Measure of Emotional Episodes with Music (MEEM): Development and psychometric evaluation of a modular instrument

One major concern for the initial scale development process is sample size. Commonly cited guidelines suggest minimum samples of 100–150 for CFA [54]. Some researchers argue that samples as small as 100 are adequate when item communalities are high [55]. Here, we evaluated communalities to determine whether our initial sample size of 100 participants per sub-experiment was adequate. Another aspect to consider is the participant to parameter ratio [56]. Considering our models ranged from 19 to 27 estimated parameters, we exceed the recommended minimum ratio of 5:1 with a sample size of 100 [56]. In each sub-experiment, we asked the participants to rate items against vignettes—one vignette for each sub-construct that captures transient state rather than a stable trait theoretically designed to elicit the target state —so we obtained multi-level data from each participant. The repeated observations were therefore not intended to increase the effective sample size but to sample the latent construct across multiple situational contexts, reducing vignette-specific variance and allowing the CFA to estimate the construct independently of any single eliciting scenario. Such multi-level data can be modelled with a CFA technique using robust standard errors that adjust for the nested data at the participant level.

For the task, participants rated the item sample for the episode construct (22 items in the case of EDR, see Supporting Material) after reading a vignette (see S2 Appendix 2). The presentation order for vignettes was randomised in each of the five surveys. Each participant was presented with all corresponding vignettes for that construct in a given survey. A participant doing the EDR survey, for example, would see three vignettes and answer the 22 EDR items after each vignette. Attention check questions were intermixed within the randomised order of items for each survey. After answering all of the items for a given vignette, participants would indicate how relevant they considered the vignette to be based on their own life experience. After completing all the items for each vignette in a given survey, the participants were routed back to prolific. Participants were not allowed to complete another sub-experiment if they had already participated in one of the five surveys.

To determine the goodness of fit of the factor structure of these previously evaluated and ranked items representing sub-constructs (2–3) for each of the five constructs, we applied multi-level confirmatory factor analysis (CFA) using the lavaan package in R [57]. For each sub-experiment (one sub-construct), observations were nested within individuals to vignettes (2 or 3 vignettes, one for each sub-construct), and the multilevel structure was modelled by specifying participants as clusters. This multi-level modelling approach allows us to evaluate whether the total variance in the observed items is due to variance in responses to the vignettes (within-person variance) versus how much the structure varies between different people. Two (FM, CB, PEP) to three (EDR, AIA) latent factors were specified, each measured initially by all available observed items (S2 Appendix 2) or by exploring a model using the three items with the highest factor loadings for each sub-construct. The purpose of the three items per latent factor was to explore a more parsimonious model following the practical guidance [58,59]. This vignette design is consistent with our theoretical premise that affectual responses during emotional episodes are context-dependent [34,36], though the present manuscript is not intended to isolate any specific situational features affecting within-person variability. Rather than treating this variability as pure measurement error, we retain within-person variance in the multi-level model and report relevant reliability metrics but do not interpret it as a within-person model. All five models were estimated using the robust maximum likelihood estimator (MLR), with participant ID specified as a cluster variable to adjust for non-independence. Considering participants responded to multiple vignettes in a random order with the same construct relevant item content, the use of MLR corrected for our nested/clustered multi-level approach, potential standard errors, and the non-normality statistic. The particular vignettes, however, are not assumed to be naturally occurring higher-order factors, but rather allow for repeated measurement occasions. Model fit was evaluated using a standard combination of absolute and comparative fit indices (, CFI, TLI, RMSEA, and SRMR). All data and analysis notebooks are available at https://zenodo.org/records/21909314. This multi-level modelling approach aligns with our underlying theory which specifies that emotional episodes, the affectual response and the contextualisation of this response, will vary depending on what situation a listener finds themselves in [34,36]. While the constructs are meant to operate in this manner, it is not known how the situational variance would manifest in a given situation for a particular person.

The sub-constructs each had one vignette (12 total) which were designed to be typical situations that would garner responses to items associated with a specific sub-construct (see S1 Appendix 1). The episode construct EDR, for example, had three vignettes corresponding to three sub-constructs: “You are attending a celebration where music is the heart of the party. You are able to join in the fun which is aided by the music throughout the event” for Enjoyment, “It was pointed out that you missed an important detail in your work which you now have to redo. You put on a music playlist to drown out self-criticism as you fix the mistake” for Distraction, and “After a long and demanding week, you take time to rest by listening to music. The stress melts away and you start to recover as you listen along” for Relaxation (see S1 Appendix 1 for all the vignettes).

