in Visual Working Memory

Finding Memory in Search: The Effect of Visual Working Memory Load on Visual Search. 1 Department of Psychology, University of Toronto

The allocation of visual short-term memory capacity: Evidence for a flexible storage mechanism

Evidence for false memory before deletion in visual short-term memory

AUTOCORRELATION AND CROSS-CORRELARION ANALYSES OF ALPHA WAVES IN RELATION TO SUBJECTIVE PREFERENCE OF A FLICKERING LIGHT

Your article is protected by copyright and all rights are held exclusively by Psychonomic Society, Inc.. This e-offprint is for personal use only and

Demonstrations of limitations in the way humans process and

Attentional spread in the statistical processing of visual displays

The role of selective attention in visual awareness of stimulus features: Electrophysiological studies

DATA MANAGEMENT & TYPES OF ANALYSES OFTEN USED. Dennis L. Molfese University of Nebraska - Lincoln

Visual Working Memory Represents a Fixed Number of Items Regardless of Complexity Edward Awh, Brian Barton, and Edward K. Vogel

Implicit memory influences the allocation of attention in visual cortex

Interference with spatial working memory: An eye movement is more than a shift of attention

Early Selection Versus Late Correction: Age-Related Differences in Controlling Working Memory Contents

Running head: PERCEPTUAL GROUPING AND SPATIAL SELECTION 1. The attentional window configures to object boundaries. University of Iowa

Introduction to Computational Neuroscience

Irrelevant features at fixation modulate saccadic latency and direction in visual search

Journal of Experimental Psychology: Human Perception and Performance

Top-down search strategies determine attentional capture in visual search: Behavioral and electrophysiological evidence

The association of color memory and the enumeration of multiple spatially overlapping sets

Supplementary material

Congruency Effects with Dynamic Auditory Stimuli: Design Implications

PSYCHOLOGICAL SCIENCE. Research Article

Morton-Style Factorial Coding of Color in Primary Visual Cortex

The Simon Effect as a Function of Temporal Overlap between Relevant and Irrelevant

Attentional enhancement during multiple-object tracking

Discrete capacity limits in visual working memory Keisuke Fukuda, Edward Awh and Edward K Vogel

Effects of Light Stimulus Frequency on Phase Characteristics of Brain Waves

Functional Fixedness: The Functional Significance of Delayed Disengagement Based on Attention Set

The path of visual attention

Perceptual grouping in change detection

Framework for Comparative Research on Relational Information Displays

The effects of subthreshold synchrony on the perception of simultaneity. Ludwig-Maximilians-Universität Leopoldstr 13 D München/Munich, Germany

Mental Rotation affects Spatial Attention: an EEG study. Introduction

The Representation of Simple Ensemble Visual Features Outside the Focus of Attention

Title of Thesis. Study on Audiovisual Integration in Young and Elderly Adults by Event-Related Potential

Object Substitution Masking: When does Mask Preview work?

The control of attentional target selection in a colour/colour conjunction task

What Matters in the Cued Task-Switching Paradigm: Tasks or Cues? Ulrich Mayr. University of Oregon

Neural correlates of short-term perceptual learning in orientation discrimination indexed by event-related potentials

Perceptual grouping in change detection

Enhanced discrimination in autism

Attentional and anatomical considerations for the representation of simple stimuli in visual short-term memory: evidence from human electrophysiology

Journal of Experimental Psychology: Learning, Memory, and Cognition

Multiple spatially-overlapping sets can be enumerated in parallel

The role of low frequency components in median plane localization

Neural Measures of Individual Differences in Selecting and Tracking Multiple Moving Objects

Tracking target and distractor processing in fixed-feature visual search: Evidence from human electrophysiology

What ERPs can tell us about Visual Working Memory. Veronica B. Perez & Edward K. Vogel. University of Oregon. Eugene, Oregon, USA

The speed of voluntary and priority-driven shifts of visual attention. *Corresponding Author. Malet Street, London, WC1E 7HX, UK

Stressing the mind: The effect of cognitive load and articulatory suppression on attentional guidance from working memory

Dissociable neural correlates for familiarity and recollection during the encoding and retrieval of pictures

Toward the Neural Mechanisms of Reduced Working Memory Capacity in Schizophrenia

Visual working memory for simple and complex visual stimuli

When memory is not enough: Electrophysiological evidence for goal-dependent use. of working memory representations in guiding visual attention

What matters in the cued task-switching paradigm: Tasks or cues?

