How Satisfying Is Satisfaction Research?

After reviewing more than 70 studies with the SWD dependent variable, we note three problems that prevent progress toward better understanding of how it varies across countries. The problems are conceptual, theoretical and statistical.

First, studies of SWD are not having the impact they should in mature democracies at least in part because of the legacy of debate about the concept behind it. The SWD measure grew out of a field of study focused on the stability of democratic regimes and was used to gauge regime support and legitimacy (Easton, Reference Easton1965, Reference Easton1975; see also Canache et al., Reference Canache, Mondak and Seligson2001; Linde and Ekman, Reference Linde and Ekman2003; Linde and Dahlberg, Reference Linde, Dahlberg, Bågenholm, Bauhr, Grimes and Rothstein2021). Yet it has become a standard question in surveys in consolidated democracies, and indeed most of the research using the question is now on cross-national datasets from these countries. Most of those working with SWD data from these countries justify the data's use by arguing that it provides a valid measure of how democracy is practised in a specific place, as experienced by the citizen (see, for example, Bernauer and Vatter, Reference Bernauer and Vatter2012; Curini et al., Reference Curini, Jou and Memoli2012). However, there seems to be continuing concern that SWD may measure support for democracy as a form of governance, on the one hand, or measure support for the current government, on the other.

We agree with recent interventions that argue SWD should be interpreted as measuring a concept between these two extremes (Valgarðsson and Devine, Reference Valgarðsson and Devine2022; Norris, Reference Norris and Norris2011). In mature democracies, SWD is not a measure of preference for democracy as against other forms of government, since in nearly all these countries over 90 per cent of citizens think that democracy is best, yet aggregate SWD varies from about 40 per cent to 90 per cent. The other extreme would be to say that SWD is merely a judgment of the current authorities (Linde and Ekman, Reference Linde and Ekman2003), but this can't be, since in most countries aggregate satisfaction moves only glacially as compared with much bigger swings in government support (Christmann, Reference Christmann2018). So the field should be more confident that the SWD question is the clearest and most consistent indicator of the balance sheet for democracy within and across countries (Kölln and Aarts, Reference Kölln and Aarts2021) and the single best indicator for learning about the effects of institutions on the quality of democracy.Footnote ¹

The second problem involves the tension between macro-level concerns and individual-level data. The issue is not that analysis proceeds at the macro level, with macro-level determinants linked to macro indicators such as SWD. We do not quarrel with macro work confined to the macro level (Devine and Turnbull-Dugarte, Reference Devine and Turnbull-Dugarte2021) or that includes only socio-demographic controls (Quaranta and Martini, Reference Quaranta and Martini2016). The problem is that many recent studies include individual-level attitudes and characteristics as determinants of SWD right alongside the macro factors. There is often a suggestion that country-level institutions and contextual factors operate causally through individual-level experiences and attitudes. However, these arguments are usually incompatible with the statistical model deployed in the empirical work. Moreover, theory is usually only provided for the one or two novel macro-individual linkages that are the focus of a given paper (for example, direct democracy for Bernauer and Vatter, Reference Bernauer and Vatter2012; or the “party choice set” for Dassonneville and McAllister, Reference Dassonneville and McAllister2020).

The notable exception to the general lack of attention to the path from macro influences through to SWD is Christmann (Reference Christmann2018), who uses individual-level data to calculate a composite “democracy perceptions index” and shows that “the effect of ‘objective’ performance is almost completely mediated by subjective evaluations of it” (85). This is, in essence, a reduced-form version of the model we present. We take up the challenge he presents in his concluding paragraph: “It would be an interesting contribution to disaggregate the various attributes of democratic quality and test what exactly drives the relationship [of macro factors] with SWD” (88).

It is impossible at present to cumulate findings across studies because the empirical-theoretical specifications—the macro and individual-level factors chosen for inclusion in a particular study—are so varied. Our Appendix Table A1 inventories 21 papers examining contextual or institutional influences on SWD. We show a matrix of theoretically plausible determinants of SWD and catalogue which variables were present in those papers’ empirical specifications. There is little commonality across studies in either the macro variables or the individual-level variables that have substantive attitudinal-behavioural content. This means that conclusions about the macro-level factors will vary across studies because they affect only the residual variance after the individual factors are partialled out. This may sound like a technical concern, but it is in fact deeply theoretical.

