Neural Evidence for Feature-based Distractor Inhibition Open Access

In the visual world that surrounds us, there is an enormous number of stimuli that have the potential to attract our attention, such as email pop-up notifications while reading this research article or flashing advertisements while driving. Because the processing capacity of our visual system is limited, processing distracting information comes at the price of missing task-relevant information. Thankfully, our attention is not solely guided by stimulus-driven processes. Our visual system is equipped to handle distraction (Luck, Gaspelin, Folk, Remington, & Theeuwes, 2021), for instance, via learning of distractor regularities.

Accumulating evidence supporting that learning of distractor regularities reduces interference comes from distractor location learning (Hanne, Tünnermann, & Schubö, 2023; Theeuwes, Bogaerts, & van Moorselaar, 2022; Liesefeld & Müller, 2021; Failing, Feldmann-Wüstefeld, Wang, Olivers, & Theeuwes, 2019; Ferrante et al., 2018; Wang & Theeuwes, 2018). Typically, researchers used a variant of the additional singleton task (Theeuwes, 1991, 2018), where participants were asked to search for a shape singleton while ignoring a color singleton distractor. The distractor was more likely to be presented at one location (high-probability location) without informing the participants. Results showed reduced RTs for a high-probability distractor compared with when the distractor appeared at less likely locations (low-probability locations), indicating that observers can extract spatial regularities to reduce distractor interference. Several authors proposed that the likely distractor location is suppressed preemptively (Huang, Donk, & Theeuwes, 2023; Wang, van Driel, Ort, & Theeuwes, 2019), suggesting that at the level of the priority map, the weights of the most likely distractor location are attenuated to reduce distractor interference before presentation of the search display (Theeuwes et al., 2022). These findings align with the hierarchical framework by Luck and colleagues (2021), which proposes that spatial gain control can operate before attentional priority is computed to reduce interference of the high-probability distractor.

In addition to modulating distractor interference via spatial gain control, the hierarchical framework proposed by Luck and colleagues (2021) suggests that the visual system can also rely on feature gain control. However, compared with the large body of research on distractor location learning, distractor feature learning is less well understood, and the neural mechanism by which attentional priority is adapted is still under debate.

Earlier studies have shown that distractor interference is reduced when the same distractor color is presented throughout the experiment, leading to the signal suppression hypothesis (Gaspelin & Luck, 2018b; Gaspelin, Leonard, & Luck, 2015, 2017). The signal suppression hypothesis states that salient distractors elicit a strong “attend-to-me signal” (Sawaki & Luck, 2010) but can be actively suppressed via a control mechanism. In addition, it has been shown that distractor interference is reduced when a distractor color is presented repeatedly over a longer period compared with when the distractor color is newly introduced (Ramgir & Lamy, 2023; Vatterott, Mozer, & Vecera, 2018; Vatterott & Vecera, 2012). For example, in Vatterott and Vecera (2012), the color of a distractor remained the same throughout a block but changed between blocks. The results showed distractor cost at first, which disappeared in the second half of the block, indicating that observers needed to gain experience with the distractor color to reduce distractor interference (Vatterott et al., 2018; Gaspelin et al., 2015).

Furthermore, others have shown that observers can extract probabilistic feature regularities to reduce distractor interference (Golan & Lamy, 2022; Stilwell, Bahle, & Vecera, 2019). Stilwell and colleagues (2019) used an additional singleton task in which the distractor could appear in one of five colors, with one color being more likely than the others. The RTs were faster when the distractor appeared in a high-probability color than a low-probability color, indicating learning of distractor feature regularities to reduce distractor interference. The authors assumed that the weights within the priority map were adjusted based on experience with the distractor. Hence, suppression of distractor features might be initially reactive (i.e., distractor captures attention before it is inhibited), but with experience, it transitions to a more preemptive form of control (Gaspelin, Gaspar, & Luck, 2019; Stilwell et al., 2019; Vatterott & Vecera, 2012).

Electrophysiological measures can be used to examine the temporal dynamics of attention by analyzing the lateralized ERP components N2pc and distractor positivity (P_D). The N2pc is a negative-going deflection observed at posterior electrode sites contralateral to the attended stimulus typically elicited 200–300 msec after stimulus onset. The N2pc is considered an index of attentional allocation to a visual stimulus (Gaspelin et al., 2023; Luck, 2012; Luck & Hillyard, 1994). In contrast to the N2pc, the P_D is a positive-going deflection at contralateral posterior electrode sites, elicited between 100 and 500 msec after a salient distractor is presented laterally, alongside a target at the vertical midline (Gaspelin et al., 2023; Liesefeld, Liesefeld, & Müller, 2022; Weaver, van Zoest, & Hickey, 2017; Gaspar & McDonald, 2014; Kiss, Grubert, Petersen, & Eimer, 2012; Hickey, Di Lollo, & McDonald, 2009). The exact nature of the P_D is still under debate, but it is considered an index of attentional inhibition (Gaspelin et al., 2023). Earlier studies observed a P_D after the typical N2pc time window, most likely indicating suppression of the distractor after an initial shift of attention (Liesefeld et al., 2022; van Moorselaar, Daneshtalab, & Slagter, 2021; van Moorselaar & Slagter, 2019; Liesefeld, Liesefeld, Töllner, & Müller, 2017). There have also been reports of an early P_D emerging before the N2pc time window, which may suggest a suppression mechanism that operates preemptively to prevent an attentional shift toward the distractor (Stilwell, Egeth, & Gaspelin, 2022; Weaver et al., 2017; Sawaki & Luck, 2010). These latter findings are in favor of the signal-suppression hypothesis (Gaspelin & Luck, 2018b; Gaspelin et al., 2015, 2017). Thus, the P_D can occur in a variable time range with a more flexible timing than the N2pc (see Gaspelin et al., 2023, for a review).

Most EEG studies on the learning of distractor feature regularities also manipulated distractor location regularities (van Moorselaar et al., 2021; van Moorselaar, Lampers, Cordesius, & Slagter, 2020; van Moorselaar & Slagter, 2019). van Moorselaar and colleagues (2021) used the traditional additional singleton task, in which the features of the salient distractor were either fixed or varied randomly between trials (van Moorselaar et al., 2021). Spatial distractor probabilities were only applied in the second half of the experiment, so we focus here on the no-spatial bias blocks in the first half. The results showed smaller distractor cost in the fixed feature condition than in the mixed feature condition. In addition, at the neural level, an early P_D and an earlier peak of the late P_D were observed with fixed features, indicating that learning of a fixed distractor feature facilitates distractor inhibition (see also Gaspelin et al., 2023). However, in the study by van Moorselaar and colleagues (2021), the target feature was also predictable. Thus, the EEG results may reflect not only the inhibition of distractor features but also the up-weighting of target features (van Moorselaar, Huang, & Theeuwes, 2023; Gaspar & McDonald, 2014). Another study used steady-state visual evoked potential to examine the neural response to a learned distractor color before search display onset. The results showed a suppressed response to a learned distractor color compared with other colors, suggesting that distractor feature learning might operate preemptively (Carlisle & Zhang, 2021). However, although distractor feature learning has been demonstrated to be reliable, its neural mechanism is largely unknown.

