Does democratization or democratic stability critically depend on certain cultural characteristics – that is, that the citizenry holds particular values and beliefs? The literature counts several ‘culturalist’ explanations of democracy, or lack thereof. German, Italian and Japanese cultural characteristics were previously argued to inhibit well-functioning democratic regimes,Footnote 1 but following decades of democratic consolidation after WWII, these arguments have declined in popularity. Likewise, some scholars have argued that cultural characteristics of different African, Middle Eastern and East Asian countries do not square easily with democracy.Footnote 2 However, such culturalist explanations are often drawn from quite limited or selective observations, and fare poorly when evaluated against more systematic evidence.Footnote 3 Still, there exists a more refined cultural explanation of democracy that relates not to some stable national or religious characteristic, but rather, more universally, to the extension of liberal, freedom-oriented or so-called self-expression values. Below, we reinvestigate this proposed relationship, but find no support for any causal effect of self-expression values on democracy. Instead we highlight how their correlation may be explained by democracy – which often emerges from coups or international interventions – gradually inducing the ‘learning’ of self-expression values within the population.
Despite having previously been contested by important contributions,Footnote 4 the self-expression values explanation of democracy remains commonly acknowledged.Footnote 5 The 2005 book by Ronald Inglehart and Christian Welzel (henceforth IW) presented a widely cited, and arguably the most developed, theoretical framework on this.Footnote 6 They argue that several indicators, for example post-materialistic aspirations and generalized trust, tap the same dimension, measuring what they call ‘self-expression values’ – the extent to which individual choice and freedom are appreciated and prioritized. Further, IW propose that when citizens adopt self-expression values, they increasingly view autocratic regimes as illegitimate, eventually forcing regime liberalization. Meanwhile, citizens in democracies experiencing strengthened self-expression values will work harder to protect their regimes. Hence, self-expression values supposedly breed democracy.
Cross-country correlations do, indeed, indicate such a positive relationship. Figure 1 (top plot) shows the correlation between levels of self-expression values (early 1990s) and democracy a decade later.
Fig. 1 Top: Levels of self-expression values (early 1990s) and democracy (late 1990s/early 2000s). Bottom: Changes in values and democracy from the early 1990s to the early 2000s Note: self-expression is measured by IW’s (p. 150) index, ranging from 0 to 1 and drawing on the WVS. Democracy is measured by (reversed, seven-year lagged) Freedom House Index scores, ranging from 1 to 7.
Although often given a causal interpretation, this correlation could, however, stem from many different processes.Footnote 7 Another look at the data indicates that values may not causally affect regime type after all; there is no positive relationship between changes in self-expression values and in democracy from the early 1990s to the early 2000s (Figure 1, bottom plot).Footnote 8 Several countries, such as South Africa, Taiwan and Slovakia, experienced considerable changes toward democracy without value changes in their populations, whereas Peru experienced political liberalization and lower self-expression values. Furthermore, many countries, including Jordan and Russia, experienced increases in self-expression values without democratization. The high correlation between self-expression values and democracy levels could instead be due to different demographic, political-historical or socio-economic factors systematically influencing both values and regime type. The correlation could also come from democratic institutions affecting values, at least – as our results below indicate – in the longer run. We add to the literature by designing empirical tests that account for such issues, and thus contribute to resolving a long-standing debate on whether self-expression values causally determine the chances of democratization or democratic survival. Previous contributions have highlighted that the positive effect of values on regime type is sensitive, for example, to the choice of democracy measure and other specification issues such as lag length on independent variables.Footnote 9 Still, previous studies have relied on very limited data material and – by current standards – fairly crude model specifications.Footnote 10
We employ multiple imputation to correct for potential sample-selection biases and use panel data techniques in order to better deal with omitted variable bias and endogeneity of self-expression values. We also run models that account for the slow-moving nature of political regimes and, particularly, self-expression values. In sum, we find no evidence that self-expression values cause democracy. Moreover, self-expression values enhance neither democratization chances nor democratic survival. Further, we do not find effects from any sub-components of self-expression values, such as generalized trust. Rather, we find that values are endogenous to countries’ experiences with democracy, corroborating the so-called institutional learning hypothesis. This has implications for debates on the viability of ‘democracy without democrats’, that is, whether democratization and subsequent consolidation are possible without democratic political cultures.Footnote 11 Our findings suggest that not only can democracy emerge and endure without democrats, but that democracy breeds democrats in the long run. Thus we propose that (1) democratization processes are spurred by factors other than self-expression values and (2) the observed cross-country correlation partly reflects democracy’s effect on popular values.
We first review the argument that self-expression values enhance democracy, before presenting our alternative theoretical account, data and design, and finally the empirical analysis.
