The Results section is organized into two parts: The first reports the full manual-RT analysis, which reveals a certain, strikingly altered characteristic of ASD individuals’ response behavior, but offers no explanation of how this alteration is brought about. The second part presents an analysis of the eye-movement data, with particular focus on the oculomotor dynamics in the critical, altered condition, effectively testing, and deciding between, alternative accounts of the underlying cause of altered perceptual decision-making in individuals with ASD.
All RT analyses were performed on individuals’ median RTs after excluding trials on which a participant made an incorrect response (approximately 3% of trials on average).
Figure 2 depicts the RTs, and error rates, for the different distractor conditions (distractor absent, distractor in rare region, distractor in frequent region), separately for the ASD and the TD group. Numerically, RTs were slower in the ASD than in the TD group (distractor-present trials: 1173 vs. 1076 ms; distractor-absent trials: 1108 vs. 990 ms). However, a mixed-design ANOVA with the factors distractor condition and group failed to reveal this difference to be significant [main effect of group: F(1, 42) = 0.74, p = 0.39, ηp2 = 0.017, BFincl = 0.67]. Importantly, the main effect of distractor condition was significant, F(2, 42) = 101.53, p < 0.001, ηp2 = 0.71, BFincl > 1000, but not the interaction with group, F(1,42) = 1.17, p = 0.32, ηp2 = 0.03, BFincl = 0.37.
To break down the distractor-condition effect, we analyzed (1) the overall amount of distractor interference, defined as the RT difference between distractor-present trials (averaged with equal weight for trials with a distractor in the rare and, respectively, the frequent region) and distractor-absent trials, and (2) the distractor-location (probability-cueing) effect, defined as the RT difference between trials with a distractor in the rare and, respectively, the frequent region (Sauter et al. 2018; Zhang et al. 2019). Distractor interference was somewhat lower in the ASD group (124 vs. 156 ms in the TD group), but the difference was not significant [Welch two-sample t test: t(38.01) = − 1.49, p = 0.14, BF10 = 0.72]. Importantly, there was a significant probability-cueing effect [50 ms, t(43) = 4.31, p < 0.001, BF10 = 253], which did, however, not differ significantly between the ASD and TD groups [47 vs. 53 ms; Welch two-sample t test: t(40.77) = − 0.23, p = 0.83, BF10 = 0.30]; in fact, the Bayesian analysis provides substantial evidence in favor of no difference. Thus, the individuals with ASD learnt to proactively suppress distractors in the frequent distractor region as effectively as TD individuals—at variance with the former being compromised in their ability to acquire the distractor-distribution ‘prior’ (i.e., with the hypo-prior account).
Figure 3 depicts the RTs, and average error rates, on distractor-absent trials, depending on whether the target appeared in the rare or the frequent distractor region, for the two participant groups (ASD and TD). For this analysis, trials on which the target appeared at the same location as a distractor on the previous trial were removed, to rule out confounding of the target position effect by carry-over of inhibition of the distractor location on the preceding trial (see coincidence effects depicted in Fig. 4). A mixed-design ANOVA with the factors target condition and group revealed RTs to be significantly slower for targets appearing in the frequent versus the rare distractor region [1055 vs. 1035 ms; F(1, 42) = 4.67, p = 0.036, ηp2 = 0.10, BFincl = 1.67]—replicating previous studies using this paradigm (Wang and Theeuwes 2018a, b; Zhang et al. 2019). However, this target-position effect did not differ significantly between the ASD and TD groups [25 vs. 15 ms; F(1, 42) = 0.28, p = 0.60, BFincl = 0.34]—indicating that both groups acquired the same strategy of generally suppressing any singleton targets and distractors in the frequent distractor region, in line with suppression operating at the level of the search-guiding priority map (see Liesefeld and Müller 2019, 2020; Sauter et al. 2018, 2019; 2020).
Carry-Over of Reactive Distractor Inhibition Across Trials
Given this, we went on to test the alternative hypothesis that individuals with ASD react more strongly to rare, as compared to frequent, distractor events—by examining carry-over of reactive inhibition placed on the distractor location on a given trial onto targets appearing at the same location on the subsequent trial. That is, we analyzed the distractorn–1–targetn position coincidence trials, focusing exclusively on distractor-absent trials n (to assess pure carry-over of inhibition uninfluenced by any dynamics set in motion by the presence of a distractor on trial n; see Zhang et al. 2019). Figure 4 presents the distractor–target coincidence effect—the difference in RTs between distractor-absent trials n on which the current target did versus did not appear at the location of the distractor on the preceding trial n – 1—separately for the two groups.
