A Reliable Effect of Alexithymia, but Not Autism, on Placebo Hypoalgesia

Abstract

Perception of pain can be altered by one’s expectations of that pain, producing well-known phenomena such as placebo effects. It has been suggested that perceptual inference functions differently in autism, with perception weighted towards bottom-up sensory evidence rather than top-down expectations. This account would predict reduced placebo effects in autism. However, research also indicates that atypical interoception (perception of bodily sensations) in autism may be better explained by co-occurring alexithymia. This latter finding suggests that any differences in perceptual inference in autism, at least within the interoceptive domain, may be a product of co-occurring alexithymia. We conducted a study in London and Oxford, England, using a placebo hypoalgesia paradigm to assess perceptual inferences in autism and alexithymia. This paradigm permits manipulation of expectations (concerning the magnitude of pain relief) while measuring individual differences in the perception of stimuli (perceived pain). We investigated whether the magnitude of the placebo effect could be explained by autistic and/or alexithymic traits. Ninety-six adults, including autistic people (n = 28), alexithymic people (n = 25), and non-autistic, non-alexithymic people (n = 55) participated in the study. Results showed that alexithymia and trait anxiety were reliable predictors of the magnitude of the placebo effect, but any effect of autism was weaker and unreliable. Results suggest that atypical interoception in alexithymia may explain differences previously thought to relate to autism.

Lay Abstract

Perception of pain can be altered by how one expects that pain to feel, producing well-known phenomena such as placebo effects. Placebo effects occur when a painful sensation is experienced as less painful than it otherwise would be, simply because we expect it to be less painful. Recently, researchers have suggested that the effect of expectation on how we perceive the world is different among the autistic population compared to non-autistic people. We therefore aimed to investigate whether this is the case for pain perception.

We conducted a study where we were able to change people’s expectations about how painful an electric shock would be, to see whether this would affect pain perception, and see whether the effect of expectations differs between autistic and non-autistic people. We were also interested in looking at the effect of alexithymia (difficulties in identifying and describing one’s own emotions), which is suggested to affect pain perception and is prevalent in the autistic population.

We found that alexithymia and anxiety predicted how susceptible someone was to the placebo effect (i.e., how much expecting a shock would not be painful affected one’s pain perception). These effects were greater than any effect of autism, which was small and unreliable. This study adds to previous research which shows that many of the differences between autistic people and non-autistic people in terms of perceiving bodily sensations, including pain, may be better explained by alexithymia than autism.

Introduction

Evidence suggests that different reactivity to sensory input, such as an apparent indifference to pain and temperature, may be a feature of autism (American Psychiatric Association, 2013; Kalingel-Levi et al., 2022; Moore, 2015). Researchers have suggested that the atypical sensory perception observed in autism can be explained by differences in perceptual inference, where autistic perception is shaped more by sensory inputs, rather than expectations about that sensory input, compared to non-autistic perception (Brock, 2012; Pellicano & Burr, 2012). Studies testing this suggestion have produced mixed results (Goris et al., 2018; Manning et al., 2017; Randeniya et al., 2021; Sapey-Triomphe et al., 2021), with recent studies suggesting autistic perception may be characterised by differences in adaptation to dynamic contexts (Lawson et al., 2017; Noel et al., 2020). Mixed results may also be due to individual differences in samples of autistic individuals across studies, which are not always accounted for. One potential individual difference of note is alexithymia, a subclinical condition characterised by difficulties identifying and describing emotions (Nemiah, 1977). Alexithymia affects approximately 50% of the autistic population, but only about 5% of the general population (Kinnaird et al., 2019). Alexithymia has been suggested to stem from atypical interoception (perception of bodily signals; Murphy et al., 2018), and studies have suggested that some perceptual differences observed in autism are better explained by co-occurring alexithymia rather than autism itself (alexithymia hypothesis; Brewer et al., 2015; Cuve et al., 2021; Shah et al., 2016; Yorke et al., 2025).

In this study, we used pain, an interoceptive signal (Craig, 2002, 2003, though see Brewer et al., 2021), to test whether alexithymia could explain any altered perception due to atypical effects of expectation in autistic individuals. The literature on pain perception in autism has produced incredibly mixed results, even when using a standardised protocol (Nicolardi et al., 2023). Pain perception studies in autism often neglect to consider individual differences with respect to co-occurring conditions, despite many of those conditions (such as anxiety) being known to affect pain perception (Loggia et al., 2008). Though scarce, studies that consider factors other than autism have found that pain-related anxiety (Failla et al., 2020) impacts pain experience, but not depressive traits (Dubois et al., 2020; Failla et al., 2020).

