Experience of Groove Questionnaire: German Translation and Validation

The current study aimed at translating and adapting the EGQ into German and validating the questionnaire with a large German sample. The central question was whether the two-dimensional factor structure of the English original version could be confirmed in the German translation. Furthermore, the present study offers first insights into convergent validity and discusses the implications for the construct of groove stemming from the validity measurements.

The EGQ is an inventory that can be used for the participants’ self-reported groove experiences while listening to a piece of music. The original version contains two scales: Urge to Move (M) and Pleasure (P), comprising three items each: M1: “This music evokes the sensation of wanting to move some part of my body”; M2: “This music is good for dancing”; M3: “I cannot sit still while listening to this music”; and P1: “Listening to this music gives me pleasure”; P2: “I like listening to this music”; P3: “This music makes me feel good”, respectively (Senn et al., 2020, p. 65). Both scales are highly correlated (r = .80) and show high internal consistency (Cronbach’s α_{Urge to Move} = .92, α_Pleasure = .97).

The development of the original English questionnaire was conducted in four steps: First, 25 candidate items were collected from the groove literature or formulated by the authors to measure the constructs Urge to Move and Pleasure. Second, these items were rated by 15 experts in groove and rhythm research (concerning the following questions: Which construct is captured? How certain am I about this? How relevant is the item?). The five least appropriate items were discarded, and two new items proposed by the respondents were added to the list. At this point, a first listening experiment was conducted in which N = 56 listeners rated eight musical stimuli using the set of 22 items (see Senn et al., 2020, p. 64, for the list of stimuli). An exploratory factor analysis was carried out to reveal the underlying factor structure. For each of the resulting two scales, three items were selected that best captured the underlying constructs. Finally, these six items were used in a second, larger listening experiment in which N = 197 participants rated the same 8 musical stimuli. A confirmatory factor analysis (CFA) was conducted and reliability (internal consistency measured by Cronbach’s α) as well as content validity were assessed (for more detailed information on the development of the EGQ see Senn et al., 2020).

The German translation of the six items was obtained by employing a four-step-procedure similar to that used by Lin et al. (2021). For an overview of the procedure, see Figure 1.

In Phase 1, the original questionnaire was translated into German by two professional translators (German being their mother tongue). Specific guidelines for the translation of scientific items for a psychological test were retrieved from Hambleton and Zenisky (2011) and recapitulated for the translators in a two-page summary.

In Phase 2, these two preliminary translations were discussed by the Hanover research team (i.e., N. Düvel, P. Labonde, and R. Kopiez) and an additional expert in questionnaire development. The two translations were consolidated into a third preliminary German version of the EGQ. This version was sent to the Lucerne research team (i.e., T. Bechtold, O. Senn) for further comments and approval.

In Phase 3, the third preliminary German version was sent to another professional translator (English as the mother tongue) for back-translation into English. The same translation guidelines as in Phase 1 were attached.

In Phase 4, the Hanover research team discussed the differences between the back-translation (obtained from Phase 3) and the original version deducing inappropriate expressions in the third preliminary German version. Taking all versions into account, a pre-final German version was proposed and subsequently sent to the Lucerne research team for comments and approval. After clarification of some minor issues, this version was declared to be final. For a list of all six items, see Figure 4.