The Neural Fate of Task-Irrelevant Features in Object-Based Processing

Human Variation in Overriding Attentional Capture

Distinct Capacity Limits for Attention and Working Memory Evidence From Attentive Tracking and Visual Working Memory Paradigms

A study of the effect of auditory prime type on emotional facial expression recognition

Mental representation of number in different numerical forms

Redundancy gains in pop-out visual search are determined by top-down task set: Behavioral and electrophysiological evidence

Profiling Attention s Pure Effect on the Sensory-Evoked P1 and N1 Event-Related Potentials of Human Electroencephalography

Posner s Attention Test

Transcranial direct current stimulation modulates shifts in global/local attention

Overcoming the three-item limit: Gestalt grouping principles explain increases in change detection capacity. Justin Halberda Johns Hopkins University

Supplementary materials for: Executive control processes underlying multi- item working memory

Visual Search: A Novel Psychophysics for Preattentive Vision

Grouping by similarity is mediated by feature selection: evidence from the failure of cue combination

LEARNING DURING SLEEP: AN INDIRECT TEST OF THE ERASURE-THEORY OF DREAMING

Transcription:

Vision Research 85 (2013) 20 25 Contents lists available at SciVerse ScienceDirect Vision Research journal homepage: www.elsevier.com/locate/visres Processing statistics: An examination of focused and distributed attention using event related potentials Shruti Baijal a,b, Chie Nakatani c,b, Cees van Leeuwen c,b, Narayanan Srinivasan a,b, a Centre of Behavioural and Cognitive Sciences, University of Allahabad, India b Laboratory for Perceptual Dynamics, Brain Science Institute RIKEN, Japan c Laboratory for Perceptual Dynamics, KU Leuven, Belgium article info abstract Article history: Received 21 December 2011 Received in revised form 19 September 2012 Available online 4 October 2012 Keywords: Focused attention Distributed attention Statistical summary Working memory CDA Human observers show remarkable efficiency in statistical estimation; they are able, for instance, to estimate the mean size of visual objects, even if their number exceeds the capacity limits of focused attention. This ability has been understood as the result of a distinct mode of attention, i.e. distributed attention. Compared to the focused attention mode, working memory representations under distributed attention are proposed to be more compressed, leading to reduced working memory loads. An alternate proposal is that distributed attention uses less structured, feature-level representations. These would fill up working memory (WM) more, even when target set size is low. Using event-related potentials, we compared WM loading in a typical distributed attention task (mean size estimation) to that in a corresponding focused attention task (object recognition), using a measure called contralateral delay activity (CDA). Participants performed both tasks on 2, 4, or 8 different-sized target disks. In the recognition task, CDA amplitude increased with set size; notably, however, in the mean estimation task the CDA amplitude was high regardless of set size. In particular for set-size 2, the amplitude was higher in the mean estimation task than in the recognition task. The result showed that the task involves full WM loading even with a low target set size. This suggests that in the distributed attention mode, representations are not compressed, but rather less structured than under focused attention conditions. Ó 2012 Elsevier Ltd. All rights reserved. 1. Introduction Corresponding author at: Centre of Behavioural and Cognitive Sciences, University of Allahabad, Allahabad 211 002, UP, India. E-mail address: nsrini@cbcs.ac.in (N. Srinivasan). Perceivers are limited in their ability to focus attention on items or regions. Common tasks, such as tracking of moving targets among non-targets, can only be performed successfully with up to four targets (Awh & Pashler, 2000; Cavanagh & Alvarez, 2005; Pylyshyn & Storm, 1988). Such tasks involve attention, not only to select the relevant targets but also to maintain object information in short-term memory as they move among non-targets. Therefore, the limitation of focused attention is considered to be closely related to the limitation in access to WM/short-term memory (Cowan, 2005; Luck & Vogel, 1997; Makovski & Jiang, 2007). Not all types of attentional tasks, however, seem to be subject to restrictions in the number of items that can be processed simultaneously (Ariely, 2001; Chong & Evans, 2011; Chong & Treisman, 2005a). Ariely (2001), for example, demonstrated that participants could accurately report statistical information such as the mean size of more than four circles, while they were poor at reporting the size of individual circles. Performance in the mean size judgment task is little, if at all, affected by variation in the number or density of items (Chong & Treisman, 2005a), exposure duration or delay (Chong & Treisman, 2003) and difficulty of selection (Chong & Treisman, 2005b). These findings contrast with those of WM/visual short-term memory (VSTM) studies (Baddeley, 1997; Franconeri, Alvarez, & Enns, 2007; Luck & Vogel, 1997; Makovski & Jiang, 2007), for which a close relation to focused attention was assumed. Some authors have argued that focused attention and limited capacity are still sufficient for reliable statistical judgments, if these are based on a small but representative subsample of the items on display. Myczek and Simons (2008) simulated the expected accuracy of the mean size judgment of eight circles based on random samples of one to three circles. Their model reached a level of accuracy well above chance level, even when means were computed based on one or two sampled items. The results of the model fitted well with human performance over arrays of eight circles. This led authors to conclude that the results on the mean judgment task could be obtained by using focused attention strategies that do not violate VSTM capacity restrictions using subsampling. Chong et al. (2008, Experiment 2) tested whether the mean is computed from a smaller number of samples rather than from all 0042-6989/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.visres.2012.09.018

