Autistic and non-autistic adults exhibit comparable sequence learning
To test our hypothesis that we would observe reduced sequence learning in the autistic group, as indexed by a reduced RT difference between predictable and unpredictable conditions relative to the control group, we submitted RTs to a RM-ANOVA with within-subject factor condition (easy-predictable, difficult-predictable, and unpredictable) and between-subjects factor group (ASD, control). This revealed a significant main effect of condition (F(2,122) = 28.804, p < 0.001, η² = 0.028, BFincl = 1.862e+8) (Fig. 2a), no main effect of group (F(1,61) = 0.518, p = 0.474, η² = 0.008, BFincl = 0.423) and no interaction between condition and group (F(2,122) = 0.429, p = 0.652, η² = 0.000, BFincl = 0.182). Post-hoc Bonferroni-corrected t-tests demonstrated that RT was significantly lower for the predictable-easy (mean (standard error) = 660.350 (11.438)) compared to the predictable-difficult ( = 697.318 (12.062), t(61) = -6.707, p < 0.001, d = -0.845) and unpredictable conditions ( = 687.624 (11.316), t(61) = -5.156, p < 0.001, d = -0.65). Although RTs for the unpredictable and predictable-difficult conditions differed - with lower RTs for the unpredictable condition – this difference did not reach statistical significance (t(61) = 2.018, pbonf = 0.137). In sum, we observed lower RTs for the predictable-easy compared to the unpredictable condition, suggesting that sequence learning enabled participants to speed their responses. However, we observed no significant speeding for the predictable-difficult condition, thus raising the possibility that the sequence was too challenging for participants to learn. Finally, the lack of a main effect of group suggested that the groups do not differ with respect to sequence-learning. This result was strengthened by Bayesian independent t-tests, with the BF indicating moderate evidence for H0 for the easy (BF01 = 3.193) and anecdotal evidence for the difficult (BF01 = 2.371) predictable conditions (Figs. 3a-3b).
A RM-ANOVA with within-subject factor condition (easy-predictable, difficult-predictable and unpredictable) and between-subjects factor group (ASD, control) and IES as dependent variable revealed a main effect of condition (F(2,122) = 25.078, p <0.001, η² = 0.036, BFincl = 1.585e+7) and no main effect of group (F(1,61) = 0.064, p = 0.801, η² = 0.001, BFincl = 0.509) or group by condition interaction (F(2,122) = 0.377, p =0.687, η² = 0.001, BFincl = 0.137) (Fig. 2b). Post-hoc tests demonstrated that IES were significantly lower for the predictable-easy ( = 685.502 (12.222)) compared to both the unpredictable ( = 719.985 (12.938), t(61) = -5.357, p < 0.001) and the predictable-difficult condition ( = 734.790 (15.226), t(61) = -5.942, p <0.001). IES for the unpredictable compared to predictable difficult conditions did not significantly differ (t(61) = -2.155, p = 0.099, Cohen’s d = 0.271). However, a Bayesian paired t-test provided weak evidence that IES for the unpredictable blocks differed from the predictable difficult condition (BF01 = 0.625, Fig. 3c). Thus, RT, after correcting for number of errors, was higher during the predictable difficult, relative to the unpredictable, condition adding support for a lack of sequence learning during the difficult-predictable condition. The lack of a main effect of group, or interaction between group and condition suggests that the groups did not differ in the ability to execute the appropriate action. In sum, we did not find evidence to support the hypothesis that, relative to controls, autistic adults exhibited decreased sequence learning.
