Validation of the German Glasgow Sensory Questionnaire and replication of sensory processing differences in students with higher and lower Autism-Spectrum Quotient

Zeisel, Annalena; Thiel, Tobias; Gaigg, Sebastian B.; Roessner, Veit; Ring, Melanie

doi:10.1186/s12888-023-04903-9

Research article
Open access
Published: 14 June 2023

Validation of the German Glasgow Sensory Questionnaire and replication of sensory processing differences in students with higher and lower Autism-Spectrum Quotient

Annalena Zeisel¹,
Tobias Thiel¹,
Sebastian B. Gaigg²,
Veit Roessner¹ &
…
Melanie Ring ORCID: orcid.org/0000-0001-7579-1829¹

BMC Psychiatry volume 23, Article number: 426 (2023) Cite this article

1482 Accesses
1 Citations
4 Altmetric
Metrics details

Abstract

Background

The Glasgow Sensory Questionnaire (GSQ) gives insight into sensory processing differences (hypo- and hyper-sensitivity across modalities), which is a clinically defining characteristic of autism spectrum disorder (ASD). Because there is no validated German version of this instrument, this study aimed at validating the German GSQ. Further, a replication of the GSQ’s sensory processing differences was intended.

Methods

University students of Technische Universität or Universitätsklinikum in Dresden, Germany, were recruited via email distribution or the university homepage and 297 German-speaking students completed the online survey, comprising the German GSQ, Autism-Spectrum Quotient (AQ) and Symptom-Checklist (SCL-90). For validation of the German GSQ, confirmatory factor analyses followed by exploratory factor analyses were applied.

Results

The German GSQ has moderate to low validity, good to acceptable reliability, and a different internal structure from the original GSQ. Replicating the sensory processing differences in students with higher and lower AQ was not successful.

Conclusions

Results indicate that the GSQ, developed especially for individuals with ASD, is less informative for the general population if there are not enough individuals with higher AQ scores in the sample.

Peer Review reports

Background

Imagine you are at home: the light in your room has been flickering barely noticeable for weeks, a light breeze comes through the door, and you hear the dull car noises outside. Just another day in an ordinary apartment, for most. What if, for you, the flickering feels like fireworks, the breeze is freezing cold, and the noise upsets you? Every individual perceives identical sensory stimuli differently. Some seek them out specifically, simply accept them, or find them irritating and unpleasant. And for some, such “ordinary” stimuli can become unbearable.

Autism spectrum disorder (ASD) is a neurodevelopmental disorder that is characterized by difficulties in social communication and interaction (A) and restricted, repetitive behavioral patterns (B) [1]. In addition, the fifth version of the Diagnostic and Statistical Manual of Mental Disorders now recognizes that ASD is very commonly associated with sensory processing differences including both hyper- and hyposensitivity [1]. This means that there are individuals with ASD who perceive sensory stimuli differently, either as particularly intense (e.g., background noises seem especially loud and lead to distraction) or much weakened (e.g., hurting oneself but feeling no pain).

A number of studies have documented both sensory hyper- and hyposensitivity in autism using primarily self-report questionnaires [7, 18, 23, 28, 40, 43]. However, qualitative interviews [57, 69] as well was psychophysiological [47] and neuroimaging [12, 25, 26] evidence confirms that individuals with ASD often experience their sensory environment differently. In the general population there is evidence of an association between subclinical autism-like traits as measured with the Autism-Spectrum Quotient (AQ; [6]) and sensory processing differences as captured by the Glasgow Sensory Questionnaire [51]. Sensory processing differences in ASD can affect all sensory modalities, including vision [4, 12, 37, 55], hearing [27, 37, 56, 69], taste [8, 37, 41], olfaction [3, 8, 15, 37], tactile perception [5, 37], vestibular perception [33, 37], and proprioception [37, 47]. Senses can be affected individually, or processing differences can occur in multiple modalities [54].