S. Baijal et al. / Vision Research 85 (2013) 20 25 21 the items in the display. They replicated their previous study (Chong & Treisman, 2005a), except that they tested their participants with either the whole display as used in the original study or with samples of one or two items drawn from the display used in their original study. The findings of Chong et al. (2008) suggest that rather than a restricted sample, stimuli from the entire display contribute to the estimation of the mean. Statistical judgments, therefore, seem to involve a process of distributed attention involving computations over a large number of items. These findings of Treisman and her colleagues (Chong & Evans, 2011; Chong & Treisman, 2003, 2005a) have led to the notion of distributed attention. While typically items in VSTM are stored as coherent objects (object files in Treisman s terminology) when using focused attention, items might be stored in the VSTM at a more primitive level in the distributed attention mode. This level consists of features (Treisman, 2006) or proto-objects (Pylyshyn, 2003), content that is pooled to obtain summary statistics. By contrast, however, Alvarez and colleagues (Alvarez & Oliva, 2009; Franconeri, Alvarez, & Enns, 2007), proposed that in the distributed attention mode, objects are still integrally represented in VSTM, but their representation is compressed such that some information will inevitably be lost. Reporting summary statistics involves pooling information over a large number of items, such that this still can reliably be done. Treisman (2006), for example, proposed that parallel and automatic processes lead to computation of statistical properties of a group of items from their feature-level representations. While these studies (Chong & Treisman, 2003, 2005a, 2005b; Chong et al., 2008) have provided arguments in favor of the distributed attention hypothesis for statistical judgment, they do not fully reveal the nature of the representations used in distributed attention tasks. Different types of representation have been proposed for computing statistical judgments in the distributed attention mode; feature level representations (Treisman, 2006): proto-objects (Pylyshyn, 2003) and, alternatively, compressed objects (Alvarez & Oliva, 2009; Franconeri, Alvarez, & Enns, 2007). Feature-level representations and proto-objects are less structured than object files. We may assume that their formation involves activation of a feature map containing unbound features from all items in a display. Therefore, WM load would be high, even in the set-size 2 condition. Alvarez and colleagues (Alvarez, 2011; Brady & Alvarez, 2011), on the other hand, proposed that in distributed attention, the WM representation is compressed. The statistics over the compressed representation tend to be more accurate as they are obtained by pooling across independent sources of noise from the local measurements of the display (Alvarez & Oliva, 2009) and/or detecting outliers and guiding attention to appropriate targets (Alvarez, 2011). Only the statistical representation might be stored in WM resulting leading to low load (pertaining to retention of only one item or representation), less than the load in the focused attention mode for set-size 2. It is possible that the working memory load might depend on compression rate; if it were high, WM load would be low, and if it were low, WM would be high regardless of set size. It is also possible that the individual object information is not lost (after computing the statistical summary) and is also encoded along with the statistical estimate especially with low set sizes and this would also result in higher WM load. Examination of WM load and its determinants is critical to understanding the nature of representation in the distributed attention mode. For this purpose we used an event related potential measure of visual short term memory called Contralateral Delay Activity (CDA). CDA is a slow negative ERP component recorded at posterior electrode locations and is defined as a difference waveform between the ipsilateral and the contralateral sides of the brain during a delay period. CDA is proposed to reflect the load of VSTM. In Vogel and Machizawa (2004), participants were asked to detect changes between two images separated by a delay of about one second. The CDA amplitude increased as the number of items increased but reached a limit at about three to four items and the authors suggested that it represents the resources used to keep the representations active in VSTM (also see Drew & Vogel, 2008). While there is some debate on the precise interpretation of the amplitude of CDA, this measure is particularly relevant to the present study due to its sensitivity to the measures of number of objects held in working memory (Vogel & Machizawa, 2004), efficiency of attentional filtering (Jost et al., 2011; Vogel, McCollough, & Machizawa, 2005; Woodman & Arita, 2010), and precision of their visual representations (Machizawa, Goh, & Driver, 2012). Most importantly, CDA amplitude represents the precision (coarse or fine) with which object information is retained in the working memory, especially when the number of objects are relatively low (Machizawa, Goh, & Driver, 2012). Different amounts of precision with different number of items might be needed for statistical estimates compared to object identification. In the current study, we computed ERP measures from EEG recordings while observers engaged in either of two tasks: In the Mean Task, they estimated mean size over 2, 4 or 8 target disks; in the Member Task, they performed a recognition task over individual target disks. We calculated CDA to examine the nature of processing and the types of representation involved in the two tasks. The member task is similar to the one used by Vogel and colleagues and hence we expected CDA amplitude to increase as the number of disks increases from 2 to 4 and to asymptote for set-size 8. If participants use sub-sampling (Myczek & Simons, 2008) for the mean task, then participants would need to remember at best two items and hence CDA amplitude would be similar to or lower than the CDA amplitude of set-size 2 in the Member Task for all set sizes in the Mean task. Similarly, if the compressed estimate of mean size (Alvarez & Oliva, 2009; Ariely, 2001; Franconeri, Alvarez, & Enns, 2007) is remembered, that would predict CDA amplitudes in the Mean Task to be low (possibly, lower than set-size 2 of the member task as only one summary representation would be retained). On the other hand, if a feature-level representation of all objects were stored along with statistical estimate in WM, we would expect CDA amplitude to be higher (i.e., similar to set sizes 4 and 8 in the Member Task). 2. Method 2.1. Participants Ten female and three male volunteers (mean age = 21 years, SD = 2.5 years) with normal or corrected-to-normal vision provided written consent. All participants received 1000 yen per hour as compensation for their participation. The research ethics committee of RIKEN had accepted the experimental procedure. 2.2. Stimuli The stimulus display consisted of fifteen colored disks presented each on the left and right visual fields (Fig. 1). The disks were colored in red (RGB [220, 71, 105]) or green (RGB [94, 158, 12]), and were luminance matched (9.4 cd/m 2 ). Background was gray (RGB [192, 192, 192]) and its luminance was 26.6 cd/m 2. The diameter of the disks ranged from 0.29 to 2.29, separated by a step-size of 0.19, which yielded 11 sizes. These sizes were chosen pseudo-randomly for fifteen circles to be placed in a hemi-field. The circumferences of any two disks were separated by at least of 0.19. Either the left or the right visual half field was indicated as the task-relevant half field by a pre-cue, a red-colored (left or right)

