Serial dependence in orientation is weak at the perceptual stage but intact at the response stage in autistic adults

Abstract

Recent studies have suggested that autistic perception can be attributed to atypical Bayesian inference; however, it remains unclear whether the atypical Bayesian inference originates in the perceptual or post-perceptual stage or both. This study examined serial dependence in orientation at the perceptual and response stages in autistic and neurotypical adult groups. Participants comprised 17 autistic and 23 neurotypical adults. They reproduced the orientation of a Gabor stimulus in every odd trial or its mirror in every even trial. In the similar-stimulus session, a right-tilted Gabor stimulus was always presented; hence, serial dependence at the perceptual stage was presumed to occur because the perceived orientation was similar throughout the session. In the similar-response session, right- and left-tilted Gabor patches were alternately presented; thus serial dependence was presumed to occur because the response orientations were similar. Significant serial dependence was observed only in neurotypical adults for the similar-stimulus session, whereas it was observed in both groups for the similar-response session. Moreover, no significant correlation was observed between serial dependence and sensory profile. These findings suggest that autistic individuals possess atypical Bayesian inference at the perceptual stage and that sensory experiences in their daily lives are not attributable only to atypical Bayesian inference.

Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5; American Psychiatric Association, 2013) as atypicality in the social domain (social communication and interaction) and nonsocial domain (repetitive patterns of behavior, interests, or activities, and hyperreactivity or hyporeactivity to sensory input). In particular, atypical sensory perception has recently attracted attention because it was included as a core autistic trait for the first time in the DSM-5. Extensive research has shown that autistic children and adults experience atypical perception in multiple sensory domains and adopt various behaviors to cope with it in their daily lives (Leekam, Nieto, Libby, Wing, & Gould, 2007; Tomchek & Dunn, 2007). Nevertheless, there is still a lack of studies demonstrating which component of sensory processing is atypical in autistic individuals compared to neurotypical individuals. 

Over the last decade, researchers have focused on Bayesian inference to account for the differences in sensory processing between autistic and neurotypical individuals. Bayesian inference, also known as predictive coding, is a computational scheme for perceptual inference based on prior sensory experience (Aitchison & Lengyel, 2017; Friston, 2010; Rao, 1999; Rao & Ballard, 1999). It is theorized that perception (posteriors) is based on the integration of sensory input (likelihood) and belief of the input (priors). The extent to which perception is influenced by sensory input or prior belief depends on precision. Higher precision of the sensory input relative to the prior belief will result in veridical perception, whereas lower precision of the sensory input will bias perception toward the prior belief. After the integration of the sensory input and prior belief, the discrepancy between the two, known as the prediction error, is carried to higher brain areas for generating better beliefs. This scheme is implemented at multiple levels of the perceptual hierarchical mechanism (Rao & Ballard, 1999). 

Pellicano and Burr (2012) first suggested that autistic individuals possess atypical Bayesian inferences in perception. They suggested that the low precision of prior beliefs (hypo-priors) was responsible for the atypical perception of autistic individuals, leading to a tendency to perceive the world more veridically rather than being biased toward prior experience. Subsequently, several studies proposed various accounts that differ from the hypo-priors account, such as the high precision of sensory inputs (Brock, 2012), overfitting of predictions (Van de Cruys, de-Wit, Evers, Boets, & Wagemans, 2013; Van de Cruys et al., 2014), and an imbalance of precision between sensory inputs and prior beliefs (Lawson, Rees, & Friston, 2014). Although discussion about the most plausible account is beyond the scope of this study, these studies suggest that atypical Bayesian inference in sensory processing in ASD results in autistic perception that is biased toward sensory inputs. The atypical Bayesian inference theory is attractive to many researchers because it can also account for the cognitive, behavioral, and social characteristics of ASD (Palmer, Lawson, & Hohwy, 2017), yet empirical studies are still lacking. 