Sensory processing differences in ASD often go along with sensory seeking (e.g., repetitive sniffing or touching of objects or seeking out visual stimulation) and/or sensory defensive behaviors (e.g., avoiding noisy environments) [14, 40, 43], and are thought to contribute to the development of mental health conditions such as anxiety [2, 21, 22, 31, 46], which commonly co-occur with ASD [30, 34, 65]. Especially individuals with ASD who have difficulty to react adequately to sensory input end up in stress, pain, or anxiety [2, 21, 24, 64]. In fact, 14%-26% of individuals with ASD are affected by a comorbid anxiety disorder [49]. Gillot and Standen [22] report that in adults with ASD elevated anxiety correlated with sensory stress.

Besides anxiety, in adults with ASD, unpleasant sensory sensitivity was related to fear, anger, social difficulties, and escape from certain situations, followed by avoidance, isolation, and depression [57]. Also, Lundqvist [38] found a relation between tactile over-responsiveness and impaired social interaction. Furthermore, hyper- and hyposensitivity led to feelings of sickness, emotional unease, or inability of multisensory processing due to fixation to one sensory input in an adult general population with more autism-like traits [48]. Last, hyper- and hyposensitivity were related to social deficits in individuals with ASD of all ages [62]. However, sensory hypersensitivity was not only related to negative experiences. Adults with ASD reported that heightened sensory experiences also resulted in calming, pleasure, fascination, and possibly went along with newfound abilities and even with professional activities, such as being an excellent chef due to distinct sensitivity regarding the gustatory sense [57]. Whether sensory input is perceived as positive or disturbing depends on the respective situation [57].

To gain insight into sensory processing differences in adults, previous research has used different self-report questionnaires such as the Adolescent/Adult Sensory Profile [9], the Sensory Perception Quotient [60], or the Sensory Sensitivity Questionnaire [44]. The Adolescent/Adult Sensory Profile was not specifically developed for individuals with ASD, and all listed questionnaires cover no more than five different sensory modalities. This has motivated the development of the Glasgow Sensory Questionnaire (GSQ; [48]) as a self-report measure of hyper- and hyposensitivity across all seven sensory modalities for individuals with ASD. It consists of 14 subscales concerning seven sensory modalities (visual, auditory, gustatory, olfactory, tactile, vestibular, proprioception) and the two domains hypersensitivity and hyposensitivity. The GSQ is available in its original language, English [48], and has been validated already in Japanese [59, 63], French [51], Danish [35], and Chinese [67]. It strongly correlates with the AQ [6] in all these translations [31, 35, 48, 51, 59, 63, 66, 67] indicating that, regardless of language, more autism-like traits go along with greater sensory processing differences as indicated by higher scores on the GSQ [31, 48, 51, 63, 67]. In addition, adults with ASD scored higher on the GSQ compared to typically developing adults [35, 59].

Regarding the GSQ, Sapey-Triomphe et al. [51] reported a stronger correlation between the domains hyper- and hyposensitivity, higher hypersensitivity scores in relation to hyposensitivity scores, and a more consistent pattern of sensory processing differences for individuals with higher AQ scores compared to individuals with lower AQ scores. Sapey-Triomphe et al. [51] referred to these results as “sensory profiles”, because the AQ and GSQ scores were related to each other in a particular way, so called “profiles”. This study refers to it as “sensory processing differences”. These findings might implicate that individuals with more autism-like traits experience more sensory processing differences (both hyper- and hyposensitivity) than individuals with fewer autism-like traits. Importantly, the GSQ seems also sensitive to sensory processing differences in other conditions such as attention deficit and hyperactivity disorder, anxiety, history of mental illness, and migraine. Individuals with these conditions also scored higher on the GSQ [31, 45], which is important to bear this in mind when interpreting the GSQ.

