1. Introduction
During the height of the Eurozone crisis, European citizens, particularly those in the bailed-out debtor countries, lost support for their political institutions, identified as satisfaction with democracy (SWD) and trust in political institutions. What's behind this collapse in political support is a fundamental debate in the political support literature: to what extent was it a function of the dire economic situation (Foster and Frieden, Reference Foster and Frieden2017; Devine, Reference Devine2021), or a more intrinsic loss of faith about the workings of the domestic, democratic system of governance (Armingeon and Guthmann, Reference Armingeon and Guthmann2014; Armingeon et al., Reference Armingeon, Guthmann and Weisstanner2016; Sánchez-Cuenca, Reference Sánchez-Cuenca2017)? This question speaks to the concern that national-level democracy is being hollowed out by the influence of actors beyond the domestic sphere, resulting in greater detachment from domestic, governing institutions (Scharpf, Reference Scharpf1999; Mair, Reference Mair2013; Laffan, Reference Laffan2014; Vowles and Xezonakis, Reference Vowles and Xezonakis2016). This was seen to be exemplified by the Eurozone crisis, where the economic interventions by the European Commission, International Monetary Fund (IMF) and European Central Bank (ECB), known collectively as the Troika, imposed a set of structural reform programs upon a number of European Union (EU) countries. This led to claims that democratic governance in Europe had been relegated to being a system of “democracy without choice”, a “democratic void”, or suffering from “democratic impotence” (Alonso, Reference Alonso2014; Sánchez-Cuenca, Reference Sánchez-Cuenca2017).
Whilst recent empirical contributions conclude that interventions resulted in a decline of political support, understood as both trust in political institutions and SWD, over and above the existing economic distress (Armingeon and Guthmann, Reference Armingeon and Guthmann2014; Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019), there are crucial missing gaps within this literature. First, the majority of analyses (though see Armingeon and Guthmann, Reference Armingeon and Guthmann2014) are conducted at the aggregate level, leaving “uncharted territory” at the individual level (Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019). The result is that whilst we have firmly established that interventions affect political support, we do not know whether this is a causal relationship, if it operates at the individual level, or through which mechanisms. Second and related, we do not have a convincing answer to whether the mechanism is one of economic distress, concerns over democratic legitimacy, or something else entirely. These questions address issues of democratic governance under interdependence and how people process such information.
In this paper, we address both of these gaps by triangulating analyses at the aggregate and individual-level to provide an empirical contribution to our understanding of the intervention-induced collapse in political support. First, we contribute novel individual-level causal evidence of the negative impact of economic interventions on political support and economic perceptions from a quasi-experiment in Portugal, in which a press release announcing an intervention (and the conditional reform measures required of the Portuguese state) occurred during survey fieldwork. Second, we consider the hypothesis that the link between interventions and political support is through subjective economic evaluations, what we call the economic cue thesis: news of intervention acts as an information shock, which catalyzes a learning process that updates individuals’ understanding of how the economy is performing and, more importantly, how the economy is likely to perform in the near future. We study this with a mediation analysis through a structural equation model on aggregate Eurobarometer (EB) data from all EU countries between 2005 and 2015 (293 country-years), which decomposes the overall effect of intervention into its direct and indirect components. This allows us to quantify both whether intervention influenced economic perceptions or political support and how much of the effect of interventions on the latter operates through economic evaluations. Due to there being no survey questions for the “reduced maneuverability” mechanism, we can only test this through “objective” measures such as national interest rates, and so our evidence on this mechanism is indirect.
The results of these two complementary studies show that interventions exhibited a negative causal impact on both political support and economic perceptions. The mediation analysis shows that the negative effect of intervention from the EU is partially mediated by updates in economic priors, supporting the economic cue thesis. Yet, approximately two-thirds of the effect of the intervention is through other mechanism(s). As such, whilst we are confident that the economic route explains a sizeable proportion of the collapse in political support, we attribute the rest to a combination of the “democratic impotence” mechanism, some unidentified causal path, as well as random error. Whilst the theoretical models point toward democratic impotence, the data available do not permit us to isolate this path. Still, our analysis allows us to provide a unique assessment of the extent to which interventions affected political support and how much is via the rational updating of economic priors. Finally, we show that objective economic conditions (such as unemployment change, GDP, and overall economic “misery” (Kriesi, Reference Kriesi2016)) only affect political support via their influence on economic perceptions. Overall, we provide robust and rigorous causal evidence for the effect of economic interventions on political support and quantify the extent and pathways through which they operate.
In the next section, we present the existing literature and our hypotheses. We then describe our data and methods and present our results, and conclude with a brief discussion.
