Repetitive Behaviours in Williams Syndrome: A Cross‐Cultural Comparison Between the United Kingdom and Japan

ABSTRACT

Background

Williams syndrome (WS) is a relatively rare neurodevelopmental condition characterised by distinctive cognitive and behavioural phenotypes, including restricted and repetitive behaviours (RRBs). Although cross‐cultural studies suggest that caregiver reports of autism‐related RRBs may be culturally subjective, little is known about whether caregiver‐reported RRB profiles in WS are similar across cultural contexts. Additionally, because RRB profiles may vary with age, examining cross‐sectional age‐related patterns is important. This study explored between‐country variation and cross‐sectional age‐related patterns in caregiver‐reported Repetitive Behaviour Questionnaire (RBQ) scores among individuals with WS in the United Kingdom and Japan.

Methods

Eighty primary caregivers of individuals with WS from Japan (n = 40) and the United Kingdom (n = 40) completed the Repetitive Behaviour Questionnaire (RBQ). Bayesian negative binomial regression models were used to examine between‐country differences and cross‐sectional age‐related associations in caregiver‐reported RBQ total, sensory/motor and sameness/circumscribed interests scores.

Results

UK caregivers reported directionally higher Total RBQ and Sameness/Circumscribed Interests scores than Japanese caregivers, with the clearest contrast observed for the Sameness/Circumscribed Interests domain. No clear evidence of a between‐country difference was found for caregiver‐reported Sensory/Motor Behaviours scores, although a weak directional tendency toward higher UK scores was observed. Moreover, age showed weak cross‐sectional negative trends for total RBQ and Sensory/Motor Behaviours scores, although these estimates were imprecise and were attenuated in the verbal mental age‐adjusted analyses.

Conclusions

These findings are consistent with previous studies of caregiver‐reported RRBs among autistic individuals, suggesting that insistence on sameness and circumscribed interests may be reported differently across cultural contexts. In WS, caregiver‐reported RBQ profiles may vary by cultural context and show tentative cross‐sectional age‐related patterns, highlighting the need to consider contextual and developmental factors when interpreting parent‐report measures of these behaviours.

1. Background

Williams syndrome (WS) is a relatively rare genetically determined neurodevelopmental condition caused by the microdeletion of approximately 28 genes in one copy of chromosome 7 (Meyer‐Lindenberg et al. ²⁰⁰⁶; Zitzer‐Comfort et al. ²⁰⁰⁷). Its prevalence ranges from 1 in 7500 to 1 in 20 000 (Meyer‐Lindenberg et al. ²⁰⁰⁶; Strømme et al. ²⁰⁰²). WS is associated with distinctive social, cognitive and behavioural phenotypes, which have been extensively studied. For example, individuals with WS tend to exhibit hypersociability (Jones et al. 2000), characterised by a heightened interest in interacting with both familiar and unfamiliar people (Doyle et al. 2004; Frigerio et al. ²⁰⁰⁶; Järvinen‐Pasley et al. ²⁰¹⁰; Porter et al. ²⁰⁰⁷; Riby and Hancock ²⁰⁰⁸; Riby et al. ²⁰¹⁴). Sensory processing differences are also common (Glod et al. 2020; Hirai et al. ²⁰²⁵; John and Mervis ²⁰¹⁰) alongside a propensity toward heightened anxiety (Leyfer et al. 2006; Stinton et al. ²⁰¹²), compulsive behaviours (Semel and Rosner 2003) and restricted and repetitive behaviours (RRBs; Rodgers et al. ²⁰¹²). Repetitive and stereotypical movements have been reported in 86% of individuals with WS (Davies et al. 1998) and can be related to sensory processing (Riby et al. 2013) as well as an intolerance of uncertainty (Glod et al. 2019). Moreover, Huston et al. (2022) examined repetitive thoughts and behaviours in a sample of individuals with WS using structured clinical interviews and standardised scales, reporting a broad range of repetitive phenomena in this population. Collectively, these findings underscore the clinical relevance of characterising RRBs in WS.

Behavioural phenotypes can be influenced by genetic and environmental factors, including cultural norms (Chiao 2018). Research on how culture influences the expression of WS‐associated phenotypes remains limited. To date, only one cross‐cultural study has examined whether cultural norms influence how the social phenotype in WS arises from specific genetic underpinnings (Zitzer‐Comfort et al. ²⁰⁰⁷). The study highlights cross‐cultural differences in global sociability associated with WS in Japan and the United States. These findings are expected, given that differences in Japanese and US social and cultural norms create sociability differences in the general populations of both countries. Therefore, it remains unclear whether culture shapes behavioural phenotypes in nonsocial domains, such as the RRBs that manifest in individuals with WS, as it does in the social‐cognitive domain. Existing research regarding other associated neurodevelopmental conditions sheds light on this research question. A recent cross‐cultural study of autism across five countries (Greece, Italy, Japan, Poland and the United States) evaluated nonverbal communication/socialisation, repetitive behaviour/restricted interests and communication based on caregivers' responses (Matson et al. 2017). The results indicate cultural variation in caregiver‐reported RRB impact, with more impactful repetitive behaviours and restricted interests reported by caregivers in Poland and the United States compared with Greece and Italy, as well as greater cultural variation in caregiver‐reported RRB profiles than in social and communication profiles. Matson et al. (2017) suggest that behaviours related to socialisation and communication may be universal, whereas caregivers' interpretations of RRBs may be more culturally subjective. Therefore, cultural norms may shape parents' interpretation of RRBs associated with WS. Investigating this relationship can contribute to our understanding of this relatively rare condition and inform support strategies for individuals with WS.