As shown above, the GSQ is a helpful tool to capture sensory processing differences. However, to this day there is no German version of the GSQ. Therefore, for the purpose of this study, the English version of the questionnaire was translated, and a large German University student sample was asked to fill in the GSQ as well as the German AQ [20]. The two main goals of this study were: (a) the validation of the German version of the GSQ according to the Universalist model of cross-cultural adaptation [29] and (b) the replication of sensory processing differences of individuals with higher or lower AQ scores as reported in the original English version by Robertson and Simmons [48] and in the French version of Sapey-Triomphe et al. [51]. Regarding the first aim, consistent with past research, it was expected that construct validity, reliability, and internal structure of the German GSQ are comparable to the original English version [48]. Regarding the second aim, it was expected that the German student sample would show similar sensory processing differences as the French sample depending on whether participants scored higher or lower on the AQ [51]. More precisely, a stronger correlation between hyper- and hyposensitivity, higher hypersensitivity scores in relation to hyposensitivity scores, and a more homogeneous pattern of sensory processing differences in all seven sensory modalities for individuals with higher compared to individuals with lower AQ scores were expected.

Methods

Participants

A sample of 297 German University students was recruited from October 2020 to May 2021 as part of a larger research project with the aim of improving the diagnostic process of anxiety disorders in individuals with ASD. The majority of the sample was recruited via the student email distribution list to 24,408 students including students of the Technische Universität Dresden (TUD) and of the Universitätsklinikum Carl Gustav Carus Dresden (UKD). The remaining participants were recruited via the homepage for online registration for experiments of the Faculty of Psychology of the TUD, the central experimental server of psychology of the TUD, the homepage for students of the UKD, the homepage of the Student Representatives of the UKD, Facebook groups, eBay, word of mouth, and private recruitment of friends and family. Participants were included if they reported being a student of at least 18 years, spoke German fluently, and if they fully completed the questionnaires. Participants were excluded if they reported consuming cannabis more than five times a year or if they used any other illegal drugs within the past year. Inclusion and exclusion criteria as well as demographic data were determined prior participation via a screening questionnaire. Following email and online recruitment, 490 students got in touch. Due to different aspects (minimum of subjects for inclusion of 300, gender ratio, and deadline), only 348 individuals were screened with 310 students fulfilling inclusion criteria. Six of these individuals did not report back, so access to the online survey was sent out to 304 students. Of those, seven individuals did not fill in the survey (reasons e.g., no time, private reason). Therefore, the total sample included 297 German speaking students with varying educational qualifications and fields of study (176 women, 121 men, mean age: 23.5 ± 3.9 years, range: 18–52 years, see Table 1). Because the current study took place during the Covid-19 pandemic, perceived psychological distress was measured using the Global Severity Index (GSI) of the Symptom-Checklist 90-R (SCL-90-R; [16]). Students reported an average GSI T-score of 47.8 (≥ 60 = increased psychological distress). Further, at screening, 16.5% of the students reported having been affected by at least one physical illness (e.g., migraine, thyroid disease, neurodermatitis) and 12.8% by at least one mental illness (e.g., depression, anxiety, eating disorders) in their lives. Detailed information is presented in Tables 1 and 2.

