While feminism is understood as an emancipation movement, aiming at the abolition of the marginalization and oppression of women, and enforcing the democratization of gender relations and the critique of male domination, antifeminism may be defined as an organized opposition to feminist emancipation efforts. Henninger et al. (2021), following Birsl (2020) and Lang and Fritzsche (2018), highlight the ideological position of antifeminism, “which is concerned with the opposition to […] processes of sociopolitical liberalization and denormalization of gender relations, as well as with the maintenance of heteronormative relations of domination” (Lang and Fritzsche, 2018, p. 340, translated by the authors).

For Birsl (2020), the field of action of feminist emancipation efforts is not limited to the hierarchical order of bipolar gender stereotypes, but rather it permeates all social conditions. She identifies antifeminism as a “thin ideology” with an antisemitic bias: it is based on a juxtaposition of progressive efforts to address women’s and LGBTI* concerns, and on the preservation of the heteronormative family as the “primordial cell” of society in neoliberalism.

Antifeminism, like modern antisemitism, emerged in the 19th century as an antimodernist resentment specifically directed against the egalitarian promises and contradictory dynamics of liberal and pluralistic modernity (Volkov, 2001; Hessel and Misiewicz, 2020). Feminists are personified as guilty of fundamental, contradictory changes in gender relations and families in the course of modernity. The term antifeminism was defined in a essay by Dohm (1902), who also interpreted it in the context of the rise of modern antisemitism. A key difference has been pointed out though (Hessel and Misiewicz, 2020): Unlike antisemitism, the rejection of women’s emancipation was almost universally shared and hardly recognized, thus fulfilling an important mobilizing and bridging function between the bourgeois and conservative camps. According to historian Planert (1998), antifeminism, characterized by its opposition to feminist movements, can be distinguished both from sexism, a term focusing on gender-based discrimination, and from misogyny, a pervasive devaluation of women deeply entrenched in societies.

When looking at these analyses and definitions of antifeminism, clear connections to the concept of (right-wing) authoritarianism arise. Modern theories of authoritarianism typically distinguish three dimensions: authoritarian aggression, authoritarian submission, and conventionalism (Altemeyer, 1981, 1996). The authoritarian syndrome may be defined as a person’s affinity for rigid ideologies that “allow them to submit to an authority, to share in its power, and to call for others to be devalued in the name of the system” (Decker et al., 2022, p.28). Antifeminist, misogynist, and sexist acts can thus all serve to fulfill an authoritarian need and establish a clear moral distinction between “good” and “bad” behavior (Manne, 2019, p. 146). Moreover, by humiliating women and trying to assign them a certain role, hierarchical power relations can be stabilized. We make use of this connection between antifeminism and authoritarian and right-wing extremist attitudes in the validation process, expecting them to be related, and yet distinct phenomena.

The major focus of German and Austrian antifeminism research is the study of organized antifeminist actors. However, antifeminism can manifest in many different ways. Some forms remain almost unnoticed as they manifest in inconspicuous phenomena, such as the non-perception of the everyday oppression and marginalization of women, the devaluation and segregation of the “female sphere” (e.g., in the low social status of “women’s stuff” and reproductive work), the defense of the traditional gender order against feminist interests, and the rejection of the “feminization” of society. One main aim of this study is to provide a measurement instrument to monitor antifeminist attitudes in the general population, thus extending current research by enabling the investigation of antifeminism in all social strata.

While a lot of scales exist to measure gender role sexism in general population samples, only little research has been conducted on antifeminism specifically. We conceptualize these two constructs to be very close, but distinct, as antifeminism goes beyond attitudes towards parenthood or the division of labor, addressing a general opposition to the emancipation of women in many different spheres. However, it makes sense to take a look at existing scales on gender role sexism and the results they produced to place antifeminism in a nomological network and make predictions about the behavior of our scale when it comes to sociodemographic variables, e.g., sex, place of residence in East and West Germany, educations, age, income and religious affiliation.

