Based on the above-mentioned consensus definition, RWE consists of a pattern of attitudes characterized by ideas of inequality on the one hand and a dismissive attitude towards the democratic state on the other (Decker & Brähler, 2006). Three dimensions relate to the historical-political ideology of RWE (in Germany) and includes the support of a right-wing authoritarian dictatorship, national chauvinism and the trivialization of National Socialism. The other three dimensions reflect an ethnic-racist view of humanity and encompass xenophobia, anti-Semitism and Social Darwinism, which attribute different values to peoples and groups within the population and between people. A qualitative study accompanying the Thuringia Monitor, in which respondents were asked about their associations with individual statements to capture RWE-A, essentially confirms the theoretical framework of the subdimensions (Best & Salomo, 2014). Derived from theory, definition and facets of RWE-A there are two different suggestions for its internal structure of the six equal first-order dimensions. The authors of the Mitte-survey continue to follow the original delineation of a political and a social higher-order dimension of RWE-A, each one conceptualized by three first-order dimensions as described above (Zick & Mokros, 2023). The authors of the Thuringia Monitor as well as of the Leipzig Authoritarianism study suggest a two dimension second-order structure, namely with an ethnocentrism factor, which binds the national chauvinist and xenophobic attitudes, and a neo-Nazi ideology factor, which binds the Social Darwinism, trivialization of National Socialism, anti-Semitism, and support of right-wing authoritarian dictatorship (Best & Vogel, 2022; Decker et al., 2024). Nevertheless, all of these studies conceptualize one common higher-order factor RWE-A consisting of all six dimensions summed up by 18 items.

The 18-item scale operationalizes RWE-A by means of statements (see Figure 1) to which respondents express their disagreement or agreement on a 5-point Likert scale (“fully disagree”, “mostly disagree”, “partly agree/partly disagree”, “mostly agree” and “fully agree”). The original German wording as well as the English translation we are following here can be found in Heller et al. (2022). The survey method of the former Mitte-survey (until 2012) and the following Leipzig Authoritarianism study are paper-and-pencil interviews, of the Mitte-survey since 2014 computer-assisted telephone interviews. The differences between the two survey methods have repeatedly appeared to have an impact on response behavior and the results (Faulbaum & Mokros, 2023; Heller et al., 2024). Heller et al. (2020) confirmed the reliability and validity of the RWE-A construct for the Leipzig survey in 2018 and reflected the need to examine it over time. The fundamental question is whether respondents understand the items in the same way and whether the structure of the constructs analyzed is identical across surveys. Only then results of the respective study series will be comparable (Leitgöb et al., 2023). The structural uniformity of a measurement instrument over time can be assessed empirically using measurement invariance analyses.

Before testing the measurement invariance of the second-order factor model, it is necessary to test the invariance of the first-order factors which is a prerequisite for the assessment of invariance of the second-order factor model (Rudnev et al., 2018).²2

The procedure in Rudnev et al. (2018) is a bit different to ours. They tested combinations of metric and scalar invariances of first-order and second-order models while our procedure is stepwise: At first, tests of the configural, metric and scalar invariances of the first-order factor model and at second, tests of the same invariance restrictions for the second-order factor model. In our view, this stepwise strategy of testing multiple group confirmatory factor models is more informative according to the requirements for scalar invariance of the second-order factor model.

For all model estimates, the full-information-maximum-likelihood method (FIML) was used to utilize all available information in the data (Reinecke, 2014). No residual correlations between the items are specified. Table 4 in the Appendix shows the comparison of the three model variants with the group-specific and total χ²-values of the simultaneous analyses as well as the goodness-of-fit measures RMSEA and CFI. The specifications of the configural, metric and scalar model are nested and allow χ²-difference tests. The χ²-values of the group samples can be reliably compared with each other, as the group sizes hardly differ with exception for the year 2021 (2014: N = 2,004; 2016: N = 2,003; 2019: N = 2,002; 2021: N = 1,748; 2023: N = 2,026). The total size of the data set across the five Mitte-surveys is N = 9,783. It is noticeable that the model fit for 2023 is worse compared to previous years. If the sample size is large (> 300) the χ2-difference is known to reject models even when violations are minor (Chen, 2007). According to Chen et al. (2005) a difference in CFI between the model variants not larger than .01 implies that the model fit of the most parsimonious model can be accepted. The CFI differences in Table 4 of the Appendix allow us to accept the scalar model variant of the first-order factor model as a basis for the analysis of the second-order factor model.

Latent mean differences are significant for all six first-order factors across all surveys. With the sample from 2014 as the reference year, the latent means are decreasing from 2016 to 2021 whereas the latent means in 2023 are increasing compared to 2014 (see Appendix, Table 5). This result confirms the increasing tendency of right-wing attitudes in the last couple of years.