2. EU intervention and political support
The intervention of the EU into the domestic economies of its member states began at the height of the Eurozone crisis in 2010 when, faced with insurmountable borrowing costs, the Greek government became the first country to sign a financial bailout package with the Troika. Memorandums of Understanding instigating processes of intervention and fiscal monitoring between the Troika were later signed by Ireland (2010), Portugal (2011), Spain (2012), and Cyprus (2013). Italy, although not a formal signatory of an agreement with the Troika, was also informally subjected to intervention by the ECB—as was Spain—in 2011. The then president of the ECB (Jean-Claude Trichet) clandestinely wrote to each country's head of government requiring a battery of fiscal reforms and internal devaluation measures to be implemented in order for the countries to qualify for future financial assistance.Footnote 1 Submitting their citizens to the conditions demanded by European creditors, the governments of the intervened-in member states are viewed as having abdicated their duty to act responsively to the policy preferences of their voters via the prioritization of economic responsibility to their member state peers (Alonso, Reference Alonso2014; Laffan, Reference Laffan2014; Turnbull-Dugarte, Reference Turnbull-Dugarte2020b).
The EU's decisions to intervene in member state economies has had a profound effect on party politics, electoral behavior, and public opinion, shaping our broad understanding of the workings of electoral democracy. It is argued to have shaped the “quality” and legitimacy of democracy in Europe (Morlino and Quaranta, Reference Morlino and Quaranta2016; Sánchez-Cuenca, Reference Sánchez-Cuenca2017). Issue salience, particularly on the dimension of EU integration (Katsanidou and Otjes, Reference Katsanidou and Otjes2015; Otjes and Katsanidou, Reference Otjes and Katsanidou2017; Turnbull-Dugarte, Reference Turnbull-Dugarte2020a), has been changed, party systems reconfigured and responsiveness reduced (Traber et al., Reference Traber, Giger and Häusermann2018).
Arguably the most significant effect, and certainly of most relevance to our argument, is that intervention transformed citizens’ attachments to their political system—such as SWD and trust in institutions, which we refer to as political support. Using a range of methods, data sources, and indicators, the literature has established that the Eurozone crisis shattered political support, particularly in Southern Europe, and through multiple mechanisms. Armingeon et al. (Reference Armingeon, Guthmann and Weisstanner2016) have highlighted, for instance, how citizens in debtor countries have become detached from their political systems, ultimately severing the ties of legitimacy between the governed and governors; in doing so, dividing the Union between debtor and creditor countries. Armingeon and Guthmann (Reference Armingeon and Guthmann2014) also report a notable collapse in both the level of SWD and trust in the national parliament amongst the bailout recipient states. Ruiz-Rufino and Alonso (Reference Ruiz-Rufino and Alonso2017) find similar evidence from a difference-in-difference estimation, which shows that the intervened-in countries experienced much lower levels of democratic satisfaction than creditor countries, which even increased in satisfaction (such as Germany). Schraff and Schimmelfennig (Reference Schraff and Schimmelfennig2019) provide evidence of the same negative effect, with variation in intensity depending on country, using a generalized synthetic control model; a negative effect which, however, fades over time as economic conditions improve. Whilst the vast majority of work finds a negative intervention effect, this conclusion is not uncontested, with some finding that there was a muted effect on support (Magalhães, Reference Magalhães2016), and intervention in fact bolstered support for democracy as a regime principle (Cordero and Simón, Reference Cordero and Simón2016). With respect to the interventions, then, the literature generates a clear hypothesis:
H1: EU economic intervention reduces satisfaction with democracy and political trust at the national level.
2.1 EU interventions, economic evaluations, and citizen updating
The main question that remains in the literature is what links external economic interventions with political support. This is an important omission as it tells us about the nature of political support and the role of governing institutions (domestic or otherwise) in fostering or eroding it. Devine (Reference Devine2021) outlines three pathways. Two fall under the reduced maneuverability thesis Footnote 2, which posits that intervention by a supranational or international institution ties the hands of domestic governments limiting their ability to offer distinct policy alternatives to their voters that diverge from the conditions demanded of them by their external creditors (Alonso, Reference Alonso2014). This reduced maneuverability leads to a situation in which voters are no longer able to choose the policy direction of the state resulting in a “democracy without choice”. It is this lack of choice that distinguishes the maneuverability-constraining nature of EU intervention, and the welfare state retrenchment packages required by the intervening institutions, from bread and butter austerity politics.
Throughout the years that followed the financial crisis a number of states in Europe, both within and outside of the Eurozone, adopted austerity as a standard policy response (Blyth, Reference Blyth2013). Austerity, and the democratic discontent that it led to (Huebscher et al., Reference Huebscher, Sattler and Wagner2021), was not unique to the intervened-in countries. What was unique to the intervened-in states, however, was the democratic legitimacy of reforms. Whilst British, German, and Dutch citizens, to give illustrative examples, were subjected to austerity measures that were self-imposed by their governments, the citizens of the intervened-in states were subjected to these policies primarily because of external pressure (Javier, Reference Javier2013; Ruiz-Rufino and Alonso, Reference Ruiz-Rufino and Alonso2017). A lack of choice undermines citizens’ faith in representative democratic institutions which leads to a reduction in levels of political support and SWD (Ruiz-Rufino and Alonso, Reference Ruiz-Rufino and Alonso2017), political trust (Torcal, Reference Torcal2014), and a reduction in participation with the political and electoral process (Turnbull-Dugarte, Reference Turnbull-Dugarte2020b). This can be either through citizens directly perceiving this constraint on maneuverability or only the lack of responsiveness of partiesFootnote 3 and the political class.