22 S. Baijal et al. / Vision Research 85 (2013) 20 25 Participants were comfortably seated in a sound-attenuated chamber with dim ceiling lights. The monitor screen (Trinitron MultiScan G520, SONY, Tokyo, Japan) was placed at a distance of 155 cm. A response box (Etalcia tenkey box, Elecom, Osaka, Japan) was placed next to their arm rest, in immediate reach of the participant. After the electrodes were fixed, a practice session preceded the experiment. Each experimental trial began with the presentation of a redcolored central fixation cross for duration of 1000 ms. Subsequently, a red-colored arrow (pre-cue) was presented for 200 ms. The pre-cue pointed towards the left visual field in half of the trials and in the other half towards the right visual field. The cue was followed by a bilateral display of colored disks presented for 300 ms. After a 1000 ms delay period, two test disks were displayed; the participants chose one of the test disks as their response and rated the confidence of their response by pressing a corresponding key on the response box. Six keys on the response box were labeled 1 to 6, indexing the scale for confidence ratings. Out of these, any one of 3 keys on the left-side (labeled 1 3) could be used to choose the left-sided test disk and vice versa for choosing the right-sided test disk as response. In the Mean task condition, participants judged which of the two test disks matched the mean size of target disks of the previous display. In the Member task condition (or recognition task), participants judged which of the two test disks was presented previously on the cued visual field (i.e. was a member). The Mean and the Member tasks both had three target setsizes; there were 96 trials per target set-size. The member or mean judgment tasks were presented in a counterbalanced manner in alternating blocks, four blocks for each task. This yielded 576 trials for the experimental session. The number of pre-experimental practice trials was 120 (20 trials per condition). 2.4. EEG: apparatus and processing Fig. 1. Illustration of experimental task a trial began with the presentation of a fixation-cross followed by a cue. The stimulus display contained target (red) and non-target (green) disks. After a delay of 1000 ms, observers reported the mean target size in the Mean task condition by selecting one of the test disks. In the Member task condition, they reported which of the test disks was a member of the target disks. For each response, confidence rating was given. arrow (RGB [220, 71, 105]) presented at the center of the display. The task-relevant (cued) visual field included 2, 4 or 8 red disks of different sizes. The remaining disks were green, that is, non-targets. The task-irrelevant visual field contralateral to the cue contained equal numbers of red and green disks as the cued side. The average size of the non-target disks on either side was never equal to the average size of the target disks. The stimulus display never contained a disk of which the size matched with the average size of the target-set. The stimulus display was followed by a probe display which consisted of two red disks. The size of one of the disks matched the correct answer (which will be explained below), while the other was either larger or smaller than the correct disk by 0.39. 2.3. Procedure Stimulus presentation was performed using SuperLab Pro version 4.0 (Cedrus Corporation, San Pedro, CA). A 19-channel EEG system (EEG1100, Nihon Kohden, Tokyo, Japan) with a 19 tin electrodes (ElectroCap, Electro-Cap International, Inc., Eaton, Ohio) cap was used to record EEG activity. Reference electrodes were placed on the left and the right ear lobes, which were digitally linked. EOG electrodes were attached to the right and left temples for horizontal EOG (HEOG), as well as above and below the left eye for vertical EOG (VEOG). Data were digitized at 500 Hz (0.1 100 Hz analog bandwidth). The EEG data was subjected to independent component analysis (ICA) to remove artifacts; the procedure was able to identify independent component for eye-blink EOG, MEG, and AC related components. These artifact components were excluded from signal reconstruction. The reconstructed EEG signal was then band-pass filtered (0.1 30 Hz). The participants sometimes failed to keep fixation on the center location. Based on EOG records, we manually excluded trials with saccades or eye movements away from fixation within 200 ms from the stimulus onset irrespective of the experimental condition for all participants. For detecting horizontal eye movements, HEOG activity was observed for any step-like voltage deflections of the positive or negative voltage. In other words, a positive or negative spike with an amplitude that remained constant for a small duration before it falls off to the original value constituted a horizontal eye movement. After artifact-rejection, we ensured that each subject had at least 10 trials available for averaging. Based on EOG records, we manually excluded trials with saccades or eye movements away from fixation within 200 ms from the stimulus onset. This procedure resulted in removal of around 11% of trials in the mean task and 10% of trials in the member task. The EEG was segmented from 1500 ms to +2500 ms from cueonset only for correct trials. The interval between 1500 ms and cue-onset was used as baseline. Each segment was baselinecorrected and then averaged from the cue onset to compute ERPs and further processed to obtain CDA waveforms. CDA was prominent over posterior electrode locations. As in previous CDA studies (Drew & Vogel, 2008; Vogel & Machizawa, 2004), we selected the posterior occipital, parietal and temporal electrode sites O1, O2, P3, P4, T5 and T6 for CDA analysis. Since CDA is a difference wave between ERPs from cued and un-cued hemispheres, we computed contralateral-ipsilateral difference waves for each of these pairs of electrodes (i.e., O1/2, P3/4, and T5/6) for the mean and members tasks. Based on CDA peak time of the previous studies (Drew & Vogel, 2008), the mean amplitude during the early delay period (the period between 500 and 900 ms from cue onset in Fig. 3, which corresponds to 0 400 ms from delay onset) was taken as CDA peak activity. We also analyzed the CDA waveforms 250 500 ms as well as 900 1500 ms from cue onset. 3. Results 3.1. Behavioral performance and subjective ratings Percentage accuracy was computed for each of the conditions and submitted to a [2 (task type) 3 (set-size)] analysis of

