Keywords

4.1 The World Values Survey and European Values Survey Data

Launched in 1981, the World Values Survey (WVS) is a series of nationally representative surveys conducted in nearly 100 countries, covering almost 90 per cent of the world's population, using a common questionnaire on the attitudes of the world's population towards religion, politics, economics, society, education, prejudice, gender and sexuality and the family. The WVS is the largest non-commercial, cross-national, time-series survey of human beliefs and values ever conducted, and currently includes interviews with nearly 400,000 respondents (Inglehart, 2020). The Website of the World Values Survey currently states:

The World Values Survey (www.worldvaluessurvey.org) is a global network of social scientists studying changing values and their impact on social and political life, led by an international team of scholars, with the WVS association and secretariat headquartered in Stockholm, Sweden. The survey, which started in 1981, seeks to use the most rigorous, high-quality research designs in each country. The WVS consists of nationally representative surveys conducted in almost 100 countries which contain almost 90 percent of the world’s population, using a common questionnaire. The WVS is the largest non-commercial, cross-national, time series investigation of human beliefs and values ever executed, currently including interviews with almost 400,000 respondents. Moreover the WVS is the only academic study covering the full range of global variations, from very poor to very rich countries, in all of the world’s major cultural zones. (https://www.worldvaluessurvey.org/WVSContents.jsp)

The current study uses the well-established methodology of analysing data from international surveys, again in the World Values Survey, as already presented in detail in the study by Tausch et al., (2014). We would like to emphasise that, in addition to comparing percentages and means in cross-tabulations, the present study makes particular use of the method of partial correlations and promax factor analysis. As can be seen in Tausch et al. (2014), promax factor analysis is particularly suitable for extracting dimensions of variables that may be correlated with each other from a dataset with many variables. Table 4.1 of our paper shows the date of the WVS samples as well as the sample size N.

Table 4.1 Our surveys from the World Values Survey and the European values study
Full size table

The latest version of the WVS had the following name: WVS Cross-National Wave 7 spss v4 0.zip; and the identical version of the joint EVS/WVS is accessible through two data service points: EVS/GESIS: via the GESIS Data Collection at GESIS—Leibniz Institute for the Social Sciences (data download page); and WVSA: via the WVS website.

4.2 Methodology

Our research attempt is of course guided by the vast traditions of mathematical-statistical analysis in opinion survey research (see Tausch et al., 2014, see furthermore Abdi, 2003; Basilevsky, 2009; Brenner, 2016; Browne, 2001; Fabrigar et al., 1999; Hedges et al., 2014; Kline, 2014; Knippenberg, 2015; McDonald, 2014; Mulaik, 2009; Suhr, 2012; Yeşilada et al., 2010).

Our main statistical calculations relied on simple cross tables, comparisons of means, bi-variate and partial correlation analyses, factor analysis (oblique factor rotations based on promax factor analysis) (Abdi, 2003; Babones, 2014; Basilevsky, 2009; Blalock, 1972; Browne, 2001; Cattell, 2012; Ciftci, 2010, 2012, 2013; Clauß & Ebner, 1970; Fabrigar, et al., 1999; Finch, 2006; Gorsuch, 1983; Harman, 1976; Hedges et al., 2014; Kline, 2014; Rummel, 1970; Suhr, 2012; Tabachnick et al., 2001; for a condensed survey, see also Tausch et al., 2014). For the algorithm of partial correlation analysis and promax factor analysis, we refer our readers to IBM-SPSS (2014); Hendrickson et al., (1964) and Morrison (1976).

This being said, a few more specifications are necessary for the readers interested in getting to know more details of the methodologies used in this work. So, our methodological approach is within a more general framework to study global values with the methodology of comparative and opinion-survey based political science (Brenner, 2016; Knippenberg, 2015; Inglehart, 2018a, 2018b, 2020). Our methodology of evaluating the opinions of global publics from global surveys is in addition based on recent advances in mathematical statistical factor analysis (Basilevsky, 2009; Hedges et al., 2014; Kline, 2014; McDonald, 2014; Mulaik, 2009). Such studies allow to project the underlying structures of the relationships between the variables.