The most commonly used measure to assess gender role attitudes is the Gender-role attitudes (ISSP 94; Braun, 1999). Consisting of eleven items on three dimensions, it was first used in the International Social Survey Programme (ISSP) in 1994 and administered in 24 countries. 4-6-item versions of the scale had been used before in the American General Social Survey (GSS), the German General Social Survey (ALLBUS), the British Social Attitudes Survey as well as the Polish General Social Survey (Braun, 1999). The scale focuses on women’s participation in the labor market and traditional vs. egalitarian family structures and it does not specifically address the issue of antifeminism (Walter, 2018, pp. 47). Some interesting observations regarding sociodemographic groups were made though: as expected, women generally hold more egalitarian attitudes than men. More egalitarian views on the division of labor were also prevalent in the former eastern compared to the western German states (Baier, 2014). This was ascribed to the working policies of the former socialist German Democratic Republic (GDR), that facilitated and encouraged women to enter the work force. Moreover, time is a factor on many levels as well: Period effects in studies carried out in western industrialized countries show a liberalization in more recent years and younger birth cohorts seem to be more egalitarian than older ones (Baier, 2014; Mays, 2012; Brewster and Padavic, 2000; Bolzendahl and Myers, 2004; Braun et al., 1994; Endrikat, 2003; Hofäcker and Lück, 2004), though there is evidence for a return to more traditional attitudes in younger cohorts, at least in East Germany (Heller et al., 2024).

A negative correlation between education and traditional gender role concepts or sexist attitudes has been demonstrated several times (Thornton et al., 1983; Davis and Robinson, 1991; Rhodebeck, 1996; Bryant, 2003; Bolzendahl and Myers, 2004). However, according to Bolzendahl and Myers (2004), maternal employment, contrary to expectations, did not lead to a reduction in gender role sexist attitudes among women. Perceptions of economic deprivation also showed to be irrelevant for women’s gender role orientation (ibid.). This corresponds with other studies showing mixed and inconclusive results regarding the effect of the individual economic situation on political attitudes (Rippl and Baier, 2005).

Thornton et al. (1983) and Glick et al. (2002) empirically demonstrated that highly religious people tend to hold more conventional gender role views. Mays (2012) could only find a weak effect for the relationship between religiosity and sexist attitudes, although this could also be a consequence of their operationalization of religiosity, which was surveyed exclusively via churchgoing frequency.

Stober (1995) constructed a 20-item, balanced scale to measure antifeminist attitudes specifically. It was validated using a sample consisting of 102 US citizen. The scores showed high reliability, factorial validity as well as high correlations with related constructs (RWA, sexism, and openness to experience). Contrary to expectations, no differences between men and women were observed.

Simon and Kohl (2023) recently developed a scale on anti-egalitarian beliefs consisting of two subscales with five items each, antifeminism and naturalization of gender differences. Confirmatory factor analysis yielded a good model fit and structural equation models show moderate positive relationships to related phenomena (i.e., belief in a just world, right-wing authoritarianism, social dominance orientation).

Based on these results as well as the theoretical considerations outlined above, we expect antifeminism to show moderate to high correlations to gender role sexism, authoritarianism as well as right-wing extremist attitudes. Furthermore, we expect antifeminism to be higher among male participants, participants with a lower level of educational attainment, and older participants. Unlike gender role sexism, we expect values to be higher in participants living in East Germany, as we conceptualize antifeminism to be closer to authoritarianism than gender role sexism; authoritarianism, like right-wing extremism, is generally more prevalent in the East German population today. We do not expect differences with regards to income. As results regarding religious affiliation have been inconclusive, we are only investigating this on an exploratory basis. Furthermore, we are going to test a central precondition of latent mean comparisons before investigating differences based on sociodemographic variables, namely that of measurement invariance. Measurement invariance describes the ability of a scale to capture a certain construct equally well in all groups under investigation. Results and interpretation as well as comparison of mean scores might be heavily biased, if measurement invariance is not established (Meredith and Millsap, 1992; Van de Vijver, 2018).