The German translation of the EGQ was then tested and validated in an online survey based on the platform SoSci Survey (https://www.soscisurvey.de). The entire survey was presented in the German language. After reading a short introduction and giving informed consent, participants created a personalized ID (as proposed by Pöge, 2008). This procedure was required to collate the anonymously gathered datasets from the main study with the responses from the retest so that the study would be in compliance with the current European Data Protection Regulation (Deutsche Gesellschaft für Psychologie e.V., 2016; Hanover University of Music Drama and Media, 2017). Participants then indicated their gender, age, native language, as well as their language used in everyday life, level of education, and profession. Before starting the main part of the survey, participants were asked which kind of playback device they used (loudspeakers, headphones, or internal speakers of their laptop or smartphone). Then participants adjusted the volume of their devices by following a procedure developed by Wycisk, Kopiez, and Wolf (2018) to control for unwanted changes in playback volume during the experiment. While listening to short audio clips of white noise, participants had to perform a simple counting task of stimuli of different volume above and below their perceptual threshold as an objective measure. If the playback volume was too high, participants heard and counted more stimuli than they were supposed to hear with volume settings desired by the researchers. Therefore, participants who entered numbers higher than a certain threshold were given immediate feedback to decrease their volume and restart the task. If the volume was too low, they counted too few stimuli and were subsequently asked to increase their volume and repeat the task until the desired count was achieved. This ensured that participants were able to play back the audio example with an appropriate volume. Following this test, eight audio examples (the same as in Experiment 2 from Senn et al., 2020; see Table 1 in the present paper) were presented to the participants. From a wide variety of genres, the examples were selected to elicit extreme reactions on the dimensions Pleasure and Urge to Move by Senn et al. (2020, pp. 50–51). The Soul classic “Superstition” by Stevie Wonder and “Bala” by Bassekou Kouyate (focusing the Ngoni, a lute from West Africa) were expected to score highly on both Pleasure and Urge to Move. “Tchip Tchip” by Cash & Carry (also known as “Chicken Dance”) as well as “Hamdouchi” by lute player Maleem Mahmoud Guinia from Marocco in collaboration with American jazz saxophonist Pharoah Sanders were expected to elicit high Urge to Move but low Pleasure ratings. In case of “Tchip Tchip,” the low Pleasure ratings were expected due to the low prestige of the music (and the associated dance) and the squawky sound. In case of “Hamdouchi” they were expected because of the intense timbre with dissonant multiphonics. “Sunrise,” by the Bulgarian women’s choir, Angelite, and the excerpt from Mahler’s Symphony No. 3, were expected to elicit low Urge to Move due to fairly irregular pulse, but high Pleasure due to pleasant sonority. The high-energy free jazz recording “Machine Gun” by the Peter Brötzmann Octet (with no regular rhythmic pulse) and “My Pal Foot Foot” by the rock band The Shaggs (with discordant guitar playing and very loose rhythmic organisation) were expected to score low on both Urge to Move and Pleasure.

Each excerpt was 30 seconds long; the order of the musical stimuli was randomized across participants. They rated these music excerpts using three different questionnaires: (1) the new German translation of the EGQ with its two factors Urge to Move and Pleasure (six items); (2) the two factors Prototypical Aesthetic Emotions and Animation taken from the Aesthetic Emotions Scale (Aesthemos; Schindler et al., 2017; 13 items, all items available from the original German version); and (3) the Drum Pattern Quality Scale (DPQS) by Frühauf et al. (2013). These questionnaires were chosen to examine the content validity of the EGQ and situate it within the context of currently available comparable measures. The two factors of the Aesthemos were chosen as they are related to the two EGQ factors with regard to their content but measure more generic facets of a pleasurable and invigorating experience in response to a piece of art. A comparison of these factors could indicate if an additional inventory is necessary in the particular case. Moreover, the Aesthemos shows good psychometric properties. Second, the DPQS was chosen for the validation process of this study because it measures an aspect of performance quality concerning rhythmical aspects of the music (as does the EGQ). It not only measures the groove experience but also considers the overall quality rating of a drum pattern performance. Furthermore, it is one of the few inventories using multiple items, therefore, making it more comparable to the EGQ. Finally, both the DPQS and the Aesthemos are available in the German language. Consequently, these scales did not have to be translated to be used in this study. This precluded problems regarding the content and construct validity of a translation of comparable questionnaires into German. The DPQS uses the following items for the evaluation of drum pattern performance: “(a) ‘The interpretation (timing) of this music suits the rhythm’ (interpretation); (b) ‘This music animates me to move my body’ (animation); (c) ‘This music is well-performed’ (performance); (d) ‘I like this music’ (liking); (e) ‘This music is of high aesthetic quality’ (quality)” (Frühauf et al., 2013, p. 250). All items were used in the original German version. Each of the three questionnaires was presented on a separate page (pages in the above-mentioned order: EGQ, Aesthemos, and DPQS), along with the corresponding audio example on each page. After completing the three questionnaires for each of the eight musical examples, the playback volume was controlled again with the same procedure as used before. Finally, participants answered the 18 items for the General Factor of the Goldsmiths Musical Sophistication Index (Gold-MSI; Müllensiefen, Gingras, Musil, & Stewart, 2014; Schaal, Bauer, & Müllensiefen, 2014). This questionnaire was used as a descriptive measure to characterize the given sample. Participants took on average 13.2 minutes (SD = 5.2, median = 11.8, interquartile range = 4.6 minutes) to complete the survey.