A mixed-design ANOVA with target condition (target n in frequent vs. rare distractor region) and group (ASD vs. TD) as factors revealed both main effects to be significant [target condition: F(1, 42) = 17.02, p < 0.001, ηp2 = 0.29, BFinc > 1000; group: F(1, 42) = 7.75, p = 0.008, ηp2 = 0.16, BFinc = 53]; importantly, the interaction also turned out significant, F(1, 42) = 9.53, p = 0.004, ηp2 = 0.19, BFinc = 76. The distractor–target coincidence effect (i.e., the carried-over inhibition) was substantial and significant only for targets in the rare region [228-ms effect, t(43) = 4.1, p < 0.001, BF10 = 124; frequent region: 14-ms effect, t(43) = 0.96, p = 0.34, BF10 = 0.25]. Further, it was equally small (and non-significant) for both groups if the current target coincided with a preceding distractor in the frequent region (ASD: 11-ms effect; TD: 17-ms effect); by contrast, for targets following a distractor in the rare region, the effect was more than five times larger for ASD than for TD individuals (384 vs. 71 ms, Welch two sample t test: t(33.2) = 3.05, p = 0.004, BF10 = 10). The negligible coincidence effect for targets in the frequent region is as expected: this region is proactively (tonically) suppressed, curtailing any additional re-active (phasic) inhibition in response to an actual distractor occurring there (Zhang et al. 2019). Similarly, as the rare distractor region is not tonically suppressed, the increased coincidence effect for this region is also as expected. However, the size of the group difference is striking: individuals with ASD display what looks like a ‘qualitative’ increase in reactive inhibition to unexpected distractors. This effect is not a chance finding: we still find a trace of it on the next trial,Footnote 2 when the spatially coincident distractor (with the current target) occurred two trials back (distractorn–2–targetn): here, a significant 159-ms effect for individuals with ASD [t(19) = 2.1, p = 0.046, BF10 = 1.5] compares with a 35-ms effect for TD individuals.
Thus, the only, and striking, difference between the ASD and TD groups is that, in ASD, the coincidence effect was considerably larger, by a factor of at least five, when the target appeared at a previous distractor location, but only if the target and the previous distractor appeared in the rare distractor region. There are two possible explanations for this pattern: (1) Stronger reactive inhibition placed on—or ‘inhibition of return’ (IOR, Klein 2000; Posner and Cohen 1984) to—a previous distractor location in the rare region, which would be consistent with reports of stronger IOR in individuals with ASD in a Posner-type cueing paradigm (Rinehart et al. 2008). Or (2), a reactively strengthened belief that any salience signal (indicative of the presence of a singleton item) arising at the same location as the distractor in the rare region on the previous trial must actually be caused by a distractor, rather than a target. From the perspective of a drift–diffusion model (e.g., Ratcliff et al. 2016), there may be a reactively strengthened decision bias towards identifying the item at such a location as a ‘distractor’, and against identifying it as a response-relevant ‘target’—involving, say, a shift of the starting point of the evidence accumulation process closer to a ‘distractor’ decision and away from a ‘target’ decision. Although appearing similar to account (1), account (2) is subtly different: it is not the allocation of attention to the previous distractor location in the rare region as such that is altered (i.e., more strongly inhibited), but the amount of evidence that needs to be accumulated by focal-attentional processing of the item at this location to arrive at a target decision.
Deciding between these alternatives is not possible based on the RT results alone, as RTs represent only the end of a whole chain of processes leading up to the final response. However, we can gain insight into this chain by analyzing the patterns of eye movements on a given (type of) trial. In what follows, we summarize the main findings of the eye-movement analysis, following prior eye-movement studies of attentional capture (Di Caro et al. 2019; Geyer et al. 2008; Wang et al. 2019), though with a focus on the critical (ASD vs. TD differential) RT effects in the present study. Details of the eye-movement analyses and findings are presented in Supplementary.
Previous ‘attentional-capture’ studies of distractor-location probability-cueing effects using similar stimuli to the present study have found a reduced likelihood of oculomotor capture by distractors at frequent versus rare distractor locations, and potentially expedited disengagement of the eye from a distractor that had summoned a saccade at frequent versus rare locations. Also, on distractor-absent trials, saccade latencies to targets appearing at the frequent distractor location were delayed (Wang et al. 2019). Overall, this is consistent with the view that distractor-location probability learning involves proactive inhibition of the likely distractor locations, and, possibly, faster disengagement of attention, and the eye, from distractors at these locations. Accordingly, we first examined whether the ASD and TD groups would show similar effect patterns on distractor-present and -absent trials.