Autistic and non-autistic adults exhibit comparable surprise-related slowing
To test our second hypothesis that, relative to non-autistic controls, autistic participants would exhibit a reduction in surprise-related slowing - as indexed by a reduced RT difference between surprising and unsurprising trials - we submitted RTs to a RM-ANOVA with within-subject factors surprise (surprising trials, unsurprising trials), condition (easy-predictable, difficult-predictable) and between-subjects factor group (ASD, control). We observed main effects of surprise (F(1,61) = 34.144, p < 0.001, η² = 0.017, BFincl = 1.166e+13), condition (F(1,61) = 34.144, p < 0.001, η² = 0.017, BFincl = 4.526e+13) and a surprise by condition interaction (F(1,61) = 72.325, p < 0.001, η² = 0.022, BFincl = 3.714e+8). Post-hoc comparisons revealed an increase in RT for surprising compared to unsurprising trials for the easy-predictable (surprising: = 703.782 (11.744), unsurprising: = 649.278 (11.575), mean difference = 54.340; t(61)= 9.850, p < 0.001, d = 1.241) (Fig. 4a) but not the difficult-predictable condition (surprising: = 694.875 (12.011), unsurprising: = 697.835 (12.241), mean difference = -3.656; t(61) = -0.663, p = 1.000, d = -0.083) (Fig. 4b). Crucially, no main effect of group (F(1,61) = 0.493, p = 0.485, η² = 0.008, BFincl = 0.326), surprise by group (F(1,61) = 0.797, p = 0.375, η² = 0.000, BFincl = 0.267), condition by group (F(1,61) = 0.795, p = 0.980, η² = 0.000, BFincl = 0.199) or surprise by condition by group (F(1,61) = 0.493, p = 0.485, η² = 0.000, BFincl = 0.088) interactions were observed. To ensure that the lack of group difference could not be attributed to differences in baseline speed, we re-ran the analysis with baseline-corrected mean RT scores. This did not change the observed pattern of results, with no main/interaction effect(s) of group observed (all p-values > 0.05, all η² < 0.001, all BFincl < 1). Indeed, no differences in motor execution overall were observed between groups (Supplementary Results).
Surprise-related slowing scores for the two groups in the easy-predictable condition were compared using a Bayesian independent t-test. The BF01 was equal to 3.778, indicating the data were approximately 3.8 times more likely under the hypothesis that groups did not differ with respect to surprise-related slowing and providing moderate evidence for H0 (Fig. 3d).
Finally, although IES scores varied significantly across condition (F(1,61) = 16.538, p < 0.001, η² = 0.011, BFincl = 106.780), surprise (F(1,61) = 43.141, p < 0.001, η² = 0.031, BFincl = 1.905e+7) and condition by surprise (F(1,61) = 56.538, p < 0.001, η² = 0.026, BFincl = 2.649e+7), no main/interaction effect(s) of group were observed (all p-values > 0.05, all η² < 0.001, all BFincl < 1) (Suppl. Figs. 2a-2b).
Trial-by-trial surprise did not differ between groups
The observed results demonstrated typical surprise-related slowing in autistic individuals. However, the above analyses collapse data across all trials within each condition and cannot detect differences in the temporal progression of surprise-related slowing, nor reveal differences between the groups in the speed of acquisition of surprise-related slowing. Trial-by-trial surprise was therefore included as a predictor in a multiple regression analysis, alongside trial number and condition. Standardized beta values (β) for the main and interaction effect(s) of predictors (Table 2) were compared using one-sample t-tests to determine if they were significant predictors of RT. β values that significantly differed from zero were averaged across each group and compared using standard and Bayesian independent sample t-tests. If differences in the temporal progression of surprise-related slowing existed between groups, we would expect to observe a significant difference in β values relating to the interaction between surprise and trial number. However, no differences in β values were observed between groups for this interaction (t(61) = 1.130, p = 0.263, d = 0.287, BF01 = 2.260), nor for β values for condition (t(61) = -0.022, p = 0.983, d = -0.005, BF01 = 3.868), trial-by-trial surprise (t(61) = 0.905, p = 0.369, d = 0.229, BF01 = 2.739) or surprise by condition (t(61) = 1.191, p = 0.238, d = 0.302, BF01 = 2.130). In summary, no differences in the temporal evolution of surprise-related slowing were observed between groups.
ADOS severity scores as a predictor of surprise-related slowing and sequence-learning
Focusing specifically on the easy-predictable condition, where surprise-related slowing and sequence-learning effects were observed, correlation analysis showed that ADOS scores were not a significant predictor of surprise-related slowing (r = -0.324, F(1,26) = 3.042, p = 0.093) or sequence-learning (r = 0.070, F(1, 26) = 0.129, p = 0.723). Furthermore, neither AQ nor TAS scores were significant predictors of behavioural measures (Supplementary Results).