1 INTRODUCTION

Our daily life consists of many circumstances where we must switch between cognitive tasks and adapt our behavior to new situations. To accomplish this, visual selective attention can be tuned to current task settings so that task-relevant stimuli are prioritized and task-irrelevant stimuli are ignored (Maunsell & Treue, 2006; Olivers et al., 2011). This attentional set affects neural processing from early in the visual hierarchy (Battistoni et al., 2017; Moore & Zirnsak, 2017) and prioritizes relevant stimuli while preventing or overcoming selection of distractors (see e.g., Ansorge et al., 2011; Kiss et al., 2013; Sawaki & Luck, 2013; Wykowska & Schubö, 2011). It relies on mnemonic representations of target and distractor templates and facilitates both target processing and the suppression of stimuli known to be task-irrelevant (see e.g., Arita et al., 2012; Feldmann-Wüstefeld & Schubö, 2016; Reeder et al., 2017; Vatterott & Vecera, 2012).

Proactive attentional set can be indexed in its impact on task completion, but it can also be directly assessed in brain activity prior to stimulus onset. For example, the power and phase of pre-stimulus alpha-band EEG oscillations predict subsequent detection rate and discrimination performance (Busch et al., 2009; Chaumon & Busch, 2014; van Dijk et al., 2008; Hanslmayr et al., 2007). The preparatory establishment of attentional control seems to be reflected in a systematic decrease of posterior alpha power (Capotosto et al., 2016; Clayton et al., 2018; Mathewson et al., 2014), in line with the idea that alpha desynchronization reflects a release from inhibition (Klimesch, 2012) and thus induces a state of perceptual readiness (Hanslmayr et al., 2011; Mathewson et al., 2012; Sawaki et al., 2015). Although some studies have challenged the direct relation between post-stimulus alpha-band power and attentional selection (Antonov et al., 2020; Gundlach et al., 2020; Zhigalov & Jensen, 2020), a recent study has provided convincing results supporting the notion that pre-stimulus alpha-band power directly impacts post-stimulus attentional selection (van Zoest et al., 2021). According to their findings, van Zoest et al. (2021) suggested that increased power of pre-stimulus alpha-band reflects an advanced suppression mechanism which results in less attentional capture and less need for distractor suppression after stimulus onset.

Alpha oscillations are also associated with cognitive flexibility in task switching paradigms. In paradigms requiring shifts in attentional strategy, cues identifying a task shift elicit a reduction of frontal alpha power, and this has been linked to the need to adjust attentional control settings in switch trials (e.g., Foxe et al., 2014; Gladwin & de Jong, 2005; Poljac & Yeung, 2014; Proskovec et al., 2019). This “task-set reconfiguration” requires effort (Mayr & Keele, 2000; Monsell, 2003; Monsell & Mizon, 2006), but has a direct benefit: when participants are given sufficient time between cue and task onset, the cue reduces task-switching costs (Monsell & Mizon, 2006).

Human observers are therefore equipped with a dynamic top-down control system that directs attention according to current task goals. However, other influences on attention can interfere with this optimization. In particular, attentional selection is strongly shaped by prior experience (Awh et al., 2012; Failing & Theeuwes, 2018; Ferrante et al., 2018). Aside from effects of prior reward (e.g., Anderson et al., 2011; Feldmann-Wüstefeld et al., 2016; Hickey et al., 2010), attention is biased toward stimuli that have been predictive, even when this predictive power has explicitly ended (Feldmann-Wüstefeld et al., 2015; Kadel et al., 2017; Le Pelley et al., 2011; O’Brien & Raymond, 2012). In Feldmann-Wüstefeld et al. (2015), for example, selection history was manipulated by having two groups of participants complete different categorization tasks with the same stimuli. One group categorized the shape of stimuli, the other the color, and both subsequently completed the same visual search task. In the search task, participants showed a pronounced attentional bias toward a task-irrelevant distractor defined in the feature dimension that had been predictive in the categorization task. Subsequent research has shown that this sustains even when participants are explicitly told that the tasks are unrelated and when the tasks are completed on different days. The lingering bias disappears only after several hundred visual search trials have been completed (Kadel et al., 2017).

Selective attention is thus sensitive to proactive top-down control on one hand and selection history on the other. How are these mechanisms related? How do we reconcile situations in which selection history is in conflict with top-down control settings? Can top-down control compensate for effects of selection history?