Distractor-Location Effects on Distractor-Present Trials
Both participant groups required overall fewer eye movements to acquire the response-critical target information with a distractor present in the frequent versus the rare distractor region [3.76 vs. 3.91 fixations, F(1, 37) = 10.11, p < 0.01; this effect did not significantly interact with group]. The main reason for this was that oculo-motor capture of the first saccade by the distractor was reduced [25 vs. 35%, F(1, 37) = 9.90, p < 0.01] and, in a trade-off, first eye movements going straight to the target were increased with a distractor in the frequent, versus one in the rare, region [28 vs. 21%; F(1, 37) = 38.01, p < 0.001],Footnote 3 for both groups. This pattern is consistent with learnt proactive inhibition of the frequent distractor region. Further, the first distractor fixations were marginally shorter on distractors that appeared in the frequent region, as compared to distractors in the rare region [203 vs. 213 ms: F(1, 36) = 3.82, p = 0.06]. This indicates that disengagement of attention from a distractor in the frequent region was expedited. Thus, replicating Wang et al.’s (2019) prior study with unimpaired young participants, both proactive inhibition of locations in the frequent distractor region, preventing oculo-motor capture by a distractor in the first instance, and, if capture prevention failed, expedited disengagement of the eye from a distractor in the frequent region contribute to the acquired (overall) RT probability-cueing effect. This is the case equally for the ASD and the TD group.
Distractorn−1–Targetn Spatial-Coincidence Effect on Distractor-Absent Trials
On coincidence trials, the percentage of first saccades directed straight to the target tended to be increased overall for targets following a distractor in the rare distractor region [an additional 9% as compared to non-coincident trials, vs. an additional 1% in the frequent region, F(1, 35) = 2.71, p = 0.11]. If anything, this effect was stronger in individuals with ASD (an additional 13% in the rare region vs. 1% in the frequent region); that is, compared to TD individuals (an additional 5% in the rare vs. 2% in the frequent region), a greater percentage of their first saccades went directly to a target located in the rare region [rare region, coincident vs. non-coincident trials: t(19) = 2.07, p = 0.052].Footnote 4 The increased proportion of first saccades directed to the target effectively rules out increased cross-trial IOR of the preceding distractor location as an explanation for why ASD individuals took so long to respond to a target that followed a distractor at the exact-same (‘coincident’) location in the rare region. Increased IOR would have greatly reduced the ‘priority’ of the rare distractor location, decreasing the likelihood of attention, and the eye, returning there on trial on trial n. However, if anything, a target at this location was more, rather than less, likely to attract the first saccade, which is inconsistent with a—due to cross-trial IOR—lowered attention-attracting power of this item in ASD versus TD individuals.
However, while individuals with ASD were somewhat more likely to direct their first saccade to the target in the rare distractor region, the average number of fixations they took to make a response decision on coincidence, relative to non-coincidence, trials tended to be increased when the target appeared in the rare region [3.93 vs. 3.29 fixations, t(19) = 1.89, p = 0.07], but not for targets in the frequent region [3.42 vs 3.35 fixations, t(19) = 0.70, p = 0.49]; for TD individuals, by contrast, the number of fixations (on coincidence vs. non-coincidence trials) was comparable with targets in the rare region [3.31 vs. 3.33 fixations, t(16) = − 0.08, p = 0.93], as well as targets in the frequent region [3.41 vs. 3.44 fixations, t(16) = − 0.36, p = 0.72]. Thus, individuals with ASD required more eye movements to arrive at a response decision when the target followed a distractor at the ‘coincident’ location in the rare region, accounting for their slowed RTs on such coincidence trials. This is the case even though their first saccade was, if anything, more likely to be directed straight to the target.
Given that the first eye movement cannot account for the increased ‘coincidence’ effects with spatially coincident distractors and targets in the rare region, we conducted a more detailed analysis of the extended oculomotor scanning patterns on such trials to uncover why individuals with ASD require more eye movements. In particular, we partitioned participants’ scanning behavior on all distractor-absent trials into three categories based on the ‘target fixation pattern’ (treating multiple fixations in a row on the target as a single target fixation, in each category). Category 1 consisted of trials on which the target was fixated a single time before response, henceforth referred to as ‘single final target fixation’ trials; category 2: trials on which the target was fixated a single time, followed by multiple fixations on other items before response, ‘single non-final target fixation’ trials; and category 3: trials on which the target was fixated more than once, with at least one fixation of a non-target item in between, ‘target re-fixation’ trials.