Table 1 Demographical data: age, gender, education

Full size table

Table 2 Mental and physical health

Full size table

Composition of the subgroups for replicating sensory processing differences

To differentiate between individuals with higher or lower AQ scores in this study, a cut-off score of 26 was used, following previous studies [51, 70]. This cut-off value has been suggested to have acceptably high sensitivity and specificity to identify levels of autism-related traits in general population samples that may be clinically relevant (see [50]). In the present study, 262 students had an AQ < 26 points and 35 students an AQ ≥ 26. The 35 students with higher AQ scores were compared with a matched subsample of 35 students with lower AQ scores, derived from the group of 262 students. Individuals with low AQ scores were selected in a way that the two sub-groups of 35 students each did not differ in terms of gender, χ²(1) = 0.00, p = 1.00, φ = 1.00, and age, t(68) = -0.26, p = .79, d = -0.06, 95% Confidence Interval (CI) [-0.53, 0.41] and were similar in their SCL-90 GSI scores and in the occurrence of any mental disorders. Overall, subgroups did not differ in their demographic data, U _min = 490.00, Z = -1.46, p _min = .15, r = -0.17, their reported mental and physical health, U _min = 542.50, Z = -1.26, p _min = .29, r = -0.15, nor in their perceived psychological distress, t(68) = -0.74, p = .46, d = -0.18, 95% CI [-0.65, 0.29]. For detailed information on demographic and health data as well as absolute values see Tables 1 and 2.

Procedure

Participants filled in the questionnaires online via LimeSurvey [52], which took about 45–60 min. Participants were reimbursed for their time with 10€ or Psychology students were offered 1 course credit.

Measures

All questionnaires used were self-report measures in German versions. Out of a total of 12 questionnaires, only the AQ [20], the GSQ (self-developed German translation) and the SCL-90 [19] were considered in the current study.

The Autism-Spectrum Quotient (AQ; [6]) is a self-report questionnaire frequently used to examine autism-like traits in individuals with and without ASD [32, 39, 50, 58, 70]. The AQ consists of 50 items, summed up to five subscales (social skills, attention switching, attention to detail, communication, imagination) with 10 items each. Individuals indicate the extent to which they agree with each statement on a 4-point Likert scale (I strongly agree to I strongly disagree). Each item is scored 1 (autism-like) or 0 (not autism-like). Total scores range between 0 to 50 points with higher scores indicating a higher expression of autism-like traits. For each participant, the AQ total score and the five AQ subscale scores were calculated. Regarding reliability, the Cronbach’s alpha of the AQ for the current sample was 0.88, so the internal consistency of the questionnaire is satisfying.
The Glasgow Sensory Questionnaire (GSQ; [48]) was especially developed for adults with ASD and is appropriate for measuring sensory processing differences in individuals with and without ASD. It contains 42 items on 14 subscales concerning seven sensory modalities (visual, auditory, gustatory, olfactory, tactile, vestibular, proprioception) with six items each. Out of these six items, three each relate to the domains hypersensitivity and hyposensitivity. Individuals indicate the extent to which they experience the content of each statement on a 5-point Likert scale (0 = Never to 4 = Always). Total scores range between 0 to 168 points. The original English version has an internal consistency of r = 0.94 and reasonable content validity [48]. In the present study, the GSQ total score, 14 GSQ subscale scores, GSQ scores of the seven sensory modalities, and GSQ scores of the two domains hyper- and hyposensitivity were computed.
The Symptom-Checklist 90-R (SCL-90-R; [16]) is an inventory for individuals aged 12 and older to assess perceived psychological distress over the past seven days. On 90 items, individuals are asked to rate how much they experience the content of each statement on a 5-point Likert scale (0 = Not at all to 4 = Extremely). Across all items, a global parameter, the Global Severity Index (GSI), can be calculated to measure the overall psychological distress. For the GSI, T-scores for students exist for interpretation (60–64 = slightly, 65–69 significantly, 70–74 strongly, 75–80 very strong increased psychological distress). The German SCL-90-R [19] was used, with an internal consistency of r = 0.96 for students GSI and a disputed factorial validity [19].

Translation and cross-cultural adaptation

Following permission of the authors (Robertson & Simmons, University of Glasgow), the GSQ was translated into German. A translation-back-translation was done by two German/English bilingual psychologists followed by a discussion and revision of the translation in the research team. To enable an intercultural equivalent validation of the GSQ, the aspects of the Universalist model of cross-cultural adaptation [29] were considered including conceptual, item, semantic, operational, measurement, and functional equivalence.

