Examining the diametric model of autistic and psychotic traits through temporal perception Authors: Roy Ramati, Yaffa Yeshurun, Ahmad Abu-Akel Published: 2026 ABSTRACT Temporal visual processing deficits are well-documented in psychosis spectrum disorders and linked to core symptoms including hallucinations and cognitive disorganization. However, studying these mechanisms in isolation may overlook the influence of co-occurring autistic traits, which are increasingly recognized within psychosis populations. We hypothesized that psychosis-proneness and autistic traits would interactively affect temporal crowding. We investigated temporal crowding using an orientation-estimation task in neurotypical participants across two experiments (N1 = 81, N2 = 93). Participants viewed sequences of three randomly oriented stimuli at varying stimulus onset asynchronies (SOA: 200–400ms) and reported the middle item’s orientation. Mixture models examined performance in terms of encoding precision, guessing rates, and substitution errors. The Community Assessment of Psychic Experiences (CAPE) and Autism-Spectrum Quotient (AQ) measured trait expression. Response surface analysis (RSA) mapped trait interactions. RSA revealed that balanced expressions of both traits (particularly when both were elevated) were associated with significantly reduced temporal order errors. These findings align with the diametric model, which suggests compensatory interactions between psychosis-proneness and autistic traits. Our findings extend this model to mid-to-high level perceptual mechanisms, suggesting that opposing cognitive tendencies can create mutual compensation at the perceptual level, with implications for understanding symptom heterogeneity across both conditions.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-58526-7. Introduction Temporal visual perception—the ability to parse a dynamic visual input into discrete events and objects, assign them to the correct temporal order, and accurately encode the identity of these events with sufficient fidelity—is essential for constructing coherent experience and guiding adaptive behavior1–3. These demands on the perceptual system can be dissociated into (i) temporal sequence encoding (i.e., determining which item occurred when4; typically measured through temporal order judgment (TOJ) tasks at very short temporal intervals (< 150ms); and (ii) stimulus identity encoding (i.e., the precision in which stimulus features are maintained5;. Both processes are essential for constructing a coherent temporal experience, and failure in either can impair temporal visual perception6. Deficits in temporal visual perception are well-documented among individuals with psychosis spectrum disorders (PSD), both in judging the temporal order of rapidly presented stimuli4,6,7 and in backward masking tasks where brief temporal intervals (< 150ms) impair target identification8–11. Such impairments have been linked to core symptoms including hallucinations12, disorganized cognition13,14, and impaired social perception15,16. These deficits reflect widespread anomalies in early-level visual processing, often attributed to magnocellular pathway dysfunction11,17–20. Consistent with this account, individuals with schizophrenia exhibit reduced contrast sensitivity21, reduced sensitivity to magnocellular-biased stimuli22,23, and diminished cortical response to magnocellular stimuli19. Further, they exhibit strong spatial crowding effects24. Moreover, individuals with schizophrenia exhibit deficits in visual integration and perceptual organization18,25–29, suggesting that temporal and spatial integration deficits may reflect a common underlying dysfunction in binding information into coherent percepts. Hence, understanding the mechanisms shaping temporal visual processing—particularly how temporal order encoding and stimulus identity encoding interact— is crucial for elucidating perceptual disturbances along the psychosis spectrum. However, studying these mechanisms in isolation may overlook co-occurrence with autism spectrum disorder (ASD), which is increasingly recognized within psychosis populations. ASD is diagnosed in approximately 24% of individuals with schizophrenia, and up to 9% of those with first episode psychosis30–33. This co-occurrence extends beyond clinical diagnoses to encompass subclinical trait expressions (i.e., minor manifestations of disease, and single trait expressions that are associated with either disorder without meeting clinical criteria for a disorder34;. Accordingly, several studies have already emphasized the importance of examining these two disorders together33,35,36. One of the theoretical frameworks developed to account for this co-occurrence is the diametric model, initially suggesting that ASD and PSD represent etiologically and phenotypically opposite extremes of a unidimensional diametrical continuum, characterized by distinct and opposing patterns of social cognition and behavior37,38, See35 for other models. Under this view, social cognition and mentalizing abilities were conceptualized as underdeveloped in ASD (e.g., diminished theory of mind and an overall tendency towards mechanistic thinking), and hyper-developed in PSD (e.g., excessive theory of mind and an overall tendency towards hypertrophied mentalistic thinking). However, recent behavioral, neural, and genomic findings demonstrate that autism and psychotic traits modulate the same cognitive systems in opposing directions, yet their co-occurrence yields unique profiles characterized by compensatory or “normalizing” effects. For instance, Abu-Akel et al.39 demonstrated that while autism and psychotic traits independently impaired perspective-taking, their co-occurrence resulted in typical performance, suggesting that opposing mentalizing tendencies can modulate each other. Similarly, Abu-Akel et al.40 found that psychosis-proneness increased attentional capture by irrelevant salient stimuli, while autistic traits reduced such capture, with balanced expressions of both traits producing more adaptive, intermediate responses. Moreover, and consistent with the diametrical model, other researchers reported that individuals with elevated levels of both schizotypal and autistic traits demonstrated greater emotion recognition accuracy than those with elevations in only one trait41, and that the interaction between autistic and schizotypal traits was associated with better executive functioning and social functioning42–45. Given that the existing literature examining both conditions simultaneously has predominantly focused on higher-level social cognition and attentional control, with only limited investigation of perceptual processes35, investigating temporal perception while simultaneously