In the second step we examined the invariance of the second-order factor model for RWE-A. In all model variants the scalar model specification of the first-order model based on the results of the first step are applied. In the configural variant of the second-order factor model all second-order factor loadings are free across the samples as well as the intercepts of the first-order factors. Due to some large modification indices of the residual covariances of the first-order factors we additionally admitted the residual covariances between national chauvinism (F2) and xenophobia (F4) and between trivialization of National Socialism (F3) and anti-Semitism (F5) for all samples. The configural model shows an acceptable model fit (see Table 1). The decision to correlate the factor residuals can be explained by the fact that national chauvinism and xenophobia share an enduring ethnocentric attitude in which the own nation is considered superior to others, while anti-Semitism and the trivialization of National Socialism are more strongly linked by their historical references. The release of the residual covariance at factor level takes account of the overlap in content between the sub-dimensions without making excessive model adjustments at item level (overfitting). If item residuals were correlated instead, this would be accompanied by a stronger model fit, which would weaken a theoretical interpretation, especially as the items also contain methodological artifacts and specific effects that do not necessarily affect the theoretical construct. Additionally, model results contain no hints of significant model modifications on the item level.

For the metric model the second-order factor loadings are restricted to be equal across the samples and in the scalar variant latent mean differences of the second-order factor RWE-A are additionally estimated. Overall, restrictions of the metric and scalar variant have little impact on the model fits. Due to the parsimoniousness of the scalar model the RMSEA is slightly higher compared to the metric or the configural model variant. Differences between the CFI are all below .01 (see Table 1) so that it is possible to substantially discuss the mean differences of RWE-A obtained from the scalar model variant.

Table 2 provides the latent mean changes and variances of the construct RWE-A over the period from 2014 to 2023 according to the scalar model of the multiple group comparison in Table 1. Due to statistical identification the latent mean of RWE-A has to be fixed at zero for one group (here the first sample of 2014). The estimated mean values for the following years 2016 to 2023 show the changes compared to reference year 2014. The level of RWE-A decreases until 2021 (unstandardized estimates are between -.036 and -.069, standardized values are between -.084 and -.166), but increase in 2023 (unstandardized estimate is .138, standardized value is .249). The variance of RWE-A between 2016 and 2021 is between .144 and .184 and fluctuates on a low level. In contrast, the variance for 2023 shows a significantly higher value (.307). Even a multi-group comparison limited to just two periods (e.g., 2021 and 2023) shows a significant increase in RWE-A.

Next, we examine the influence of selected exogenous demographic variables on RWE-A that showed continuously significant differences on the agreement or disagreement with RWE-A over time. Therefore, Table 3 shows the influence of the exogenous variables age, education, region, and political self-positioning on the construct RWE-A. Estimates are fixed across some surveys for Age (2014–2019), Education (2014–2021) and Political Position (2014–2021). For Sex estimates are fixed across all surveys (2014–2023).

The younger the respondents, the more likely they agree to RWE-A. Age has the strongest effect in 2023 (unstandardized: -.121, standardized: -.167). Sex has no effect on RWE-A. The effect of education is largely stable over years (unstandardized: -.208, standardized: -.371) and declines slightly in 2023 (unstandardized: -.154, standardized: -.231). People with a higher level of education are tending recently more to RWE-A, which will be discussed later in more detail. People living in Western Germany have a slightly lower level of RWE-A than those living in the former socialist Eastern Germany, which also remains stable over time (unstandardized: -.071, standardized: -.069).

Finally, we validated the construct of RWE-A based on political self-positioning and voting preference. Political self-positioning, measured using the so-called right-left scale, has a robust effect on RWE-A over the years 2014 to 2021 (see Table 3): The more respondents position themselves as right-wing with their political opinion the more likely they hold RWE-A (unstandardized: .192, standardized: .363). However, a look at the approval ratings (most recently Zick et al., 2023) shows that the influence of political self-positioning is not linear for RWE-A, but rather corresponds to a ‘golf club curve’: People who position themselves on the “left” of the political spectrum tend to agree more often, those who position themselves “more to the left” are particularly unlikely to agree, followed by those who see themselves “right in the center”. Then agreement increases significantly among those who position themselves “more to the right” and even more so among those who position themselves on the “right”.

Voting preferences are measured by the so-called Sunday poll³3

In Germany, elections are generally held on a Sunday.

We analyzed a theory-based construct of RWE-A through large survey studies conducted in Germany since 2006, specifically focusing on the last five survey years of the Mitte-survey, using multiple group comparisons. For this analysis, a confirmatory factor model was developed based on 18 items encompassing six first-order factors: support of a right-wing dictatorship, national chauvinism, trivialization of National Socialism, xenophobia, anti-Semitism and Social Darwinism. Additionally, a superordinate second-order factor representing RWE-A was included.

Various sub-facets of RWE-A form a coherent and stable construct. The tested 18-item scale effectively captures this construct with reliability and validity. It is noteworthy how clearly the sub-dimensions, which may initially seem quite distinct, are interconnected. It remains to be seen how effectively the scale depicts and records RWE-A in other countries, especially to what extent the sub-dimension of trivializing National Socialism might be examined. For example, Jungkunz et al. (2024) successfully tested a scale of RWE-A with fewer sub-dimensions, and without a direct reference to a National Socialist past, across three Western nations.

Moreover, the results underline the importance of monitoring RWE-A as a coherent ideological construct over time. Future research should examine the applicability and adaptability of the measurement model in international contexts and investigate how shifts in contextual factors such as political discourse and polarization may influence the expression and normalization of ideological subdimensions and right-wing extremist attitudes as a whole.