Abstract

We bisected the sequence of processing into operations taking place before or after the engagement of visual–spatial attention during a difficult search task using event-related potentials. We were able to assign variance in RTs associated with experimental factor effects to phases of processing by examining stimulus-locked (SLpcN) and response-locked (RLpcN) posterior contralateral negativity. Participants searched for a gray square with one gap among gray squares with two gaps. The number of displayed items (set size) and the number of response alternatives were varied. Both experimental manipulations affected the onset latency of the RLpcN, whereas the SLpcN showed small or no latency effects, suggesting they had effects after the initial deployment of attention. Moreover, amplitude effects in the RLpcN and SLpcN behaved similarly. Most importantly, different aspects of the RLpcN dissociated the experimental manipulations: Set size primarily affected processing between RLpcN onset and peak amplitude of the RLpcN, whereas the number of response alternatives affected the onset latency and the latency of peak amplitude of RLpcN. These results show how RLpcN activity can dissociate factor effects that are not separable with SLpcN activity during difficult search.

INTRODUCTION

Visual search has been a powerful experimental tool for the study of perception and cognition, as well as the role of attention, in processing from stimulus presentation to the response. Two general search modes have been proposed: parallel search and serial search (Treisman & Gelade, 1980). Parallel search (or feature search) was defined as the processing of all presented information simultaneously when the target item is sufficiently salient or unique in some way that it “pops out.” Serial search, on the other hand, was postulated when the target item is difficult to discriminate from the distractors (i.e., when the target is not sufficiently unique compared with the distractors) and requires item-by-item processing to tease them apart. Much of the work following the proposed parallel–serial search dichotomy has demonstrated that visual search operates more on a continuum and that search tasks are not always easily placed in one of two categories (Wolfe, 1994, 1998; Humphreys & Muller, 1993; Duncan & Humphreys, 1989; Wolfe, Cave, & Franzel, 1989). For this reason, visual search can instead be described in terms of efficiency, where pop-out search is “very efficient” and strict serial search is “very inefficient” (Wolfe & Gray, 2007; Wolfe, 1994, 1998).

The purpose of the present work was to isolate processes engaged during difficult (or inefficient) search by combining mental chronometry and electrophysiological measures, which allowed us to track the temporal sequence of events and bisect them into major subphases. Difficult search has been studied extensively using behavioral measures (e.g., Wolfe, 1998; Duncan & Humphreys, 1989; Treisman & Gelade, 1980), but research examining difficult search using EEG has been more limited. In most of the work using EEG in visual search, the targets and distractors are chosen to reduce the variance in the time taken to find the target, which is particularly useful when using the event-related potential (ERP) technique. When a target is easy to find, for example, because of a clear difference in color or shape relative to distractors, attention is consistently deployed to the target quickly. Event-related averaging relative to the onset of a search display containing a lateral stimulus to which attention is deployed (i.e., either in the left or right visual field) generally produces a clear N2pc component. The N2pc has been shown to be a useful index of the deployment of visual spatial attention to lateral stimuli (see Luck & Kappenman, 2012, for a review) and is characterized by a greater negativity at posterior sites contralateral to the attended visual field (relative to ipsilateral sites), with an onset latency of approximately 200 msec and a return to baseline approximately 300 msec after stimulus onset (Eimer, 1996; Luck & Hillyard, 1994a, 1994b).

Luck and Hillyard (1990) measured the N2pc during pop-out and difficult search conditions (see also Luck & Kappenman, 2012). In the pop-out search task, the target was a triangle with a horizontal dash, and the distractors were triangles without a dash. Because the target possessed an additional, unique, feature compared with distractors, it was easily located. For the difficult search task, the target and distractor definitions were inversed: The target was a triangle with no dash, and the distractors were triangles with a dash. Now, the target was defined by the absence of a feature and significantly more difficult to find (see Treisman & Souther, 1985). For pop-out search, a discrete N2pc component was observed from approximately 200 to 300 msec. For the difficult search task, the N2pc onset was still at approximately 200 msec; however, the component was sustained and did not return to baseline. The authors suggested attention was engaged on candidate stimuli, one after the other, until the target was found, varying component offset latency from trial to trial and producing a sustained wave in the aggregate (Luck & Kappenman, 2012). In contrast to pop-out search, the first stimulus selected during a difficult search task is often a distractor, rather than the target.1 Thus, the time at which the target was found was more variable during difficult search, and the components underlying visual search, including the N2pc, were smeared, resulting in the observed sustained activity. Therefore, examining this component via stimulus-locked averaging provides little information regarding the underlying mechanisms of difficult search because the onset of N2pc will reflect engagement on the first attended stimulus, whether it is a target or a distractor. Moreover, the resulting sustained component (the stimulus-locked posterior contralateral negativity [SLpcN]) was likely a combination of multiple N2pc responses mixed with sustained posterior contralateral negativity (SPCN) activity, where the latter is an index of visual short-term memory (VSTM) or visual information maintained for subsequent processing after the N2pc but with similar ERP characteristics (Jolicœur, Brisson, & Robitaille, 2008; Dell'Acqua, Sessa, Jolicœur, & Robitaille, 2006; Vogel & Machizawa, 2004). Although pop-out stimuli are useful to ensure a component associated with visual spatial attention tightly time-locked to stimulation, the associated search tasks are generally very easy and not representative of a large class of search conditions, leading to difficult search. In the present work, we sought to extend the usefulness of N2pc/SPCN (or SLpcN) to a broader class of search conditions.

A response-locked counterpart to the SLpcN with a similar scalp distribution and magnitude was recently reported and named the RLpcN (response-locked posterior contralateral negativity; Drisdelle & Jolicœur, 2019). Results from a pop-out search task showed two stimulus-locked components (the N2pc and the SPCN) and a response-locked counterpart, marked by what appears to be a single contralateral negative activity with an onset that varies depending on the duration of processing after the initial deployment of attention. Drisdelle and Jolicœur (2019) argued, and presented evidence, that the initial portion of the RLpcN waveform represents mainly N2pc activity (i.e., the initial deployment of attention), whereas later portions of the RLpcN represent mainly SPCN activity (subsequent processing of visual information in VSTM). In other words, this response-locked component is a time-varying mixture of both components and elucidates the time course of these processes relative to the response (the lateralized readiness potential has been examined using the same comparison; see Osman, Moore, & Ulrich, 1995). Because processing after the initial deployment of attention varies with RT, the duration of the RLpcN also varies with RT (i.e., more time passes between the onset and the response for longer RTs), making this component a useful tool to bisect processing responsible for RT variations. Using an easy pop-out search task, Drisdelle and Jolicœur (2019) observed RLpcN waveforms with onsets corresponding to the initial deployment of attention (ranging from around 500 to 300 msec before the response depending on the condition). The search task was to locate a square unique in color (either orange or green) and to indicate whether a gap in this square was on the top face or not, among three distractors (blue squares with gaps as well). In their second analysis, electrophysiological data were separated into tertiles according to RT both between trials (within-subject) and between participants (between-subjects). The onset of the RLpcN was later (i.e., closer to the response) when RTs were shorter, because less time passed between the onset of the initial engagement of attention (RLpcN onset) and response execution. Importantly, very little differences in component onset latencies were observed for the N2pc and SPCN, suggesting that RT variation was primarily determined by the duration of processes occurring after the onset of engagement of attention on the target. Overall, Drisdelle and Jolicœur (2019) demonstrated the usefulness of examining both stimulus-locked and response-locked data to pinpoint the locus of RT variability within the stream of information processing relative to the engagement of visual–spatial attention.

Here, we extended the approach described in Drisdelle and Jolicœur (2019) in two ways. First, we included explicit experimental manipulations expected to affect processing after the deployment of visual–spatial attention. Second, we examined difficult search, where variability in the time to find the target was expected to smear factor effects in the SLpcN, but not in the RLpcN. Moreover, the SLpcN should not show a clear demarcation in the EEG signal between the N2pc and the SPCN (as observed in pop-out visual search tasks), similar to the results of Luck and Hillyard (1990). We aimed to understand better the sequence of operations using experimental manipulations that were expected to affect different stages of processing and modulate the latency of RLpcN. We show that, by performing both stimulus-locked and response-locked averaging, we were able to elucidate the dynamics of cognitive processes occurring after the initial deployment of attention (see also Cosman, Arita, Ianni, & Woodman, 2016; Töllner, Rangelov, & Müller, 2012; Hackley, Schankin, Wohlschlaeger, & Wascher, 2007).

Our visual search paradigm consisted of a bilateral display of gray items containing a target square that had a side with one gap among distractor squares that had a gap in each of two sides. A cue presented before the onset of the search display indicated the visual hemifield containing the target. We manipulated the time required to find and identify the visual target via a manipulation of the number of items in the display (set size). Because the target was similar to the distractors, we expected discrimination between the target and distractors would require focal attention and produce a posterior contralateral negativity. Adding more distractors would lengthen search time, likely because attention would be engaged on some distractors before finding the target. This sequence was expected to generate a set of rapidly overlapping N2pc–SPCN sequences, which would be observed as a single sustained component for both stimulus-locked and response-locked averages. No differences in onset of the SLpcN across different set sizes were expected because the time to deploy attention to the first item(s) in the attended hemifield was not expected to depend on the number of items in the display. We also manipulated the time required to select a response by varying the number of response alternatives associated with the gap location in the target square. We expected that response selection would begin after the target had been found, and so there should be no differences between the levels of this manipulation in the stimulus-locked average (SLpcN). For the RLpcN, we expected a longer duration between component onset (initial deployment of attention) and the response when more processing was required after the initial attentional deployment, that is, when the response selection process would take more time, as well as when more items were in the search display.

Two experiments were conducted. In Experiment 1, there were two, four, or six items in each hemifield and there were either two or four response alternatives. A second experiment was conducted to generalize and replicate the results obtained on Experiment 1. In this second experiment, there were either one, two, three, or four items in each hemifield and the response-selection difficulty manipulation was similar to that in Experiment 1.

EXPERIMENT 1

The goal of the first experiment was to identify markers of the duration of the inspection of displayed items during difficult visual search from the process of response selection once the target was found. The number of items (set size) and the difficulty of response selection (number of response alternatives) were manipulated. Both stimulus-locked (SLpcN) and response-locked (RLpcN) activities were examined to observe processing associated with the initial deployment of visual spatial attention (onset of the SLpcN and RLpcN) and subsequent cognitive processing up until the response. The task induced a relatively slow search process because all stimuli were of the same color (gray) and did not possess unique features, which made the target difficult to discriminate from distractors. Thus, stimuli were likely selected and processed individually (or in small groups) to find the target. For this reason, we expected to observe a sustained lateralized stimulus-locked component similar to that obtained by Luck and Hillyard (1990) as well as a response-locked counterpart (RLpcN; Drisdelle & Jolicœur, 2019).