measuring ASD and PSD provides an opportunity to observe opposing biases, that may exert reciprocal compensatory interactions. When studied in isolation, some individuals with ASD exhibit enhanced temporal resolution, typically reflected in shorter temporal integration windows and heightened sensitivity to rapid stimulus changes and decreased capability to integrate information over time46–50. Further, enhanced motion perception was observed in children with ASD, suggesting a more efficient extraction of dynamic visual information than controls51. However, these enhancements are not universal; several studies show typical or even reduced temporal sensitivity, depending on task demands, stimulus complexity, or individual profiles52–56. These findings underscore the notion that visual perception in ASD is not characterized by a single deficit or advantage, but by heterogeneous and context-dependent differences. Thus, although both disorders affect temporal perception, their combined impact and potential interactions remain unclear, as autistic and psychotic traits have not been measured simultaneously. Moreover, previous studies focused on low-level temporal processes while higher-level perceptual processes remain largely unexplored. Temporal crowding refers to impaired target identification by preceding and succeeding distractors that appear in the same spatial location5,57. Unlike visual masking which operates at briefer intervals (< 150ms58–61; temporal crowding spans longer durations (170-475ms) and has similar magnitude with central and peripheral vision62. Importantly, temporal crowding impairs both target encoding—reflected in reduced encoding precision—and temporal order encoding—reflected in increasing erroneous report of distractors63. The fact that this dual impairment occurs even with relatively long inter-item intervals and to a similar degree regardless of eccentricity suggests it may reflect higher-level perceptual processes. Therefore, investigating this phenomenon can extend our understanding of perceptual disturbances in PSD and ASD beyond previously reported low-level deficits. Here, we investigated whether psychosis-proneness and autistic traits interactively affect temporal crowding using an orientation-estimation task. By employing mixture-model analysis, we estimate both temporal order encoding and stimulus identity encoding, providing novel examination of higher-level temporal visual processing. Importantly, the present study adopted a dimensional framework and examined autistic traits and psychosis-proneness in a neurotypical sample, rather than clinically diagnosed ASD or PSD. Although these trait dimensions are conceptually and clinically distinct from categorical diagnoses, they are thought to reflect subclinical variation related to liability for ASD and PSD related phenotypes. As such, studying these traits in the general population may provide insights into mechanisms relevant to clinical conditions while avoiding some of the condition-related confounds such as chronic illness, active symptomatology and pharmacological treatment. The findings should therefore be interpreted within a dimensional framework, with implications that may extend to clinical populations. Given documented effects of both psychosis-proneness and autistic traits on visual perception, we expected both to affect temporal crowding. Critically, because individuals with high co-occurrence of both traits have demonstrated improved performance in the socio-cognitive and attentional domains, suggesting potential trait compensation39,40, in this study, we expected their interactions to yield a similar pattern of effects. To examine these interactions, we employed response surface analysis (RSA64–66, which allows examination of both balanced and biased trait expressions, potentially unraveling compensatory mechanisms. Method Participants Participants were recruited online via Prolific (www.prolific.com), a research-recruitment platform with a pre-registered international participant pool. The study was posted on Prolific’s marketplace, where eligible users (i.e., participants filtered by Prolific’s prescreening tools according to our eligibility criteria detailed below) enrolled to participate. No additional recruitment channels (e.g., social media, mailing lists, or university websites) were used. Eligibility criteria: Ages 18–35 years, English fluency, normal or corrected-to-normal vision, and no reported psychological, mental, or physical health conditions. The upper age limit of 35 years was set to minimize age-related changes in visual perception. Visual functions relevant to the present task—including contrast sensitivity, temporal resolution, and processing speed—begin to show measurable decline from the late thirties onward67. Restricting the sample to young adults therefore reduced unwanted variance unrelated to the trait dimensions of interest. Several technical restrictions were applied to ensure data quality. Participation was restricted to desktop and laptop computers via Prolific’s device filters; mobile and tablet devices were not permitted. The task was required to run in either Chrome or Firefox browsers, with the experiment enforcing fullscreen mode for the duration of the session. Participants were instructed at the outset to sit 50–70 cm from the screen, and screen distance was verified through the virtual chinrest calibration procedure68. Using G*Power69, an a priori power analysis for a full factorial multiple regression with two predictors (autistic traits and positive psychotic experiences), indicated a target sample of 95 participants, assuming α =.05, power=.80, and a medium effect size (f2 ≥ 0.15)40. After exclusion procedures (detailed below), Experiment 1 included 81 participants (M/F = 32/49; Mean Age = 26.22 (SD = 4.06)), and Experiment 2 included 93 participants (M/F = 36/57; Mean Age = 27.59 (SD = 4.37)). Participants’ submissions were accepted if the entire task and questionnaires were completed and at least two of three attention checks were answered correctly. The attention checks were instructed-response items embedded within the questionnaire battery (e.g., “Please mark ‘definitely agree’ for this item”). Three such items were inserted at evenly spaced intervals across the 70 trait questionnaire items (one every 20 items), spanning both the AQ and the CAPE-positive scales. Participants were compensated £6.75 for their participation. Multiple submissions by the same participant were prevented automatically by Prolific, which restricts each verified user account to a single submission per study via the participant’s unique Prolific ID. The Prolific ID was also collected at the start of each session for verification. For Experiment 2, we applied two