The mere possibility for top-down control seems to do little in negating the effect of selection history. We have recently found that the opportunity for trial-wise top-down preparation (enabled by pretrial cueing) will not override selection history effects (Kadel et al., 2017, Exp. 1 and 2). However, little is known about the mechanisms underlying proactive top-down preparation in situations with a selection history bias. The purpose of the current study is therefore to directly index the preparation of attentional control settings and determine if this preparation, when present, can compensate for individual selection history. We tracked proactive control in pre-stimulus alpha power (Schneider et al., 2021; van Zoest et al., 2021) and we looked at the post-stimulus ERP in switch trials to identify the effect of this preparation and the effect of selection history on stimulus processing. In the ERP, the early distractor positivity (early Pd; Hickey et al., 2009; Sawaki & Luck, 2010; Weaver et al., 2017; van Zoest et al., 2021) was employed to track rapid, stimulus-triggered suppression of irrelevant stimuli. The subsequent N2pc (Eimer, 1996; Luck & Hillyard, 1994a, 1994b) was used to index changes in the attentional resolution of attended stimuli. Both components emerge as voltage differences across visual cortex ipsilateral and contralateral to eliciting stimuli.

We manipulated selection history by having participants complete two intermingled types of trial. In one trial type, they categorized stimuli that varied in color and shape (Figure 1a). Half of the participants (color-categorization group) were required to categorize the uniquely-colored stimulus (blue vs. green), while the other half (shape-categorization group) were required to categorize the uniquely-shaped stimulus (triangles vs. pentagons). In the other type of trial, all participants completed a visual search task that required them to attentionally select a uniquely shaped target and ignore a uniquely colored distractor (Figure 1b). For the color-categorization group, the predictive dimension in the categorization task was task-irrelevant and potentially distracting in the search task. For the shape-categorization group, in contrast, the predictive dimension in the categorization task was also relevant in the visual search task.

(a) Exemplary displays in the categorization task. Participants in the color-categorization group had to press one button for a green and another button for a blue circle. Participants in the shape-categorization group had to press one button for a pentagon and another button for a triangle. Participants were naïve to their group assignment when the experiment started and had to learn on a trial-and-error basis by receiving immediate auditory feedback in incorrect trials. (b) Exemplary displays in the search task. Both groups searched for the diamond-shaped target and reported the orientation (horizontal vs. vertical) of the embedded line. In 60% of the trials, an additional color distractor was presented (right panel) which had to be ignored

To provide the opportunity for proactive attentional control, the categorization and search tasks were performed within the same experiment, but the sequence of trial types changed between blocks (Figure 2a). In random-sequence blocks, the tasks were intermingled unpredictably so that no task-specific proactive preparation was possible. In fixed-sequence blocks, the tasks alternated in a fixed, predictable pattern. Our expectation was that participants would proactively reconfigure their attentional control settings in fixed-sequence blocks, where such preparation was possible, and that this would be reflected in preparatory oscillations in the alpha frequency band and in effects on behavior, Pd, and N2pc.

(a) Schematic depiction of the two task sequences used in the experiment. Each rectangle represents one experimental block and each of the small letters below the rectangles represents one trial. Letters “C” represent categorization trials, and letters “S” represent search trials. Participants completed 20 blocks in each task sequence. In fixed-sequence blocks, categorization trials and the search task trials alternated in fixed sequences of two trials per task. In random-sequence blocks, trials of both tasks alternated in random order. Panels B and C are included for clarity with the focus on the switch trials and they are redundant to the complete presentation of the data depicted in panels D and E. (b) Mean response times in switch trials in the categorization task for the color-categorization (blue) and the shape-categorization group (black), separated for fixed and random task sequences. Error bars represent standard errors of the mean. (c) Mean response times in switch trials in the search task for the color-categorization (blue) and the shape-categorization group (black), separated for fixed and random task sequences. Solid lines show RTs for distractor-absent trials, dashed lines represent RTs for trials with an additional color distractor. Error bars represent standard errors of the mean. (d) Mean response times in the categorization task for the color-categorization (blue) and the shape-categorization group (gray), separated for fixed-sequence (left bars in each panel) and random-sequence blocks (right bars in each panel). Bars with darker colors show RTs for switch trials, bars with lighter colors show RTs for repetition trials. Error bars represent standard errors of the mean. (e) Mean response times in the search task in the color-categorization group (blue) and the shape-categorization group (black) in fixed-sequence (left panel) and random-sequence blocks (right panel). Filled bars show RTs for trials with an additional color distractor, unfilled bars show RTs for distractor-absent trials. Error bars represent standard errors of the mean