On the other hand, there is the subjective economic cue thesis (Foster and Frieden, Reference Foster and Frieden2017; Devine, Reference Devine2021). This theory maintains that economic intervention by external actors provides a cue to citizens that engenders information acquisition and learning regarding both the state of the economy as well as the competence of those in power to effectively manage the country's economic problems. When individuals observe their country in need of emergency financial support from supranational actors, they are cued to update and revise their (negative) economic evaluations and expectations. When economic times are hard, citizens tend to be less satisfied with democracy and less trusting, with the reverse being true when times are good (Friedrichsen and Zahn, Reference Friedrichsen and Zahn2014; Foster and Frieden, Reference Foster and Frieden2017; Christmann, Reference Christmann2018). Moreover, theories of economic voting assume that citizens’ electoral and political choices are shaped by their evaluations of economic performance (Duch and Stevenson, Reference Duch and Stevenson2008) and the same has been observed in the case of political support, with negative economic conditions correlating with lower levels of trust and SWD (Quaranta and Martini, Reference Quaranta and Martini2017; Christmann, Reference Christmann2018).
Whilst both of the theoretical mechanisms linking intervention and political support assume political learning, they diverge on what priors are being updated by the information shock. The reduced maneuverability thesis posits that it is the priors regarding the constraining nature of external actors which are updated, whilst the economic cue thesis argues that it is subjective economic evaluations which are updated. We test the hypothesis that economic interventions update economic evaluations in both quasi-experimental and observational settings. If we do not find that interventions update economic evaluations, this suggests either that citizens do not update their views or that the link between interventions and political support is not through economic evaluations:
H2: EU economic intervention leads to worse economic evaluations
We can also extend this logic further. Not only would we expect that interventions affect economic evaluations, we would also expect that at least some of the effect of interventions operates through their effect on evaluations; in other words, the effect is mediated. This is an important area of contention within the extant literature on the effects of external intervention. Whilst Ruiz-Rufino and Alonso (Reference Ruiz-Rufino and Alonso2017) posits that the negative correlation between intervention and political support is the product of citizens’ realization that external actors are constraining the responsiveness of domestic institutions to their preferences, Devine (Reference Devine2021) and Schraff and Schimmelfennig (Reference Schraff and Schimmelfennig2019) maintain that political support is determined by the negative economic expectations. We thus directly test whether the effect of the intervention on political support is mediated by economic evaluations. In doing so, we are able to isolate how much of the effect of the intervention is transmitted via economic perceptions.Footnote 4
H3: Negative effect of EU intervention on political support is mediated by updates in economic evaluations (economic cue thesis)
3. Data and methods
Empirically, we employ a two-stage analytical approach using EB data. First, to test H1 and H2 we use an unexpected event design (Muñoz et al., Reference Muñoz, Falcó-Gimeno and Hernández2020) to provide quasi-experimental evidence of the causal effect of intervention on both political support and economic evaluations at the individual level. A press release regarding an intervention was announced during EB survey fieldwork in Portugal, providing us with an experimental setting where some respondents were interviewed before the press release and some were interviewed after. The data therefore provides the ideal setting to examine the effect of the as good as random exposure to the announcement regarding the EU's economic intervention on both economic perceptions and political trust at the individual level. Whilst the possibility of intervention existed before the confirmation announcement, the announcement still operated as an informational shock and we test extensively for pre-treatment trends in our dependent variables to show that the announcement led to a stark and significant discontinuity (Figures A2, A3, and A7). We also reject related issues such as non-random sampling (Figure A4).
Our second stage of the analysis turns to the full EB data in order to examine H3. The data comprise 293 country-year observations for all 28 EU member states between 2005 and 2015.Footnote 5 Using this aggregate-level data, we provide a mediation analysis using a generalized structural equation model which isolates the effect of EU intervention on SWD and the potential mediating role of economic perceptions. Specifically, we test whether EU intervention exhibits a direct effect on SWD, or if the relationship is mediated, with intervention serving as an information shock that leads to an updating of the public's assessment of the state of the economy. The following sections detail the specific operationalization and estimation techniques of each empirical test alongside the main results.