Crucially, in studies on distractor feature learning, the high-probability color is not only more likely but also more likely to repeat in successive trials than the low-probability colors. Previous research showed that selection on a previous trial can affect selection on a subsequent trial (Ramgir & Lamy, 2022; Kristjánsson & Ásgeirsson, 2019; Thomson, D'Ascenzo, & Milliken, 2013; Maljkovic & Nakayama, 1994, 1996), a phenomenon of selection history known as intertrial priming. While statistical learning develops over time with exposure to the distractor (Hanne et al., 2023), intertrial priming is more short-lived. In their seminal article, Maljkovic and Nakayama (1994) asked participants to search for a uniquely colored diamond among differently colored diamonds. The target and distractor colors could either remain the same or swap colors between trials. Participants responded faster when the target and distractor colors were repeated in two consecutive trials compared with when they were changed between trials, indicating that encountering the same features improves task performance. In an additional experiment, Maljkovic and Nakayama (1994) examined the time frame within which intertrial priming operates. The authors found that feature repetition improved search performance even when the same target feature was encountered five to eight trials in the past, suggesting that not only immediately preceding trials, but also more distant trials can influence task performance. Because the trial sequence was unpredictable, the authors assumed that it was unlikely that the participants held the target color in memory and actively prepared for an upcoming trial. Instead, the results were interpreted as evidence for a passive memory process driving intertrial priming. The results also highlight that recent experiences with a target feature benefit attentional selection separately from goal-directed guidance.

To date, it is still controversial at which processing stage intertrial priming improves target detection in visual search tasks (Ramgir & Lamy, 2022; Kristjánsson & Ásgeirsson, 2019; Kruijne & Meeter, 2015). For instance, intertrial priming has been suggested to modulate attentional priority by adjusting the weights of the target and distractor features. Alternatively, intertrial priming might benefit memory retrieval after priority has been computed, thereby speeding up response-related processes. In a recent review, Ramgir and Lamy (2022) proposed that intertrial priming modulates how fast a stimulus is selected or rejected. The authors focus on perceptual rather than on response-related processes, assuming that intertrial priming might influence the processing of the target and distractor after attentional priority has been computed. In either case, intertrial priming is a phenomenon that is well known to facilitate target detection and might also play a role in studies on distractor feature learning in visual search tasks.

Indeed, in their study on distractor feature learning, Golan and Lamy (2022) observed that the decrease in RTs in trials with a distractor presented in a high- compared with a low-probability color was partly due to intertrial priming. In addition, van Moorselaar and Slagter (2019) reported a reduced P_D when the distractor was repeatedly presented, although, in their study, the location was repeated rather than a nonspatial feature. In contrast to van Moorselaar and Slagter (2019), other studies did not provide evidence that the P_D was influenced by intertrial priming (Gaspelin et al., 2023; van Moorselaar et al., 2021; Feldmann-Wüstefeld & Schubö, 2016). In addition, Feldmann-Wüstefeld and Schubö (2016) did not observe a modulation of the target N2pc amplitude by distractor repetition. Hence, it is not yet clear what role intertrial priming plays in the learning of statistical distractor regularities and under which conditions it influences the P_D and N2pc amplitude.

In addition, it is not yet well understood whether learned distractor feature regularities transfer to later cognitive processes involved in processing these learned features, such as visual working memory (VWM; Wang, Cong, & Woodman, 2023; Won, Venkatesh, Witkowski, Banh, & Geng, 2022). Previous studies have provided some evidence supporting this idea. For instance, the effect of distractor feature learning persists for some trials in an extinction phase, in which learning is no longer possible (Golan & Lamy, 2022). Thus, distractor feature learning might last long enough to interact with later cognitive processes, thereby impairing VWM performance for a feature previously associated with a frequently occurring distractor. Furthermore, a recent study by Won and colleagues (2022) showed an impact of learned distractor suppression on memory performance. The authors used a high-powered study design to examine memory precision based on a single-probe trial per participant. Participants performed an additional singleton task and were asked to recall the last presented distractor color after a systematically manipulated number of trials. With more trials and, thus, more time to learn how to handle distraction, memory precision was reduced. This suggests that distractor suppression operates at the sensory level to prevent readout to other cognitive processes (Won et al., 2022). In a more recent study, Wang and colleagues (2023) implemented a color change detection task (Luck & Vogel, 1997, 2013) to assess VWM capacity after statistical learning of a target color in an (independent) visual search task. The results showed that higher VWM capacity was correlated with more pronounced learning benefits observed with a high-probability target color. Thus, individual VWM performance seems to be a central factor in statistical learning of target regularities. On this basis, it might be that learned distractor feature regularities transfer to later cognitive processes, such as VWM performance.