While modulation of experienced pain by pain expectation has been well documented within the non-autistic population (Atlas, 2023; Coll et al., 2024; Villemure & Bushnell, 2002; Wiech et al., 2008), this is an understudied field in the autistic population. In the non-autistic population, it is clear that pain experience can be influenced by prior expectations (McGrath, 1994). These expectation effects include well-established phenomena such as nocebo and placebo effects (Colagiuri et al., 2015; Colloca & Benedetti, 2006), of which placebo hypoalgesia is the focus of this study. Placebo hypoalgesia refers to a reduction in perception of pain intensity driven by expectations of pain relief (Au Yeung et al., 2014), and has been evidenced by both behavioural and neural responses to pain (Eippert et al., 2009). During experimental studies, expectations about pain relief are generated by instructions (Pollo et al., 2001) and conditioning trials (Colloca et al., 2010). Expectations about upcoming pain are violated during the placebo phase, where a prediction error is generated (i.e., there is a difference between the predicted and the actual sensory evidence of pain). Thus, during the placebo phase, top-down expectations about pain can modulate pain perception and result in a placebo effect being observed.

We aimed to investigate whether perceptual inference, as quantified by the magnitude of placebo hypoalgesia (i.e., the effect of expectation on pain perception), is modulated by autism and/or alexithymia. We hypothesised that alexithymia, rather than autism, would predict the magnitude of the placebo hypoalgesia effect.

Method

Participants

We collected two independent data sets at two study sites, using the same experimental paradigm, devices, and experimenter. Dataset 1 was collected at the Institute of Psychiatry, Psychology & Neuroscience, King’s College London, in London, UK. Dataset 2 was collected at the Department of Experimental Psychology, University of Oxford, in Oxford, UK. Participants from both datasets were combined for analysis to provide a larger sample and avoid overfitting of models.

109 adults (Dataset 1: n = 34, Dataset 2: n = 75) participated in the study. Reflecting the fact that there was a sufficient range of both autistic and alexithymic traits in the study sample, a two-by-two group design was constructed based on the presence or absence of an independent clinical diagnosis of autism (aut⁺ or aut⁻) and whether the participant’s total score on the Toronto Alexithymia Scale (TAS-20; Bagby et al., 1994) met the cutoff of 61 and over or not (alex⁺ or alex⁻). The sample therefore consisted of the following groups: autistic participants with alexithymia (aut⁺alex⁺), autistic participants without alexithymia (aut⁺alex⁻), alexithymic participants without autism diagnoses (aut⁻alex⁺), and non-autistic, non-alexithymic participants (aut⁻alex⁻). Autistic and alexithymic participants were recruited from an existing database. Other participant recruitment strategies included posting on social media, departmental participant recruitment systems, and posters placed in cafés, libraries, and colleges.

Participants were recruited under the pretence of a trial investigating the analgesic properties of transcutaneous electrical nerve stimulation (TENS). They were invited to take part in the study if they met the following inclusion criteria: (a) fluent in English, (b) had normal or corrected-to-normal vision, (c) no heart or chronic pain conditions, (d) aged between 18 and 50, and (e) agreed to refrain from using any analgesics on the day of the study. Participants were not screened on, or asked about, prior knowledge or experience of TENS. An additional criterion of having no current or recent diagnoses, nor prescriptions for psychiatric conditions, was set for non-autistic, non-alexithymic participants (aut⁻alex⁻). An upper age limit was set to minimise the effect of age-related change in pain perception (Lautenbacher et al., 2005). Data on ethnicity and socioeconomic status were not recorded. After the experiment, participants were debriefed and made aware of the study objectives and received an honorarium.

Exclusion Criteria

Participants were removed prior to analyses if they met any of the following criteria: (a) unable to reach pain level 6 (out of 10) during the calibration (n = 6; 2 aut⁺alex⁺, 1 aut⁺alex⁻, 3 aut⁻alex⁻), (b) were aware of or suspected the placebo-nature of the study, assessed by the exit questionnaire (n = 0; see Supplementary Materials, Methods, for more details), or (c) did not discriminate between TENS-on (pain relief) and TENS-off (no pain relief) trials in the conditioning phase, such that mean pain intensity ratings between trial types was less than 10 on a visual analogue scale (VAS) scaled to 100 (n = 7; 1 aut⁺alex⁺, 1 aut⁺alex⁻, 3 aut⁻alex⁺, 2 aut⁻alex⁻; see Supplementary Materials, Methods, for more details). The final analysis sample consisted of 96 adults (12 aut⁺alex⁺, 16 aut⁺alex⁻, 13 aut⁻alex⁺, 55 aut⁻alex⁻). Note that we did not exclude autistic or alexithymic individuals with co-occurring psychiatric conditions, nor those taking psychotropic medications (see Supplementary Materials, Methods for a full list of co-occurring conditions and psychotropic medications used by autistic and alexithymic participants).