After completion, participants who provided their email address (which was saved separately from their data set) and agreed to participate in a retest received an invitation one week later. The retest consisted of only some parts of the main study: informed consent, ID generation (to enable matching of IDs from test and retest), playback device information, and volume test. Participants then listened to the same eight musical examples, which were rated using the German version of the EGQ. The retest also concluded with a second control of the playback volume.

The required number of participants was calculated to ensure a minimum statistical power of 1-ß = 80% at a maximum alpha error rate (false positive result) of 5%. The test-retest-reliability was expected to be very high, around r = .90. Furthermore, the correlation between the three questionnaires (as a measure of convergent validity) was expected to be around r = .70. The necessary sample size for the null hypothesis significance testing employing the classical power analysis (Cohen, 1988) would be estimated to be very low for the expected high correlations near 1. Nevertheless, relying on the procedure of classic power analysis in this scenario and only inviting the number of participants suggested by this method was likely to result in large confidence intervals which imply vague point estimates (Maxwell, Kelley, & Rausch, 2008). Therefore, instead of this classical power analysis, the method of Accuracy in Parameter Estimation (AIPE) was employed to control for the width of the confidence interval of the considered measure (Maxwell et al., 2008). For a desired test-retest reliability of r = .90 with a confidence interval width (CIW) of 0.1, N = 62 participants were required. For the desired convergent validity of r = .70 with CIW = 0.2, N = 104 participants were required (calculated by the R package MBESS; Kelley, 2007, 2020; R Core Team, 2020; RStudio Team, 2020).

After collecting data from about 100 participants, the goodness-of-fit measure root mean squared error of approximation (RMSEA) for the confirmatory factor analysis was estimated to be .14 with a 90% confidence interval (CI) of [.08; .20] (reporting a 90% CI is a standard for the RMSEA; Abell, Springer, & Kamata, 2009, p. 160). Therefore, the value was larger than the threshold of .08 to .10 for a mediocre fit (MacCallum, Browne, & Sugawara, 1996, p. 134) and larger than .067 (with 90% CI of [.052, .082]) from the English version of the questionnaire (Senn et al., 2020, p. 57; see section Confirmatory Factor Analyses and Factor Structure in the present paper for further explanation of the RMSEA). To decrease the CI on the measure of RMSEA and to see how many participants needed to be recruited to run a CFA with satisfactory accuracy, another AIPE analysis was calculated. Assuming the RMSEA = .067 of Senn et al. (2020) to be a good approximation of the population RMSEA, the target for the 90%-CIW was 0.07. This opened the possibility of the 90%-CI excluding a RMSEA value higher than .10 and including the value from Senn et al. (2020) with RMSEA = 0.067. The following analysis resulted in 364 additionally required participants. Therefore, we aimed at collecting data from more than 500 participants in total, allowing for the exclusion of outliers and other erroneous data points.

Data were collected by two methods: First, the invitation was circulated by means of mailing lists, social media, and similar methods of convenience sampling. Most respondents had a high educational level and were interested in or had experience with music (see Table S2 in the online Supplementary Material at mp.ucpress.edu for full information). Second, further participants were acquired by a commercial sample provider (mo’web research, 2020) and remunerated for their participation. For this commissioned sample, balanced gender distribution and a minimum age of 18 years were demanded. The typical participants in the providers service to participate in such a study are people with lower level of education or elderly people. These participants were expected to level out the distribution biases of age and educational level, for instance, from the convenience sample back to the representative German norm and were therefore found appropriate for our study.

In total, 570 people participated in the online survey. These data were filtered in a three-step procedure before starting the analysis. An overview of the procedure is provided in Figure 2. All filtering steps and subsequent analyses (if not indicated otherwise) were conducted in RStudio (RStudio Team, 2020) using R (R Core Team, 2020). All employed R packages are mentioned in the corresponding sections of this paper.