The current study followed two goals. First, we wanted to learn more about the neural mechanisms associated with distractor feature learning. To this end, we measured EEG while participants performed an adapted variant of the additional singleton task. In this task, the color distractor appeared more often in one color than in the remaining colors. If participants learn this feature regularity to reduce distractor interference, we expect to find less distractor interference for distractors that appear in the high-probability color compared with the low-probability colors (Golan & Lamy, 2022; Stilwell et al., 2019). We implemented a search task where participants had to search for a shape singleton (diamond among circles or vice versa), and the distractor could appear in one out of five colors. Previous studies that used a similar mixed-feature design have observed a P_D occurring with a latency of 300–364 msec (Liesefeld et al., 2022; van Moorselaar et al., 2021; McDonald, Green, Jannati, & Di Lollo, 2013). Therefore, at the neural level, we assume that any potential distractor modulations would appear as changes in a late P_D time window. Furthermore, in earlier studies where the distractor feature was predictable, a reduction in the late P_D time window has been found, indicating that participants prepared for an upcoming distractor thereby decreasing interference (Abbasi, Henare, Kadel, & Schubö, 2023; Liesefeld et al., 2022). On the basis of these findings, we reason that participants learn to prepare for the high-probability color and, as a result, less suppression might be required when the distractor appears in the more likely color. Thus, we expect to observe a smaller P_D to lateral distractors elicited by a high-probability distractor color compared with low-probability colors (see also van Moorselaar & Slagter, 2019, 2020; Heuer & Schubö, 2020). Because the high-probability color was more often repeated in successive trials than a low-probability color, we divided the high-probability trials into repeated and nonrepeated trials to examine the impact of feature repetition on search performance. Furthermore, in an exploratory analysis, we investigated whether repeated exposure to the high-probability distractor color in successive trials modulates distractor suppression with extensive practice. If suppression becomes more efficient with increasing exposure to the distractor and, thus, more time to learn about the distractor features (see also Gaspelin et al., 2019; Stilwell et al., 2019; Vatterott & Vecera, 2012), we expect to observe a shift in the P_D to an earlier time window. Thus, we compared the P_D elicited by repeated high-probability trials in both early and late time windows between the first and second half of the experiment. Finally, if learned feature suppression facilitates target selection, we expect to find a larger target-elicited N2pc in trials with a high-probability distractor color compared with low-probability colors.

Second, we examined whether distractor feature learning affects VWM performance. To this end, we implemented a change detection task using distractor colors identical to those used in the learning task. If learned feature suppression affects VWM performance by reducing sensory processing of the high-probability color, we expect to observe lower performance for high-probability colors than for low-probability colors in the second but not the first run of the memory task.

The sample size was based on Stilwell and colleagues (2019) and Golan and Lamy (2022), who showed distractor feature learning to be a medium–large effect (dz = 0.49 and 0.87). A power analysis in G*Power (Faul, Erdfelder, Lang, & Buchner, 2007) with an alpha level of .05, a power of .80, and the more conservative dz of .49 indicated that at least a sample size of 28 is needed to find the expected distractor feature learning effect using a one-tailed paired t test. Considering that the power analysis was based on behavioral results, we decided to increase the number of participants to 30 to increase the power at the neural level.

Four participants were replaced due to dropout during the experiment, behavioral performance below chance level (accuracy < 50% in one of the two tasks), or excessive eye movements in EEG (>25% of trials were rejected; see details below), resulting in the planned sample size (M = 22.5 years, 15 female and 15 male participants). Participants were required to give written informed consent before the experiment and received either course credit or monetary compensation (10 €/hr) for their participation. The study was approved by the local Ethics Committee of the Department of Psychology at Philipps-University Marburg and conducted in accordance with the Declaration of Helsinki. Visual acuity and color vision were tested using an OCULUS Binoptometer 3 (OCULUS Optikgeräte GmbH).

Participants were seated comfortably in a sound-attenuated, electrically shielded, and dimly lit room. They viewed a 22-in. VPixx VIEWPixx monitor (VPixx Technologies Inc.) at a distance of 100 cm. For stimulus control, E-Prime 2.0 software (Psychology Software Tools, Inc.) was used on a Windows 7 PC. Behavioral responses were recorded with a gamepad (Speedlink Thunderstrike Gamepad). The study consisted of a learning task and a change detection task.

In the learning task, the visual search display contained eight stimuli, either seven circles (1.95° diameter) and one diamond (1.55° × 1.55°) or vice versa. The stimuli were arranged in an imaginary circle (11.14° diameter) around a fixation point (0.17° diameter), with a distance of 4.3° between the center of two adjacent stimuli. In the stimuli, either a vertical or a horizontal line was embedded (1.26° × 0.06°, gray: RGB 108, 108, 108). In distractor-absent trials, the outlines of all stimuli were gray (RGB: 108, 108, 108). In distractor-present trials, one of the nontargets had a different color (color singleton), randomly chosen from five possible colors (green: RGB 0, 142, 0; yellow: RGB 167, 93, 0; blue: RGB 0, 111, 234; red: RGB: 255, 0, 0; magenta: RGB: 210, 11, 213).

In the change detection task, colored squares (1.66° × 1.66°) were randomly presented at 16 candidate locations evenly distributed around the fixation point (0.17° diameter) in an imaginary 4 × 4 matrix (innermost square: 4.12° × 4.12°; distance between two stimuli: 0.8°). In addition to the five distractor colors used in the learning task, we included another color (purple: RGB 133, 62, 252). Any improvement in this neutral color in the second run can be interpreted as resulting purely from a practice effect. In both experiments, all stimuli were presented against a black background. All colors were measured to have the same luminance (∼25 cd/m²) using a Minolta LS-110 luminance meter (Konica Minolta Holdings Inc.).

The trial procedure and sample displays are shown in Figure 1. The learning task was an adapted variant of the additional singleton task. Each trial started with a fixation for 600 msec, followed by a visual search display for another 200 msec. Participants were asked to search for the shape singleton (circle or diamond) and to report the orientation of the embedded line (horizontal or vertical) as quickly and correctly as possible while ignoring the color singleton. The trial was terminated by pressing one of the two shoulder buttons with the left or right index fingers or after a maximum of 1200 msec. Assignment of the buttons (horizontal vs. vertical) was counterbalanced across participants. The intertrial interval was randomly jittered between 700 and 1200 msec. The task consisted of 1,512 trials in 28 blocks of 54 trials each. In two thirds of all trials, a color singleton distractor appeared. The distractor was presented in one of five colors in two thirds of the distractor-present trials (high-probability color). In the remaining distractor-present trials, the distractor color was chosen from the other four colors with equal probability (low-probability color). Thus, the high-probability color appeared eight times more likely than one of the low-probability colors. The colors were counterbalanced across participants. The target and the distractor were presented only on the horizontal or vertical axis, and their locations were balanced within the experiment. A practice block was implemented before the experiment, which participants could repeat once if the accuracy of the first practice block was below 50%. They needed to reach an accuracy above 50% to continue with the experiment. Participants could take a short break within and after each block. After every 10 blocks, they could take a longer break.

The change detection task was run once before and once after the learning task. In the change detection task, a memory array consisted of four colored squares, which was presented for 250 msec after a fixation of 1000 msec. Participants were required to remember the colors and their locations. After a retention interval of 1000 msec, a colored item appeared on the probe display at one of the four locations presented in the memory array. Participants were instructed to respond as quickly and accurately as possible whether the color of the item at that location had changed or not. The trial was terminated by pressing one of the two shoulder buttons with the left or right index fingers. Assignment of the buttons (change vs. no change) was counterbalanced across participants. The task consisted of 288 trials in four blocks of 72 trials each. Each color had an equal probability to be tested (48 trials per color). The probed item changed its color in half of the trials. In the first run of the task, participants completed 24 practice trials and needed to reach an accuracy above 50% to continue with the experiment. The practice block could be repeated once.