Some studies have shown that autistic participants’ responses are insusceptible to prior experiences using perceptual tasks such as a time interval reproduction task (Karaminis et al., 2016), binary classification of images as either faces or houses (Lawson, Mathys, & Rees, 2017), and a serial two-tone frequency discrimination task (Lieder et al., 2019). These findings can be interpreted to mean that autistic individuals veridically perceive the external world independently of prior perceptual experiences; that is, Bayesian inference at a perceptual stage is atypical in ASD. However, performing perceptual tasks also involves post-perceptual processing stages, such as categorical judgments and motor responses. Therefore these findings can also be interpreted to mean that autistic individuals accurately perform judgments or responses in perceptual tasks independently of prior performance. In other words, Bayesian inference at the post-perceptual stage appears to be atypical in ASD. 

To clarify these two possibilities, we focused on serial dependence, a behavioral bias in which visual decisions in the present moment are distorted by the history of stimuli seen in the recent past. Serial dependence is observed in various visual features such as shape (Manassi, Kristjánsson, & Whitney, 2019), numerosity (Corbett, Fischer, & Whitney, 2011), emotional expression (Liberman, Manassi, & Whitney, 2018), spatial position (Manassi, Liberman, Kosovicheva, Zhang, & Whitney, 2018), and orientation (Fischer & Whitney, 2014). A previous study demonstrated that serial dependence complies fully with the predictions of the Bayesian inference model (Cicchini, Mikellidou, & Burr, 2018). Serial dependence in orientation is adequate for examining two-stage Bayesian inferences because many researchers have accumulated evidence that it originates at both the perceptual (Cicchini, Mikellidou, & Burr, 2017; Fischer & Whitney, 2014; St. John-Saaltink, Kok, Lau, & de Lange, 2016) and post-perceptual stages (Ceylan, Herzog, & Pascucci, 2021; Fritsche, Mostert, & de Lange, 2017; Pascucci et al., 2019). 

Cicchini et al. (2017) used a unique paradigm in their second experiment, dividing it into two stages. This paradigm is based on the assumption that serial dependence in orientation occurs only if the current orientation is similar to the previous one (Fischer & Whitney, 2014; Fritsche et al., 2017). They separated serial dependence into perceptual and response stages by controlling the stimulus orientation and instructing participants alternately to perform direct or mirror reproduction of a perceived orientation in the adjustment method. If all stimuli were extracted around the diagonal orientation, the perceived orientations were similar, whereas the response orientations were dissimilar in every trial, then serial dependence at the perceptual stage was presumed to occur. If stimuli were extracted around the diagonal or antidiagonal orientation in every odd or even trial and the response orientations were similar in every trial, whereas perceived orientations were dissimilar, then serial dependence at the response stage was presumed to occur. 

The aim of this study was to clarify whether atypical Bayesian inference in ASD originates in the perceptual or post-perceptual stage or both. Therefore we examined serial dependence in orientation in autistic and neurotypical individuals using the paradigm of Cicchini et al. (2017), which enabled us to divide it into the perceptual and response stages. In addition, we assessed participants’ sensory profiles using the Adult/Adolescent Sensory Profile (AASP; Brown, Tollefson, Dunn, Cromwell, & Filion, 2001) and examined the association between these two-stage serial dependences and the AASP to clarify whether atypical Bayesian inference in ASD is related to sensory experiences in their daily lives. 

Methods

Participants

Autistic participants were recruited by making contact via a mailing list of an autistic social community. Following the past recruitments for the other experiments from the autistic community, the target number of autistic participants was set to approximately 20 during recruitment. This number was not based on an a priori sample size calculation. Consequently, 21 Asian autistic adults were recruited by making contact via the community mailing list. The second author confirmed the diagnosis of ASD using the Autism Diagnostic Observation Schedule-2 (ADOS-2; Lord et al., 2012). Nineteen met the cutoff criteria for diagnosis based on the ADOS-2. One was excluded because of advanced age (>70 years). One was excluded because of too many erroneous responses (39% of the responses deviated by more than 90° from the presented orientation). The final sample comprised 17 autistic adults (35% female and 65% male) with a mean age of 38.3 years (SD = 8.7 years; range, 23–53 years). As for neurotypical participants, 24 Asian adults were recruited by a Japanese participant recruitment company (Agekke Inc., Tokyo, Japan). They responded to screening questionnaires indicating that they had no diagnoses of neurodevelopmental disorders, mental disorders, intellectual disabilities, or physical disabilities. One was excluded because of failure to meet the cutoff criteria for the Autism spectrum quotient (AQ) score. The final sample comprised 23 adults (48% female and 52% male), with a mean age of 37.9 years (SD = 9.6 years; range, 21–56 years). There were no significant differences in age, sex, or intelligence quotient (IQ) between the autistic and neurotypical groups (Table 1). Socioeconomic status was not recorded. This study was approved by the local ethics committees of the University of Tokyo (approval number: 19-249) and Teikyo University (approval number: 507, 575) and performed in accordance with the ethical standards of the Declaration of Helsinki. Written informed consent was obtained from each participant. 