Examining cross‐cultural differences in RRBs has important clinical implications. When behavioural assessment tools are applied across cultural contexts without adequate standardisation, measurement bias may be introduced, potentially affecting diagnostic decision‐making and intervention planning for individuals with WS (see also, Rogler ¹⁹⁹³). The present study focuses on the United Kingdom and Japan as comparison groups for several reasons. First, the only prior cross‐cultural study of behavioural phenotypes in WS compared Japanese and US samples (Zitzer‐Comfort et al. ²⁰⁰⁷), positioning Japan as a natural context for extending this line of inquiry to a new cultural comparison. Second, an established body of WS research and normative Repetitive Behaviour Questionnaire (RBQ) data exists in the United Kingdom (e.g., Riby et al. ²⁰¹³; Rodgers et al. ²⁰¹²), providing a well‐characterised reference group. Third, Japan and the United Kingdom represent meaningfully contrasting cultural environments with respect to norms around behavioural conformity, collectivism versus individualism and parental reporting styles—dimensions that may influence the expression, perception and caregiver reporting of RRBs in neurodevelopmental conditions.

Beyond between‐country differences, age‐related patterns in caregiver‐reported RRB profiles among individuals with WS are also important. Previous studies have reported age‐related decreases in social responsiveness among Japanese individuals with WS using the SRS‐2 (Hirai et al. 2024) and developmental reductions in temper tantrums in individuals with WS (Rice et al. 2015). In contrast, sensory‐profile findings suggest that age‐related reductions in sensory sensitivity are not consistently observed in individuals with WS (Hirai et al. 2025). These findings raise the possibility that different components of caregiver‐reported RBQ scores may show different cross‐sectional age‐related patterns.

The current study aimed to examine (1) whether caregiver‐reported RBQ profiles associated with WS differ between the United Kingdom and Japan and (2) whether these caregiver‐reported profiles show cross‐sectional age‐related associations. This aim is addressed using the RBQ (Turner 1995, 1999), which measures the prevalence, frequency and duration of repetitive behaviours. We hypothesised that caregiver‐reported RBQ scores associated with WS would differ between the United Kingdom and Japan, mirroring suggestions from studies of other neurodevelopmental conditions such as autism. Moreover, because prior work has reported age‐related decreases in social responsiveness (Hirai et al. 2024) and temper tantrums (Rice et al. 2015), whereas sensory findings appear less consistent in WS (Hirai et al. 2025), we further hypothesised that caregiver‐reported RBQ scores might show cross‐sectional age‐related decreases.

2. Method

2.1. Participants

The participants were the primary caregivers of 80 individuals with WS who were aged 3–42 years (M = 10.3 years, SD = 7.41; 34 males, 46 females). Of these, 40 caregivers were from Japan and 40 from the United Kingdom, with the samples matched according to the sex and age of the individuals with WS [Japan: 16 males, 24 females, M = 10.1 years, SD = 7.8; the United Kingdom: 18 males, 22 females, M = 10.5 years, SD = 7.0; age‐matched by country, t(78) = 0.23, p = 0.82] (Table 1). Participants were recruited through WS‐specific support and community routes in both countries: via the Williams Syndrome Foundation (WSF) and family‐based WS support networks in the United Kingdom, and at a music camp organised by the non‐profit organisation Smirhythm in Japan. In the United Kingdom, molecular genetic confirmation of WS is required for WSF membership. However, because recruitment also occurred through family‐based WS support networks and the present study records did not include individual‐level diagnostic documentation or recruitment‐route information for each participant, the exact number of UK participants with molecular genetic confirmation could not be verified. In Japan, participants were recruited through a music camp for individuals with WS organised by Smirhythm. Although participation in the camp involved administrative confirmation of disability status, detailed information regarding the specific diagnostic procedures for WS, including genetic testing, was not available for all participants. Given the rarity of WS, the sample size was determined pragmatically on the basis of recruitment feasibility. Although the achieved sample permitted exploratory cross‐cultural comparison, it was not large enough to estimate small‐to‐moderate between‐country effects or interaction effects with high precision. This study was performed in line with the principles of the Declaration of Helsinki. Ethical approval for the UK sample was obtained from the School of Psychology Ethics Committee at Newcastle University. Ethical approval for the Japanese sample was obtained from the Ethics Committee of Jichi Medical University. Written informed consent was obtained from all participants and/or their parents or legal guardians, as appropriate, prior to participation. Data were collected in 2011 and 2012 in the United Kingdom and in 2016 in Japan.

TABLE 1.

Descriptive statistics of participants' characteristics.

Measure	UK (n = 40) M (SD)	Japan (n = 40) M (SD)
Number of male/female participants	18/22	16/24
Chronological age (years)	10.5 (7.0)	10.1 (7.8)
Age range (years)	4–39	3–42
Verbal mental age (months)	122.6 (24.0) a n = 26 b	122.5 (46.9) c n = 16 b
RCPM scores	15.5 (3.55) n = 26	14.5 (4.27) n = 16

Abbreviations: M, mean; n, number of participants; RCPM, Raven's Coloured Progressive Matrices; VMA, verbal mental age; WS, Williams syndrome; SD, standard deviation.

^a British Picture Vocabulary Scale‐II.

^b Number of participants with available VMA and RCPM data.

^c PVT‐R, Picture Vocabulary Test‐Revised.

2.2. Materials

Participants completed the RBQ (Turner 1995, 1999), a parent‐report measure designed to assess the frequency, severity and range of repetitive behaviours. The original RBQ was selected to enable direct comparison with prior UK studies of RRBs in individuals with WS that used this measure (Riby et al. 2013; Rodgers et al. ²⁰¹²). Revised versions of the RBQ (e.g., RBQ‐2 and RBQ‐3) were developed primarily for use with autistic populations and differ in item content and factor structure from the original; using the original RBQ therefore ensured consistency with the existing WS literature. The RBQ version used in the present study comprised 30 scored items. The Total RBQ score was calculated as the sum of Items 1–30, yielding a possible range of 0–60 after recoding scores of 3 to 2 where applicable. Factor‐derived subscale scores were calculated following the two‐factor scoring approach reported by Honey et al. (2012). Sensory/Motor Behaviours was calculated from 12 items (Items 1–8, 10–12 and 28; score range 0–24), and Insistence on Sameness/Circumscribed Interests was calculated from 15 items (Items 6, 14–26 and 29; score range 0–30), with Item 6 contributing to both subscale scores.

