Assessing the relationship between absolute pitch and autistic traits using a novel continuous slider scale

Abstract

Background

Evidence regarding the link between absolute pitch (AP) and autistic traits is currently inconclusive. Previous subclinical measurements of autistic traits using discrete response options may mask subtle variations, contributing to discrepancies between studies. This study employs a novel slider for continuous measurements to reveal AP-related subtle variations along autistic traits that discrete scales might overlook.

Methods

We contrasted autistic traits measured using a novel continuous slider with those measured using a traditional discrete scale in a larger sample (N = 120) than previous similar studies, with musicians and non-musicians stratified into high, medium, and low-AP proficiency groups. Data were collected between July 2023 and April 2024. Participants indicated their agreement/disagreement on the Autism spectrum Quotient (AQ) either (i) discretely by selecting from four fixed choices or (ii) continuously by adjusting a visual slider along a 0 to 100% continuum. Additionally, cognitive tasks associated with AP ability, including pitch adjustment, musical proficiency, and relative pitch ability, were assessed.

Results

A significant correlation between conventionally measured autistic traits and slider adjustments validated the slider’s efficacy. A higher autistic score in AP musicians was revealed across social/communication domains using continuous relative to discrete scales. Notably, continuous measurement identified a heightened autistic trait in the AQ subscale imagination in AP musicians that was not found in discrete measures. Among all cognitive abilities assessed in AP musicians, only pitch-related skills predicted autistic traits, suggesting similarly enhanced pitch functions.

Conclusion

These findings support the hypothesized link between AP and autistic traits, highlighting the need for further validation of continuous measurement scales to better understand the co-occurrence of specific human traits.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-025-03419-z.

Introduction

Absolute pitch (AP) is the ability to identify musical tones without a referent. The prevalence of AP is estimated to be around 1–4% in the general population [1, 2], with a higher incidence ranging from 4 to 15% observed in musicians [3, 4], tone-language speakers [5], and individuals with autism [6]. The development of AP ability requires both a genetic component and a critical developmental period during which music exposure often occurs [3, 4, 7, 8]. As such, AP is a relatively stable trait usually present in musically-trained individuals and is not easily trainable compared to relative pitch (RP), which is the ability to identify musical intervals given a reference tone. Over the years, numerous studies have examined the behavioral, anatomical and functional brain differences underlying AP ability with much focus on processing of musical pitch, pitch intervals, and pitch categorization [for a review, see 9]. In addition to musical domains, some studies have found that AP ability extends to influence processing in non-musical domains, including most notably language processing such as lexical tone perception [7, 10, 11]. Further, AP effects have been reported in the domains of visuospatial processing [12], stream segregation [13, 14], and auditory working memory [15,16]. Interestingly, it has been suggested that individuals with AP may be associated with a tendency towards certain cognitive styles or personality traits similar to those of autistic individuals, though inconsistencies currently remain [17, 18].

The cross-domain link between AP ability and subclinical autistic traits is supported by some behavioral and neural studies. Autism spectrum disorder (ASD) encompasses a group of neurodevelopmental disorders characterized by challenges in social communication, repetitive behaviors, sensory hypersensitivity, and difficulties adapting to unexpected change [19]. Similar to the development of AP ability, both genetic and environmental factors play important roles in ASD. Evidence supporting a putative link between AP and autistic-like traits came from studies indicating a higher proportion of AP (i.e., 11%) in individuals with autism compared to typically-developing individuals [6, 20, 21]. Among AP possessors, an increased prevalence of autism and Asperger syndrome has also been documented [22, 23]. In addition to prevalence, some studies have indicated increased autistic traits in musicians with AP. Specifically, musicians with AP have been shown to exhibit higher scores on self-rated autistic trait subscales across the socio-communicative and imagination domains, with higher autistic traits correlated with higher accuracy on AP identification task [17, 18, 24].

Additionally, AP individuals have demonstrated enhanced abilities in extracting a visual target shape embedded in a figure or an auditory target melody within a multi-melody background [12, 14, 25], an orientation towards local details similarly observed in individuals with autism [26, 27]. This local-oriented processing bias is also supported at the neural level, with AP listeners exhibiting reduced connectivity in whole-brain functional networks both structurally [28] and during natural music/audiobook listening [29, 30] and resting state EEG [18], suggesting a widely underconnected brain with reduced functional integration. Such regional hyperconnectivity coupled with global hypoconnectivity have been similarly observed in autistic individuals, particularly with respect to the default mode network [31, 32]. From a theoretical perspective, the link between AP and autism is attributed to shared characteristics in perceptual processing and/or cognitive style. The enhanced perceptual functioning theory and the theory of veridical mapping connect AP and autism with enhanced low-level perception and a tendency for one-to-one mapping between similar perceptual or cognitive structures [33–35]. For instance, individuals with AP demonstrate superior recall of the direct correspondence between musical pitches and verbal labels, similar to the heightened recall for pairings of tones and animal pictures observed in autistic children [36]. Similarly, a detail-oriented processing style, emphasized in accounts such as weak central coherence account [37], the enhanced perceptual functioning theory [35], and the empathizing-systemizing theory [38], may underlie the cognitive profiles of both groups. Along the same line, Chin [39] proposes a developmental model for AP, suggesting that children with a tendency towards details are more likely to develop AP ability. While these findings support a shared local processing style in AP and autism, inconsistencies persist as detailed-oriented performance is not consistently observed across all domains (e.g., auditory but not visual) in individuals with autism or AP [25].

While existing evidence suggests a putative link between AP and autism, the observed effects tend to be small and based on small samples. Specifically, studies reporting a higher proportion of AP among individuals with autism are primarily based on case studies or case reports [for a review, see 40]. Among the few studies indicating an increased incidence of autistic traits among AP possessors, the sample sizes are relatively limited (ranges from 13 to 33 per group) with variations in the AP cutoff criteria [i.e., 17, 18; for details refer to Supplementary Table S1]. Additionally, among studies that have reported higher autistic traits in musicians with AP, the effect has been generally weak or non-significant, with small-to-moderate effect sizes (e.g., p = .058, η² = 0.15; p = .081, d = − 0.44; [17, 18]). The reported AP effects among the five Autism-spectrum quotient (AQ) subscales were inconsistent, with some reporting an AP effect on the imagination and attention switching autistic subscales [17, 18], while others found an effect only on the socio-communicative domain [24]. Evidently, systematic investigation based on larger samples is currently needed to evaluate the putatively shared mechanisms underlying AP and autism.

One potential contributing factor to current inconclusive link between AP and autism is the use of discrete rating scales in measuring autistic traits. Notably, previous studies assessing autistic traits in AP individuals has used the Autism spectrum Quotient (AQ) questionnaire [41], which uses four discrete response options for score determination. The AQ questionnaire consists of 50 forced-choice items covering five autistic subscales including: imagination, social skills, communication, imagination, attention to detail, and attention switching using various statements. Participants are required to rate how strongly each statement applies to them via a single forced-choice response with four discrete rating options of, “definitely agree”, “slightly agree”, “slightly disagree” and “definitely disagree.” Scoring is given as either 1 or 0 depending on the directionality of the statement, regardless of the degree of agreement (i.e., “slightly” or “definitely” receive the same score). Importantly, such equal-appearing interval scale design may not capture subtle differences along autistic traits, which may contribute to the small and inconsistent effects observed between AP and autistic traits. In addition, previous comparison is often based on dichotomous groups of AP musicians and RP/non-AP musicians based on some AP cutoff score criteria, leaving those with intermediate levels of “quasi-AP ability” often unexamined. Given that both AP ability and autistic traits are thought to exist on a continuous spectrum in the general population, studying it from the perspective of a continuous scale may provide further insights to the potentially link. Furthermore, a slider version may offer practical advantages, such as being more feasible and accepted to individuals who may find that four response options do not sufficiently capture their experiences. As such, we explored whether a continuous visual slider could provide results comparable to the conventional AQ and potentially reveal subtle variations in autistic traits. By providing a greater range of response options, the slider may facilitate the detection of fine-grained differences along autistic traits that might be less apparent with discrete scales. This may help provide insights into the current inconsistent findings regarding autistic-like traits associated with AP.

The aim of this study is to examine the association between AP ability and autistic traits using two different AQ measurement methods in a relatively large sample of listeners (N = 120) with varying levels of AP proficiency. We compared autistic traits measured using the standard Autism spectrum Quotient questionnaire [41], which employs discrete rating scales, to those measured using a continuous slider across three groups differing in AP abilities: high-AP (≥ 70%), quasi-AP (≥ 20% and < 70%), and low-AP (< 20%). This approach aims to cover the full range (0-100%) of AP proficiency rather than simply categorizing participants into AP versus non-AP (or RP) groups. Additionally, we conducted measurements related to AP abilities including pitch adjustment, musical aptitude, and relative pitch tasks. We expect to see a link between AP ability and autistic traits, with the continuous scale providing comparable results and possibly capturing subtle variations compared to discrete response options. We refrain from making a priori hypotheses regarding which of the five autistic subscales might demonstrate significance, given the inconsistent findings from previous studies.

Methods

Participants

One-hundred and twenty normal-hearing listeners aged 18 to 42 (mean age 23.26 ± 4.35 years; 61 females) participated. All participants were native speakers of Mandarin Chinese with normal hearing (< 20 dB HL pure tone thresholds at octave frequencies 125–8000 Hz; MA 25e; Maico Diagnostics, Minneapolis, USA). Handedness was assessed using the Edinburgh Handedness Inventory [42]. Most of the participants were students at National Central University or enrolled in nearby universities. For the musician group, qualifying participants were required to have commenced formal music training at or before the age of seven (mean age = 5.69 ± 1.39) and continued to take formal music lessons for greater than 7 years (> 7 years of musical activity, and > 4000 h of cumulative lifetime practice). All musicians were currently active in playing a musical instrument for at least 5 h per week for at least one year prior to the study. The non-musician group was defined as individuals with less than 1 year of experience with formal music lessons (except for compulsory, school-based music courses). All participants completed the full version of the Montreal Music History Questionnaire (MMHQ Mandarin version; [43]) in the lab which assessed their demographic background and detailed information regarding musical training. As an assessment of perceptual musical skills, all participants also completed the Advanced Measures of Music Audiation (AMMA) musicality assessment online, which includes tonal and rhythm subtests [44]. Written informed-consent was obtained from each participant in accordance with the Research Ethics Committee at National Taiwan University, Taiwan. All procedures were approved and performed in accordance with the guidelines of the Research Ethics Committee at National Taiwan University, Taiwan. The study was conducted between July 2023 and April 2024.

Auditory stimuli generation and apparatus

All testing was conducted in a double-walled, steel, acoustically isolated chamber (interior dimensions 2.5 × 2.5 × 2 m; Industrial Acoustics Company). Stimuli were generated using MatLab on an ASUS Vento PC and presented at a sampling rate of 44.1 kHz through 16-bit digital-to-analog converters (Creative Sound Blaster X-Fi Titanium). Sounds were presented through Sennheiser headphones (HD380 Pro, Wedemark, Germany) at 70 dBA.

Measures and tasks

AP screening task

We assessed participants’ AP ability using an in-house AP screening task modeled after our previous study [13]. The task included two separate blocks of 50 trials each: one with pure tones and the other with piano notes. Each auditory stimulus lasted 1000 ms with 100 ms rise and decay ramps. In the pure tone block, stimuli were randomly selected from frequencies corresponding to musical notes between C2 and B6 (65.4–1975.5 Hz, equal-tempered scale; A4 = 440 Hz). To avoid immediate pitch comparison, consecutive tones were chosen such that they differed by at least two octaves plus one semitone [3]. After each tone, a 600 ms burst of white Gaussian noise occurred 600 ms later, followed by a 1000 ms response window. The noise served to interfere with echoic memory, and the strict response time aimed to minimize the use relative pitch strategies. Participants were instructed to identify each note by its musical name (e.g., Do or F#). The piano block used the same frequency range and spacing constraints, presenting piano tones instead of pure tones in a separate set of 50 trials. Scoring followed the approach of Baharloo et al. [3], giving 1 point for exact note identification and 0.5 points for responses within one semitone (e.g., identifying C# as C). Unlike Baharloo et al. [3], who assigned 0.75 points for 1-semitone error, we applied a slightly stricter criterion. Octave errors were not penalized, consistent with findings that AP and non-AP individuals do not significantly differ in octave identification [45]. The expected chance-level score on this task was 8.3 points, equivalent to approximately 16.72%.

Pitch adjustment task

We employed a Graphical User Interface (GUI) with a slider-based design to assess participants’ ability to adjust the pitch of a pure tone to match a visually-presented musical note name on the screen (e.g., Do#). The pitch adjustment task provides a finer-resolution (i.e., 1 Hz) measure of AP ability than the AP screening, which relies on categorical naming. On each trial, participants manipulated an unlabeled GUI slider to adjust the pitch of a tone so that it matches the frequency of the target note. The slider covered a fixed range of 3/4 of an octave but was randomly offset relative to the target frequency on each trial. Target notes were randomly chosen from the 2nd through 5th octaves. The range of 3/4 octave, rather than a full octave, ensured only one correct solution per trial. The slider offered two step sizes, 10% and 1% of its range, with a minimum resolution of 1% (approximately 0.09 semitones). During the 30-second adjustment period, participants could press a button on the GUI to listen to the target tone as many times as needed. Once satisfied with their adjustment, they pressed another button to submit their response. At the start of each trial, the slider was reset to the middle of the slider. Each of the 12 musical notes was tested in a randomized order across three adjustment sessions. Performance was quantified as the mean absolute difference between the adjusted tone and the target note in semitone units.

Relative pitch task

Relative pitch (RP) ability was assessed given previous reports that AP ability covaries with RP ability. Relative pitch ability was assessed using a three-interval forced choice design via GUI similar to the procedure used by Hove et al. [46]. The stimuli consisted of piano tones randomly selected from the tonal scale within two sharp/flat notations based on Western music conventions: G-major, F-major, E-minor, and D-minor scales in the 3–5th octave range. The musical intervals were randomly selected from musical intervals comprising major thirds, perfect fifth, and minor seventh. On each trial, one musical interval was randomly selected and presented twice sequentially with an ISI of 300 ms. All tones were 500 ms in duration, with 20 ms linear rise-decay ramps. All intervals consisted of ascending melodic intervals, in which the second tone of each interval was always higher in frequency than the first. The intervals were selected with the constraints that intervals for adjacent trials were from two different tonal scales and differed by more than 2 semitones. The participant’s task was to identify the interval presented by choosing one of three buttons on the screen labeled “3,” “5,” or “7,” with the number indicating the musical interval corresponding to major thirds, perfect fifth, and minor seventh, respectively. Each participant was allowed to practice for 12 trials at most with feedback before completing the test session of 24 trials. No feedback was given at any time during the actual test.

Advanced measures of music audiation (AMMA)

An online graphical user interface (GUI) version (https://giamusicassessment.com) of the AMMA test [44] was used as an objective assessment to assess an adult’s perceptual musical aptitude across tonal and rhythm dimensions. The AMMA consists of 30 same-difference judgment questions, ranging in difficulty level, using four rating options of “different tonal”, “different rhythm”, “same”, or “not sure.” For each trial, the listener heard two melodic sequences via headphones, where the two melodies could be either identical, or contains a change in the rhythm or tonal dimension (but never both). The listener was required to judge whether the two sequences were the same, different, or not sure. In addition, for the “difference” response, the listener had to judge whether the difference occurred for the tonal dimension or rhythm dimension. No feedback is provided at any point. The AMMA test yields a tonal, a rhythm, and a composite score combining tonal and rhythm by giving 1 point for correct judgments (same/different) and 0 point for “not sure” responses. The entire test took approximately 20 min to complete.

Autistic traits assessment

The two versions of the AQ assessment were given to participants during two different visits to the lab, with the AQ slider version administered during the first visit and the standard AQ questionnaire during the second visit. The time between visits ranged from one week to no more than one month.

Adult autism-spectrum quotient (AQ)

The Adult Autism-spectrum Quotient (AQ) questionnaire developed by Baron-Cohen et al. [41] was used to measure autistic traits (Mandarin version; https://docs.autismresearchcentre.com/tests/ AQ_Mandarin.pdf). The AQ questionnaire, though cannot be used as a diagnostic tool by itself, has been validated against clinical diagnosis [47]. The AQ questionnaire consists of 50 questions across 5 subscales of autism: imagination, communication, attention to details, social skills, and attention switching. Each subscale consisted of 10 judgement questions on a 1–2 sentence description of an autistic trait within the subscale. Participants were required to using discrete rating scale of “definitely agree”, “slightly agree”, “slightly disagree”, and “definitely disagree” for each question in a self-paced manner. Half of the questions were designed such that an agree response indicates higher autistic traits and the other half of the questions were designed to elicit a disagree response. For each AQ subscale, raw scores ranged from 0 to 10 points, (i.e., 1 point given for “definitely/slightly agree”, and 0 point given for “definitely/slightly disagree”) depending on the direction of autistic traits. The total AQ score is 50 points (5 subscales × 10 points/subscale), with higher score indicating higher AQ traits. In terms of the five autism subscales, higher AQ scores indicate greater difficulties in imagination, social skills, communication, attention switching and increased attention to details.

Slider adjustment version of the AQ

A Graphical User Interface (GUI) slider design is employed to assess autistic traits on a continuous scale. All 50 questions were identical to the original AQ questionnaire [41] except for the following changes to the response option. Instead of using a 4 discrete categories of response options, a continuous slider that covered the range from 0 to 100% was designed. On each trial, one question from the original ASD questionnaire was presented as text following the same order as in the original questionnaire. Participants were instructed to indicate the response self-paced by using a mouse to adjust the slider icon on a continuous slider ranging from 0% representing “Definitely Disagree” to 100% representing “Definitely Agree.” Upon completion of each question, participants press the NEXT button to answer the next question. The slider icon was initialized at 50% (i.e., neutral) and reset to the middle at the beginning of each trial. All 50 questions were tested following the original ASD questionnaire. For scoring of the AQ slider, the total score was computed by averaging the user-adjusted percentage values across all 50 items. Subscale scores were computed by averaging the adjusted percentage values for the 10 items corresponding to each of the five AQ subscales. As a preliminary validation for the AQ slider version, a correlational analysis was performed between the total score obtained from the standard AQ questionnaire score and the AQ slider version. Figure 1 shows a representation of the GUI slider design used to assess autistic traits.

Fig. 1.

A screenshot of the graphical user interface slider design used to assess autistic traits on a continuous scale. The participant used a mouse to place the slider icon to indicate subjective response (0 to 100%) of each ASD statement at a self-paced rate. All statements were identical to the original ASD questionnaire and presented in the same order as described by Baron-Cohen et al. [41]

Statistical analysis

All statistical analyses were performed using MatLab (2015a, MathWorks, USA) and SPSS 18.0. (IBM). An alpha level of p < .05 (two-tailed) was set a priori to determine statistical significance. A power analysis using the open-source software G*Power (version 3.1; [48]) indicated that with a total sample size of 120 and an alpha level of 0.05, it would yield a statistical power of 0.84 to detect a medium effect size (f = 0.25) in an ANOVA. AP screening scores from each group were assessed for normality using the Shapiro-Wilk’s test, which revealed no evidence of violations of the normality assumption (high-AP group: W = 0.948, p = .063, low-AP group: W = 0.949 p = .072), except for the quasi-AP group (W = 0.939, p = .045). Based on visual inspection of the Q-Q plot and the fact that ANOVA is quite robust against violations of normality, parametric tests were used for statistical analysis. One-way ANOVA with AP group as a between-subject factor was performed for each measurement and task related to AP ability (AQ score, AQ adjustments, pitch adjustment, relative pitch, AMMA). Post-hoc analysis was conducted using t-test with multiple comparisons corrected using Bonferroni corrections (family-wise α = 0.05). Pearson’s correlations were computed to estimate the relationship between autistic traits and musical abilities and musician experience (AMMA, training hours, age of onset of musical training).

Results

Participants’ characteristics

Table 1 summarizes the demographic and musical characteristics of the participants. Based on their AP screening performance, musicians and non-musicians were categorized into three AP proficiency groups: High-AP (≥ 70% accuracy), Quasi-AP (≥ 20% and < 70%), and Low-AP (< 20%; see Fig. 2A). These AP proficiency cutoff criteria align with those commonly used in previous studies [49, 50]. To confirm that the musician groups were comparable in terms of their musical training backgrounds (i.e., only AP ability differs), we conducted independent samples t-tests to examine onset age and musical training hours. No significant differences emerged between the high-AP and quasi-AP groups on either the age at which formal music training commenced (t(78) = 1.96, p = .053) or cumulative training hours (t(78) = − 0.34, p = .74). As expected, the three groups showed significant differences in terms of AP screening score (F(2, 117) = 491.08, p < .001, η_p² = 0.893, Fig. 2B) and pitch adjustment accuracy (F(2, 117) = 122.20, p < .001, η_p² = 0.676). Mean absolute deviation from target tones (MAD; z-transformed) for pitch adjustment correlated significantly with AP screening scores (r = − .790, p < .001; Fig. 2C), validating the AP proficiency level. Post-hoc comparisons with Bonferroni correction (α/3 = 0.0167) revealed significant differences in AP screening scores between all pairs of groups, with the high-AP group scoring higher than the quasi-AP group followed by the low-AP group (all ts > 12.10, ps < 0.001; see Supplementary Table S2). Likewise, pitch adjustment accuracy significantly differed among the groups in the same pattern (all ts > 3.98, ps < 0.001; see Supplementary Table S3). Since AP and relative pitch (RP) abilities are often related, we also examined RP performance using an interval identification task. There was a significant group difference in RP ability (F(2, 117) = 22.24, p < .001, η_p² = 0.252), with the AP groups overall performing above 60% at identifying RP intervals. Bonferroni corrected post-hoc comparisons showed that the high-AP group outperformed both the quasi-AP and low-AP groups (all ts > 4.20, ps < 0.001), while no difference emerged between the quasi-AP and low-AP groups (see Supplementary Table S4).

Table 1.

Participants’ characteristics

	High-AP	Quasi-AP	Low-AP	Statistical tests
No.	40	40	40
Age	23.20 ± 3.98	23.88 ± 5.16	22.93 ± 2.63
Gender (female/male)	18 M/22F	9 M/31F	24 M/16F
Handedness (right/left)	37R/3L	39R/1L	36R/4L
Starting age (yrs)	5.38 ± 1.48	6.00 ± 1.36	N/A	t = 1.96, p = .053
Years of playing (yrs)	15.09 ± 6.01	12.90 ± 4.73	N/A	t = 1.93, p = .057
Cumulative training hours	12,963.95 ± 9196.01	13,617.43 ± 8088.95	N/A	t = − 0.34, p = .74
Primary training style	Classical	Classical	N/A
Primary instrument	Piano	Piano	N/A
AP Screening score (%)a	82.08 ± 8.22	40.30 ± 15.58	9.45 ± 4.16	F = 491.08, p < .001
Score range (max = 100)	(71–96)	(22–69)	(4–17)
Pitch Adjustment Test b	-4.15 ± 1.23	-2.97 ± 1.41	0.00 ± 1.00	F = 122.20, p < .001
Relative Pitch Task (RP)	82.71 ± 17.72	64.09 ± 21.71	55.10 ± 16.88	F = 22.24, p < .001
AMMA Total score	67.08 ± 6.29	60.08 ± 10.51	57.23 ± 8.88	F = 16.91, p < .001
Tonal score	33.55 ± 3.15	29.40 ± 5.71	27.58 ± 4.90	F = 8.42, p < .001
Rhythm score	33.53 ± 3.43	30.68 ± 5.13	29.65 ± 4.41

^aAP group cutoff criteria is based on AP screening score (combined performance on pure tone and piano notes)

^bAccuracy on pitch adjustment test is based on deviation from target-note frequency in units of semitone and normalized in reference to the low-AP group

Fig. 2.

Performance on AP identification and pitch adjustment for the high-AP, quasi-AP, and low-AP groups. A. Cutoff criteria according to scores on the AP proficiency screening task for the high-AP (> 70%), quasi-AP (20–70%), and low-AP (< 20%) groups. B. Mean accuracy on AP identification task for pure tone, piano, and total (combined) for high-AP, quasi-AP, and low-AP groups. C. Relationship between AP screening accuracy and pitch adjustment performance. Precision on pitch adjustment was quantified by the standardized mean absolute deviation (zMAD) from target-note frequency (semitone units). Individual dot represents individual subject. Red: high-AP; Blue: quasi-AP; Gray: low-AP; ***p < .001

Association between AP ability and autistic traits based on conventional AQ score (discrete scale)

First, we assessed whether having AP is associated with higher autistic traits using conventional Autism-spectrum Quotient questionnaire [41], which is based on a discrete scale. The mean total AQ score was higher in the high-AP group (mean ± SD: 24.6 ± 7.5) compared to the quasi-AP (21.7 ± 5.8), and low-AP groups (20.8 ± 4.9). Notably, there were 8 listeners in the high-AP group and 2 listeners in the quasi-AP group that reached an AQ score above 32, the clinically significant cutoff for autism without intellectual disability according to Baron-Cohen et al. [41]. A one-way ANOVA on the total score on the AQ revealed a significant difference between AP groups on AQ score, F(2, 117) = 4.14, p = .018, η_p² = 0.066 (Fig. 3A). Post-hoc analysis revealed that the high-AP group scored significantly higher on AQ score than the low-AP group (t(78) = 2.66, p = .010), whereas no difference exists between the high-AP and quasi-AP group (t(78) = 1.97, p = .053) or between the quasi-AP and low-AP groups (t(78) = 0.68, p = .50). To examine the relationship between AP ability and autistic traits, Pearson correlations were conducted between AQ scores and two measures of AP ability: AP screening accuracy and pitch adjustment precision. A significant positive correlation was found between AQ scores and AP screening accuracy (r(118) = 0.279, p = .002), indicating that individuals with higher AP screening scores had higher autistic traits (Fig. 3B). In addition, a significant negative correlation was revealed between AQ scores and pitch adjustment error (r(118) = ‒0.227, p = .013; Fig. 3B), suggesting that individuals with more precise pitch adjustment (i.e., smaller deviations from the target-note frequency) also exhibited higher AQ scores. Together, these results suggest that listeners with greater AP ability, whether assessed through categorical pitch naming or fine-grained pitch adjustment, are associated with higher autistic traits. Table 2 shows the mean and standard deviation of AQ total score and scores on the 5 AQ subscales for the high-AP, quasi-AP, and low-AP groups. A one-way ANOVA revealed no AP group differences on any of the AQ subscales (all ps > 0.05; see Fig. 3C; Table 2).

Fig. 3.

Mean scores on the AQ and AQ subscales as a function of AP ability. A. Mean AQ score for the high-AP, quasi-AP, and low-AP groups. Individual dot represents individual subjects. B. Correlations between AQ scores and two measures of AP ability: AP screening accuracy (left) and pitch adjustment performance (right). Pitch adjustment performance is indexed by the mean absolute deviation to target-note frequency (zMAD), with smaller deviations indicating higher precision. Scores on the five AQ subscales for the high-AP, quasi-AP, and low-AP groups. Error bar represents ± 1SD. Red: high-AP; Blue: quasi-AP; Gray: low-AP; n.s. not significant

Table 2.

Mean scores on AQ subscales for the high-AP, quasi-AP, and low-AP groups

AQ subscale	High-AP Mean (SD)	Quasi-AP Mean (SD)	Low-AP Mean (SD)	Statistical Test
Social skill (SS)	5.0 (2.3)	3.9 (2.1)	4.0 (2.0)	F(2, 117) = 3.00, p = .053
Attention switching (AS)	5.9 (2.1)	5.4 (2.0)	5.3 (1.9)	F(2, 117) = 0.93, p = .40
Attention to detail (AD)	5.8 (1.6)	5.9 (1.9)	5.3 (2.2)	F(2, 117) = 1.03, p = .36
Communication (CO)	3.9 (2.3)	3.1 (2.0)	2.9 (1.5)	F(2, 117) = 2.95, p = .056
Imagination (IM)	3.9 (2.2)	3.2 (1.6)	3.2 (1.7)	F(2, 117) = 2.12, p = .13
Total	24.6 (7.5)	21.7 (5.8)	20.8 (4.9)	F(2, 117) = 4.14, p = .018*

*Mean and standard deviation (SD) of autism-spectrum quotient (AQ) factor scores based on Baron-Cohen et al. [41] for the high-AP, quasi-AP, and low-AP groups. *p < .05

Association between AP ability and autistic traits based on a continuous scale measurement of AQ

To examine whether there might be a stronger link between AP ability and autistic traits when assessed using a continuous scale, we first performed a validation of percentage autistic traits measured via a slider against the conventional ASD score [41]. A high correlation was obtained between autistic traits measured on a continuous scale (i.e., slider) and discrete scale, r(118) = 0.911, p < .001, validating the use of a continuous scale on assessing autistic traits. This suggests that listeners who score higher on AQ score via conventional AQ score were also likely to score higher on autistic traits via a slider scale. Next, we assessed whether using a continuous scale to measure AQ traits may better capture the link between AP and autistic traits relative to conventional discrete AQ scores. Figure 4A shows the mean autistic traits measure (from 0 to 100%; with 100% indicates highest autistic traits) for the high-AP, quasi-AP, and low-AP groups. A one-way between group ANOVA reveals a significant AP effect on percentage autistic traits as measured via a continuous scale, F(2, 117) = 5.84, p = .004, η_p² = 0.091. Specifically, pair-wise comparisons revealed that the AP group had higher autistic traits than the quasi-AP (t(78) = 2.67, p = .009) and low-AP groups (t(78) = 2.92, p = .005), whereas no difference on autistic traits exists between the quasi-AP and low-AP groups (t(78) = 0.17, p = .87). To assess the relationship between AP ability and autistic traits measured via the continuous slider scale, correlational analyses were performed between AQ slider score and two indices of AP ability (i.e., AP screening accuracy and pitch adjustment precision). The analysis revealed a significant positive correlation between AP screening accuracy and AQ slider scores, r(118) = 0.246, p = .007 (Fig. 4B), with listeners having higher AP screening scores associated with a greater percentage of autistic traits on the slider scale. In addition, higher precision on AP pitch adjustment, as indexed by smaller absolute deviations from user-adjusted to target-note frequency (i.e., smaller errors), was associated with higher percentage of autistic traits, r(118) = ‒0.234, p = .010 (Fig. 4B).

Fig. 4.

Percentage of autistic traits on the AQ and AQ subscales measured on a continuous slider scale as a function of AP ability. A. Percentage of autism traits measured on a continuous scale for the high-AP, quasi-AP, and low-AP groups. B. Correlations between AP screening accuracy (left) and pitch adjustment performance (right) and autistic traits based on percentage slider value. Pitch adjustment performance is indexed by the mean absolute deviation to target-note frequency (zMAD), with smaller deviations indicating higher precision. C. Percentage of autistic traits across the 5 AQ subscales for the three AP groups. Individual dot represents individual subjects. Error bar represents ± 1 S.E.M. **p < .01; *p < .05

Here, the total autistic traits was indexed by the percentage slider adjustment averaged across the 5 subscales of AQ and may have potentially masked contribution from an autistic factor. Thus, we further analyzed whether there’s any difference in AP ability on percentage of autistic traits on each autistic subscale. Compared to the conventional AQ score, a one-way between groups ANOVA based on the continuous slider revealed a significant effect of AP on autistic traits in the social skill (F(2, 117) = 3.41, p = .036), imagination (F(2, 117) = 3.33, p = .039), and communication subscales (F(2, 117) = 3.53, p = .033; Fig. 4C). Post-hoc analysis showed that across all three AQ subscales, this AP effect was mainly contributed by the high-AP group scoring higher than the quasi-AP group, with no significant differences in any other pairwise comparisons (for post-hoc details, see Supplementary Table S5). Consistent with the results obtained using conventional AQ, there was no significant effect of AP ability on the attention switching or attention to details AQ subscales (see Table 3).

Table 3.

Mean percentage of autistic traits across AQ subscales based on a continuous slider scale for the high-AP, quasi-AP, and low-AP groups

AQ subscale	High-AP Mean (SD)	Quasi-AP Mean (SD)	Low-AP Mean (SD)	Statistical Test
Social skill (SS)	50.11 (16.38)	42.42 (13.09)	43.71 (12.55)	F(2, 117) = 3.41, p = .036*
Attention switching (AS)	55.13 (13.53)	51.37 (12.45)	51.35 (12.98)	F(2, 117) = 1.12, p = .33
Attention to detail (AD)	55.38 (11.98)	54.59 (13.05)	49.21 (15.90)	F(2, 117) = 2.3o, p = .096
Communication (CO)	41.84 (14.71)	34.11 (14.45)	36.41 (10.51)	F(2, 117) = 3.53, p = .033*
Imagination (IM)	42.99 (14.86)	36.57 (9.10)	37.07 (12.71)	F(2, 117) = 3.33, p = .039*
Total	49.08 (9.99)	43.81 (7.53)	43.54 (6.63)	F(2, 117) = 5.84, p = .004**

*Mean and standard deviation (SD) of percentage autistic traits on the AQ subscales for the high-AP, quasi-AP, and low-AP groups. **p < .01; *p < .05

Apart from absolute pitch, prior literature has reported equal or superior musical skills for processing pitch, melody, and rhythm in individuals with autism [51, 52]. As such, we examine whether better perceptual processing ability to process music-related aspects of tonal and rhythm, as assessed by the musical aptitude test AMMA, is associated with higher autistic traits. Correlational analysis on the relationship between AMMA score and percentage total autistic traits revealed a significant correlation between autistic traits and tonal skill (r(118) = 0.186, p = .042), but not rhythm skill (r(118) = 0.129, p = .16; Fig. 5A). Specifically, regardless of musician experience, listeners with better tonal perception ability were associated with higher autistic traits.

Fig. 5.

Relationship between musical experience and autistic traits. A. Scatterplots depict correlation between tonal (left panel) and rhythm (right panel) ability (indexed by tonal score on AMMA) and percentage autistic traits obtained via the visual slider (B) Cumulative music training hours (computed from MMHQ) and percentage autistic traits (C) Age of onset of music training (obtained from MMHQ) and percentage autistic traits. Note that the low-AP group (gray dots) were excluded from analyses in B and C due to non-applicable training hours. Each dot represents individual subject. Red: high-AP group; Blue: quasi-AP group; Gray: low-AP group; **p < .01, *p < .05 level of significance

Relationship between musical experience and autistic traits

In addition, we assess whether musical training experience, a prevalent feature associated with most AP musicians, is linked to higher autistic traits. Musical training experience is indexed by the cumulative training hours and the age of onset at which music training began, as computed from MMHQ. The analysis revealed no significant correlation between either cumulative music training hours (r = − .050, p = .66; Fig. 5B) or age of onset of music training (r = .173, p = .125; Fig. 5C) with mean percentage autistic traits across all autistic subscales.

Discussion

Using a continuous slider to detect subtle differences along autistic traits, we found evidence of an association between AP ability and autistic traits in a large cohort of musicians and non-musicians varying in AP proficiency. To our knowledge, this is the first study to employ a continuous scale to assess autistic traits with finer resolution, ranging from 0 to 100%, differing from previous discrete options of the conventional AQ questionnaire [41]. Additionally, we employed a relatively large cohort of musicians and non-musicians with AP ability spanning a wide spectrum of proficiency (0-100%) to include individuals with intermediate levels of AP (i.e., the quasi-AP group). While AP ability was not significantly linked to any of the AQ subscales when measured with the conventional AQ questionnaire, significant associations with three autistic subscales emerged when assessed using a continuous slider. Specifically, listeners with higher AP proficiency scored higher on the imagination subscale in addition to the social and communication subscales. Compared to previous studies reporting non-significant associations between AP and autistic traits, typically with small to moderate effect sizes (e.g., p = .058, η² = 0.15; p = .081, d = − 0.44; [17, 18]), the current finding using a continuous slider revealed a statistically significant association between AP ability and AO slider scores, with a medium effect size (p = .004; η_p² = 0.091). This suggests that the slider method may capture trait differences more sensitive than traditional discrete measures, though further psychometric validation is warranted. These results were further corroborated by a significant correlation between the percentage of autistic traits and AP pitch adjustment, an index of AP ability with a finer resolution. Additionally, the AP-related enhancement in autistic traits remained regardless of changes in cutoff criteria or the use of a continuous measure of AP proficiency to determine AP status (details in Supplementary Fig. 1). Apart from AP ability, we found that better music-perception skills in processing tonal, but not rhythm materials, were associated with higher autistic traits. However, cumulative music training experience and age of onset of music training did not predict autistic traits. Overall, these findings suggest an association between AP ability and autistic traits that is observed consistently across both the slider and the conventional AQ questionnaire, highlighting the need for future exploration on the potential use of the slider version.

Unlike previous studies that applied equal interval rating to assess autistic traits, the present study utilized a continuous slider to examine the putative link between AP and autism. Our findings suggest that a continuous scaling method can capture associations in a manner comparable to a fixed rating scale. Specifically, while we did not obtain any significant differences among AP groups on the AQ subscales based on the conventional AQ questionnaire [41], differences in the imagination, social skill, and communication AQ factors emerged when a continuous slider scale was used. Discrete and continuous rating scales have been employed in psychology and related fields to assess perception of various stimuli (e.g., pain, speech naturalness), sometimes yielding divergent results [53]. To our knowledge, it has not been applied in the context of assessing autistic traits. Previous studies on AP and autistic traits have typically observed small or non-significant effects [17, 18], in addition to the inconsistent association revealed with respect to the subscale of AQ [20, 24]. One speculation for the inconsistent findings may be that the conventional AQ questionnaire relied on discrete options, which may not capture subtle differences along autistic traits. That is, answering “totally agree” and “slightly agree” are judged as identical in terms of scoring, potentially obscuring differences between the degrees of agreement. Notably, although we obtained a strong correlation between the visual slider and the traditional AQ score (r = .911), it is important to acknowledge that the AQ slider lacks formal psychometric assessment. Further validation, such as assessing inter- and intra- rater agreement and comparing with other AQ scoring procedures, is necessary. For instance, the scoring procedure developed by Austin [54] assigns 0 to 3 points to the four discrete response options on standard AQ, yielding a total score ranging from 0 to 150 points and providing a more nuanced measurement. Comparing and validating the AQ slider against these and other established AQ rating approaches across studies would allow for more comprehensive assessment of its utility. Taken together, while the use of a continuous scale, as shown here, may offer a viable method for better detection of the degree of agreement or disagreement, further psychometric validation is warranted.

Our results provide further support to the existing inconclusive link between AP and autistic traits, suggesting heightened autistic traits across the imagination, social skills, and communication domains as assessed by both discrete and continuous measures. Specifically, our findings corroborate prior studies suggesting that AP is associated with elevated levels of autistic traits, particularly those related to differences in social and communication skills typically associated with autism [24]. One speculation is that heightened sensitivity to pitch in AP individuals, similar to those observed in autistic individuals [35, 55], could potentially overwhelm them in social situations, making it difficult to filter and process social cues effectively. This may lead to challenges in social interaction and communication, particularly in noisy environments. Indeed, sensory overload, particularly concerning auditory stimuli, is commonly reported in autistic individuals, affecting higher-order functions such as social communication and emotional regulation [56, 57]. Another potential reason contributing to atypical socio-communication skills may be related to multisensory integration ability. Specifically, altered multisensory integration ability has been suggested to contribute to differences in social communication in autistic individuals, especially when the stimuli involve social information (e.g., human faces) [56, 58]. Whether AP ability also affects multisensory integration requires further investigation, particular with respect to social versus non-social information Additionally, we report that individuals with higher AP ability tend to score higher on the imagination subscale relative to low-AP individuals, suggesting less imagination, which is consistent with previous studies [17, 18]). One explanation could be that intense focus on precise pitch perception by AP listeners leads to a preference for concrete, factual information over abstract or imaginative thinking. In line with this notion, Chin’s two-factor model of AP has proposed an analytical cognitive style with reduced flexibility that scaffolds the development of AP, potentially contributing to a tendency for less imagination [35, 39]. Further research is needed to delineate the precise relationship between analytical cognitive style and imagination in individuals with AP using different tasks, particularly during critical developmental periods.

Our results did not support the detail-oriented cognitive style often speculated to accompany the co-occurrence of AP and autism. Specifically, we did not find that AP musicians exhibited heightened autistic traits on the attention-switching and attention-to-details subscales, consistent with previous findings [17, 18]. However, our results contradict studies that have reported an enhanced ability to process embedded figures or interleaved melodies among individuals with higher AP or autistic traits [12, 14, 25, 27]. Recent neural evidence further supports a shift toward higher segregation with reduced integration in AP individuals, demonstrated by diminished connectivity in both structural and functional networks among AP possessors [18, 28, 29]. One possible explanation for such discrepancies, and a limitation of the present study, is that the enhanced attention to details in AP individuals might be context-dependent. The current findings, indicating a lack of a local tendency in AP musicians, were based on discrete and continuous self-rated responses on the AQ. It is plausible that the preference for local details in AP individuals is more observable in tasks directly involving low-level sensory information such as musical pitches. Indeed, Wenhart and Altenmuller [25] demonstrated that higher AP ability was linked to reduced global-to-local interference in hierarchically structured melodies, but this effect was limited to the auditory domain, not visual. These findings collectively suggest that the autistic trait of attention to details in AP possessors may be specific to tasks involving pitch-related materials and not as observable in self-rated measures. Another contributing factor is that individual variabilities, such as the heterogeneity commonly observed within AP musicians, could potentially mediate the relationship between AP and an autistic-like cognitive style (e.g., attention to details). For instance, a recent study has demonstrated that individual variabilities in cognitive abilities such as working memory play an important role in mediating the relationship between autism and altered pitch perception [59]. This suggests that relevant cognitive abilities that likely vary within AP individuals could potentially modulate the relationship between AP and autistic traits.