Furthermore, a practical problem is that because macro-level variables can only affect country-level averages of individual-level variables, researchers are in a classic cross-national degrees-of-freedom problem that is often under-acknowledged (see Scruggs, Reference Scruggs2001; Stegmueller, Reference Stegmueller2013; also Christmann, Reference Christmann2018: 85, who acknowledges this explicitly). In studies that include a select few macro-level variables, readers are left wondering whether those are the real macro-level influences or if they are merely correlated with other plausible ones (see, for example, Magalhães, Reference Magalhães2014). Solving this problem is a matter of strong but careful theorizing about the causal processes from exogenous macro factors to citizen attitudes. Without a link from a macro factor to an individual consideration that then feeds into satisfaction, the claim about the macro influence has a “black box” quality.

The third issue is that the statistical models used in the literature, even multilevel ones, mirror this lack of careful theorizing of the macro-micro causal structure. Inclusion of individual-level attitudes and behavioural indicators that are themselves caused by contextual factors will typically produce a total effect of context and institutions that is biased downward, as Christmann (Reference Christmann2018) argues. According to the leading statistical researchers in this area: “MLM approaches do not accommodate mediation pathways with Level-2 outcomes and may produce conflated estimates of between- and within-level components of indirect effects” (Preacher et al., Reference Preacher, Zyphur and Zhang2010: 209).

To take but one example: even Wells and Krieckhaus (Reference Wells and Krieckhaus2006), who aimed to modernize the statistical methodology in SWD research, potentially run afoul of this problem. In reanalysis of the Anderson and Guillory (Reference Anderson and Guillory1997) data, they include political interest as a determinant in a model that attempts to find a link between consensus democracy and satisfaction. But it can easily be argued that consensus democracy promotes citizens’ interest in politics by preventing the disaffection produced by majoritarian structures (see, for example, Lijphart, Reference Lijphart1997; Anderson, Reference Anderson, Loewen and Rubenson2019). If so, this specification might conclude that consensus institutions do not increase SWD because those institutions’ total effect on SWD, specifically the portion mediated by interest, is not evident in the results.

In the language of multilevel mediation analysis the appropriate structure for SWD research is a 2-1-1 design (Preacher et al., Reference Preacher, Zyphur and Zhang2010; Fang et al., Reference Fang, Wen and Hau2019). Country context (level 2) affects citizens’ experiences, attitudes and behaviours (level 1), which then affect SWD (level 1).Footnote ² The individual-level attitudes, experiences and behaviours are mediators of the contextual effects. Preacher et al. (Reference Preacher, Zyphur and Zhang2010) show that the multilevel structural equation model (MSEM) is the appropriate statistical specification.

A Macro-Micro Theoretical Model of SWD

The first step is to start at the end of the chain and specify the relationship between possible mediators and the outcome of interest. What determines how satisfied are citizens with the way democracy works in their country?

According to survey response theory and research, answers to the SWD question represent spontaneous reactions, not deeply considered, crystallized attitudes (Tourangeau et al., Reference Tourangeau, Rips and Rasinski2000). Responses are constructed by sampling from “considerations” (Zaller, Reference Zaller1992). When asked how democracy “works,” then, citizens may think of a variety of different aspects of democracy and their personal experiences with it, as Kölln and Aarts (Reference Kölln and Aarts2021) demonstrate. We should therefore theorize the considerations that might be relevant as respondents answer.Footnote ³

We imagine the considerations have roughly additive effects: any one of them might add or subtract to overall satisfaction irrespective of the levels of the others. For example, if a person feels that government policy is a long distance from their ideal, that person will be somewhat less satisfied, even if they perceive the system as one that is fair, hears their political voice and results in good economic outcomes.