4. Evidence
Study 1: quasi-experimental analysis
For the quasi-experimental study, we rely on the EB 75.3 in May 2011, which occurred during the period in which Portugal requested and was granted a financial bailout package with strict conditionality from the EU institutions. Whilst Portugal requested a rescue loan on the 7th April and provided greater detail on the 5th of May, this was approved and given the go-ahead by the Eurogroup and ECOFIN ministers on the 16th of May.Footnote 6 The announcement entailed greater monitoring of budget controls, reforms of the health, housing and service systems, and a catalog of “ambitious” privatizations. This was, of course, reported in the Portuguese press, which we provide examples of in the Appendix. Essentially, those with the potential to be exposed to the announcement were being informed of the Troika-solicited reforms and conditions attached to Portugal's bailout request.
The EB fieldwork in Portugal took place between 7th May and 22nd May, providing a sample both before and after the press release. As we do not have a measure of compliance (whether individual citizens actually became aware of the news), our core estimand is the intent-to-treat (ITT) effect.Footnote 7 Given the exogenous nature of treatment allocation (exposure to the news regarding intervention and the conditions attached), we separate the sample into those interviewed before the press release (control) and those interviewed on or after the announcement was made (treatment). The control group comprises 77 percent of the sample (N = 803) whilst the remaining 23 percent (N = 245) of respondents are assigned to treatment. We estimate the effect of exposure to the news regarding intervention whilst controlling for basic pre-treatment covariates: education, age, occupation, and gender.Footnote 8 Intervention is operationalized via a simple exposure binary indicator from the 17th of May.
Whist our quasi-experimental design relies on data from only one of the five intervened-in countries, Portugal is case which is likely to be exeternally valid. Portugal represents neither an extreme example of intervention nor a contested case. The interventionist actions taken by the Troika in Greece can be viewed as an extreme example of intervention given the repeated need of external actors to provide financial rescue packages—each laced with their own conditionalities—on numerous occasions over a significant period of time. These repeated interventionist actions were not observed in the other intervened-in states. At the same time, there is consensus over Portugal's status as an intervened-in economy. Despite Spain signing a Memorandum of Understanding with the EU's financial institutions in 2012, there is domestic contention over whether or not this financial transfer equates to a “bailout” or not (Magone, Reference Magone2016). The same is true of Italy: whilst some studies include the actions of the ECB's communication with Italy as tantamount to intervention (Ruiz-Rufino and Alonso, Reference Ruiz-Rufino and Alonso2017) others do not (Devine, Reference Devine2021; Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019). As such Portugal represents a “perfect pupil” (Magone, Reference Magone2014) example of intervention. It is therefore a case where the theorized effects should be observed.
In addition, whilst there may be questions over how exogenous this treatment was,Footnote 9 we assume that whilst the potential for being bailed out existed before, the date and exact content of the bailout program would not have been known before the 16th May. Moreover, whilst prior to the intervention announcement citizens may have held pessimistic assessments of the current economic situation, our theoretical expectations are primarily concerned with prospective evaluations where intervention cues citizens that Troika-solicited austerity measures will shape economic conditions in the near future as well as update perceptions of the current economic situation. Empirically, we also observe an increasing trend in political support and economic expectations, and no trend in economic evaluations, during the pre-treatment period (presented in Figure A3). This makes our findings of a negative effect even more significant. The data, therefore, confirm our expectation that the results are not part of an ongoing trend during the fieldwork; if anything, the treatment reversed an otherwise positive trend.
To answer H1 and H2, we use three outcome variables. We use the dichotomous measure of whether the respondent “tends to trust” (1) or “tends not to trust” (0) the country's national parliament for H1. We unfortunately cannot also test for democratic satisfaction, as this is not included in the dataset. For H2, we use two indicators of economic perceptions. First, we use one of “economic evaluations”, which indicates how the respondent thinks the national economy is doing, on a four-point scale from “very good” to “very bad”. Second, we use one of economic expectations (or prospective evaluations) which asks how the economy is expected to do in the next 12 months, on a three-point scale indicating “better”, “worse”, or “the same”.Footnote 10 All of these retain the original coding in the survey; however, we reverse the coding of economic evaluations from the raw data so that the responses run from negative to positive evaluations and recode the economic expectations so that “the same” is the middle category. This is only to make the interpretation of coefficients more intuitive. Full question wordings and descriptive statistics are provided in the Appendix.
Each of our model specifications in our quasi-experimental test relies on linear regression, the results of which are presented in Figure 1. The results show that the intervention had a significant impact on trust in parliament and economic evaluations at the 95 percent level and on economic expectations at the 99 percent level. For trust, this amounts to a 6.8 percentage-point decrease in the probability of trusting parliament. Considering the mean is 27 percent, this is a substantively large change relative to the overall probability (− 25 percent relative to the sample mean). For economic evaluations, the expected change is − 0.11 on a 1–4 scale (2.8 percent of the total, 19 percent of a standard deviation); and for economic expectations, the expected change is − 0.17 on a 1–3 scale (5.6, 27 percent of a standard deviation).