Data analysis

The analysis of the results followed the approach of Sapey-Triomphe et al. [51]. Where possible, we compared our data to the original English data of Robertson and Simmons [48]. However, when there were no data available as in the case of the subscale analysis, we followed Sapey-Triomphe et al. [51] by comparing our data to the data of Ward et al. [66]. In general, unlike Sapey-Triomphe et al. [51], IBM SPSS Statistics (Version 27 and Version 28) and IBM SPSS Amos (Version 27) were used for statistical analyses. One-sided tests were used with a threshold of p < .05 for statistical significance, and for multiple comparisons Bonferroni correction was applied. For group comparisons of correlations, the data were z standardized and tested two-tailed using an online Fisher’s r-to-Z Transformation [68].

To investigate the validity and compare the present scores to the scores obtained by the original English version [48] and to the scores obtained by Ward et al. [66], Spearman’s ρ and two-tailed one-sample t tests were performed. Regarding the mentioned comparison, it should be noted that Pearson’s r (used by Robertson and Simmons [48] and Sapey-Triomphe et al. [51]) and Spearman’s ρ (used in this study) are comparable because they use the same scale from -1 to 1. However, it must always be kept in mind when making this comparison that both measures are based on different assumptions [36, 53]. In our study we used Spearman’s ρ because pre-analyses showed that the premise of linearity was not given. Therefore, the use of Pearson’s r would have underestimated the strength of the relation between the two variables in question [17]. Internal reliability was assessed with Cronbach’s alpha and item-total correlations were considered. To test the internal structure, a confirmatory factor analysis (CFA), with unweighted least squares method (ULS), was used to verify the existing model of Robertson and Simmons [48]. In addition, to check if a better model exists, exploratory factor analyses (EFA), with principal axis factoring method (PAF) and ULS, were calculated. For rotation, varimax rotation was used, and small coefficients (absolute value below 0.30) were suppressed.

Spearman’s ρ was used for correlation analyses. Subgroups were compared using Fisher’s r-to-Z transformation [68] and two-tailed unpaired t tests. For within group comparisons, two-tailed paired-samples t tests were calculated. Between subgroups, correlation matrices for the 14 GSQ subscales were compared with Fisher’s r-to-Z comparisons [68].

Results

Validation process

There were no missing data. For multiple comparisons, only significances after Bonferroni correction are reported with p _corrected < .0035 for the 14 GSQ subscales, p _corrected < .007 for the seven GSQ sensory modalities, and p _corrected < .01 for the five AQ subscales.

Validity

The GSQ total scores ranged from 11 to 89 (M = 39.8 ± 15.4 compared to 56.7 ± 23.6 in the original English sample) and the AQ total scores ranged from 3 to 41 (M = 16.1 ± 7.7 compared to 22.5 ± 10.6 in the original English sample) in the total sample (see Tables 8 and 9 in Appendix 1). The GSQ and AQ total scores correlated moderately in the total sample, ρ = 0.44, p < .001 (compared to r = 0.78, p < .0001 in the original English version), see Fig. 1a.

The means and standard deviations of the 42 GSQ items of the present data correlated strongly and moderately respectively with means and standard deviations by Robertson and Simmons [48], ρ = 0.87, p < .001 (for means), and ρ = 0.60, p < .001 (for standard deviations; see Fig. 2a). As can be seen in Fig. 2b, one-sample t tests between means of the 42 GSQ items of the current sample and means by Robertson and Simmons [48] were not significant for Item 7, t(296) = -1.05, p = .30, d = -0.06, 95% CI [-0.18, 0.05], Item 9, t(296) = -0.99, p = .32, d = -0.06, 95% CI [-0.17, 0.06], and Item 40, t(296) = -0.60, p = .55, d = -0.04, 95% CI [-0.15, 0.08]. For the means of the remaining 39 GSQ items one-sample t tests were significant, t(296)_min > 2.01, p _max < .045, d _min > 0.12, 95% CI [0.003, 0.23].