Self-Expression Values as a Force for Democracy: The Argument
Numerous scholars have proposed that popular values – while not necessarily exogenous – affect political regime type. For instance, almost 200 years ago de Tocqueville postulated that democracy in the United States reflected the liberal orientations of the American people, which, in turn, were a function of particular historical and socio-economic conditions.Footnote 12 After WWII, sociological and psychological work elaborated on variations in individual ‘democratic predispositions’. For example, Adorno et al. introduced the ‘authoritarian personality’ – rooted in perceptions of threats nurturing low self-esteem, misanthropy and dogmatic rigidity, whereas Maslow proposed two democratic orientations – emphasis on self-actualization and humanistic inclinations – to consider others as equals.Footnote 13 Further, Almond and Verba argued that stable democracies require a civic culture, ‘a pluralistic culture based on communication and persuasion, a culture of consensus and diversity, a culture that permitted change but moderated it’.Footnote 14 As noted, values are not exogenously given, and economic development is often considered a key factor in creating more tolerant, liberal, freedom-oriented or simply ‘democratic’ citizens.Footnote 15 Lipset, for instance, argues that improved educational attainment – related to economic development – makes people more tolerant toward opposition and minorities, in turn promoting democracy.Footnote 16
More recently, IW proposed a Revised Theory of Modernization (RTM), the current standard for the argument that citizens’ values systematically determine regime type.Footnote 17 RTM contains two links. The first ties socio-economic development to value changes, proposing that individuals mired in scarcity aspire to satisfy basic economic needs, while individuals whose economic needs are satisfied strive for self-realization and autonomy. The outcome of economic modernization is thus an ecological syndrome of values encompassing different orientations such as autonomy, participation, tolerance and trust; the underlying value dimension ranges from survival to self-expression orientations. The second link, conceptualized as a supply–demand relation of freedoms, ties self-expression values to democracy.Footnote 18 Democratic institutional structures provide the supply of freedoms, while demand is related to citizens’ values.Footnote 19 A tendency toward congruence is supposedly inherent, as the institutional supply of freedoms is under pressure to satisfy the masses’ demands. Two important assumptions are that mass support is crucial to regime survival, and that mass support is driven by values rather than, for example, instrumental evaluations of how regimes affect resource distribution.Footnote 20
One can further nuance the second link by distinguishing between democratization and democratic survival. Regarding democratic survival, the argument is that liberally oriented publics discipline governing elites to obey by democratic rules, and that the elites’ values resemble those of wider society. Hence, both winners and losers of democratic elections are disciplined by a liberal public and political actors drawn from such publics are more likely to act out of genuine commitment to democratic ideals. This mitigates various threats to democratic survival, such as coups and auto-coups. IW further argue that self-expression values induce democratization by eroding regime support and increasing support for pro-democratic forces in autocracies.Footnote 21 The pro-democratic forces materialize through collective action (for example, social movements and campaigns) motivated by aspirations for freedom. Collective action motivated by such aspirations supposedly has a high probability of success in engendering democratization; participants and leaders of freedom-motivated movements are dedicated, and thus are not tempted by economic benefits promised by autocrats or containable by repression in the long run.Footnote 22 As people increasingly emphasize self-expression values, autocracies face growing ‘suppression costs’ leading to intra-elite tensions and anti-regime movements. When the regime can no longer bear these costs, democratization occurs.
Yet this argument hinges on assumptions that are problematic, for example in light of recent literature on co-optation, repression and manipulation in autocracies. The expectation regarding democratization is particularly ambiguous; IW’s model is relatively straightforward, with liberal-democratic preferences being converted into institutional change through mass collective action. However, autocrats can prevent mass values from translating into collective action through different measures, and employ strategies to mitigate their impact if collective action actually materializes. Although regime capabilities of co-opting and repressing threats may depend on access to revenues, for example from natural resource extraction, Bueno de Mesquita and Smith analyze how leaders can reduce demands for revolutionary change by increasing public goods provision.Footnote 23 More generally, autocrats can co-opt selected population groups and their leaders by providing economic benefits and political positions.Footnote 24 Alternatively, leaders can underprovide ‘co-ordination goods’, such as a free press and communication technologies, thereby limiting strategic co-ordination.Footnote 25 To illustrate, Iran – a country scoring fairly high on self-expression values – has shut down internet and phone access when facing unrest.Footnote 26 Regimes also repress their oppositions, thereby enduring despite large gaps between the supply of freedoms and people’s demands.Footnote 27 Indeed, leaders anticipating democratization movements, for example because of widespread self-expression values, may have extra incentives to spend resources on co-optation or repression.
However autocrats repressing or co-opting ‘liberal’ populations constitutes only one potential reason for why only a weak (or no) causal link from values to democracy may exist. Democracy may largely emerge because of other factors than popular values – including international political developments and domestic elite strategies, but also other societal changes highlighted by traditional modernization theory, such as urbanization and increased education. Further, democratic institutions – once in place – could induce self-expression values, thus explaining the cross-country correlation. We elaborate on this below.
An Alternative Account of the Democracy–Values Relationship
Historically, democratization processes have often been triggered by international factors unrelated to domestic value orientations.Footnote 28 For example, the post-WWII democratization experiences of Japan and (Western) Germany were direct results of military defeat in WWII and external imposition by the Allied powers. Indeed, early expositions of German and Japanese political cultures questioned whether democracy could survive given their illiberal citizenries. Almond and Verba even provided survey data showing low ‘participant culture’ in Western Germany about a decade after democratization. In 1959/1960, only 19 per cent of Germans thought ‘most people can be trusted’ (compared to 49 per cent for the UK and 55 per cent for the US).Footnote 29 In 2013 (data from 6th World Values Survey (WVS) wave), 45 per cent of Germans answered this question in the positive. Hence, the current extension of self-expression values in Germany has seemingly followed rather than preceded democracy. More recently, external pressures and conditionalities imposed by Western governments and international organizations – such as the EU in post-communist Eastern European countries and donor governments and the IMF and World Bank in early-1990s Sub-Saharan African countries – have contributed to political liberalization in several countries that, according to Figure 1, (still) score low on self-expression values.Footnote 30
The literature also documents that various domestic factors affect democratization, but these factors often involve other actors than the general publicFootnote 31 . Coups d’etát constitute the most common mode of autocratic breakdown,Footnote 32 and coups in autocracies actually improve democratization prospects.Footnote 33 Less numerous party elites or military elites organizing coups expectedly face less difficult collective action problems than organizers of mass uprisings,Footnote 34 and involve actors with the requisite means to depose the ruler. Huntington, for example, describes how the first transition of the Third Wave, in Portugal, started with a military coup.Footnote 35 Although coup makers, as in Portugal, often do not intend it, coups may initiate political dynamics that ultimately induce democratization.Footnote 36 Further, transition processes mainly involving elite-led negotiations and subsequent pacts have also been common historically; popular values may have been less important than elite-led negotiation tactics and bargains in, for example, Spain and Chile.Footnote 37
Finally, even in cases where popular pressure might have played a critical role in inducing democratization – one example according to Acemoglu and Robinson was nineteenth century Britain, where the working classes pushed for franchise extensions – popular motivations need not be the desire to realize self-expression values, but rather to secure a regime allowing for progressive redistribution.Footnote 38 Although we lack survey material for the nineteenth century British working class, its material conditions and education levels would – if we accept the first link of RTM from IW – suggest that self-expression values were not widespread.