additional prescreening criteria: (i) participants must have completed at least ten prior Prolific studies, of which at least nine had been accepted, and (ii) participants who had taken part in Experiment 1 were excluded. The study was approved by the Ethics Committee of the University of Haifa (395/22; 287/19) and complied with the guidelines of the Helsinki Declaration. Participant exclusion procedure Based on previous work62,63,70, an overall performance score was calculated for each participant as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{Overall}\: \text{ performance}\: =\: 1- (\mathrm{mean}\: \text{ absolute}\: \text{ error} / 180)$$\end{document} We excluded participants whose overall performance score was lower than 0.55 (in any experimental condition). This is because the mean absolute error of participants who completely guessed the target’s orientation should be around 90° and therefore overall guessing performance score should be around 0.50. In Experiment 1, 13 participants were excluded due to low overall performance score, and another participant was excluded due to failed attention checks, resulting in a final sample of 81 participants. In Experiment 2, one participant was excluded due low overall performance score, and one participant was excluded due to an extreme outlier parameter value (see Results section), resulting in a final sample of 93 participants. Material Orientation-estimation task The task was constructed using PsychoPy and ran on Pavlovia (www.pavlovia.org). In both experiments, all the stimuli and task instructions appeared on a gray background (RGB= [128,128,128]). A sequence of three circular stimuli with an inner radius line was presented at the center of the screen. The second stimulus was designated as the target and the first and third stimuli were the distractors. The radius line’s orientation was chosen randomly from 360 possible orientations for each of the circles, with the constraint that items in the same sequence differed from one another by at least 30° (for stimuli construction, see Supplementary Material 1A). Each stimulus appeared for 50ms, with inter-stimulus intervals of either 150ms or 350ms, resulting in SOAs of 200ms or 400ms, respectively (Fig. 1). The SOA was fixed for a given trial but varied randomly across trials. The stimuli were embedded in a white noise patch. A probe appeared 500ms after the last blank display of the sequence. The task was to reproduce the orientation of the target—the second stimulus, by adjusting the orientation of the probe using the mouse to click on the probe’s circle outline. A white dot appeared with the first mouse click at the selected location. The participants could adjust their response without a time limit, and they locked their final response by pressing the space bar. In an uncrowded condition, only the target appeared followed by the probe. Feedback was given only in the practice trials. Each participant performed 80 trials on each experimental condition (crowded 200ms, crowded 400ms, uncrowded). The trial type varied randomly across the experimental session. Due to a coding error, all participants completed 241 trials instead of 240. Given that Experiment 1 yielded high guessing rates (see Results section), the stimuli in Experiment 2 were altered to be more visible (Supplementary Material 1B). Fig. 1A schematic illustration of an experimental trial in: (a) Experiment 1; (b) Experiment 2. ISI= interstimulus interval. A schematic illustration of an experimental trial in: (a) Experiment 1; (b) Experiment 2. ISI= interstimulus interval. The autism-spectrum quotient (AQ) The AQ71 is a widely used self-report questionnaire designed to quantify traits associated with the autism spectrum among neurotypical adults72,73. The AQ consists of 50 items, with 10 items representing five subscales: social skills, communication, attention to details, imagination and attention switching. Each item is rated on the 4-point Likert scale (“definitely agree,” “slightly agree,” “slightly disagree,” “definitely disagree”). Following the standard scoring procedure, each response is coded dichotomously: both response options in the autistic-trait direction (i.e., either “definitely” or “slightly” agree/disagree, depending on item polarity) are scored as 1, and the other two responses are scored as 0. The AQ-50 total score is the sum of all 50 items, ranging from 0 to 50, with higher scores indicating more autistic traits74. The total AQ score was used as the primary index of autistic traits, rather than the individual subscale scores. This decision was both theoretical and statistical: the AQ-50 was developed as a unidimensional measure of broad autistic-trait expression, and we did not have a-priori predictions about specific subscales differentially modulating temporal visual processing. Examining each AQ subscale separately would also have substantially increased the number of statistical comparisons (five AQ subscales crossed with multiple performance parameters and SOA conditions), undermining interpretability without theoretical justification. The AQ has a moderate internal consistency with a Cronbach’s α varying from 0.63 to 0.77. Experiment 1 yielded high internal consistency (α = 0.785), with moderate consistency for its subscales: social (α = 0.709); communication (α = 0.688); imagination (α = 0.384); attention to details (α = 0.589); and attention switching (α = 0.718). The AQ internal consistency for Experiment 2 was also high (α = 0.848), with moderate consistency for its subscales: social (α = 0.795); communication (α = 0.700); imagination (α = 0.585); attention to details (α = 0.658); and attention switching (α = 0.659). The community assessment of psychic experiences (CAPE) The CAPE75 is a widely used, self-report questionnaire for the assessment of psychic experiences among adults in both clinical and non-clinical settings76. It consists of 42 items assessing three dimensions: positive symptoms (20 items), negative symptoms (14 items), and depressive symptoms (8 items). Participants rate the frequency of the items on a 4-point scale (i.e., ‘never’–‘nearly always’). In this study, we utilize the positive dimension as a proxy for psychosis-proneness77 which consists of four subscales: bizarre experiences, perceptual abnormalities, persecutory ideation, and magical thinking78. The positive dimension was selected because it specifically captures attenuated psychotic-like experiences—particularly anomalous perceptual phenomena that are most directly relevant to the perceptual mechanisms under investigation6,12–14. The negative and depressive dimensions of the CAPE assess constructs (e.g., anhedonia, mood disturbance) that are less directly linked to the temporal visual processes