Most importantly, we were interested in whether proactive attentional control in fixed-sequence blocks would differ between the color- and shape-categorization groups. Participants in the shape-categorization group could rely on a similar attentional set for both the categorization and search tasks, because in both cases the target stimulus was defined in the same featural dimension. In contrast, participants in the color-categorization group attended to color targets in categorization trials, but shape targets in search trials, and therefore had to substantially reconfigure attentional control when the task switched. Our expectation was that correlates of reconfiguration would therefore emerge prominently when participants in the color-categorization group completed switch trials in fixed-sequence blocks. If participants are able to properly reconfigure in this circumstance, we expected this to benefit their attentional control in fixed-sequence blocks. As a result, in the search task, the color-categorization group should demonstrate less attentional capture in fixed-sequence blocks relative to random-sequence blocks.

2 METHOD 2.1 Participants

Forty volunteers (9 male) participated in the experiment for course credit or monetary payment (8€/h). Written consent for participation was obtained before the experimental session. All but two participants were right-handed and all had normal or corrected-to-normal vision. Eight participants had to be excluded from analysis due to excessive eye movement artefacts in EEG data (over 25% of the trials; see below for details). Of the remaining 32 participants (6 male), 16 were assigned to the color-categorization group (mean age ± SD: 23.5 ± 2.5 years) and 16 to the shape-categorization group (mean age ± SD: 23.3 ± 2.5 years).

2.2 Stimulus and apparatus

Participants were seated in a comfortable chair in a dimly lit, electrically shielded and sound attenuated room and responded via a customizable keypad (Ergodex DX1) held on their lap. Two response buttons on the left half of the pad were used in the categorization task and two separate buttons on the right half of the pad were used in the search task. Participants used the thumb and ring finger of their left hand to respond during the categorization task and the index and middle finger of their right hand to respond during the search task. Task presentation was controlled via E-Prime 2.0 (Psychology Software Tools, Inc.) on a standard PC under Windows XP. Stimuli were presented on a 22″ LCD-TN screen (Samsung Syncmaster 2233) at a viewing distance of 100 cm. For auditory feedback, two stereo speakers were positioned behind the screen, each on one side (Logitech Z120 2.0).

In both tasks, the display consisted of eight objects of 2.3° visual angle placed equidistant from the screen center in a circular search array on a dark gray background (CIELAB coordinates with reference white point of D65: L* = 25.32, a* = 0, and b* = 0; distance screen center to stimulus center: 6.3°; horizontal eccentricity: 5.7°). The display was not color-calibrated, but we measured and matched the luminance of the stimuli (~27–30 candela). In the categorization task, the display contained six neutral distractor stimuli (gray circles, L* = 46.44, a* = 0, and b* = 0) and two unique objects (see Figure 1a). One unique object had a distinct color, either green (L* = 61.62, a* = −56.72, and b* = 51.06) or blue (L* = 28.59, a* = 40.83, and b* = −65.28). The other unique object had a distinct shape, either a triangle or a pentagon. This type of stimulus, which differs from its surroundings in a single featural dimension, is known as a singleton. The color and shape singletons were presented at any of eight equidistant locations with exactly one neutral distractor between them. All singleton combinations (blue/triangle, blue/pentagon, green/triangle, green/pentagon) were presented equally often in all possible locations.

In the search task, the target was a diamond-shaped singleton with a horizontal or vertical line inside (see Figure 1b). Neutral distractor stimuli were gray circles (L* = 46.44, a* = 0, and b* = 0) that contained a gray oblique line tilted 45° to the left or right. In 40% of the search trials, the target was presented with seven neutral distractors (distractor-absent trials, Figure 1b). In the remaining 60% of trials, a color singleton in red (L* = 39.56, a* = 49.69, and b* = 29.59, circular shape) with an embedded oblique line appeared with one neutral distractor separating it from the target (distractor-present trials, Figure 1b). Target and distractor appeared with equal likelihood in each of the eight stimulus positions and equally often in the same side of the visual field (distractor-present, same side) as on opposite sides (distractor-present, opposite sides). Each trial began with a centrally presented gray fixation cross (0.6° visual angle) 500 ms before the stimulus display, which remained on the screen throughout the trial.