Fig. 1. Marginal effect of intervention on trust and economic views.
This is how the average citizens’ economic perceptions and levels of political support have been affected by the intervention shock. It is of course possible that this effect is moderated by variables such as partisanship. That said, we expect trust in parliament (as opposed to trust in government) to be less influenced by partisanship. In the case of Portugal, the mainstream left and right parties were both involved in the process of EU intervention. The Sócrates-led Socialist government engaged in initial talks with the Troika institutions in May 2011 and the EU-demanded reforms were adopted via the approval of the Coelho-led coalition government of center-right parties. Importantly, the socialists abstained in the vote with only the hard-left voting against. Effectively, all mainstream parties were complicit in the introduction of EU intervention (Fernandes et al., Reference Fernandes, Magalhães and Santana-Pereira2018). It is for this reason that we also present the null effect of intervention on trust in government in Appendix Table A5; whilst this does caveat our results, we are cautious given the high chance of partisan reasoning and our inability to test or control for that possibility. Whilst we are unable to test for the potential moderating effect of partisanship due to the absence of any relevant variables in the EB dataset, Okolikj and Hooghe (Reference Okolikj and Hooghe2020) show that during the same time period in Europe, partisanship exhibited only a very minor effect in biasing individual's economic evaluations with government-supporting and opposition-supporting voters changing their perceptions of economic performance largely symmetrically. It is also important to note that the average effect on the population as a whole is substantive.
This caveat aside, these results indicate that intervention had an immediate and sizable impact on trust in parliament, evaluations of the current state, and expectations of the future performance of the economy. The effect is twice as large on economic expectations as evaluations, indicating that of relatively more importance is the updating of how the economy is expected to perform rather than how bad it currently is. This suggests that the updating of economic priors is not necessarily about the current state of the economy—since most will likely have more important and direct sources of information, such as income or employment—but rather that current conditions are not likely to improve any time soon. These results support H1 and H2, showing that the randomly allocated exposure to news regarding intervention has a negative effect on both political support and economic perceptions.
We provide robustness and sensitivity tests for the assumptions regarding, amongst other concerns, excludability, ignorability, and spurious time trends (Muñoz et al., Reference Muñoz, Falcó-Gimeno and Hernández2020). We cater to each of these concerns in turn. First, we reduce the bandwidth to three days either side of the intervention to rule out time trends and the influence of hard-to-reach groups which are surveyed at the end of fieldwork. Second, we construct a number of placebo dates, such as the median date of the control group (including and excluding the treated individuals) and the 5th and 6th days of fieldwork. We complement this with randomization inference by conducting a permutation test, which reproduces the analysis by randomly assigning the “treatment” date to any of the dates in the survey: from 1000 iterations, the placebo effects are normally distributed around zero. Third, we change the unit of analysis to all 28 then-member states and show that there are no significant negative effects in any other countries.Footnote 11 Fourth, the treatment date is only significant for our outcome of interest. Finally, we implement nearest-neighbor matching with a range of specifications. Whilst these results at times lose significance due to the reduced sample size, the point-estimates are almost always identical to our reported result. The full output and discussions of these tests are reported in the Appendix.
The results of our quasi-experimental analysis are important for establishing the plausibility of a causal effect of the interventions on support and economic perceptions, an issue which has previously been unstudied at the individual level (Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019). However, we improve the generalizability of this precise causal estimation in the next section by broadening our model to all countries in the EU and isolating the mechanism through which the interventions operated.
Study 2: mediation analysis
The section that follows applies generalized simultaneous equation modeling (GSEM) in order to carry out a mediation analysis. Relying on the aggregate level EB data from all EU countries between 2005 and 2015 (see Table A8 for survey waves included in analysis), we model simultaneous equations that estimate the impact of intervention on SWD. SWD is measured via a four-point scale indicating those who are “Not at all satisfied” (1) to “Very satisfied” (4). The mediating variable of interest, subjective economic perceptions, is measured in a similar four-point fashion, indicating those who believe the economic situation in their country is “Very bad” (1) to “Very good” (4).
The primary explanatory variable capturing intervention is a dichotomous indicator for the country-years in which a member state was intervened. This operationalization replicates that prevalent in the literature (Ruiz-Rufino and Alonso, Reference Ruiz-Rufino and Alonso2017; Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019; Turnbull-Dugarte, Reference Turnbull-Dugarte2020b) and includes Greece and Ireland post-2010, Portugal post-2011, Spain post-2012, and Cyprus post-2013.Footnote 12 We do not include Italy in the main intervention model since it was not a formal signatory of a Memorandum of Understanding. Including Italy (post-2011)—and other states in receipt of economic bailout packages before the Eurozone crisis—in the intervention group does not alter the main findings.
