Fostering support for non-democratic rule?

In democratic contexts, performance evaluations are often conceptualized to affect support for the incumbent government rather than for the political regime itself (see, e. g., Citrin 1974). Due to the often far reaching fusion of incumbents and regime in political systems that lack regular alternation in office, however, this distinction is rarely made for non-democratic contexts.

Another reason for implementing formally democratic institutions can be found in authoritarian rulers’ efforts to solicit the cooperation of regime elites or opposition forces (Boix and Svolik 2013; Gandhi and Przeworski 2007). Apart from these internal rationales, there are also external reasons for controlled political liberalization: Since the Cold War, democracy has emerged as the only legitimate form of political rule across the globe, and economic assistance from Western countries is now often conditioned on holding free and fair elections. Providing purportedly democratic multiparty elections may thus also be a way to gain international legitimacy and secure external economic assistance while still retaining autocratic power (Levitsky and Way 2010a, pp. 16–20).

Of course, political liberalization may not always be temporally prior to changes in regime support: Non-democratic rulers may also choose to embark on a course of controlled political liberalization as a reaction to dwindling popular support, i. e. changes in the degree of political liberalization may follow changes in the level of regime support. However, the expectation would then still be that regime support increases as a result of this liberalization and not the other way around – it seems highly unlikely that non-democratic rulers would choose to liberalize as a reaction to increasing popular support, i. e. that liberalization occurs after regime support has increased.

While it would be more faithful to the process nature of controlled political liberalization to analyze the relationship between changes in the degree of political liberalization and changes in the level of regime support, suitable longitudinal data are not available. The hypothesized causal relationship therefore can only be examined using cross-national variations in the levels of political liberalization and regime support, and all hypotheses are formulated accordingly.

Conceptually, regime support and evaluations of democratic performance are two distinct constructs. While democratic performance evaluations refer to a specific attitude based on cognitive assessments of the regime’s democraticness (and nothing else), regime support refers to a more general and diffuse attitude that may be based not only on judgments of the regime’s democratic performance but also on judgments of its economic performance, the perceived legitimacy of the regime, the popularity of the ruler, and so forth.

It is also conceivable that regime support affects democratic performance evaluations: if citizens are satisfied with their political regime – for whatever reasons – they might also evaluate this regime as being more democratic. While this endogeneity problem cannot be resolved empirically with the data available, the theoretical considerations outlined above (rational-choice approach) as well as prior research strongly suggest that evaluations of democratic (as well as economic or administrative) performance are indeed causally prior to regime support and not the other way around.

Huhe and Tang (2016) have recently (and independently of this contribution) also suggested a moderating effect of this type. They, however, do not offer a coherent theoretical argument for this effect. Empirically, they find that pro-democratic value orientations do exert a stronger negative influence on political support in non-democracies than in democracies.

All political systems not listed as “electoral democracies” by Freedom House for the respective survey years were classified as non-democracies. Four of these non-democracies were excluded due to the apparent instability of their political regimes (Egypt, Thailand) or the lack of state monopoly (Libya, Palestine (West Bank)) during the survey fieldwork period.

To avoid structures in a single country (China) dominating the results, all countries are weighted equally for all analyses. Net sample size remains unaffected by this weighting.

Robustness checks using instead either the Afrobarometer or the Arab Barometer data did not yield substantially different results.

Robustness checks using changes in inverted Freedom House scores over the past 5 and past 10 years, respectively, as a measure of liberalization did not yield substantially different results.

All three items load highly on a single factor both in the pooled sample and in individual countries (standardized factor loadings generally >0.5). Measurement invariance across all countries, however, could not be established. This is a common issue in cross-national comparative research (see, e. g., Ariely and Davidov 2011; Coromina and Davidov 2013; Davidov et al. 2014). While a lack of measurement invariance makes an interpretation of the results methodologically problematic, this issue cannot be resolved with the data currently available.

For individual countries, alpha ranges from 0.563 (Honduras) to 0.842 (Morocco). Apart from Honduras, only Uganda (alpha = 0.598) and Kenya (alpha = 0.602) yield similarly poor reliability values. While alphas below 0.65 or even 0.7 are often considered problematic, they are not uncommon with very short scales such as the one used here (e. g., Cortina 1993; Hair et al. 2010, p. 125). Since dimensionality analyses (CFA) confirm the unidimensionality of the scale, even the poor values for Honduras, Uganda, and Kenya are deemed acceptable.

Of course, social desirability and political fear are always problematic when conducting survey research in non-democratic political systems. However, only small minorities (an average of 4.8%) of respondents in African non-democracies appeared suspicious during the interviews (the highest proportions being 16% in Algeria, 11% in Sudan, and 10% in Cameroon); no more than 4% of respondents in East Asian non-democracies seemed insincere in answering the interviewer’s questions; and a maximum of 10% of interviewers reported difficulties in asking questions about politics in Latin American non-democracies (in Venezuela; less than 5% in Honduras and Nicaragua). While surely not definitive (and of course disregarding those refusing to be interviewed altogether), these numbers do provide some indication about the willingness of citizens in non-democracies to respond to survey questions and can be interpreted as a sign that their answers can be deemed reasonably valid. In addition, prior studies report only weak correlations between measures of political fear and political support (Chen and Shi 2001; Shi 2001; Yang and Tang 2010), corroborating the proposition that political fear has no major effect on response behavior even in non-democracies.

Unfortunately, no comparable measures for socioeconomic status or household income were available for all four surveys.

A “2-1-1” mediation depicts an independent variable on level 2 (in this case: the degree of system-level political liberalization), a mediating variable on level 1 (in this case: individual-level democratic performance evaluations), and a dependent variable on level 1 (in this case: individual-level regime support) (see, e. g., Zhang et al. 2009).

In order not to underestimate the standard errors for the level-1 slopes, the slopes of the main individual-level variables of interest – democratic performance evaluations and democratic value orientations – were estimated including a (significant) random component. Despite some of the slopes of the level-1 control variables having significant variance on level 2 as well, these slopes were not estimated with a random component in order to keep the model as parsimonious as possible. This is justified for two reasons: one, the effects of the control variables are only of minor interest here; and two, the overall fit of the model is barely improved by including these random components (cf. Hox 2010, pp. 54–59; Snijders and Bosker 2012, pp. 155–161).

As a consequence, Model 1d in Table 3 contains two effects of democratic performance evaluations on regime support: the first one is the “regular” level-1 effect and the second one is the level-2 mediation effect.