Current methodology of the social sciences makes it clear that besides factor analysis, there are also other powerful tools of multivariate analysis available to test complex relationships between an independent variable and independent variables (Tabachnik, & Fidell, 2001; Abdi, 2003; Babones, 2014; Basilevsky, 2009; Browne, 2001; Clauß & Ebner, 1970; Fabrigar, et al., 1999; Hedges et al., 2014; Kline, 2014; Suhr, 2012; Tabachnick et al., 2001; for a condensed survey, see also Tausch et al., 2014). In our case, we also used partial correlation analysis. Omitted variable bias indeed is a serious problem in the discipline (see Tausch et al., 2014).

4.3 Promax Factor Analysis

In the vast literature, surveyed in Tausch et al., (2014), there are two ways to add together the results from the different components, making up either an UNDP-type of performance Index indicator: simply adding the results together, or first grouping them together to various subcomponents, and only from there to arrive at the final results. Our multivariate analysis greatly relies on factor analysis (see Tausch et al., 2014; and IBM Documentation SPSS Statistics, 2023; and Universität Zürich Methodenberatung, 2023). Factor analysis combines groups of interval-scaled variables into meaningful factors that are as independent of each other as possible. It can also be used to discover structures in the data (see IBM Documentation SPSS Statistics, 2023; and Universität Zürich Methodenberatung, 2023).

Concerning factor analysis and the so-called oblique rotation of the factors, which are underlying the correlation matrix, we also refer our readers to important literature on the subject (Abdi, 2003; Browne, 2001). The IBM-SPSS routine chosen in this context was the so-called promax rotation of factors (Browne, 2001; Fabrigar et al. 1999; Suhr, 2012; Yeşilada et al., 2010), which in many ways must be considered to be the best suited rotation of factors in the context of our research today.Footnote 1 Formulated in plain everyday language, the mathematical procedures of the rotation of factors which best represent the dimensions underlying a correlation matrix are necessary to make the structure simpler and more reliable.

The problem which factor-analysis is solving can be described as follows: can the variables under consideration here be represented in mathematically reduced dimensions, and what percentages of the total reality are thus reproduced, and how are these dimensions related to each other? And what is the relationship of the underlying variables with these dimensions? Is there indeed such a “factor” or “dimension” as religiosity, and how does it affect phenomena like “trust in the police” or “Antisemitism”? Is there, apart from it, also something like “Accepting Gender Equality”, and also something like “class” or “status”, which influences “trust in the police” or “Antisemitism”, independent from the other “factors”? Promax factor analysis is a well-established multivariate and mathematical variety among the general techniques of factor analysis, which extracts the underlying dimensions from the matrix of correlations between the variables and precisely answers the questions just raised above.Footnote 2 It was amply described in recent literature (Finch, 2006; Tausch et al., 2014, see, furthermore Gorsuch, 1983; Harman, 1976; Rummel, 1970). As already stated, Promax factor analysis is the most appropriate technique of factor analysis in public opinion survey studies today (Finch, 2006; Ciftci, 2010, 2012, 2013; Ciftci & Bernick, 2015). Factor analysis—in our case promax factor analysis—also allows the researcher to use the mathematical model for the development of a new measurement scale for the new dimensions, derived in the research process (Tausch et al., 2014). In modern social indicators research, such new scales are called “parametric indices”.

Factor analysis is therefore primarily used for data structuring and data reduction. On the one hand, grouping variables into factors facilitates interpretation, and on the other hand, a single factor or a few factors can be used instead of a large number of variables in further analyses. Factor analysis includes a number of different procedures, some of which have different aims (see IBM Documentation SPSS Statistics, 2023; and Universität Zürich Methodenberatung, 2023).

4.4 Testing Levels of Significance

The Universität Zürich Methodenberatung (2023) underlines, among others the following advices for the practice of factor analysis:

  1. 1.

    The Kaiser–Meyer–Olkin Measure of Sampling Adequacy is a statistic that indicates the proportion of variance in the variables that might be caused by underlying factors. High values (close to 1.0) generally indicate that a factor analysis may be useful with the data. If the value is less than 0.50, the results of the factor analysis probably won't be very useful. The rule of thumb is that the KMO value should be at least 0.60 in order to proceed with factor analysis. The literature generally suggests 0.50 as a lower acceptable limit.

  2. 2.

    Bartlett's test of sphericity tests the hypothesis that the correlation matrix is an identity matrix, which would indicate that the variables are unrelated and therefore unsuitable for structure detection. Small values (less than 0.05) of the significance level indicate that a factor analysis may be useful with the data.

  3. 3.

    In addition, Bartlett's test can be used to test the null hypothesis that the variables are completely uncorrelated. However, this test assumes that the data are normally distributed. The Eigenvalue of a factor indicates how much of the total variance of all variables is explained by that factor. SPSS normalises the total variance to be explained to the number of variables.