Each episode construct was probed with appropriate items obtained from expert content evaluations ([36]; 22 for EDR, 14 for FM 16 for CB, 20 for PEP, 15 for AIA). Each item is worded to reflect current or recent experiences (e.g., The music made me feel calm or relaxed; see S2 Appendix 2) and it is expected that the instrument would be administered to people during or immediately following experiences with music. Considering assumed response behaviour, participants indicate whether they are experiencing the measured attribute described by each item on a five point scale from Strongly disagree (1) to Strongly agree (5), following best practice recommendations [40,60].

One content domain, the AIA’s Curiosity sub-construct had less than three items following the expert content evaluations. Typically, if there are less than three items for a supposed factor then there are potential issues that relate to factor complexity and no zero elements in the coefficients ([61], p. 244). These were, however, not issues with the factor case. Nevertheless, to circumvent potential issues of only having two items as indicators of a sub-construct before conducting the CFA, we included borderline unacceptable items from the expert content evaluations to bolster the number of items in this sub-construct.

As a demographic variable, we collected an indication of participants’ musical interest and experience through one item from the Ollen Musical Sophistication Index (OMSI; [62]). Common use of this single criterion indicator is the result of work of investigating two popular music sophistication indexes to determine a single best estimating item to indicate overall sophistication [63]. The resulting OMSI item, “In terms of your musical interest and expertise, which title best describes you?”, was best and is widely used to demarcate between musician and non-musician groups or between those who are assumed to have little, moderate, or high levels of interest in music. There are six self-selected options, “non-musician”, “music loving non-musician”, “amateur musician”, “serious amateur musician”, “semi-professional musician”, and “professional musician”. In this article, we use the OMSI variable to indicate between musician and non-musician groups.

There were 87 items in total assessed across the five CFA processes (see S2 Appendix 2 for detailed descriptives). The outcome of the CFA for the EDR sub-experiment, all three sub-constructs appear to have at least three suitable item candidates with factor scores ranging from .54–.79 for Relaxation, .73–.84 for Enjoyment, .64–.77 for Distraction. The Connection–Belonging two factor structure between the Group Cohesion / Socialisation and Reducing Loneliness sub-constructs saw candidate items for the former ranging from .61–.85 while candidates ranged from .84–.89 for the latter. The two factor structure for the Focus–Motivation construct was supported but items diverged slightly from the prior expert classification [36]. The Energy Control sub-construct had two strongly loaded items from the Focus sub-construct which both mention motivation; (F1) “I used the music to motivate myself”, and (F7) “The music motivated me to finish the task I had to do” (see S2 Appendix 2). With this in mind, the top three candidate items for Energy Control ranged from .84–.87 (M1, M3, & F1/M9) and suitable candidate items ranged between .70–.87 for Focus. Strong loading candidate items for the Personal Emotional Processing construct ranged .62–.82 for Reflection / Coping sub-construct and .64–.85 for the Expressing Feelings sub-construct. Finally, the Aesthetic–Interest–Awe three factor structure had suitable items with strong factor loadings for Being Moved / Spirituality ranging .61–.89 and .62–.82 for the Aesthetics sub-construct. The Curiosity sub-construct was identified coincidentally by a factor with the two items which were originally identified by the experts as best. These items had strong loadings (.83 for I2 and .85 for I1, see S2 Appendix 2) and, despite adding additional items, were again the top two indicators for the Curiosity sub-construct.