First, incomplete datasets (n = 37) were discarded. One participant declared an age of six years and was therefore also excluded from the data set. Therefore, at this point in the study, we had 532 participants. Second, the time to answer the main section of the study (eight stimuli, for each there were three questionnaires on three separate pages, resulting in 24 pages) was calculated for each participant. Participants were excluded from the data set if they took longer than 48 minutes for the 24 pages (i.e., longer than 2 minutes per questionnaire) in that section of the survey. This was the case for n = 16 participants, who were then excluded. The fastest participant took seven minutes for the 24 pages (= 17 seconds per questionnaire on average), which was considered to be realistic. Therefore, no participants were excluded due to implausibly short durations for the completion of the survey. At this point, 516 participants remained in the data set. In a third step, multivariate normality (MVN) was computed to check if the data set met the statistical assumption for the CFA or if the data set deviated from MVN (Sedlmeier & Renkewitz, 2013, p. 680; Tabachnick & Fidell, 2014, p. 666). MVN was assessed by means of the R package MVN (Korkmaz, Goksuluk, & Zararsiz, 2014), using Mardia's (1970) computational method. This analysis revealed that the dataset did not fulfil the requirements of MVN (skewness as well as kurtosis deviated noticeably from normality). Thus, we excluded the outliers contributing to the violation of MVN (n = 61) and repeated the test for MVN. The filtered data showed normal kurtosis, p = .32. However, MVN regarding skewness could not be reached: The data differed statistically significant from skewness = 0, p < .001.

After the three steps of data filtering, the data set consisted of N = 455 participants, which was then the final sample used in the subsequent analyses. Histograms of responses to the six items of the EGQ are displayed in Figure S1 of the online Supplementary Material.

The final sample (N = 455) consisted of a convenience sample (n = 99) and a sample acquired by the commercial provider mo’web (n = 356). To ascertain the correctness of analysing the two subsets as one larger data set, the two subsets were compared (see Table S2 in the online Supplementary Material at mp.ucpress.edu for full descriptive information as well as inference statistical comparisons of the samples).

Participants from the convenience sample were younger than participants from the mo’web sample. All in all, the age of all participants (M = 43.8 years, SD = 15.1) resembled the mean age of Germany’s population (44.5 years, estimated for the 31 December 2019; Statistisches Bundesamt [Destatis], 2020). The difference between this sample and the mean age of the German population showed no distinct deviation from zero using a two-tailed one-sample t-test, t(454) = -0.976, p = .329, d = -0.046 with 95% CI [-0.138, 0.046]. Participants from the convenience subsample showed, on average, a higher educational level than participants from the mo’web subsample. Similarly, musical sophistication was higher in the convenience than in the mo’web sample (Cohen’s d = 1.01). The musical sophistication of our sample can be compared to the German norm sample for the Gold-MSI surveyed by Schaal et al. (2014, p. 445). The entire sample of the present study had a general musical sophistication level of 66.8 (SD = 23.3) whereas the German norm sample showed a general musical sophistication value of 70.4 (SD = 19.94). The mean of our sample corresponded to the 45th percentile of the norm sample. To conclude, the degree of musical sophistication of our sample differed from the German norm sample; however, only with a small but statistically noteworthy effect based on a two-tailed one-sample t-test, t(454) = -3.27, p < .001, d = -0.153 with 95% CI [-0.246, -0.061].

To ensure that the data were more representative for the German population, both sub-datasets were merged into one dataset. All further analyses were conducted based on this joint dataset from the convenience sample as well as the mo’web sample.

The Aesthemos consisted of 42 items grouped into 21 subscales with two items each. These subscales could be grouped into either seven empirical factors as a result of an exploratory factor analysis (Schindler et al., 2017, pp. 24–25) or into four sub-classes identified by literature review and regarding theoretical constructs (Schindler et al., 2017, pp. 28–32). All items relevant for the present study are displayed in Table S3 of the online Supplementary Material at mp.ucpress.edu.

Concerning prototypical aesthetic emotions (i.e., appreciation, other-praising, and self-transcendent emotions; Schindler et al., 2017, p. 11), we surveyed the following four subscales with two items each: (1) Feeling of Beauty/Liking, (2) Fascination, (3) Being Moved, and (4) Awe. The factor score for Prototypical Aesthetic Emotions was calculated as the mean of the eight items from these four subscales (Table 2; for more details, see Table S4 of the online Supplementary Material at mp.ucpress.edu). Additionally, for Factor 4 (Animation), three subscales were surveyed: (5) Enchantment, (9) Vitality, and (10) Energy. Due to a technical problem, one of the items for Enchantment (Item 18: “was enchanted”) was missing from data collection. Therefore, the factor score for Animation was calculated as the mean of the five remaining items.