For the manipulation affecting response selection, participants had either two or four response alternatives after the selection of the target (two: the gap was on top of the square or not; four: which side of the square had a gap). We expected that the number of response alternatives would have different effects on the onset latency of the SLpcN and the RLpcN. Because the onset of the SLpcN represents the engagement of attention to the first selected item (which, in difficult search, could be the target or a distractor), we did not expect any modulations of SLpcN onset latency, given that response selection should logically take place after the onset of attention. We did, however, expect an earlier RLpcN onset when response selection required more processing time (i.e., there were more response alternatives), reflecting time elapsed after the initial engagement of attention.

We also manipulated the number of items in each hemifield (either two, four, or six). As with the response selection manipulation, set size was not anticipated to affect the onset of the SLpcN. We did expect the set size manipulation to affect the onset of the RLpcN relative to the response, given that a larger set size would be associated with a longer mean search time, thereby lengthening the period between the initial deployment of visual spatial attention and response execution.

In addition to latency, we examined the amplitude of the components for each manipulation. We expected that the stimulus-locked and response-locked components reflect a combination of continued visual search (when the first item selected and identified is not the target) as well as subsequent processing/maintenance of selected stimuli in VSTM (SPCN activity). The SPCN is characterized as a negativity contralateral to the attended hemifield, after the N2pc in time, and has been shown to reflect maintenance of information in VSTM (Maheux & Jolicœur, 2017; Mazza & Caramazza, 2011; Jolicœur, Sessa, Dell'Acqua, & Robitaille, 2006).2 Researchers have suggested that this component reflects an intermediate processing buffer between the initial deployment to the relevant stimuli (reflected by the N2pc) and processing involved in identification and response selection (e.g., Maheux & Jolicœur, 2017). For example, when a certain attribute of the target presented in the visual field must be reported (e.g., a forced-choice response task where participants reported the identity of digits; Jolicœur et al., 2008), information regarding the attributes of the target must be extracted, and it is suggested that this process requires passage through VSTM. Following this logic, when a task is relatively simple, there is less of a need for further processing of the target after the initial deployment of attention. However, when target identification is more difficult, such as in the present work, more time and processing will be required to determine the identity of the selected item. A smaller SPCN would therefore be observed for a simple task, whereas when the task is more difficult, SPCN amplitude would be larger (reflecting a higher probability of information passage into VSTM for further processing). In the present task, distinguishing the target from distractors was relatively difficult, and items were likely selected, submitted for further processing, and identified, one after another (although not necessarily in discrete stages). We therefore hypothesize that the probability of passage into VSTM for further processing of every selected item would very likely be at ceiling. For this reason, we did not expect any amplitude effects in the SLpcN or RLpcN (reflecting sustained processing during visual search) between set sizes. Finally, because the number of response alternatives was not anticipated to affect activity until after a target was located and selected (i.e., after SPCN-related activity within the SLpcN), we did not anticipate a difference in amplitude in either stimulus-locked or response-locked activity.

Methods

Participants

Forty-seven adults volunteered and were compensated $20 for participating in a procedure vetted by the Ethics Committee of the Faculty of Arts and Science at Université de Montréal. They reported having no neurological or psychiatric problems, were not taking psychoactive medication, and had normal (or corrected-to-normal) visual acuity as well as normal color vision. Participants completed a questionnaire regarding personal information and a signed consent form before beginning the experiment. Thirty-two adults (age: M = 22.52 years, SD = 4.14 years; 27 women; two left-handed) were kept for the final analysis, and 15 participants were excluded from the final analysis (see the EEG recordings section for details).

Procedure

Participants were seated in a dimly lit room at 57 cm from a computer screen on which the stimuli were displayed. The experiment was programmed under MATLAB (Mathworks, 2009) and the Psychophysics Toolbox (Kleiner et al., 2007; Brainard, 1997; Pelli, 1997).

Visual Stimuli

An illustration of the stimulus display can be seen in Figure 1. The display consisted of two hemifields (on the left and right of fixation) each containing gray squares (size: 1° × 1°) placed in a 4 × 4 grid (i.e., 16 possible locations). In each hemifield, one square had a gap on one side (on the cued side this stimulus was the target), and the remaining squares (the distractors) had two gaps on different sides (gap size: 0.33°). Before the presentation of the squares, a cue (a pair of arrows, one above and one below fixation) indicated the task-relevant hemifield. A square never occurred in the corners of the 4 × 4 grid, resulting in 14 possible stimulus locations on each side of fixation or hemifield (28 positions in total). Each grid cell was 2° × 2° in size. The center of the box, within each grid cell, was jittered (0.3°) so that it could be between 0.7° and 1.3° from the center of the grid vertically and horizontally. Boxes could not be within 10% of their grid cell outer limit, so that, if there was a box in the neighboring grid cell, they were not too close and could not overlap. More specifically, there was a boundary of 0.2° (0.7°–0.5° = 0.2° and 1.3° + 0.5° = 1.8°, where 0.5° is the distance from the center to the edge of the stimuli [1° / 2 = 0.5°]). Stimuli were dispersed among the grid positions randomly. For each trial, the display had either two, four, or six stimuli on each side of fixation consisting of a stimulus with one gap and either one, three, or five stimuli with two gaps. The number of items presented in each hemifield was always equal. There were four possibilities for the stimulus with one gap (gap on top, right, left, or bottom) and six possibilities for stimuli with two gaps (gaps on top bottom, top right, top left, right left, right bottom, or left bottom). The target stimuli possibilities were presented with equal probability (p = .25), and the distractors with two gaps were selected among the six possibilities at random. For the square with one gap on the uncued side, the gap location was selected among the three other possibilities than that of the target on the cued side.

Figure 1. 

Stimulus events in Experiments 1 and 2. All stimuli were gray, and the hemifield containing the target was indicated by an informative cue near fixation. Distractors had two gaps, whereas the target only had one. The task required finding the square with one gap and performing a speeded button-press response in either a 2AD response or 4AD response task. Accuracy feedback was presented after the response (see text for further details).

Figure 1. 

Stimulus events in Experiments 1 and 2. All stimuli were gray, and the hemifield containing the target was indicated by an informative cue near fixation. Distractors had two gaps, whereas the target only had one. The task required finding the square with one gap and performing a speeded button-press response in either a 2AD response or 4AD response task. Accuracy feedback was presented after the response (see text for further details).

Visual Search Task

The experiment was split into two parts for all participants: Half of the experiment required a two-alternative discrimination (2AD) response, whereas the other half required a four-alternative discrimination (4AD) response. The order of these two halves (2AD–4AD vs. 4AD–2AD) was counterbalanced across participants. For each response alternative task, there were two possible response mappings. For the 2AD condition, half of the participants (n = 16) pressed the “v” key for a gap on top and the “b” key for a gap on any of the other three sides (mapping 2AD-1), and this assignment was reversed for the other half (n = 16; mapping 2AD-2). For the 4AD condition, half of the participants (n = 16) pressed “c” for up, “v” for right, “b” for down, and “n” for left (mapping 4AD-1), whereas the other half (n = 16) pressed “n” for up, “b” for right, “v” for down, and “c” for left (mapping 4AD-2).3 Response hand was manipulated between participants (16 participants were instructed to respond with the right hand, and the other 16 participants with the left hand).

There was a delay of 2000 msec before the first trial at the start of each block. A fixation cross (+) was then presented at the center of the screen, and participants were instructed to maintain fixation and refrain from blinking during the trial. Trials were initiated by pressing the spacebar. After an average of 500 msec (± 200-msec jitter) after trial initiation, the fixation cross was presented with two arrows (two lesser than [<] or greater than [>] signs placed vertically above and below the fixation cross; 600 msec ± 200-msec jitter), signaling which hemifield (left or right of fixation) to search. Between the arrows and the onset of the visual search display, there was a brief delay (500 msec ± 200-msec jitter). The visual search display was presented for a maximum of 5000 msec, during which participants located the target stimuli on the cued side (the square with one gap) and made their response. For the 2AD block, participants indicated whether the gap was on the top of the square or not. For the 4AD block, participants indicated the location of gap (top, right, down, or left). Feedback (500 msec) was a cross (+) for a correct response, a dash (-) for an incorrect response, and a vertical line (|) if a response was not made within 5000 msec (timeout). For each number of response alternatives (2AD or 4AD), there was a practice block of 16 trials followed by four experimental blocks of 96 trials (384 experimental trials per response alternative block or 768 trials overall).

The orders of response alternative blocks (two levels: 2AD first or second), 2AD response key mapping (two levels: 2AD-1, 2AD-2), 4AD response key mapping (two levels: 4AD-1, 4AD-2), and response hand (two levels: left or right) were counterbalanced between participants. For each participant and each response alternative block (2AD or 4AD), the number of items (three levels: 2, 4, or 6 in each hemifield), gap position on the target (four levels: up, right, down, or left), and cued side (two levels: left or right) gave a 3 × 4 × 2 within-subject design. Each of the 24 possible combinations in the within-subject design was presented four times in each block of 96 trials.