Because we ultimately want to link institutional differences to the quality of democracy, our theoretical specification of intervening individual-level considerations must be exhaustive. We need to account for all the attitudes and behaviours that might have a statistically significant effect on SWD so that it is plausible that all contextual-institutional effects will operate through these individual-level variables.

It is important to note that any particular institutional or contextual factor may have different effects on different individual-level considerations. Institutions are best understood as trade-offs, so the modelling must allow for this. While it may be enlightening to learn about the reduced-form total effect of an institution on SWD, the more satisfying approach we propose will enable researchers to see how a given institution may contribute positively in some respects, while detracting in other ways.

We propose the following exhaustive set of considerations—individual-level criteria for democracy—that may affect SWD. In doing so, we are drawing on a huge literature recently summarized by Linde and Dahlberg (Reference Linde, Dahlberg, Bågenholm, Bauhr, Grimes and Rothstein2021). For each consideration, we provide some representative citations that demonstrate the linkage (or themselves review a literature) to support our claim that these cover the important factors in citizens’ judgments of democracy. Perhaps there are others, but it is at these ten that we drew the line; others can be proposed, but surely most can be subsumed into one of the ten.

Citizens will think democracy works well if the following are present:

1. 1. Engagement. Citizens understand politics and are engaged with it when they wish to be. Furthermore, engagement indicates participation in what we broadly think of as democratic deliberation—even if this is relatively passive participation—which will likely lead to a better understanding of the rationale for political decisions and policies (Warren, Reference Warren2017). From a psychological perspective, unfamiliar objects, or ones about which there is confusion, are viewed more negatively than familiar ones (Garcia-Marques et al., Reference Garcia-Marques, Prada and Mackie2016). Engagement is also precondition for active participation in the process of representation, even minimally in terms of voting (Quintelier and Van Deth, Reference Quintelier and Van Deth2014; Chang, Reference Chang2018).

2. 2. Representation. Citizens are represented in the formal representative institutions of parliaments and governments, including the total party system. That is, at least some elite political actors, usually parties and their legislators, are acting in those citizens’ interests. In instrumental terms, this reduces the risk that policy will end up a long way from a citizen's ideal point (Anderson, Reference Anderson and J2010; André and Depauw, Reference André and Depauw2017; Ferland, Reference Ferland2021; Urbaniti and Warren, Reference Urbinati and Warren2008). This consideration is where winner-loser status enters our model (see Nadeau et al. [Reference Nadeau, Daoust and Dassonneville2021] for a review of the literature).

3. 3. Voice. Citizens’ views are voiced in the political dialogue, including in parliaments but also in the courts, the bureaucracy, the media, and civil society. They feel heard in elite discourse, even if it is others who are speaking (Merkley et al., Reference Merkley, Cutler, Quirk and Nyblade2019; Hoerner and Hobolt, Reference Hoerner and Hobolt2020). Dahlberg and Holmberg (Reference Dahlberg and Holmberg2014), Reher (Reference Reher2015) and Ezrow and Xezonakis (Reference Ezrow and Xezonakis2011) find that the closer citizens are to their preferred party, the higher their SWD, irrespective of whether their party is in government.

4. 4. Ideal policy. Actual policy is, or citizens predict it will be, close to citizens’ ideal point in a multidimensional space (Curini et al., Reference Curini, Jou and Memoli2012; Dahlberg and Holmberg, Reference Dahlberg and Holmberg2014; Ferland, Reference Ferland2021; Mayne and Hakhverdian, Reference Mayne and Hakhverdian2017). This consideration explicitly involves policy outputs, independent of how policy gets made.

5. 5. Well-being. Citizens are materially secure (Kölln and Aarts, Reference Kölln and Aarts2021; Nadeau et al., Reference Nadeau, Daoust and Arel-Bundock2020). We expect that citizens care primarily about their own and their family's well-being and that they almost universally use their own perceptions of security and well-being as a proxy for a judgment about the political system's effect on them (Quaranta and Martini, Reference Quaranta and Martini2016, Reference Quaranta and Martini2017).Footnote ⁴

6. 6. Fairness. Irrespective of policy outputs and representation, democracies are built on fairness, and so we expect perceptions of fair treatment to strongly influence satisfaction with how democracy works, for both selfish and altruistic reasons (Magalhães, Reference Magalhães2016; Magalhães and Conraria, Reference Magalhães and Conraria2018; Mutz and Mondak, Reference Mutz1997). Fairness of the electoral and judicial systems are important elements of this overall judgment. The extent of social protection may be an indicator of fairness that influences SWD (Lühiste, Reference Lühiste2014).