The same 3‐to‐2 recoding was applied to the factor‐derived subscale scores where applicable, following Honey et al. (2012). Items 9, 13, 27 and 30 did not load onto either factor in the validation study and were therefore not included in the subscale scores; however, these items contributed to the Total RBQ score. Accordingly, the Total RBQ score was calculated independently of the two subscales and was not the sum of the Sensory/Motor Behaviours and Insistence on Sameness/Circumscribed Interests scores.

The RBQ demonstrates excellent psychometric properties, including high interrater reliability (κ = 0.99) and test–retest reliability (M = 0.83; Turner ¹⁹⁹⁹). For the Japanese sample, a back‐translated version was used (by MH and KA with support from JR), following standard back‐translation procedures to support conceptual and linguistic equivalence. Verbal mental age (VMA) data were not available for all participants because the vocabulary assessments were obtained only when additional cognitive testing formed part of the original data‐collection protocol and was feasible within each setting. The VMA subset was defined by availability of auxiliary cognitive assessments rather than by RBQ scores or prespecified behavioural criteria. Consequently, VMA data were available for a subset of 42 participants (the United Kingdom: n = 26; Japan: n = 16). In the United Kingdom, VMA was assessed using the British Picture Vocabulary Scale, 2nd Edition (BPVS‐II; Dunn et al. ¹⁹⁹⁷). In Japan, VMA was assessed using the Picture Vocabulary Test‐Revised (PVT‐R; Ueno et al. ²⁰⁰⁸). These data were used only in sensitivity analyses. Nonverbal ability was also assessed using Raven's Coloured Progressive Matrices (RCPM) in the subset for whom auxiliary cognitive assessments were available; these scores are reported descriptively in Table 1 but were not included in the primary or sensitivity models.

2.3. Procedure

In the United Kingdom, participants were recruited via the WSF and family‐based WS support networks. Study announcements and flyers were distributed through these channels, and families opted into participation. These data were collected as part of a larger programme of research; recruitment materials informed families that one component of the study concerned RRBs. Questionnaires were administered in paper format and sent to families by post. The accompanying participant information sheet included contact details for the research team, so that families could seek clarification on any questionnaire items if needed. Completed questionnaires were returned by post to the research team in the United Kingdom. In Japan, participants were recruited at a music camp organised by the non‐profit organisation Smirhythm. Prior to the camp, parents were informed that they would be asked to complete a questionnaire concerning their child's characteristics. At the camp, parents were provided with the paper version of the RBQ and completed it on site. Research team members were present during questionnaire completion and could answer procedural questions; however, they did not provide interpretive guidance on how specific behaviours should be rated. For individuals with WS for whom additional assessment was feasible, RCPM (Raven et al. 1990; Sugishita and Yamazaki ¹⁹⁹³) and the PVT‐R (Ueno et al. 2008) were administered on site during the camp. Completed questionnaires and assessment data were collected by the research team in Japan.

2.4. Data Analysis

All statistical analyses were conducted using R (version 4.4.2) with the brms package (version 2.22.0) for Bayesian regression modelling. Parts of the data analysis code were developed with the assistance of large language models (ChatGPT [OpenAI] and Claude [Anthropic]). All code was reviewed, validated and executed by the authors, and the authors take full responsibility for the analysis and its results.

2.4.1. Data Preparation

Prior to analysis, we examined the distributional properties of the RBQ scores to guide model selection. Variance‐to‐mean ratios (VMR) revealed substantial overdispersion across all three outcome measures (Total = 6.29, Sensory/Motor Behaviours = 2.81 and Sameness/Circumscribed Interests = 4.28), violating the equidispersion assumption of Poisson regression and motivating the use of alternative discrete‐response models. The proportion of zero scores was relatively low across outcomes (5%–11%), indicating that zero‐inflated models would not offer added value over standard overdispersed models for these nonnegative integer‐valued RBQ sum scores.

To improve model convergence and facilitate interpretation, age was centred at the sample mean (10.31 years) and scaled by dividing by 10, such that a one‐unit change represented a decade of age. Country was contrast‐coded with Japan serving as the reference category, allowing for direct interpretation of UK effects relative to the Japanese baseline.

2.4.2. Statistical Models

Given the overdispersed, nonnegative integer‐valued nature of the RBQ sum scores, we employed single‐level Bayesian generalised linear models using negative binomial regression (NB2 parameterisation and log link) as our primary analytic approach. Although RBQ scores are bounded sums of ordinal item responses rather than unbounded event counts, the negative binomial model was used as a pragmatic approach to accommodate overdispersion in the observed score distributions. The negative binomial distribution extends the Poisson by incorporating an additional shape parameter to account for extra‐Poisson variation; in the present study, it was used as a pragmatic model that provided an empirically adequate approximation to the overdispersed RBQ sum‐score distributions. The main model included Country (Japan as reference; UK vs. Japan), Age (mean‐centred and scaled in decades) and their interaction as predictors of each RBQ outcome (Total, Sensory/Motor Behaviours and Sameness/Circumscribed Interests).

We specified weakly informative priors to provide modest regularisation while allowing the observed data to drive posterior inference. The intercept prior was a normal distribution centred on the log of the observed mean for each outcome with a standard deviation of 1 (computed after centring Age), reflecting reasonable uncertainty about baseline rates while keeping predictions plausible. Regression coefficients were assigned normal priors centred at 0 with a standard deviation of 0.3, expressing mild scepticism about large effects while remaining open to meaningful associations. The shape parameter controlling overdispersion received an Exponential (1) prior, which favours smaller φ (i.e., allows substantial overdispersion) but permits larger φ if supported by the data. Sensitivity checks with Half‐Normal (0,1) or Gamma (2,1) priors yielded robust conclusions.