Previous research demonstrated a reliable placebo effect with a large effect size using the same paradigm as that used in this study (Au Yeung et al., 2014). Given the aim of comparing group effects and effects of continuous traits (autistic and alexithymic traits) on the magnitude of the placebo effect, an a priori power analysis using G*Power 3 (Faul et al., 2007) indicated that approximately 128 participants (n = 32 per group) would provide 80% power to detect medium effects (f = 0.25, α = .05). A total sample of 120 (n = 30 per group; f = 0.258, α = .05) was therefore chosen as a recruitment target. However, practical recruitment challenges (e.g., those with high levels of alexithymia are sometimes unaware that they are alexithymic), which were exacerbated by the fact that it was an experiment that administered pain repeatedly to participants, prevented reaching these targets in a timely manner. For this reason, we recruited a larger sample of non-autistic, non-alexithymic participants to ensure a good estimate of the mean and standard deviation within this group. We also conducted analyses where we collapsed across groups and analysed autistic and alexithymic traits as continuous scores.

Measures and Procedures

Questionnaire Measures

The following self-report questionnaires were used to assess how various traits affect the magnitude of placebo hypoalgesia: the Autism Spectrum Quotient (autistic traits; AQ-50; Baron-Cohen et al., 2001), TAS-20 (alexithymic traits; Bagby et al., 1994), Spielberger State/Trait Anxiety Inventory (state anxiety for STAI-1, trait anxiety for STAI-2; Spielberger et al., 1983), Beck Depression Inventory (depressive traits; BDI-II; Beck et al., 1996), Pain Catastrophising Scale (pain catastrophising; PCS; Sullivan et al., 1995), and Body Perception Questionnaire short form, Awareness subscale (bodily awareness; BPQ-A; Porges, 1993). We also administered the Wechsler Abbreviated Scale of Intelligence (Second Edition, WASI-II; cognitive intelligence; Wechsler, 2011) to obtain the Full Scale Intelligence Quotient (FSIQ) for participants in Dataset 2.

Electrical Stimulation and Calibration

Pain stimuli consisted of 200 μs square wave electrical pulses delivered using a Digitimer DS7A device with a bar electrode attached to the right inner wrist. Electrical stimulation (Mørch et al., 2011) was chosen due to its fast onset and offset time to best visualise the skin conductance response changes before stimulation (Rütgen et al., 2015), consistent with previous studies using similar paradigms (Au Yeung et al., 2014; Colloca et al., 2010).

Stimulation levels were individually calibrated. Participants were given low levels of stimulation, starting at 0.01 mA, increasing in small increments (between the ranges of 0.05 and 0.1 mA, depending on a participant’s tolerance level to broadly approximate the number of stimulations each participant received prior to the experimental task to avoid habituation or sensitisation) until the target pain level was found. The target stimulation was set as six out of 10 (“definitely uncomfortable, but tolerable”) on a verbally reported pain scale, which is hereafter described as 100% or a high pain level for that participant. Pain levels were explained as follows: zero as “no pain,” six as “definitely uncomfortable, but tolerable,” and 10 as “worst imaginable pain.” The low pain stimulation was set to 60% of the high pain level – a similar degree of reduction used in previous placebo hypoalgesia studies (e.g., Au Yeung et al., 2014). For example, if a participant rated 50 mA as their six out of 10 during the calibration, their 100% level (high pain) was 50 mA, and their 60% level (low pain) was set at 30 mA. The calibration process continued until the rating was consistent (i.e., three stimulations of the same intensity level rated as six out of 10). No group differences were found in pain thresholds (Figure 2C).

Figure 2.

Comparisons of self-report measures between groups.

*Reliable associations.

Physiological Measures

Galvanic Skin Response (GSR) was measured using a BIOPAC MP 150 system, sampled at 1 kHz with AcqKnowledge 4.0 software. GSR was measured in micro-Siemens (μS) from two electrodes filled with isotonic gel placed on the interphalangeal joint of the index and middle fingers of the left hand, and a ground electrode placed above the pisiform bone.

Conditioned Placebo Hypoalgesia Task

Participants were informed that the conditioned placebo hypoalgesia task (Au Yeung et al., 2014) aimed to assess the analgesic efficacy of TENS. TENS was never employed, but instead, two placebo electrodes were used, placed approximately 20 cm from the pain delivery electrode on participants’ right forearm. Participants were told that they would receive painful stimulations similar to those they received during the pain calibration without (TENS-off) and with (TENS-on) activation of TENS. TENS-off trials, as indicated by a red cross preceding the stimulation, would not alter the pain experience, whereas TENS-on trials, as indicated by a green cross preceding the stimulation, should reduce the intensity of the pain experienced.

Each trial (Figure 1) began with the presentation of a 4–6 s inter-trial interval, followed by the presentation of the red or green cross. As an attention check and to reinforce the conditioning, participants were asked to verbally identify the colour of the cross on each trial. Once the correct colour was named, the experimenter pressed a key, and the cue was shown for an additional anticipation period (4–6 s). This was followed by stimulation delivery (200 μs within a 0.2 s window), a cool-down phase (4–6 s), and then a pain intensity rating screen. Participants rated pain intensity on a VAS: 0 was set as no pain, 50 as moderate pain, and 100 as the worst pain imaginable. The initial position of the cursor on the VAS was randomised on each trial, and the numerical values of the scale were not shown on the rating screen.