In both tasks, participants were instructed to keep their gaze fixed on a fixation point throughout a trial. Following each response, a warning message was presented for 1000 msec if the participants answered incorrectly or too slowly (>1200 msec). Feedback on mean accuracy and RTs were given after each block. At the end of the experiment, participants completed a postexperimental questionnaire. They were queried as to whether they had noticed any regularity in the learning task. Regardless of their response, participants were instructed to provide estimates for the probability of each distractor color occurring. The estimates should add up to 100% (Vicente-Conesa, Giménez-Fernández, Luque, & Vadillo, 2023).

The data were preprocessed using R Version 4.0.0 (R Core Team, 2020), and the statistical analyses were performed using JASP 0.17.1 (JASP Team, 2023). For the analysis of RTs, trials with incorrect responses and trials in which the RTs exceeded 2.5 SDs of the condition-specific RTs calculated for each participant per block were removed. Trials in which no response occurred within the specified time interval of 1400 msec after search display onset were considered incorrect trials. In the change detection task, the condition with a high-probability color had only a small number of 48 trials, whereas the condition with low-probability colors had four times as many trials. To account for this imbalance in trial numbers, the trials in the low-probability condition were bootstrapped, generating 1000 resamples by replacement sampling with approximately the same number of trials as in the high-probability condition. From the resamples, mean accuracy and RTs were calculated. When accuracy was submitted to statistical tests, we applied the variance-stabilizing arcsine transformation to account for the fact that the data were bounded. In the figures, the nontransformed data are plotted. In the learning task, there was a considerably larger number of trials (336 trials in the low-condition and 672 trials in the high-condition) and, thus, no bootstrapping was performed. The data were analyzed using variance analyses (repeated-measures ANOVAs) and planned contrasts. A p value below .05 was considered significant. If sphericity was violated, Greenhouse–Geisser correction was applied.

The EEG was recorded from 64 Ag/AgCI slim active signal electrodes with a BrainAmp amplifier (actiCAP, Brain Products) at a sampling rate of 1000 Hz (high cutoff: 250 Hz, low cutoff: 0.016 Hz) during the learning task. The electrodes were placed following the International 10–20 system, and impedances were reduced below 5 kΩ. The vEOG was recorded by deriving bipolar recordings from an electrode placed below and above the left eye (AF7) and hEOG was recorded from electrodes FT9 and FT10 placed near the outer canthi of each eye. The EEG signal was referenced online to FCz.

The EEG data were preprocessed using custom scripts written in MATLAB Version 2022b (Mathworks) and the toolbox Fieldtrip (Oostenveld, Fries, Maris, & Schoffelen, 2011). JASP 0.17.1 (JASP Team, 2023) was used for statistical analyses. All reported time windows are time-locked to the onset of the visual search display. The EEG data were rereferenced offline to the average of the mastoid electrodes (TP9 and TP10) and band-pass filtered at 1–35 Hz. Next, the continuous EEG data were segmented to cover an epoch of 700 msec, starting 200 msec before search display onset and ending 500 msec after the search display appeared. Trials removed in the behavioral analyses were also excluded from EEG processing. Segments in which participants performed eye movements (hEOG > ±50 μV) or blinked (vEOG > ±80 μV) within the critical time window of −200 msec to 350 msec were removed. In addition, segments containing voltages exceeding 80 μV at the parieto-occipital electrodes of interest (PO3/4 and PO7/8) within the same time window were excluded. After preprocessing, 80% of artifact-free trials remained, with values ranging between 77% and 82% across the three experimental conditions (high, low, and absent). The remaining epochs were then baseline-corrected using a time interval of 200 msec preceding search display onset.

We computed the ERPs separately for the three distractor conditions within a time window of −200 to 500 msec. We focused on trials in which either the target was presented on the horizontal midline (target lateral) and the distractor above or below the fixation point, or vice versa (distractor lateral). Contra- and ipsilateral mean activity of the electrodes of interest was calculated separately for distractor- and target-elicited waveforms and for each distractor condition. We determined the time windows for the statistical analyses as a range of 25 msec around the cross-conditional peaks (high- and low-probability trials) of the difference waveforms (contralateral − ipsilateral) within prespecified time intervals (N2pc: 200–300 msec, P_D: 200–400 msec); see van Moorselaar et al., 2021; Feldmann-Wüstefeld & Vogel, 2019, for a similar approach). This approach resulted in a time window of 224–274 msec for the N2pc and 276–326 msec for the P_D. In an additional exploratory analysis described below, we analyzed the positivity observed in an earlier time window. For this earlier P_D time window, we used a prespecified time interval of 100–200 msec and a time window of 129–179 msec for statistical analysis. If sphericity was violated, Greenhouse–Geisser correction was applied.

First, we examined whether participants extracted the feature regularities to reduce distractor interference. Hence, a one-way repeated-measures ANOVA was performed on mean RTs (see Figure 2) with the factor Distractor Condition (high, low, absent). The results revealed a main effect of Distractor Condition, F(2, 58) = 43.30, p < .001, η_p² = .60. Planned contrasts showed that the mean RT was reduced when the distractor was presented in a high-probability color (M = 689.11 msec, SE = 12.83) compared with a low-probability color (M = 701.60 msec, SE = 12.37); t(58) = 5.11, p ≤ .001, indicating distractor feature learning. RTs were still fastest when no distractor was presented (M = 678.88 msec, SE = 13.54); t(58) = 4.19, p ≤ .001. No speed-accuracy trade-off was detected.

Due to the probability manipulation, the distractor repeatedly appeared in the same color in several consecutive trials. To examine whether feature learning effects were driven by intertrial priming, we conducted a similar analysis as described above, separating the high-probability trials into trials in which the high-probability color repeated from the previous trial (high repeated) and trials in which a low-probability color or a distractor-absent trial was presented in the previous trial (high nonrepeated; see Figure 3). A one-way repeated-measures ANOVA with the factor Distractor Condition (high repeated, high nonrepeated, low) revealed a significant main effect, F(1.64, 47.63) = 15.44, p < .001, η_p² = .35. Planned contrast showed faster RTs when the distractor appeared repeatedly in a high-probability color (M = 687.82 msec, SE = 12.88) compared with a low-probability color (M = 701.60 msec, SE = 12.37); t(58) = 5.17, p < .001. However, there was no significant difference between repeated and nonrepeated high-probability trials, t(58) = 0.83, p = .411, and also the nonrepeated high-probability trials were faster (M = 690.03 msec, SE = 12.97) than the low-probability trials, t(58) = 4.35, p < .001. Thus, the observed decrease in distractor interference with a high-probability distractor color cannot be explained by priming from the previous trial.

