Music may have a sweet spot: too predictable becomes boring, while too unpredictable stops feeling musical. This study asks whether that sweet spot shifts for people with autistic or psychotic-like traits—and the answer is surprisingly not supported by the data.

People prefer music with an intermediate level of predictability: not so predictable as to be boring, yet not so unpredictable that it ceases to be music. This sweet spot for predictability varies due to differences in the perception of predictability.

The study tested whether psychotic and autistic traits were associated with a higher preference for predictability in music. Participants heard twenty-nine pre-composed music excerpts scored for complexity by musical experts, and a participant’s preferred predictability corresponded to the peak of the inverted U-shaped curve between music complexity and liking—a Wundt curve.

The sweet spot varied between individuals, but the study found no support for those variations being associated with autistic and psychotic traits. Predictability has a sweet spot: too much unpredictability is stressful, while excessive predictability is fatiguing.

People use compensatory strategies to regulate predictability in the environment, such as engaging in exploration to cope with boredom or seeking information to make an unfamiliar situation more predictable. What one person considers comfortable, someone else may find either too monotonous or too chaotic, so the optimal level depends on the unpredictability perceived by the individual.

Higher perceived unpredictability is proposed to be a causative factor in both Autism Spectrum Disorder and psychosis. Excessive unpredictability is stressful and unpleasant, and people are willing to pay to avoid it. Unpredictable stimuli cause high levels of arousal and a high strain on attentional and cognitive resources.

Attention and learning rates increase in unpredictable environments because unexpected events signal that statistical relationships are not fully learned. When relationships change so rapidly that learning is no longer worth the effort, high unpredictability can produce the unpleasant feeling of not being in control.

To reduce distress, people engage in behaviours that turn unpredictable situations more predictable. An intermediate level of predictability should produce the highest amount of stimulation without causing aversion, and should therefore be experienced as the most pleasurable.

Gold and colleagues showed that people like songs with intermediate levels of predictability better than songs either low or high in predictability. This inverted U-shaped relationship between liking and predictability has been found in music, visual texture patterns, geometric shapes, and online web pages.

In the visual domain, a preference for intermediate levels of predictability has even been found in infants. The experience of predictability reflects subjective perceptions of predictability, which only partly relate to objective features such as the number of tones in a song or edges in a painting.

People vary in their perception of stimuli as predictable or unpredictable. Forming probabilistic predictions based on one’s model of statistical properties in music is thought to be central to music perception. Differences in long-term or online learning of musical regularities lead to differences in expectations; a music expert can make more accurate predictions and thereby experience less unpredictability than a non-musician.

Differences in subjective perception could explain why individuals differ in preferred predictability when exposed to identical levels of objective unpredictability. The explanation has implications for psychosis and Autism Spectrum Disorder because symptoms of both disorders have been separately attributed to overestimations of uncertainty.

According to these theories, overestimations of uncertainty arise from excessive prediction errors, meaning the deviation between prediction and outcome. A prediction error can signal that learning is incomplete, that a change has happened and requires updating, or that inherent environmental randomness limits how much learning can improve prediction.

As unpredictability increases and changes become more frequent, the rate of error signals also increases. Overestimating unpredictability should cause a corresponding surge in aversion and distress. Psychotic and autistic symptoms have been linked to experiencing increased distress in unpredictable situations.

The prediction was that individuals with psychotic and autistic traits would prefer higher levels of predictability, with sweet spots skewed toward more predictable stimuli or environments. The study aimed to investigate whether tendencies toward psychosis and Autism Spectrum Disorder were related to a higher preference for predictability in music.

The researchers chose to focus on music preferences, although they expected that a higher preference for predictability might be observable in other domains. They opted for a naturalistic music setting, using music composed and performed by humans, to increase the chance of capturing ecologically valid behaviours reflecting real-life responses to unpredictability in music.

If preference for predictability influences music preferences, how much a listener enjoys a piece should depend on the listener’s perceived level of predictability. Because not all acoustic features contribute equally to the experience of predictability, subjective evaluations might capture relevant aspects better than an assessment based on acoustic features.

The main study therefore focused on subjective evaluations, while also exploring whether the results would replicate with an objective measure of predictability. A participant’s preferred predictability was represented by the peak of an inverted U-shaped curve between preference and predictability, also called a Wundt curve; individual differences appeared as lateral shifts in those peaks.