targeted here77,79. This focus is also consistent with prior work on psychosis-proneness and visual perception, which has predominantly used positive-symptom measures, allowing the present findings to be situated within that literature. Experiment 1 yielded high internal consistency (α = 0.865) and acceptable for its subscales: Bizarre experiences (α = 0.726); Perceptual abnormalities (α = 0.694); Persecutory ideation (α = 0.719); Magical thinking (α = 0.746). In Experiment 2, the internal consistency was also high (α = 0.901), and from acceptable to good for its subscales: Bizarre experiences (α = 0.828); Perceptual abnormalities (α = 0.773); Persecutory ideation (α = 0.737); Magical thinking (α = 0.749). Procedure Participants began each experiment with a brief demographic questionnaire to obtain information regarding sex, age, and Prolific ID. No additional demographic information was collected. Subsequently, participants performed the “virtual chinrest” procedure68 to control for stimuli size. The orientation-estimation task began with a practice phase of 17 trials. These trials were easier than the experimental trials, such that the SOA was longer, and the stimuli were presented without noise. The practice phase ended with nine practice trials, identical to the experimental trials. In the entire practice phase, participants received feedback based on their performance via a red dot (i.e., poor performance) or a green dot (i.e., good performance). Each participant performed 80 trials for each condition (crowded 200ms, crowded 400ms, uncrowded) in a random order. Following the task, participants were redirected automatically to an online survey containing both the AQ and the CAPE questionnaires. Participants were instructed to answer all questions honestly and without time limitations. Statistical analysis Task performance measures and model fitting The orientation error was defined as the angular difference between the target’s orientation and the participants reported orientation for each trial (e.g., if the target’s orientation was 60° and the participant adjusted the probe to 80°, the response error for that trial was + 20°). Response errors ranged from − 180° to 180°. The orientation errors from the crowded trials were analyzed separately for each SOA condition of each participant, using the two-misreport mixture model80(Eq. 2). The model estimates four parameters: (a) the width (SD) of a von Mises distribution (normal-circular distribution) of errors, that is centered on the target’s true orientation (i.e., error = 0). The SD conveys the quality of target’s encoding precision (larger SD’s= lower precision); (b) The height (g) of a uniform distribution of errors, reflecting the participants guessing rate. The g parameter is determined mainly by the signal-to-noise ratio (SNR), with higher g indicating lower SNR62,70. (c) The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:1$$\end{document} parameter represents the rate of mistakenly reporting the orientation of the first distractor (i.e., the distractor that preceded the target). (d) The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:2$$\end{document} parameter represents the rate of mistakenly reporting the orientation of the second distractor (i.e., the distractor that followed the target). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:1$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:2$$\end{document} are frequently referred to as substitution errors. These errors reflect erroneous encoding of the items’ temporal order (i.e., the participant reports the first or the third stimulus instead of the second). Thus, a higher rate of substitution errors indicates worse temporal order encoding. Substitution errors are modeled by additional von Mises distributions, with the same SD, centered around the distractors’ orientation. 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:p\left(\theta\:\right)=\left(1-g-{\beta\:}_{1}-{\beta\:}_{2}\right){\varphi\:}_{\sigma\:}\left(\theta\:\right)+\frac{g}{2\pi\:}+{\beta\:}_{1}{\varphi\:}_{\sigma\:}\left({\theta\:}_{1}^{*}\right)+{\beta\:}_{2}{\varphi\:}_{\sigma\:}\left({\theta\:}_{2}^{*}\right)$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document} represents the orientation error; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varphi\:$$\end{document} indicates the von Mises distribution with mean equal to zero (i.e., zero error) and standard deviation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\sigma\:$$\end{document} (SD); The θ1*and θ2* represents the errors relative to the orientation of the distractors. For the uncrowded condition which did not include distractors, the error distributions were fitted with the standard mixture model81(Eq. 3) that has only two free parameters (g, SD): 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:p\left(\theta\:\right)=(1-g)\:{\varphi\:}_{\sigma\:}\left(\theta\:\right)+g/2\pi\:$$\end{document} The models were fitted using the MemToolbox developed by Suchow et al.82(https://doi.org/10.1167/13.10.9). As can be seen in Fig. 2, both models fit the data well. Fig. 2Mean error distributions and model fit (in black) for the different conditions of each experiment. Mean error distributions and model fit (in black) for the different conditions of each experiment. Temporal crowding analysis We conducted one-way (SOA) repeated-measures ANOVAs for each parameter using the following R packages: tidyverse83, rstatix84. Response surface analysis RSA uses a polynomial regression model (Eq. 4) to visualize and map, in a 3D space, the response surface pattern of the 2 interacting predictors (i.e., positive psychosis-proneness and autistic traits) with task performance measures (i.e., model parameters), and to examine whether the combined effect of the interacting terms is linear or curvilinear along two axes: the line of congruence (LOC) where the predictors’ scores match each other, and line of incongruence (LOIC) where the scores do not match—one predictor is dominant over the other64,65. 