In this study, we set out to validate the Leipzig Antifeminism Short Scale (LAF-S), an instrument to examine antifeminism and its spread where it is not visible in organized form, but still manifest in attitudes: in different groups in the middle of society. Our analysis is an extension of the work by Höcker et al. (2020): using more advanced statistical methods, we aim to introduce the first valid and reliable short scale for antifeminism to an international readership. The scale itself is designed to monitor antifeminist tendencies efficiently in large-scale, general population samples. By testing measurement invariance and then comparing latent mean scores across different sociodemographic groups and social strata (gender, age, education, income, religious affiliation, and region of residence), we start to shed light on the distribution of antifeminist beliefs and thus lay a foundation to explore the dynamics that cause these kinds of antidemocratic attitudes.

The sample was taken from a national, biennial representative survey conducted by the University of Leipzig, Germany, the ‘Leipzig Studies on Authoritarianism’ [former Leipzig ‘Mitte’ (Centre) Studies]. The study is designed to monitor political positions, opinions and attitudes in the German general population since 2002. Data collection for this particular study was realized by an independent institute for opinion (USUMA) from May to June of 2020. A multi-stage randomization procedure following the ADM-sampling method was performed to ensure representativeness of the sample. In 2020, 5,418 households were contacted using 258 sampling points throughout Germany. Of these households, 5,389 were eligible to participate, i.e., non-vacant and with at least one inhabitant meeting the inclusion criteria (at least 14 years of age and a sufficient ability to understand spoken and written German). In case of multi-person households, Kish selection grid was used to randomly select the target person. All participants provided informed consent; in case of minors, at least one next of kin, caretaker, or guardian provided additional consent. Sociodemographic information was obtained in a face-to-face setting. The participants were then handed a questionnaire to fill out on their own with the interviewer still present to answer any queries. Response rate was at 46.8%, leading to a total sample size consisted of N = 2,503. As missing values did not exceed 5% on any single item (Schafer and Graham, 2002), participants with missing values in the questionnaire were excluded for the analyses (n = 43). Table 1 gives an overview of the sociodemographic characteristics of the sample. When comparing sex, education, and age groups, to data provided by the Federal Statistical Office of Germany (2020) only minor deviations could be observed, including a slight overrepresentation of female and older participants as well as an underrepresentation of participants with a low level of educational attainment. Keeping these deviations in mind, we consider the sample to be representative of the German population. Thus, no weights were applied in the analysis.²

The items for the Leipzig Antifeminism Short Scale (LAF-S) were taken from an 11-item antifeminism scale with three dimensions: antifeminism, gender role sexism, and pro-feminism. In this study, we focus on the antifeminism subdimension of the scale, as it captures a distinctly political dimension with clear connections to other antimodern resentments. It consists of four items and were developed by Charlotte Höcker, Hannah Eitel, and Oliver Decker as part of an 11-item scale with three dimensions.³ All items were scored on a 4-point scale ranging from 1 = “I completely disagree” to 4 = “I completely agree.” Table 2 comprises the original item wording as well as an English translation.

To establish convergent validity, a well-validated, nine-item short scale for authoritarian attitudes, the “Kurzskala Autoritarismus” [KSA-3; Beierlein et al. (2014); validated by Heller et al. (2022)] was used. It measures authoritarian attitudes based on the three dimensions proposed by Altemeyer (1981, 1996) using three items each: authoritarian submission as a person’s tendency to follow the rule of a strong leader, authoritarian aggression as the willingness to punish (socially) deviant behavior and conventionalism as the adherence to established rules of conduct and reluctance to change. Answers were scored using a 5-point Likert-type scale.

Subsequently, right-wing extremism was measured using the “Fragebogen zur rechtsextremen Einstellung – Leipziger Form” [FR-LF; Decker et al., 2013; validated by Heller et al. (2020)]. It is based on a definition established at a consensus conference of leading German social scientist (Kreis, 2007) and captures the following dimensions using three items each: chauvinism, antisemitism, antifeminism, xenophobia, belittling the crimes of national-socialism, and agreement to a right-wing dictatorship. A 5-point Likert-type scale is used for scoring. All item wordings in English and German may be found in the Supplementary material.