Even if democracy is not caused by self-expression values, but emerges for the above-discussed reasons, democracy and values may still correlate strongly. According to ‘institutional learning theory’, individuals’ values, preferences and behavior are heavily influenced by the institutional environments within which they operate. Hence, democratic institutions establish particular practices, norms and values as the going standards, and these standards ultimately shape people’s values and preferences. This implies that citizens, by living under democratic institutions over time, internalize ‘democratic values’. When becoming accustomed to channeling demands through democratic institutions and experiencing democratic competition and debate, people become more tolerant, trusting and liberal, and this may explain the democracy–self-expression values correlation.Footnote 39
The above proposition builds on an assumption that self-expression values, such as genuine tolerance and generalized trust, are cognitively complex and inherently difficult to learn.Footnote 40 Superficial support of freedom and democracy can be acquired through, for instance, international diffusion, but fully internalizing self-expression values requires having practiced democratic behavior. Theories of social identity suggest that individuals are inherently inclined to distrust strangers and to be intolerant of people from other backgrounds. Yet persistent experience of engaging with different fellow citizens gives people the opportunity to apply a ‘sober, second thought’; values such as tolerance are therefore ‘learned best when citizens are exposed to the rough-and-tumble of democratic politics’.Footnote 41
Different democratic-institutional aspects may help instill self-expression values. A free media stimulates open discussion and advances civil liberties through the dissemination of knowledge; elections create awareness among citizens of their roles as political participants; and the growth of civil society organizations, and citizens’ active practice in such organizations, strengthens political involvement and induces tolerance and trust through contact with others.Footnote 42 Finally, several democracies have actively worked to promote democratic norms and values through ‘civic education’, teaching democratic citizenship in educational institutions or community workshops.Footnote 43
The literature offers several historical examples of such ‘democratic learning’. Mishler and Rose report democratic-institutional learning mechanisms in post-Communist Eastern European countries, and Lindberg proposes that repeated (but often initially flawed) elections in Sub-Saharan African countries have induced democratic values.Footnote 44 Returning to the German case, Rohrschneider shows how political elites in East and West Germany, although they share a similar pre-war history and culture, developed different values because of different post-war institutional experiences. The relatively authoritarian German political culture was replaced by growing acceptance of democratic values among West German elites after years of experience with democratic institutions.Footnote 45 This change did not occur in communist East Germany.
Hence, self-expression values may be a consequence of, rather than a critical force for, democracy. Importantly, even if we accept a modernization framework, modernization theorists from Lipset onwards have suggested multiple channels through which economic changes affect regime type – including class composition, cross-cutting cleavages and urbanization – and RTM’s focus on self-expression values as the critical channel is not necessarily valid. Moreover, as we elaborate on in the following section, the empirical evidence supporting RTM rests on shaky methodological foundations.
Existing Empirical Evidence and Threats to Inference
Some recent systematic analyses studying the effects of mass attitudes and values on democracy exist, and not all reach the same conclusion.Footnote 46 Yet Inglehart and Welzel’s studies are seminal contributions.Footnote 47 Using data from the WVS, they very clearly conclude that self-expression values result from processes of modernization and induce democracy. In contrast, Seligson and Hadenius and Teorell report that the findings are sensitive to the selection of control variables and the choice of democracy measure.Footnote 48 Inglehart and Welzel have responded vociferously to these criticisms, contending that ‘[t]here is a remarkably strong correlation between self-expression values and effective democracy. The evidence indicates that the causal linkage is mainly from self-expression values to democracy’.Footnote 49 Below, we show that this interpretation is unwarranted.
IW’s inferences are mainly based on cross-country variation – as holds true for the literature in general – but also on pooled data with countries observed at two time points, and a model investigating changes between two points in time.Footnote 50 These designs are not unreasonable, given the issue of available survey data. However, they provide an insufficient basis for drawing strong causal inferences. Strictly speaking, all the positive ordinary least squares (OLS) coefficients tell is that self-expression values are more widespread in countries with democratic regimes (controlling for selected variables), while survival values are more widespread in autocracies. This may seem like methodological purism; given the plausible argument in RTM and cross-country correlation, can we not conclude that self-expression values induce democracy? Unfortunately, we cannot. The income–democracy literature provides some important indications why.
Following Lipset, several scholars replicated the income–democracy correlation using more data, alternative controls and more complex estimation techniques.Footnote 51 The correlation was – often implicitly – taken as evidence that income enhances democratization prospects and stabilizes existing democracies. However, Przeworski and Limongi altered the conventional knowledge, reporting that income only enhances democratic stability and not democratization.Footnote 52 Even more dramatically, Acemoglu et al. argue that the income–democracy relationship is not causal at all; the correlation disappears when controlling for country-fixed effects.Footnote 53 Although later analysis suggests that this stems from fixed effects (FE) models being inefficient for slow-moving variables, it casts doubt on classic modernization theory’s most central hypothesis.Footnote 54 One interpretation is that different historical experiences – for instance concerning institutions built during colonial rule – explain why some countries embarked on paths of economic development and democracy while others did not.
Inferring from cross-country correlations to a causal relationship is at least equally problematic for values and democracy.
First, omitted variables may systematically affect both democracy and self-expression values.Footnote 55 Country-specific factors – related to particularities in political histories, religious traditions or geographic locations – may crucially impact the population’s ideals and the accountability and transparency of government. If so, the values–democracy relationship could be spurious. There may, for instance, be trajectories in the political histories of the UK or the Netherlands that explain why these countries currently have both liberally oriented populations and democratic regimes.
Secondly, the correlation may be due to democracy enhancing self-expression values, following the democratic learning argument above. This may induce endogeneity biases in estimates if not accounted for. IW challenge institutional learning theory, claiming that emancipative values are largely exogenous to democratic institutions. Admittedly, they discuss reverse causality and conduct Granger tests, lagging independent variable(s) and controlling for the lagged dependent. However, their specifications are problematic.Footnote 56 First, when using democracy (measured in 1997–2002) as the dependent variable, lagged democracy (early 1980s) is measured long before lagged values (early 1990s), since ‘1990 measures refer to a point in time at which democracy was a moving target in many countries’.Footnote 57 We consider this a problematic (and arbitrary) choice. These tests, for example, do not reflect the possibility that early 1980s democracy affects early 1990s values, and fail to control for the relevant democracy level. This choice also affects the sample, as early 1980s democracy cannot be (appropriately) measured for new states emerging in the early 1990s. Further, the Granger tests reported to disconfirm institutional learning theory include only nineteen countries, mostly developed countries where democratic institutions had long existed (thereby, possibly, already having exhausted institutional learning effects). Further, IW’s tests do not include multiple controls.Footnote 58
Thus democratic institutions affecting self-expression values may still explain the correlation taken as evidence of RTM. Below, we employ panel models with similar lags for all independent variables, expand the sample and control for multiple confounders. Moreover, we use generalized method of moments (GMM) models that are constructed to handle endogenous regressors, which IW’s OLS models are not.