4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:z\:=\:b0\:+\:b1x\:+\:b2y\:+\:b3{x}^{2}+\:b4xy\:+\:b5{y}^{2}+\:e$$\end{document} For each performance measure within each SOA condition, we examined four coefficients (a1-a4) that characterize the response surface: (a) a1 coefficient (b1 + b2; slope of the LOC) indicates whether matching predictor scores resolve in higher or lower performance measure scores; (b) a2 coefficient (b3 + b4+b5; test for curvature in LOC) reflects whether the performance measure score increases or decreases as predictors match at increasingly high or low scores; (c) a3 coefficient (b1-b2; slope of the LIOC) indicates if one type of predictor mismatch (e.g., high psychosis proneness-low autism) produces different outcomes than the other type (e.g., low psychosis proneness-high autism); and (d) a4 coefficient (b3-b4 + b5; curvature of LOIC) indicates whether the performance measure score increased or decreased as predictors diverged64. Prior to analysis, CAPE-positive and AQ total scores were standardized to z-scores (ZCAPEp and ZAQ, respectively), ensuring both predictors were placed on a common scale, and reducing multicollinearity. In the model, x represents ZAQ scores, y represents ZCAPEp scores, and z represents temporal crowding parameters. The x- and y-axes range from negative to positive scores, and 0 reflects the scale midpoint. Model fit was evaluated using R² values. RSA focused on the difference between the SOA conditions (200ms-400ms) for each parameter, and individual SOA condition models were performed when such a model reached significance. The analysis was conducted using the RSA R package (Version 0.9.1066. Because the RSA involved four pre-specified omnibus models testing the 200ms−400ms difference for each mixture-model parameter (SD, g, β1, β2), we applied a Benjamini-Hochberg false discovery rate (FDR) correction at (q=0.05) within each experiment to control for multiple comparisons. Adjusted p-values are reported throughout alongside the original p-values. As a confirmatory analysis, we additionally fit a linear mixed-effects model with the same polynomial trait structure but with SOA modeled as a within-subject factor and a random participant intercept; the model specification, rationale, and full results are reported in the Supplementary Material. Results Trait questionnaire descriptive statistics Distributions of AQ and CAPE-positive scores in both samples are presented in Fig. 3. In Experiment 1 (N = 81), AQ scores ranged from 11 to 39 (M = 23.68, SD = 6.77, Skew = 0.23), and CAPE-positive scores ranged from 21 to 61 (M = 32.49, SD = 7.77, Skew = 0.96). In Experiment 2 (N = 93), AQ-50 scores ranged from 4 to 46 (M = 24.47, SD = 7.91, Skew = 0.18), and CAPE-positive scores ranged from 22 to 64 (M = 33.74, SD = 9.39, Skew = 1.34). The AQ distributions were approximately normal in both samples, whereas the CAPE-positive distributions showed mild positive skew. Independent-samples Welch’s t-tests revealed no significant differences between the experiments in either AQ (t(171)=−0.71, p=.476, Cohen’s d =−0.11) or CAPE-positive scores (t(171)=−0.95, p=.339, Cohen’s d =−0.14), indicating that the two samples were comparable in trait expression. Fig. 3Distributions of trait questionnaire scores in Experiment 1 (N = 81; blue) and Experiment 2 (N = 93; orange). (a) Distribution of Autism-Spectrum Quotient (AQ) total scores. (b) Distribution of Community Assessment of Psychic Experiences positive-dimension scores (CAPE-p). In panels (a) and (b), bars depict density histograms (20 bins) and curves depict kernel density estimates; vertical dashed lines mark the sample means. (c) Joint distribution of AQ total and CAPE-p scores, with each point representing one participant. Distributions of trait questionnaire scores in Experiment 1 (N = 81; blue) and Experiment 2 (N = 93; orange). (a) Distribution of Autism-Spectrum Quotient (AQ) total scores. (b) Distribution of Community Assessment of Psychic Experiences positive-dimension scores (CAPE-p). In panels (a) and (b), bars depict density histograms (20 bins) and curves depict kernel density estimates; vertical dashed lines mark the sample means. (c) Joint distribution of AQ total and CAPE-p scores, with each point representing one participant. Temporal Crowding analyses (Table 1; Fig. 4) In both experiments, the SOA effect on guessing rate (g) was significant, with higher guessing rates at the shorter SOA. Additionally, there was a significant difference between the uncrowded and 400ms condition. The high guessing rate observed in Experiment 1 indicated a low SNR, suggesting difficulties in distinguishing the target from the external noise. To verify that the stimuli modifications in Experiment 2 successfully improved the SNR, we compared the average guessing rates between experiments. Independent samples t-tests revealed significantly lower guessing rates in Experiment 2 compared to Experiment 1 across all experimental conditions: Uncrowded condition (ΔM = 0.286; t(172) = 11.57, p<.001, Cohen’s d = 1.76), 200ms condition (ΔM = 0.279; t(172) = 8.99, p<.001, Cohen’s d = 1.37), and 400ms condition (ΔM = 0.282; t(172) = 10.20, p<.001, Cohen’s d = 1.55). Analysis of the target identity encoding parameter (SD), in both experiments, also revealed a significant SOA effect. SD was significantly larger in the 200ms than 400ms condition, suggesting that temporal crowding decreases the precision of target encoding, replicating previous studies of temporal crowding62. Additionally, SD in the 400ms condition remained significantly larger than in the uncrowded condition, demonstrating persistent temporal interference even at longer intervals. As for the encoding of temporal order, we start with testing whether the participants made substitution errors even with the relatively long temporal intervals employed here (200-400ms). One-sample t-tests (collapsed across SOA) revealed that both first distractor (β1) and second distractor (β2) substitution rates were significantly greater than zero in Experiment 1 (β1: M = 0.013, t(80) = 7.78, p<.001, Cohen’s d = 0.86; β2: M = 0.012, t(80) = 5.53, p<.001, Cohen’s d = 0.61) and Experiment 2 (β1: M = 0.008, t(92) = 7.63, p<.001, Cohen’s d = 0.78; β2: M = 0.005, t(92) = 6.95, p<.001, Cohen’s d = 0.71). Hence, the participants demonstrated impaired temporal order encoding even when the items in the sequence were separated by relatively long temporal intervals. However, when examining substitution errors as a function of SOA, we found no significant effects of SOA on either β1 or β2 substitution rates. Thus, while participants reliably confused the temporal order of the stimuli, these substitution errors did not vary as a function of the temporal separation between stimuli, differing from previous studies of temporal crowding63. We note, however, that this specific pattern of SOA effects—a significant effect on guessing rate and no significant effect on substitution errors—was observed before in Sahar and Yeshurun (Experiment 262. We also observed this specific pattern of results in data of studies in our lab that are not yet published. It seems that some participants, when confused about the temporal order (which likely occur more often with the short SOA) prefer to guess rather than take the risk of a substitution error. Such a scenario would indeed result in a significant SOA effect with guessing but not substitution errors. Still, we do not know what leads the participants