In order to establish greater generalizability of the LAF-S and to meaningfully compare latent mean scores, we seek to establish invariance across several different demographic groups, that may be found in Table 1. Sex was assessed in three categories: male, female, and divers. As only one person fell into the divers category, only male and female participants were included in the group comparison. Six age groups of roughly the same size were differentiated. Regarding the current place of residence, we differentiated between the states that were part of the former GDR (East Germany) and those that were not (West Germany). The districts of Berlin that used to be part of West Germany were coded accordingly. To make results more accessible, educational attainment was converted four different categories: low education (equivalent of less than 10 years of schooling), medium (equivalent of approx. 10 years), and high (equivalent of more than 10 years of schooling), as well as participants still in school. The latter were excluded for group comparison. Household equivalent income was assessed in four groups: ≤1,000 EUR, >1,000–2,000 EUR, >2,000–3,000 EUR, and >3,000 EUR. Five categories were used to differentiate religious affiliations: Protestant, Catholic, Muslim, and no religious affiliation as well as a residual category ‘other’ containing Jewish, Buddhist, Evangelical, Christian-Orthodox and other religious affiliations.

As the scale under investigation was scored with just four response categories, correct scaling is an issue. Simulation studies by Rhemtulla et al. (2012) could show that Likert-type scales with less than five response categories should be treated as ordinal instead of metric to insure correct estimation of models in (multi-group) confirmatory factor analyses [(MG)CFA]. To assess the scale’s psychometric properties, we thus first evaluated standard item statistics for ordinal data, namely median and mode. Additionally, we looked at response distributions across the four response options for each item. To make results more accessible, we also reported common indicators such as mean, standard deviation, skewness and kurtosis as well as item-total-correlation, and item difficulty. Typically, skewness and kurtosis are used as indicators for normality with a cut-off of >|2| for a nonnormal distribution (Pituch and Stevens, 2016). Item difficulty should range between 0.20 and 0.80 and item-total-correlation should be above 0.30 (Moosbrugger and Kelava, 2012). While item difficulty is not necessarily a meaningful indicator when it comes to measuring attitudes and beliefs, we include it here for comparative purposes.

We then performed confirmatory factor analysis to verify the scale’s unidimensionality. To account for the ordinal character of the data, we used the weighted least square mean and variance adjusted estimator (WLSMV) that is specifically designed for ordinal observed variables, as it does not make any assumptions about their distribution (Li, 2016).⁴ Instead of item intercepts, threshold estimates are used based on polychoric correlations. As there are no general recommendations for model fit indices and cut-off values in models with ordinal indicators yet, the following standard criteria were used in their scaled version to evaluate model fit (Schermelleh-Engel et al., 2003). χ² and χ² divided by the degrees of freedom should be as low as possible. For the Comparative-Fit-Index (CFI) as well as the Tucker-Lewis-Index (TLI) values >0.95 were rated as acceptable and values >0.97 as good. For the root mean square error of approximation (RMSEA) and the standardized root mean square residual (SRMR), we used a cut if of <0.05 for a good fit and <0.08 for an acceptable fit. Finally, factor loadings for each item as well as three indicators for internal consistency were calculated: Regular Cronbach’s Alpha that is based on Pearson correlations, ordinal alpha, that uses polychoric correlations instead (Zumbo et al., 2007) and McDonald’s omega (McDonald, 1999). Convergent validity is assessed by calculation correlations to right-wing extremism and its sub-dimensions as well as authoritarianism.

To establish measurement invariance, traditional approaches use MGCFA on increasingly restrictive, nested models. Following recommendations by Svetina et al. (2020) on invariance testing for ordinal data, three types of models were compared. (1) a configural model with no restrictions imposed, (2) a threshold model in which item thresholds were set equal across groups, (3) a final model in which both thresholds and loadings were equal between groups⁵. For each step, the decline in model fit indices is reported to assess whether invariance holds. Once again, no common cut-off values exist for ordinal data, as they depend on multiple factors like the number of groups, the number of factors or the number of observations per group. We thus focused on changes in the CFI, as the χ²-difference-test is known to be highly sensitive in large samples (ΔCFI should be <0.01; Cheung and Rensvold, 2002). Changes in χ²-values as well as RMSEA are reported as well.