In this study, we observed a relatively high estimated prevalence of participants who scored above the AQ cut-off in our cohort of Mandarin-speaking AP musicians. Specifically, eight participants in the high-AP group and two in the quasi-AP group scored above 32 on the AQ, the clinically significant cutoff for autism without intellectual disability according to Baron-Cohen et al. [41], though we note that these participants have not been assessed in relation to the diagnostic criteria outline in DSM-IV-TR. Conversely, no participants in the non-musician group exceeded this cutoff. Among the ten individuals surpassing the cutoff, two had received official autism diagnoses, with one having an exceptionally high AP screening score of 97 (details in Supplementary Table S6). The estimated prevalence of participants who score above the cutoff of 32 based on self-reported autistic traits on the AQ (12.5%; 10 out of 80 musicians) in the present study was comparatively higher compared to other studies (e.g., 1/16 in [17]; 0/31 in [18]). Specifically, a chi-square test of independence was performed to examine the relationship between AP category and meeting the cut-off for autism without intellectual disability on the AQ. The association was significant, χ²(2, N = 120) = 11.35, p = .003, indicating that individuals with higher AP ability were more likely to meet the cut-off for autism without intellectual disability according to Baron-Cohen et al. [41]. One speculation might be that tone-language expertise (i.e., all participants were native Mandarin Chinese speakers), with its emphasis on pitch contour, contributed to the observed increased prevalence of clinically significant autistic traits. While tone language experience has been linked to a higher prevalence of AP in tone-language countries [5, 7] its relationship with autism or subclinical autistic traits is currently unclear. In terms of the prevalence of autism across countries, a recent meta-analysis has identified the highest ASD prevalence in North America, followed by Europe, and the lowest in Taiwan [60]). While this may suggest that autism could be lower in tone language speakers (e.g., Taiwan) compared to non-tone language speakers (e.g., North America), other factors such as methodological issues, socio-economic factors, and culturally specific social cognitive styles could contribute to variations in ASD prevalence [61]. Future studies can compare autistic traits or performance on tasks assessing autistic traits among AP individuals who are speakers of tone or non-tone languages to gain further insights into the relationship between tone-language and autistic traits.

Limitations and suggestions

Although this study aimed to capture the whole range of AP ability from 0 to 100% AP proficiency within our sample of 120 musician and non-musician participants, several limitations should be acknowledged. The primary limitation is the use of a visual analog slider to assess AQ, which, at its current stage, lacks psychometric assessments. Further validation such as inter- and intra- rater agreement and comparisons with other established methods of AQ scoring approaches are warranted. Second, the demographics of our participants are relatively homogeneous. This study intentionally recruited individuals with similar music and educational backgrounds (e.g., most were university students) to ensure that differences observed were primarily due to AP ability. However, this homogeneity limits the generalizability of our findings to other populations, such as children and older adults, who may also possess AP and exhibit different relationships with autistic traits. It would be interesting in future research to examine how the link between AP ability and autistic traits changes across the life span. Additionally, the primary instruments for all musicians in our study was the piano. Therefore, the findings may be specific to pianists and not generalizable to musicians who play other instruments. Indeed, previous research suggests that AP ability is stronger for an individual’s primary instrument [62] and this may affect the interpretation of the present results.

Additionally, there are methodological concerns regarding the administration and use of the two AQ measures. This study assesses autistic traits using self-reported data, either through traditional questionnaire or a visual slider, which may limit the objectivity of the findings. For instance, participants’ own subjective feelings, motivations and interpretation of their behavioral tendencies may bias their responses. Moreover, the order in which the two version of the AQ were administered, in which the slider version of the AQ was always given first, may have introduce potential order effects. Although we tried to space the two visits by at least one week (and no more than one month), participants may have become familiar with the type of questions asked, which could potentially affect their responses on the second visit. This possible carry over effects should be considered when interpreting the results. Lastly, there are some statistical concerns, as a few p-values were just below 0.05, and some significance tests (e.g., the AQ subscales) were conducted without correction for multiple comparisons. Therefore, the risk of type I errors should be considered when interpreting the results.

Conclusion

In summary, our results support an association between AP ability and autistic traits in a large sample, with comparable results captured by both continuous measurement and the conventional AQ questionnaire. The utilization of a continuous rating scale aligns with outcomes derived from conventional AQ questionnaires, indicating that AP ability is predominantly linked with autistic traits in social and communication domains. Employing a continuous scale reveals an additional link to imagination in AP musicians. AP possessors exhibit superior tonal and rhythm skills, with only pitch-related skills predicting autistic traits. These findings underscore the utility of a continuous scale in evaluating variations across cognitive traits, offering deeper insights into the potential shared and/or distinct neurocognitive mechanisms underlying savant abilities and autism.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AD

Attention to Detail

AMMA

Advanced Measures of Music Audiation

AP

Absolute Pitch

AQ

Autism Spectrum Quotient

AS

Attention Switching

ASD

Autism Spectrum Disorder

CO

Communication

DSM

Diagnostic and Statistical Manual of Mental Disorders

EEG

Electroencephalography

GUI

Graphical User Interface

IM

Imagination

MMHQ

Montreal Music History Questionnaire

RP

Relative Pitch

SD

Standard Deviation

SEM

Standard Error of the Mean

SS

Social Skill

MAD

Mean Absolute Deviation

Author contributions

I. H.: conceptualization, investigation, methodology, formal analysis, visualization, manuscript writing; Z.H.: investigation, data collection, formal analysis, figure preparation; All authors reviewed the manuscript.

Funding

This research was funded by the National Science and Technology Council, Taiwan, Grant Number: NSTC112-2410-H-008-065-MY2 to I.H.H.

Data availability

The datasets generated during and/or analysed during the current study are available in the Open Science Framework repository [https://osf.io/u7ze4/](https:/osf.io/u7ze4).

Declarations

Ethics approval and consent to participant

All procedures performed in the study involving human participants were in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. All experiment protocols were approved and performed in accordance with the guidelines of the Research Ethics Committee at National Taiwan University, Taiwan (NTU-REC No. 202105EM029). Informed consent was obtained from all individual participants included in the study.

Consent for publication

The results presented in this manuscript have not been published elsewhere, nor are they under consideration (from any of the authors) by another publisher.

Competing interests

The authors declare no competing interests.