EEG Recordings

The EEG were recorded with 64 Ag–AgCl electrodes using the International 10–10 system (at sites Fp1, Fpz, Fp2, AF7, AF3, AFz, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8, FT7, FC5, FC3, FC1, FCz, FC2, FC4, FC6, FT8, T7, C5, C3, C1, Cz, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPz, CP2, CP4, CP6, TP8, P9, P7, P5, P3, P1, Pz, P2, P4, P6, P8, P10, PO7, PO3, POz, PO4, PO8, O1, Oz, O2, and Iz; Sharbrough, 1991) placed on an elastic cap. Data were recorded using the BioSemi ActiveTwo system and Actiview (BioSemi B. V.). Signals at two external electrodes placed on the left and right mastoids were averaged after recording and used for offline referencing. A horizontal EOG (HEOG), used to measure horizontal eye movements, was defined as the voltage difference (calculated offline) between two external electrodes placed on the external canthi. A vertical EOG (VEOG), used to measure eye blinks and vertical eye movements, was defined as the voltage difference (calculated offline) between an external electrode placed below the left eye and Fp1 (located above the left eye). EEG analyses were done with MATLAB with toolboxes EEGLAB (Delorme & Makeig, 2004) and ERPLAB (Lopez-Calderon & Luck, 2014), using a custom GUI designed by our laboratory. EEG data were high-pass filtered at 0.01 Hz and low-pass filtered at 30 Hz, and EOG data were high-pass filtered at 0.1 and low-pass filtered at 10 Hz. To correct blink-related ocular activity, an independent component analysis (ICA; Makeig, Bell, Jung, & Sejnowski, 1996) was applied to the continuous data (for the procedure details, see Drisdelle, Aubin, & Jolicoeur, 2017). After segmentation (described in the following section), artifacts not corrected with ICA were identified using the VEOG (deflection < 50 μV within a time window of 150 msec) for vertical eye movements and the HEOG (deflection < 35 μV within a time window of 300 msec) for horizontal eye movements. A visual inspection of the averaged lateralized HEOG activity for each participant was then performed, and only participants with less than a 5-μV deviation toward the target (contralateral minus ipsilateral activity) were kept. For trials with seven or fewer channels containing artifacts (activity exceeding ± 100 μV in amplitude during the trial), those channels containing artifacts were interpolated, only for that trial, using the EEGLAB spherical spline interpolation function. When there were more than seven channels containing artifacts, the trial was rejected. Only trials with a correct response were included for the final analysis.

Fifteen participants were excluded from the final analysis for the following reasons: Seven participants had too many saccades (five participants had an HEOG deflection > 5 μV and two participants lost over 50% of trials because of saccade-related activity), seven participants had low accuracy (60% and below), and one participant did not complete the task.

EEG Segmentation

Each trial kept for the final analysis produced two epochs, one segmented relative to the stimulus onset and the other relative to the response. Stimulus-locked epochs used a time window from −200 msec prestimulus to 1000 msec poststimulus, and response-locked epochs used a time window from −2000 msec preresponse to 200 msec postresponse. The time window of response-locked time window was selected to include activity based on the group mean RT (M = 1539 msec, SD = 240 msec), so the average moment of the initial deployment of attention (onset of the SLpcN) was included in the epoch. Both stimulus-locked and response-locked segmentations were baseline corrected using the mean voltage during a 200-msec period immediately before stimulus presentation.

Measurements and Statistics

The SLpcN and RLpcN components are lateralized components that were obtained by subtracting ipsilateral activity (activity from electrodes over the left hemisphere for left targets and activity from electrodes over the right hemisphere for right targets) from contralateral activity (activity from electrodes over the left hemisphere for right targets and activity from electrodes over the right hemisphere for left targets). Time windows for the amplitude in activity of interest were selected based on a time window centered at the peak of the grand-averaged waveform (see Figure 3).4

For amplitude, the SLpcN component was measured from 500 to 900 msec; and the RLpcN, from −650 to −100 msec. For latency, estimates were based on the fractional area latency of jackknife curves, for each type of average (SLpcN and RLpcN). The fractional area latency is the time at which the cumulative area between a curve and a fixed amplitude (often 0 μV) is at a certain percentage of the area in a given window (for a more detailed description and validation of the jackknife approach with fractional area measurements, see Kiesel, Miller, Jolicœur, & Brisson, 2008). In our application, we computed the area over the specified window (for component onsets: from 200 to 650 msec for the SLpcN and from −1800 msec to the response for the RLpcN) and then found the latency at which the specified percentage of a total area was accumulated when sweeping from left to right (30% for the SLpcN and 40% for the RLpcN). The area used for the SLpcN and RLpcN was between the curve and 0 μV.

ANOVAs were used to evaluate the patterns of activity in stimulus-locked and response-locked epoch averages separately as well as for behavioral analyses. When sphericity was violated, Greenhouse–Geisser corrections for nonsphericity were applied.5 Reported effect sizes for significant ANOVA results were determined using the generalized eta squared (ηG2; Bakeman, 2005; Olejnik & Algina, 2003). For behavioral analysis, RTs below 150 msec were removed, as well as RT outliers, using the recursive method described in Van Selst and Jolicoeur (1994). For the analysis examining experimental conditions, Bonferroni-corrected pairwise comparisons were used to decompose interactions. Behavioral and electrophysiological data were subjected (separately for accuracy and RT and for all components) to 2 (Number of response alternatives: 2AD or 4AD) × 2 (Block order of response alternatives: 2AD first or 4AD first) × 3 (Set size: two, four, or six visual items in each hemifield) mixed ANOVAs. Confidence intervals were reported with means. For within-subjects confidence intervals with a 95% confidence level, the Cousineau–Morey method was used (Baguley, 2012; Morey, 2008; Cousineau, 2005).

To assess trial by trial-to-trial variance in RT of our within-subjects experimental manipulations (number of response alternatives and set size), an estimate of the within-subjects standard deviation was calculated. First, the variance of each condition, separately for each participant, was obtained and then averaged across participants. The square root of these average variances was then taken, giving an estimate of standard deviation.

Results

Behavioral Results

Accuracy.

Mean accuracy and 95% within-subjects confidence intervals for each condition in Experiment 1 can be seen in Table 1 and Figure 2. Accuracy was higher when there were two response alternatives in comparison with when there were four, F(1, 30) = 20.30, p < .0001, ηG2 = .15. Accuracy decreased as set size increased, F(2, 60) = 127.2, p < .0001, ηG2 = .33; all pairwise differences were significant, ts(31) > 7.52, ps < .0001. A significant interaction between Set size and the Number of response alternatives was observed, F(2, 60) = 8.84, p = .001, ηG2 = .01, which appears to be caused by a larger decrease in accuracy with an increase in set size for the 4AD condition compared with the 2AD condition (see Figure 2). An interaction was also observed between the Order of the response alternative blocks (i.e., whether the 2AD or 4AD was presented first) and the Number of response alternatives, F(1, 30) = 4.78, p = .04, ηG2 = .04, which was driven by a larger difference in accuracy between response alternative conditions when the 4AD block was performed first (2AD vs. 4AD: acc = .88 vs. .77, respectively; Δ = .11) compared with when the 2AD block was performed first (2AD vs. 4AD: acc = .84 vs. .80, respectively; Δ = .04) and likely a practice effect, where participants tend to be more accurate in the second half of the experiment. All other effects and interactions were not significant (all Fs < 1).

Table 1. 
Results from Experiment 1
Response AlternativesSet SizeAccuracyRT (msec)
M95% CIM95% CIEst. of SD
2AD .92 ±.020 1025 ±60 ±329 
.85 ±.023 1302 ±60 ±486 
.80 ±.021 1513 ±86 ±603 
  
4AD .87 ±.023 1498 ±60 ±549 
.79 ±.019 1824 ±63 ±695 
.71 ±.029 2073 ±67 ±826 
Response AlternativesSet SizeAccuracyRT (msec)
M95% CIM95% CIEst. of SD
2AD .92 ±.020 1025 ±60 ±329 
.85 ±.023 1302 ±60 ±486 
.80 ±.021 1513 ±86 ±603 
  
4AD .87 ±.023 1498 ±60 ±549 
.79 ±.019 1824 ±63 ±695 
.71 ±.029 2073 ±67 ±826 

Accuracy (mean proportion correct) and mean RT (msec) with 95% within-subjects confidence intervals (95% CI) for each combination of the number of response alternatives (2AD or 4AD) and set size (2, 4, or 6 items). Estimations of within-subject standard deviation (Est. of SD) are also provided for RT (msec).

Figure 2. 

Accuracy (mean proportion correct; left) and RT (sec; right) in Experiment 1 as a function of the number of response alternatives and set size (two, four, or six items in each hemifield). Error bars show 95% within-subjects confidence intervals.

Figure 2. 

Accuracy (mean proportion correct; left) and RT (sec; right) in Experiment 1 as a function of the number of response alternatives and set size (two, four, or six items in each hemifield). Error bars show 95% within-subjects confidence intervals.

RT.

Mean RTs with 95% within-subjects confidence intervals for Experiment 1 can be seen in Table 1 (which also shows estimates of within-subjects SD) and Figure 2. Participants were quicker when there were two response alternatives compared with when there were four, F(1, 30) = 150.76, p < .0001, ηG2 = .47. There was also a significant increase in mean RT as the number of items presented increased, F(2, 60) = 223.84, p < .0001, ηG2 = .39. This effect is supported by t tests, which show that participants were slower when there were more items (all pairwise differences were significant: all ts(31) > 10.42, ps < .0001). The number of response alternatives interacted with set size, F(2, 60) = 3.50, p < .05, ηG2 = .004, which was driven by a slightly larger set size effect in the 4AD condition compared with the 2AD condition, as can be seen in Figure 2. The number of response alternatives also interacted with the order with which the response alternative blocks were presented, F(1, 30) = 10.07, p = .003, ηG2 = .06. The interaction reflects a larger RT difference between the 2AD and 4AD conditions when the 4AD block was presented first (2AD vs. 4AD: 1269 vs. 1921 msec, respectively; Δ = 652 msec) compared with when the 2AD block was presented first (2AD vs. 4AD: 1291 vs. 1676 msec, respectively; Δ = 385 msec), likely because of a practice effect, similar to accuracy. All remaining effects and interactions were not significant (all Fs < 2.02, all ps > .15).

Electrophysiological Results

Stimulus-locked (SLpcN component).

The stimulus-locked average was marked by a sustained posterior contralateral negativity (SLpcN) beginning roughly 200 msec after stimulus onset (see Figure 3). For the mixed ANOVA (2 [Number of response alternatives: 2AD or 4AD] × 2 [Order of response alternative blocks: 2AD first or 4AD first] × 3 [Set Size: two, four, or six items in each hemifield]), there were no significant main effects or interactions for all conditions (amplitude: all Fs < 1; latency: all Fs < 3.06, all ps > .056; see Table 2). Thus, as predicted, there were no differences between conditions for the SLpcN.

Figure 3. 

Results from Experiment 1. Grand-averaged SLpcN (left) and RLpcN (right) contralateral minus ipsilateral activity at electrodes PO7/PO8. Amplitude measurement windows used for statistical analyses are indicated in gray at the top (SLpcN: 500–900 msec; RLpcN: −650 to −100 msec). The top row of figures shows waveforms for each display set size (two items: pink lines; four items: red lines; six items: dark red lines). The bottom row of figures shows waveforms for either two (blue lines) or four (green lines) response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes only.