Table 1.

 

Demographic variables and statistics for comparison of the two groups comparison. Notes: AASP = Adult/Adolescent Sensory Profile, ADOS-2 = Autism Diagnostic Observation Schedule-2, AQ = Autism Spectrum Quotient, IQ = intelligence quotient.

 

Instruments

Autism diagnostic observation schedule, second edition

The ADOS–2 is a standardized, semistructured observational assessment used to evaluate autistic characteristics such as communication, social interaction, play, and restricted and repetitive behaviors (Lord et al., 2012). Module 4 (for adolescents and adults with fluent speech) was used in the present study. Only the autistic group received the ADOS–2 to confirm the diagnosis of ASD. The cutoff score was >7. 

Autism spectrum quotient

The AQ is a 50-item, self-administered questionnaire developed by Baron-Cohen, Wheelwright, Skinner, Martin, and Clubley (2001) to assess the degree to which an adult with normal intelligence has traits associated with ASD. The Japanese version (Wakabayashi, Baron-Cohen, Wheelwright, & Tojo, 2006) was used in this experiment. The total score was used as a screening tool in neurotypical adults. The cutoff score was <33. The AQ can be subdivided into five subscales: social skill, attention switching, local detail, communication, and imagination. 

Adult/Adolescent sensory profile

The AASP is a 60-item self-administered questionnaire that assesses the everyday sensory experiences of adults and adolescents in four sensory quadrants: low registration, sensation seeking, sensory sensitivity, and sensation avoiding (Brown et al., 2001). Low registration reflects a disregard for or a slow response to sensation because of high neurological thresholds. Sensation seeking reflects counteractive responses to high neurological thresholds and encompasses pleasure derived from rich sensory environments and behaviors that create sensations. Sensory sensitivity represents distractibility, difficulty screening stimuli, and discomfort with sensation owing to low neurological thresholds. Sensation avoiding includes behaviors that limit exposure to stimuli as counteractive responses to low neurological thresholds. 

Stimuli and apparatus

The participants sat at a desk in a dimly lit room and fixed their head on a chin rest 85 cm from a 22-inch CRT display (Mitsubishi Diamondtron Flat RDF22PII; Mitsubishi Electric, Tokyo, Japan) at a refresh rate of 75 Hz. Stimuli were presented on a computer (Mac Pro ME253J OS X 10.11.6; Apple, Inc., Cupertino, CA, USA) with MATLAB R2019a (MathWorks) using the Psychophysics Toolbox extensions (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997). A fixation dot (0.5°) was presented at the center of the display on a black background. A Gabor stimulus (spatial frequency of 0.33 cycles per degree [cpd], peak contrast of 25% Michelson, 1.5° SD Gaussian contrast envelope, 500 ms) was presented either rightward or leftward of fixation (6° horizontal, 4° vertical eccentricity). A mask stimulus (random noise filtered at 0.3 cpd and windowed in 1.5° SD Gaussian contrast envelope, 1000 ms) was presented at the same location as the Gabor stimulus to minimize negative aftereffects. A response bar (0.6° wide white bar windowed in 1.5° SD Gaussian contrast envelope) was presented for the adjustment methods. 