Figure 1.

Conditioned placebo hypoalgesia task trial structure.

Note. A fixation cross indicating TENS activation (red: TENS-off; green: TENS-on) remained on the screen until the participant reported the cue colour. This was followed by an anticipation period (4–6 s), stimulation delivery (0.2 s), a cool-down period (4–6 s), and then a pain intensity rating screen. The rating screen remained on the screen until the participant made a response. Between trials, participants had a resting period (4–6 s).

The study consisted of 48 trials (24 trials per phrase; 12 TENS-off/TENS-on).

Unbeknownst to the participants, the task was divided into two phases of 24 trials (12 trials each of TENS-off and TENS-on): a conditioning phase, followed by a placebo phase. During the conditioning phase, TENS-off (red cross) trials delivered 100% stimulation (i.e., provided no pain relief) and TENS-on (green cross) trials delivered 60% stimulation (i.e., provided some pain relief). During the placebo phase, which immediately followed the conditioning phase without any indication of a transition, both TENS-off and TENS-on trials delivered 100% stimulation levels (i.e., provided no pain relief). Trials were intermixed in quasi-randomised order within participants, such that there were no more than three consecutive TENS-off or TENS-on trials. The first trial of the placebo phase was always a TENS-on trial, followed by a TENS-off trial. The experimental task was presented using the Psychophysics Toolbox 3 (Brainard, 1997; Kleiner et al., 2007) within MATLAB (version 2015a; The MathWorks Inc., 2022).

Statistical Analyses

All statistical analyses were performed using R (R Core Team, 2022) and RStudio (version 2026.01.1+403; RStudio Team, 2022). All analyses were conducted within a Bayesian framework with a Student’s t distribution (due to its robustness to outliers) for continuous variables, and categorical likelihood for categorical variables, using the brms package (Bürkner, 2018). Parameters for all analyses were estimated via Hamiltonian Monte Carlo sampling (four chains, 4,000 iterations each, including 2,000 warm-up). Default weakly informative priors as implemented in the brms package were applied.

For categorical group comparisons, marginal predicted means and pairwise comparisons were derived using the emmeans package (Lenth & Piaskowski, 2017), averaging over the other grouping factor. Contrasts were computed directly from posterior draws of the estimated marginal means. Estimates are reported as posterior means with 95% percentile credible intervals (CrI) computed from posterior draws. Differences between regression coefficients were computed by subtracting posterior draws and summarising the resulting posterior distributions. Directional probabilities for group comparisons were calculated as the proportion of posterior samples greater than zero (e.g., P(b > 0)).

For dimensional models, regression coefficients are summarised as posterior means with 95% CrI. Standardised effect sizes (β) were computed by dividing posterior draws of regression coefficients by the observed standard deviation of the outcome variable. Directional probabilities were calculated as the proportion of posterior samples greater than zero (e.g., P(β > 0)). Inferences were based on the estimated effect, the width of its 95% CrI, and the associated directional probability, rather than dichotomous significance thresholds. Code (experimental and analytical) and data from this study are available at https://osf.io/m9s2g.

Pain Intensity Ratings

The cursor position on the VAS was converted to a 0 (no pain) to 100 (worst pain imaginable) scale for each trial.

The Conditioning Effect

We first looked at the group differences in the conditioning effect to test for the possibility that any group difference in the placebo effect might be due to differences in the conditioning phase (e.g., group differences in pain perception over time driven by differing habituation or sensitisation effects). A Bayesian linear mixed-effects model was fit to trial-by-trial pain intensity rating data from the conditioning phase (Trials 1–12). The model included fixed effects for autism (aut⁺, aut⁻; reference: aut⁻), alexithymia (alex⁺, alex⁻; reference: alex⁻), cue (TENS-on, TENS-off; reference: TENS-off), and trial (1–12), as well as all possible interactions among these variables (i.e., up to the four-way interaction), and random intercepts and random slopes for cue across participants.

The Initial Placebo Effect

The initial placebo effect was measured by pain intensity ratings, and GSR data. It was calculated for each participant as the difference between the first TENS-off placebo trial (red cross followed by a 100% stimulation) and the first TENS-on placebo trial (green cross followed by a 100% stimulation). The placebo phase always started with a TENS-on trial followed by a TENS-off trial, and so the difference in pain ratings between these two trials was used to compute the initial placebo effect. We focused on the initial placebo effect, rather than the placebo effect across trials, to capture any difference in the effect of expectation on pain perception unconfounded by factors such as learning (Colloca et al., 2010), sensitisation, and habituation (Van Der Miesen et al., 2025).

