How Trends in Democratic Support Come About

Proponents of the democratic deconsolidation hypothesis allude to contemporary economic and socio-cultural transformations as origins of the alleged growing disenchantment with democracy in Western societies. These propositions suggest the young generation is the vanguard of democratic decline (Denemark, Donovan and Niemi Reference Denemark, Donovan, Niemi, Denemark, Mattes and Niemi2016a; Foa and Mounk Reference Foa and Mounk2016; Foa and Mounk Reference Foa and Mounk2017a). Recent birth cohorts are seen to be the most susceptible to growing democratic disaffection because they have been hit the hardest by recent economic crises in several European countries, which may have undermined their confidence in the prevailing political system (Mounk Reference Mounk2018). Given their socialization experiences in the age of individualization and digitalization, the natural properties of democracy's decision-making processes might be at odds with the preferences of today's young generation. Whereas recent birth cohorts have grown accustomed to highly individualized consumer products and a fast-reacting media environment, democracy's institutions are slow by design and do not necessarily respond to individual preferences; they instead prescribe collectively binding decisions (Gurri Reference Gurri2018; Streeck Reference Streeck2016). As a consequence, today's young generation may become more disenchanted with democratic politics and potentially more open to alternative forms of governing.

However, the expectation of growing democratic fatigue, especially among the young, contradicts modernization theory, which posits that democracy's attitudinal foundations will grow stronger from one generation to the next (Alexander and Welzel Reference Alexander and Welzel2017; Kirsch and Welzel Reference Kirsch and Welzel2019; Norris Reference Norris2017; Norris and Inglehart Reference Norris and Inglehart2019). According to modernization scholars, in the wake of a universal value change, assertive orientations that are more prevalent among older cohorts shift toward self-expressive, emancipative orientations that are more prevalent among younger generations who grew up in secure material environments. As a result, the young generation may be more critical of political authorities and institutions but more steadfast in their support for the principles of self-governance.

Note that, in these discussions, scholars occasionally refer to the ‘young generation’ without clearly defining generational boundaries or work with differing sets of cohort operationalizations (for example, Foa and Mounk Reference Foa and Mounk2017a, 6; Foa and Mounk Reference Foa and Mounk2017b; 10; Norris and Inglehart Reference Norris and Inglehart2019, 36). This vagueness partly reflects the fact that various macro-level mechanisms are at play. Influences such as the 2008 economic crisis hit a well-defined cohort of individuals during their formative phase, but other influences such as the ‘fifth wave’ of information technology (Gurri Reference Gurri2018) characterize the formative years of a larger and less clearly defined cohort. While democracy-undermining influences may be more forceful the later a person was born, there is no clear criterion to delineate younger from older birth cohorts. Nonetheless, we can conclude that skeptics and advocates of the democratic deconsolidation hypothesis disagree about the differences in democratic support between older and younger birth cohorts and, by implication and more fundamentally, about the prospects that generational replacement holds for mass support for the democratic system of governing.

Less controversial than the presence of generational disparities is the claim that democracy is facing challenges that may shake confidence in the democratic polity, regardless of generational affiliation. Hence, another conceivable pathway of eroding support for democracy is that the recent economic and political crises and other processes, such as the diminishing steering capacities of nationalized democracies, may have undermined democratic support in the form of period effects – that is, uniform decreases from one point in time to the next irrespective of an individual's position in the life cycle and generational affiliation (Mounk Reference Mounk2018; Streeck Reference Streeck2016). What remains disputed, however, is whether these period effects merely affect short-term political orientations such as party preferences or whether they have begun to erode foundational attitudes toward democracy such as support for the principles of self-governing. This dispute is partly because the basic empirical facts on crucial aspects of Western societies are contested. This uncertainty reflects various shortcomings in the literature.

Strengthening the Reliability and Validity of Findings in a Contested Field of Research

The topic's longitudinal nature calls for survey questions that were asked repeatedly over a relatively long period of time. Yet, time-series data on democracy-related questions are scarce. Hence, it remains unclear whether the findings presented by prior research indicate short-term fluctuations or long-term trends (van der Meer Reference van der Meer2017). Responding to calls from all sides of the debate for more comprehensive survey data, this study uses recently published cross-national data that allow us to properly identify how democratic support has developed over time and across generations (Foa and Mounk Reference Foa and Mounk2016, 10; Foa and Mounk Reference Foa and Mounk2017a, 10; Voeten Reference Voeten2017, 1).

Although most time series do not cover long time periods, the respective surveys often contain multiple instruments to assess individual-level political support (Mattes Reference Mattes2018). While a large number of indicators reflects the complex nature of democracy, it also hinders the comparability of studies, thereby contributing to the ambiguity of existing research. Moreover, considering the vast array of analytical options, each with possibly different outcomes, it is no wonder that authors frequently suspect that some findings in the scholarly controversy were reported selectively in a way that fits the authors' claim (Alexander and Welzel Reference Alexander and Welzel2017; Norris Reference Norris2017, 5; Voeten Reference Voeten2017, 1).

In this study, we respond to the challenges for knowledge accumulation that result from researchers' ample degrees of freedom in a contested field of social inquiry in the following way. Before obtaining the data, we documented our analytical strategy and all indicators we considered relevant for assessing democratic support in a public pre-analysis plan.Footnote ¹ What is more, although limitations of space prohibit reporting the entire body of evidence in this study, we transparently report all our findings in an easy-to-use interactive companion website. These procedures bind our hands to intentional or unintentional misuses of analytical discretions and provide the reader with unbiased and unrestricted access to the collected evidence (Wuttke Reference Wuttke2019).