It is common in the literature to isolate target and distractor processing in visual search by selectively presenting targets and distractors on the vertical meridian of the display (e.g., Hickey et al., 2009). When the target is presented on the vertical midline, this supports discrete identification of lateralized distractor processing in the ERP within a condition. However, there is an associated cost: if target and distractor positions are randomized in a dense search array, target-vertical trials occur rarely and ERPs are based on relatively few trials. With this in mind, we have not adopted this design in the current study, instead identifying variance in target and distractor processing through comparison of results across physically identical conditions. Within-condition ERPs are therefore based on many trials without the need for a long, exhausting experiment.

2.3 Procedure 2.3.1 General trial procedure

Categorization and search trials started with a fixation cross for 500 ms that was followed by the stimulus display for 200 ms. A blank screen with a central fixation cross was subsequently displayed for up to 1800 ms, indicating that participants should respond while maintaining fixation. A correct response within that time interval triggered the beginning of the inter-trial-interval (1000 ms). An erroneous or missing response led to acoustic feedback in form of a low buzzing tone.

2.3.2 Feedback-guided learning phase

Participants started the experiment by completing a block of 64 categorization learning trials. Participants were informed that in each trial one stimulus would be different in color and another would be different in shape. They were asked to respond by pressing either the upper or lower response button with their left hand and told that errors were followed by a buzzing tone. They were not told which stimulus was assigned to which button press, or that only one dimension was response predictive. Instead, they had to use the acoustic feedback to find out which dimension was response relevant and how the two possible stimuli within that dimension were mapped to the response keys (see Kadel et al., 2017, for details). Participants in the color-categorization group learned to respond to color singletons and ignore shape singletons, pressing one key for a blue singleton and another for a green singleton. Participants in the shape-categorization group learned to respond to shape singletons and ignore color singletons, pressing one key for a triangle and another for a pentagon. The assignment of response buttons was varied across participants and response accuracy and speed were emphasized equally. In the first 32 trials of the learning phase, stimulus presentation was prolonged to 500 ms to facilitate learning. Participants proceeded to the next block when accuracy was >75%, otherwise they had to repeat the block. On average, participants performed 2.23 blocks of the learning categorization task (SD = 1.39) before shifting to the mixed practice phase.

2.3.3 Mixed practice phase

This block was performed after the initial learning and combined 32 categorization learning task trials (as described above) with 32 search task trials in a random order. In search task trials, all participants responded to the orientation of the line embedded in the diamond shape target by pressing either the left or right response-board button with their right hand. Stimulus-response mapping was counterbalanced over participants within each of the color- and shape-categorization groups. In this phase of the experiment, stimuli were presented for 1000 ms in search trials in order to facilitate learning. As in the learning phase of the experiment, participants proceeded to the next block of trials when accuracy was >75%.

2.3.4 Main experiment

The main experimental session was performed the next day and EEG was recorded throughout. Participants were informed that the task was to be performed in two types of experimental blocks. In random-sequence blocks, categorization and search trials were intermingled in a random, unpredictable order (with the limitation that no more than four trials of one task could follow each other). In fixed-sequence blocks, trials of the categorization and search task alternated in a fixed and predictable sequence of exactly two trials per task. The task trial sequence (fixed or random) was identified on the screen before the block started.

In total, participants completed 40 blocks of 64 trials each, 20 in each task sequence, 2560 trials in total, and 1280 trials in each of the categorization and search tasks. In the search task, 512 of the 1280 trials were distractor-absent trials, where in the remaining trials the distractor was presented either on the same side as the target (384 trials) or on the opposite side (384 trials). Immediate auditory feedback was given after incorrect responses. After errors, participants took a forced break of at least 8 s, and they were given performance feedback (RT and accuracy) after each block. Participants were prompted to take longer breaks of several minutes on two occasions.

2.4 EEG recording

EEG was recorded from 64 Ag-AgCl active electrodes (actiCAP by Brain Products GmbH, Munich, Germany). Electrodes were placed according to the international 10–10 system. Vertical EOG (vEOG) was recorded from Fp1 and an electrode placed below the left eye, and horizontal EOG (hEOG) was recorded from electrode positions F9 and F10. Impedances were kept below 5 kΩ. All electrodes were referenced to FCz during recording and re-referenced offline to the average of all electrodes. The signal was recorded with a BrainAmp amplifier (Brain Products, Munich, Germany) at a sampling rate of 1000 Hz and high pass filtered at 0.016 Hz and a low pass filtered at 250 Hz (−3 dB cutoff, Butterworth filter, 30 dB/oct roll-off).