S. Baijal et al. / Vision Research 85 (2013) 20 25 23 Fig. 2. (a) Task performance (percentage accuracy), and (b) confidence rating (1 = not at all confident, 6 = absolutely confident) in the Mean and Member tasks for set-sizes 2, 4, and 8 (Error bars = SE). variance (ANOVA) and post hoc t-tests. All post hoc t-tests had Newman-Keuls corrections applied to their alpha levels. There was a main effect of task type, F(1, 12) = 78.86, p <.001 showing more accurate responses for the Mean task (76%) compared to the Member task (65%), as well as a main effect of set-size, F(1,12) = 37.14, p <.001. Post-hoc t-tests showed that overall accuracy reduced from set-size 2 to 4, t(12) = 4.65, p <.01 and from set-size 4 to 8, t(12) = 7.42, p <.001. Fig. 2a shows the significant interaction effect between task type and set-size, F(2, 24) = 30.35, p <.001. Post-hoc tests revealed that while the accuracy scores in the Member task reduced from set-size 2 to 4, t(12) = 8.39, p <.001 and from set-size 4 to 8, t(12) = 9.78, p <.001, there was no set-size dependent change in the accuracy scores in the Mean task (Fig. 2a). The results showed a reduction in accuracy with increase in the number of targets in the Member task, while the accuracy during the Mean task remained high across all set-sizes, which is consistent with other studies (Ariely, 2001; Chong & Treisman, 2005a). Confidence ratings for correct responses on the Mean and Member tasks were submitted to a 2 3 ANOVA. For one participant the data were not properly recorded and therefore could not be included in the analysis. There was a main effect of task, F(1,11) = 9.37, p <.05; participants were more confident during the Mean task than the Member task. There was a main effect of set-size, F(2, 22) = 31.95, p <.001. The post hoc tests showed that the participants were more confident in responding to set-size 2 Fig. 3. Difference waveform (contralateral ipsilateral ERPs) at the temporal site for the (a) mean task and the (b) member task. ERPs are shown from 300 ms before onset of cue (0 ms). Vertical lines indicate onset of stimulus (200 ms) and that of delay period (500 ms). Broken vertical line separates early (500 900 ms) and late (900 1500 ms) periods of CDA. In the Mean task, there is no set-size related difference in the amplitudes of the early and the late CDA. However, in the Member task, the early CDA amplitudes increased with set-size from two to four and asymptotes thereafter.

