It has long been argued that democratic regimes and public support for them are mutually reinforcing: that high levels of public support ensure democracies remain strong and that experience with democratic governance generates robust public support (see, e.g., Easton Reference Easton1965; Lipset Reference Lipset1959). But the evidence for either part of this claim has been decidedly mixed. Countries with greater democratic support have been found to become stronger and more stable democracies (e.g., Inglehart and Welzel Reference Inglehart and Welzel2005, 251–4) and just the opposite (Fails and Pierce Reference Fails and Pierce2010, 182–3). Similarly, studies have alternately found that more experience with democracy yields more democratic support (e.g., Fails and Pierce Reference Fails and Pierce2010, 183) or instead that long-established democracies are suffering from democratic fatigue (e.g., Foa and Mounk Reference Foa and Mounk2017).

One important reason for these mixed results is the difficulty of measuring democratic support over time and across many countries. Public support for democracy cannot be directly observed, and its incorrect measurement will limit inferences about the relationships between public opinion and institutional development. Furthermore, the survey data available across countries and over time on support for democracy—or indeed most topics in public opinion—are sparse and incomparable, greatly hindering broadly comparative research. Recent pioneering studies have sought to overcome the hurdle of sparse and incomparable data by developing latent variable measurement models of public opinion (see Caughey, O’Grady, and Warshaw Reference Caughey, O’Grady and Warshaw2019; Claassen Reference Claassen2019; Solt Reference Solt2020). A pair of prominent recent works took advantage of this latent variable approach to measure democratic support for over one hundred countries for up to nearly three decades and assess, respectively, its consequences for and roots in democratic change (Claassen Reference Claassen2020a; Reference Claassen2020b). The first of these works concluded, supporting the classic argument, that mass support had a positive influence on democratic change, especially the endurance of democracy (Claassen Reference Claassen2020a, 127–30). The second directly contradicted the classic argument, concluding that democratic change has a thermostatic effect on public support—that is, that rather than generating its own support, deepening democracy provokes a backlash and it is instead democratic backsliding that calls forth greater public support (Claassen Reference Claassen2020b, 46–50).

The models employed in these studies’ analyses, though, do not account for uncertainty in their measurement of democratic support. Because they are unobserved, latent variables are inherently accompanied by measurement uncertainty. To leave this uncertainty unacknowledged is to make the implausible assumption that the latent variables are measured perfectly, an assumption that distorts both statistical and substantive inference (see, e.g., Crabtree and Fariss Reference Crabtree and Fariss2015).

Here, we reexamine the classic arguments about support for democracy and democratic change that were tested in these two pieces while correcting this oversight. In addition to incorporating the measurement uncertainty, we sought to reduce it by expanding considerably the survey data drawn on and reestimating democratic support for 144 countries for up to 33 years between 1988 and 2020. Our analyses reveal that the significant relationships between public support and democratic change disappear once measurement uncertainty is taken into account, both in replications with the studies’ original data and in our extension analyses that incorporate additional data. That is, simply taking into account measurement uncertainty—making no change to the specification of the models—reveals there is no empirical support for either claim put forward in these two works: declining democratic support does not signal subsequent democratic backsliding, and changes in democracy do not spur a thermostatic response in democratic support.