Due to another technical problem, the Aesthemos items were only collected from 16 of the 99 participants from the convenience sample. However, the items were surveyed completely for the mo’web sample resulting in overall N = 372 responses to the Aesthemos items (Table 2; for more details Table S4 of the online Supplementary Material at mp.ucpress.edu). The participants in the subsamples did not differ noticeably in their responses to the Aesthemos items (neither on the factor of Prototypical Aesthetic Emotions nor on the factor of Animation).

In a second step, analysis on the level of subscales was conducted. Subscale-scores were calculated as the mean of the two corresponding items (for Subscale (5) Enchantment only the one employed item could be used). These scores are presented in Figure S3 for each of the eight musical examples (see Table 1 for the acronyms of the stimuli) as well as the average for all eight musical stimuli.

To examine the factor structure of the six EGQ items, a confirmatory factor analysis was conducted. Three different models were examined regarding their fit to the empirical data. The first model was the target model as reported for the English version of the EGQ by Senn et al. (2020). It consisted of two correlating factors, Urge to Move and Pleasure with three items each. The second model was developed to account for the possibility of two non-correlating and therefore independent factors (same distribution of items as in Model 1). A third model was constructed to test if the groove experience could be operationalized by just one underlying factor. Analyses were conducted in RStudio using the package lavaan (Rosseel, 2012).

The reported fit indices of the CFA follow the general recommendations by Kline (2016, p. 269): χ² with degrees of freedom and p values are reported as well as values for RMSEA, CFI, and SRMR. They are supplemented by the TLI and BIC coefficients. The χ² goodness-of-fit statistic tests the null hypothesis that the data covariance matrix and the reproduced covariance matrix are the same (Abell et al., 2009, p. 158 f.). Therefore, rejecting the null hypothesis is a sign of poor fit. Additionally, it should be considered that the χ² value is directly influenced by the sample size. The comparative fit index (CFI) is an addition to the χ² statistic and also takes the respective degrees of freedom into account. It can range between 0 and 1 with a higher value indicating a better fit (Abell et al., 2009, p. 159). The benchmark for an acceptable fit is .90 and for an excellent fit .95. The Tucker-Lewis Index (TLI) regards the ratio between the χ² statistic and the degrees of freedom. It also ranges between 0 and 1, and the same benchmarks are applied as for the CFI (Abell et al., 2009, p. 159; Hu & Bentler, 1999, p. 27). The Bayesian information criterion (BIC) is unstandardized and can, therefore, only be interpreted by comparing the fit between different models. Smaller values indicate better fit (Abell et al., 2009, p. 168). The RMSEA is based on the degree of noncentrality of the χ² statistic. If the model fits the data perfectly, then RMSEA is 0; thus, a smaller value indicates a better fit (Abell et al., 2009, p. 159 f.). A good fit is suggested for values of RMSEA < 0.05; an acceptable fit for RMSEA would be < 0.08 (Kenny, Kaniskan, & McCoach, 2015, p. 488). The standardized root mean square residual (SRMR) takes into account the discrepancy between the data covariance matrix based on estimated parameter values. A value of 0.10 or smaller is interpreted as a good fit (Kline, 2016, p. 278).

Test-retest reliability of the EGQ was calculated from n = 65 participants. In addition, internal consistency of the scales was reported by calculating Cronbach’s α for the two factors. Third, construct validity was examined by calculating correlations between the two EGQ factors, the two factors of the Aesthemos, and the DPQS. The results are discussed in terms of convergent validity.

Participants evaluated the eight musical stimuli using the following three scales: the German EGQ, a part of the Aesthemos (Schindler et al., 2017) and the Drum Pattern Quality Scale (DPQS, Frühauf et al., 2013). The descriptive results of the responses for all eight stimuli are presented in Table S5 in the online Supplementary Material at mp.ucpress.edu. Descriptive results averaged across all eight stimuli are presented in Table 2. Values for the entire sample are reported. Separate values and statistical analysis for the two sub-samples (convenience and mo’web) can be found in the extended version of Table 2 in the online Supplementary Material (Table S4).