When a similar analysis is carried out using the three (or two in the case of the Curiosity sub-construct) items scoring highest in factor loadings, good model fits are obtained across the board (EDR TLI = 0.975, =51.85, p < .001, CFI = 0.984, RMSEA = 0.062, SRMR = 0.038, FM TLI = 0.985, =16.58, p < .035, CFI = 0.992, RMSEA = 0.073, SRMR = 0.018, CB TLI = 0.986, =13.85, p < .086, CFI = 0.992, RMSEA = 0.060, SRMR = 0.041, PEP TLI = 0.979, =15.92, p < .044, CFI = 0.989, RMSEA = 0.070, SRMR = 0.027, and AIA TLI = 0.966, =54.76, p < .000, CFI = 0.979, RMSEA = 0.086, SRMR = 0.051). The full breakdown of the items, sub-constructs and metrics for each construct is available at https://zenodo.org/records/21909314. All models demonstrated acceptable to excellent fit. CFI values ranged from .979 to .992 and TLI values ranged from .966 to .986, exceeding the recommended threshold of .95 [66]. RMSEA values ranged from .060 to .086, with four of five models meeting the stringent criterion of .06 and all falling within the acceptable range of .08 [67]. SRMR values ranged from .018 to .051, all well below the recommended cutoff of .08 [66]. The chi-square tests were significant for two models (EDR and AIA), which is common with adequate sample sizes and is not considered problematic when other indices indicate good fit [68].

To explore measurement invariance across vignettes, we compared models with increasingly restricted models (configural, metric, and scalar invariance) for each construct [69]. With our vignette design, we expect that the configural invariance for each construct will be achieved to demonstrate that the items represent the latent constructs similarly. We also anticipate that the measurement invariance holds for metric invariance, as it would suggest comparable relationships with the latent variables across the vignettes. However, the more restrictive scalar invariance is likely to be unachievable as the vignettes themselves produce variation in the intercepts even if the structure is otherwise similar. We analysed measurement invariance with a robust comparison between the original and each restricted model (see [70]). All constructs pass the configural invariance test, suggesting that the overall structure is valid across the vignettes when coefficients are allowed to freely vary (EDR p = .942, FM p = .612, CB p = .995, PEP p = .977, AIA p = .121, each p value comparing the full model with configural model with robust comparison). Moving on to metric invariance tests, where loadings are fixed to be the same across the vignettes, 3 out of 5 constructs pass this comparison (FM p = .897, CB p = .156, and AIA p = .614). However, none of the constructs passed scalar invariance tests, where not only loadings are constrained to be equal across the vignettes but the intercepts are also fixed to be equal across vignettes. The scalar invariance could be attempted to be recovered by allowing some items to be released from the constraints in the model, but our theorising about the vignettes suggested that the item loadings can vary across the vignettes despite them maintaining the overall structure across the construct, so neither scalar nor metric invariance was a desirable quality in our design.

While our empirical process has indicated suitable candidate items for each of the constructs, the reduction of content has influenced the existing conceptualisation for some of the sub-constructs from the prior assumptions [36]. Our choice of approach, both in terms of methodological approach and retention of the 3 best items, has emphasised narrower aspects of the sub-constructs. With this consideration, it is important to reflect on whether the items still target the intended content domains stipulated by Kirts and colleagues [36].

Energy Control, as a sub-construct of Focus–Motivation, was originally operationally defined as “Music used to control levels of intensity or drive, pumping up, motivation, stimulation” [36], but the resulting best items primarily referred to a narrower concept akin to “motivation”. Hence we have relabelled the sub-construct hereafter as Motivation to address this. A similar result occurred with two facets of Aesthetics–Interest–Awe; “Being Moved / Spirituality” and “Aesthetics” sub-constructs. Originally, “Being Moved / Spirituality” referred to “Music used to connect to a higher ideal or abstract feeling of non-social connection, religious activity, feeling or seeking strong emotions” ([36]) but the resulting items only contain spirituality related items, hence we relabel the sub-construct as Spirituality. Similarly, Aesthetics referred to “Music used to achieve or create a desired ambience, fit to the ‘mood’” but the best items refer primarily to a narrower concept of beauty. We therefore labelled the sub-construct as Beauty. The last relabelling is for the Reflection / Coping sub-construct of Personal–Emotional–Processing which originally refers to when “Music [is] used to remember or to associate to other objects, provide a sense of comfort, processing experiences”. The resulting best items from the empirical evaluations refer to comfort rather than to coping or reflection, so we renamed the sub-construct Comforting. Finally, for consistency, we simplified an existing label from Group cohesion/socialisation to Group bonding to avoid confusing labels. The purpose for relabelling of the sub-constructs is to acknowledge that the empirical data and the analysis procedure has influenced, mainly trimmed and focussed, the emerging latent constructs. To retain clarity, we have updated the labels of these specific sub-constructs to reflect the more narrow aspects of the content which emerged following Experiment 1.