7. 7. Effectiveness/efficiency/policy output. Citizens believe that the system discourages corruption and waste and is able to produce welfare-maximizing collective goods. That is, citizens have perceptions about the extent to which authorities get things done, whether or not the voter agrees with the things that get done (Adserà et al., Reference Adserà, Boix and Payne2003; Christmann, Reference Christmann2018; Dahlberg and Holmberg, Reference Dahlberg and Holmberg2014; Magalhães, Reference Magalhães2014; Quaranta and Martini, Reference Quaranta and Martini2017).

8. 8. Accountability. Citizens believe the system is arranged such that citizens can express their displeasure with government performance so that parties and persons are removed from positions of authority when their performance is wanting (Aarts and Thomassen, Reference Aarts and Thomassen2008; Powell, Reference Powell2000; Przeworski et al., Reference Przeworski, Stokes and Manin1999).

9. 9. Responsiveness. Citizens believe that politicians respond to changes in citizens’ preferences with good judgment and then suitable legislative and executive action (Armingeon and Guthmann, Reference Armingeon and Guthmann2014; Soroka and Wlezien, Reference Soroka and Wlezien2010; Torcal, Reference Torcal2014; Wlezien and Soroka, Reference Wlezien and Soroka2012).

10. 10. Transparency. Citizens believe that it is possible to observe the actions of governments and other authorities and that their actions are justified by public reasons (Baume and Papadopoulos, Reference Baume and Papadopoulos2018; Bentham, Reference Bentham and Schofield1990; Hollyer et al., Reference Hollyer, Rosendorf and Vreeland2019). Voters are likely to see a lack of transparency as a shield for corruption, which will diminish satisfaction (Anderson and Tverdova, Reference Anderson and Tverdova2003).Footnote ⁵

These are concepts, of course, and even the most complete surveys do not have ideal measurement of all of them. Nonetheless, we aim to build an estimation model of satisfaction where these are the only systematic individual-level (attitudinal-experiential-behavioural) causes of feelings of democratic satisfaction. To take one institutional factor, for example: that a proportional electoral system makes each citizen less likely to think their vote is wasted and more likely to feel well represented, not that a proportional system has an effect over and above attitudes about wasted votes and whether the person is represented. We think that this macro-to-micro causal structure is, in fact, what all scholars have in mind when they include macro factors in models of SWD. Doing so more explicitly and using an appropriate statistical representation of that empirical model enables the opening of the black box of the relationship between institutions and SWD.

Model Validation

In order to validate this ten-determinants individual-level model, we estimate an empirical version of it with a simple individual-level regression of SWD on measures of these factors. Later on, we integrate this individual-level model with a macro-model, but for now the idea is simply to show that these are all important determinants of SWD. The data are from round six of the European Social Survey, 2012 (ESS), chosen because it was focused on perceptions of the institutional elements of democratic systems and therefore is the cross-national survey dataset that has the most complete set of our ten individual-level factors.Footnote ⁶

Table 1 gives the correspondence between the theoretical variables above and the best available measures in the ESS data.

Table 1.

Theoretical Variables and Operationalizations

In the online appendix (Table A2), we present correlations to show that these variables measure distinct attitudes. An ordinary least squares (OLS) regression predicting SWD is reported in Table 2. It is estimated using ESS respondents with complete data on all of these 10 measures from the following 23 countries: Belgium, Switzerland, Czech Republic, Germany, Denmark, Spain, Estonia, Finland, France, Hungary, Ireland, Iceland, Israel, Italy, Lithuania, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Sweden, United Kingdom. We excluded Albania, Cyprus, Bulgaria, Kosovo, Russian Federation, and Ukraine because their elections were not sufficiently free or fair to be considered as mature democracies to which our theory applies, and we had concerns with the interpretation of some of the survey questions in these countries.