Figure 3. 

Results from Experiment 1. Grand-averaged SLpcN (left) and RLpcN (right) contralateral minus ipsilateral activity at electrodes PO7/PO8. Amplitude measurement windows used for statistical analyses are indicated in gray at the top (SLpcN: 500–900 msec; RLpcN: −650 to −100 msec). The top row of figures shows waveforms for each display set size (two items: pink lines; four items: red lines; six items: dark red lines). The bottom row of figures shows waveforms for either two (blue lines) or four (green lines) response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes only.

Table 2. 
Results for the SLpcN in Experiment 1
Response AlternativesSet SizeSLpcN Onset Latency (msec)SLpcN Amplitude (μV)
M95% CIM95% CI
2AD 416 ±12 −3.10 ±0.60 
413 ±16 −2.90 ±0.52 
399 ±13 −2.61 ±0.46 
  
4AD 427 ±14 −2.88 ±0.50 
408 ±14 −3.13 ±0.51 
414 ±17 −3.03 ±0.46 
Response AlternativesSet SizeSLpcN Onset Latency (msec)SLpcN Amplitude (μV)
M95% CIM95% CI
2AD 416 ±12 −3.10 ±0.60 
413 ±16 −2.90 ±0.52 
399 ±13 −2.61 ±0.46 
  
4AD 427 ±14 −2.88 ±0.50 
408 ±14 −3.13 ±0.51 
414 ±17 −3.03 ±0.46 

Mean (M) measurements of onset latency (msec) and amplitude (μV) with 95% within-subjects confidence intervals (95% CI), for each combination of the number of response alternatives (2AD or 4AD) and set size (2, 4, or 6 in each hemifield).

Response-locked (RLpcN component).

The response-locked average was marked by a sustained posterior contralateral negativity (RLpcN) beginning roughly 1800 msec before the response (see Figures 3 and 4). RLpcN latency means with the corresponding 95% within-subjects confidence intervals, separated by set size and number of response alternatives, are given in Table 3.

Figure 4. 

Results from Experiment 1. Grand-averaged RLpcN (contralateral minus ipsilateral activity) waveforms time-locked relative to the response at electrodes PO7/PO8. The top (filled lines) shows waveforms for different set sizes when there were two response alternatives (2AD; two items: pink lines; four items: red lines; six items: dark red lines), whereas the bottom (dashed lines) shows the same comparison but for four response alternatives (4AD). Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Figure 4. 

Results from Experiment 1. Grand-averaged RLpcN (contralateral minus ipsilateral activity) waveforms time-locked relative to the response at electrodes PO7/PO8. The top (filled lines) shows waveforms for different set sizes when there were two response alternatives (2AD; two items: pink lines; four items: red lines; six items: dark red lines), whereas the bottom (dashed lines) shows the same comparison but for four response alternatives (4AD). Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Table 3. 
Results for the RLpcN in Experiment 1
Response AlternativesSet SizeRLpcN Onset Latency (msec)RLpcN Latency of Peak Amplitude (msec)
M95% CIM95% CI
2AD −489 ±80 −314 ±19 
−621 ±74 −333 ±19 
−761 ±111 −347 ±20 
  
4AD −812 ±64 −540 ±17 
−906 ±65 −552 ±19 
−935 ±63 −551 ±23 
Response AlternativesSet SizeRLpcN Onset Latency (msec)RLpcN Latency of Peak Amplitude (msec)
M95% CIM95% CI
2AD −489 ±80 −314 ±19 
−621 ±74 −333 ±19 
−761 ±111 −347 ±20 
  
4AD −812 ±64 −540 ±17 
−906 ±65 −552 ±19 
−935 ±63 −551 ±23 

Mean (M) measurements of onset latency (msec) and latency of peak amplitude (msec) with 95% within-subjects confidence intervals (95% CI), for each combination of the number of response alternatives (2AD or 4AD) and set size (2, 4, or 6 in each hemifield).

The RLpcN had an earlier onset relative to the response (i.e., the component had a longer duration), when participants had four response alternatives compared with two response alternatives, F(1, 30) = 48.92, p < .0001, ηG2 = .21, and with an increase in set size, F(2, 60) = 13.85, p < .0001, ηG2 = .09. To decompose the set size effect, paired t tests were conducted, and significance was corrected for multiple comparisons (Bonferroni correction significance threshold: p = .016). Results show a significant difference between Set Sizes 2 and 6, t(31) = 5.36, p = .0001, and between Set Sizes 2 and 4, t(31) = 3.41, p = .002, and a marginally significant difference between Set Sizes 4 and 6, t(31) = 2.02, p = .05. All other onset latency effects and interactions were not significant (all Fs < 2.36, all ps > .12).

Visual inspection of the curves suggests that, in addition to a difference in RLpcN onset, there could also be a difference in latency patterns between experimental manipulations in the peak amplitude of the RLpcN. The same jackknife method previously described to measure latency onset was used. We calculated the latency at which 50% of the total area under the curve from 900 to 200 msec was reached for the 4AD condition (centered at peak) and 50% of the total area under the curve from 700 msec to the response (centered at peak) for the 2AD condition.7

The rationale for selecting different windows for the 2AD and 4AD conditions was that the curves do not have the same shape and likely represent a combination of processes with separable latencies. Set size appears to affect the onset of attention (RLpcN onset, as expected) to a greater extent than the latency of the RLpcN peak amplitude, where the curves for each set size appear to converge to a peak approximately 400 msec before the response (Figure 3, top right). This does not seem to be the case for the number of response alternatives: The onset latency effect seems to persist beyond the latency of the peak amplitude (Figure 3, bottom right). Because the RLpcN likely represents overlapping activity from both experimental manipulations (set size and number of response alternatives), different parts of the curve could represent different processes that are possibly separable. The pattern of means (see Table 3) supports our measurement window selections and corresponds nicely with the latency of the peak amplitude of the RLpcN for each condition shown in Figure 4. The ANOVA designs used to measure the latency of the RLpcN peak amplitude were the same as those used to measure RLpcN onset latency.

The RLpcN had an earlier latency of the peak amplitude when there were more response alternatives by approximately 215 msec, F(1, 30) = 402.33, p < .0001, ηG2 = .75 (see Figure 3, bottom right), and when there were more distractors by approximately 20 msec, F(2, 60) = 3.74, p = .03, ηG2 = .02 (see Figure 3, top right). To decompose the set size effect, paired t tests were conducted, and significance was corrected for multiple comparisons (Bonferroni correction significance threshold: p = .016). Unlike RLpcN onset latency, which showed differences between all set sizes (save one marginal effect), there was only a significant difference in latency of the RLpcN peak amplitude between Set Sizes 2 and 6 (Δ = 22 msec), t(31) = 3.13, p = .004. No difference was observed between Set Sizes 2 and 4 (Δ = 15 msec), t(31) = 1.91, p = .07, or between Set Sizes 4 and 6 (Δ = 7 msec), t(31) = 0.74, p = .46. There was a marginally significant three-way interaction, F(2, 60) = 3.28, p = .056. For all other main effects and interactions, F < 1.61 and p > .21.

Results from the mixed ANOVA show no significant effects or interactions between conditions for RLpcN amplitude. A marginal effect was observed for the two-way interaction between the Number of response alternatives and Set size, F(2, 60) = 2.56, p = .09. For all other amplitude main effects and interactions, F < 1.

Discussion

No difference in SLpcN latency was observed for the number of items presented (two, four, or six items in each hemifield) or for the number of response alternatives (either 2AD or 4AD; see Table 2 and Figure 3). The onset of the SLpcN likely reflected the time at which attention was engaged on the first item(s) selected during visual search, which was not necessarily the target, giving a component similar in latency for all set sizes. It is likely that the sustained component observed in the stimulus-locked average (SLpcN) was a combination of activity associated with deploying attention from item to item (or small groups of items) as well as to subsequent processing for the selected items. In other words, the SLpcN was likely a combination of overlapping N2pc and SPCN activity. Furthermore, because the onset of the SLpcN was not affected by set size or by the number of response alternatives but a difference in RT was observed for these manipulations, the affected processes were likely downstream of the initial deployment of attention (see also Drisdelle & Jolicœur, 2019; Cosman et al., 2016; Töllner et al., 2012; Hackley et al., 2007).

We were able to observe effects on downstream processing in the response-locked averages: More time elapsed from the onset of the RLpcN to the response when a condition required more search time (more items) or a more complex response selection (more response alternatives). Our results therefore suggest that examining ERPs time-locked to the response can reveal effects and separate processing into substages that were otherwise unobservable in the stimulus-locked averages. Interestingly, the results shown in Figures 3 and 4 suggest that the latency effects observed for set size and response alternative manipulations occurred primarily during different phases of the RLpcN. For the number of response alternatives, the mean difference in latency between four and two response alternatives was 260 msec at onset and 217 msec at peak, suggesting that a large part of the processing associated with response selection occurred between the latency of the RLpcN peak and the response. Regarding the difference at onset, if processing associated with the selection of a response occurred, in large, between the peak amplitude of the RLpcN and the response, a similar latency farther into the waveform is expected because a constant difference would persist (here, we would be starting at the response and moving backward). For the set size manipulation, the difference in latency was much larger and farther from the response (RLpcN onset Δ between Set Sizes 2 and 6: 197 msec) than closer (RLpcN peak Δ between Set Sizes 2 and 6: 22 msec). Although there was a small significant difference between set size conditions at the latency of the RLpcN peak amplitude, the curves for all three set sizes appear to peak at approximately the same latency. For both experimental manipulations, there was more variance at onset than at latency of the peak amplitude. The more variable measurements at the onset of the RLpcN are possibly caused by accrued variance from the moment to which the curves were time-locked (the response) because of an increase in various overlapping processing or trial-to-trial noise.

We suggest that visual search/item processing primarily affected the time course of processing from RLpcN onset to the peak amplitude and response selection complexity generally occurred later in the processing stream (see Figures 3 and 4). In other words, the number of response alternatives manipulation affected curve latency closer to the response, demonstrated by the separation between the two conditions at the latency of the peak amplitude. This in turn suggests that some of the RT variation because of this experimental manipulation was occurring between the latency of the peak RLpcN amplitude and response. For the significant difference in latency of RLpcN peak amplitude for the set size manipulation, one possibility is that the distractors surrounding the target were causing a small amount of interference with response selection. Thus, when there were more items, there was a higher chance for interference from distractors nearby, shifting the latency of the peak amplitude slightly with an increase in set size.

