Democratization and Political Tolerance in Seventeen Countries: A Multi-level Model of Democratic Learning

¹ In some countries the list of groups was modified to adjust the groups to the local context.

² Instead of the standard Likert format used in tolerance questions, which records both the direction and intensity of responses, the WVS uses a dichotomous format where responses are limited to “yes” (tolerant) or “no” (intolerant). This format may underestimate the actual degree of tolerance, even though the patterns found here approximate those revealed by other studies using different surveys and question formats (Shamir 1991; Duch and Gibson 1993; Sullivan et al. 1993). In addition, the amount of variance in the dependent variable is likely diminished, which means that our analyses are biased against finding statistically significant relationships. While this under-estimates the t-values of coefficients, it does have the desirable side-effect of establishing an especially rigorous hurdle for rejecting the null hypothesis.

³ Are “criminals” fugitives from justice, ex-felons, political criminals, or the “criminals” who run the country, for example?

⁴ To determine whether individuals who selected criminals were substantially different from those who selected another group, we estimated a logit model where we regressed a dummy variable indicating whether respondents chose criminals (coded 1 if the individual selected criminals as their least liked group and 0 otherwise) on the micro-level variables used to predict political tolerance (see Table 2). There is a slight propensity for people to select criminals if they: participate less, value speech less, are rightists, value conformity, are female, older, and more educated. However, altogether the nine predictors in the model account for very little variance in selecting criminals versus other groups. Using the pooled data, the pseudo-R² for the entire set of predictors is .02; estimating this model for individual countries, pseudo-R² ranges from .000 to .05. Overall, this analysis suggests that respondents who selected criminals as their least-liked group are not that different from those who selected other groups.

⁵ The WVS also asks respondents whether they would let their least-liked group “teach in our schools.” This indicator was not used because teaching in schools has not been interpreted in most democratic coutries (e.g., the U.S.) as a basic political freedom. Moreover, in several nations, some groups on the list (communists in Central Europe, for instance) have denied civil liberties to ordinary citizens for much of the 20th century. It is thus conceivable that within a specific historical context, even a democratic citizen would reach the conclusion that a fascist or communist, for example, should not be permitted to teach in schools in order to protect democratic institutions. Some constitutions, such as Germany’s, specifically mention that citizens must be loyal to the democratic state in order to occupy a politically sensitive post. We hasten to add that we do not wish to justify intolerance (for that is what the denial of such political liberties constitute, regardless of the justification for such restrictions). What we would like to assure is that we are focusing on the least controversial elements of Dahl’s polyarchy-the right to demonstrate and to hold public office. We note however, that the teaching in schools item is moderately correlated with holding demonstrations (r = .37) and public office (.52) and our results do not change if this item is included in the political tolerance scale.

⁶ Substantial percentages of publics report either “having done” or “would do” the following: signing a petition (70 percent), joining a boycott (43 percent) and attending a demonstration (54 percent) in the pooled sample.

⁷ We use the abbreviation HLM as a common shorthand reference, even though hierarchical modeling refers to nonlinear as well as linear modeling. We used the hierarchical generalized linear model in HLM 5.04 which is designed for use with discrete ordered dependent variables to estimate models in Tables 2, 3 and 4.

⁸ HLM simultaneously controls for the effects of macro and micro-level variables.

⁹ The correlation between Dem. Longevity and Human Dimension is .62, .28 between Dem. Longevity and Federalism, and .41 between Human Dimension and Federalism.

¹⁰ We used Freedom House’s (Freedom House 2000) evaluations of the 1994 civil liberties and political rights of each country because the waves for the world values surveys were conducted between 1995 to 1997. These results are available from the authors upon request.

¹¹ We dropped the conformity variable from this analysis because HLM would not converge when all micro-level variables are included in the model.