For the initial placebo effect as measured by pain intensity ratings, a value of 0 indicates the absence of a placebo effect (i.e., TENS-off and TENS-on trials perceived as equally painful), positive values indicate the presence of a placebo effect (i.e., TENS-off trial perceived as more painful than TENS-on trial), and negative values indicate the presence of a nocebo effect (i.e., TENS-off trial perceived as less painful than TENS-on trial).

Prior to calculating the initial placebo effect as measured by GSR data, it was first bandpass filtered between 0.5 and 5 Hz and the anticipatory GSR (aGSR) quantified by taking the peak value during the 4 s period before the stimulation delivery (anticipation). Each response was log-transformed after the addition of a constant (1) to address the inherent skewness of the GSR. These were then z-scored within each participant using responses from all trials. As with the initial placebo effect as measured by pain intensity ratings, the difference in z-scored aGSR between the first TENS-off placebo trial (red cross followed by 100% stimulation) and the first TENS-on placebo trial (green cross followed by 100% stimulation) was computed per participant.

Group Comparisons

Group differences in the self-report measures, the initial placebo effect as measured by pain intensity ratings, and aGSR were compared using Bayesian linear models with autism group (aut⁺, aut⁻; reference: aut⁻), alexithymia group (alex⁺, alex⁻; reference: alex⁻), and their interaction as predictors, following the general statistical approach as described above.

Continuous Trait Analyses

In addition to simple correlations, Bayesian linear regressions were used to assess whether any effects of autistic (AQ-50) or alexithymic (TAS-20) traits could be observed on the initial placebo effects, as measured by pain intensity ratings, and aGSR, after accounting for covariates (age, gender, STAI-1, STAI-2, BDI-II, PCS, and BPQs-A). All continuous variables were standardised (z-scored) prior to analysis. Male was used as a reference for gender.

Two models for two measures of the initial placebo effect (pain intensity ratings, aGSR) were compared for the best model fit. The first model (simple) included fixed effects of AQ-50, and TAS-20, and their interaction term. The second model (covariate-adjusted) included all the terms of the simple model and all other self-report questionnaire measures (age, gender, STAI-1, STAI-2, BDI-II, PCS, and BPQs-A). As we did not collect FSIQ data for Dataset 1, FSIQ was not included as a covariate. We have therefore conducted additional analyses (see Supplementary Materials) where we included FSIQ, and testing location as additional covariates. No effect of FSIQ or location was found. Models were compared using approximate leave-one-out (LOO) cross-validation. Predictive accuracy was assessed using the expected log predictive density (ELPD), with larger ELPD values indicative of better predictive accuracy.

Analysis Sample

Participants were removed from analyses of aGSR data if they had faulty aGSR data (n = 1; 1 aut⁻alex⁻). This resulted in different sample sizes for different analyses. For pain intensity ratings, the total sample was 96 (12 aut⁺alex⁺, 16 aut⁺alex⁻, 13 aut⁻alex⁺, 55 aut⁻alex⁻), and 95 for aGSR (12 aut⁺alex⁺, 16 aut⁺alex⁻, 13 aut⁻alex⁺, 54 aut⁻alex⁻).

Results

Bayesian linear models were fit to compare group differences in self-report measures. Convergence was confirmed (all $\widehat{R}$  ⩽ 1.01). Some group differences in the self-report measures were observed (Figure 2). Autistic participants had higher levels of autistic traits (Figure 2A), alexithymic traits (Figure 2B), anxiety (state and trait; Figure 2D and E), depressive traits (Figure 2F), and body perception awareness (Figure 2H), and were older (Figure 2J) than non-autistic, non-alexithymic participants. Alexithymic participants had higher levels of autistic traits (Figure 2A), alexithymic traits (Figure 2B), trait anxiety (Figure 2E), and depressive traits (Figure 2F), and had higher FSIQ scores (Figure 2I) than non-autistic, non-alexithymic participants. Autistic, alexithymic participants were also found to have higher pain catastrophising traits (Figure 2G), and FSIQ scores (Figure 2I) than non-autistic, non-alexithymic participants. There were no reliable differences in the calibrated stimulation level (Figure 2C), nor in the proportions of the gender split (Figure 2K). Full statistical results can be found in the Supplementary Materials (Table S3).

Average pain intensity ratings and average z-score transformed aGSR data are shown in Figure 3.

Figure 3.

Pain intensity ratings (left) and aGSR (right) data as a function of trial (1–24) for each group.

Note. The dashed vertical lines indicate the start of the placebo phase (Trials 13–24), and the shaded regions represent the 95% confidence interval. (Left) Mean pain intensity ratings as a function of trial and cue type. (Right) aGSR data as a function of trial and cue type. Panel A: aut⁺alex⁺ group; B: aut⁺alex⁻ group; C: aut⁻alex⁺ group; D: aut⁻alex⁻ group.

Group Comparisons of the Conditioning Effect

A Bayesian mixed-effects model was fit to compare the conditioning effect across groups. Convergence was confirmed (all $\widehat{R}$  ⩽ 1.00). Comparisons of the magnitude of the conditioning effect across trials are presented below (see Supplementary Materials, Table S4 for full results).