Table 2.

Determinants of SWD (OLS regression)

Note: ESS data 2010–2012. N = 27,819. Standard errors adjusted for country clusters. SWD mean = 0.55; sd = 0.25 (all variables rescaled to 0-to-1). Bold coefficients are more than 1.95 times their standard errors.

Table 2 shows that these are ten important and separable determinants of satisfaction with democracy. System outputs (income, health and education) and fairness, transparency and responsiveness appear to be the stronger influences. We note that the only variable that has no effect is the very poor measure of voice, which asks whether the respondent feels closer to a political party. We expect that a better measure, such as the common question asking if a respondent has a “say in government” (not asked in ESS 2012), would be significantly linked to SWD.

Readers will note that this model contains no so-called control variables, particularly socio-demographics. Socio-demographics are potentially influential at a higher level of causality. Omitting them does not threaten our inferences at either the 2-1 or 1-1 levels, because they are orthogonal to the contextual factors and prior to the individual-level factors that we argue are driving SWD. A bicameral system, for example, cannot affect individuals’ socio-demographic characteristics.

To confirm that the theoretical model of ten major determinants of SWD (at the conceptual level) applies quite generally, we show in online appendix Table A3 an analogous regression estimation on German and Canadian election studies data (not the ESS data). The models include some similar and some different operationalizations, but they are all directed at measuring the theoretically specified categories elaborated above. The results align closely with those in Table 2, showing that these are indeed generally applicable determinants of SWD.

Macro Context and Institutional Influences on the ten Considerations

Then which institutions and contextual factors affect these considerations? Careful theorizing is necessary to justify a statistical model that embodies claims about the causal influence of the many possible institutional and contextual factors on these ten attitudinal-behavioural considerations, especially in this low-degrees-of-freedom situation. Of course, each macro determinant may affect only one or some of the individual-level determinants of SWD. Obviously, the list of institutional and contextual factors is virtually endless, and scholars have covered much ground, so we aim only to paint a panoramic picture of the possibilities. We enumerate what seem to us the most important macro-level influences, grouping them into three broad categories, and list under each one the variables that we ultimately include in the estimation models. We also point out which individual-level determinants the macro factors are likely to affect which of the ten individual-level determinants. For simplicity, we suggest here only the linkages that we included in our final models, limited of course by data availability. Variables that appear in our estimation models are in bold in the text below, and the individual-level variables they affect are in italics.

A Micro-Macro Empirical Model of SWD

Our theoretical model implies that we should construct an empirical model and use an estimation procedure that allows macro-level factors to influence country-level averages of individual-level attitudes and behaviours. These individual determinants then combine additively and probabilistically to determine responses to the SWD survey question.

We employ the framework of multilevel generalized structural equation modelling, or MSEM (Preacher et al., Reference Preacher, Zyphur and Zhang2010). Intuitively, this simply involves two layers of regression equations, where the level 1 (individual) equation is a regression of SWD on its influences:

The level 2 equations are, in effect, ten separate regressions of the country averages of these citizen attitudes/behaviours on the country-level factors that theory suggests will influence them. Each regression has an n of 23 (countries). For example:

It should be noted that this approach is more similar to “two-step” multilevel methods (Jusko and Shively, Reference Jusko and Phillips Shively2005) than to most of the multilevel modelling on multi-country survey datasets. Crucially, the degrees of freedom are low and estimates likely not very robust, as Bryan and Jenkins (Reference Bryan and Jenkins2016) carefully demonstrate.

One issue is the causal structure at the higher, macro level. Electoral systems are causal influences on party systems, for example, but we think both are potential determinants of some of the individual-level factors. To simplify, we separated hard, constitutional-institutional factors, which change only rarely or glacially, from contingent political, social and economic factors whose average levels may be affected by institutions but which also show a great deal of variation even when institutions are stable. The electoral system falls in the first category and the number of electoral parties in the second. In general, we included measurements from the second category because they show more variation and because the links between institutions and these more contingent factors are the subject of voluminous research by other scholars. We suggest that even more detailed theoretical work should accompany subsequent efforts to model the influences of these factors on SWD.