To examine whether participants had noticed the color regularity, we analyzed their postexperimental reports, in which they indicated the frequency with which they assumed the distractor had appeared in a particular color. The estimates for each low-probability color were averaged within each participant, facilitating a more direct comparison with the high-probability color. Overall, 40% of the participants indicated that they had noticed a regularity in the learning task. Regardless of their response, the average estimate from all participants indicated that they assumed that the distractor had appeared in 25% of trials in a high-probability color and in 20.9% of trials in a low-probability color. For those participants who indicated noticing a regularity in the learning task, the estimated percentages were similar (21.25% high, 19.6% low). Given that the estimated percentages are around the baseline for random guessing at 20%, these results suggest that the participants did not notice the feature regularity.

To examine whether the learning of distractor feature regularities in the visual search task affected VWM performance in the change detection task, we submitted the mean accuracy and mean RTs to two-way repeated-measures ANOVAs with the factors Distractor Condition (high, low) and Session Run (first, second). For both dependent measures, the main effect of Session Run was significant, accuracy: F(1, 29) = 7.47, p = .011, η_p² = .205; RTs: F(1, 29) = 17.08, p < .001, η_p² = .371. More specifically, accuracy increased and RTs were faster in the second run (accuracy: M = 76.46%, SE = 1.73; RT: M = 655.04 msec, SE = 16.57; see Figure 4, right column) compared with the first run (accuracy: M = 72.57%, SE = 1.64; RT: M = 702.25 msec, SE = 19.20; see Figure 4, left column), indicating an improved performance when the task was repeated after the search task. The main effect of Distractor Condition was significant only for the mean RTs, accuracy: F(1, 29) = 0.52, p = .48, η_p² = .018; RTs: F(1, 29) = 8.15, p = .008, η_p² = .219, with faster RTs when the probe item was presented in a low-probability distractor color (M = 670.22 msec, SE = 18.0) compared with a high-probability distractor color (M = 687.07 msec, SE = 16.47). Most importantly, in both dependent measures, the interaction effect between Condition and Session was not significant, accuracy: F(1, 29) = 0.92, p = .345, η_p² = .031; RTs: F(1, 29) = 0.71, p = .406, η_p² = 0.024, indicating no modulation of VWM performance by distractor feature learning.

To analyze whether the distractor P_D was modulated by distractor feature learning, we submitted the mean amplitude observed in the P_D time window to a two-way repeated-measures ANOVA with the factors Laterality (contralateral, ipsilateral) and Distractor Condition (high, low). The results showed a significant main effect of Laterality, F(1, 29) = 33.76, p < .001, η_p² = .538, but no difference between distractor conditions, F(1, 29) = 0.86, p = .363, η_p² = .029. Importantly, an interaction between both factors was detected, F(1, 29) = 10.12, p = .003, η_p² = .259. This interaction was due to a smaller P_D being found when the distractor appeared in a high-probability color (M = 0.59 μV, SE = 0.13) compared with a low-probability color (M = 0.97 μV, SE = 0.17; see Figure 5C).

Next, we investigated whether the differences found in the P_D time window (276–326 msec) were driven by intertrial priming. Thus, we separated the high-probability trials into trials that were preceded by a high-probability trial (high repetition) and trials that were preceded by a low-probability or distractor-absent trial (high nonrepetition; see Figure 6). We submitted the mean amplitudes to a two-way repeated-measures ANOVA with the factors Laterality (contralateral, ipsilateral) and Distractor Condition (high repetition, high nonrepetition, low). The results revealed a significant main effect of Laterality, F(1, 29) = 30.54, p < .001, η_p² = .513, but no reliable differences between distractor conditions, F(2, 58) = 0.406, p = .668, η_p² = .014. Interestingly, the interaction was significant, F(1.58, 45.76) = 5.44, p = .012, η_p² = .158. Planned contrasts showed a smaller P_D for repeated trials (M = 0.47 μV, SE = 0.16) compared with low-probability trials (M = 0.97 μV, SE = 0.17); t(58) = 3.29, p = .002. There was no significant difference between nonrepeated trials (M = 0.69 μV, SE = 0.14) and low-probability trials, t(58) = 1.87, p = .066, as well as between the repeated and nonrepeated high-probability trials, t(58) = 1.42, p = .162. These results indicate that the observed effects of distractor feature learning on distractor suppression were partly due to intertrial priming.

In an exploratory analysis, we examined whether participants' ability to prepare for the likely distractor color improved with increasing repetition and increasing exposure to the distractor. To this end, we analyzed the activity in the same P_D time window as above (276–326 msec), and also in an earlier time window (129–179 msec). We did this analysis for the repeated high-probability trials, as in these trials, both intertrial priming and feature learning become effective. If suppression builds up with repeated exposure, we expect to see a modulation in the ERP pattern when comparing the first and second halves of the experiment (see Figure 7). We performed a two-way repeated-measures ANOVA with the factors Laterality (contralateral, ipsilateral) and Experimental Half (first, second).

For the late P_D time window (276–326 msec), there was a significant main effect of Laterality, F(1, 29) = 8.57, p = .007, η_p² = .23, but no difference between the experimental halves, F(1, 29) = 1.23, p = .276, η_p² = .041. Crucially, the interaction effect was significant, F(1, 29) = 4.92, p = .035, η_p² = .145, indicating that the P_D was modulated across time: The P_D was smaller in the second half of the experiment (M = 0.29 μV, SE = 0.19) than in the first half (M = 0.64 μV, SE = 0.17).

In addition to the differences in the P_D time window, the data also showed differences in an earlier time window. In some studies, this early time window has been related to an inhibitory process that occurs before the first shift of attention (Stilwell et al., 2022; Weaver et al., 2017; Sawaki & Luck, 2010). For the early P_D time window (129–179 msec), we observed significant main effects of Laterality, F(1, 29) = 13.21, p = .001, η_p² = .313, and Experimental Half, F(1, 29) = 11.52, p = .002, η_p² = .284. The positivity was found to be larger in the second half (M = 0.53 μV, SE = 0.11) than in the first half of the experiment (M = 0.16 μV, SE = 0.13); F(1, 29) = 5.88, p = .022, η_p² = .169, indicating temporal modulation of this early positivity.