Procedure

The experimental paradigm, illustrated in Figure 1, was a close replication of the second experiment by Cicchini et al. (2017). In each trial, the Gabor stimulus was presented for 500 ms, followed by a masking stimulus for 1000 ms. Participants were instructed to reproduce either the orientation of the Gabor stimulus in every odd trial or its mirror (flipped about the vertical axis) in every even trial by pressing the left or right keys and setting the orientation of the response bar. The color of the fixation dot (yellow or magenta) informed the participants whether they had to perform a direct or mirror reproduction. 

Figure 1.

 

Schematic illustration of procedure.

The stimulus orientation was manipulated to create two conditions, run in different sessions. In the similar-stimulus session, a right-tilted Gabor stimulus (from 30° to 60° in steps of 3°) was always presented. The response bar was presented in different hemifields in odd and even trials to minimize the unexpected effects of the last reproduction on the current one. Serial dependence at the perceptual stage was presumed to occur in this condition because the perceived orientations were similar, whereas response orientations were dissimilar. In the similar-response session, in odd trials a right-tilted Gabor stimulus was presented (from 30° to 60° in steps of 3°) and in even trials a left-tilted one was presented (from −60° to −30° in steps of 3°). The response bar was presented constantly at 6° below fixation to maximize the response history effects. Serial dependence at the response stage was presumed to occur under this condition because the response orientations were similar, whereas the perceived orientations were dissimilar. 

Each participant completed two sessions, each comprising 154 trials. The session order was counterbalanced between the participants. The locations of the stimuli (right or left) were fixed for each participant and counterbalanced between them. It took approximately 25 minutes to complete one session. A 10-minute break was allowed between the two sessions. An 11-trial practice session was conducted before each session. Assessments and questionnaires were administered before or after the experiment. 

Data analysis

The trials were subdivided according to the current orientation. A linear regression of the responses to the current orientation on the previous orientations was fitted to the subdivided trials. Slope values for each current orientation were averaged to obtain an index of dependence on the previous trial, referred to as “serial effect.” A higher serial effect score indicates a larger serial dependence. Before fitting, trials in which the orientation difference between the current and previous trials exceeded 18° were excluded (16.5% of all data) following Cicchini et al. (2017) because serial dependence in orientation occurs only if the current orientation is similar to the previous one (Fischer & Whitney, 2014; Fritsche et al., 2017). Trials in which the response deviated more than 90° from the current orientation were also excluded (1.4% of all data) because these trials were regarded as mistaken mirror responses in odd trials or mistaken non-mirror responses in even trials. Trials in which the response deviated significantly from the subdivided trials were also excluded as outliers (1.2% of all data), following the Grubbs’ test (Grubbs, 1950). In the similar-stimulus session, the participants were required to make a mirror response in every even trial, whereas the perceived orientations were similar throughout the session. To retrieve the sequence of the perceived orientations, the responses in every even trial were reversed in the analysis. In the similar-response session, the stimulus orientation was right- or left-tilted in every odd or even trial, whereas the requested responses were similar throughout the session. To retrieve the sequence of requested responses, the stimulus orientations in every even trial were reversed in the analysis. 

In addition, the SD of the response orientations was calculated as an index of the reciprocal of the precision of orientation perception, referred to as “response variability.” SD values obtained in each of the subdivided trials were averaged thereafter. A higher response variability score represented a lower precision of orientation perception. Unlike obtaining the serial effect, trials in which the orientation difference between the current and previous trials exceeded 18° and trials in which the response deviated significantly were included in the SD calculation. Trials in which the response deviated more than 90° from the current orientation were excluded (1.4% of all data). 

Because the two sessions (similar-stimulus and similar-response sessions) were assumed to reflect two-level serial effects (perceptual and response levels), the serial effect and response variability for each session were defined as dependent variables and the group (autistic or neurotypical adult group) was defined as an independent variable. Therefore a two-sample Welch's t-test was conducted for serial effects and response variabilities to compare the two groups for each session. 

Results

Group comparison of instruments

Table 1 presents the statistics of responses to the instruments from the two groups. Significant differences were found for AQ and three subscale scores on the AASP. Unexpectedly, the autistic adult group showed a significantly lower sensation seeking score on the AASP than the neurotypical adult group. 