2.5 Data analysis 2.5.1 Behavioral data

The first trial of each block was rejected from analysis, as were trials with incorrect responses and trials with outlier RT (>2 SD from mean RT calculated separately for each participant and separately for each block and each task). This led to exclusion of 9.30% of trials in the shape-categorization group and 9.22% in the color-categorization group.

2.5.2 EEG data

Brain Vision Analyzer (Brain Products, Munich, Germany), the Fieldtrip toolbox (Oostenveld et al., 2011) and custom scripts for Matlab R2019a (Mathworks, http://www.mathworks.com) were used for off-line EEG data processing.

Event-related potentials

EEG was segmented into 700-ms epochs time-locked to the display onset, including a 200-ms pre-stimulus baseline. Vertical EOG (vEOG) was calculated as the difference between Fp1 and the electrode placed below the left eye, and horizontal EOG (hEOG) was calculated as the difference between electrodes F9 and F10. The four channels were filtered using a low-pass filter of 35 Hz. Trials with eyeblink (vEOG > ±80 μV), or horizontal eye movements (hEOG > ±35 µV step criterion) within the first 350 ms after stimulus onset were excluded from analysis. Channels with activity > ±80 µV in the first 350 ms after stimulus onset in a trial were also excluded. The first trial of each block was rejected from analysis, as were trials with incorrect responses. Participants with less than 75% artifact-free trials were excluded from further analysis (8 participants). The remaining 32 participants had 88.1% artifact-free trials on average. In total, 12.0% of trials had to be excluded in the shape-categorization, and 11.8% in the color-categorization group.

To quantify the early Pd and N2pc in both categorization and search tasks, mean contralateral and ipsilateral activity in the ERP was calculated for electrodes PO7/PO8 and cross-conditional effect peaks were identified. A 40 ms window was centered on this peak latency and amplitude measures reflect the mean across this interval.

For the categorization task, the N2pc peak emerged at 224 ms. For the search task, Pd and N2pc measurements were separated across the conditions identified in Figure 4. In distractor-absent trials, the N2pc peaked at 246 ms (Figure 4a). When the target and distractor were in the same visual hemifield the early Pd peaked at 127 ms and the N2pc peaked at 236 ms (Figure 4b). When the target and distractor were in opposite visual hemifields, the positive-polarity early Pd expresses as a negative-polarity peak, because the ERP is locked to the location of the target stimulus (such that a positivity contralateral to the distractor emerges as a negativity contralateral to the target). In this condition, the early Pd peaked at 142 ms and the N2pc peaked at 294 ms (Figure 4c).

In addition to mean amplitude, onset latency of the N2pc component was analyzed in search task trials (distractor absent, distractor-present same side, distractor-present opposite sides) using a jackknife-based approach (Kiesel et al., 2008; Miller et al., 1998; Smulders, 2010). Conditional onset was defined as the point where 50% of maximum N2pc amplitude was reached (Kiesel et al., 2008). Relevant statistics are corrected for the jackknife procedure and this is identified with the subscript “c”.

Time-frequency analysis

Oscillatory activity was analyzed over a 3000 ms epoch beginning 2000 ms before display onset. Trials excluded from ERP analyses were also excluded from time-frequency analysis. Before performing time-frequency analysis, data were downsampled to 500 Hz. The spectral analysis of the zero-padded time series was performed using a Fast-Fourier transformation. Zero-padding was done using the “nextpow2” function which returns the smallest power-of-two larger than the length of the time series. The power spectrum of the EEG was computed within a 500 ms Hanning window which moved in steps of 30 ms, so that every bin represented data from 250 ms before and after the nominal latency. This analysis was performed for frequencies 2 to 32 Hz with a resolution of 1 Hz. Trials were sorted according to the factors task sequence (fixed vs. random sequence), and task repetition (task switch vs. task repetition), resulting in four conditions per participant. Power values of each frequency at each time point and electrode were averaged separately for each condition.

Because task-set reconfiguration was not required in repetition trials, our expectation was that differences in task-set reconfiguration should emerge as a difference in pre-stimulus alpha power between switch and repetition trials, but only for fixed-sequence blocks. In random-sequence blocks, there was no opportunity for participants to know when task repetitions would occur, and accordingly no opportunity for preparation.