If invariance may be assumed in the final model, latent mean scores can be meaningfully compared. To this end, the latent mean is set to zero in one of the groups in the final model, and latent means of the other groups are calculated relative to that reference group (Wu and Estabrook, 2016). Latent mean deviations from the reference group are then tested for significance using Wald t-test. p-values and standard errors are reported for each comparison. As it is still common practice in most psychological and social science research to compare observed sum or mean scores instead of latent means, we will additionally be reporting and comparing observed mean scores using ANOVA and post hoc TukeyHSD tests, to see if this may impact the results.⁶

RStudio 2025.05.1 Build 513 with R version 4.4.1 (2024-06-14 ucrt) was used for all analysis. Lavaan version 0.6–7 was used for (MG)CFA and latent mean comparison.

Table 3 shows descriptive item statistics of the LAF-S’s four items. Mean scores were relatively low with standard deviations indicating a possible bottom effect. Yet, both skewness and kurtosis fell under the cut-off of <|2| (Pituch and Stevens, 2016), indicating a normal distribution. Item difficulty fell in the medium level, ranging from 0.42 to 0.47 and item total correlation can be judged as excellent, exceeding the cut-off of 0.40 by far (Moosbrugger and Kelava, 2012).

Table 4 depicts the model fit indices of the CFA. It is visible that all indices stay within their respective cut-offs, with only the RMSEA exceeding the 0.08 mark by far. Kenny et al. (2015) could show that RMSEA tends to be high in models with only small degrees of freedom as this one. Overall, model fit can thus be judged as acceptable.

Table 5 shows factor loadings and internal consistency based on this estimation method. Factor loading ranged from 0.75 for Item 3 to 0.83 for Item 2. Internal consistency can be judged as good: Regular Cronbach’s alpha was at 0.81, ordinal alpha turned out a little higher at 0.86 and McDonald’s Omega was at 0.82.

In a first validation study by Höcker et al. (2020), convergent validity was reported by calculating Pearson correlations to adjacent constructs, namely right-wing extremism and authoritarianism. Their results are depicted in Table 6. Correlations reached from 0.27 (with authoritarian aggression and authoritarian submission) to 0.51 (with the sum score for right-wing extremism). These medium to high connections can be seen as an indicator of convergent validity. Regarding the related construct measuring gender role sexism, a high correlation of 0.53 was found. Correlations between gender role sexism and right-wing extremism and authoritarianism, respectively, were markedly lower than those of antifeminism with the two.

Results of analysis of measurement invariance may be found in Table 7. Following the cut-off of ΔCFI <0.01, invariance may be assumed in all conditions. Meaningful, i.e., unbiased, comparisons of latent mean scores across these groups are thus possible.

Results of the comparisons of latent mean deviations relative to a set reference group as well as comparisons of observed mean scores using ANOVA and post hoc TukeyHSD tests can be found in Table 8. Male participants showed significantly higher latent mean scores in antifeminism than female participants. Participants living in the former eastern German states scored higher than those in the former West. A gradient was observed regarding the educational background. Those participants with low educational attainment showed the highest endorsement of antifeminism—significantly higher than the reference group a medium level of educational attainment. Those participants with higher education on other hand exhibited significantly lower values than the reference group. No significant differences were found with regard to age groups. Regarding income, no differences between the group with an income of up to 1,000 EUR and the group with more than 1,000–2,000 EUR or the group with more than 2,000–3,000 EUR were found. The highest income group with more than 3,000 EUR exhibited significantly lower values compared to that with less than 1,000 EUR. Finally, results regarding religious affiliation suggest highest values of antifeminism in Muslim participants. Differences between Catholic and Protestant participants were significant on a 5%-level when latent mean scores were compared, with Catholic participants showing slightly higher values. While results of comparison of observed mean scores generally pointed in the same direction with only slight differences in significance levels, these differences between Catholic and Protestant participants did not turn out to be significant in the comparison of observed mean scores.