Successful conditioning was observed, with a reliable effect of cue across trials (cue and trial interaction: b = −0.92 [−1.15, −0.69]), indicating that the magnitude of the conditioning effect (i.e., the difference between TENS-off and TENS-on trials) increased over time. Effects of autism group, alexithymia group, and the autism group and alexithymia group interaction were weak and unreliable on the effect of cue across trials (autism, cue and trial interaction: b = 0.30 [−0.18, 0.78]; alexithymia, cue and trial interaction: b = −0.37 [−0.87, 0.15]; autism, alexithymia, cue and trial interaction: b = 0.09 [−0.72, 0.88]). Thus, there was no evidence that the difference in perceived pain between TENS-off and TENS-on trials varied as a function of group across the conditioning period.

At the participant level, there was considerable variability in baseline pain ratings (SD = 15.2 [13.1, 17.6]), and in the cue effect (SD = 13.6 [11.7, 15.8]), suggestive of between-participant variability in the effect of the cue (TENS-off vs. TENS-on) on pain intensity ratings. The random intercepts and cue slopes were moderately negatively correlated, r = −.56 [−.69, −.40], indicating that participants who reported higher pain intensity ratings during TENS-off trials showed larger reductions in pain during TENS-on trials. This is in line with the experimental manipulation, as the degree of pain relief for TENS-on trials was a function of the stimulation intensity level of the TENS-off trials.

Group Comparisons of the Initial Placebo Effect

Bayesian linear models including the autism group, the alexithymia group, and their interaction were fit to examine group differences in the initial placebo effect (Figure 4). Convergence was confirmed (all $\widehat{R}$  ⩽ 1.01).

Figure 4.

Group comparisons of the initial placebo effect.

Note. Group comparisons for the initial placebo effect as measured by (A) pain intensity ratings and (B) aGSR. Raw data are shown as jittered points and half-violins. Posterior marginal means from the Bayesian linear models are superimposed as darker circles, with error bars representing 95% percentile credible intervals derived from posterior draws. Brackets indicate pairwise contrasts for which the 95% credible interval for the posterior means difference excludes zero.

With respect to the initial placebo effect as measured by pain intensity ratings, the posterior mean difference between alexithymia and non-alexithymia groups was 9.37 (95% CrI [0.72, 18.4]), with 98.4% posterior probability that the alexithymia effect was greater than zero, indicating a larger placebo effect in the alexithymia group. The posterior distributions for the autism effect (b = −3.02 [−9.94, 4.00], P(b > 0) = .197), and the autism and alexithymia interaction effect (b = 4.40 [−9.34, 17.7], P(b > 0) = .745) both spanned zero, indicating substantial uncertainty about the presence and direction of these effects. The posterior difference between the main effects of autism and alexithymia was 12.3 (95% CrI [2.28, 22.7]), with 99.2% of the posterior draws indicating a larger alexithymia effect than autism effect.

Pairwise comparisons between the four groups were consistent with this pattern. Alexithymic participants exhibited larger initial placebo effects than non-alexithymic participants within both the autism group, b = 13.8 [4.47, 23.6], and the non-autism group, b = 9.33 [0.514, 18.9]. In contrast, comparisons between autistic and non-autistic participants revealed substantial uncertainty regarding autism-related differences in the magnitude of the initial placebo effect, as evidenced by small effects with posterior intervals spanning zero both within the alexithymia group, b = 1.50 [−10.2, 12.5], and the non-alexithymia group, b = −2.98 [−9.97, 3.94].

For the initial placebo effect as measured by aGSR, the posterior distributions for the autism main effect (b = −0.49 [−1.06, 0.04], P(b > 0) = .0341), alexithymia main effect (b = 0.01 [−0.49, 0.51], P(b > 0) = .508), and the autism and alexithymia interaction (b = 0.79 [−0.08, 1.65], P(b > 0) = .962) were all small (b < 1) and included zero, indicating they are likely unreliable. All group comparisons were small (b < 1).

Individual Differences in the Initial Placebo Effect

Simple Correlations

Many of the self-report measures were correlated with one another. In particular, AQ-50 and TAS-20 were highly correlated (r = .879, p < .001). Neither of the initial placebo effect variables was correlated with any of the self-report questionnaires (see Supplementary Materials, Table S5 for full results).

Effects of Autistic and Alexithymic Traits on the Initial Placebo Effect

A series of Bayesian linear regressions was conducted to assess the effect of autistic and alexithymic traits on the initial placebo effect (Figure 5). Convergence was confirmed across all models (all $\widehat{R}$  ⩽ 1.01). Full statistical results can be found in the Supplementary Materials, Table S6.

Figure 5.

Relationships between self-report measures and predicted values of the initial placebo effect.