The RLpcN curves for the 2AD and 4AD response alternatives also differed in overall shape, where the 4AD condition had a shallower slope during onset compared with the 2AD condition (see Figure 3). To avoid distorted measurements, we used a separate time window in our measurement of the latency of peak RLpcN amplitude for each of the response alternative conditions. This difference in shape was likely because of a more variable peak amplitude for the 4AD condition compared with the 2AD condition, reflecting increased variance in the time required to select a response. The estimates of within-subjects RT SDs (Table 1) corroborate this suggestion: RTs were more variable for the 4AD condition compared with the 2AD condition.

There were also no differences between conditions (all manipulations) for SLpcN and RLpcN amplitude, suggesting that the memory trace, or information passed on for further processing, did not differ according to the number of items presented or the number of response alternatives.

EXPERIMENT 2

Experiment 2 replicated and generalized the most important new results observed in the RLpcN in Experiment 1. In Experiment 1, the number of response alternatives (2AD or 4AD) produced a large latency effect at the point of peak amplitude in the RLpcN, where the difference in latency for set size was much smaller. In other words, more time passed between the time of peak amplitude of the RLpcN and the response when there were more response alternatives, whereas the RLpcN waves for different set sizes peaked at approximately the same latency. These results suggest that the peak amplitude of the RLpcN component could correspond with the time at which attention was engaged on the target (likely with some presently unknown lag). Once the target was found, we would expect little further influence of the number of displayed items on RT. However, variations in processing time in the response selection portion of the task would now become relevant and cause variations in component behavior in the interval between peak RLpcN amplitude and the response.

These new findings suggest that analyses of the RLpcN waveforms can provide useful indices of fundamental underlying mechanisms mediating performance in difficult search. As such, it is important to verify and extend these results. In Experiment 2, we simplified the response alternative manipulation by removing the different response key mapping alternatives used in Experiment 1, which had no consequence on RTs. We retained the 2AD versus 4AD manipulation, which produced clear (and interesting) results in Experiment 1.

Moreover, in Experiment 1, no difference in SLpcN or RLpcN amplitude was observed between set sizes. As previously mentioned, we had anticipated that the onset latency of the SLpcN would not vary with display set size because distinguishing the target from distractors was relatively difficult, and so it is possible each item had to be attended and submitted for further processing to be classified as a target or a distractor. Nonetheless, we wondered if a smaller SLpcN and/or RLpcN amplitude might be observed if the display contained a single item in each visual hemifield, for a variety of reasons outlined in the General discussion. As such, and to explore the set size factor further, we changed the possible set sizes in Experiment 2 to one, two, three, or four items in each hemifield. The reduction in set sizes will also reduce the probability of distractor interference on the target, and so we expected the latency of the RLpcN peak amplitude set size effect to decrease in magnitude. The experiment was otherwise similar to Experiment 1.

Methods

Participants

Thirty-one adults were compensated $20 for participating in the experiment, which had been vetted by the Ethics Committee of the Faculty of Arts and Science at Université de Montréal. Inclusion and exclusion criteria were the same as in Experiment 1. Twenty-four adults (age: M = 23.21 years, SD = 3.46 years; 12 women; six left-handed) were kept for the final analysis, and seven participants were excluded (details are presented in the Data collection and analysis section).

Procedure, Stimuli, and Task

The procedure was the same as Experiment 1 except for the following modifications. The display set sizes were one, two, three, or four (rather than two, four, or six) stimuli in each hemifield. When the task had two possible response alternatives (top or not), participants responded by pressing the “v” key for a gap on the top line segment of the square or “b” for all other gap locations. When the task had four possible response alternatives (“on what side is the gap located”), participants responded with “c” for left, “v” for top, “b” for down, or “n” for right. These were the only response key mappings in Experiment 2. Finally, to maximize the number of trials, the cue duration was shortened from 600 to 200 msec, and all jitters were reduced to 100 msec (instead of 200 msec). The duration between the cue and stimulus onset was 600 msec (with a jitter of 100 msec). For each response alternative block (2AD or 4AD), participants completed a practice block (16 trials) and five experimental blocks (96 trials each; total of 480 experimental trials per response alternative condition), giving a total of 960 experimental trials.

Data Collection and Analysis

EEG recordings and analysis pipelines were the same as Experiment 1. Statistical analyses were also the same as Experiment 1 with the exception that, for significant effects of set size, when the pattern of means suggested a linear relationship, polynomial contrasts were calculated instead of paired t tests. For amplitude measurements, the SLpcN component was measured from 450 to 850 msec, and the RLpcN from −650 to −150 msec. For component latency, the same jackknife method described in Experiment 1 was used. For SLpcN onset latency, the same measurement window (200–650 msec) and area percentage (30% between the curve and 0 μV) were used as in Experiment 1. For RLpcN onset latency, a measurement window from −1700 msec to the response was used, and the area calculated for each waveform was 40% between the curve and 0 μV. To measure differences in the latency of the RLpcN peak amplitude, as in Experiment 1, we used two time windows, one for the 2AD condition (50% of the area under the curve from −600 to 100 msec) and one for the 4AD condition (50% of the area under the curve from −700 msec to the response). Behavioral data (separately for accuracy and RT) and electrophysiological data (for both components, amplitude and latency measures) were subjected to 4 (Set size: one, two, three, or four items in each hemifield) × 2 (Number of response alternatives: 2AD or 4AD) × 2 (Block order: 2AD first or 4AD first) mixed ANOVAs.

Using the same exclusion criteria described in Experiment 1, seven participants were excluded from the final analysis for the following reasons: Five participants had excessive saccades (HEOG deflection > 5 μV), one participant had low accuracy, and one participant did not follow the instructions on how to respond.

Results

Behavioral Results

Accuracy.

Mean accuracy (proportion correct) and 95% within-subjects confidence intervals for each condition in Experiment 2 can be found in Table 4 and Figure 5. Accuracy was higher for the 2AD condition than for the 4AD condition, F(1, 22) = 11.82, p = .002, ηG2 = .04. Accuracy decreased as the number of displayed items increased, F(3, 66) = 45.51, p < .0001, ηG2 = .19. Polynomial contrasts corroborated what can be seen in the pattern of means (Table 4), namely, a monotonic relationship in which accuracy decreased as set size increased (linear: F(1, 22) = 64.09, p < .0001). The quadratic and cubic contrasts were not significant (quadratic: F(1, 22) = 1.48, p = .24; cubic: F < 1). The number of response alternatives and set size manipulations did not interact, F(3, 66) = 0.27, p = .76. All other effects and interactions were not significant (all Fs < 1).

Table 4. 
Results from Experiment 2
Response AlternativesSet SizeAccuracyRT (msec)
M95% CIM95% CIEst. of SD
2AD .95 ±.021 768 ±52 ±233 
.93 ±.012 965 ±38 ±322 
.89 ±.014 1147 ±49 ±427 
.86 ±.015 1283 ±72 ±506 
  
4AD .92 ±.017 923 ±40 ±301 
.89 ±.016 1171 ±42 ±420 
.86 ±.019 1362 ±48 ±530 
.82 ±.026 1538 ±65 ±642 
Response AlternativesSet SizeAccuracyRT (msec)
M95% CIM95% CIEst. of SD
2AD .95 ±.021 768 ±52 ±233 
.93 ±.012 965 ±38 ±322 
.89 ±.014 1147 ±49 ±427 
.86 ±.015 1283 ±72 ±506 
  
4AD .92 ±.017 923 ±40 ±301 
.89 ±.016 1171 ±42 ±420 
.86 ±.019 1362 ±48 ±530 
.82 ±.026 1538 ±65 ±642 

Accuracy (mean proportion correct) and mean RT (msec) with 95% within-subjects confidence intervals for each combination of the number of response alternatives (2AD or 4AD) and number of items (1, 2, 3, or 4 in each hemifield). Estimations of within-subjects standard deviation (Est. of SD) are also provided for RTs (msec).

Figure 5. 

Accuracy (mean proportion correct; left) and mean RT (sec; right) with 95% within-subjects confidence intervals for each combination of the number of response alternatives (2AD or 4AD) and number of items (one, two, three, or four items in each hemifield) for Experiment 2.

Figure 5. 

Accuracy (mean proportion correct; left) and mean RT (sec; right) with 95% within-subjects confidence intervals for each combination of the number of response alternatives (2AD or 4AD) and number of items (one, two, three, or four items in each hemifield) for Experiment 2.

RT.

Mean RTs with 95% within-subjects confidence intervals for Experiment 2 can be seen in Table 4 (which also shows estimates of within-subjects SD) and Figure 5. For the number of response alternatives, mean RT was shorter for the 2AD condition compared with the 4AD condition, F(1, 22) = 47.41, p < .0001, ηG2 = .21. RT increased with an increase in set size, F(3, 66) = 235.27, p < .0001, ηG2 = .52, with significant linear, F(1, 22) = 292.49, p < .0001, and quadratic, F(1, 22) = 10.71, p = .003, relationships. The cubic contrast was not significant (F < 1). An interaction was observed between the number of response alternatives and set size, F(3, 66) = 4.4, p = .02, ηG2 = .008. The interaction reflected a larger effect of set size when participants had four response alternatives compared with when they only had two, as can be seen in Figure 5. An interaction was also observed between the number of response alternatives and which response alternative block was performed first, F(1, 22) = 7.69, p = .01, ηG2 = .04. The difference in RT between the 2AD and 4AD blocks was larger when the 4AD block was performed first (2AD: 1002 msec; 4AD: 1293 msec; Δ = 291 msec) compared with when the 2AD block was presented first (2AD: 1080 msec; 4AD: 1204 msec; Δ = 124 msec). A marginally significant three-way interaction was observed, F(3, 66) = 2.78, p = .08. For all other main effects and interactions, F < 1.

Electrophysiological Results

Stimulus-locked (SLpcN component).

Waveforms showing the main effects for the SLpcN (by number of response alternatives and by set size) are shown in Figure 6 (left), and latency and amplitude measurements are given in Table 5.

Figure 6. 

Results from Experiment 2. Grand-averaged SLpcN (left) and RLpcN (right) contralateral minus ipsilateral activity at electrodes PO7/PO8. Measurement windows used for statistical analyses are indicated in gray (SLpcN: 450–850 msec; RLpcN: −650 to −150 msec). The top row shows waveforms for each display set size, either one (pink lines), two (red lines), three (dark red lines), or four (gray lines) items. The bottom row shows waveforms for either two (blue lines) or four (green lines) response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Figure 6. 