Group comparison of serial effect and response variability

Figure 2 shows the mean and 95% confidence interval (CI) of the serial effects for each session. For the similar-stimulus session, the mean serial effect was lower in the autistic adult group (M = 0.042) than in the neurotypical adult group (M = 0.142). As seen from the 95% CI, only the serial effect in the neurotypical adult group deviated significantly from zero. Welch's two-sample t-test showed a significant difference between the two groups (t_(30.91) = 2.45, p = 0.020, Hedges’ g = 0.79). These results indicated that serial dependence was observed only in the neurotypical adult group for the similar-stimulus session. By contrast, for the similar-response sessions, the mean serial effect in the autistic adult group (M = 0.096) was equivalent to that in the neurotypical adult group (M = 0.105). A significant deviation from zero in both groups and no significant difference between the two groups (Welch's two-sample t-test, t_(31.86) = 0.21, p = 0.834, Hedges’ g = 0.07) indicated that serial dependence was observed in both groups for the similar-response session. 

Figure 2.

 

Results of the serial effect for two sessions and two groups. Small dots represent individual scores. Bars and error bars represent means and 95% confidence intervals.

Figure 3 shows the mean and 95% CI of response variability for each session. There were no significant differences between the two groups in the similar-stimulus session (autistic adult group, M = 8.26; neurotypical adult group, M = 8.67; Welch's two-sample t-test: t_(32.58) = 0.58, p = 0.568, Hedges’ g = 0.18) and the similar-response session (autistic adult group, M = 7.64; neurotypical adult group, M = 8.27; Welch's two-sample t-test: t_(32.17) = 0.79, p = 0.433, Hedges’ g = 0.25). These results indicated that the precision of orientation perception was equivalent between the two groups, regardless of the session. 

Figure 3.

 

Results of the response variability for two sessions and two groups. Small dots represent individual scores. Bars and error bars represent means and 95% confidence intervals.

Correlation between serial effect and AASP

A correlation analysis between the serial effects of the two sessions and the four subscale scores in the AASP was conducted to clarify whether atypical Bayesian inference in ASD was related to sensory experiences occurring in the participants’ daily lives (Table 2). The serial effect for the similar-stimulus session was moderately but significantly correlated with the sensory sensitivity and sensation avoiding subscale scores, whereas the serial effect for the similar-response sessions was not correlated with any of the four subscale scores. However, the significant correlations in the similar-stimulus session seemed spurious because significant differences between the autistic and neurotypical adult groups were observed in both the serial effect for the similar-stimulus session and the subscale scores in the AASP. Therefore partial correlation coefficients adjusted for group differences were calculated, and no significant correlation was found for any of the variables (Table 2). These results indicate that serial dependence in orientation perception is irrelevant to sensory experiences in the participants’ daily lives. In addition, the correlation of the serial effects for the similar-stimulus and similar-response sessions was not significant, indicating that the serial dependence for the two sessions was independent. 

Table 2.

 

Pearson's correlation coefficients and partial correlation coefficients between two serial effects and AASP subscale scores. Notes: AASP = Adult/Adolescent Sensory Profile.

 

Correlation between serial effect and AQ

A correlation analysis between the serial effects of the two sessions and the total and subscale scores of the AQ was conducted to confirm whether the difference of the serial effect for the similar-stimulus session in the group comparison was also observed as a function of the self-administered scale of autistic traits, and to specify which components of autistic traits were related to the serial effects (Table 3). Similar to the group comparison results, the total score of the AQ was significantly correlated with the serial effect for the similar-stimulus session (r = −0.39, p = 0.012), but not for the similar-response session (r = −0.20, p = 0.225). As for the subscale scores, the imagination subscale score was significantly correlated with the serial effect for the similar-stimulus session (r = −0.35, p = 0.025), although it was comparable to the correlation coefficients for the other subscale scores (around −0.30). 

Table 3.

 

Pearson's correlation coefficients between two serial effects and AQ. Notes: AQ = Autism Spectrum Quotient.