Research Design, Imputation Model and Data
In contrast to previous studies employing cross-section data, we mainly use panel data to study whether self-expression values affect democracy. This has three major advantages. First, it expands the number of data points. Secondly, it allows us to control for country-fixed effects, which reduces the risk of omitted variable bias, and may also – although we are not focusing on this here – alleviate measurement equivalence problems.Footnote 59 Thirdly, it allows accounting for the endogeneity of values. Importantly, if self-expression values actually cause democracy, we should identify covariance along the temporal dimension within (and not only between) countries.
Imputation Model
Our dependent variable and most controls have decent time series. Yet one challenge is the lack of temporal variation on values. As IW, we employ WVS data, supplying it with European Values Survey observations when available. These data contain five survey waves – from 1981 to 2009 – and many countries have participated in fewer. Hence, we run a multiple imputation model employing Amelia II, which accounts for the time-series–cross-section data structure, to predict self-expression values for years not covered by WVS waves (and other missing data points).Footnote 60 This allows us to use panel data models, but it also has other substantial benefits.
The conventional response to missingness in the values–democracy literature is listwise deletion; units with ≥1 missing are dropped. Even if data is missing completely at random (MCAR), listwise deletion generates inefficiencies. Moreover, missingness is often systematic, leading also to selection biases; previous estimates of how values relate to democracy may therefore be wrong.Footnote 61 The WVS, for instance, contains an overweight of rich, democratic countries among those participating in all waves. Certain types of dictatorship could be selected out of the sample, and perhaps at particular points in time, potentially leading us to mis-estimate the general effect of values on democracy. When conducting imputation we assume data are missing at random (MAR), meaning the missingness pattern only depends on variables included in the imputation model and not on non-included predictors or the variable with missing data itself. Although MAR is a strict assumption, we seek to mitigate violations of it by including proxies of self-expression values from regional barometers and numerous other relevant predictors (for example, repression and development proxies, see Appendix Sections AI–AIII). While unobserved predictors might remain, we have thus tried to mitigate concerns that the data are not missing at random when composing our model. In any case, MAR is a weaker assumption than MCAR, which must hold for listwise deletion in order to yield unbiased results.
As noted, multiple imputation may potentially contribute to minimizing selection biases and increasing efficiency. Hence, one could argue that we should have constructed imputed datasets for all countries. But, in order to provide a minimum of direct information for predicting values, we retain only countries that participated in ≥1 WVS wave and run most models on countries with ≥2 non-imputed observations on the Self-Expression Index (SEI; see below). Our imputation-model specification allows for country-specific (second-order polynomial) time trends. Indexes, fractions, and other restricted variables have their theoretical minimum and maximum values as bounds, and unrestricted variables have their empirically observed minimum and maximum values. We incorporate numerous variables in the imputation model to increase predictive power and satisfy the MAR condition. Notably, we include questions on values from the Afro-, Arab-, Asia- and Latinobarometer Waves. These surveys include SEI component questions, measured for many years missing in the WVS (see Appendix Sections AII–AIII). Below, we run regression models on five imputed datasets to mitigate the influence of particular predictions, and calculate average coefficients and imputation-corrected errors. Importantly, we have tested different imputation-model specifications, and our results are very robust (see Appendix Section AIX).
Although listwise deletion likely yields more misleading results,Footnote 62 we impute numerous observations, and our results’ validity therefore hinges on the imputation model’s performance. Does it fairly precisely predict SEI for years not measured by WVS? We try to ensure the answer is ‘yes’, for instance by including various predictors. But ultimately, this is an empirical question, and different tests (see Appendix for details) do indicate that the imputed data are trustworthy. First, imputed SEI values could largely be driven by certain other variables (for example, socio-economic resources) that are also included in our regressions; hence, the ‘null results’ below might stem from high collinearity. Yet sensitivity tests using different control sets yield stable results, and variance inflation factor tests indicate that collinearity is, in fact, unproblematic.
Secondly, our null results might emanate from incorporating additional uncertainty on the imputed data points; we follow standard prescriptions and calculate imputation-corrected errors. How much the SEI standard errors are inflated by such uncertainty varies a lot with the model specification (see Appendix Tables A6–A7; there is a 44.3 per cent increase on average). Despite this, our null results are retained, except in two models, even when completely ignoring this uncertainty.
Thirdly, visual inspections indicate that the imputation model produces ‘sensible’ predictions. Finally, over-imputation tests show surprisingly good results – we cannot know whether the imputation model provides correct predictions for missing SEI values, but it very accurately reproduces actual values. In sum, the imputation model predicts well, and different tests indicate that problems with the imputation procedure are not driving our results.
Operationalization of Main Variables
IW favor a substantive democracy definition, as opposed to defining democracy according to formal institutions such as contested elections. Their Effective Democracy Index (EDI) is constructed by multiplying the Freedom House Index (FHI) by a measure of corruption or rule of law. The resulting index, they argue, is a more realistic indicator of how democracy functions, capturing whether governing elites actually safeguard civil and political rights. We use EDI – multiplying FHI by ‘Control of Corruption’ from World Governance Indicators, as in IW – when replicating and investigating the robustness of their results. Yet EDI suffers from serious validity and reliability problems,Footnote 63 and Teorell and Hadenius report that the EDI–values correlation is largely driven by EDI’s corruption component.Footnote 64 Although proponents of EDI claim that it is the most valid measure of democracy,Footnote 65 we also present models using FHI, which taps a broad, substantive democracy concept,Footnote 66 and should thus be appropriate for testing RTM. We also conduct robustness tests with other democracy measures, such as the Polity Index.