As in the case of the quasi-experimental design, we present our mediation analysis via the potential outcomes framework (Imai et al., Reference Imai, Keele and Yamamoto2010b, Reference Imai, Keele, Tingley and Yamamoto2011). The purpose of the mediation analysis is to assess how much of the treatment effect of intervention (T) on our potential outcome, SWD (Y), is caused by updates in perceptions of economic performance (M). We estimate three parameters: the average total effect (ATE), the average mediated effect (AME), and the average direct effect (ADE) of the EU's economic intervention on political support. The ATE represents all of the variation in SWD that can be explained by EU intervention and is equal to the sum of the ADE and AME (Imai et al., Reference Imai, Keele and Tingley2010a, 312; Imai et al., Reference Imai, Keele, Tingley and Yamamoto2011, 769).
The total effect can be defined as TE = SWDc,t(T(1))−SWDc,t(T(0)), where T indicates that a country(c) was exposed to EU intervention at time (t). Obviously for a single data point both T(1) and T(0) cannot be observed so the ATE relies on the average difference in means between units exposed to treatment (intervention) and those not.
We are interested in deconstructing the ATE into its distinct components: AME + ADE. The AME is indicative of the impact of EU intervention that is mediated by changes in perceived economic conditions. Since economic evaluations (M) can be updated by exposure to intervention, there will be two potential values: M c,t(1) when economic perceptions are exposed to treatment (T=1) and M c,t (0) when economic intervention is absent (T=0). The ME measures the variation in SWD caused by alterations in the value of the mediating variable if the value of treatment was held constant. SWDc,t(T, M(1)) indicates the outcome variable when exposed to intervention T(1), whilst SWD c,t(T, M(0)) indicates the outcome variable when T(1) but the value of the mediating variable has the M value when T(0). The AME indicates the variation in SWD caused by a unit change in economic perceptions that was brought about by intervention whilst also considering the direct effect of intervention on SWD.Footnote 13 If intervention exhibits no effect on subjective economic evaluations (i.e., M(1) = M(0)), the mediation effect is equal to zero (Imai et al., Reference Imai, Keele and Tingley2010a, 311) and this would indicate that intervention does not operate via the economic cues thesis but rather other mechanisms. Formally, the mediation effect can be summarized as ME = SWD c,t(T, M(1))−SWD c,t(T, M(0)).
Finally, the ADE represents the variation in SWD that can be explained by all other alternative channels that link EU intervention to political support, such as the reduced maneuverability or “democracy without choice” thesis (Alonso, Reference Alonso2014). The direct effect is formalized as: DE = SWD c,t(T(1), M(T))−SWD c,t(T(0), M(T)). This indicates the impact of intervention on SWD whilst holding the effect of economic perceptions constant. The ADE indicates the average effect of intervention on SWD that is not explained via mediation through subjective evaluations of the economy.
Given the variables available in the data, we are only able to provide a concrete test of the mediation caused by the economic updating hypothesis. The remaining influence of intervention once the mediating effect of subjective economic performance is considered, the ADE, captures all remaining avenues of influence. Given the theorized mechanisms, we can assume that the ADE is capturing the influence of the reduced maneuverability thesis as well as other potentially unaccounted for avenues. The model controls for objective economic performance, country fixed effects, and time trends.Footnote 14
Figure 2 displays the key coefficients of the models of our two equations as a causal path diagram. The full output is reported in the Appendix (see Table A17). The coefficients displayed in the figure illustrate the effect of EU intervention and changes in unemployment on both economic evaluations and SWD. The unemployment rate captures the country-level (percent) change in unemployment between time t and the year prior (t − 1). Unemployment is taken as the main indicator of objective economic conditions alongside intervention although other measures of macroeconomic performance are also considered (see Figure 4). We take this to be the strictest test, since unemployment is the indicator most closely related to the lived economic experiences of individual citizens (unlike, for instance, changes in GDP). The model displays a significant direct effect of intervention on economic evaluations and SWD. In the case of the macroeconomic control, however, the results show that the changes in unemployment effect economic evaluations but not SWD. This suggests that any effect of unemployment on our core-dependent variable is mediated through its effect on economic evaluations. To demonstrate this, we turn to the results of the mediation analysis presented in Figure 3. The illustration shows the ATE, AME, and the ADE of exposure to EU intervention and a one percentage-point increase in the unemployment rate on SWD.
Fig. 2. Causal path diagram of the generalized structural equation models.
Fig. 3. Direct, indirect, and total effect on SWD.
Fig. 4. Comparing mediation effects.
As expected, the total effect of both intervention and unemployment is significant and negative. Exposure to intervention is associated with a sizeable and significant decrease in SWD as too is an increase in unemployment. Compare the effects of intervention to that of changes in unemployment. First, there is a clear difference in magnitude. Whilst the ATE of both intervention and changes in unemployment exhibit a significant effect on SWD, the substantive effect of intervention is notably greater. The level of SWD in countries undergoing intervention is, on average, 0.44 lower. The average value of SWD (scaled 1–4) in the sample is 2.46 so a 0.44 change is a notable effect size (11 percent of the total, or 113 percent of a standard deviation). A one percentage-point increase in unemployment is associated with a .026 decrease in SWD. As an illustration, Greece experienced an increase in the unemployment rate between 2011 and 2012 of close to seven percentage-points. Taking the ATE of unemployment, this seven percentage-point change would translate into a 0.18 decrease in SWD which equates to 4.6 percent of the total or 46.67 percent of a standard deviation.
Beyond the effect of intervention and taking changes in unemployment as an indicator of macroeconomic performance, it is clear that the relationship between unemployment and SWD is entirely mediated by updates in citizens’ perception of the state of the economy. Recall that the AME indicates the internal value of the total effect that is explained via updates in economic perceptions. The AME of unemployment is significant and negative, whereas the ADE (that which captures the direct effect of changes in unemployment) is indistinguishable from zero. In other words, there is no other avenue for the effect of objective economic performance (unemployment) on SWD once the impact such changes have on economic perceptions is considered. In the case of EU intervention on the other hand, there is a clear direct effect on SWD. Even when considering the impact that intervention has on political support as a result of changes in subjective assessments of the economy, EU intervention leads (directly or through other mechanisms) to a sizeable decline in SWD.
Importantly, the mediated effect of changes in the unemployment rate are echoed across alternative operationalizations of objective economic performance. Figure 4 depicts the ATE, AME, and ADE of changes in the budget deficit, public debt, or government borrowing rates, which are used as substitute indicators in the place of changes in unemployment in a step-wise fashion. In terms of their effect on political support once taking into account the impact of intervention, only changes in the level of unemployment and public debt have any non-zero effect on SWD, whereas changes in the budget deficit or government borrowing rates don't have any impact. The null effect in the case of the cost of borrowing is of note given that the fiscal flexibility of the state is often constrained by the accessibility of the state purse to access liquidity on the financial markets. Whilst interest rates might be a measure of objective maneuverability, the lack of any effect on SWD suggests that any correlation between state access to capital on the financial markets is not an economic signal that citizens rely on to update their prior understanding of the state of politics in the country.Footnote 15
As in the case of the unemployment rate, the significant negative effect of government debt on political support appears to be entirely mediated by updates in citizens’ perception of the economy. The ADE is indistinguishable from zero once the impact of debt on subjective evaluations is considered. This is largely consistent with Becher and Donnelly (Reference Becher and Donnelly2013), who show that objective economic conditions affect voting behavior via the mediated impact they have on citizens’ subjective evaluation of the state of the economy. The results we present here suggest that the role of economic conditions in shaping citizens’ level of SWD operates in a similar fashion. This reconciles existing work which debates whether support is driven by perceptions or conditions; we argue both, but that economic conditions work through perceptions rather than via a direct effect. Once the indirect effects of each of our macroeconomic variables are considered, none of them exhibit an independent direct effect on SWD. This is notably different from the effect of economic intervention. Across all of the model specifications using different alternatives to operationalize objective economic malaise, the independent ADE of EU intervention retains its significant negative effect. As a result, we are confident that intervention exhibits an important direct effect on political support beyond that which is mediated by subjective evaluations regarding the state of the economy.
To ensure the robustness of our findings related to intervention, we carry out three additional sensitivity tests. First, we estimate a jackknife test removing each intervened-in country from the sample in a step-wise manner (Figure A12). Our results from this test are twofold. First, we observe that the negative effects of intervention from a single country are not influencing the overall results. Second, this provides greater external validity to our quasi-experimental test in Portugal: the effects are similar in all scenarios. Second, we re-estimate our main model to include a control indicating the country-years during which a country held a legislative or presidential election. Elections, as windows of opportunity for voters to exercise political choice and trigger alternations in government, can increase political support. The country-years in which elections took place could therefore have impacted the relationship between EU intervention and SWD. Including this in our model, however, shows that this is not the case, with our findings remaining unchanged. Third, we replicate the mediation analysis on the individual-level data from Portugal (Figure A20). As in the case of our cross-national analysis, we observe that around one-third of intervention's effect on political trust is mediated via economic evaluations, whereas the remaining effect of intervention operates via alternative mechanisms, amongst which, we can assume are concerns regarding state maneuverability.
In real terms, our results provide mixed evidence in support of H3. On the one hand, we find that a noteworthy proportion of intervention's effect on political support is mediated by economic evaluations. This suggests that intervention serves, as we hypothesized and in line with the economic cue thesis, as an information shock that updates citizens’ priors regarding the current and future performance of the economy. On the other hand, however, and unlike the objective performance indicators, a large(r) part of the variation in political support explained by intervention occurs because of mechanisms other than the economic cue thesis. Unfortunately, it is not possible to test these other mechanisms due to data limitations. We expand on this in the following section.