Collapsed across coincident and non-coincident trials, the total dwell time spent fixating the target was longest in the ‘single final’ condition for both groups, and similarly short in the ‘single non-final target fixation’ and ‘target re-fixation’ conditions, F(2, 62) = 23.0, p < 0.001. At the same time, the behavioral RT was shortest in the ‘single final’ condition, and longest in the ‘target re-fixation’ condition, with the ‘single non-final’ condition in between [F(2, 62) = 123.9, p < 0.001]. The slow RTs in the ‘target re-fixation’ condition and the relatively slow RTs in the ‘single non-final target fixation’ condition likely reflect the fact that, on the first visit to the target, the target was mis-identified as a non-target/distractor, as a result of which oculomotor inspection moved elsewhere before either returning to the target or its vicinity to extract the response-relevant information (here, the orientation of the bar inside the target shape). Zhaoping and Guyader (2007) described a similar oculomotor pattern in a study of saliency-driven attention allocation in standard visual search in the absence of salient distractors: even though the target, as the most salient item, attracted the first eye movement, oculomotor scanning went on to other, non-target items before eventually returning to the target and responding.
Looking at the RTs as a function of the fixation pattern with a target in the rare distractor region, and comparing the critical coincident versus non-coincident trials (Fig. 5), RTs do not differ much between coincident and non-coincident trials in either the ‘single final’ (1038 vs 952 ms) or the ‘target re-fixation’ condition (1847 vs 1797 ms); they only differ in the ‘single non-final target fixation’ condition, where there is a substantial cost in responding to coincident, relative to non-coincident, targets (1915 vs 1352 ms). The reason for this may be that after sampling the target and mis-identifying it as a distractor or non-target, it took longer to accumulate information about the response-critical target feature without returning to the target (i.e., from eccentric vision). Since both groups show essentially the same pattern, this cannot explain why, in the ASD group, responses were particularly slow to coincident targets in the rare distractor region.
Accordingly, since the reaction- and dwell-time measures did not differ between the two participant groups, the explanation for this differential pattern must lie in the frequencies with which ASD and, respectively, TD individuals produced a particular eye-movement pattern. Indeed, examining the proportions of distractor-absent trials on which the target appeared in the rare distractor region, with the different target-fixation patterns (see Fig. 6) revealed the frequencies of these patterns to differ between coincident and non-coincident trials for the two groups: Both groups showed near-equivalent frequencies of the three target-fixation patterns on non-coincident trials (‘single-final target fixations’ being most frequent, at 77%, and ‘single non-final’ and ‘target re-fixation’ patterns being relatively infrequent, at 8 and 15%, respectively). However, in the ASD group, the percentage of trials with the ‘single final’ pattern was reduced on coincident, as compared to non-coincident, trials and those with the ‘single non-final’ and, most markedly, the ‘target re-fixation’ patterns were increased (51% ‘single final’, 15% ‘single non-final’, and 33% ‘target re-fixation’). No such change in the frequencies of target-fixation patterns (for coincident vs. non-coincident trials) was evident in the TD group (79% ‘single final’, 10% ‘single non-final’, and 10% ‘target re-fixation’). Chi-square tests revealed the distributions of target-fixation patterns to differ significantly between coincident and non-coincident trials in the ASD group [Χ2(2, N = 1525) = 16.6, p < 0.001], but not the TD group [Χ2(2, N = 1607) = 0.44, p = 0.80].
This effect pattern suggests that the main reason for why individuals with ASD took so long to make a response to a target (on trial n) appearing at the previous (trial n − 1) distractor location in the rare region is that, although their eyes were initially attracted to the singleton target as efficiently as in TD individuals, they then moved away from the target before eventually returning there (‘target re-fixation’ trials) or its vicinity (‘single non-final target fixation’ trials) and making the response decision. This is inconsistent with IOR-based accounts, whether they assume cross-trial IOR to be increased for the distractor location on the preceding trial n − 1 or within-trial IOR for the location first inspected on the current trial n. At variance with increased cross-trial IOR for the preceding distractor location, the (trial n) target at this location did attract the eye as rapidly in ASD as in TD individuals; and at odds with increased within-trial IOR, ASD individuals returned nearly as readily as TD individuals to the ‘coincident’ target location after having inspected it first (i.e., on “target re-fixation” trials).
Thus, instead, this pattern is likely to reflect a carried-over (from the preceding trial) bias against making a ‘target’ decision to the item occupying the location of the preceding distractor in the rare region. That is, for individuals with ASD, having detected a distractor at this location on trial n − 1 disproportionately strengthens their prior belief that this location contains a distractor on the subsequent trial n. As a result, when first attending to/fixating the ‘coincident’ target location on trial n, they tend to mis-identify it as a distractor. Consequently, their search is redirected elsewhere and returns only later, after not having found the target at other display locations, to the ‘coincident’ target position.