Our quantitative analysis began with models with a greater number of independent variables than those described above, but it became clear that only a few macro factors had significant effects on country levels of the various individual determinants. With only 23 countries, models of this kind cannot include extraneous variables. We do not have space to narrate our process of elimination, but it was informed by a review of the literature on each factor separately. In the results below, then, we present country-level models that include significant macro influences as well as ones for which there was very strong theoretical justification to expect an effect.

As a preliminary, Figure 1 gives the country means on satisfaction with democracy for the countries in our main dataset (ESS 2010–2012). It shows that the range in which macro factors can have their impact in the countries in the estimation sample is .38: from .36 (Slovenia) to .74 (Switzerland).

Figure 1.

Satisfaction with democracy by country, 11-point scale 0–1

Source: European Social Survey

Macro-Level Estimates

The presentation of our results here is atypical, but we could not think of a more economical way to present ten estimations, with different combinations independent variables, and still allow the reader to read the text alongside the results, avoiding ten separate tables. The unusual thing to note is that standard errors appear as superscripts above the coefficient.

(Measurement details are in the online appendix. For these country-level averages of individual-level survey questions, the concept appears underlined to begin the subsection, with the operationalized individual-level variable in italics and the range of national averages in parentheses. Coefficients are presented in line, with standard errors as superscripts. Note that all of the dependent variables here are scaled zero to one.)

Political Engagement (Interest, .29 [Lithuania] to .64 [Denmark]) = .49 -.12^(.037) Majority Government -.07^(.021) Government Alternation  + .01^(.011) Effective Number of Parties (ENP) votes + .001^(.0003) district magnitude (DM) + .001^(.0004) Population per MP. Root mean squared error (rmse) =.33

A country's level of political interest is affected negatively by both the presence of a single-party majority government and by low district magnitude. Surprisingly, interest is higher where governments have not been recently changed by elections, measured as the number of alternations over the last three elections. Interest is higher in countries with a higher number of citizens per member of parliament.Footnote ⁷

Policy Representation (Voted for a Party in Government, .16 [Portugal] to .47 [Finland]) = .26  +.16^(.061) Parties in Government. rmse=.47

The proportion of citizens who are winners varies, sensibly and strongly, with the number of parties in government. As the parties in government measure ranges from under .07 in the United Kingdom, Spain and Portugal to over .65 for Slovakia, Czech Republic, Israel and Belgium, the effect reaches a maximum of about .13. More encompassing government coalitions allow more voters to feel like winners. No other variables affect the proportion of winners.

Voice Representation (Closer to a Political Party, .27 [Poland] to .71 [Denmark]) = .35 -.03^(.057) ENP votes +.11^(.004) ENP seats -.004^(.003) ENPseats*ENPvotes − .05^(.026) Government Alternation. rmse=.49

The proportion of citizens who feel closer to one of the political parties increases with the effective number of campaigning parties. It declines as governments are changed more regularly in national elections. Unfortunately, the measure of voice representation is the weakest of the 10 determinants; we expect a better measure might be linked to more macro factors.

Well-Being (Feeling about Household Income , .42 [Hungary] to .87 [Denmark]) = .32 -.76^(.401) Economic Growth -.01^(.003) Social Expenditure  +.005^(.0009)GDP per capita$‘000 -.002^(.004) GINI. rmse=.13

Not surprisingly, subjective economic well-being is related to the national economy, static and dynamic, as well as to social expenditure, but not to inequality.

Fairness (Courts Treat Everyone the Same, .30 [Portugal, Slovakia] to .91 [Denmark, Norway]) = .10  +.09^(.01) Corruption -.01^(.004) GINI +.50^(.300) Gender Equality. rmse=.27

Feelings about procedural and legal fairness are related predictably to the presence of corruption and also to social (gender) inequality. Economic inequality is also linked to more negative perceptions of fairness.