In the total sample as well as in students with higher AQ scores, none of the 14 GSQ subscale means correlated significantly with the data by Ward et al. [66], ρ _max < 0.35, p _min > .02, after Bonferroni correction. In students with lower AQ scores, the subscale tactile hypersensitivity mean correlated with the data by Ward et al. [66], ρ = 0.46, p = .0028. Of the seven sensory modality means, none correlated with the data by Ward et al. [66] when comparing total samples as well as individuals with higher and lower AQ scores in both samples, ρ _max < 0.29, p _min > .04, after Bonferroni correction. Details on correlations are presented in Table 10 in Appendix 1.

Reliability

The internal consistency for the German GSQ was satisfying. For the 42 items, Cronbach’s alpha was 0.88 in the total sample, 0.85 in individuals with higher AQ scores, and 0.86 in individuals with lower AQ scores. For the 14 subscales, Cronbach’s alpha was 0.87 in the total sample, 0.82 in students with higher AQ scores, and 0.85 in students with lower AQ scores. Corrected item-total correlation was 0.36 on average for the total sample, with correlations under 0.20 for Items 17, 22, and 36. It was 0.33 for students with higher AQ scores (with a negative correlation for Item 35), and 0.34 for students with lower AQ scores. Detailed item-total correlations can be found in Table 11 in Appendix 1.

Internal structure

A visualisation of the tested model is presented in Fig. 5 in Appendix 2. The model shows the internal structure of the GSQ as presented by Robertson and Simmons [48]. All GSQ items were assigned to one of the 14 GSQ subscales. In a next step, each of these 14 GSQ subscales was assigned to one of the two domains hyper- or hyposensitivity. Our assignment corresponds to the structural construction of the questionnaire by Robertson and Simmons [48]. In order to test the model, a Standardized First-Order Confirmatory Factor Model of the GSQ’s 14 subscales was computed in a first step, followed by a Standardized Second-Order Confirmatory Factor Model of the GSQ’s two domains based on the first-order model. Twelve participants were excluded from CFA from the total sample (N = 297) because they were identified as outliers due to Mahalanobis distance, p < 10^–3.^{Footnote 1} For the remaining 285 students, the model solution for CFA in Amos was not permitted because six error variances were negative for the second-order factor model for the two domains hyper- and hyposensitivity. Therefore, we set these negative variances to 0 and re-ran the analyses. In more detail, standardized regression weights for the two domains hyper- and hyposensitivity were on average 0.88, range = 0.91. (vestibular hyposensitivity)—0.49 (olfactory hyposensitivity). For the 14 subscales, standardized regression weights were on average 0.44, range = 0.09 (Item 17, olfactory hyposensitivity)—0.72 (Item 23, gustatory hypersensitivity). The covariance between the two domains was 0.25, the correlation was 0.84. The Goodness of Fit Index (GFI) was 0.94 and the χ² /df was 0.80 (χ² = 520.5, df = 810). Overall, the present data did not seem to fit the model presented in Fig. 5 in Appendix 2 perfectly. Then, an EFA with the 42 GSQ items in the sample of the remaining 285 students was performed to extract the underlying factor structure of the German GSQ version. The ratio of subjects per variable was 6.8. Since PAF and ULS achieved relatively similar results, only the results of the ULS are reported. The Kaiser–Meyer–Olkin measure of sampling adequacy was 0.82 (> 0.60), showing that data were appropriate for factor analysis. Bartlett’s test of Sphericity was significant (χ² > 2620, p < .001), indicating that correlations between variables were sufficient. The determinant was 5.913E-5 (> 0.00001), so multicollinearity was not a problem. Measures of sampling adequacy (ideally > 0.60) were below 0.60 for Items 22 (tactile hypersensitivity), 17, and 36 (both olfactory hyposensitivity) and, therefore, these items should be removed from the scale. There were 29 (3%) nonredundant residuals (ideally < 60% for a good model). The average communalities across the 42 items were 0.31 before and 0.40 after rotation, with communalities under 0.30 after rotation for Items 5, 9, 17, 18, 22, 29, and 38. This means that on average the extracted factors moderately predicted participants’ responses to the items. Fourteen factors with an eigenvalue ≥ 1 could be identified which accounted for 39.5% of the total variance after rotation, which is rather low. The 14 factors were also confirmed by the scree-plot. However, as the classification of the items to the subscales differed from the one in the original English version [48], the factor loadings on the 14 factors did not reflect the original structure. Factor loadings can be found in Table 12 in Appendix 1.