Results from Experiment 2. Grand-averaged SLpcN (left) and RLpcN (right) contralateral minus ipsilateral activity at electrodes PO7/PO8. Measurement windows used for statistical analyses are indicated in gray (SLpcN: 450–850 msec; RLpcN: −650 to −150 msec). The top row shows waveforms for each display set size, either one (pink lines), two (red lines), three (dark red lines), or four (gray lines) items. The bottom row shows waveforms for either two (blue lines) or four (green lines) response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Table 5. 
Results for the SLpcN in Experiment 2
Response AlternativesSet SizeSLpcN Onset Latency (msec)SLpcN Amplitude (μV)
M95% CIM95% CI
2AD 415 ±28 −1.44 ±0.51 
406 ±11 −2.37 ±0.55 
399 ±17 −2.54 ±0.41 
405 ±15 −2.47 ±0.72 
  
4AD 431 ±28 −1.45 ±0.48 
407 ±11 −2.43 ±0.49 
399 ±15 −2.48 ±0.48 
418 ±19 −2.34 ±0.62 
Response AlternativesSet SizeSLpcN Onset Latency (msec)SLpcN Amplitude (μV)
M95% CIM95% CI
2AD 415 ±28 −1.44 ±0.51 
406 ±11 −2.37 ±0.55 
399 ±17 −2.54 ±0.41 
405 ±15 −2.47 ±0.72 
  
4AD 431 ±28 −1.45 ±0.48 
407 ±11 −2.43 ±0.49 
399 ±15 −2.48 ±0.48 
418 ±19 −2.34 ±0.62 

Mean (M) measurements of onset latency (msec) and amplitude (μV), with 95% within-participant confidence intervals (95% CI), for each combination of the number of response alternatives (2AD vs 4AD) and set size (number of displayed items: 1, 2, 3, or 4 in each hemifield).

An effect of set size was observed for the SLpcN amplitude, F(3, 66) = 7.04, p = .0004, ηG2 = .06, but not latency, F(3, 66) = 2.77, p = .07.8 The amplitude effect appears to be driven by a smaller SLpcN amplitude for the one-item condition (see Figure 6, top left, and Table 5). To decompose this effect, a repeated-measures ANOVA comparing the one-item condition to the mean of the two-, three-, and four-item conditions, as well as a repeated-measures ANOVA comparing the two-, three-, and four-item conditions (without the one-item condition), were conducted. Results show that the amplitude of the SLpcN was smaller for the one-item condition compared with the mean of the two-, three-, and four-item conditions, F(1, 23) = 20.13, p = .0002, ηG2 = .12, and that there was no significant difference between the two-, three-, and four-item conditions (F < 1). All other effects and interactions for amplitude (from the mixed ANOVA) were not significant (all Fs < 1.31, all ps > .26). For latency, a significant interaction between which response alternative block was first and the number of response alternatives was observed, F(1, 22) = 4.36, p = .05, ηG2 = .02. Regardless of which block (2AD or 4AD) was presented first, no significant difference between the 2AD and 4AD conditions was observed (2AD: t(11) < 1; 4AD: t(11) = 2.24, p = .047; Bonferroni correction significance threshold: p = .025), but there appears to be an interaction because of a larger difference in latency between the response alternative conditions when the 4AD block was presented first (2AD: 399 msec; 4AD: 420 msec; Δ = 21 msec) compared with when the 2AD block was presented first (2AD: 414 msec; 4AD: 407 msec; Δ = 7 msec). All other effects and interactions for latency (from the mixed ANOVA) were not significant (all Fs < 1.17, all ps > .29), including the interaction between set size and the number of response alternatives, F(3, 66) = 0.35, p = .64.

Response-locked (RLpcN component).

Waveforms showing the main effects for the RLpcN (by number of response alternatives and set size) are shown in Figure 6 (right) and by set size separately for the 2AD and 4AD conditions in Figure 7. Means for RLpcN onset latency and the latency of the peak amplitude of the RLpcN are shown in Table 6.

Figure 7. 

Results from Experiment 2. Grand-averaged RLpcN (contralateral minus ipsilateral activity) waveforms time-locked relative to response at electrodes PO7/PO8. The top (filled lines) shows waveforms for different set sizes when there were two response alternatives (one item: pink lines; two items: red lines; three items: dark red lines; four items: gray lines), whereas the bottom (dashed lines) shows the same comparison but for four response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Figure 7. 

Results from Experiment 2. Grand-averaged RLpcN (contralateral minus ipsilateral activity) waveforms time-locked relative to response at electrodes PO7/PO8. The top (filled lines) shows waveforms for different set sizes when there were two response alternatives (one item: pink lines; two items: red lines; three items: dark red lines; four items: gray lines), whereas the bottom (dashed lines) shows the same comparison but for four response alternatives. Waveforms were low-pass filtered at 10 Hz for visualization purposes.

Table 6. 
Results for the RLpcN in Experiment 2
Response AlternativesSet SizeRLpcN Onset Latency (msec)RLpcN Latency of Peak Amplitude (msec)
M95% CIM95% CI
2AD −331 ±115 −189 ±55 
−442 ±86 −240 ±26 
−510 ±92 −242 ±29 
−577 ±73 −246 ±29 
  
4AD −508 ±140 −316 ±44 
−550 ±101 −347 ±26 
−690 ±90 −347 ±18 
−681 ±110 −330 ±37 
Response AlternativesSet SizeRLpcN Onset Latency (msec)RLpcN Latency of Peak Amplitude (msec)
M95% CIM95% CI
2AD −331 ±115 −189 ±55 
−442 ±86 −240 ±26 
−510 ±92 −242 ±29 
−577 ±73 −246 ±29 
  
4AD −508 ±140 −316 ±44 
−550 ±101 −347 ±26 
−690 ±90 −347 ±18 
−681 ±110 −330 ±37 

Mean (M) measurements of onset latency (msec) and latency of peak amplitude (msec), with 95% within-subjects confidence intervals (95% CI), for each combination of the number of response alternatives (2AD vs 4AD) and set size (number of displayed items: 1, 2, 3, or 4 in each hemifield).

For the number of response alternatives, the onset latency of the RLpcN was earlier (relative to the response) when there were four response alternatives compared with when there were two, F(1, 22) = 15.84, p = .0006, ηG2 = .08, with no difference in amplitude, F(1, 22) = 1.06, p = .32. There was a significant difference in onset latency and amplitude across set sizes (amplitude: F(3, 66) = 11.03, p < .0001, ηG2 = .10; onset latency: F(3, 66) = 7.59, p = .0002, ηG2 = .10). Polynomial contrasts showed a significant linear trend for RLpcN onset latency, with an earlier onset when more items were presented, F(1, 22) = 17.11, p < .001 (see Figure 6 and Table 6), and no significant quadratic or cubic relationships (all Fs < 1). To decompose the set size effect observed for RLpcN amplitude, a repeated-measures ANOVA comparing the one-item condition to the mean of the two-, three-, and four-item conditions as well as a repeated-measures ANOVA comparing the two-, three-, and four-item conditions were conducted. A significant difference between the one-item condition and the mean of all three other set sizes (two, three, and four items) confirmed that the RLpcN was significantly smaller when only one item was presented, F(1, 23) = 31.2, p < .0001, ηG2 = .2 (one item: M = −1.09 μV, 95% CI = ±0.33 μV; mean of two, three, and four items: M = −2.34 μV, 95% CI = ±0.33 μV), whereas no difference was observed between the two-, three-, and four-item conditions (F < 1). Returning to the mixed ANOVA, no other effects or interactions were significant for either amplitude (all Fs < 2.1, all ps > .16) or onset latency (all Fs < 1).

Similar to Experiment 1, the latency of the RLpcN peak amplitude was examined. An earlier latency of the peak RLpcN amplitude was observed for the 4AD condition compared with the 2AD condition, F(1, 22) = 50.72, p < .0001, ηG2 = .26, similar to the onset latency results and replicating the results in Experiment 1. However, the difference in latency of the peak amplitude by set size was only marginally significant, F(3, 66) = 3.04, p = .06, ηG2 = .04, which, according to the pattern of means (Table 6), was likely driven by Set Size 1 (Figure 6). No other effects or interactions were significant for the latency of the peak RLpcN amplitude (all Fs < 1.01, all ps > .36).

Discussion

The results from Experiment 2 suggest the same loci of factor effects for both manipulations (number of response alternatives and set size) as in Experiment 1. First, no SLpcN latency effects were observed for the number of response alternatives or set size conditions, suggesting that both manipulations affected processes occurring after the initial deployment of spatial attention. When examining results relative to the response (RLpcN), the 4AD condition began earlier than the 2AD condition. The ANOVA results from both latency measures (onset and latency of the peak amplitude) and corresponding waveforms (Figures 6 and 7) suggest that, for the number of response alternatives, both moments in the waveform were varying, on some level, together. As can be seen in Figure 6, the peak amplitude of the RLpcN was closer to the response for the 2AD condition than for the 4AD condition. As in Experiment 1, although the difference between the 2AD and 4AD conditions was slightly larger at onset (Δ = 142 msec) than at peak (Δ = 106 msec), most of the difference in RT variance between both waveforms was because of the difference between the latency of the peak amplitude and the response persisting into the onset of the RLpcN (when reading the waveforms from the response). These findings dovetail nicely with those from Experiment 1 and confirm that the latency of the peak RLpcN provides an index of the processing time taken after the engagement of attention on the target (presumably the last attended item, assuming self-terminating visual search).

We found a monotonic increase in RLpcN onset latency as set size was increased, in line with the behavioral RT results (see Tables 4 and 6 and Figures 5 and 6). Importantly, as in Experiment 1, the RLpcN waves obtained generally converged at about the same time (relative to the response) for all set size conditions, and the difference in latency of the peak amplitude between set sizes failed to reach significance in Experiment 2 (p = .06). This pattern of results is consistent with the hypothesis that the time of peak latency is related to the end of the search process and the beginning of response selection.