To strengthen the validity of the substantive findings and to increase the informational value of testing the mechanisms that may underlie particular patterns of attitude shifts, this study employs statistical techniques to separate distinct attitudinal trends. Attitudinal trends may stem from period effects that reflect shifts in the marginal distribution of attitudes, for example in response to external shocks that do not vary by age or generational affiliation. Alternatively, such trends may reflect generational disparities that affect the distribution of attitudes within a society over time through generational replacement. Time may also play a role as life-cycle effects that are hard to disentangle from generational effects from a cross-sectional perspective. Yet, life-cycle effects usually have no ramifications for attitudinal trends at the societal level and are thus of limited relevance to the state of democracy (Norris Reference Norris2017, 9f; Voeten Reference Voeten2017, 5).Footnote ²

Hence, each temporal effect has distinct implications for the future distribution of attitudes. Because these temporal effects relate to distinct theoretical arguments, disentangling cohort, period and life-cycle effects advances testing the theoretical mechanisms proposed in the debate on democratic deconsolidation. Finally, aggregate-level stability when observed in descriptive studies may overshadow countervailing attitudinal dynamics of period and cohort effects that could be revealed when separating these effects. While prior studies have often used less demanding methods (for example, Denemark et al. Reference Denemark, Mattes and Niemi2016b; Foa and Mounk Reference Foa and Mounk2017b; Norris Reference Norris2017; Norris and Inglehart Reference Norris and Inglehart2019; Voeten Reference Voeten2017) that run the risk of conflating temporal effects, there are good reasons to employ statistical techniques that discern the distinct temporal pathways that may underlie attitudinal trends toward democratic deconsolidation.

Data and Research Design

We analyze data from the European Values Survey, including the recently released fifth wave (EVS 2019). We consider all European countries in the analyses that are classified as consolidated democracies according to the Polity IV index plus France. Because attitude trends may differ across countries, we report the results for each country separately. Evidence on seven additional European democracies is reported in an interactive Shiny Web Application that accompanies this study (http://democracy.alexander-wuttke.de).Footnote ³

Bearing in mind the wealth of relevant indicators and the space limitations, we only report the development of selected indicators in the main text. Indicator selection was guided by data availability (indicators must have been surveyed in at least three EVS survey waves) and substantive arguments. Given their character as the ‘backbone of the measurement of support for democracy’ (Mattes Reference Mattes2018), it does not come as a surprise that citizens' regime preferences are central to current scholarly debates (Foa and Mounk Reference Foa and Mounk2016; Voeten Reference Voeten2017). EVS employs the following questions to capture them: ‘I'm going to describe various types of political systems and ask what you think about each as a way of governing this country. For each one, would you say it is a very good, fairly good, fairly bad or very bad way of governing this country? (1) Having a democratic political system (2) Having a strong leader who does not have to bother with parliament and elections (3) Having experts, not government, make decisions according to what they think is best for the country (4) Having the army rule the country.’ In the main text, we use this question to report on respondents' preferences for democratic and authoritarian systems of government, respectively. As the second concept of key interest, we analyze the development of institutional trust, which has also played a role in the current debate (Voeten Reference Voeten2017). For compact reporting, we created an unweighted summary index of confidence in institutions that are affected by or involved in the process of democratic decision making: parliaments, justice systems and civil service bureaucracies (see the interactive Appendix for results on single indicators).

The main text focuses on these primary indicators, but we complement the reported evidence with condensed conclusions on substantively related secondary variables that tap into more specific attitudes towards democracy and other cultural and political orientations that are relevant to democratic stability. The full results on the latter are available in the interactive Appendix. The Appendix also reports tabulated results and findings on (mostly statistically insignificant or small) life-cycle effects. The main text focuses on period and generational disparities because they are critical to the over-time development of aggregate popular support for democracy.

As a main analytical strategy for discerning age, period and cohort effects, we employ generalized additive models (GAM, Grasso Reference Grasso2014). We include the survey years as fixed effects for the period effects and recode age to a three-level categorical variable (15–29, 30–59, 60+).Footnote ⁴ Often referred to as the ‘problem of generations’ (Mannheim Reference Mannheim1970), operationalizing birth cohorts is rarely straightforward. Generational boundaries are seldom self-evident, and their specification is even more complex in cross-national analyses since formative phases and events may differ across countries. Because ill-specified thresholds run the risk of hiding meaningful patterns, we refrain from deriving a comprehensive categorization of cohorts. Instead, we estimate smoothed nonlinear cohort effects that retrieve any cohort commonalities among groups of individuals born in temporal proximity. This exploratory approach is particularly useful for the purpose of this study, as we focus on the attitudes of the young as one select birth cohort. Doing so allows us to examine whether recent birth cohorts differ from older birth cohorts without setting rigid a priori boundary specifications, and thus minimizes the risk of overlooking meaningful disparities related to a person's time of birth.

To assess the magnitude of attitudinal dynamics, we visualize predicted cohort and period effects with simultaneous intervals (Simpson Reference Simpson2018) based on the observed values approach (Hanmer and Kalkan Reference Hanmer and Kalkan2013). Because age, period and cohort (APC) analyses are sensitive to modeling choices, we replicated the analyses using the robust hierarchical APC method (rHAPC), which are reported in the Appendix (Bell and Jones Reference Bell, Jones, Burton-Jeangros, Cullati, Sacker and Blane2015). The rHAPC is more conservative concerning generational effects than the GAM and tends to underestimate generational effects (Bell and Jones Reference Bell and Jones2018), particularly when they are nonlinear such as when one cohort stands apart from preceding generations. We consider GAM to be the primary model and consider the results as less robust when they are not supported by both models. Hence, when interpreting the results, we focus on cases in which the evidence is consistent across models, and we mention substantial inconsistencies.