Accountability (Are Governments Punished in Country , .31 [Italy] to .77 [Denmark]) = .69  +.00^(.000) DM -.04^(.029) Gov't Alternation -.02^(.001) Gov't Non-electoral Change -.02^(.041) Federal., rmse=.29

Macro factors are only weakly related to feelings about whether governments are punished for doing a bad job. We detect an effect in a counterintuitive direction, where countries that experience more government turnover have citizens who are less likely to agree that governments are punished for doing a bad job.

Responsiveness (Governments Change Policy in Response to People, .30 [Spain] to .65 [Switzerland]) = .49 -.03^(.021) Majority Government -.06^(.013) Gov't't Alternation  +.17^(.049)Parties in Gov't -.01^(.025) Federal. rmse=.24

Government alternation is strongly related to perceptions of responsiveness, but negatively, such that more frequent changes are linked to perceptions that government does not change policy. While perhaps this is counterintuitive, it is quite possible that citizens change governments frequently when they think governments do not respond to public opinion. We note that, so far, government turnover is associated with the negative side of four individual-level determinants (engagement, voice, accountability, responsiveness). The number of parties in government (measured 0 to 1) is positively linked to perceptions of responsiveness across countries (Dassonneville and McAllister, Reference Dassonneville and McAllister2020). Federalism and majority government make no difference to feelings about government responsiveness.

Transparency (Governments Explain Decisions, .33 [Spain, Italy] to .69 [Switzerland]) = .25 -.03^(.046) Majority Gov't -.06^(.025) Gov't Alternation  +.005^(.002) Transparency -.02^(.042) Federal. rmse=.26

Countries with more frequent government alternation have citizens less likely to feel that their government explains its decisions. As we would expect, one-party majorities are perceived as explaining their decisions less fully. Countries rated high on overall transparency by experts have citizens that share that view. Again, federal countries are no different.

Effectiveness (State of Health and Education , .42 [Portugal] to .74 [Denmark, Finland]) = .52 -.007^(.004) GINI +.003^(.001) GDPper capita $‘000  +.005^(.004) Social Expenditure. rmse=.19

Judgments of effective government provision of services are linked, negatively, to inequality, positively to the country's wealth, and weakly positively to the amount of social spending.

Policy Proximity (Average absolute distance from country L-R mean , .12 [Ireland] to .23 [Israel]) = .19  +.05^(.028) Parties in Gov't -.001^(.003) ENPvotes -.04^(.011) Federal -.001^(.0004) Transparency -.0003^(.0001) Population per MP. rmse=.14

Greater citizen polarization, surprisingly, is influenced by having more parties in government but not with more parties getting votes. Federal countries and more transparent ones turn out to have citizens more tightly clustered around the centre. And the more people per legislator, the closer the average citizen is to the country's mean citizen policy position.

Table 3 presents a summary of these findings. The results show how our empirical model permits the same macro factors to have separate, even opposite, effects on SWD through different intervening individual factors. We note, for example, that the number of parties in government has a positive effect on mean feelings about representation and responsiveness but a negative effect on citizens’ average closeness to the ideological centre of the country (polarization). Since feelings of representation and responsiveness increase SWD but polarization decreases it, the total effect of the number of parties in government on SWD is in the combination of these mediating effects. And this, of course, accords with previous research and theory about multi-party proportional systems versus two- (or two-plus) party systems.

Table 3.

Summary of Macro Effects on Country Averages of Individual Determinants of SWD

A Multilevel Structural Equation Mediation Model for SWD

We now put the two halves of the model together. The macro part and the individual-level part of the model are estimated concurrently using the ten individual-level determinants and then separate equations for the macro-micro relationships from the previous section (Preacher et al., Reference Preacher, Zyphur and Zhang2010). Estimates are from a multilevel structural equation model with free country-level variance. The MSEM is computation-intensive but is now conveniently estimated natively in Stata and also in R using the ‘lavaan' library.