Analysis of the skewness of the responses to the six items of the EGQ revealed that particularly the three Urge to Move items showed skewed distributions (skewness = .40, .53, .45, respectively—therefore, right-skewed) compared to the three items of the pleasure-scale (skewness = -.04, .04, -.08, respectively).

A scatterplot of the ratings for Urge to Move and Pleasure is displayed in Figure 3. In addition to the single observations (in semi-transparent symbols), the large symbols represent the mean ratings of the eight musical examples with 95% confidence intervals. To facilitate the comparison of these findings with the results from the English original version of the EGQ, the corresponding Figure from Senn et al.’s (2020) Experiment 2 is cited in Figure 3b of this study.

Confirmatory factor analyses were calculated for all three models: first, the target model (two correlating factors); second, two independent factors; and third, one factor. The results with corresponding measurements for the goodness of fit are displayed in Table 3. Comparing the three models regarding their χ² goodness-of-fit statistic, we concluded that Model 1 with the smallest χ² value indicates the best fit by this measure. Model 1 showed a CFI of .99, which exceeded the benchmark of .95 for an excellent fit (Hu & Bentler, 1999, p. 27). Models 2 and 3 did not reach the benchmark for the CFI. Similar to the CFI, Model 1 reached the excellent fit regarding the TLI (benchmark: .95), and Models 2 and 3 failed to reach an acceptable fit (benchmark: .90). Regarding the BIC, Model 1 showed a better fit to the empirical data than did Models 2 and 3. The RMSEA values of all three models tested here exceeded the benchmark of 0.08. In comparison, Model 1 showed by far the smallest RMSEA of all three models. The 90% CI of Model 1 included the benchmark of 0.08 for an acceptable fit. Models 1 and 3 met the benchmark of the SRMR (0.10) and therefore showed a good fit evaluated by this criterion.

To sum up the results of the confirmatory factor analyses, Model 1 showed the best fit to the collected data, evaluated by all the fit indices. CFI and TLI attested an excellent fit, and SRMR a good fit. The χ² statistic and the RMSEA did not show the desired results but achieved the best values for Model 1 compared to the other two models.

The results of the CFA for the target model are displayed in Figure 4. Each factor coefficient was high (> .93), and the two factors Urge to Move and Pleasure correlated strongly (at r = .90).

All reliability measures were obtained by the software jamovi (The jamovi project, 2020).

To measure the test-retest-reliability, we had some participants repeat the evaluation of the eight musical stimuli with the EGQ after at least one week. Participants took the retest on average 11.5 days after the first data collection (see Table S2 in the online Supplementary Material at mp.ucpress.edu for details regarding the sub-samples). The correlations between test and retest for both factors were as follows: Urge to Move: Pearson’s r = .867 (p < .001, n = 65), 95% CI [.790, .917]; Pleasure: Pearson’s r = .908 (p < .001, n = 65), 95% CI [.853, .943]. Therefore, the German version of the EGQ showed excellent test-retest reliability. For scatterplots of test and retest for the two factors of the EGQ, see Figure S4 in the online Supplementary Material at mp.ucpress.edu.

As a measure of internal consistency, Cronbach’s α was calculated (see Figure 4). For the Urge to Move scale, mean internal consistency was α = .967. For all three items, removal of one item did not improve the internal consistency. For the Pleasure scale, mean internal consistency was α = .986; removal of one item also did not improve the internal consistency.

The Aesthemos and the Drum Pattern Quality Scale were employed in this study to determine convergent validity with the EGQ. Therefore, correlations between the questionnaires were calculated (see Table 4). The two factors Urge to Move and Pleasure of the EGQ correlate strongly with Factors 2 (Prototypical Aesthetic Emotions) and 4 (Animation) from the Aesthemos (all correlations .83 < r < .87) and with the DPQS (r = .78 and .87 for Urge to Move and Pleasure, respectively).

This study presents a German translation and further validation of the Experience of Groove Questionnaire (EGQ). The model structure of the original questionnaire could be confirmed. The reliability of the German EGQ was estimated to be high: Responses were reliable over time (high test-retest reliability), and responses to the two scales were highly consistent (high Cronbach’s α). These high internal consistencies indicate that some of the items from both scales might be redundant. Nevertheless, future additional validation of the questionnaire should be observed before shortening the EGQ.