Interestingly, there was a smaller amplitude for the one-item condition compared with displays with two, three, or four items (see Figure 6, top). On the basis of our results and those of Experiment 1, we put forward two possible explanations for these results. First, the increase in contralateral activity for two or more items compared with a single item could be because of an increase in the need for focused attention when the target must be isolated from nearby distractor(s). Our results could also be described by serial and limited-capacity parallel processing visual search theories, in which we would suppose participants were processing about two items at a time (unless there was only one item). Both proposed accounts for our results are described in more detail in the General discussion.

GENERAL DISCUSSION

We showed it is possible to identify and dissociate processing stages affected by factor manipulations that cause RT variations during difficult visual search using electrophysiological measures. In two experiments, we found sustained stimulus-locked and response-locked lateralized ERPs related to visual attention (response-locked: Drisdelle & Jolicœur, 2019; stimulus-locked: Luck & Hillyard, 1990). We suggest these sustained components reflected attentional engagement to selected items (the onset of both components reflecting the deployment of attention to the first selected item) with subsequent processing up until the response likely reflecting overlapping ERPs associated with selective attention and visual working memory operations in both cases. This is unlike pop-out search, where there is generally a clear demarcation between these processes (N2pc vs. SPCN).

For pop-out search, Drisdelle and Jolicœur (2019) sorted EEG data into tertiles according to both the participant mean RTs and trial-by-trial RTs (within each participant and condition). In this earlier work, we showed the leading edge of the RLpcN was earlier, relative to the response, when participants were slower and/or trials had a longer RT. Importantly, very small or no latency effects were observed in the corresponding stimulus-locked components (N2pc and SPCN). The results therefore suggested that, during the pop-out search task, processes contributing most strongly to variations in the eventual RT (e.g., the processing time during working memory and response selection phases) occurred downstream of attentional engagement (N2pc) and the engagement of subsequent processing in VSTM (SPCN). The present work goes beyond our previous work by introducing factor manipulations designed to modulate RT during a difficult search task after the deployment of attention to a first item, allowing us to understand better the stream of processes recruited. We showed that, by examining lateralized posterior ERPs time-locked to the response, we can separate processing into substages using electrophysiological signatures that were otherwise indistinguishable from one another when activity was time-locked to the stimulus.

In the present experiments, the latency of the SLpcN, a component we suggest reflects a combination of attentional engagement and subsequent processing in VSTM (i.e., a combination of N2pc and SPCN observed during easy search tasks), was not modulated by either the number of displayed items or the number of response alternatives. Importantly, however, clear RT effects were observed, with an increase in RT as we increased the number of items or the number of response alternatives, suggesting that both factors affected processing after the initial deployment of attention. Clear latency effects were observed in the response-locked counterpart (RLpcN): The onset of the RLpcN was earlier when there were either more items or more response alternatives. We interpret this earlier latency, relative to the response, as an increase in processing time after the engagement of spatial attention in the search display.

Our results also suggest that it may be possible to separate processing taking place before target selection (i.e., visual search until the target is located) from processing taking place after (e.g., response selection) using the time course of the RLpcN. First, in both experiments, the target was difficult to discriminate from the distractors, making search difficult and increasing search time when there were more distractors. We showed an electrophysiological marker of this increase in search time: RLpcN was earlier, relative to the response, with increasing set size, in both experiments.

Second, in both experiments, RLpcN onset was also earlier, and RTs were longer, with an increase in the number response alternatives (4AD vs. 2AD). The curves for both experiments (Figures 3 and 4 for Experiment 1 and Figures 6 and 7 for Experiment 2) show a divergence in latency of the peak amplitude of the RLpcN for the number of response alternatives and a general convergence between set size levels. The convergence of the different set size waveforms to a single point suggests that this moment (i.e., the peak amplitude of the RLpcN) is possibly related to target selection. Thus, if we suppose that response selection can only begin after the target is selected, then the time after the peak amplitude of the RLpcN would reflect any additional time needed to perform either a two-alternative or four-alternative response. The pattern of means at onset and peak latency corroborate this idea. For the number of response alternatives, the difference in latency was similar at both time points for the 2AD and 4AD conditions (because a large part of the difference at peak latency persists as a constant into the onset of the component). For set size, however, the difference in latency between the smallest and largest set sizes (between two and six items in Experiment 1 and between one and four items in Experiment 2) was larger at onset than at latency of the peak amplitude. In short, we are suggesting that visual search and target selection occurred between RLpcN onset and peak amplitude, whereas response selection occurred later.

Given our results, we propose the following stream of processing. First, activity from the stimulus to the onset of the SLpcN (the N2pc in pop-out visual search) would reflect the processing time required for any initial sensory and perceptual processing up until when attention is deployed to the first item, which, in the case of difficult search, could be a target or a distractor. Second, a sustained SLpcN component in the stimulus-locked average reflects visual search for the target, likely including more shifts in attention as well as item encoding, selection, and any subsequent processing of the selected items (i.e., the SPCN in pop-out visual search), until the target is identified. Because of the large variance in search times (because the target will generally be found after inspecting different numbers of distractors, across trials), it is not possible to see the termination of visual search from the SLpcN (which would be reflected by the offset of the SLpcN if we assume self-terminating search and that the memory trace fades at this moment). From the response, we can observe activity after the deployment of attention to the first item (RLpcN onset) up until the response. The processing time between RLpcN onset and peak amplitude appears to reflect visual search and target selection (reflected by the difference in onset latency between set size conditions up until the convergence at RLpcN peak amplitude). Finally, response selection ensues later in the curve, causing a latency shift in both the onset and peak amplitude of the RLpcN depending on the number of response alternatives.

We will now consider the search process itself in a more fine-grained fashion. In Experiment 1, the amplitudes of the RLpcN and SLpcN were not modulated by the number of displayed items (either two, four, or six), whereas in Experiment 2, the SLpcN and RLpcN showed smaller amplitudes for one item compared with the two-, three-, or four-item conditions. We put forward two possible explanations for these results.

First, our results could be because of an increase in the need for focused attention when a target must be selected among distractors. Several researchers have observed a larger N2pc in response to pop-out targets where distractors are also present (e.g., Luck, Girelli, McDermott, & Ford, 1997; Luck & Hillyard, 1994a, 1994b). If this were the case, we would perhaps also expect increasing contralateral negative activity with an increasing number of distractors, which would increase further the need for focused attention. However, we observed a plateau at two items, suggesting that, according to this account, the increase in activity would be because of the presence of distractors and not the number of distractors.

A second possibility is that a maximum of two items could be processed at the same time. For example, mental operations during visual search could overlap with both items being processed at different stages (Wolfe & Gray, 2007; Pashler, 1994). We tentatively suggest the following scenario: While an item was being processed in VSTM, mechanisms of attention were already deployed to preview and select the next item for further processing, which would give a constant load of two items (see Salthouse, 1985, for an example of preview during typing, and the “car wash” model of guided search proposed by Wolfe & Horowitz, 2004). Thus, each item must be selected and submitted for further processing before moving onto the next, until the target was identified, and while one item was being identified as either a distractor or a target (a capacity-demanding stage of visual search), attention could be deployed to a second item so as to begin preview. In this view, our work converges nicely with previous work by Pashler (1994) who used a serial response paradigm with or without preview of subsequent items. He showed that a preview of one item increased the rate of processing compared with no preview and that additional preview items had little effect. This model may be appropriate for some aspects of our results given the equivalence of SLpcN amplitudes for set sizes of two and greater but a smaller amplitude for Set Size 1. Alternatively, processing of both items could occur in parallel, at the same stage of processing, with a limited capacity of two items (Hulleman & Olivers, 2017). The present results do not allow us to distinguish between these alternatives, and so more research is required.

Conclusion

Using both SLpcN and RLpcN components, we were able to fractionate RT into subprocesses and understand better the temporal organization of the underlying cognitive processes determining RT in difficult search. Our results suggest that, by examining posterior lateralized activity from both the stimulus and the response, we can identify the timing of processes associated with attending to the first processed item (stimulus to SLpcN onset), visual search up to target selection (RLpcN onset to latency of the RLpcN peak amplitude), and response selection (occurring after visual search and affecting both RLpcN onset latency and latency of the peak amplitude).

Acknowledgments

We thank Pia Amping for programming and Mathieu Charbonneau, Martine Desjardins, Pascale Forget, Maude Laflamme, Christine Lefebvre, Wanseo Kim, Anne Monnier, Amélie Renaud-Lavallée, Amour Simal, Antoine Slegers, and Leslie Wu for their assistance in data acquisition. This work was supported by the Natural Sciences and Engineering Research Council of Canada Discovery Grant (title: “Cognitive Neuroscience of Selective and Central Attentional Control”), the Canada Research Chairs program, and the Canada Foundation for Innovation.

Reprint requests should be sent to Pierre Jolicoeur, Département de Psychologie, Université de Montréal, Pavillon Marie-Victorin, CP6128, Succursale Centre-Ville, Montréal, Québec, Canada H3C 3J7, or via e-mail: pierre.jolicoeur@umontreal.ca.

Notes

1. 

The goal of the present work was not to address any questions regarding the serial or parallel nature of the difficult search induced by our task. For ease of exposition, we describe possible serial-processing scenarios, although we do not wish to claim that our work excludes other possibilities.

2. 

Also called the contralateral delay activity by Vogel and Machizawa (2004) and the contralateral negative slow wave by Klaver, Talsma, Wijers, Heinze, and Mulder (1999).

3. 

We verified that the different mappings within the 2AD or 4AD conditions did not produce RT effects. Two-sample t tests revealed no effect on RT for the different key mapping within the 2AD condition, t(30) = 0.11, p = .91 (mapping 2AD-1: M = 1285 msec, 95% CI = 161 msec; mapping 2AD-2: M = 1275 msec, 95% CI = 103 msec), or within the 4AD condition, t(30) = 0.03, p = .98 (mapping 4AD-1: M = 1800 msec, 95% CI = 137 msec; mapping 4AD-2: M = 1797 msec, 95% CI = 185 msec). Therefore, this factor was not considered in subsequent analyses.

4. 

Large onset latency effects, reflecting the probability that attention was deployed, contribute to the earlier amplitude differences in the RLpcN. We capture these earlier amplitude differences in latency measures and focus on amplitudes when each waveform reached a minimum, which provides a basis for comparison that is less likely to be influenced by latency effects.

5. 

We corrected the p values but report uncorrected degrees of freedom.

6. 