IW consider self-expression values to consist of different related aspirations, such as autonomy, participation, tolerance and trust, at one end of a dimension and survival values at the other. WVS contains different indicators potentially tapping this dimension. To replicate IW, we use their additive index (SEI) based on five indicators. The first indicator measures generalized trust; the second the propensity to engage in civic action, reporting whether respondents have signed (or would consider signing) a petition; the third tolerance of others’ liberty, as indicated by tolerance of homosexuality; the fourth feelings of happiness; and, the fifth post-materialistic aspirations of liberty. The resulting additive SEI ranges from 0 to 1.Footnote 67
Empirical Analysis
Among WVS-participating countries, those with more widespread self-expression values are generally also more democratic. However, countries with such values, for example Sweden or the United States, have often been democratic for many decades, and the correlation may be due to different mechanisms. If self-expression values are indeed a key determinant of democracy, countries with (predominantly) such values should mainly experience democratization rather than sliding toward autocracy, and countries dominated by survival values should not predominantly experience democratization. Although the first pattern holds in the sample, the second does not: For countries scoring below the median on SEI, 80 per cent of the changes were toward more democratic regimes (using FHI) between the second and third (alternatively the fourth if not participating in the third) WVS waves. The equivalent number is even higher (about 85 per cent) for the lower quartile of SEI observations. Furthermore, Figure 1 showed that positive or negative changes in SEI were not systematically associated with particular changes in FHI either. However these are only bivariate patterns, and we investigate the values–democracy relationship more thoroughly below.
Do Self-Expression Values Cause Democracy?
IW find that self-expression values enhance democracy, and report this as robust to controlling for different variables. Taking their analysis as our starting point, we replicate their Model 5 from p. 199. This model (A1, Table 1) uses EDI as the dependent variable, and Vanhanen’s Index of socio-economic resources and years under democracy as controls.Footnote 68 We include only fifty-eight countries, whereas IW have sixty-one – for example, we do not include East and West Germany as different countries with both having assigned Germany’s post-reunification democracy scores, but exclude East Germany.Footnote 69 Our cross-section OLS model replicates the strong positive relationship. SEI is significant at 0.1 per cent, and the point estimate indicates that a one-unit increase is associated with about a 1.4 point higher EDI, more than EDI’s range.Footnote 70 The result is further strengthened when substituting socio-economic resources with log GDP per capita. Still, the results are fairly sensitive to other minor adjustments, such as using rule-of-law-based EDI (only significant at 5 per cent) or modifying the lag structure by a few years.Footnote 71
Table 1 Replicating and Adjusting Models in Inglehart and Welzel (Reference Inglehart and Welzel2005), accounting for Omitted Variable and Sample Selection Biases
Note: standard errors in parentheses. Models A1–A6 are OLS on cross-country data. Models A7–A10 are OLS PCSE with errors adjusted for panel-specific AR(1) autocorrelation and heteroskedastic panels. Following IW, the dependent variable is measured in the late 1990s or early 2000s in OLS models; control variables are measured for similar years as IW, and SEI and independent FHI variables are lagged seven years. Models A5–A10 include imputed data; coefficients are averaged over five imputed samples and errors are imputation corrected. Imputed observations in A6–A7 are measured in 1995. Maximum time series (dependent variable) in Models A7–A10 is 1988–2009; independent variables are lagged seven years. +p<0.10, *p<0.05, **p<0.01, ***p<0.001
Generally, IW find somewhat weaker evidence of a relationship when using FHI as the dependent variable. Model A2 replicates their model of values and ‘formal democracy’, controlling for socio-economic resources and lagged democracy.Footnote 72 SEI is positively related to FHI, although significant only at 10 per cent.
Many proposed determinants of democracy are highly sensitive to including and excluding particular controls.Footnote 73 Also when testing the effect of SEI, several controls are relevant, but we follow IW’s proposed list.Footnote 74 Thus Model A3 adds measures of religious composition (Protestants – Muslims), years of schooling, and ethnic fractionalization alongside socio-economic resources and democratic tradition.Footnote 75 Whereas IW always find a significant coefficient at 0.1 per cent, SEI is significant only at 5 per cent in A3, and the coefficient size is reduced by 37 per cent from A1. Furthermore, when substituting EDI – a highly problematic measure of democracy – with FHI, Model A4 yields no relationship (t=1.44).
Still, the results could be affected by sample-selection biases. As discussed, selection into WVS waves is likely not completely random; democratic and liberal Western countries are, for instance, over-represented in Models A1–A4. Thus we conducted the described multiple imputation procedure to adjust for selection biases, thereby increasing the number of countries from fifty-two to ninety-two.Footnote 76 Models A5 (EDI) and A6 (FHI) are otherwise similar to A3 and A4. When including imputed data, even the EDI-based result turns insignificant (t=1.14), and SEI drops dramatically in size and t-value when FHI is the dependent variable (t=0.16). Hence, selection biases may have contributed to the result reported by IW.
Cross-country OLS models utilize limited information – and arguably the wrong type of information (see below) – since they disregard developments within countries over time. The imputation model allows us to predict SEI in years it is missing, and we run OLS models with panel-corrected standard errors (PCSE) using country-year as the unit.Footnote 77 Models A7 (EDI) and A8 (FHI) include similar controls to the cross-section models to facilitate comparison. Including time-series variation strengthens the result again; SEI is now significant at 0.1 per cent for EDI, as the standard error drops when including time-series information. Still, SEI remains insignificant (t=0.58) when using FHI.Footnote 78
IW proposed more controls than are included in Table 1, namely the Gini income inequality index, government welfare expenditure – army expenditure and exports per capita. We excluded these above due to few degrees of freedom, but including them is less problematic when using imputed data and time-series information. The extended models are B1 (EDI) and B2 (FHI) in Table 2. All models in Table 2 employ samples including only countries with ≥2 non-imputed SEI observations, but this is not critical as the results are retained when including countries with one SEI observation or all WVS-participating countries (see Appendix Section BIII). While B1 and B2 still show that the choice of democracy measure influences estimates – the t-value of SEI is 3.49 for EDI and 0.64 for FHI – both coefficients shrink when also controlling for inequality, public spending and trade.