2.4.3. Model Fitting and Diagnostics

Models were estimated using Hamiltonian Monte Carlo with four chains of 4000 iterations each, including 1000 warmup iterations. To ensure adequate exploration of the posterior distribution, particularly in the tails, we set the adapt_delta parameter to 0.98 (and increased max_treedepth to 12, when necessary), providing more conservative step‐size adaptation and reducing the risk of divergent transitions. Convergence was evaluated with standard diagnostics: all R‐hat values were below 1.01, indicating excellent mixing across chains, and effective sample sizes exceeded 1000 for all parameters, suggesting adequate posterior sampling. Visual inspection of trace plots confirmed good mixing and stationarity. Notably, no divergent transitions were observed during sampling, suggesting adequate exploration of the posterior geometry.

Model adequacy was evaluated through posterior predictive checks, comparing observed data distribution with distributions generated from the posterior predictive distribution in terms of overall density patterns, measures of central tendency and dispersion and extreme values such as maxima. Additionally, predictive accuracy was further evaluated using leave‐one‐out cross‐validation with moment matching. These assessments indicated that the negative binomial specification provided a better fit than the standard Poisson models, which failed to accommodate overdispersion. The zero‐inflated negative binomial models provided no meaningful improvement over the negative binomial models, consistent with the low prevalence of zero scores; therefore, the simpler negative binomial specification was retained.

2.4.4. Sensitivity Analyses

Recognising that VMA could confound cross‐cultural comparisons, we conducted sensitivity analyses to evaluate the robustness of our findings. For the subset with complete VMA data (n = 42), models were extended to include standardised VMA and its interaction with country. The standardisation of VMA improved numerical stability and facilitated interpretation of VMA effects, allowing us to examine whether the between‐country contrasts in caregiver‐reported RBQ scores remained in the same direction after accounting for potential differences in verbal development between the two samples. To examine whether child sex influenced RBQ scores or accounted for any cross‐cultural differences, we additionally fitted sex sensitivity models in which child sex (coded as a binary variable: male vs. female, with female as the reference category; hereafter, MF) was included as an additive covariate alongside the primary predictors. Specifically, the model took the form Outcome ~ Country × Age10 + MF, with Japan as the reference category for Country, and Age10 denotes age mean‐centred and scaled per decade (as described in Section 2.4.1). An exploratory model incorporating a two‐way interaction (Country × MF) was also fitted to assess whether the magnitude of between‐country differences in caregiver‐reported RBQ scores varied by child sex.

2.4.5. Effect Size Reporting

Results are presented as incidence rate ratios (IRRs; more precisely, multiplicative ratios of the modelled expected RBQ sum score) with 95% credible intervals, providing an intuitive interpretation of effect sizes on the multiplicative scale of the negative binomial log‐link model. In this context, IRRs represent the multiplicative change in the modelled expected RBQ score for a one‐unit change in a predictor. For the country comparison (the United Kingdom vs. Japan), IRRs above 1 indicate higher scores in the United Kingdom, whereas values below 1 indicate higher scores in Japan. For age effects, IRRs above 1 indicate increases in caregiver‐reported RBQ scores with age, whereas values below 1 indicate decreases in caregiver‐reported RBQ scores with age. For interaction terms, IRRs above 1 indicate that the effect of one variable is amplified by the other, whereas values below 1 indicate attenuation of effects. To complement interval estimates, we also report the posterior probability that each effect is positive, denoted as P[·] (e.g., P[UK > Japan], P[Male>Female]), which provides a direct probabilistic statement about the direction of each effect on the outcome scale. This Bayesian approach offers a more nuanced interpretation than traditional null hypothesis significance testing, allowing readers to assess the strength of evidence on a continuous scale rather than relying on arbitrary significance thresholds. Because three conceptually related RBQ outcomes were examined, the analyses should be interpreted as exploratory and descriptive rather than as confirmatory tests of independent hypotheses. We did not apply a formal multiplicity correction; instead, we emphasise effect sizes, credible intervals, directional consistency across outcomes and the preliminary nature of the findings.

3. Results

Caregiver‐reported RBQ scores showed directional between‐country differences and weak cross‐sectional age‐related associations. Across domains, scores tended to be higher in reports from UK caregivers than from Japanese caregivers (Table 2). Directional posterior probabilities for country comparisons were generally moderate to high (≈0.67–0.94), although several 95% credible intervals included 1.0, indicating uncertainty in the precise magnitude of the differences.

TABLE 2.

Mean (SD) scores on outcome variables.

Outcome variable	UK (n = 40)	Japan (n = 40)
Total RBQ score (Items 1–30; range: 0–60)	11.8 (9.32)	8.98 (6.18)
Sensory/Motor Behaviours (range: 0–24)	4.53 (3.44)	4.05 (3.45)
Sameness/Circumscribed Interests (range: 0–30)	5.98 (5.75)	4.25 (2.94)

Note: The Total RBQ score was calculated as the sum of Items 1–30 and was not calculated as the sum of the two‐factor‐derived subscales. Items 9, 13, 27 and 30 contributed to the Total RBQ score but not to either subscale. Item 6 contributed to both subscale scores but was counted once in the Total RBQ score.

Abbreviations: n, number of participants; SD, standard deviation.

3.1. Total RBQ Scores

In the full sample (N = 80; without VMA control), UK caregivers reported directionally higher Total RBQ scores than Japanese caregivers (IRR = 1.23, 95% CrI [0.92, 1.64]; P[UK>Japan] = 0.92). Age was weakly negatively associated with Total scores (IRR per decade = 0.93, 95% CrI [0.72, 1.21]; P[decrease] = 0.72) (Figure 1A, Table 3).

FIGURE 1.

Estimated cross‐sectional age‐related trends in RBQ scores: (A) Total RBQ score based on Items 1–30, (B) Sensory/Motor Behaviours subscale score and (C) Sameness/Circumscribed Interests subscale score. The left and right panels show the Japanese and UK samples, respectively. The grey shaded area represents the 95% credible interval.

TABLE 3.

Bayesian negative binomial regression results for total RBQ scores (all participants, N = 80).