  4. 4.

    The so-called “Kaiser criterion” (also known as the “eigenvalue rule”) states that only factors with an eigenvalue greater than 1.0 should be extracted. SPSS selects the number of factors strictly according to this criterion, unless the user specifies a fixed number of factors. The screeplot shows the number of factors on the x-axis and their Eigenvalues on the y-axis. If the factors are random, the slope is flat.

  5. 5.

    The factor loading of a variable is the correlation between the variable and the factor. Theoretically, values between −1 and + 1 are possible. The amount of factor loading indicates how closely a variable is related to a factor: Values close to 0 indicate that there is little relationship. The higher the value, the stronger the correlation.

  6. 6.

    In order to assess the factor loadings and the assignment of the variables to the factors, the rotated component matrix is considered. Factor loadings below ±0.20 should not be considered. If an item does not load higher on any factor, it is recommended to remove the item and run the analysis again. Factor loadings of ±0.30 to ±0.40 are minimally acceptable, but higher values are desirable (especially with small samples and a small number of variables).

4.5 Parametric Indicators

Our indicators are so-called “parametric indicators,” which—in every day plain language—combine the data with the methods of multivariate statistical analysis (see Tausch et al., 2014). Such a parametric indicator relies on advanced statistical methods, such as principal components analysis (see, again, Tausch et al., 2014). Such an analysis extracts an overriding indicator, mathematically best representing the component variables and their correlation matrix. Our parametric indices thus rely on the original survey respondents of the survey, and calculate the country results, based on principal components factor scores.

Our statistical calculations were performed by the routine and standard IBM-SPSS statistical program (IBM-SPSS XXVIII).Footnote 3 Footnote 4 Since both our data and the statistical methods used are available around the globe, any researcher can repeat our research exercise with the available open data and should be able to reproduce the same results as we did.

4.6 Error Margins

For the calculation of error margins of the representative opinion survey, readers are referred to the easily readable introduction to opinion survey error margins, prepared by Cornell University Roper Center (2017). Readers more interested in the details are also being referred to Langer Research Associates (n.d.)Footnote 5 On the basis of the methodological literature on opinion surveys this website makes available a direct opinion survey error margin calculator. It is important to recall that, for example at a ficticious 5% distrust rate in the Government, error margins for our chosen samples of around 1.000 representative interview partners for each country are + −1.4%. A 10% distrust rate, the error margin is  + −1.9%: and at a distrust rate of 15% the error margin is + −2.2%; see Langer Research Associates (n.d.) That error margins differ according to reported rates of, say, distrust in the police, is an important fact of opinion survey research theory, often forgotten to be mentioned in the public debate. Keeping in line with standard traditions of empirical opinion survey research (Tausch et al., 2014), for all analysed groups and sub-groups, a minimum sample size of at least 30 respondents per country had to be available to be able to attempt reasonable predictions (Clauß & Ebner, 1970) (Table 4.2).

Table 4.2 Maximum ranges of variation for survey results (the probability of error is 5%)
Full size table

4.7 Dimensions and Variables from the World Values Survey and European Values Survey

In a brave new world of fully available social science data, it would certainly have been easier to find an appropriate design for our multivariate analyses. Admittedly, the impartial observer and analyst will very quickly notice that the complete data set of the World Values Survey varies considerably from wave to wave, and that for the analysis of global respondents’ attitudes towards religion, there is far more data available in the World Values Survey wave 2010–2014 than in the latest wave from 2017 onwards.

Another important limitation for the design of the present analysis is that for some EU countries that are of particular interest for the present study, such as the Republic of Austria, unfortunately no data are available for the period 2010–2014. The following list shows the variables used in the multivariate factor analysis. For the World Values Survey wave 2017–2022, we used two indicators of homonegativity, namely rejection of homosexual neighbours and rejection of parenthood by homosexual couples.

In general, our chosen variables well reflect the variables which were used in other studies, surveyed in our Chap. 3. Of particular relevance, the research by Janssen and Scheepers (2018) provided very valuable insights for the present study, especially its treatment of religious particularism and religious salience.