The three latent factors corresponded to the supposed sub-constructs representing EDR, Enjoyment (E), Distraction (D), and Relaxation (R), and were captured with three items for each latent factor. The standardised coefficients are shown in Fig 2. The CFA model was excellent, all measures of fit provided evidence that the latent factors and the items capture the variance of the statements well (=89.9, df = 72, p = .074, TLI = 0.976, CFI = 0.984, RMSEA = 0.050, SRMR = 0.046). Items related to enjoyment such as E1, E6 and E7 loaded strongly on the same factor, as expected. The items predicted to be strong for the Distraction sub-scale, such as D1, D6 and D7, received positive loadings to the same latent factor. Also the Relaxation sub-scale obtained consistent positive loadings from the relaxation items (R1, R3, and R7). Tests of measurement invariance compared the model across the different groups, here (and for all of the other scales) we use the vignettes as our grouping. With an equal factor structure across the groups (i.e., configural invariance), the EDR scale passes this test based on the model fit. Adding stipulations that the factor loadings should be equal across groups (metric invariance) and that both loadings and intercepts should be equal (scalar invariance) were also tested. The EDR scale passes the metric invariance test using the robust (p = .43) but fails the scalar invariance test (p < .001).

The scale for Focus–Motivation (FM) consists of two sub-scales with each represented by three items. The standardised loadings of the CFA model for FM is shown in Fig 3 but the means and standard deviations of ratings are found in S3 Appendix 3. The model provided an excellent fit to the data (=20.56, df = 16, p = .196, TLI = 0.987, CFI = 0.993, RMSEA = 0.053, SRMR = 0.033) on all metrics. The items for the motivation and focus sub-scales were equally relevant with both sets of items obtaining high positive loadings within their respective latent factors. This scale passes the metric invariance test (p = .55) but fails the scalar invariance test (p < .01).

The Connection–Belonging (CB) scale consists of two latent factors, each represented by 3 items with sub-scales labeled as Group bonding (G) and reducing Loneliness (L). Fig 4 suggests that the standardized loadings indicate the correct items being highly loaded on each latent factor. The mode fit is excellent on all fit metrics (=22.14, df = 16, p = .139, TLI = 0.978, CFI = 0.989, RMSEA = 0.062, SRMR = 0.040). The CB scale also passed the metric invariance test (p = .928) and passed the scalar invariance test (p = .289).

The scale for Personal Emotion Processing contains two latent factors Expression (X) and Comforting (C), each represented by three items that loaded clearly onto appropriate factors (see Fig 5). Model fit is very good (=26.87, df = 16, p = .043, TLI = 0.961, CFI = 0.979, RMSEA = 0.082, SRMR = 0.040) although the absolute fit is borderline acceptable ( should provide a fit above p < .05) and RMSEA value exceeds the suggested threshold of 0.08). Considering the high comparative fit indices (CFI and TLI) and the other absolute fit index (SRMR) being very good and under the recommended values, we can interpret the model as providing a good fit. The PEP scale passed the metric invariance test (p = .279) but failed to pass the scalar invariance test (p < .01).

The scale for Aesthetic–Interest–Awe consists of three latent factors, with one consisting of two items and the other two having 3 items. The sub-scales were labeled Spirituality (S), Curiosity (I), and Beauty (B). Looking at the CFA model output (Fig 6 for standardised loadings), we observe healthy positive loadings for the Spirituality, Curiosity, and Beauty sub-scales. Overall, the CFA model yields a very good fit (=80.10, df = 51, p = .006, TLI = 0.961, CFI = 0.976, RMSEA = 0.076, SRMR = 0.046), even though the absolute fit index () fails to support the model. This seems to be due to just 2 items representing the Curiosity sub-scale (I1 and I2) and the poorer fit of the item B5. Additionally, the Aesthetic–Interest–Awe scale fails to pass the metric invariance test (p = .022) and also fails to pass the scalar invariance test (p < .01). Looking at the loading structure in detail, it is particularly the item B5 that causes minimal issues to the model. However, there is no major gain from eliminating this item from the model to only obtain metric invariance.

Next, we applied discriminant validity to the scales, and used the updated criteria from Rönkkö and Cho [75] for nested mode comparison to check whether latent variable correlations are sufficiently low so that the sub-constructs can be considered to represent distinct constructs. This calculation was done using additional functionality to lavaan library added by SemTools [76]. Table 1 demonstrates the overlap of the sub-scales within the five scales. All nine pairwise comparisons yielded significant chi-square differences ( range: 21.1–189.9, all p < .001), indicating that no two factors could be collapsed into a single latent factor. Even though one scale, namely PEP, receives high interfactor correlation (.745 [.609–.881]), this is considered sufficiently differentiated, and the rest are firmly between .32 and .70. These results collectively suggest that sub-constructs within each scale are sufficiently discriminable from each other.

As a variant of discriminant validity, we also obtained Heterotrait-Monotrait Ratio (HTMT [77]; sometimes called disattenuated correlation using parallel reliability, see [75]) scores across sub-scales as HTMT has been shown to be less biased estimator of discriminant validity than McDonald’s or the comparisons [78]. We calculated HTMT ratios for sub-scales (shown in Table 2), which yielded a range of values between 0.272 to 0.809 with a median of 0.484. As the HTMT values under .85 indicate adequate discriminant validity [77], and hence the sub-scales do not significantly overlap. To test the discriminant validity between the five scales, we would need the second order model to have provided a good starting point for the model. As the model fit was poor, exploring the discriminant validity at the level of these high level constructs is not really recommended, but just for the sake of comparison, calculating the HTMT ratios for the five scales produced values between 0.459 and 0.871 with a median of 0.692. The highest HTMT value was observed between PEP and EDR (0.871), suggesting similarities in item wordings and potential overlap between these two scales. However, this interpretation should be qualified by the caveat that the underlying structural model is intended as a diagnostic tool rather than a reliable representation of the data.

To calculate reliability (composite reliability) for all sub-scales, we use the first-order model to calculate reliability and discriminant validity indices. This allowed us to calculate the composite reliability scores (for McDonald’s see [79]; see also [80]) for all 12 sub-scales simultaneously using semTools [76]. The scores range from 0.769 [CI 0.739–0.794] (D) to 0.903 [CI 0.891–0.915] (C), and all but two (D, B) provide good reliability (>0.80) and those two (D = 0.769 [CI 0.739–0.794] and B = 0.783 [CI 0.761–0.804]), acceptable reliability.

The results, presented in Table 3, suggest that most sub-scales occupy distinct locations within the affective circumplex, with correlation patterns that align intuitively with their conceptual meanings. For example, Enjoyment is positively correlated with positive-valence emotions and negatively correlated with negative-valence emotions, and is primarily associated with higher arousal. In contrast, Distraction is less positively valenced and shows small but positive correlations with negative-valence quadrants (r = 0.12 and 0.14 for high- and low-arousal negative valence, respectively). Relaxation shows strong associations with positive-valence, low-arousal emotions, as expected. More comparisons, than what we are able to perform here, will be needed across a variety of real-world applications, methods, and comparable instruments to faithfully review the convergent validity between MEEM and affectual states.

Among other limitations, there are specific limitations related to our methodological decisions and analytical choices. To situate and contextualise the emotional experiences participants had in mind when rating the items, we employed the experimental vignette method (see [51]). However, we used a single vignette for each sub-scale within an episode, whereas a factorial vignette design could have been employed to disentangle and diagnose the effects of different vignette elements [85]. Using vignettes in our experimental design, however, highlights another concern for ecological validity. While we achieved good markers of reliability, generalising the experimental vignette results to real world scenarios should be approached with healthy scepticism. Such practice is common when determining the suitability of an instrument in a particular study design or for a specific purpose. In the short term, those who wish to apply MEEM should seek to review and detail validity evidence along with their primary study aims. The community can build upon the validity evidence provided here, by investigating different real-world scenarios, testing effects of administration duration from the episode experience, and assessing response patterns between laboratory and ESM or EMA designs. While the limitation is an issue for the present work, in the future meta reviews of studies using MEEM will assist in developing informed guidelines for administration in a wide array of ecological settings.

In our analyses, we adopted a confirmatory approach to factor analysis (confirmatory factor analysis, CFA), as the items and constructs were informed by two rounds of expert evaluation, including rankings of item relevance for each sub-construct [36]. A more traditional approach would have been to conduct an exploratory factor analysis (EFA) followed by CFA. However, given the extensive prior knowledge of the items and constructs, our primary aim in the first experiment was to evaluate whether the observed factor structure was consistent with theoretical predictions and to select the best-performing items to represent each factor [44,46]. For that reason, we chose an exploratory CFA approach in the first experiment to 1) assess whether our assumed structure from the theory (see [34]) and content validity (see [36]) was coherent with the data we gathered here, and 2) to explore models with different collections of items, in our case we kept this simple and only took the top three items from each sub-construct.