We provide results in the online appendix (Table A4). The model performs well. The country-level variance of this model conditional on the macro factors influencing SWD through the individual-level attitudes is a scant .0011, which is one-tenth the country-level variance of a constant-only model, and the p-value from the likelihood ratio test for the necessity of the multilevel structure is an insignificant .18, up from .0007 in the constant-only model. We conclude that specification of the structural model obviates the multilevel model, in effect trading the multilevel model's description of variance across levels for the more satisfying substantive account of country-level variation as influenced by theoretically specified determinants.

Readers may hope for a comparison of this MSEM with the typical random-effects multilevel model that includes macro and individual factors together. This is not possible because we have 17 candidate macro factors and only 23 countries. We entered all 17 in a standard random-effects model, and the estimates were all over the place, with very high t-statistics and unreasonably large coefficient values. When we enter those same 17 macro variables one by one, as the only macro factor on top of our ten individual-level determinants, 16 are statistically insignificant; only economic growth shows a positive relationship with SWD. That further corroborates our claim that the ten individual-level determinants are exhaustive and sit between macro factors and SWD.

Table A4, presented in the online appendix, is hardly worth a look, as the macro-level regressions are essentially identical to those presented above in the body of the article. And the individual-level model coefficients are nearly identical too.Footnote ⁸ This result reflects our theoretical model of a two-step path of influence.Footnote ⁹ Instead, we present in Table 4 the calculations of the total indirect effects of the macro variables on SWD through the individual-level determinants. For ease of interpretation, we also give the estimate (in bold) of their maximum effect—that is, from the minimum to the maximum of the macro/institutional variable. This is the most movement in SWD we can expect from the “worst” to the “best” country on a given macro variable.Footnote ¹⁰ The table shows the most powerful macro influences at the top, descending to the non-influential factors—that is, ranked by absolute value.

Table 4.

Maximum Total Indirect Macro Effects

Note: Calculated from Stata SEM procedure's post-estimation total effects; country range of SWD is .38 (.36 to .74); estimates are from a non-multilevel model—see note 10.

The biggest influences on citizens’ judgments of their democracy are not political institutions, just as most recent SWD studies have shown. Even contingent political factors (number of parties, single-party government) are not terribly powerful. The top of the list is populated by socio-economic contextual factors (see also Dahlberg and Holmberg, Reference Dahlberg and Holmberg2014), topped by wealth and clean government (see also Quaranta and Martini, Reference Quaranta and Martini2016; Dassonneville and McAllister, Reference Dassonneville and McAllister2020; Nadeau et al., Reference Nadeau, Daoust and Arel-Bundock2020). For the countries in the data furthest apart on these measures, we expect a difference in satisfaction of nearly one-quarter of the range of satisfaction across countries. If we add the effect of economic inequality (GINI) and of government social expenditure, these four macro factors account for two-thirds of the difference in satisfaction with democracy between, say, Sweden and Lithuania (across only 300 km of the Baltic Sea!). Of course, these factors don't necessarily travel together, as we find Hungary or the Czech Republic relatively low on wealth and transparency but not among the countries high in income inequality. Notable also, as we saw above, is that the influence of these economic and output measures comes, sensibly, mostly through citizens’ feelings about their well-being and the quality of health and education.

The first contingent political factor to appear is the measure of government change through elections—government alternation. It has perhaps a counterintuitive total effect whereby countries that have changed governments in all of the last three elections are less satisfied with democracy than countries that have had less alternation. The obvious explanation is, in fact, that the causal arrow goes the other way, and government alternation is a result of feelings of dissatisfaction, as Ezrow and Xezonakis’ (Reference Ezrow and Xezonakis2016) findings strongly suggest. We note that the implied endogeneity mandates further model development in future research.

The political results of institutional differences have smaller effects but usually in the expected direction. The number of parties in government has a net positive effect on SWD and is even stronger given that the majority government variable is a special case of the number of parties in government, so the combined effect is that more parties in government lead to more satisfied citizens. Even this, the most important political factor, is in truth not hugely influential given its relatively small variation; this is partly because we saw that it has a positive effect through responsiveness and representation but a negative effect through ideological distance from the centre, illustrating the trade-offs inherent in many institutional alternatives.