Parameter	Estimate	95% CrI	Incidence rate ratio (IRR)	95% CrI (IRR)	Posterior probability
Intercept	2.23	[2.01, 2.47]	9.30	[7.47, 11.8]	—
Country (UK vs. Japan)	0.21	[−0.09, 0.50]	1.23	[0.92, 1.64]	92.0% (P[UK > Japan])
Age (per 10 years)	−0.073	[−0.33, 0.19]	0.93	[0.72, 1.21]	72.0% (P[IRR < 1])
Country × age	−0.039	[−0.40, 0.33]	0.962	[0.67, 1.40]	41.0% (P[IRR > 1])
Shape	2.08	[1.39, 2.98]	—	—	—

Note: Models are Bayesian Negative Binomial (log link). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10), and Age was centred at the sample mean (10.31 years). Rate ratios (IRR) > 1 indicate increases and < 1 decreases. Posterior probabilities report directional evidence (e.g., P (IRR > 1) for UK > Japan). The Negative Binomial shape parameter indexes overdispersion (NB2; larger values = less overdispersion; Var(μ) = μ + μ ² /shape). Credible intervals including 1 indicate uncertainty about a nonnull effect.

Abbreviations: CrI, credible interval; RBQ, Repetitive Behaviour Questionnaire.

In the VMA‐available subsample (N = 42), the UK–Japan contrast in caregiver‐reported Total RBQ scores remained in the same direction but was attenuated and imprecisely estimated (IRR = 1.18, 95% CrI [0.77, 1.79]; P[UK>Japan] = 0.79). The cross‐sectional age association was near 0 (IRR per decade = 0.98, 95% CrI [0.65, 1.47]; P[decrease] = 0.54). VMA showed a weak negative association (IRR = 0.93, 95% CrI [0.64, 1.33]; P[decrease] = 0.66). Finally, evidence for Country × Age and Country × VMA interactions was limited (credible intervals including 1.0; directional probabilities ≈ 0.5–0.6) (Table 4).

TABLE 4.

Bayesian negative binomial regression results for total RBQ scores with VMA control (VMA Subsample, N = 42).

Parameter	Estimate	95% CrI (Estimate)	IRR	95% CrI (IRR)	Posterior probability
Intercept	2.24	[1.98, 2.49]	9.39	[7.24, 12.06]	—
Country (UK vs. Japan)	0.166	[−0.256, 0.579]	1.180	[0.774, 1.785]	79.0% (P[UK > Japan])
Age (per 10 years)	−0.020	[−0.432, 0.384]	0.980	[0.649, 1.469]	54.0% (P[IRR < 1])
VMA (z‐score)	−0.077	[−0.441, 0.288]	0.926	[0.643, 1.333]	66.0% (P[IRR < 1])
Country × Age	0.032	[−0.399, 0.470]	1.033	[0.671, 1.600]	56.0% (P[IRR > 1])
Country × VMA	−0.019	[−0.425, 0.378]	0.981	[0.654, 1.463]	46.0% (P[IRR > 1])
Shape	2.15	[1.42, 3.12]	—	—	—

Note: Models are Bayesian Negative Binomial with log link (NB2). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10), with Age centred at the sample mean (10.31 years). IRR > 1 indicates an increase and < 1 a decrease. Posterior probabilities report directional evidence (e.g., P [IRR > 1]). The NB2 shape parameter indexes overdispersion (larger values = less overdispersion).

Abbreviations: CrI, credible interval; IRR, incidence rate ratio; RBQ, Repetitive Behaviour Questionnaire; VMA, verbal mental age.

3.2. Sensory/Motor Behaviours

Without controlling for VMA, the between‐country contrast in caregiver‐reported Sensory/Motor Behaviours scores was modestly in the UK direction (IRR = 1.09, 95% CrI [0.79, 1.50]; P[UK>Japan] = 0.71). Age showed a notable negative trend (IRR per decade = 0.83, 95% CrI [0.62, 1.10]; P[decrease] = 0.90) (Figure 1B, Table 5).

TABLE 5.

Bayesian negative binomial regression results for sensory/motor behaviours (all participants, N = 80).

Parameter	Estimate	95% CrI	Incidence rate ratio (IRR)	95% CrI (IRR)	Posterior probability
Intercept	1.40	[1.19, 1.60]	4.05	[3.29, 4.95]	—
Country (UK vs. Japan)	0.086	[−0.24, 0.41]	1.09	[0.79, 1.50]	71.0% (P[UK > Japan])
Age (per 10 years)	−0.19	[−0.48, 0.10]	0.83	[0.62, 1.10]	90.0% (P[IRR < 1])
Country × Age	−0.05	[−0.45, 0.35]	0.95	[0.64, 1.42]	40.0% (P[IRR > 1])
Shape	2.27	[1.31, 3.80]	—	—	—

Note: Models are Bayesian Negative Binomial with log link (NB2). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10), with Age centred at the sample mean (10.31 years). IRR > 1 indicates an increase and < 1 a decrease. Posterior probabilities report directional evidence (e.g., P [IRR > 1]). The NB2 shape parameter indexes overdispersion (larger values = less overdispersion).

Abbreviation: CrI, credible interval.

When controlling for VMA (N = 42), the between‐country contrast in caregiver‐reported Sensory/Motor Behaviours scores remained weakly in the UK direction (IRR = 1.11, 95% CrI [0.71, 1.74]; P[UK>Japan] = 0.67), whereas the cross‐sectional age association was attenuated (IRR per decade = 0.95, 95% CrI [0.62, 1.47]; P[decrease] = 0.59). VMA again showed a weak negative association (IRR = 0.92, 95% CrI [0.63, 1.36]; P[decrease] = 0.67). Evidence for interaction effects (Country × Age, Country × VMA) was inconclusive (Table 6).

TABLE 6.

Bayesian negative binomial regression results for sensory/motor behaviours with VMA control (VMA subsample, N = 42).