To compute the difference in preparatory alpha between switch and repetition trials in fixed and random-sequence blocks, we contrasted power values in each time and frequency bin using [(switch−repetition)/(switch + repetition)] × 100 individually for each posterior channel (O1/2, PO7/8, PO3/4, P7/8, P5/6, P3/4, P1/2, Oz, POz and Pz). Next, power values were averaged over channels, separately for participants in each group. Finally, the power values of switch and repetition trials were forwarded to a statistical analysis based on a cluster-based permutation test with 5000 permutations. A cluster-defining threshold of α < .01 was employed (corresponding to a critical t-value of 2.95; Maris & Oostenveld, 2007).

We used cluster-based permutation tests for the data in the frequency range of 6 to 32 Hz and in the time range of 800-ms pre-stimulus interval. Focusing the analysis on this range of data involved 27 frequency bins and 28 time bins in the analysis. Analysis was based on the average power spectrum of 17 posterior channels and compared the power spectrum of switch and repetition trials separately for fixed and random trial sequences. According to this, in each type of trial sequence and in each group, the labels of switch and repetition trials were permuted. In this analysis, a significant cluster indicated that the corresponding frequency power differed significantly between switch and repetition trials. To subsequently test whether this switch-vs-repetition effect differed in color- vs shape-categorization groups, the normalized power differences between switch and repetition were compared between the groups. In this analysis, the labels of color-categorization and shape-categorization were permuted. This analysis employed a cluster-based permutation test with 5000 permutations. A cluster-defining threshold of α < .05 was employed (corresponding to a critical t-value of 2.04).

3 RESULTS

A core motivating hypothesis for the study was that preparation would differ between shape- and color-categorization groups in task switch trials. Accordingly, we constrained the analysis of the ERP to switch trials. To provide a comprehensive description of participants’ performance, RT analysis was conducted for all data.

3.1 Categorization task 3.1.1 Behavioral results

(See Figure 2b,d) RT and accuracy were analyzed in a 3-way ANOVA with a between-subject factor for categorization group (color-categorization group vs. shape-categorization group) and within-subject factors for task sequence (fixed vs. random sequence) and task repetition (task switch vs. task repetition). All reports in the results section identify mean plus/minus standard error of the mean (M ± SEM).

A main effect of task sequence emerged, with faster responses in fixed-sequence blocks (M = 541 ± 10 ms, M = 590 ± 9 ms), F(1,30) = 176.71, p < .001, = 0.85. An additional main effect of task repetition was detected, with faster responses when the task repeated (M = 530 ± 7 ms, M = 601 ± 12 ms), F(1,30) = 153.02, p < .001, = 0.84. Task repetition interacted with task sequence: the benefit of task repetition was greater in fixed-sequence blocks (fixed-sequence: ΔM(switch-rep) = 76 ± 6 ms, random-sequence: ΔM(switch-rep) = 67 ± 7 ms), F(1,30) = 7.32, p = .01, = 0.20.

In line with the idea that participants in the color-categorization group would most benefit from the opportunity to prepare, task sequence interacted with categorization group: participants in the color-categorization group benefited from fixed task sequence more than did participants in the shape-categorization group (color-categorization group: ΔM(rand-fix) = 70 ± 6 ms, shape-categorization group: ΔM(rand-fix) = 29 ± 4 ms), F(1,30) = 30.84, p < .001, = 0.51). Similarly, task repetition interacted with categorization group: participants in the color-categorization group benefited from task repetition more than did participants in the shape-categorization group (color-categorization group: ΔM(switch-rep) = 84 ± 10 ms, shape-categorization group: ΔM(switch-rep) = 59 ± 6 ms), F(1,30) = 4.55, p = .04, = 0.13. Although participants in the color-categorization group responded nominally faster than participants in the shape-categorization group (M = 548 ± 13 ms, M = 584 ± 13 ms), this effect failed to reach significance, F(1,30) = 3.58, p = .07, = 0.11. No other effects emerged (all ps > .1).

The only reliable effect on accuracy was a main effect of task repetition (Mrepetition = 98.4 ± 0.2% vs. Mswitch = 96.3 ± 0.5%), F(1,30) = 40.18, p < .001, = 0.57.