The expected association was a shift toward more predictable music for people with psychotic and autistic traits. Three hundred and twenty-six participants were recruited through the online platform Prolific and at the campus of UiT—The Arctic University of Norway.

Participants came from the general population because psychotic and autistic symptoms are distributed along continua in the general population. Five participants were excluded for failing quality-control checks. The stimuli were selected from a pool of instrumental music excerpts whose properties were characteristic of popular music, including pop, rock, jazz, world music, or a mixture of these styles.

The excerpts were assumed to be unknown to eight musical experts, and excerpts frequently played in broadcast media were excluded. The experts rated each excerpt on overall complexity on a one-to-ten scale, and each excerpt’s complexity score was the average of the experts’ ratings.

In this study, complexity was treated as the inverse of predictability: elements that are more difficult to predict lead to more prediction error. The final stimulus pool contained twenty-nine excerpts, with complexity scores from 2.625 to 8.625 and durations from 38 to 75 seconds.

Table one lists the twenty-eight music excerpts used in the task, including each artist, song, presentation block, expert-rated complexity, and average liking from three hundred twenty-one participants. Complexity is based on ratings from eight experts, while liking was reported on a zero-to-one-hundred visual analogue scale.

The table matters because it documents the stimulus set and shows how excerpts were distributed across blocks, with two warm-up trials from block one omitted. Block one included two warm-up trials, and the excerpts were presented in random order within each block.

Participants listened to all musical pieces in their entirety and then rated how much they liked each piece on a visual analogue scale from zero, disliked very much, to one hundred, liked very much. Questionnaire items were presented between blocks to avoid fatigue. Autistic traits were measured with the twenty-eight-item Autism Spectrum Quotient, or AQ-short, which had previously been validated using non-clinical samples.

Positive symptoms of psychosis, such as paranormal beliefs, were measured using the twenty frequency items in the positive subscale of the Community Assessment of Psychic Experiences scale, or CAPEp. Three control questions reflected common misconceptions about psychosis, such as whether a participant believed in kidnappings by aliens.

The internal consistencies of AQ-short and CAPEp were very good, with alpha values of point eight five and point eight six, respectively. The main analysis included only participants whose preferred level of complexity could be determined according to pre-registered criteria. A preferred level of complexity corresponded to the peak, or apex, of an inverted U-shaped curve between preference and complexity scores—a Wundt curve.

Each peak was determined from a fitted quadratic model, using quadratic regression analyses between the music excerpts’ complexity scores and each participant’s preference ratings. The quadratic component was calculated by squaring the complexity scores.

An inverted U-shape has a negative quadratic component, and the further that component is below zero, the sharper the peak. Participants with a quadratic component larger than negative zero point one were excluded. This exclusion criterion traded off sample size against measurement error: peaks with a quadratic term near zero could vary widely because of random errors, while reducing sample size also reduces statistical power.

An a priori power analysis selected the negative zero point one cutoff to ensure a convex parabola, reducing a hypothetical sample of two hundred to one hundred fifty-nine, with eighty-one point five percent power to detect a correlation of point two. Figure two illustrates two participant-level quadratic fits between music complexity and liking.

In the left panel, the inverted-U pattern meets the pre-registered criteria for a Wundt curve, with the dashed line marking the parabola’s peak as the preferred complexity level. In the right panel, the positive quadratic component produces a U-shaped relationship rather than a Wundt curve, so the dashed line marks a bounded maximum used only in exploratory analyses.

The study calculated partial correlations between preferred complexity and autistic and psychotic traits separately, while controlling for mood and ACE-IQ sum scores. These tests were one-sided because the hypothesis predicted that autistic and psychotic traits would correlate with a preference for simpler music, with peaks toward the lower end of the complexity scale.

Exploratory analyses examined preferred complexity in relation to other indices and tested whether traits were associated with more variable liking responses. Because preferred complexity was not normally distributed, with a Shapiro-Wilks test below point zero zero one, Kendall’s rank correlation was chosen as the non-parametric test.

The study also replicated the main analysis by replacing musical experts’ complexity scores with Wiener entropy, also known as spectral flatness. Wiener entropy reflects an excerpt’s noisiness, or the uniformity of its power spectra, and is therefore an objective measure of complexity.

Entropy was calculated by dividing each excerpt into fifty-millisecond segments and analysing the segments’ frequency spectra. Reducing the segments to twenty milliseconds did not meaningfully change the results, and the same exclusion procedures produced a sample of one hundred eighty-three participants.