Parameter	Estimate	95% CrI (Estimate)	IRR	95% CrI (IRR)	Posterior probability
Intercept	1.42	[1.16, 1.67]	4.14	[3.19, 5.31]	—
Country (UK vs. Japan)	0.104	[−0.344, 0.556]	1.110	[0.709, 1.744]	67.0% (P[UK > Japan])
Age (per 10 years)	−0.047	[−0.473, 0.386]	0.954	[0.623, 1.473]	59.0% (P[IRR < 1])
VMA (z‐score)	−0.083	[−0.468, 0.303]	0.920	[0.625, 1.355]	67.0% (P[IRR < 1])
Country × Age	−0.047	[−0.500, 0.413]	0.954	[0.607, 1.510]	58.0% (P[IRR < 1])
Country × VMA	−0.029	[−0.452, 0.394]	0.971	[0.636, 1.484]	55.0% (P[IRR < 1])
Shape	1.92	[1.31, 2.75]	—	—	—

Note: Models are Bayesian Negative Binomial with log link (NB2). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10) with Age centred at the sample mean (10.31 years). IRR > 1 indicates an increase and < 1 a decrease. Posterior probabilities report directional evidence (e.g., P [IRR > 1]). The NB2 shape parameter indexes overdispersion (larger values = less overdispersion).

Abbreviations: CrI, credible interval; IRR, Incidence rate ratio; VMA, verbal mental age.

3.3. Sameness/Circumscribed Interests

Without controlling for VMA, this domain showed the strongest directional between‐country difference in caregiver‐reported scores (IRR = 1.28, 95% CrI [0.94, 1.74]; P[UK>Japan] = 0.94). The cross‐sectional age association was near 0 (IRR per decade = 1.08, 95% CrI [0.84, 1.39]; P[increase] = 0.71) (Figure 1C, Table 7).

TABLE 7.

Bayesian negative binomial regression results for sameness/circumscribed interests (all participants, N = 80).

Parameter	Estimate	95% CrI	Incidence rate ratio (IRR)	95% CrI (IRR)	Posterior probability
Intercept	1.45	[1.21, 1.69]	4.26	[3.35, 5.42]	—
Country (UK vs. Japan)	0.25	[−0.06, 0.55]	1.28	[0.94, 1.74]	94.0% (P[UK > Japan])
Age (per 10 years)	0.072	[−0.18, 0.33]	1.08	[0.84, 1.39]	71.0% (P[IRR > 1])
Country × Age	−0.12	[−0.49, 0.26]	0.89	[0.62, 1.30]	27.0% (P[IRR > 1])
Shape	1.52	[1.08, 2.11]	—	—	—

Note: CrI, credible interval. Models are Bayesian Negative Binomial with log link (NB2). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10), with Age centred at the sample mean (10.31 years). IRR > 1 indicates an increase and < 1 a decrease. Posterior probabilities report directional evidence (e.g., P [IRR > 1]). The NB2 shape parameter indexes overdispersion (larger values = less overdispersion).

In the VMA‐available subsample, after adjusting for VMA, the UK–Japan contrast in caregiver‐reported Sameness/Circumscribed Interests scores remained in the same direction but was attenuated and imprecisely estimated (IRR = 1.15, 95% CrI [0.76, 1.74]; P[UK>Japan] = 0.75). The cross‐sectional age association was near 0 (IRR per decade = 1.02, 95% CrI [0.68, 1.52]; P[increase] = 0.54). VMA showed a weak negative and imprecisely estimated association (IRR = 0.91, 95% CrI [0.62, 1.30]; P[decrease] = 0.70). Evidence for Country × Age and Country × VMA interactions was inconclusive (Table 8).

TABLE 8.

Bayesian negative binomial regression results for sameness/circumscribed interests RBQ scores with VMA control (VMA subsample, N = 42).

Parameter	Estimate	95% CrI (Estimate)	IRR	95% CrI (IRR)	Posterior probability
Intercept	1.48	[1.21, 1.74]	4.39	[3.35, 5.70]	—
Country (UK vs. Japan)	0.142	[−0.272, 0.553]	1.153	[0.762, 1.740]	75.0% (P[UK > Japan])
Age (per 10 years)	0.019	[−0.388, 0.417]	1.019	[0.678, 1.516]	54.0% (P[IRR > 1])
VMA (z‐score)	−0.100	[−0.473, 0.263]	0.905	[0.623, 1.301]	70.0% (P[IRR < 1])
Country × Age	0.068	[−0.371, 0.504]	1.070	[0.692, 1.655]	62.0% (P[IRR > 1])
Country × VMA	−0.074	[−0.481, 0.334]	0.929	[0.619, 1.396]	64.0% (P[IRR < 1])
Shape	1.58	[1.12, 2.19]	—	—	—

Note: Models are Bayesian Negative Binomial with log link (NB2). Country is coded UK vs. Japan (Japan = reference). Age effects are reported per 10 years (Age10), with Age centred at the sample mean (10.31 years). IRR > 1 indicates an increase and < 1 a decrease. Posterior probabilities report directional evidence (e.g., P (IRR > 1)). The NB2 shape parameter indexes overdispersion (larger values = less overdispersion).

Abbreviations: CrI, credible interval; IRR, incidence rate ratio; RBQ, Repetitive Behaviour Questionnaire; VMA, verbal mental age.

3.4. Model Validation

Posterior predictive checks indicated that the negative binomial models adequately capture the observed distributions, including density, variance and upper tails. Leave‐one‐out cross‐validation favoured the negative binomial models over the Poisson models in the full sample, with ΔELPD values of +116.3 for Total RBQ scores, +28.8 for Sensory/Motor Behaviours scores and +48.5 for Sameness/Circumscribed Interests scores. These results supported the negative binomial specification over the Poisson model, consistent with the observed overdispersion. Comparisons of the negative binomial and zero‐inflated negative binomial models showed negligible ELPD differences in both the full sample [ΔELPD, negative binomial—zero‐inflated negative binomial: Total RBQ score = −0.94; Sensory/Motor Behaviours = +0.30; Sameness/Circumscribed Interests = +0.80] and the VMA subset [Total RBQ score = −0.64; Sensory/Motor Behaviours = +0.08; Sameness/Circumscribed Interests = +0.40]. Because these differences were close to 0, there was no evidence that adding a zero‐inflation component improved predictive performance; the simpler negative binomial models were therefore retained. All models demonstrated good Markov Chain Monte Carlo (MCMC) diagnostics (R‐hat ≤ 1.01; large effective sample sizes; no divergent transitions).

3.5. Sex Sensitivity Analysis

The sex distribution of individuals with WS was comparable across countries (Japan: 16 males, 24 females; the United Kingdom: 18 males and 22 females). Descriptively, males showed higher RBQ scores than females in both countries across all three outcomes (Total: Japan males M = 11.25, SD = 7.64; Japan females M = 7.46, SD = 4.73; UK males M = 13.78, SD = 11.39; UK females M = 10.18, SD = 7.37).

In the sex‐adjusted model (Country × Age10 + MF), child sex showed evidence of a positive association with Total RBQ scores (male vs. female: IRR = 1.35, 95% CrI [1.01, 1.81]; P[Male>Female] = 0.977). Evidence was more modest for the Sensory/Motor Behaviours (IRR = 1.32, 95% CrI [0.96, 1.80]; P[Male > Female] = 0.955) and Sameness/Circumscribed Interests subscales (IRR = 1.29, 95% CrI [0.95, 1.77]; P[Male > Female] = 0.947), with credible intervals that included 1.0. Critically, the country contrast remained essentially unchanged after controlling for child sex (Total: IRR = 1.23, 95% CrI [0.91, 1.64]; Sensory/Motor Behaviours: IRR = 1.09, 95% CrI [0.80, 1.49]; Sameness/Circumscribed Interests: IRR = 1.27, 95% CrI [0.93, 1.73]), indicating that the between‐country differences in caregiver‐reported RBQ scores were not attributable to between‐sample imbalances in sex composition.

An exploratory model incorporating a Country × Sex interaction provided little evidence that sex‐related differences in RBQ scores varied between the UK and Japanese samples (Total: IRR = 1.09, 95% CrI [0.73, 1.64]; Sensory/Motor Behaviours: IRR = 1.02, 95% CrI [0.67, 1.57]; Sameness/Circumscribed Interests: IRR = 1.18, 95% CrI [0.78, 1.80]; P[interaction > 1] ≈ 0.54–0.78). LOO‐IC comparisons indicated marginally better predictive fit for the sex‐additive model relative to the base model (ΔELPD: Total = +2.0, Sensory/Motor Behaviours = +1.3, Sameness/Circumscribed Interests = +1.0), though differences were small relative to their standard errors and do not warrant strong conclusions.

3.6. Summary

Across the three RBQ outcomes, caregiver‐reported scores were directionally higher in the United Kingdom than in Japan, with the clearest country contrast observed for the Sameness/Circumscribed Interests domain. In the VMA‐available subsample (Japan: n = 16; United Kingdom: n = 26), the country contrasts remained in the same direction but were attenuated and imprecisely estimated, with wide credible intervals. These VMA‐adjusted analyses should therefore be interpreted as sensitivity analyses rather than definitive evidence that between‐country differences are independent of verbal developmental level. VMA showed weak negative and imprecise associations with RBQ scores across domains. Country × Age and Country × VMA interactions were also inconclusive; thus, the present data do not provide strong evidence for country‐specific age‐related or VMA‐related slopes. Additionally, caregiver‐reported RBQ scores tended to be higher for male than female participants across both countries, particularly for Total RBQ scores; however, controlling for child sex did not materially alter the between‐country contrasts, and no clear evidence was found for a Country × Sex interaction.

4. Discussion

The current study found directional evidence of between‐country differences in caregiver‐reported Total RBQ and Sameness/Circumscribed Interests scores, whereas no clear evidence of a between‐country difference was observed for caregiver‐reported Sensory/Motor Behaviours scores. Overall, UK caregivers reported higher RBQ scores than Japanese caregivers. Caregiver‐reported Total RBQ and Sensory/Motor Behaviours scores showed weak directional cross‐sectional age‐related negative trends, but the evidence was insufficient to draw firm conclusions. Moreover, the total RBQ scores for UK individuals in this study were comparable to those reported in previous studies of individuals with WS in the United Kingdom (Riby et al. 2013; Rodgers et al. ²⁰¹²).

In line with our predictions and consistent with a recent cross‐cultural study of caregiver‐reported RRBs in autistic individuals (Matson et al. 2017), we found directional evidence for between‐country differences in caregiver‐reported Total RBQ and Sameness/Circumscribed Interests scores. Matson et al. (2017) suggest that cultural factors may influence how parents perceive RRBs relative to other autism‐related behaviours. Our findings extend this idea cautiously by suggesting that caregiver‐reported RBQ scores in individuals with WS may show directional between‐country variation, particularly for Sameness/Circumscribed Interests.

These differing patterns between the Sensory/Motor Behaviours and Sameness/Circumscribed Interests subscales may reflect the distinct nature of each domain. Sensory/Motor Behaviours are often considered lower order RRBs and may reflect more biologically rooted sensorimotor responses that are less dependent on caregiver interpretation or cultural expectations. In contrast, Sameness/Circumscribed Interests reflect higher order RRBs, which involve cognitive rigidity, routine‐based behaviours and narrow interests that may be more influenced by environmental context and cultural expectations. This may help explain why the clearest directional country contrast was seen for caregiver‐reported Sameness/Circumscribed Interests, whereas the Sensory/Motor Behaviours contrast was weak and imprecisely estimated. One possible interpretation is that caregiver reports of sameness and circumscribed interests may be shaped by culturally patterned expectations about daily routines, behavioural flexibility and the salience of individual preferences. However, because the present study did not directly measure cultural values, parenting practices or response styles, this explanation should be regarded as tentative. Future research should directly assess these contextual factors when examining cross‐cultural variation in caregiver‐reported RRBs.

We observed weak cross‐sectional negative trends of age on both total and Sensory/Motor Behaviours scores; however, these estimates were imprecise and were substantially attenuated after adjusting for VMA and should therefore be interpreted with caution. These patterns are broadly consistent with prior studies reporting age‐related reductions in social responsiveness as measured by the Social Responsiveness Scale, Second Edition (Hirai et al. 2024), as well as reductions in physical aggression and temper tantrums, though verbal aggression does not appear to decrease before 19 years of age (Rice et al. 2015). However, sensory‐profile findings suggest that age‐related reductions in sensory sensitivity are not consistently observed in individuals with WS (Hirai et al. 2025). The present Sensory/Motor Behaviours findings should therefore be interpreted as preliminary and not as definitive evidence of developmental decline. Notably, little evidence of a comparable age‐related trend was observed for Sameness/Circumscribed Interests scores, suggesting that the two RRB components may follow different cross‐sectional age patterns. The mechanisms underlying this potential dissociation remain unclear. Longitudinal studies with larger samples would be required to determine whether these cross‐sectional patterns reflect genuine developmental change.

This study has several limitations. First, we assessed caregiver‐reported RBQ profiles only in individuals with WS from the United Kingdom and Japan and found directional between‐country differences in Total RBQ and Sameness/Circumscribed Interests scores. A previous study has identified cross‐cultural differences in social phenotypes, such as global sociability between individuals with WS from Japan and the United States (Zitzer‐Comfort et al. ²⁰⁰⁷); however, the mechanisms underlying these differences remain unclear. Future research should conduct cross‐cultural studies on both social and nonsocial phenotypes to elucidate these mechanisms. Second, data were provided by primary caregivers, which introduce potential reporting bias. Cultural differences in parental attitudes toward children with WS may have influenced reporting. In addition, respondent‐level demographic information—including parental sex (i.e., whether the respondent was the mother or father), parental age, educational level and marital status—was not systematically available for either sample and therefore could not be examined, although these characteristics may influence the perception and reporting of behavioural symptoms. Future research should consider alternative assessment methods and the systematic collection of respondent‐level demographic information to address these potential sources of bias. Third, we were unable to determine whether any participants had a codiagnosis of autism spectrum disorder (ASD), as no standardised ASD assessments were administered and this information was not available for either sample. ASD co‐occurrence has been reported in approximately 10% of individuals with WS (Richards et al. 2015), and given the phenotypic overlap between WS‐associated features and ASD symptomatology, the potential contribution of ASD codiagnosis to observed group differences cannot be fully excluded. Future research would benefit from incorporating standardised ASD assessments as part of participant characterisation. Fourth, medication data were not collected for either sample; accessing such information for the UK participants would have required separate NHS research ethics approval. Certain medications (e.g., stimulants) may influence the frequency or presentation of repetitive behaviours, and this represents a further uncontrolled variable. Fifth, genomic deletion size data were not available for either sample. Although the typical WS deletion is relatively consistent in size (~1.55 Mb), atypical deletions exist and could potentially influence cognitive and behavioural outcomes; their absence therefore represents an uncontrolled variable in the interpretation of between‐country differences. Sixth, although a back‐translated Japanese version of the RBQ was used, formal tests of cross‐cultural measurement invariance or psychometric equivalence were not conducted. Accordingly, the observed between‐country differences in RBQ scores may reflect not only behavioural variation but also differences in item interpretation, response style or scale functioning across languages and cultural contexts. Seventh, the two samples were also recruited and assessed using different procedures (postal opt‐in recruitment via the WSF and family networks in the United Kingdom vs. on‐site camp‐based recruitment and assessment in Japan). These procedural differences may have influenced participation, caregiver reporting or the availability of auxiliary measures, and therefore represent a further source of between‐country heterogeneity. In addition, the UK data were collected in 2011–2012, whereas the Japanese data were collected in 2016. Although both datasets were collected before the present analyses and used comparable questionnaire methods, this temporal gap may have introduced cohort or period effects, including changes in awareness of neurodevelopmental conditions or caregiver reporting practices over time. Eighth, the modest sample sizes constrain statistical precision, as reflected in the wide credible intervals observed for several parameters. Future studies with larger samples would improve the precision of effect estimates and help clarify patterns that remain directional but uncertain in the present data. Finally, although participants in both countries were identified as having WS through WS‐specific support or community settings, individual‐level documentation of diagnostic procedures was not available for all participants. In the United Kingdom, molecular genetic confirmation is required for WSF membership; however, because recruitment also occurred through family‐based WS support networks and the study records did not document the recruitment route or molecular diagnostic status for each participant, the exact number of genetically confirmed UK participants could not be verified. In the Japanese sample, detailed information on diagnostic procedures, including molecular testing, was also not available for all participants. This uncertainty in diagnostic verification warrants additional caution in the interpretation of between‐country comparisons.

Despite these limitations, this study is, to our knowledge, the first to examine caregiver‐reported RBQ profiles among individuals identified as having WS in the United Kingdom and Japan. The findings provide directional evidence for between‐country differences in caregiver‐reported Total RBQ and Sameness/Circumscribed Interests scores. However, no clear evidence of a between‐country difference was observed for caregiver‐reported Sensory/Motor Behaviours scores. Moreover, we observed weak cross‐sectional age‐related negative trends for both Total RBQ and Sensory/Motor Behaviours scores, suggesting possible age‐related patterns in caregiver‐reported scores, although these estimates were imprecise and should not be interpreted as evidence of developmental decline. Given that cultural norms can influence the diagnostic process of psychiatric conditions (Rogler 1993), future research should examine how cultural norms affect the expression, perception, and caregiver reporting of WS‐associated behavioural phenotypes.

Funding

This study was funded by the Pfizer Health Research Foundation, Japan; Japan Society for the Promotion of Science (JSPS) KAKENHI (Grant Numbers: 18H01103, 21K18554, and 21KK0041); and a Grant‐in‐Aid for Scientific Research on Innovative Areas (Grant Number: 15H01585) awarded to MH. The grant‐giving institutions played no role in the design, collection, analysis and interpretation of data and the writing of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.