Table 2 Accounting for Country-Fixed Effects and Endogeneity of Values
Note: standard errors in parentheses. Data are from fifty-eight countries with ≥2 SEI observations. Coefficients are averaged over five imputed samples, and errors are imputation corrected. Errors are adjusted for panel-specific AR(1) autocorrelation and heteroskedastic panels in PCSE models, and clustered on country in FE models. Arellano-Bond and system GMM models consider socio-economic resources, public spending and SEI to be endogenous. Maximum time series (dependent variable) is 1988–2009; independent variables are lagged seven years. +p<0.10, *p<0.05, **p<0.01, ***p<0.001
There are thus legitimate reasons to question whether self-expression values enhance democracy, especially if one does not consider EDI a valid measure. However, even scholars who argue that EDI is valid should not conclude that there is a causal effect. The above results may be influenced by democracy affecting values, or by country-specific characteristics systematically relating to both values and regime type. If self-expression values indeed enhance democracy, we should also observe a relationship in FE models controlling for country-specific characteristics. The FE model on EDI (B3) reports a small, insignificant (t=0.61) SEI coefficient, as does Model B4 using FHI (t=0.21). Since the FE models only incorporate within-country variation, they guard against omitted variable bias related to time-invariant, country-specific factors that influence regime type. Still, FE models are inefficient when estimated for slow-moving variables such as SEI, potentially leading to Type II errors.Footnote 79 Thus to ensure we are not falsely discarding any effect, we tested more efficient random effects (RE) models. Yet Models B5 and B6 do not identify any effect of SEI either.
The positive cross-country correlation between self-expression values and democracy may therefore be spurious. Borrowing from an analogous argument in Acemoglu et al. on income and democracy, deeper contextual variables could have had a crucial impact on both citizens’ intrinsic ideals and the accountability of their governments.Footnote 80 Some societies, such as the Scandinavian, may have embarked on development paths where people gradually acquired liberal-democratic values and political institutions became increasingly transparent and accountable, while others, such as China or Ethiopia, embarked on paths of repressive governments and citizens valuing authority and tradition. These divergent development paths could result from differences in the characteristics of institutions that were adopted at critical junctures in history, for instance during colonial rule.Footnote 81
A second explanation of the spurious relationship concerns the spatial clustering of democracy and of particular values – IW, for example, speak of different ‘cultural zones’. Regarding strategies for transitions to (and consolidation of) democracy, populations and rulers are more likely to learn from and emulate those in neighboring countries.Footnote 82 Different regional hegemons – such as the US in the Americas and Russia in Central Asia – also have different regime type preferences for their respective neighboring countries.Footnote 83 Hence (culturally similar) neighboring countries often also have similar regime types. To investigate whether this explains the above-identified omitted variable bias, we added IW’s cultural zone dummies to the OLS PCSE Models B1 and B2 (see Appendix Table B12). Indeed, this draws the SEI coefficient from B1 (EDI) much closer to the FE coefficient in B3 (which is 0.04). More specifically, the coefficient drops from 0.32 to 0.15, and is now only significant at 10 per cent. This suggests that regional diffusion mechanisms contribute to explaining the spurious relationship between values and regime type. Regarding the PCSE model using FHI, the SEI coefficient becomes negative (but remains insignificant) when including cultural zone dummies.
A third possible explanation relates to divergent histories of democratic rule. Some countries have long histories of elections, civil liberties protection, and other democratic institutions and practices, whereas others do not. A democratic history is among the most robust determinants of current democracy,Footnote 84 and self-expression values may – as institutional learning theory suggests – have been shaped by citizens experiencing and enjoying democratic rights, open debate and ‘compromise politics’ over multiple decades. Hence, a prior democratic history could induce a spurious relationship between self-expression values and current democracy. The models above control for IW’s ‘democratic tradition’, but this has a peculiar construction; it only registers democratic experience for current democracies. Furthermore, it simply counts the number of years with a Polity score ≥6, and does not distinguish between differential impacts from living under high- and low-quality democracies. Although manipulated elections may induce learning of democratic norms,Footnote 85 living in a high-quality democracy such as Sweden or Denmark plausibly enhances self-expression values more than living in a ‘flawed democracy’. Hence, we substitute Democratic Tradition with Democratic Stock from Gerring et al. – which, importantly, incorporates the graded nature of democracy – in the OLS PCSE Models B1 and B2.Footnote 86 Consequently, SEI drops by 61 per cent and remains far from significant (t=0.25) when FHI is the dependent variable, whereas it drops by 55 per cent and becomes insignificant (t=1.61) when EDI is dependent (see Appendix Table B11). Thus values have no clear effect on democracy – independent of the democracy measure used – once pre-existing democratic history is appropriately controlled for.
These results on how democratic history shapes both the prospects for current democracy and self-expression values point further toward the above-discussed endogeneity problem. Previous studies using cross-country correlations to infer that self-expression values produce democracy may simply have picked up a reverse causal effect, which is congruent with the above-discussed literature on institutional learning. The endogeneity of values should thus be accounted for before concluding whether they affect regime type. To mitigate such endogeneity biases, we run Arellano-Bond GMM models. Importantly, these models account for both the endogeneity of values and country-fixed effects.Footnote 87 Furthermore, FE models are less efficient when time series are relatively short. Although our FE models do draw on time series of twenty-one years, Arellano-Bond models are arguably more appropriate. We ran several Arellano-Bond specifications, for instance varying the lag structure, but did not find any positive effect of SEI on either EDI or FHI. For example, Models B7 (EDI) and B8 (FHI) in Table 2 – using seven-year lags and including countries with SEI observations from ≥2 WVS waves – report minuscule t-values of 0.02 and −0.27, respectively. Still, Arellano-Bond models perform sub-optimally for slow-moving variables. Blundell and Bond, using Monte Carlo simulations, show that system GMM models perform better in such contexts.Footnote 88 These models still account for country-specific effects, and augment the Arellano-Bond instrumentation strategy when dealing with endogenous regressors. Interestingly, later studies report that the null results on income and democracy from Acemoglu et al. are altered – instead identifying a positive effect – when using system GMM models.Footnote 89 Likewise, such models recover an effect of education on democracy in system GMM models where FE models yield none.Footnote 90 While values may also affect regime type, our FE and Arellano-Bond models fail to pick this up because values and democracy levels change slowly, possibly leading us to conduct Type II errors. This is, however, not the case. Self-expression values do not affect democracy according to the system GMM Models B9 (EDI; t=0.97) and B10 (FHI; t=−0.15).Footnote 91
All (non-)results in Models B3–B10 are robust. SEI remains insignificant when removing outliers and influential observations; when using the Polity Index as the democracy measure; when testing one-, five- or eight-year lags; and when applying different sample inclusion criteria. The results are also retained when adding year dummies to control for time trends. Throughout, we have operationalized values with SEI from IW. We re-ran our models using the values index from Norris, but still find non-significant results.Footnote 92 In sum, when accounting for endogeneity and country-fixed effects, neither the choice of democracy measure nor other specification choices matter; there is simply no evidence that self-expression values causally affect regime type.