The linear mixed model confirmed Wundt curves between liking and complexity in the sample of one hundred eighty-one participants, with a significant positive linear effect and a significant negative quadratic effect. There were individual differences in preferred complexity, indicated by variation in the peaks of the parabolas.

The peaks had a median of 4.96 and a mean of 4.97, with a standard deviation of 1.791. The sample’s preferred levels of complexity spanned the entire complexity range of the music. Figure four is a histogram of each participant’s preferred level of musical complexity, with the vertical axis showing the count of participants.

The distribution shows substantial individual variation, but a large group preferred the lowest complexity level offered, while many others clustered around the midpoint. The authors report a median preferred level of four point nine six and a mean of four point nine seven, highlighting why individual differences matter when interpreting the overall Wundt-shaped liking pattern.

Contrary to expectations, preferred complexity was neither associated with autistic traits nor with psychotic-like experiences. The partial correlation between AQ-short scores and preferred complexity was non-significant, with Kendall’s tau equal to point zero three nine and p equal to point seven eight four.

The partial correlation between CAPEp scores and preferred complexity was also non-significant, with Kendall’s tau equal to point zero three three and p equal to point seven four three. Regression showed that CAPEp, AQ-short, music experience, and BAIS could not explain individual differences in preferred complexity, with R squared equal to point zero two one and adjusted R squared equal to negative point zero zero one.

Replacing experts’ complexity ratings with entropy scores replicated all results from the main analysis. Figure three shows overlapping histograms of mean CAPEp scores, in pink, and AQ-short scores, in blue. CAPEp values are concentrated mainly around the lower end of the scale, while AQ-short scores extend across a broader range, with many observations between about one point seven and two point eight.

This matters because it makes the sample’s variation in psychotic and autistic traits visible; the authors report that twenty-nine point eighty-three percent exceeded the AQ-short clinical cut-off, and forty-four point two percent exceeded the CAPEp ultra-high-risk cut-off.

The study investigated whether psychotic and autistic traits were related to preferring predictable music, based on the reasoning that preference is modulated by how much predictability the individual experiences. Participants varied in how much predictability among music excerpts they preferred, as shown by variations in the peaks of the inverted U-shaped relationships between music complexity and liking.

Contrary to predictions, there was no support for variations in the sweet spot for predictability being associated with psychotic or autistic traits. The lack of support was unexpected, and this was the first investigation into the relationship between psychotic-like experiences and a preference for predictability.

The results do not refute the notion of a general predictability preference in psychosis. CAPEp scores were skewed toward the lower end of the scale, making it possible that null results occurred because of a low prevalence of psychotic traits in the sample.

The findings indicate that non-clinical samples may not be sufficient to detect a relationship between tendencies toward psychosis and a preference for predictability in music. Patients with delusions self-report a higher preference for predictability than healthy controls, while no evidence was found for an association between predictability preference and delusion-proneness in the general population.

The music task did not include a measure of perceived predictability because the aim was not to test for overestimation of unpredictability. Overestimation of unpredictability cannot be inferred directly from liking responses without also controlling for other factors that influence emotional reactivity.

People may prefer predictable music because unpredictable music causes aversion, but aversion may result from experiencing high unpredictability or from having a lower aversion threshold. Measuring perceived predictability is necessary if future studies seek to interpret similar findings through computational theories of overestimation of uncertainty in psychosis and Autism Spectrum Disorder.

The study could not control for the possibility that younger people like more complex music because gender and age were not linked to participant responses. Measuring the peaks of the inverted U-shaped curves between liking and predictability showed that the sweet spot for predictability in composed music varies between individuals.

There was no support for psychotic or autistic traits being associated with liking predictable music. This was the first investigation of an association between psychotic traits and predictability preferences, while previous research on autistic traits and predictability preferences was scarce.

The results do not refute the general notion of a preference for predictable experiences in either psychosis or Autism Spectrum Disorder. The findings instead suggest that relationships between traits and predictability preferences may be difficult to observe with stimuli having high ecological validity, and that a large range of stimulus predictability is needed to account for large variations in the sweet spot for music predictability.

People differed substantially in the complexity of music they liked best, but autistic and psychotic-like traits did not explain those differences. The result highlights both the limits of this test and the need to measure perceived unpredictability directly.