24 S. Baijal et al. / Vision Research 85 (2013) 20 25 compared to set-size 4, t(11) = 10.65, p <.001 and 8, t(11) = 8.55, p <.001. There was also a significant interaction effect of task type set-size, F(2, 22) = 6.33, p <.01. The post hoc tests showed that during the Member task the participants were more confident in set-size 2 compared to 4, t(11) = 14.65, p <.001 and 8, t(11) = 18.59, p <.001 and in set-size 4 compared to 8, t(11) = 3.95, p <.05. Similarly, in the Mean task, the participants were more confident in set-size 2 compared to 4, t(11) = 9.58, p <.001 and 8, t(11) = 11.83, p <.001 and in set-size 4 compared to 8, t(11) = 8.26, p <.001. Most importantly, participants were more confident in the Mean task compared to the Member task with set-size 4, t(11) = 6.38, p <.001 and set-size 8, t(11) = 8.07, p <.001. The results showed that the participants were more confident during the mean than during the Member task only with higher set-sizes (Fig. 2b). 3.2. CDA results Data from two participants were excluded from the statistical analysis of ERPs as their amplitudes were more than two standard deviations away from the group mean. 1 The behavioral results of all those whose ERP data was selected showed the same pattern as those obtained with all the thirteen participants. The difference ERPs of the Mean and Member tasks are shown in Fig. 3. A negative offset started around 300 ms after the cue onset (100 ms from stimulus onset), which was considered as CDA. Since CDA is a slow activity, we set two time windows for our analysis: an early delay period (500 900 ms) and a late delay period (990 1500 ms) with all time/latencies marked from cue onset in the current text, unless noted. They could be translated to latencies from stimulus onset by deleting 200 ms. Analysis time window is a potential issue of debate. Previous CDA studies used varieties of time windows (Drew & Vogel, 2008; Machizawa, Goh, & Driver, 2012; Vogel & Machizawa, 2004). Time window of our interest are in the early part of CDA, since Baijal and Srinivasan (2011) have demonstrated that working memory consolidation in the mean task occurs earlier than the member task. The mean amplitudes from these periods were submitted to a 3 (regions on the scalp: occipital, temporal and parietal) 2 (task types) 3 (set-sizes) repeated measures ANOVA. The pre-delay period activity (200 500 ms) was also analyzed; however no significant effects were observed. The mean amplitude of the late delay period (900 1500 ms in Fig. 3) showed no significant main effects or interactions suggesting that the late CDA amplitudes do not differ between the two tasks in the late delay period. Based on CDA peak time of the previous studies (Drew & Vogel, 2008), the mean amplitude during the early delay period (500 900 ms in Fig. 3, which corresponds to the period 300 700 ms from stimulus onset) was taken as CDA peak activity. The results of a 3 2 3 repeated measures ANOVA with the mean amplitude of CDA waveforms in the 500 900 ms from cue onset showed a main effect of scalp regions, F(2,20) = 5.45, p <.05. Post-hoc tests showed an overall larger amplitude of temporal, t(10) = 4.59, p <.05 and occipital, t(10) = 3.06, p = 0.05 compared to parietal region. Fig. 3 depicts the interaction of task type set-size effects, F(2, 20) = 4.33, p <.05. The post hoc tests revealed that during the Member task, the increase in CDA amplitude from set-size 2 to 4 was marginally significant, t(10) = 4.10, p =.06. There was a significant increase in CDA amplitudes from set-size 2 to 8 in the Member task, t(10) = 5.49, p <.05 with no significant difference in amplitude between set-size 4 and 8 (p >.1). This indicates an increase in the CDA amplitude with the number of targets in the Member task up to 4. Within the Mean task, there was no change in amplitude as a function of set-size (p >.4). Specifically, CDA amplitudes of set-size 2 were higher in the Mean task compared to the Member task, t(10) = 3.92, p <.05 (Fig. 4). These results further suggest differences in early CDA amplitudes between Mean and Member tasks. 4. Discussion Previous studies showed that mean size judgments could successfully be performed over an array of items after a short exposure, even if the number of items is beyond the capacity limits of focused attention (Ariely, 2001; Chong & Treisman, 2005a). Do such judgments involve a distributed attention mode (Alvarez & Oliva, 2009; Chong et al., 2008; Franconeri, Alvarez, & Enns, 2007; Treisman, 2006)? If so, what kind of representations is used? We hypothesized that WM loading could indicate the type of representation in the distributed attention mode; WM loading would be low regardless of set size, if only the statistical estimate is stored in WM (Treisman, 2006) or if a high rate of target compression is used (Alvarez & Oliva, 2009). The WM load would be high regardless of set sizes, if target compression is low, or if a feature-level or proto-object representation would be stored in WM. As a measure of the working memory load, we used CDA amplitude. As expected, the CDA amplitude increased with set-size in the Member Task, which is the focused attention task. The amplitude increased from set-size 2 to 4, but did not increase further for the set-size 8. This replicated the pattern of results in Vogel and Machizawa (2004). The authors considered that saturation of the amplitude reflected the close link between CDA amplitude and the working memory limitation, which is around 4 (Luck & Vogel, 1997). In the Mean Task, on the other hand, the amplitude was as high as the maximum level obtained with the member task, regardless of set size. Although there are discussions over what CDA amplitude reflects (Cowan, 2005; Drew & Vogel, 2008; Ikkai, McCollough, & Vogel, 2010; Woodman & Arita, 2010; Vogel & Machizawa, 2004), the simplest interpretation of the results would be that the working memory was full across all set sizes. In particular, at set-size 2, the amplitude was higher in the Mean than in the Member task conditions. The result favors feature-level representations. 1 We performed analysis on accuracy data from participants included in the ERP analysis (that is excluding the two participants). The results were similar to those obtained with all the participants. The overall accuracy in the mean task (75%) was greater than that in the member task (65%), F(1,10) = 65.42, p < 0.001. There was a significant main effect of set-size, F(1,10) = 27.07, p < 0.001 revealing a reduction in accuracy with an increase in set-size. In addition, there was a significant interaction effect of task and set-size, F(1,10) = 28.82, p < 0.001. There was a set-size dependent reduction in accuracy in the member task as the accuracy reduced from set-size 2 to set-size 4, t(10) = 8.40, p <.001 and from set-size 4 to set-size 8, t(10) = 8.59, p <.001; however, the accuracy in the mean task remained high irrespective of the set-size. Fig. 4. Early CDA amplitude as a function of task type and set-size.

