3. The Inner Logic: An Intergroup Approach to the Populist Mentality in Europe

We first also included Manicheism in our model, but this dimension was largely uncorrelated with the rest of our dimensions, and thereby decreased model fit. We therefore decided to exclude Manicheism from our analysis. This exclusion led to a more focused model, as all dimensions related to the specific conflict between two groups: the corrupt elite and the pure people. Indeed, Manicheism is a more general attitude towards any type of societal conflict with a “good” pole opposed to an “evil” one. Hence, a Manichean outlook could apply to any kind of strongly held beliefs. For example, Bertsou and Caramani (2022) found that items measuring Manichean attitudes loaded positively on both the populist and the elitist scale, while these two concepts are expected to be negatively correlated.

Configural invariance is the lowest level and requires that each construct is measured by the same items. It does not guarantee, however, that the measurement properties are the same. Meaningful comparisons require at least metric invariance, tested by constraining the factor loadings between the observed items and the latent variable to be equal across the compared groups. If metric invariance is established, one may assume that respondents in the different samples interpret the items in the same way, although it is still uncertain if the construct is measured in the same way. The third and highest level of measurement invariance is scalar invariance, tested by constraining both the factor loadings and the indicator intercepts to be equal across groups. If scalar invariance is established, one may assume that respondents use the scale in the same way in each group; thereby offering the possibility of comparisons of latent and observed means across samples. In order to test for the different levels of invariance, we progressively imposed new constraints on loadings and intercepts and relied on ∆CFI to evaluate model fit between two successive models, using the ΔCFI ≤ 0.01 criterion as suggested by Cheung and Rensvold (2002). Given the ordinal data of our observed variables, CFA was performed using the diagonally weighted least squares (DWLS) estimator (Kline, 2015).

For this last scale, we only have the two first-order factors. Since adding a second order factor would make the CFA model unidentified, we were forced to restrain our analysis and limit the CFA to two dimensions.

As a robustness test, we ran exploratory factor analyses using different solutions. An inspection of eigenvalues pointed towards a 2-factor model, where items measuring people homogeneity all loaded on a different factor than those from the other three subdimensions. This offers additional evidence in support of the fact that people homogeneity seems to be a country-specific measure with a tenuous relationship with populist mentality. People-centrism is therefore a potentially problematic measure of the populist mentality, due to its questionable internal validity as a populist subdimension and its limited cross-national comparability.

We clustered standard errors at country level as we deal with data coming from multiple surveys, where dependence within the data is commonly observed (due to shared sampling techniques, questionnaire wording, interviewers, or other aspects of survey design). These dependencies may bias estimation of standard errors and lead to invalid statistical inferences. Standardized cluster errors also account for heteroscedasticity.