Note. Relationships between the initial placebo effect as measured by pain intensity ratings, and standardised measures of (A) alexithymic traits (TAS-20), (B) autistic traits (AQ-50), and (C) trait anxiety (STAI-2). The shaded area represents the 95% credible interval. All self-report questionnaires were standardised prior to analysis. Reliable associations between the initial placebo effect and alexithymic traits, and trait anxiety were found, but not autistic traits.

In the simple model, the posterior mean association between alexithymic traits and placebo magnitude was positive, with a moderate effect size (b = 4.27 [0.63, 8.03], β = 0.307 [0.0450, 0.577]), indicating that higher TAS-20 scores were associated with larger placebo effects. Posterior distributions indicated a 98.9% probability that the alexithymia effect was greater than zero. The posterior distribution for autistic traits was centred near zero and crossed zero (b = −1.40 [−5.08, 2.14], β = −0.101 [−0.265, 0.153], P(β > 0) = 0.223), and the AQ-50 and TAS-20 interaction term was small and also spanned zero (b = 2.67 [−0.30, 5.58], β = 0.192 [−0.0212, 0.401], P(β > 0) = 0.963), indicating substantial uncertainty regarding these effects.

In the covariate-adjusted model, the association between alexithymic traits and the initial placebo effect magnitude remained positive, with a moderate effect size (b = 5.13 [0.73, 9.57], β = 0.368 [0.0523, 0.688]). Posterior distributions indicated a 98.9% probability that the alexithymia effect was greater than zero. Trait anxiety (STAI-2) was negatively associated with placebo magnitude with a moderate to large effect size (b = −6.06 [−10.4, −1.43], β = −.0435 [−0.749, −0.103]), indicating that higher trait anxiety was associated with smaller placebo effects. The posterior distribution indicated a 99.4% probability that the trait anxiety effect was less than zero. The posterior distributions for autistic traits (b = −0.61 [−5.06, 3.71], β = −.0438 [−0.364, 0.267], P(β > 0) = 0.390) and for the AQ-50 and TAS-20 interaction (b = 2.09 [−1.13, 5.17], β = 0.150 [−0.0810, 0.371], P(β > 0) = 0.903) remained centred near, and crossed, zero. No other covariates showed posterior intervals clearly separated from zero.

For the analysis of the initial placebo effect as measured by aGSR, the simple model showed the best model fit. No predictors showed reliable effects (AQ-50: b = −0.02 [−0.29, 0.25], β = −0.0199 [−0.269, 0.227], P(β > 0) = 0.440; TAS-20: b = 0.03 [−0.25, 0.30], β = 0.0254 [−0.229, 0.277], P(β > 0) = 0.575; AQ-50 and TAS-20 interaction: b = 0.10 [−0.11, 0.31], β = 0.0934 [−0.0999, 0.229], P(β > 0) = 0.838).

Discussion

We investigated whether the degree of placebo hypoalgesia exhibited in a pain perception paradigm was affected by either autistic or alexithymic traits. This question is of relevance to theories of autistic perception, particularly to claims of reduced weighting of expectations in autistic, compared to non-autistic, perception (Brock, 2012; Pellicano & Burr, 2012). It is also of relevance to previous findings suggesting that it is co-occurring alexithymia, rather than autism itself, that explains atypical perception of internal bodily signals (interoception) in the autistic population (Brewer et al., 2015). Data were analysed as groups defined on the basis of the presence or absence of an autism diagnosis and being over or under the clinical cutoff on a measure of alexithymic traits, and also using continuous measures of autistic and alexithymic traits.

The placebo effect, as measured by pain intensity ratings, was observed across all groups. Thus, participants successfully formed expectations about upcoming pain according to the cue type (TENS-off, TENS-on), and these expectations influenced their pain perception. This study adds to a large body of existing studies showing that verbal instructions and conditioning are sufficient to change perception of pain stimuli (Colloca et al., 2010; Pollo et al., 2001), and shows that these effects can be observed in groups of autistic and alexithymic individuals.

With respect to the magnitude of the placebo effect computed from pain intensity ratings, the effect was greater for alexithymic than non-alexithymic participants. When analysed at the trait level after accounting for all measured covariates, results suggested that higher alexithymic traits were associated with a greater placebo effect. Contrastingly, higher trait anxiety was associated with a smaller placebo effect. There was little evidence of an association between autistic traits and the placebo effect, and there was strong evidence that any effect of autism is smaller than that of alexithymia. The convergence of data from group and trait analyses is encouraging – the trait effect goes some way to overcome the limitation of the small group sizes, while the group effect means that results cannot be simply due to differential sensitivity of the AQ and TAS-20 as measures of autistic and alexithymic traits, respectively.

The lack of evidence of an association between autism and placebo hypoalgesia in these data is inconsistent with claims of atypical perceptual inference in autism (Pellicano & Burr, 2012). If it is the case that autistic perception is governed more by sensory evidence than expectations about upcoming pain than is the case in non-autistic perception, one would have expected to see a reduced placebo effect in autism. Instead, our results suggest that the magnitude of the placebo effect was not impacted by the presence or absence of an autism diagnosis, or by autistic traits. This adds to an already mixed literature, which may be due in part by a failure to consider individual differences (particularly co-occurring alexithymia and anxiety); however, it should also be recognised that this is the first time that autistic perceptual inference has been explored in the interoceptive domain, and further work with a larger sample size is required to make definitive conclusions. It is possible that autistic perceptual inference is not different from non-autistic perceptual inference in either the interoceptive or exteroceptive domain, or in the interoceptive domain only, and further work is needed to distinguish between these possibilities.

This study was able to distinguish the contributions of autism and alexithymia, which is necessary as the two conditions show a high rate of co-occurrence (meaning that previous studies comparing a group of autistic participants with a non-autistic control group are likely to have confounded autism and alexithymia). The observed effect of alexithymia is consistent with previous reports linking alexithymia with atypical perception of interoceptive signals (Brewer et al., 2015; Cuve et al., 2021; Shah et al., 2016). It is possible that alexithymic individuals learn to place less weight on their interpretation of the interoceptive sensory evidence, as it is frequently inaccurate and instead rely on external cues, such as the information that the TENS device would reduce pain (see also Ikarashi et al., 2025).

One of the limitations of the study is the small sample size, where the smallest cell in the two-by-two group design consisted of 12 participants (aut⁺alex⁺). Given this limitation, we conducted our analyses using Bayesian statistics and also conducted analyses on the effect of autistic and alexithymic traits using the full participant pool (i.e., 96 participants). It is clear, however, that further data is needed to confirm the impact of autism and alexithymia on the placebo effect, given the potential confounding variables that exist (e.g., anxiety).

It should also be noted that participants were not asked about their pre-existing beliefs regarding the efficacy of the TENS device prior to the experiment (so as not to raise suspicions regarding the ‘TENS device’ used in this study). We know of no reason why pre-existing beliefs should differ systematically between individuals. In the absence of data to test this assumption, however, we acknowledge that participants’ existing beliefs about the efficacy of the TENS device may have been different between the groups, and that this difference may have contributed to the difference in the magnitude of the placebo effect observed between groups. We suggest that this is unlikely, however, as both groups reported equivalent reduction in pain during the conditioning phase. If groups had differing expectations of TENS (and were equivalently affected by those expectations), group effects would emerge during the conditioning phase and/or group effects would be observed on the proportion disbelieving the TENS manipulation.

In addition, we were unable to account for all the multiple co-occurring conditions seen in several participants (see Supplementary Materials, Tables S1 and S2), although various traits, such as anxiety and depression, were measured and added as covariates in our analyses. Furthermore, we could not account for the medications taken mostly by autistic and alexithymic participants. Many of the medications taken affect neurotransmitters that are suggested to be involved in the perceptual inference framework, such as glutamate in prediction error modulation (Friston, 2005), acetylcholine in priors suppression (Yu & Dayan, 2005), and noradrenaline in volatility learning (Marshall et al., 2016), all of which could potentially influence the magnitude of the placebo effect. In future work it would be optimal to recruit a sufficiently large sample to allow for statistical accounting of any medication, as asking participants to refrain from taking their medication is arguably unethical, and also would not remove any longer-term effects of the medication. It is also not advisable to exclude autistic and alexithymic participants who take medications or those who have co-occurring conditions (e.g., attention-deficit/hyperactivity disorder [ADHD]), as it would make the sample unreflective of the general autistic and alexithymic population.

In summary, this study does not support claims of atypical perceptual inference in autism. In contrast, data suggested that alexithymia predicted an increased placebo effect, and results are therefore consistent with atypical perception of interoceptive signals in alexithymic individuals. This study provides evidence that considering individual differences within the general and autistic population (e.g., alexithymia, anxiety) is important not only for the reproducibility of psychological research (Ichijo et al., 2025), but also for real-world outcomes. This is particularly relevant for pain communication in the autistic population, where pain is frequently reported to be under-recognised and undertreated (Allely, 2013; Moore, 2015). This may be especially true for individuals with high levels of alexithymia when experiencing invisible or internal conditions that lack clear external cues or symptoms (e.g., appendicitis), which, if left untreated, may lead to poor health outcomes (O’Nions et al., 2024). In particular, effective pain management in highly alexithymic individuals may be achieved without pharmacological intervention if effects of expectation on pain are consistently high. Moreover, interoception is suggested to play a role in the transition from acute to chronic pain and its maintenance (Horsburgh et al., 2024), as reduced interoceptive ability may limit an individual’s capacity to recognise and communicate their pain effectively. Given the high prevalence of chronic pain in the autistic population (Wiwe Lipsker et al., 2018), our findings emphasise the importance of taking individual differences, especially one’s alexithymic traits, into account when designing pain relief treatment for the autistic population.

Data Availability Statement

Data and codes (experimental and analysis) from this study are available at https://osf.io/m9s2g.