S. Baijal et al. / Vision Research 85 (2013) 20 25 25 Machizawa, Goh, and Driver (2012) demonstrated that working memory load/ CDA amplitude was high when object information had to be retained with greater precision even when the set-size was low. It should be noted that working memory load in their study regards object files formed using focused attention. It may also be possible that greater difficulty in representing objects with high precision (i.e., discriminating between finer orientations) compared to lower precision (i.e., discriminating larger changes in orientation) would have led to increased CDA amplitudes. Our study provides new suggestion that nature of representations, statistical and object-based, could load working memory differently. In addition, at least for small set sizes both representations needed to compute statistical estimates as well as individual object information could both be available in WM resulting in higher CDA amplitudes with small set sizes (Alvarez, 2011; Brady & Alvarez, 2011) given that both could compete for common WM resources (Baijal & Srinivasan, 2011). It should be noted that the results by themselves do not specify the exact nature of representation used for mean judgment. It is the under-structured feature-based nature of the representations that explains the current result. Therefore, the results do not categorically exclude other types of representation, such as coarse representations (Bays & Husain, 2008; Franconeri, Alvarez, & Enns, 2007) and proto-objects (Pylyshyn, 2003). In terms of the nature of attentional mode used for statistical judgments, the CDA result clearly speaks against the sub-sampling theory. Sub-sampling, that is focusing on one or two items for performing the mean task cannot account for the CDA results with the Mean task since it would predict low CDA amplitudes regardless of set sizes. In contrast, we obtained high CDA amplitudes (representing maximum loading of working memory) for all the set sizes used in the study. This result rules out the sub-sampling hypothesis that argues that the better performance in the Mean task is based on one or two items at best. The difference between CDA amplitudes even with set size 2 that is well within the limits of focused attention or working memory indicates that there is a significant difference in processing mechanisms underlying the two tasks. Could a non-attentional factor such as task difficulty, be a cause of the CDA results? Though there still is a debate over the interpretation of CDA amplitude, previous studies seem to predict lower amplitude for the easier task and/or greater processing efficiency (Ikkai, McCollough, & Vogel, 2010; Jost et al., 2011; Woodman & Arita, 2010). This contradicts with the current results, which show higher amplitude in the easier task compared to the more difficult task. It is to be noted that this difference in CDA amplitude for set size 2 was present even though the behavioral performance and confidence rating for the set sizes was not different between the Mean and the Member tasks. Hence, the differences in CDA amplitudes between mean and member tasks cannot be simply attributed to task difficulty or behavioral performance in the mean task compared to the member task. Though the precise nature of the representations used for mean judgment is still unclear, the current CDA results, at the very least, showed that the representations involved in the statistical judgment of objects are different from those for focused attention tasks. The representation was not a sub-sample or a highly compressed version of the all targets, but was less-organized pieces or featural representations that would fill up the working memory. Acknowledgments We thank Dr. Preeti Verghese and two anonymous reviewers for their helpful comments. References Alvarez, G. A. (2011). Representing multiple objects as an ensemble enhances visual cognition. Trends in Cognitive Sciences, 15(3), 122 131. Alvarez, G. A., & Oliva, A. (2009). Spatial