3.1.2 ERP results

(See Figure 3) Focusing on the switch trials, the N2pc was analyzed in a 2-way ANOVA with between-subject factor for categorization group and within-subject factor for task sequence. Figure 3 further separates the data as a function of the location of the non-predictive singleton, but statistical analysis was collapsed across this factor. As evident in Figure 3, color singletons create an early positive-polarity effect in the lateral ERP. This early lateral positivity emerges contralateral to the location of the color singleton for both color-categorization and shape-categorization groups without differing between these groups. This appears to reflect the “positivity posterior contralateral” component (PPC), which is thought to reflect stimulus salience and does not vary as a function of whether an eliciting stimulus is a target or distractor (Corriveau et al., 2012). Consistent with the idea that preparation would improve attentional resolution of the target, analysis of the N2pc identified a main effect of task sequence, with the N2pc larger in fixed-sequence blocks (M = −0.96 ± 0.21 µV, M = −0.73 ± 0.21 µV), F(1,30) = 5.73, p = .02, = 0.16. No other effects emerged (all ps > .1).

Grand-average of difference waves recorded at parieto-occipital electrodes PO7 and PO8, elicited by predictive singletons in switch trials in the categorization task, locked to the location of the color singleton in the color-categorization group (blue lines) or locked to the location of the shape singleton in the shape-categorization group (black lines). The upper panels represent the waveforms when singletons appear on the same side and the lower panels represent the waveforms when singletons appear on the opposite sides for fixed (left panels) and random-sequence blocks (right panels). For illustration purposes, EEG waveforms were filtered using a low-pass Butterworth filter with high cutoff frequency of 35 Hz (12 dB/oct)

3.2 Search task 3.2.1 Behavioral results

(See Figure 2c,e) RT and accuracy were analyzed in a 4-way ANOVA with a between-subject factor for categorization group (color-categorization group vs. shape-categorization group) and within-subject factors for distractor-presence (distractor present vs. absent), task sequence (fixed vs. random sequence), and task repetition (task switch vs. task repetition).

In analysis of RT, a main effect of distractor-presence emerged, with slower responses when the distractor was present (M = 640 ± 12 ms, M = 667 ± 13 ms), F(1,30) = 84.09, p < .001, = 0.74. An additional main effect of task sequence emerged, with faster responses in fixed-sequence blocks (M = 639 ± 13 ms, M = 667 ± 12 ms), F(1,30) = 44.31, p < .001, = 0.60. The main effect of task repetition was also significant, with faster responses in task repetition trials (M = 625 ± 11 ms, M = 681 ± 15 ms), F(1,30) = 68.37, p < .001, = 0.70.

Categorization group interacted with distractor-presence: the distractor cost was larger in the color-categorization group (ΔM = 37 ± 5 ms, ΔM = 17 ± 3 ms), F(1,30) = 10.82, p = .003, = 0.27. The history of selecting color during the categorization task appears to have increased sensitivity to color during the search task.

Categorization group also interacted with task sequence: the propensity toward faster responses in fixed-sequence blocks was accentuated in the color-categorization group (color-categorization group: ΔM(rand-fix) = 39 ± 7 ms, shape-categorization group: ΔM(rand-fix) = 18 ± 5 ms), F(1,30) = 6.29, p = .02, = 0.17. This suggests increased preparation in this group, who had to switch task sets between trial types. Note that even though the color-categorization group responded numerically faster than the shape-categorization group in the fixed-sequence blocks (Mcolor-categorization = 625 ± 19 ms, Mshape-categorization = 652 ± 19 ms, t(30) = 1.04, p = .30), the distractor cost remained marginally larger (ΔMcolor-categorization = 33 ± 5 ms, ΔMshape-categorization = 20 ± 4 ms, t(30) = 2.05, p = .05).

Task sequence interacted with task repetition: the benefit of task repetition was greater in fixed-sequence blocks than in random-sequence blocks (ΔM(switch-rep) = 59 ± 7 ms, ΔM(switch-rep) = 51 ± 6 ms), F(1,30) = 6.57, p = .02,  = 0.18. Task repetition also interacted with distractor cost: the distractor cost was smaller when the task was repeated (ΔM = 24 ± 4 ms, ΔM = 30 ± 3 ms), F(1,30) = 8.40, p = .007, = 0.22.