Validity has been determined by comparing responses to the EGQ with responses to other questionnaires: High correlations with aesthetic emotions (as surveyed by the Aesthemos) and the perceived drum pattern quality (as surveyed by the DPQS) were found. A possible reason for this high convergent validity may be the respondents’ disinterest or inability to distinguish between the subtle differences between items that capture the constructs of being animated by the music and feeling the urge to move. The same argument holds true for the differences between experiencing pleasure while listening to music and liking it. Because of the high number of items and a possible repetitiveness due to the seemingly similar constructs measured by the items, respondent fatigue may have contributed to a less nuanced answering behavior (Ben-Nun, 2008). Furthermore, the difference between liking music and experiencing pleasure when listening to music may be hard, if not impossible, to distinguish for the participants in that they were asked to distinguish between different mechanisms underlying their aesthetic judgements. Put in simplified terms, liking describes the conscious effort to process sensory sensation and pleasure describes the sensory sensation itself (Brattico, 2015; Graf & Landwehr, 2015, 2017). One could argue that this difference is observable in the divergence between physiological measurements and self-reports but relatively difficult to assess by the participants themselves. A second reason for similar answering patterns to items P1 and P3 (r = .96, 95% CI [0.95; 0.97]) assessing experienced pleasure and likability of a stimulus may be that our translation did not capture the difference of liking and experiencing pleasure well enough.

The responses to the two factors Urge to Move and Pleasure from the EGQ were analyzed for the eight musical stimuli (see Figure 3). The skewed responses to the move-scale might be based on an unbalanced choice of musical examples: The stimulus pool included at least two stimuli with no clear pulse (“Machine Gun” by the Peter Brötzmann Octet [BRO] and “My Pal Foot Foot” by The Shaggs [SHA]) and at least one with an ambiguous pulse (“Sunrise” by the Angelite & Moscow Art Trio [ANG]). We argue that participants could not synchronize to the music in these examples and therefore tended to give the lowest score on the Urge to Move items. Additionally, for a Likert scale with a low mean value, the distribution is likely to be positively skewed as the answering options (in this case) are limited to seven options. This plausibly resulted in the skewed distributions of the Urge to Move items.

The experience of an Urge to Move while listening to the music and Pleasure are strongly connected: Mean ratings per stimulus lie near the grey main diagonal in Figure 3. Musical examples ANG (“Sunrise”) and MAH (Mahler’s Symphony No. 3) deviate most strongly from the main diagonal; their pleasure-rating is higher than participants’ urge to move in response to the music. These musical examples were expected to elicit a high Pleasure but a low Urge to Move rating (see Table 1). As expected, Stevie Wonder's “Superstition” (WON) and “Bala” from Bassekou Kouyate (KOU) show a relatively high rating on both scales (although this is much more pronounced for WON than for KOU), and BRO (“Machine Gun”) and SHA (“My Pal Foot Foot”) show a low rating on both scales. While planning the development of the original EGQ, Senn et al. (2020) chose “Tchip Tchip” by Cash & Carry (CAS) and “Hamdouchi” by Guinia and Sanders (MAL) because they expected listeners to experience much urge to move in response to the music but low pleasure. In contrast to this expectation, both stimuli were rated with about the same amount of urge to move and pleasure in their study (see Figure 3b). The same proved to be true for the present study (see Figure 3a). This suggests that pleasure might be a precondition for the experience of an urge to move in response to that music (for further discussion see Senn et al., 2020, p. 58). This could also explain why the Urge to Move scale is highly correlated with the second Aesthemos factor (Prototypical Aesthetic Emotions).

Both the English and the German version of the EGQ were developed with samples reflecting the general population (see Senn et al., 2020, Experiment 2 for the English EGQ). The same musical examples were used for the two studies. The English version was shown to adhere to a factor structure with two correlating factors (Urge to Move and Pleasure) with three items each (see Figure 3 in Senn et al., 2020). The same structure could be confirmed for the German version by a CFA (see Figure 4). In the German version, the two factors correlated even more strongly than in the English version. The path coefficients for the German EGQ were even a bit larger than for the English version; uniqueness of the items was lower. The model fit for the English version was—in terms of the CFI—comparable to the German version. Regarding the RMSEA, the English version showed a better fit to the target model than the German version. The internal consistencies of the two scales of the English version were estimated to be Cronbach’s α = .92 and α = .97 for Urge to Move and Pleasure, respectively (Senn et al., 2020, p. 57). The German version even exceeded these values for both factors (see Figure 4).