In studies on distractor location learning, the target N2pc was found not to differ between high- and low-probability trials, suggesting that distractor location learning did not affect attentional allocation to the target (van Moorselaar et al., 2021; Wang et al., 2019). Yet, it is unclear whether distractor feature learning influences the target N2pc. We calculated a two-way repeated-measures ANOVA on the mean amplitudes with the factors Laterality (contralateral, ipsilateral) and Distractor Condition (high, low, distractor absent). The results showed main effects of Laterality, F(1, 29) = 7.45, p = .011, η_p² = .204, and Distractor Condition, F(1.58, 45.75) = 5.60, p = .011, η_p² = .162, and a significant interaction between both factors, F(2, 58) = 3.85, p = .027, η_p² = .117. Planned contrasts revealed a larger N2pc amplitude for the high-probability condition (M = −0.33 μV, SE = 0.09) compared with the low-probability condition (M = −0.09 μV, SE = 0.11); t(58) = 2.55, p = .013, indicating more efficient attentional allocation in trials where the distractor appeared in the more likely color. Interestingly, the N2pc did not differ when a high-probability distractor was presented compared with when no color distractor was presented (M = −0.30 μV, SE = 0.10); t(58) = 0.33, p = .74. The difference between low-probability trials and distractor-absent trials was reliable, t(58) = 2.22, p = .030.

Given that the repetition of the high-probability color in consecutive trials could also have modulated the N2pc, we further divided the trials into low-probability trials, and repeated and nonrepeated high-probability trials (see Figure 8C, right column). The results of a two-way ANOVA with the factors Laterality (contralateral, ipsilateral) and Distractor Condition (high repetition, high nonrepetition, low) showed main effects of Laterality, F(1, 29) = 8.44, p = .007, η_p² = .225, and Distractor Condition, F(2, 58) = 5.26, p = .008, η_p² = .153, but no interaction effect, F(2, 58) = 2.68, p = .077, η_p² = .085. Planned contrasts revealed a reliable difference between repeated high-probability trials (M = −0.36 μV, SE = 0.12) and low-probability trials (M = −0.09 μV, SE = 0.11); t(58) = 2.20, p = .032. The differences between the nonrepeated high-probability trials (M = −0.30 μV, SE = 0.10) and low-probability trials, t(58) = 1.73, p = .089, as well as between the two types of high-probability trials, t(58) = 0.47, p = .64, were not statistically significant. These results suggest that the observed difference in the N2pc amplitude between high- and low-probability trials was partly driven by intertrial priming.

The goal of our study was twofold. We examined the neural mechanisms of distractor feature learning by employing a variant of the additional singleton task while simultaneously recording EEG. Participants were asked to indicate the orientation of the line in the shape singleton and ignore the distracting color singleton. Because the target was a shape singleton (i.e., either a diamond among circles or vice versa, changing unpredictably across trials), participants needed to adopt singleton-detection mode to find the target. Crucially, the color singleton was more likely to appear in one of five colors, allowing participants to learn this statistical regularity of the distractor feature to reduce distractor interference (Golan & Lamy, 2022; Stilwell et al., 2019). If participants are able to learn the distractor feature regularities, we expected to observe faster RTs and a smaller P_D when the distractor appeared in the more likely color than in less likely colors. Moreover, if distractor feature learning transfers to later cognitive processes such as VWM performance, we assumed to find lower performance in a visual–spatial change detection task for the high-probability color compared with low-probability colors.

Previous studies on distractor feature learning reported a RT benefit when the distractor was presented in a high-probability color compared with low-probability distractor colors (Golan & Lamy, 2022; Stilwell et al., 2019). These results indicate that participants learned the feature regularities of the distractor to reduce distractor interference. Consistently, our results showed faster RTs with a high-probability color compared with low-probability distractor colors, providing additional evidence for distractor feature learning. According to the framework by Luck and colleagues (2021), distractor feature learning might attenuate the salience signals of the high-probability color via feature gain control. Consequently, the relative attentional priority of the high-probability distractor is reduced within the priority map, making observers less prone to distractor interference. According to these considerations, the modulation of feature gain control is a suppression mechanism that operates preemptively and that is triggered automatically with repeated stimulus exposure.

In earlier studies, in which participants could prepare for an upcoming distractor, a reduced P_D was observed relative to trials without such preparation, which was interpreted as evidence for a preemptive operating mechanism that reduces the need for distractor suppression (Abbasi et al., 2023; Liesefeld et al., 2022). Thus, we assumed to find a less pronounced P_D elicited by the high-probability distractor color compared with low-probability distractor colors. Indeed, our results showed a reliable P_D when a distractor was present, which was smaller when the distractor appeared in the more likely color. In accordance with prior findings, one can conclude that preemptive preparation reduces the need for distractor suppression with search display onset. When considering the observed benefits in RTs with a likely distractor color (Golan & Lamy, 2022; Stilwell et al., 2019; Vatterott & Vecera, 2012), the results suggest that the visual system uses distractor feature learning to reduce distractor interference, presumably by decreasing the attentional priority of the distractor within the priority map. As a consequence, the distractor requires less suppression. In addition, earlier studies reported a reduced P_D when the distractor appeared at a more likely location (van Moorselaar et al., 2020), when participants had foreknowledge of the upcoming distractor location (Heuer & Schubö, 2020), or when distractor priming occurred in successive trials (Abbasi et al., 2023; van Moorselaar & Slagter, 2019). Hence, distractor suppression is reduced not only when its likely location is anticipated but also when a distractor feature, such as color, can be predicted.

It is important to note that there is some discussion about whether the visual system can, in advance, prepare for features of upcoming distractors, thereby reducing the need for distractor suppression (Luck et al., 2021; van Moorselaar & Slagter, 2020). Because the analysis of the P_D is time-locked to search display onset, it only provides insights into the outcome of preemptive suppression processes but does not reflect preemptive processes themselves. Further research will provide a deeper understanding of the underlying mechanisms. Nevertheless, our results provide evidence that distractors can be handled more efficiently when feature regularities can be learned, reducing the need for distractor suppression upon presentation of the distractor.