to choose this response strategy in one experiment but not in the other. Importantly, in both experiments, substitution rates were reliably above zero suggesting that temporal order encoding was indeed impaired by the presence of distractors. Fig. 4Estimated mean of parameters in Experiments 1 (blue) and 2 (pink) as a function of SOA and the Uncrowded condition (UC). (a) guessing rate; (b) standard deviation; (c) first distractor substitution errors; (d) second distractor substitution errors. Error bars represent one within-subject standard error. ‘*’ indicates a significant SOA effect (p<.05). Estimated mean of parameters in Experiments 1 (blue) and 2 (pink) as a function of SOA and the Uncrowded condition (UC). (a) guessing rate; (b) standard deviation; (c) first distractor substitution errors; (d) second distractor substitution errors. Error bars represent one within-subject standard error. ‘*’ indicates a significant SOA effect (p<.05). Table 1Temporal crowding results: descriptive statistics and statistical tests.ParameterExp.200ms Mean(sd)400ms Mean (sd)Uncrowded Mean(sd)Effect of SOA(200 vs. 400)400ms vs. UncrowdedGuessing Rate (g)10.452 (0.233)0.361 (0.234)0.335 (0.225)F(1,80) = 31.56, p<.001, ηp²=0.28t(80)=−2.26, p=.026,d=0.4920.167 (0.173)0.076 (0.117)0.046 (0.075)F(1,92) = 57.12, p<.001, ηp²=0.38t(92)=−3.40, p=.001, d = 0.31Target Encoding (SD)112.319 (10.942)9.564 (2.114)8.545 (2.112)F(1,80) = 5.66,p=.020, ηp²=0.06t(80)=−4.44, p<.001,d=0.86213.139 (3.640)10.030 (2.589)8.132 (2.018)F(1,92) = 105.96, p<.001, ηp²=0.53t(92)=−11.31, p<.001, d = 1.21First Distractor Substitution Error (β1)10.016 (0.026)0.010 (0.017)—F(1,80) = 2.28, p=.135, ηp²=0.02—20.009 (0.013)0.007 (0.015)—F(1,92) = 1.12, p=.294, ηp²=0.01—Second Distractor Substitution Error (β2)10.014 (0.033)0.009 (0.017)—F(1,80) = 1.64, p=.204, ηp²=0.02—20.006 (0.012)0.004 (0.011)—F(1,92) = 0.90, p=.346, ηp²=0.01—Note. Significant results in bold font. Exp.=Experiment. Table 1Temporal crowding results: descriptive statistics and statistical tests.ParameterExp.200ms Mean(sd)400ms Mean (sd)Uncrowded Mean(sd)Effect of SOA(200 vs. 400)400ms vs. UncrowdedGuessing Rate (g)10.452 (0.233)0.361 (0.234)0.335 (0.225)F(1,80) = 31.56, p<.001, ηp²=0.28t(80)=−2.26, p=.026,d=0.4920.167 (0.173)0.076 (0.117)0.046 (0.075)F(1,92) = 57.12, p<.001, ηp²=0.38t(92)=−3.40, p=.001, d = 0.31Target Encoding (SD)112.319 (10.942)9.564 (2.114)8.545 (2.112)F(1,80) = 5.66,p=.020, ηp²=0.06t(80)=−4.44, p<.001,d=0.86213.139 (3.640)10.030 (2.589)8.132 (2.018)F(1,92) = 105.96, p<.001, ηp²=0.53t(92)=−11.31, p<.001, d = 1.21First Distractor Substitution Error (β1)10.016 (0.026)0.010 (0.017)—F(1,80) = 2.28, p=.135, ηp²=0.02—20.009 (0.013)0.007 (0.015)—F(1,92) = 1.12, p=.294, ηp²=0.01—Second Distractor Substitution Error (β2)10.014 (0.033)0.009 (0.017)—F(1,80) = 1.64, p=.204, ηp²=0.02—20.006 (0.012)0.004 (0.011)—F(1,92) = 0.90, p=.346, ηp²=0.01—Note. Significant results in bold font. Exp.=Experiment. Temporal crowding results: descriptive statistics and statistical tests. F(1,80) = 5.66, p=.020, ηp²=0.06 Note. Significant results in bold font. Exp.=Experiment. Response surface analysis results Experiment 1 Among all parameters analyzed, only the RSA model for the 200–400 difference in first distractor substitution errors (β1) was significant, accounting for 14.6% of the variance (R²=0.146, F(5,75) = 2.55, p=.035; see detailed results of the polynomial regression in Supplementary Results 2A, Table 2). In this model, two significant RSA predictors emerged (Fig. 5b). First, we observed a significant additive linear effect (a₁=0.011, SE = 0.005, 95% CI [0.001, 0.021], p=.034), indicating that matching higher scores of both traits predicted greater improvement in performance (i.e., larger decrease in substitution errors in the transition from the 200ms to 400ms) than matching low scores. Second, we observed a significant negative curvature along the line of incongruence (a₄=−0.015, SE = 0.006, 95% CI [−0.026, −0.004], p=.006), suggesting that the more the traits match one another, there was a greater reduction in substitution errors. The significant RSA effect was specific to the difference between conditions, as neither condition yielded significance (200ms: n = 81, R²=0.042, p=.681; 400ms: n = 80, R²=0.055, p=.516; Fig. 5a, c). We note that in the 400ms β1 model, one participant was excluded from the analysis due to Bollen & Jackman85 multivariate criteria, resulting in (n = 80) for said model. These findings support our hypothesis of trait compensation, as individuals with matched high traits exhibited the greatest benefit from the change in temporal intervals. However, we note that after applying Benjamini-Hochberg FDR correction across the four primary RSA models (q=0.05), this effect did not survive correction (p-adjusted=0.122; Supplementary Results, Table 5). Still, the current RSA results are consistent with significant interaction observed with the linear mixed-effects model (Supplementary Results, Table 7). Experiment 2 As for Experiment 2, only the RSA model for the difference (200–400) in second distractor substitution errors (β2) was significant, accounting for 19.6% of the variance (R²=0.196, F(5,87) = 4.237, p=.002; see detailed results of the polynomial regression in Supplementary Results 2B, Table 3). In this model, two significant RSA predictors emerged (Fig. 5e): the a3 coefficient (a₃=0.008, SE = 0.004, 95% CI [0.0003, 0.015], p=.041) and the a₄ coefficient (a₄=−0.010, SE = 0.004, 95% CI [−0.017, −0.003], p=.007). The a3 coefficient indicates that individuals with higher autism scores combined with lower psychotic traits benefited more than individuals with higher psychosis scores and lower autism scores, in the transition from the 200ms to the 400ms SOA. The a4 coefficient suggests that as both trait scores increasingly matched one another, there was a greater reduction in substitution errors between SOA conditions, than when both trait scores increasingly deviate from one another. Collectively, these findings support our hypothesis regarding trait compensation effect, as individuals with matching traits benefit more from the SOA transition. Crucially, this effect survived Benjamini-Hochberg FDR correction across the four primary RSA models (q=0.05; p-adjusted=0.008; Supplementary Results, Table 6). Fig. 5Response surface analysis (RSA) of positive psychosis-proneness and autism traits interaction on temporal crowding performance measures in Experiments 1 (a-c) and 2 (d-f). The red axis is the congruence line (LOC) representing equal expressions of autistic traits (ZAQ) and positive psychosis-proneness (ZCAPEp). The blue dotted axis is the incongruence line (LOIC), representing the relative dominance of autistic or positive psychosis-proneness. Surface color-coding represents the distractor substitution errors (β1 or β2). ‘*’ indicates a significant SOA effect (p<.05). The outer/inner black circles represent 95%/50% of the sample. Response surface analysis (RSA) of positive psychosis-proneness and autism traits interaction on temporal crowding performance measures in Experiments 1 (a-c) and 2 (d-f). The red axis is the congruence line (LOC) representing equal expressions of autistic traits (ZAQ) and positive psychosis-proneness (ZCAPEp). The blue dotted axis is the incongruence line (LOIC), representing the relative dominance of autistic or positive psychosis-proneness. Surface color-coding represents the distractor substitution errors (β1 or β2). ‘*’ indicates a significant SOA effect (p<.05). The outer/inner black circles represent 95%/50% of the sample. Examining the individual SOA conditions yielded different patterns. For the 200ms condition (Fig. 5d), the overall model was not significant (R²=0.054, F(5,87) = 0.985, p=.432), while for the 400ms condition (Fig. 5f) a significant model emerged, explaining 19% of the variance (R²=0.190, F(5,86) = 4.082, p=.002; see detailed results of the polynomial regression in Supplementary Results 2C). In this latter model, two significant predictors emerged: the a₁ coefficient (a₁=−0.003, SE = 0.002, 95% CI [−0.006, 0.000], p=.039), suggesting that when both traits matched at lower scores there were more substitution errors, than when both traits matched at higher scores; and the a₄ coefficient (a₄=0.004, SE = 0.002, 95% CI [0.000, 0.007], p=.027), suggesting that as trait scores increasingly mismatched, there were more substitution errors with the succeeding distractor, than when both trait scores matched. We note that in this model (i.e., 400ms β2 model) one participant was excluded from the analysis due to Bollen & Jackman85 multivariate criteria, resulting in (n = 92). The RSA results of Experiment 2 are consistent with the linear mixed-effects model interaction (Supplementary Results Table 8). These findings further strengthen our hypothesis regarding trait compensation effects. Discussion This study investigated how psychosis-proneness and autistic traits interact to influence temporal visual processing, particularly their effects on temporal crowding. Temporal crowding is an ideal phenomenon for this goal as it simultaneously challenges both temporal order encoding (i.e., maintaining the correct order of events) and stimulus identity encoding (i.e., achieving high encoding precision of stimulus features). We utilized the orientation-estimation task and sequential stimuli presentation with relatively long SOAs to measure temporal crowding and used the CAPE and AQ questionnaires to measure the presence of autistic traits and psychosis-proneness. Across both experiments, we observed robust temporal crowding effects, including reduced target encoding precision (increased SD) and increased substitution errors—reflecting impairments in both stimulus identity encoding and temporal order encoding, respectively. Critically, our key empirical contribution lies in demonstrating how psychosis-proneness and autistic traits interact to influence temporal order encoding, as evidenced by the reduced substitution errors in Experiment 2. In Experiment 1, the RSA results revealed that matching higher scores of both traits predicted greater performance improvement between the 200ms and 400ms conditions than matching low scores. Similarly, when trait scores increasingly matched, there was a larger reduction in substitution errors. Together, this pattern of results may suggest that individuals with matching high levels of both traits showed the greatest benefit. This pattern was also apparent in Experiment 2, in which the items’ visibility was higher. Here too, as both trait scores increasingly matched, there was a larger reduction in substitution errors between the SOA conditions in comparison to when they did not match. However, in this experiment, although individuals with higher autistic traits combined with lower psychosis-proneness benefited more from the SOA difference than those with higher psychosis scores and lower autism scores, the greatest benefit was at the intersection of the incongruence and congruence lines (0,0). Importantly, the model revealed that the performance gains were most pronounced among individuals with matching levels of autistic traits and psychosis-proneness. This observed reduction in substitution errors among individuals with balanced levels of autistic traits and psychosis-proneness supports the presence of a compensatory mechanism between these traits. This interaction pattern is illustrated, for example, by the significant RSA model in the 400ms condition of Experiment 2: We observed a bowl-shaped response surface with a concave line of incongruence. This indicates that individuals biased toward either autistic or psychotic traits showed similarly high rates of substitution errors. In contrast, these errors were significantly reduced in individuals with balanced trait expressions—especially in those scoring high on both traits, as evident by the significant negative a₁ coefficient. This pattern suggests that even under relatively less demanding temporal conditions (400ms), individuals with uneven expression of traits remain vulnerable to perceptual interference. Conversely, a balanced expression of both traits—particularly when both are elevated—confers resilience, aligned with the idea that the compensatory interaction between autistic and psychotic traits operates across varying levels of perceptual challenge. The observed reduction pattern in substitution errors under co-occurring autism and psychotic traits offers a compelling insight into how distinct neurocognitive profiles may interact to produce unexpected perceptual advantages. This finding is particularly striking given that individuals with psychosis and schizophrenia disorders typically exhibit marked deficits in temporal order encoding, characterized by selective impairments in temporal order judgments even when they can accurately detect temporal asynchrony7. Importantly, our findings demonstrate that these temporal order encoding failures extend beyond the brief intervals (< 150ms) typically examined in temporal order judgment tasks to much longer intervals (up to 400ms), suggesting that impairments of temporal order encoding in psychosis may persist across a wider temporal window than previously documented. Formerly, these deficits were widely linked to dysfunction within the magnocellular visual pathway, which is critically involved in processing rapidly changing stimuli and temporal information17–19,22. In contrast, emerging evidence suggests that autism may involve a distinct processing profile. For example, Brown and Crewther86 demonstrated enhanced parvocellular temporal efficiency in autistic children, reflected in reduced second-order VEP amplitudes––which was coupled with perceptual performance and non-verbal intelligence in autism. Although this finding concerns early visual pathways, it aligns with behavioral evidence that autistic individuals may prefer a “local-first” processing strategy, individuating elements before extracting global structure54. To the extent that the local-first preference also holds in the temporal domain, one plausible account for our findings holds that when autistic traits are equally present, this detail-oriented processing style may counteract psychosis-related temporal deficits. Given that substitution errors arise from failures in maintaining the order of temporally adjacent items, a local-first encoding strategy may indirectly support temporal order by preserving the distinctiveness of each item in the sequence. Consequently, the co-occurrence of autistic traits may transform the temporal crowding challenge from a global integration problem (where psychosis-related deficits create vulnerability) into a local discrimination task (where autism-related processing tendencies may confer a compensatory advantage). These findings have important implications for understanding how perceptual deficits may cascade to affect higher-order cognition. The current findings extend psychosis-related deficits from early visual stages to higher-level temporal processing, consistent with the hierarchical breakdown of visual processing documented in schizophrenia29. This suggests that impairments in temporal processing may impact higher-order cognition in schizophrenia spectrum disorders, potentially contributing to symptoms of cognitive disorganization13. Such temporal processing deficits may disrupt the encoding of sequential information, undermining the ability to maintain coherent representations of ongoing experiences and contributing to disturbances in thought organization and perceptual integration. This temporal disintegration is thought to play a central role in characterizing cognitive disorganization, manifesting clinically as thought disorder and perceptual fragmentation6. Importantly, however, the interaction pattern we observed suggests that autistic traits may illuminate the mechanisms underlying these higher-order disturbances. When autistic traits are present, the “local-first” processing style which individuates items earlier and more distinctly, may help preserve the temporal position of each event more effectively, thereby stabilizing perceptual representations that would otherwise be vulnerable in psychosis. In this way, autism may provide a natural contrast case that would clarify how failures in early temporal encoding may propagate upward to produce disorganized cognition in schizophrenia. Consequently, our findings suggest that the detrimental cascading effects on higher-order cognition may be mitigated when autistic traits and psychosis-proneness co-occur in balanced expressions. Furthermore, our findings have also implications regarding the nature of the relationship between autism and psychosis spectrum disorders, as they are aligned with the diametric model, positing that psychosis and autism represent opposing cognitive phenotypes38,39. These findings add to the growing body of evidence suggesting that their co-occurrence can yield a “normalizing” effect39–42; See also43–45. However, these studies have predominantly focused on higher-order cognition and social functioning. Here, we shift the level of analysis to mid-to-higher-order temporal visual processing, demonstrating for the first time how interactions between autistic and psychotic traits shape the perceptual encoding of temporal information. By targeting intermediate perceptual mechanisms that bridge sensory input and cognitive organization, our findings identify a plausible mechanistic pathway through which trait interactions may give rise to downstream cognitive and clinical phenotypes. Several limitations warrant consideration. Our study utilized a neurotypical sample of young adults, using self-report questionnaires, which allows examination of underlying mechanisms without confounds from chronic illness or medication effects but limits direct generalizability to clinical populations. Future research should examine these interactions in individuals with diagnosed disorders to validate the clinical relevance of our findings, as well as with wider age range and measures that do not rely on self-report to avoid potential reporting biases. Additionally, while we demonstrated behavioral compensatory interactions in temporal crowding, the neural mechanisms underlying these interactions remain unclear. Future neuroimaging studies could examine whether the behavioral compensatory interactions we observed are reflected in neural activity patterns, potentially identifying specific brain networks where trait compensation occurs. In conclusion, our findings suggest that the compensatory interactions between psychosis-proneness and autistic traits operate at a mid-to-high perceptual level and specifically demonstrate that balanced expressions of autistic traits and psychosis-proneness may produce compensatory behavioral effects in temporal order encoding, as evidenced by a reduction in substitution errors when both traits are elevated. Critically, these interactions emerged at relatively long temporal intervals (200-400ms)—extending previous findings on psychosis-related temporal processing deficits, which have predominantly focused on brief intervals (< 150ms) in temporal order judgment tasks. While temporal crowding impaired both temporal order encoding (i.e., substitution errors) and stimulus identity encoding (i.e., SD), the trait interactions selectively influenced order encoding, suggesting that a “local-first” processing style associated with autistic traits may specifically counteract psychosis-related failures in maintaining temporal order. Methodologically, employing response surface analysis allowed us to uncover these nuanced compensatory interactions that would remain obscured when examining either trait in isolation. This highlights the importance of considering autistic traits as a key dimension for evaluating psychosis-proneness effects on temporal processing and vice-versa77,87, providing important implications for understanding symptom heterogeneity in psychosis spectrum conditions. These findings are aligned with the diametric model of autism and psychosis and may identify a plausible perceptual mechanism through which opposing cognitive phenotypes may mutually compensate. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1