There was a marginal main effect of Set size, F(2, 60) = 3.06, p = .054, for SLpcN onset latency because of a slightly (but not significant) earlier onset for Set Size 6 compared with Set Size 2, t(31) = 2.39, p = .02 (Δ = 16 msec; Bonferroni correction significance threshold: p = .016). A marginal three-way interaction, F(2, 60) = 2.68, p = .08, was also observed. For all other onset SLpcN latency effects, F < 1.35 and p > .26.

7. 

In addition to certain general advantages (see Kiesel et al., 2008), this method produces more stable results than selecting the separate peak amplitudes of each curve for each participant and condition. We will refer to this analysis as a measure of the latency of the peak amplitude for the remainder of the article.

8. 

This marginally significant effect was because of the one-item condition (see Figure 6, top left).

REFERENCES

REFERENCES
Baguley
,
T.
(
2012
).
Calculating and graphing within-subject confidence intervals for ANOVA
.
Behavior Research Methods
,
44
,
158
175
.
Bakeman
,
R.
(
2005
).
Recommended effect size statistics for repeated measures designs
.
Behavior Research Methods
,
37
,
379
384
.
Brainard
,
D. H.
(
1997
).
The psychophysics toolbox
.
Spatial Vision
,
10
,
433
436
.
Cosman
,
J. D.
,
Arita
,
J. T.
,
Ianni
,
J. D.
, &
Woodman
,
G. F.
(
2016
).
Electrophysiological measurement of information flow during visual search
.
Psychophysiology
,
53
,
535
543
.
Cousineau
,
D.
(
2005
).
Confidence intervals in within-subject designs: A simpler solution to Loftus and Masson's method
.
Tutorials in Quantitative Methods for Psychology
,
1
,
42
45
.
Dell'Acqua
,
R.
,
Sessa
,
P.
,
Jolicœur
,
P.
, &
Robitaille
,
N.
(
2006
).
Spatial attention freezes during the attention blink
.
Psychophysiology
,
43
,
394
400
.
Delorme
,
A.
, &
Makeig
,
S.
(
2004
).
EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis
.
Journal of Neuroscience Methods
,
134
,
9
21
.
Drisdelle
,
B. L.
,
Aubin
,
S.
, &
Jolicoeur
,
P.
(
2017
).
Dealing with ocular artifacts on lateralized ERPs in studies of visual–spatial attention and memory: ICA correction versus epoch rejection
.
Psychophysiology
,
54
,
83
99
.
Drisdelle
,
B. L.
, &
Jolicœur
,
P.
(
2019
).
Stimulus- and response-locked posterior contralateral negativity bisect cognitive operations in visual search
.
Journal of Cognitive Neuroscience
,
31
,
574
591
.
Duncan
,
J.
, &
Humphreys
,
G. W.
(
1989
).
Visual search and stimulus similarity
.
Psychological Review
,
96
,
433
458
.
Eimer
,
M.
(
1996
).
The N2pc component as an indicator of attentional selectivity
.
Electroencephalography and Clinical Neurophysiology
,
99
,
225
234
.
Hackley
,
S. A.
,
Schankin
,
A.
,
Wohlschlaeger
,
A.
, &
Wascher
,
E.
(
2007
).
Localization of temporal preparation effects via trisected reaction time
.
Psychophysiology
,
44
,
334
338
.
Hulleman
,
J.
, &
Olivers
,
C. N.
(
2017
).
The impending demise of the item in visual search
.
Behavioral and Brain Sciences
,
40
,
e132
.
Humphreys
,
G. W.
, &
Müller
,
H. J.
(
1993
).
SEarch via Recursive Rejection (SERR): A connectionist model of visual search
.
Cognitive Psychology
,
25
,
43
110
.
Jolicœur
,
P.
,
Brisson
,
B.
, &
Robitaille
,
N.
(
2008
).
Dissociation of the N2pc and sustained posterior contralateral negativity in a choice response task
.
Brain Research
,
1215
,
160
172
.
Jolicœur
,
P.
,
Sessa
,
P.
,
Dell'Acqua
,
R.
, &
Robitaille
,
N.
(
2006
).
Attentional control and capture in the attentional blink paradigm: Evidence from human electrophysiology
.
European Journal of Cognitive Psychology
,
18
,
560
578
.
Kiesel
,
A.
,
Miller
,
J.
,
Jolicœur
,
P.
, &
Brisson
,
B.
(
2008
).
Measurement of ERP latency differences: A comparison of single-participant and jackknife-based scoring methods
.
Psychophysiology
,
45
,
250
274
.
Klaver
,
P.
,
Talsma
,
D.
,
Wijers
,
A. A.
,
Heinze
,
H. J.
, &
Mulder
,
G.
(
1999
).
An event-related brain potential correlate of visual short-term memory
.
NeuroReport
,
10
,
2001
2005
.
Kleiner
,
M.
,
Brainard
,
D.
,
Pelli
,
D.
,
Ingling
,
A.
,
Murray
,
R.
, &
Broussard
,
C.
(
2007
).
What's new in Psychtoolbox-3
.
Perception
,
36
,
1
16
.
Lopez-Calderon
,
J.
, &
Luck
,
S. J.
(
2014
).
ERPLAB: An open-source toolbox for the analysis of event-related potentials
.
Frontiers in Human Neuroscience
,
8
,
213
.
Luck
,
S. J.
,
Girelli
,
M.
,
McDermott
,
M. T.
, &
Ford
,
M. A.
(
1997
).
Bridging the gap between monkey neurophysiology and human perception: An ambiguity resolution theory of visual selective attention
.
Cognitive Psychology
,
33
,
64
87
.
Luck
,
S. J.
, &
Hillyard
,
S. A.
(
1990
).
Electrophysiological evidence for parallel and serial processing during visual search
.
Perception & Psychophysics
,
48
,
603
617
.
Luck
,
S. J.
, &
Hillyard
,
S. A.
(
1994a
).
Electrophysiological correlates of feature analysis during visual search
.
Psychophysiology
,
31
,
291
308
.
Luck
,
S. J.
, &
Hillyard
,
S. A.
(
1994b
).
Spatial filtering during visual search: Evidence from human electrophysiology
.
Journal of Experimental Psychology: Human Perception and Performance
,
20
,
1000
1014
.
Luck
,
S. J.
, &
Kappenman
,
E. S.
(
2012
).
The Oxford handbook of event-related potential components
.
New York
:
Oxford University Press
.
Maheux
,
M.
, &
Jolicœur
,
P.
(
2017
).
Differential engagement of attention and visual working memory in the representation and evaluation of the number of relevant targets and their spatial relations: Evidence from the N2pc and SPCN
.
Biological Psychology
,
125
,
28
35
.
Makeig
,
S.
,
Bell
,
A. J.
,
Jung
,
T.-P.
, &
Sejnowski
,
T. J.
(
1996
).
Independent component analysis of electroencephalographic data
.
Advances in Neural Information Processing Systems
,
8
,
145
151
.
Mathworks
. (
2009
).
MATLAB
(
Version R2009b-7.9.1.705
).
Natick, MA
:
Author
.
Mazza
,
V.
, &
Caramazza
,
A.
(
2011
).
Temporal brain dynamics of multiple object processing: The flexibility of individuation
.
PLoS One
,
6
,
e17453
.
Morey
,
R. D.
(
2008
).
Confidence intervals from normalized data: A correction to Cousineau (2005)
.
Tutorials in Quantitative Methods for Psychology
,
4
,
61
64
.
Olejnik
,
S.
, &
Algina
,
J.
(
2003
).
Generalized eta and omega squared statistics: Measures of effect size for some common research designs
.
Psychological Methods
,
8
,
434
447
.
Osman
,
A.
,
Moore
,
C. M.
, &
Ulrich
,
R.
(
1995
).
Bisecting RT with lateralized readiness potentials: Precue effects after LRP onset
.
Acta Psychologica
,
90
,
111
127
.
Pashler
,
H.
(
1994
).
Overlapping mental operations in serial performance with preview
.
Quarterly Journal of Experimental Psychology
,
47
,
161
191
.
Pelli
,
D. G.
(
1997
).
The videotoolbox software for visual psychophysics: Transforming numbers into movies
.
Spatial Vision
,
10
,
437
442
.
Salthouse
,
T. A.
(
1985
).
Anticipatory processing in transcription typing
.
Journal of Applied Psychology
,
70
,
264
271
.
Sharbrough
,
F.
(
1991
).
American electroencephalographic Society guidelines for standard electrode position nomenclature
.
Journal of Clinical Neurophysiology
,
8
,
200
202
.
Töllner
,
T.
,
Rangelov
,
D.
, &
Müller
,
H. J.
(
2012
).
How the speed of motor-response decisions, but not focal–attentional selection, differs as a function of task set and target prevalence
.
Proceedings of the National Academy of Sciences, U.S.A.
,
109
,
E1990
E1999
.
Treisman
,
A.
, &
Gelade
,
G.
(
1980
).
A feature-integration theory of attention
.
Cognitive Psychology
,
12
,
97
136
.
Treisman
,
A.
, &
Souther
,
J.
(
1985
).
Search asymmetry: A diagnostic for preattentive processing of separable features
.
Journal of Experimental Psychology: General
,
114
,
285
310
.
Van Selst
,
M.
, &
Jolicoeur
,
P.
(
1994
).
A solution to the effect of sample size on outlier elimination
.
Quarterly Journal of Experimental Psychology
,
47
,
631
650
.
Vogel
,
E. K.
, &
Machizawa
,
M. G.
(
2004
).
Neural activity predicts individual differences in visual working memory capacity
.
Nature
,
428
,
748
751
.
Wolfe
,
J. M.
(
1994
).
Guided search 2.0 a revised model of visual search
.
Psychonomic Bulletin & Review
,
1
,
202
238
.
Wolfe
,
J. M.
(
1998
).
Visual search
.
Attention
,
60
,
140
156
.
Wolfe
,
J. M.
,
Cave
,
K. R.
, &
Franzel
,
S. L.
(
1989
).
Guided search: An alternative to the feature integration model for visual search
.
Journal of Experimental Psychology: Human Perception and Performance
,
15
,
419
433
.
Wolfe
,
J. M.
, &
Gray
,
W.
(
2007
).
Guided search 4.0: Integrated models of cognitive systems
. In
W.
Gray
(Ed.),
Integrated models of cognitive systems
(pp.
99
119
).
New York
:
Oxford University Press
.
Wolfe
,
J. M.
, &
Horowitz
,
T. S.
(
2004
).
What attributes guide the deployment of visual attention and how do they do it?
Nature Reviews Neuroscience
,
5
,
495
501
.