ensemble statistics are efficient codes that can be represented with reduced attention. Proceedings of the National Academy of Sciences, USA, 106, 7345 7350. Ariely, D. (2001). Seeing sets: Representation by statistical properties. Psychological Science, 12, 157 162. Awh, E., & Pashler, H. (2000). Evidence for split attentional foci. Journal of Experimental Psychology: Human Perception and Performance, 26, 834 846. Baddeley, A. (1997). Human memory: Theory and practice. Hove and New York: Psychology Press. Baijal, S., & Srinivasan, N. (2011). Consolidation of statistical information of multiple objects in working memory. Attention, Perception, & Psychophysics, 73, 1733 1741. Bays, P. M., & Husain, M. (2008). Dynamic shifts of limited working memory resources in human vision. Science, 321, 851 854. Brady, T. F., & Alvarez, G. A. (2011). Hierarchical encoding in visual working memory: Ensemble statistics bias memory for individual items. Psychological Science, 22, 384 392. Cavanagh, P., & Alvarez, G. A. (2005). Tracking multiple targets with multifocal attention. Trends in Cognitive Sciences, 9, 349 354. Chong, S. C., & Evans, K. K. (2011). Distributed versus focused attention (count vs estimate). WIRES Cognitive Science, 2, 634 638. Chong, S. C., Joo, S. J., Emmanouil, T. A., & Treisman, A. (2008). Statistical processing: Not so implausible after all. Perception and Psychophysics, 70, 1327 1336. Chong, S. C., & Treisman, A. (2003). Representation of statistical properties. Vision Research, 43, 393 404. Chong, S. C., & Treisman, A. (2005a). Statistical processing: Computing the average size in the perceptual groups. Vision Research, 45, 891 900. Chong, S. C., & Treisman, A. (2005b). Attentional spread in the statistical processing of visual displays. Perception & Psychophysics, 67, 1 13. Cowan, N. (2005). Working memory capacity. New York and Hove: Psychology Press. Drew, T., & Vogel, E. K. (2008). Neural measure of individual differences in selecting and tracking multiple moving objects. Journal of Neuroscience, 28, 4183 4191. Franconeri, S. L., Alvarez, G. A., & Enns, J. T. (2007). How many locations can be selected at once? Journal of Experimental Psychology: Human Perception and Performance, 33, 1003 1012. Ikkai, A., McCollough, A. W., & Vogel, E. K. (2010). Contralateral delay activity provides a neural measure of the number of representations in visual working memory. Journal of Neurophysiology, 103, 1963 1968. Jost, K., Bryck, R. L., Vogel, E. K., & Mayr, U. (2011). Are old adults just like low working memory young adults? Filtering efficiency and age differences in visual working memory. Cerebral Cortex, 21, 1147 1154. Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390, 279 281. Machizawa, M. G., Goh, C. C., & Driver, J. (2012). Human visual short-term memory precision can be varied at will when the number of retained items is low. Psychological Science, 23, 554 559. Makovski, T., & Jiang, Y. V. (2007). Distributing versus focusing attention in visual short-term memory. Psychonomic Bulletin & Review, 14, 1 7. Myczek, K., & Simons, D. J. (2008). Better than average: Alternatives to statistical summary representations for rapid judgments of average size. Perception & Psychophysics, 70, 772 788. Pylyshyn, Z. (2003). Seeing and visualizing. Cambridge: MIT Press. Pylyshyn, Z. W., & Storm, R. W. (1988). Tracking multiple independent targets: Evidence for a parallel tracking mechanism. Spatial Vision, 3, 179 197. Treisman, A. (2006). How the deployment of attention determines what we see. Visual Cognition, 14, 411 443. Vogel, E. K., & Machizawa, M. G. (2004). Neural activity predicts individual differences in visual working memory. Nature, 428, 748 751. Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal individual differences in controlling access to visual working memory. Nature, 438, 500 503. Woodman, G. F., & Arita, J. T. (2010). Direct electrophysiological measurement of attentional templates in visual working memory. Psychological Science, 22, 212 215.