The method for matching datasets from the retests to the corresponding dataset of the main study was not error-free: Due to the seemingly incorrect entering of IDs by some participants, a small number of datasets could not be matched and got lost for the subsequent analyses of test-retest reliability (see section Measures and Procedure of the Validation Study).

The analyses revealed a relatively high RMSEA value for the goodness of fit of the CFA. This unexpected value can be traced back to extreme responses to some stimuli in the Urge to Move scale. We assume that the stimulus pool might not have been well-balanced, which could result in unexpected responses. Nevertheless, the good SRMR suggests that the high value of the RMSEA might also have been influenced by the low degrees of freedom (Kenny et al., 2015, p. 500).

Our study conducted an initial validation of the EGQ (for the original questionnaire, only construct validity was reported). However, the questionnaire should be subject to additional psychometric evaluation in different settings (for example, using stimuli from just one genre or a greater variety of genres, as well as ratings by selected demographic groups as respondents). As the personal background in music is important in shaping the mental processes involved in the experience of groove (Senn et al., 2019), it is assumed that the individual rating patterns can diverge between different cultures. A participant from a cultural background differing from the sample used in this study could, for example, have been raised in a society in which the function of music and dance (a summary of these functions from an evolutionary standpoint can be found in Richter & Ostovar, 2016) is different from the Western concept of music and dance. As a consequence, the response patterns of different cultural subgroups could diverge systematically on the factor Urge to Move. Therefore, we share the view of Witek et al. (2020) that there is a need for more cross-cultural research on this topic to further our understanding of cultural communalities and differences in the experience of groove. Due to the high internal consistency of the two scales, one could conclude that some items might be redundant for the measurement of Urge to Move and Pleasure. However, in reference to the extreme responses to some stimuli, the need for a more diverse stimulus set becomes evident: Future stimuli should vary with regard to the musical properties that are expected to influence our groove experience: genre, tempo, beat salience, event density, microtiming, rhythmic variablilty, and syncopation, for example. Furthermore, we recommend the use of the full questionnaire until these redundancies have been confirmed by further studies.

We suggest that the EGQ is of value as an independent variable for neurophysiological measurements of groove experience. Selection of musical stimuli could be pretested, and their groove characteristics could be controlled on a reliable basis. We also suggest extending the number of language versions of the EGQ so that cross-cultural differences in groove experience can be investigated in various cultures. Concerning Senn et al.’s (2020) point on the further validation of the EGQ, the overall explanatory power of the EGQ could be further enhanced by correlating results from the EGQ with physiological measurements (e.g., EEG, fMRI, pupillometry, and motion capture). Thus, this research contributes to the validation of the EGQ as well as studies of differences or similarities in subjective and objective measurements of groove.

To facilitate future research with the EGQ, the R-package groovescale has been developed (Sander & Labonde, 2020). It allows for an integration of both EGQ versions (English and German) in the R environment and was conceived using the psychTestR framework (Harrison, 2020). Based on this package, one can build a test session with participant IDs, save the results with automatically calculated scores for both factors, and include the EGQ in online experiments. For further information regarding the download, installation and functionality of the package, see its online repository (https://github.com/KilianSander/groovescale).

We consider our study to be a contribution to the enlargement of multilingualism for standard inventories in the domain of music cognition research. Over the last years it can be observed that the earlier practice of “self-made” translations of inventories (discounting all standards of translation and validation) no longer seems to be acceptable for high quality research. Instead, we find an increasing awareness that high psychometric standards should not only be met for the development of testing procedures and inventories but the same applies to the subsequent publication of versions in different languages. In the long run, we are convinced that this line of research might open new perspectives for the growing field of cross-cultural research.

The authors would like to thank Kilian Sander for his support in the development of the groovescale R-package and his valuable feedback.

A part of this paper was presented as a poster at the virtual 36th annual conference of the German Society for Music Psychology (September 3-6, 2020).

Upon request, the data and R files for analyses of this study can be sent by email for reasons of further research.