Further, an ULS for two factors to be extracted was performed, only in this case with oblique rotation, to check if the two domains hyper- and hyposensitivity could be found. The Kaiser–Meyer–Olkin was 0.82 and Bartlett’s test of Sphericity was significant (χ² > 2620, p < .001). The determinant was 5.913E-5. Measures of sampling adequacy were below 0.60 for Items 17, 22, and 36. There were 310 (36%) nonredundant residuals. Average communalities across the 42 items were 0.19 after rotation, with communalities under 0.10 after rotation for Items 5, 17, 22, 28, 36, 38, and 39 (see Table 3). This means that on average the extracted factors poorly predicted participants’ responses to the items. The total variance was 18.58%, which is quite low compared to the explained variance of 39% found by Sapey-Triomphe et al. [51], with 15.31% of variance for Factor 1 and 3.28% of variance for Factor 2 after rotation. Seventeen items loaded on Factor 1, with 13 items belonging to the domain hyposensitivity. Factor 2 included 16 negatively loaded items, 13 of them belonging to the domain hypersensitivity. Thus, the internal structure does seem to somewhat match the structure of the original English version [48] and of the French version [51], and therefore in the analyses reported below we maintain the distinction between hyper- and hypo-sensitivity to allow for comparison with these earlier studies. Individual factor loadings are shown in Table 4.

Table 3 Communalities for the 42 GSQ items for the two-factor solution in the total sample

Full size table

Table 4 Pattern matrix with factor loadings of the two-factor solution in the total sample

Full size table

Sensory processing differences in students with higher and lower Autism-Spectrum Quotient scores

Group comparison of students with higher vs. lower Autism-Spectrum Quotient scores

Students with higher and lower AQ scores differed significantly in GSQ, t(68) = -3.31, p = .002, d = -0.79, 95% CI [-1.27, -0.30], and AQ total scores, t(58.33) = -12.12, p < .001, d = -2.90, 95% CI [-3.57, -2.22], with higher scores for students with higher AQ scores (GSQ: M = 53.1 ± 15.8, AQ: M = 31.2 ± 4.1) compared to students with lower AQ scores (GSQ: M = 41.0 ± 14.9, AQ: M = 15.7 ± 6.4).

Regarding the two GSQ domains, individuals with higher AQ scores scored significantly higher in both domains than individuals with lower AQ scores (hypersensitivity: t(68) = -2.75, p = .008, d = -0.66, 95% CI [-1.14, -0.17], hyposensitivity: t(68) = -3.18, p = .002, d = -0.76, 95% CI [-1.24, -0.27]), see Table 9 in Appendix 1 for absolute values. Unpaired t tests revealed further that students with higher AQ scores scored significantly higher than students with lower AQ scores in the auditory, t(68) = -3.22, p = .002, d = -0.77, 95% CI [-1.25, -0.28], and tactile modality, t(59.69) = -3.43, p = .001, d = -0.82, 95% CI [-1.31, -0.33]. Differences regarding the remaining modalities were not significant, t _max < -2.45, p _min > .02, d _max < -0.59, 95% CI [-1.06, -0.11] (see Fig. 3).

Further, the two subgroups were compared regarding their correlation matrices. Fisher’s r-to-Z comparisons [68] between the two subgroups revealed no significant differences in correlations between the 14 GSQ subscales, Z _max < 1.71, p _min > .09, q _max < 0.43. Details on correlations between the 14 GSQ subscales are presented in Table 5.