5. Discussion
Our contributions are threefold. First, as expected and in line with H1, we provide causal evidence that exposure to economic intervention leads to a negative effect on political support. Taking advantage of the quasi-experimental nature of information shocks regarding the EU's intervention in Portugal, we find an average treatment effect of seven percentage-points. Second, we similarly show that interventions also had a negative causal impact on economic evaluations. Citizens who were informedFootnote 16 of the press release regarding the EU's pending intervention in the country were between 17 and 33 percent of a standard deviation more negative about the economy, both how it is currently performing or how it is expected to perform in the future. The effect on prospective evaluations was found to be approximately twice as large as current evaluations, suggesting that the intervention signaled that the bad times were here to stay rather than coloring perceptions of present-day economic conditions. These findings are consistent with the existing literature (Ruiz-Rufino and Alonso, Reference Ruiz-Rufino and Alonso2017; Devine, Reference Devine2021), yet contributes the closest approach to causality so far and fills the “uncharted territory” at the individual level (Schraff and Schimmelfennig, Reference Schraff and Schimmelfennig2019).
Third, we provide nuance to the debate about the political or economic—or input versus output—determinants of political support by showing that the negative consequences of the EU's fiscal monitoring programs on political support cannot be entirely accounted for by updates in economic evaluations (Scharpf, Reference Scharpf1999; Foster and Frieden, Reference Foster and Frieden2017). Relying on the aggregate data of the cumulative waves of the EB, we find that the link between objective economic conditions and political support is entirely mediated via updates in individuals’ subjective view of the economy, but that only a third of the effect of the Eurozone's bailout packages is. This suggests that whilst the overall effect of economic conditions is large, it cannot explain the effect of interventions on its own. Rather, the effect is a combination of this and also other factors that we cannot test here. Given one theorized link between economic intervention and political support is the room to maneuver thesis, we would expect this to matter as well. Whilst we test for this using aggregate indicators such as debt and interest rates, which are commonly used in the literature, and show null effects, we are cautious of interpreting this too much given the lack of individual-level data to test causality and to separate it from general economic conditions. However, there are also other potential mechanisms: did the intervention update citizens’ views about corruption or clientilism, for instance? Of course, there will be some amount of the variation that is just noise and error.
These findings have significant contributions beyond this specific context. Our finding that the effect of economic conditions on political support is entirely mediated by economic evaluations reconciles the literature in political support which has wrestled with whether economic conditions matter, with some arguing that it is almost the only factor that matters (e.g., Stimson, Reference Stimson2004; Quaranta and Martini, Reference Quaranta and Martini2016; Foster and Frieden, Reference Foster and Frieden2017), whether it is about perceptions (Armingeon and Guthmann, Reference Armingeon and Guthmann2014), and how these may be moderated by institutional context. Our results support the notion, emerging within the support literature, that conditions are only relevant through their effect on economic perceptions (see also Christmann, Reference Christmann2018). It is worth noting that both our study and Christmann (Reference Christmann2018) arrive at the same conclusion with different datasets and analytical techniques. Whilst we don't test institutional effects, this suggests that institutions are relevant at least in part through facilitating the (accurate) linking of conditions to evaluations, in much the same way that institutions link economic conditions to voting behavior (Powell, Reference Powell2000; Kayser, Reference Kayser2007).
Political support is often theorized as the outcome of input and output factors (Scharpf, Reference Scharpf1999). The former refers to rule by the people, and the latter to state or government effectiveness (rule for the people). The bulk of the literature spends considerable time reasoning between these factors and how they condition each other (Dahlberg et al., Reference Dahlberg, Linde and Holmberg2015; Magalhães, Reference Magalhães2016). The economic interventions in Europe during the Eurozone crisis present an ideal setting to study this, given the significant institutional and economic transformations. What our results show is that researchers would be remiss to dismiss support for institutions as a function of (economic) outputs. Whilst we can't quantify the importance of input factors, we have shown that economic factors cannot account for all, or even most, of political support.
Finally, the interventions are a case of external economic shocks. Research has studied the effect of these for mass politics, and more broadly how interdependence shapes individual attitudes and behavior (Kayser and Peress, Reference Kayser and Peress2012; Arel-Bundock et al., Reference Arel-Bundock, Blais and Dassonneville2019; Ahlquist et al., Reference Ahlquist, Copelovitch and Walter2020). We have provided evidence that external economic shocks do indeed have significant repercussions; alongside the implications for voting behavior and concomitant rise of challenger parties and reshaping of party politics we again find that it also matters at the more diffuse level of institutional support. Moreover, this implies that citizens do take cues from externalities, and are not bound to domestic factors.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2021.71. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/4XXD3E