Table 2 shows stronger results for other potential determinants of democracy that are entered as controls. Once accounting for country-fixed effects, Democratic Tradition is robustly related to current democracy levels, and even changes in democracy according to the GMM models. Welfare Spending – Military Spending is positive and significant in several models, as is Protestants – Muslims. Interestingly, socio-economic resources also affect democracy in many models. Although not robust, this lends some support to classic modernization theory. This finding contrasts with the null results on income and democracy from Acemoglu et al., but also with RTM as laid out by IW. RTM proposes that value changes is the transmitter of the effect of economic change on regime type; socio-economic resources should thus become insignificant once values are controlled for. Our results indicate otherwise.
Admittedly, RTM is not the only prominent theory of democratization that has failed to receive strong empirical support.Footnote 93 Nevertheless, other proposed determinants of democracy receive robust support in different analyses. Employing Extreme Bounds Analysis (on dynamic probit models), Gassebner et al. find that short-term growth, past regime transitions, OECD membership and fuel exports robustly relate to democratization, whereas income level, past transitions, former military leader as chief executive and democratic neighbors robustly explain democratic survival.Footnote 94 Indicating the relevance of the alternative explanations of democracy discussed above, this list includes international-political factors (OECD member, democratic neighborhood) and domestic elite factors (military leader, fuel exports that provide elites with resources to repress and co-opt). Some determinants are retained in panel models akin to those above; using system GMM models, Aslaksen finds that an oil-dependent economy harms democracy.Footnote 95 We re-ran our Table 2 models (see Appendix Section BIII), but including fuel exports; we also find some support for the ‘political resource curse’ thesis, although this is not robust in GMM models. Finally, concerning international-political factors, we re-ran our models including WAVE from Knutsen – registering whether or not a country’s last regime change was within one of Huntington’s reverse waves of democratization.Footnote 96 WAVE, which taps exogenous sources of democratization from the geopolitical climate and regional diffusion effects, is often highly significant. Thus, while self-expression values are unrelated to democracy, some international-political factors and features affecting the capacities and strategies of domestic elites are found to be related to democracy.
The Endogeneity Issue Revisited: Does Experience with Democracy Affect Values?
To reiterate, self-expression values have no effect on democracy once country-fixed effects and the endogeneity of values are accounted for. We have, however, not yet presented any positive evidence for the assertion that democracy causes self-expression values. To more directly investigate this issue, we also tested system GMM models with SEI as the dependent variable, and lagged Democracy and either Democratic Tradition or Democratic Stock as the independent variables (see Table 3). FHI and EDI were generally positive in different specifications, and sometimes statistically significant, but not robust. In contrast, Democratic Tradition and Democratic Stock were always positive and very robust. A long democratic history shapes values in a more liberal direction, as institutional learning theory anticipates.
Table 3 System GMM Models with SEI as Dependent Variable
Note: standard errors in parentheses. Data from fifty-eight countries with ≥2 SEI observations. Coefficients are averaged over five imputed samples, and errors are imputation corrected. FHI, EDI, Democratic tradition and Democracy stock are endogenous. Independent variables are lagged one year. Maximum time series is 1982–2009 for C1, C3 and 1982–2006 for C2, C4. +p<0.10, *p<0.05, **p<0.01, ***p<0.001
Importantly, this result holds when Democratic Tradition or Democratic Stock are treated as endogenous (as in Table 3; see Appendix Section BIV for specifications using different lag lengths, instrumentation strategies and estimators). Past regime type can, naturally, not have been affected by current self-expression values, but it could have been affected by factors correlating with current values, such as past values. Hence, there are far stronger indications that democracy causes more liberal values – at least when considering long-term experience with democracy – than that values affect democracy. As for growth,Footnote 97 democracy may assert stronger effects on citizens’ values over the longer run; cumulative historical experiences with democracy may matter the most.
Combined with the findings from the previous section, this supports our argument that the values–regime type correlation mainly stems from (1) democratization taking place because of factors unrelated to popular values and (2) subsequent experiences with democratic institutions that shape values in more liberal directions.
Extension I: Separating Democratization and Democratic Stability
We failed to find evidence that self-expression values systematically affect regime type. Nevertheless, while GMM models account for country-fixed effects and the endogeneity of values, self-expression values could affect democratization and democratic survival quite differently. Citizenries with strong self-expression values may, for example, stabilize existing democracies but have no impact on democratization. Thus we dichotomize FHI and run dynamic probit models. The choice of FHI cut-off is arbitrary,Footnote 98 and we therefore test two different thresholds provided by Freedom House (‘free’, ‘partly free’). We also use the dichotomous Democracy and Dictatorship (DD) (minimalist) regime measure.Footnote 99
The point estimates on democratization are actually negative, but insignificant, for the low-FHI threshold and for ACLP – both when employing one-year or seven-year lags (see Appendix Section BI). When applying the high-FHI threshold – thus investigating transitions from either unfree or partly free to free regimes – the point estimates are positive, but still insignificant. In sum, there is no evidence that self-expression values induce democratization. This is not very surprising to us, given the above discussion on how factors other than those highlighted by RTM, notably including international-political and domestic elite-level factors, have constituted the main drivers of many democratization experiences.