Table 5 Correlation matrix with Spearman’s ρ between the 14 GSQ subscales in the two subgroups

Full size table

Correlations within subgroups

In both subgroups, GSQ and AQ total scores correlated significantly: students with higher AQ scores: ρ = 0.48, p = .002, students with lower AQ scores: ρ = 0.38, p = .01 (see Fig. 1b). A Fisher’s r-to-Z comparison [68] between the two subgroups revealed no significant difference in these two correlations, Z = 0.49, p = .62, q = 0.12.

In individuals with higher AQ scores, the GSQ total score and none of the AQ.subscales correlated significantly. In individuals with lower AQ scores, the GSQ total score and two AQ subscales correlated significantly. Details on correlations are presented in Table 6. Further, in students with higher and lower AQ scores, only the hypersensitivity domain correlated significantly with the AQ total score. Regarding the modalities, in individuals with higher AQ scores, there were significant correlations for two modalities with the AQ total score and in individuals with lower AQ scores for one modality with the AQ total score. Detailed information on correlations can be found in Table 6. Overall, correlations were weak to moderate. Individuals with higher and lower AQ scores did not differ significantly in their correlations (see Table 6).

Table 6 Correlations (Spearman’s ρ) between GSQ and AQ

Full size table

Within subgroups, hyper- and hyposensitivity domain total scores were significantly positively correlated, with ρ = 0.63, p < .001 for students with higher AQ scores and ρ = 0.57, p < .001 for students with lower AQ scores (see Fig. 4). A Fisher’s r-to-Z comparison [68] between the two subgroups revealed no significant difference in these two correlations, Z = 0.38, p = .71, q = 0.09. Regarding the seven sensory modalities, only for the visual and the auditory modalities the hyper- and hyposensitivity scores correlated significantly for students with higher and lower AQ scores. The two subgroups did not differ in their correlations between hyper- and hyposensitivity scores. Details on correlations between the two domains for each sensory modality and on comparisons between subgroups regarding these correlations are presented in Table 7.

Table 7 Correlations (Spearman’s ρ) between the two GSQ domains hyper- and hyposensitivity

Full size table

Hyper- and hyposensitivity domain scores did not differ significantly within subgroups, t _max < 0.39, p _min > .70, d _max < 0.07, 95% CI [-0.27, 0.40]. Regarding the seven modalities, both subgroups had significantly higher hyposensitivity scores in the proprioception modality (higher AQ scores: t(34) = -5.26, p < .001, d = -0.89, 95% CI [-1.28, -0.49], lower AQ scores: t(34) = -3.85, p < .001, d = -0.65, 95% CI [-1.01, -0.28]). Further, students with lower AQ scores had significantly higher hypersensitivity scores in the auditory modality, t(34) = 3.04, p = .004, d = 0.51, 95% CI [0.16, 0.86]. Students with higher AQ scores showed significantly higher hypersensitivity scores in the tactile modality, t(34) = 3.39, p = .002, d = 0.57, 95% CI [0.21, 0.93]. There were no other significant differences between hyper- and hyposensitivity scores, t _max < -2.57, p _min > .015, d _max < -0.44, 95% CI [-0.78, -0.09], after Bonferroni correction. See Table 9 in Appendix 1 for absolute values.

Discussion

The current study aimed at (a) validating the German version of the GSQ in students with higher and lower AQ scores according to the Universalist model of cross-cultural adaptation [29] and (b) at replicating the sensory processing differences of individuals with higher or lower AQ scores as reported in the original English version by Robertson and Simmons [48] and in the French version of Sapey-Triomphe et al. [51]. In the present study, with the present sample, the validation of the German GSQ was not completely satisfying. Also, the sensory processing differences could not be replicated. In the following, the results will be discussed.