There are several important implications of these null results. They point to a need for closer attention to the conditional aspects of the classic theory (see Easton Reference Easton1965, 119–20; Lipset Reference Lipset1959, 86–9). On the one hand, the effect of democracy on public support may depend not on its mere existence but on its effectiveness (see Magalhães Reference Magalhães2014), particularly with respect to redistribution (see Krieckhaus et al. Reference Krieckhaus, Son, Bellinger and Wells2014). On the other, the influence of public support on democracy may depend on the extent to which those who support democracy are also dissatisfied with the current regime’s performance (see Qi and Shin Reference Qi and Shin2011). Similarly, the results presented here are further evidence that the survey items that are commonly employed to measure democratic support are inadequate for the task. Because these questions contain no information on respondents’ support for democracy relative to other values with which it may come into conflict (see, e.g., Graham and Svolik Reference Graham and Svolik2020; Simonovits, McCoy, and Littvay Reference Simonovits, McCoy and Littvay2022) or on whether respondents even understand the meaning of the democracy they are claiming to support (see, e.g., Kirsch and Welzel Reference Kirsch and Welzel2019; Wuttke, Gavras, and Schoen Reference Wuttke, Gavras and Schoen2020), these questions appear to miss capturing the true extent of the support among the public that democracy will actually find when public support is in fact most needed. Our results also reinforce arguments that relationships between democracy and public support unfold only over the long term (see, e.g., Welzel, Inglehart, and Kruse Reference Welzel, Inglehart and Kruse2017) and that democratic change in the short term is instead best understood as an elite-driven phenomenon (see, e.g., Haggard and Kaufman Reference Haggard and Kaufman2021).

We draw two conclusions, one methodological and one substantive. Many constructs in social science—from democracy to corruption to public opinion—are latent variables, and recent advances have made estimating them much more practicable. As latent variable measurement models become more commonly used, it is absolutely necessary for researchers who use them to incorporate the associated uncertainty into their analyses. As demonstrated here, this can be done straightforwardly using the method of composition (see Tanner Reference Tanner1993, 52; Treier and Jackman Reference Treier and Jackman2008, 215) without requiring any further change in model specification.

And, at a time when democracy is seen as under threat around the world (e.g., Diamond Reference Diamond2015), taken together, Claassen (Reference Claassen2020a; Reference Claassen2020b) sends what is ultimately a reassuring message: the fate of democracy rests with us, the public, and when democratic institutions are undermined, we will swing to their support and constitute “an obstacle to democratic backsliding” (Claassen Reference Claassen2020b, 51). Both of these assertions may well be true, but the evidence we have, properly assessed, does not support them. There is no room for complacency.

Results

Figure 1 presents the reanalyses of the hypothesis that public support influences the level of democracy (Claassen Reference Claassen2020a, Table 1). The lighter, left-hand set of results replicates the analysis of Claassen (Reference Claassen2020a), including its exclusion of measurement uncertainty by using only the point estimates of public democratic support and the other variables measured with quantified error (i.e., democracy and corruption), and they reproduce that article’s findings. The middle results introduce a single change: the uncertainty in the measurement of public support and these other variables is taken into account. In all four models, the positive coefficients for democratic support are no longer statistically significant. The darker, right-hand results also incorporate uncertainty but additionally replace the estimates of democratic support with those based on our expanded dataset; this change works to increase the number of observations analyzed as well. Although the confidence intervals shrink considerably, the coefficient estimates move much closer to zero: the hypothesis remains unsupported.

Figure 1.

The Effect of Public Support on Democracy with Uncertainty

Note: Replications of Claassen (Reference Claassen2020a), Table 1, 128.

In Figure 2, we examine the thermostatic model of democratic support per Claassen (Reference Claassen2020b, 47, Table 1, 49, Table 2). The negative coefficient estimates for change in liberal democracy in the left-hand set of results, which do not take uncertainty into account, imply that the immediate effect of an increase in the level of democracy is a decline in public support for democracy and of a decrease in democracy an expansion of support—that democratic support does indeed respond thermostatically to democracy. However, the middle results demonstrate that this thermostatic effect, too, does not hold after the measurement uncertainty is accounted for. And again, the right-hand results reveal that the additional data of our extension do not provide support for the original conclusion.

Figure 2.

The Effect of Democracy on Change in Public Support with Uncertainty

Note: Replications of Claassen (Reference Claassen2020b), Table 1, 47, and Table 2, 49. Models denoted ECM are error-correction models; those marked FD are first-difference models.

In short, the conclusions of Claassen (Reference Claassen2020a; Reference Claassen2020b) that democratic support has a positive effect on democracy and change in democracy a negative effect on change in support are not empirically supported once measurement uncertainty is taken into account, even when more data are used.