The signal suppression hypothesis proposes that a salient distractor generates an “attend-to-me” signal resulting in a high level of activation at the priority map, thereby competing for attentional selection. However, this salience signal can be actively inhibited to prevent distractor interference via modulating the feature gain of the distractor before attentional priority is computed (Stilwell et al., 2022; Gaspelin & Luck, 2018a; Gaspelin et al., 2015; Sawaki & Luck, 2010). For instance, Gaspelin and Luck (2018a) had participants search for a unique target shape among heterogeneous nontargets, with a salient distractor appearing in some trials. In randomly intermixed probe trials, letters were overlaid on the search stimuli, and participants were asked to report as many letters as possible. The authors observed fewer recalls of distractor probes than of nontarget probes, indicating that the distractor was suppressed below the baseline level. Similarly, electrophysiological results showed a distractor P_D in the typical N2pc time window but no N2pc, suggesting suppression of the distractor feature before an initial shift of attention. In contrast to studies supporting the signal suppression hypothesis, however, our results showed a P_D in a much later time window, after an initial shift of attention would have been possible. This divergent finding might be a consequence of the differences in task design and, thus, the different attentional strategies participants used to find the target. More detailed, in studies on feature-based suppression, typically search displays with a fixed target shape and a fixed distractor color presented among heterogenous nontargets are used, forcing participants to adopt feature-search mode (Gaspelin & Luck, 2018b; Bacon & Egeth, 1994). As both the target and distractor features remain fixed, participants can preemptively prepare for the upcoming distractor, which facilitates the suppression of the distractor upon presentation of the search display. In contrast, in our task, participants had to search for a singleton target, and the distractor could appear in one of five colors. Target and distractor features were less predictable, and participants had to rely on the statistical distractor regularities to prepare for a distractor in an upcoming trial. This form of distractor preparation seems less effective than the preparation described in the studies above and might, therefore, result in a different pattern of distractor suppression.

Interestingly, and consistent with the signal suppression hypothesis, we did not observe a distractor-elicited N2pc, which would have indicated that attention had been captured by the distractor. However, there was neither a distractor-elicited N2pc in high- nor in low-probability trials. Because participants could not benefit from learning in low-probability trials, this absence of the N2pc cannot be attributed to statistical learning in these trials. Some findings suggest that the N2pc does not necessarily reflect the allocation of attention to an item but rather indexes an enhancement of the attended stimulus location, which, for example, allows for the identification of stimulus features (van Moorselaar et al., 2020; Zivony, Allon, Luria, & Lamy, 2018). Because in our study a gray shape singleton target was presented with a color distractor, participants could easily distinguish the distractor from the target based on the color information, which might explain why no distractor-elicited N2pc was detected. Consistently, studies with target and distractor defined in the same feature dimension report a distractor-elicited N2pc, likely because distinguishing the target from the distractor was more difficult (Gaspelin et al., 2023; Liesefeld et al., 2022; van Moorselaar et al., 2020; van Moorselaar & Slagter, 2019).

Furthermore, we examined whether distractor feature learning affected attentional allocation to the target. The results revealed a larger target-elicited N2pc when the target was presented together with a high-probability distractor color compared with a less likely distractor color, indicating that attentional allocation to the target was little hindered by a distractor in a high-probability color. In fact, the results showed that the amplitude of the target-elicited N2pc was similarly pronounced with a simultaneously presented high-probability distractor color compared with when a distractor was absent, indicating that attentional allocation was highly efficient even with a high-probability distractor color. In distractor location learning, in contrast, the distractor regularity did not modulate the target N2pc (van Moorselaar et al., 2020; Wang et al., 2019). Distractor location learning and distractor feature learning might, however, differ at the neural level, reducing the comparability of the findings.

Because the distractor appeared more frequently in one specific color, distractors of the same high-probability color were presented more often in consecutive trials than distractors of any low-probability color. Thus, lingering influences from the previous trial on attentional allocation, known as intertrial priming (Ramgir & Lamy, 2022; Lamy & Kristjánsson, 2013; Maljkovic & Nakayama, 1994), might have partly driven our results. To investigate this possibility, we separated the trials in which a high-probability distractor was preceded by a high-probability distractor in the previous trial from trials in which it was preceded by a different trial type. Analysis of the behavioral results showed that both repeated and nonrepeated high-probability trials were reliably different from the low-probability trials, indicating that statistical learning accounted for our results. When examining mean P_D and target N2pc amplitudes, only a difference for the repeated high-probability trials was found, but no difference between the two types of high-probability trials. Thus, the observed effects of distractor feature learning at the neural level were partly due to intertrial priming.

Our findings align with the results by Golan and Lamy (2022), who also found that the observed difference between high-probability and low-probability color was partly driven by intertrial priming. To date, the neural mechanisms of intertrial priming are still under debate (Ramgir & Lamy, 2022; Kristjánsson & Ásgeirsson, 2019). On the basis of the review by Ramgir and Lamy (2022), intertrial priming facilitates the selection or rejection of a stimulus. This process is assumed to occur after attentional priority has been determined, whereas distractor feature learning is assumed to operate via feature gain control, and modulate priority computations in preparation for the upcoming distractor (Luck et al., 2021).

Transferred to our study, RTs in trials with a repeated high-probability distractor color might have decreased because participants could benefit from the control settings of the previous trial. More specifically, when the distractor feature matched the feature of the previous trial, participants could discard a distracting item faster and focus on the selected target (Ramgir & Lamy, 2022). Similarly, the P_D elicited by a high-probability distractor might have been less pronounced due to faster rejection of the distractor when the more likely distractor color was repeatedly presented. Because the distractor appeared at the four cardinal locations, it is worth noting that location repetition may have also contributed to intertrial priming to some extent.

When inspecting Figure 6B, it is noticeable that the individual data show pronounced differences in the P_D between repeated and nonrepeated trials. Hence, some individuals might have been able to use distractor feature learning to reduce distractor interference, whereas others relied on intertrial priming. van Moorselaar and colleagues (2021) divided their participants into “learners” and “nonlearners.” Learners were participants who showed impaired target selection for targets presented at a high-probability distractor location compared with low-probability distractor locations, indicating that these participants used distractor location learning to reduce distractor interference. Similarly, the handling of distractor features might differ on the individual level in the present experiment. Because, in our study, there was no high-probability location, learners could not be detected in the same way as in van Moorselaar and colleagues (2021). An alternative would have been to define participants as learners who showed faster RTs in nonrepeated high-probability trials than in low-probability trials. In this approach, however, the number of nonlearners was only a few participants, which was insufficient for a reliable statistical analysis.