 

Discussion

This study examined serial dependence in orientation at the perceptual and response stages in autistic and neurotypical adult groups to address whether atypical Bayesian inference in ASD originates in a perceptual or post-perceptual stage or both. In the similar-stimulus session, serial dependence at the perceptual stage was presumed to occur because the perceived orientations were similar while response orientations were dissimilar. In the similar-response session, serial dependence at the response stage was presumed to occur because response orientations were similar despite dissimilar perceived orientations. The group comparison results showed that serial dependence was observed only in the neurotypical adult group for the similar-stimulus session, whereas serial dependence was observed in both groups for the similar-response session. This finding was reinforced by the results of correlation between serial effect and the AQ, which showed that the total score of the AQ was significantly correlated with the serial effect for the similar-stimulus session but not for the similar-response session. 

The weak serial dependence in the autistic adult group for the similar-stimulus session suggests that Bayesian inference at a perceptual stage is atypical in ASD, which is biased toward sensory inputs. This finding reinforces previous findings that autistic perception is unsusceptible to prior experiences in perceptual tasks (Karaminis et al., 2016; Lawson et al., 2017; Lieder et al., 2019). In other words, these findings suggest that autistic individuals perceive the external world veridically independent of their prior perceptual experiences. One may suspect that the serial dependence for the similar-stimulus session does not occur at the perceptual stage, but rather at the working memory stage, given that participants store a presented Gabor patch at the working memory stage before flipping it about the vertical axis on every trial. In view of the previous finding of serial dependence in orientation that prior belief formed by high-level perceptual history biases current early-level sensory processing (Cicchini, Benedetto, & Burr, 2021), it seems reasonable that both the perceptual and working memory stages are involved in the serial dependence for the similar-stimulus session. It can thus be concluded that the serial dependence for the similar-stimulus session at least involves Bayesian inference at the perceptual stage. 

As for the similar-response sessions, the intact serial dependence was observed in the autistic adult group, which suggests that the weak serial dependence in ASD was not due to the difference in the response stage, such as motor control accuracy, decision criteria stringency, or motivation to deal with the adjustment task. This finding also suggests that not all Bayesian inference mechanisms are atypical in ASD. This is consistent with the results of a previous priming experiment, which reported that autistic traits were associated with the effect of history not on attention, measured by reaction time, but on perception, measured by pupil dilation (Pomè, Binda, Cicchini, & Burr, 2020). Although several studies have advocated the comprehensive attribution of the cognitive, behavioral, or social characteristics of ASD to atypical Bayesian inference (see Palmer et al., 2017, for review), careful empirical studies should be conducted to verify these theories. 

An interesting finding was that there was no correlation between serial effects for similar-stimulus and similar-response sessions. This result indicates that the processing stages involved in serial dependence are independent between the two sessions and supports the assumption that the serial dependence for the similar-stimulus and similar-response sessions reflects Bayesian inference at perceptual and post-perceptual stages, respectively. 

Another interesting result was the correlation analysis between the serial effect for the similar-stimulus session and the subscale scores of the AQ. We initially expected that the serial effect would not be correlated with social traits such as the social skill, communication, and imagination subscales but would be correlated with nonsocial traits such as the attention switching and local detail subscales. In fact, however, the result showed that these correlation coefficients were approximately equivalent, although only the imagination subscale score was significant. This finding suggests that Bayesian inference at the perceptual stage is not only involved with nonsocial traits but also uniformly involved with various autistic traits. 

One might suspect that the lack of serial dependence in the autistic adult group for the similar-stimulus session is attributed to the high precision of orientation perception for autistic individuals, considering previous studies that demonstrated that high precision of orientation brings about a low serial effect (Cicchini et al., 2018). This possibility is implausible because this study showed equivalent precision of orientation perception between autistic and neurotypical adult groups, as indicated by the response variability results. This finding differs from that of a previous study that demonstrated that autistic adults have superior orientation discrimination precision for oblique stimuli compared to neurotypical adults (Dickinson, Bruyns-Haylett, Smith, Jones, & Milne, 2016). Possible explanations for this might be differences in stimuli, such as spatial frequency (0.33 vs. 3 cpd), contrast (25% vs. 99%), and eccentricity (peripheral or fovea), or differences in psychophysical methods (adjustment or two-alternative forced choice). 

Interestingly, a significant serial dependence for the similar-response session was observed, regardless of group differences. This finding is contrary to that of a previous study, which found no serial dependence for similar response sessions (Cicchini et al., 2017). It seems possible that this discrepancy may be attributed to the small number of trials. Participants carried out 2000 trials per session in Cicchini et al. (2017), whereas only 154 trials were conducted in this study to ensure that autistic participants would not be exhausted. In Cicchini et al. (2017), repeated practice of the perceptual task through numerous trials appeared to improve performance and reduce response bias, thereby resulting in no serial dependence for the similar-response session. 

One unanticipated result was that no correlation was observed between the serial effects and the AASP on the condition that it was adjusted by group differences, which indicates no direct association between serial dependence in orientation perception and sensory experiences in the participants’ daily lives. These results suggest that the autistic sensory experiences are not attributable only to atypical Bayesian inferences in perception. Other factors may cause autistic sensory experiences, such as restricted and fixated attention, or traumatic events that link sensory experiences to negative emotions (Crasta, Green, Gavin, & Davies, 2023; Jeon & Bae, 2022). This finding is consistent with a previous report that laboratory-based sensory tasks are independent of sensory questionnaires (Dwyer, Takarae, Zadeh, Rivera, & Saron, 2022). We speculate that weak serial dependence in orientation for autistic individuals does not result in sensory experiences but rather results in motor sickness, fatigue, or motor disabilities because it causes constantly unstable orientation in the visual field. Further studies are needed to examine the relationship between weak serial dependence in orientation and the extensive problems that autistic individuals encounter in their daily lives. 

This study has some limitations that should be addressed in future research. The sample size was not calculated based on a power analysis but estimated based on the past recruitments for the other experiments from the autistic social community we contacted, because of the limited number of contactable autistic communities. Although the sample size for each group in this study (approximately 20) was larger than the original study (seven participants; Cicchini et al., 2017), the results of statistical tests, especially the correlation analysis, need to be interpreted with caution. The participants were limited to adults for ease of recruitment. A previous study reported that the atypicality of sensory profiles in autistic individuals changed across age groups and appeared to become more similar to neurotypical individuals over time (Kern et al., 2006); hence, the present experiment needs to be conducted with extensive age groups. This study addressed only visual perception, although recent research has reported that autistic individuals more frequently encounter problems in other modalities, such as audition and touch (Wada et al., 2023). In future studies, perceptual phenomena regarding Bayesian inference in various modalities should be used to better understand atypical Bayesian perception in ASD. This study demonstrated the atypicality of a simple Bayesian model in ASD yet could not tackle hierarchical models such as volatility (Lawson et al., 2017; Lawson et al., 2014). Future studies should address this issue by controlling for the stimulus uncertainty. 

Conclusions

This study compared serial dependence in orientation between autistic and neurotypical adult groups and demonstrated that autistic adults possess weak serial dependence at the perceptual stage and intact serial dependence at the response stage. These findings suggest that autistic individuals do not make judgments or respond independently of prior performance but veridically perceive the external world independently of prior perceptual experiences. Moreover, this study found no direct association between serial dependence in orientation perception and sensory experiences in daily life. This finding suggests that autistic sensory experiences are not attributable only to atypical Bayesian inferences. 

Acknowledgments

The authors thank Himiko Toyama for her help in the operation of the experiment. 

Supported by JST CREST “Cognitive Mirroring: Assisting people with developmental disorders by means of self-understanding and social sharing of cognitive processes” (Grant Number: JPMJCR16E2), Japan. 

Data availability: The datasets analyzed during this study are available on Open Science Framework (doi:10.17605/OSF.IO/8CG63). 

Commercial relationships: none. 

Corresponding author: Masaki Tsujita. 

Email: [email protected]. 

Address: Research Center for Advanced Science and Technology, The University of Tokyo, Tokyo, Japan.