However, and far more surprising, self-expression values do not stabilize existing democracies either. Indeed, three of our six models in Appendix Table B1 show a negative point estimate, and – although the model using seven-year lags and a high-FHI threshold shows a weakly significant association – the relationship is always insignificant at 5 per cent.
Despite the widely accepted arguments on how self-expression values enhance democratization and democratic survival, these results simply reflect that most countries have not followed the theoretically suggested patterns. One illustrative example is South Africa. Although there was mass mobilization prior to the transition to democracy, the pacted nature of the transition following apartheid – ensuring that rich, white elites would keep their property while, in practice, transferring political power to ANC elites – was critical in allowing for democratization. In contrast, Venezuela experienced a small increase in SEI leading up to Hugo Chavez’ consolidation of presidential power, but still experienced reversed democratization under Chavez with the curtailing of civil liberties and electoral manipulation. Chavez drew notable popular support from the masses because of promises of redistribution, perhaps suggesting that material benefits rather than self-expression values drive popular preferences for regime type.Footnote 100 Belarus experienced a fairly large increase in SEI from 1990 to 2008, but the regime’s ties to Russia and extensive repression helped it avoid democratization.Footnote 101
A few countries – notably ex-communist countries such as Lithuania – have followed the ‘predicted pattern’ by having high and slightly increasing SEI scores accompanied by democratization, and then consolidation. However a closer look at these countries’ experiences casts further doubt on the validity of the values–democracy link by highlighting the above-discussed importance of international factors for democratization. The democratization experiences in Central and Eastern European countries such as Lithuania were strongly influenced by linkages to (and leverage from) Western countries;Footnote 102 EU membership possibilities, and later EU accession, may have disciplined elites to commit to democratic governance. Possibly, then, increasing self-expression values in Lithuania played little effective role. Indeed, Latvia – which experienced the second steepest drop in SEI after Iran in Figure 1 (bottom scatterplot) – had an identical increase in FHI to Lithuania. Bosnia was among the few other countries that experienced both increasing SEI scores and notable democratization during the time interval in Figure 1. However, Bosnia’s democratization related to international occupation and peacekeeping after the ethnic civil war. In sum, there are few clear-cut cases that showcase the mechanisms outlined by RTM. In the cases that fit the pattern, like Lithuania (or even the Czech Republic), it is difficult to establish that the extensiveness of self-expression values was a prerequisite of popular collective action, and to identify the relative importance of popular action compared to other factors, such as external influence.
Extension II: Components of Self-Expression Values
Despite the null results above, certain SEI components could affect democracy, but not others. Indeed, the components may tap fairly different factors; perhaps, for example, social trust affects democracy, whereas life satisfaction does not.Footnote 103 Yet we find little evidence of this.
We re-ran the models in Table 2, sequentially substituting the five components for SEI (see Appendix Section BII). No FE or GMM models show even weakly significant effects. Post-materialism and happiness are always unrelated to democracy, whereas tolerance is positively related, but only in the PCSE (significant 5 per cent) and RE (10 per cent) models using EDI. However civic action – reporting whether people have signed or would consider signing a petition – is the strongest. Independent of the democracy measure used, it is significant at 1 per cent in PCSE models and at 10 per cent in RE models. This is interesting in light of our discussion of collective action problems; it is the only indicator relating directly to political actions, and not only values or perceptions. Yet civic action is also insignificant in FE and GMM models; we cannot conclude that it causally affects democracy. Importantly, signing petitions is likely endogenous, since such activities carry less risk for citizens in democracies than in autocracies.
Further, generalized trust is always negatively signed, although it is only significant in the PCSE model using FHI. This finding is particularly interesting, given the large literature on generalized trust/social capital and democracy. Whereas some authors propose that the observed positive correlation comes from high trust-generating democratic improvements, others argue that it mainly stems from citizenries living under democratic institutions building trust.Footnote 104 Corroborating Muller and Seligson, we find no evidence that trusting citizens enhance democracy levels.Footnote 105 Moreover, we test dynamic probit models and find that trusting citizens neither induce democratization nor stabilize existing democracies.
Conclusion
The analysis above finds no evidence that self-expression values enhance democracy levels, democratization prospects or democratic durability. Hence, we doubt the popular proposition that having a relatively liberal population is a requirement for viable democracy. Previous conclusions that such values have a strong effect may have been driven by a combination of sample selection, omitted variable and endogeneity biases. Critics might interject that the above results are non-credible because of the missing data and our reliance on multiple imputation – despite the good performance of our imputation model. Our conclusions are, of course, associated with uncertainty. But poor data coverage is a problem for the wider literature and is not unique to our analysis. In the absence of broader data coverage and new studies, skeptics should follow standard norms of scientific inference and conclude that there is currently little empirical evidence that self-expression values have a positive effect on democracy. In other words, the absence of actual data should not lead to the conclusion that there is a positive relationship; it is inappropriate to reject the null hypothesis that there is no relationship between values and democracy based on studies that use limited data and do not account for important sources of bias.
We document above that the observed relationship between self-expression values and democracy dwindles when including multiple relevant controls, or when accounting for country-fixed effects. The latter result is strikingly robust and suggests that deeper political-historical or geographical factors determine both a country’s level of self-expression values and democracy. Moreover, we find evidence indicating that countries’ prior historical experiences with democracy and regional diffusion effects help explain why some countries today are both democratic and have liberal populations, whereas others are autocratic with illiberal populations. We also present analyses that distinguish between democratization and democratic survival. We find that citizenries with stronger self-expression values neither enhance the survival of existing democracies nor improve democratization chances. The latter result suggests that IW’s model – which assumes that self-expression values have a strong and consistent effect on democratization through collective action – is too simple to capture the dynamics of institutional change in autocracies. In line with the recent literature on authoritarian survival, we contend that in countries where autocrats have deep pockets and control the guns, regime change requires more than a public preference for freedom and autonomy. We highlight how the cross-country correlation between self-expression values and democracy stems from democratization processes that are triggered by other factors – such as international interventions, coups or bargains between elite groups – and that the subsequently established democratic institutions gradually nurture a more freedom-oriented citizenry that is inclined to appreciate and defend democracy.