Furthermore, in an exploratory analysis, we examined whether increasing repetition and exposure to the high-probability distractor color modulate the dynamics of distractor suppression. If distractor suppression builds up over time, distractor handling should improve between the first and second halves of the experiment, which would result in a less pronounced P_D in the second half. Indeed, we found that the P_D was smaller in the second half than in the first half, highlighting that extensive practice with the high-probability distractor feature reduces the need for distractor suppression (van Moorselaar et al., 2021; van Moorselaar & Slagter, 2019). Interestingly, we also observed an early positivity in a time window that precedes the N2pc. This early positivity became more pronounced in the second half of the experiment. Although there is some discussion about whether an early positivity can be interpreted in terms of preemptive suppression (Gaspelin et al., 2023) or whether it merely reflects an early salience signal (Barras & Kerzel, 2016, 2017; Fortier-Gauthier, Moffat, Dell'Acqua, McDonald, & Jolicœur, 2012), a modulation of the early positivity across time might hint at more than a purely passive modulation of distractor suppression, providing additional evidence that exposure to the distractor is needed to handle distraction efficiently (Stilwell et al., 2019; Vatterott & Vecera, 2012). For instance, with extensive practice, distractor feature learning might combine with intertrial priming, thereby facilitating postpriority computations such as distractor rejection, as well as processes occurring before attentional priority is determined such as downregulating the feature gain of the distractor. van Moorselaar and colleagues (2021) suggested that statistical learning enhances intertrial priming effects. Although our findings tend to support the idea that feature gain control and intertrial priming operate at different processing stages, with feature gain control acting before priority computations and intertrial priming acting after priority computations, it remains unclear how distractor feature learning and intertrial priming are linked, and how they combine in the observed ERP pattern. Further research will shed more light on this issue.

Our findings also fit into the predictive processing framework proposing that the brain constantly makes predictions about incoming sensory input to interact with its environment, because it has no direct access to the external world (Slagter & van Moorselaar, 2021; van Moorselaar & Slagter, 2019, 2020; Friston, 2009). In this framework, distractor interference can be considered a consequence of these predictions, as salient distractors produce strong sensory input and large prediction errors. With statistical learning, however, precision expectations can develop, allowing the high-probability distractor signals to be downregulated (Slagter & van Moorselaar, 2021). Hence, distractor interference is reduced when distractor features such as a likely distractor color can be predicted. This explains the observed reduction in RTs and the smaller P_D we observed with a high-probability distractor color.

In our learning task, the color distractor was the only colored item among gray stimuli. Thus, an alternative strategy to reduce distractor interference would be to down-weight the distracting color dimension, making learning at the featural level unnecessary. According to the dimension-weighting account (DWA; Liesefeld et al., 2024; Liesefeld & Müller, 2019, 2021), our visual system is structured hierarchically. The DWA suggests that in a search task, stimulus features (e.g., red, green, blue) are extracted and represented as saliency signals on feature salience maps. These feature salience maps are subsequently combined into dimensional maps (e.g., color), which can be weighted before being integrated into an overall priority map. This theory does not preclude weighting at the featural level but predominantly emphasizes weighting at the dimensional level. Thus, in our study, a possible strategy would have been to down-weight the color dimension to complete the search for the shape target without interference from the color distractor, which would have rendered distractor feature learning unnecessary in our experiment. However, it might be that the different colors in our experiment might not actually represent different features but at least in part also different dimensions, as color is considered to be multidimensional (Liesefeld & Müller, 2019, 2021; Liesefeld, Liesefeld, & Müller, 2019; Liesefeld, Liesefeld, Pollmann, & Müller, 2019). Hence, our study might provide evidence for dimension-specific suppression rather than feature-specific suppression, and future research needs to investigate whether distractor feature learning can be observed when the distractor is defined within a dimension whose substructures are somewhat clear (e.g., orientation; Liesefeld, Liesefeld, & Müller, 2019).

Lastly, we examined whether distractor feature learning in a visual search task influences later cognitive processes such as VWM performance. Surprisingly, there was no difference in VWM performance between high- and low-probability colors, suggesting that VWM performance was not modulated by distractor feature learning. However, this conclusion should be treated cautiously, as the null effect could be attributed to the possibility that the current design was not sensitive enough to detect changes. For instance, previous studies used a longer duration of the memory array (e.g., 5000 msec) combined with eye-tracking (Lancry-Dayan, Ben-Shakhar, & Pertzov, 2023; Lancry-Dayan, Nahari, Ben-Shakhar, & Pertzov, 2021). They found that newly learned familiar items gained more gaze fixation than unfamiliar items in the first hundreds of milliseconds after the memory array appeared, but this pattern was quickly reversed after 1000 msec. Returning to our study, the short durations of the memory array (250 msec) and the fixed delay period (900 msec) may lack the temporal resolution needed to detect a potential difference between the distractor colors. In addition, it might be that the learning task and the change-detection task were too different, and the change in context between the two tasks prevented an effect of distractor feature learning on VWM performance (see also Britton & Anderson, 2020). Alternatively, learning of distractor regularities, which has been found to be unaffected by VWM load (Bogaerts, van Moorselaar, & Theeuwes, 2022; Gao & Theeuwes, 2020a, 2020b), might be independent of VWM processes. Thus, a template of the distractor feature held in VWM (Carlisle, 2023; Arita, Carlisle, & Woodman, 2012) might be needed to modulate VWM performance.

In conclusion, the current study provides additional evidence that distractor feature learning can be used to reduce interference from a salient distractor. The P_D results suggest that at the neural level, less suppression is required when the distractor is more likely to appear in a particular color. Our results indicate that the visual system reduces the interference of a salient distractor via distractor feature learning and, at the same time, makes use of intertrial priming benefits. In addition, it seems that the learned distractor features are not maintained in VWM, indicating that the brain adapts flexibly to changes in the visual context.

The authors would like to thank Annegret Biechele and Lea Marie Schmitt for assistance with data collection.

Corresponding author: Sizhu Han, Fachbereich Psychologie, AG Allgemeine und Biologische Psychologie, Philipps-Universität Marburg, Gutenbergstraße 18, Marburg, 35032, Germany, e-mail: [email protected].

The data used for statistical analyses are openly available on Open Science Framework (OSF): https://osf.io/kbw5h/.

Aylin A. Hanne: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Validation; Visualization; Writing—Original draft; Writing—Review & editing. Sizhu Han: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Validation; Visualization; Writing—Review & editing. Anna Schubö: Conceptualization; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Writing—Review & editing.

This research was supported by the Deutsche Forschungsgemeinschaft (German Research Foundation; https://dx.doi.org/10.13039/501100001659): SFB/TRR 135, TP B3 (grant number: 222641018), and GRK 2271, Project 9 (grant number: 290878970).

Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance.