Cross-Cultural Measurement Invariance in the Personality Inventory for DSM-5. M.A. Sorrel et al. Psychiatry Research, July 22 2021, 114134. https://doi.org/10.1016/j.psychres.2021.114134
Highlights
• It is again evidenced that the loading structure of the PID-5 is very complex
• Partial scalar invariance was supported which allowed for factor means comparisons
• The domain where the differences were greatest was psychoticism
• The influence of non-invariant items found to be was minimal
Abstract: The validity of cross-cultural comparisons of test scores requires that scores have the same meaning across cultures, which is usually tested by checking the invariance of the measurement model across groups. In the last decade, a large number of studies were conducted to verify the equivalence across cultures of the dimensional Alternative Model of Personality Disorders (DSM-5 Section III). These studies have provided information on configural invariance (i.e., the facets that compose the domains are the same) and metric invariance (i.e., facet-domain relationships are equal across groups), but not on the stricter scalar invariance (i.e., the baseline levels of the facets are the same), which is a prerequisite for meaningfully comparing group means. The present study aims to address this gap. The Personality Inventory for DSM-5 (PID-5) was administered to five samples differing on country and language (Belgium, Catalonia, France, Spain, and Switzerland), with a total of 4,380 participants. Configural and metric invariance were supported, denoting that the model structure was stable across samples. Partial scalar invariance was supported, being minimal the influence of non-invariant facets. This allowed cross-cultural mean comparisons. Results are discussed in light of the sample composition and a possible impact of culture on development of psychopathology.
Keywords: personality disordersDSM-5PID-5factor analysismeasurement invariance
3. Discussion
The present study provides evidence for the measurement invariance of the PID-5 across the cultures examined. This pattern concurs with Thimm et al. (2016)’s results which indicated that the scale was also invariant across United States and Norwegian samples. Since Thimm et al. analyzed university students only, they claimed that a more heterogeneous sample with a larger variety of age, educational level, and socioeconomic status should be analyzed in the future to test the cross-cultural stability of the structure of PID-5. Although university students were also analyzed in the present study, two samples came from the general population. Considering that findings from the present paper are more generalizable, the limitation of the composition of the sample is somewhat surpassed. Note also that no sharp differences in the structure were found between university (French-speaking samples) and community samples (Spanish and Catalonian samples). As in Thimm et al.’s study, some intercepts had to be released. These results imply that if different groups are to be compared, the more appropriate approach would be comparing the latent factor means. However, as the practical influence of the non-invariant items was shown to be minimal, observed scores might be also interpreted with caution. This adds to previous evidence showing that PID-5 domain scores can also be compared by sex and clinical status (Bach et al., 2017; Suzuki et al., 2019).
3.1. What could Explain the Differences in Intercepts and Means?
Differences in the intercepts and the latent means can be caused by two reasons. The first one would be variations on the sample composition. Note that samples from the community settings (Spanish and Catalonian ones) generally presented the lowest mean differences, and that some of the two largest differences were observed between the French and Belgium (university setting) and both Spanish and Catalonian samples for the Disinhibition domain. Besides, there were large differences in the age distribution. French-speaking samples were younger than the Spanish and Catalonian ones. It is a well-established fact that age has an impact on Disinhibition and related constructs as Sensation seeking (Steinberg et al., 2008), with younger subjects showing higher levels of Disinhibition (Zuckerman, 1994). That would be against other authors like Debast et al. (2018) who state that PID-5 is mostly age neutral. Samples also differed in terms of sex distribution with French-Speaking samples being composed mostly by females, and Spanish and Catalonian samples showing a more balanced distribution. Since women score higher on Neuroticism (Costa et al., 2001) and lower on Disinhibition and Sensation Seeking (Zuckerman, 1994), differences between all French-Speaking samples and Spain and Catalonian can be expected on these two domains. This pattern was observed for the comparisons with Belgium and France, but not with Switzerland. Besides, the previous study by Suzuki et al. (2019) explicitly examined the issue of gender invariance measurement for PID-5, finding that the scale was equivalent across sex. In summary, there is some room to support the hypothesis that the differences in the means may be due to variations in sample composition.
The second explanation of the differences across samples implies the presence of a real impact of the culture in the development of psychopathology. From this standpoint, culture may play a role in determining the exact behavioral and contextual formulation of some maladaptive behaviors and psychopathological manifestations (Terracciano & McCrae, 2006). The fact that differences on factor means were not homogeneous across samples with a similar composition (French-Speaking university samples), and that the samples from the same country (Spain) were generally similar in their latent means, suggests that culture may play a relevant role to explain differences on personality disorders. The five samples came from four different countries (i.e., Belgium, France, Spain, and Switzerland). The most widespread framework for comparing cultures is the six-dimensional classification generated by Hofstede (e.g., Hofstede, 2011). This model postulates six dimensions: power distance, uncertainty avoidance, masculinity vs. femininity, long term vs. short term orientation, and individualism vs. collectivism. Arguably, this last dimension has received the most attention in previous personality literature (e.g., Mulder, 2012; Triandis, 2001). According to Hofstede Insights dataset (Hofstede, 2018), Belgium, France, and Switzerland score similarly in the Individualism-collectivism dimension (75, 71, and 68, respectively), while the score for Spain is markedly lower (51). It has been suggested that a possible consequence of the individualistic culture on personality is to encourage the development of distinct/unique attitudes, self-definition, and striving to attain personal goals (Mulder, 2012). This is somewhat congruent with the fact that France obtained the highest mean in Disinhibition and Spain the lowest. The existence of cohort effects is often cited as an argument for the influence of social and cultural factors on personality. For example, the antisocial behavior profile maintains an upward trend and has doubled in value since World War II (Kessler et al., 1994). However, it should be recalled here that studies linking culture and personality are still scarce, so that the hypotheses put forward should be taken with caution.
3.2. Concluding Remarks on the PID-5 Factor Loading Structure
The loading solution was very similar to that reported by the two available meta-analyses (Somma et al., 2019; Watters and Bagby, 2018). It is again evidenced that the factorial structure of PID-5 is complex, with multiple cross loadings. Specifically, 30 out of 100 possible secondary loadings had a factor loading greater than .20 in absolute value. Only 8 out of the 25 facets (32%) did not present any secondary loading greater than .20 in absolute value (i.e., Emotional lability, Separation insecurity, Submissiveness, Withdrawal, Intimacy avoidance, Manipulativeness, Irresponsibility, and Unusual beliefs and experiences). The most complex facets were Rigid perfectionism and Risk taking. This complex structure might have serious implications for assessment utility (discriminant validity) and theory (Crego et al., 2015; Watters et al., 2018). While some of these secondary loadings can be supported from a theoretical standpoint (e.g., Risk-taking has been related to Negative Affect, Disinhibition and Antisocial behaviour patterns; Aluja et al., 2007), in general, it draws attention to the fact that a revision of the instrument would probably be necessary to reach the most discriminant structure with theoretical meaning.
3.3. Limitations and Future Directions
The present work is not without limitations. The reported fit values are adequate considering previous literature on the PID-5, but again serve to illustrate that it is difficult to obtain excellent fit values when exploring factor structure in the areas of personality and personality disorders. Authors such as Hopwood and Donnellan (2010) have argued that it is to be expected that it would not be easy to achieve excellent model-data fit, given how easy it is to find items with similar phrasing or other methodology artifacts (e.g., acquiescence). It is expected that modelling these factors will lead to a better fit (Abad et al., 2018). Yet, it is important to remember that this is one of the sources of validity evidence available to support the interpretation and use of the scale scores. The fact that adequate evidence of criterion-referenced validity for the PID-5 scores is reported in prior research also contributes to that purpose (Al-Dajani et al., 2016). As for the analysed samples, all of them came from European Western cultures. In the personality field, Allik et al. (2017) and Aluja et al. (2020) found that similar cultures seem to have similar mean personality profiles. The present study supports this idea also would apply in the context of pathological personality, as most of the effect sizes of differences between countries did not reach a medium effect size, and latent means for Spanish and Catalonian samples were generally similar. It is possible that incorporating non-European Western cultures could change this pattern of cross-cultural stability of the structure, although some other studies suggest a stability beyond western cultures (Rossier et al., 2008). Another limitation is that the current study used only a nonclinical sample. However, the available research seems to indicate that this is not a major concern since prior studies found that the results at the domain level obtained in non-clinical samples might be generalized to clinical populations (Bach et al., 2018).