A marginal three-way interaction between distractor-presence, task sequence, and categorization group emerged: participants showed a numerically smaller distractor cost in fixed-sequence blocks, but only in the color-categorization group (color-categorization group: ΔMfixed = 33 ± 5 ms, ΔMrandom = 40 ± 6 ms; shape-categorization group: ΔMfixed = 20 ± 4 ms, ΔMrandom = 14 ± 3 ms), F(1,30) = 3.89, p = .058, = 0.11.

No other effects on RT emerged (ps > .1) and the only reliable effect on accuracy was an improvement in repetition trials (M = 96.6 ± 0.4%, M = 95.7 ± 0.6%), F(1,30) = 5.21, p = .03, = 0.15.

3.2.2 ERP results. Distractor-absent trials (Figure ) Target-elicited N2pc

Focusing on the switch trials, the N2pc was analyzed with a 2-way ANOVA with a between-subjects factor for categorization group and within-subject factor for task sequence. This identified a trend toward an interaction of categorization group and task sequence, with the effect of task sequence more pronounced in the color-categorization group (Mfixed = −1.16 ± 0.30 µV, Mrandom = −0.78 ± 0.28 µV) than in the shape-categorization group (Mfixed = −1.47 ± 0.30 µV, Mrandom = −1.58 ± 0.28 µV), F(1,30) = 3.27, p = .08, = 0.10. No other effects emerged (ps > .1). N2pc onset did not reliably vary in any analysis (ps > .1).

3.2.3 ERP results. Distractor-present trials: target and distractor in same hemifield (Figure ) Distractor-elicited early Pd

We focused on switch trials in analysis of the early Pd. A 2-way ANOVA with a between-subject factor for categorization group and a within-subject factor for task sequence identified a main effect of categorization group: early Pd was larger in the shape-categorization group (M = 0.98 ± 0.14 µV, M = 0.57 ± 0.14 µV), F(1,30) = 4.36, p = .045, = 0.13. A main effect of task sequence also emerged, with the Pd larger in fixed-sequence blocks (M = 0.99 ± 0.14 µV, M = 0.57 ± 0.12 µV), F(1,30) = 5.25, p = .03, = 0.15, as well as a trend toward an interaction of categorization group and task sequence, with the effect of task sequence more pronounced in the shape-categorization group (Mfixed = 1.36 ± 0.20 µV, Mrandom = 0.60 ± 0.18 µV) than in the color-categorization group (Mfixed = 0.61 ± 0.20 µV, Mrandom = 0.53 ± 0.18 µV), F(1,30) = 3.36, p = .08, = 0.10. Separate comparisons between the early Pd in fixed and in random-sequence blocks for each group showed that the marginal interaction of categorization group and task sequence was driven by the shape-categorization group (t(15) = 3.58, p = .003) rather the color-categorization group (t(15) = 0.28, p = .78).

Target-elicited N2pc

Focusing on the switch trials, the N2pc was analyzed with a 2-way ANOVA with a between-subject factor for categorization group and within-subject factor for task sequence. This identified a significant interaction of categorization group with task sequence, with the effect of task sequence more pronounced in the color-categorization group (Mfixed = −1.18 ± 0.32 µV, Mrandom = −.69 ± 0.30 µV) than in the shape-categorization group (Mfixed = −1.19 ± 0.32 µV, Mrandom = −1.48 ± 0.30 µV), F(1,30) = 5.77, p = .02, = 0.16. No other effects emerged (ps > .1) and there was no effect on N2pc onset latency (ps > .1).

3.2.4 ERP results. Distractor-present trials: target and distractor in opposite hemifield (Figure ) Distractor-elicited early Pd

As noted above, the Pd in this condition expresses as a negativity contralateral to the target (and thus a positivity contralateral to the distractor). Focusing on the switch trials, a 2-way ANOVA with a between-subject factor for categorization group and a within-subject factor for task sequence identified a significant interaction of categorization group with task sequence, with the effect of task sequence more pronounced in the shape-categorization (Mfixed = −1.31 ± 0.22 µV, Mrandom = −0.80 ± 0.18 µV) than in the color-categorization group (Mfixed = −0.54 ± 0.22 µV, Mrandom = −0.78 ± 0.18 µV), F(1,30) = 6.39, p = .02, = 0.18. This significant interaction was further analyzed by comparing the early Pd in fixed and random-sequence blocks for each group using two dependent sample t tests. This analysis showed that the amplitude of the early Pd significantly differed between fixed and ran