World Values Survey, 2010–2014:

  • Democracy: Civil rights protect people’s liberty against oppression

  • Democracy: People choose their leaders in free elections

  • Democracy: Religious authorities interpret the laws

  • Democracy: Women have the same rights as men

  • Disagree: all religions should be taught in public schools

  • Disagree: people who belong to different religions are probably just as moral as those who belong to mine

  • Disagree: the only acceptable religion is my religion

  • Distrust: People of another religion (B)

  • Favouring income inequality

  • For state ownership of business

  • Important child qualities: religious faith

  • Never attend religious services

  • Not important in life: Religion

  • Reject neighbours: Homosexuals

  • University is not more important for a boy than for a girl.

World Values Survey, 2017–2022:

  • Not important in life: Religion

  • Important child qualities: religious faith

  • Reject neighbours: Immigrants/foreign workers

  • Reject neighbours: Homosexuals

  • Homosexual couples are not as good parents as other couples

  • Men don't make better political leaders than women do

  • University is equally important for a boy and for a girl

  • Men don't make better business executives than women do

  • Willingness to fight for country

  • Democracy: Religious authorities interpret the laws.

  • Democracy: People choose their leaders in free elections.

  • Democracy: Civil rights protect people’s liberty against oppression.

  • Democracy: Women have the same rights as men.

  • Importance of democracy

  • Justifiable: Political violence

  • Gender—female

  • Year of birth.

4.8 Cross-National Data

In the explanation of the partial correlations of homonegativity, due emphasis was also given to dependency and world system approaches to development, which received a large-scale empirical confirmation in an earlier study (see also Tausch et al. (2013)). Earlier, well-known datasets for these investigations were obtained from Ballmer-Cao et al. (1979); Müller et al. (1988), Tausch (2012, 2019) and Tausch et al. (2013).

The present data set, which we used in our analysis, is available in EXCEL format data in Tables 5 and 7 at https://www.researchgate.net/publication/374631532_Homonegativity_28_09_2023_EXCEL_PUBLIC_ACCESS_Data_for_the_publication_Homonegativity_and_religiously_motivated_political_extremism_A_study_based_on_World_Values_Survey_data_from_88_countries_and_terr.

4.9 The Empirical Research Design

Our empirical research design is heavily influenced by the availability of open access international data on the phenomenon of homonegativity. In a first round of analysis, we wanted to know from the available OSCE hate crime data, which to our knowledge have not yet been analysed in the literature, what exact patterns of homonegative hate crimes exist in the countries of the European Union and what exact percentages of total societal hate crimes these homonegative hate crimes already account for. This huge database was analysed using our IBM SPSS, 29 statistical software.

Using the World Values Survey database and its 2017–2022 version and the 1981–2016 longitudinal edition, we first looked at the bivariate correlations of homonegativity at the individual level among the entire global population, and whether or not there are differences in these correlation patterns among global Roman Catholics, global Muslims and global Orthodox. Are there differences in the drivers of homonegativity for the total population, global Roman Catholics, global Muslims and global Orthodox? For example, does worship attendance among global Roman Catholics trigger a higher level of homonegativity than, say, worship attendance among global Orthodox?

We then decided to look at the bivariate correlations of global population homonegativity at the individual level with variables of xenophobia and racism.

We then proceeded to analyse the pattern of partial correlations of homonegativity in the countries of the world system with key socio-economic indicators at the country level. The UNDP Human Development Index (and its square to control for possible non-linear effects) was held constant as the key indicator of existential security. We also calculated the partial correlations of the homonegativity of the world population with key indicators from the World Values Survey (2010–2014), holding constant age & sex & highest level of education attained.

For the remainder of Chap. 5, which is devoted to the bivariate and multivariate empirical results, we conducted a Promax factor analysis of the drivers of homonegativity with data from the World Values Survey, Longitudinal_1981_2016 and the World Values Survey, 2017–2022. In each case, we present tests of significance, percentages of explained variance for each variable (Extraction (explained variance; 0.0 = 0%; 1.0 = 100%)), Eigenvalues and explained variances, and screetests for each factor analytic model, and then present factor structure matrix loadings, component correlations and country factor scores.

In Chap. 6 we analyse the levels and relationships between homophobia, extremist religiously motivated homophobia and extremist religiously motivated and potentially violent homophobia in the countries of the world. We then compute a parametric index of tolerant gender social norms and democracy (TGSNDI) based on the factor analysis of Chap. 5, and we present the weights for the factor analysis scores.

We also analyse the partial correlations of the Tolerant Gender Social Norms and Democracy Index (TGSNDI) in the countries of the world system with key country-level socio-economic indicators, again holding existential security (the UNDP Human Development Index) and its square constant.