In Experiment 1, we observed a clear reduction in conceptual breadth from the large pool of items identified by experts to a smaller set of items that best captured the latent factors [36]. Although such reduction is often necessary when developing scales with relatively few items [58], this narrowing process may introduce construct drift and thus pose a risk to instrument validity. To account for such issues, we retested all of the scales with a new sample of participants in Experiment 2. Again, we employed CFA but our use of the technique in Experiment 2 followed a more conventional confirmatory approach that is common for establishing whether the structure uncovered in an explored model can be replicated [44,46]. In light of alternative statistical approaches, such as network analysis [86], the limitation of our approach is that it is a preliminary assessment of the structure under these experimental conditions. Future work should build upon the development of the MEEM scales to test causal systems with such network approaches like temporal time series [87]. Nevertheless, the CFA approach we chose has the advantage of producing a focused and optimised instrument upon which future work can further investigate validity and reliability.

In Experiment 2, we conducted CFA to corroborate the factor structure using an independent dataset. One specific limitation concerns interpreting the two item sub-scale (Curiosity, see S2 Appendix 2). While we mitigated potential issues with identification by adding supplementary items, those who wish to use the AIA module should seek to add another indicator for the Curiosity sub-scale. Our recommended best option would be the third highest loading item from Exp. 1 (I3, see S2 Appendix 2) as a candidate to reinvestigate and mitigate potential identification issues. At this stage, we also examined convergent validity using two external measures (HAAS and GEMIAC). Although it would be informative to extend the convergent validity to instruments that capture a wider range of functional uses of music (e.g., the Healthy – Unhealthy Music scale [15]; Brief Music in Mood Regulation scale [17]; Music Use and Background Questionnaire [10]; Barcelona Music Reward Questionnaire [12]), all of these instruments assess general, trait-like tendencies rather than experiences tied to specific situations or contexts. None of the trait based measures would be suitable to provide validity evidence for states [39,88]. A limitation of the present development process concerns the lack of preexisting state-based instruments which can serve as a benchmark in the particular topic area. In particular, our review of convergent validity lacks the necessary comparisons with instruments that capture similar constructs, let alone a diverse array of methods. Longitudinal studies using MEEM could provide a similar trait assessment to compare with common trait measures. This would not address the emergence of particular states in dynamic and contextualised situations nor provide a strong validity or reliability assessment for the state based purpose.

A useful direction for future research would be to examine the descriptive schemes associated with the theoretical model in terms of their constituent components, potentially using a factorial allocation of these components. These descriptive schemes encompass central aspects such as listening modes, attention, and musical meaning (see [34]), which may provide more concrete conceptual handles for specifying and understanding the underlying drivers of emotional experiences associated with music. A related avenue for comparison, that could offer a potentially informative perspective for convergent validity, is using the BRECVEMAC framework to assess multiple features and mechanisms at the same time [22,89]. Although BRECVEMAC is a theoretical framework rather than a psychometrically validated instrument, it provides a useful conceptual account of how surface features of music may give rise to emotional responses. Such work, however, would require a congealing of different epistemic perspectives to categorise the emotions observed through a methodological framework that captures the dynamic, procedural, emotional episodes people have within musical situations.

Finally, we acknowledge the limitations of our samples. All participants were based in the UK, were English speakers, and therefore represent a Western, educated, industrialised, rich, and democratic (WEIRD [90]) population. Although the sample departs from typical student samples in terms of age (M = 45 years) and exhibits a relatively balanced gender distribution, these characteristics do not mitigate the broader limitations regarding cultural representativeness. Substantial further work is required to examine the extent to which the five proposed scales and their measurement generalise across cultural contexts. We anticipate variation both within Western societies and, more markedly, when the instrument is applied in non-Western settings. While the theoretical framework was developed with the intention of integrating functions of music that have been discussed across diverse cultural literatures, we emphasise that the current operationalisation, as well as its construct validity, should be regarded as provisional outside the present cultural context and will require re-evaluation, adaptation, and potentially reconceptualisation in other cultural regions.