Introduction
The popularity of welfare state expansion and retrenchment is a much-discussed topic. Pierson (Reference Pierson1996) argued that parties could claim credit for the popular expansion of the welfare state in the post-war period, but that since the 1970s structural pressures on government finances have pushed governments to retrench. Because so many citizens had become dependent on the welfare state, this was considered unpopular and today the new politics of the welfare state are all about avoiding blame. Nevertheless, parties of the left and right have implemented austerity (Hübscher, Reference Hübscher2016) and some even argue that parties of various ideological backgrounds have converged on their economic policy agenda in the direction of retrenchment (Bremer and McDaniel, Reference Bremer and McDaniel2020; Ezrow and Hellwig, Reference Ezrow and Hellwig2014; Lynch, Reference Lynch2019). Pierson (Reference Pierson1996) called this an era of ‘permanent austerity’.
Of course, there are exceptions and varying intensities across countries and years. There have been notable episodes of ‘emergency Keynesianism’, where governments stepped in with massive support for the financial sector and stimulus packages (Hall, Reference Hall, Miles and Lake2013), and the extent to which there is consensus across political parties on ‘permanent austerity’ has varied. However, if indeed there are cases in which parties collectively converge on an austerity consensus, how does this impact its popularity among the public? There are plausible reasons to believe that voters oppose austerity, and, moreover, that some voters object to retrenchment more than others: some austerity-induced cuts to welfare spending have had highly concentrated and unequally distributed costs (Fetzer, Reference Fetzer2019). However, preferences are not only formed by one’s material position. From models of mass preference formation, we learn that the discursive context matters. If there are permanent one-sided cues about a policy, which the permanent austerity thesis implies, this impacts public opinion (Zaller, Reference Zaller1992). Combining variation in voters’ aversion to austerity with the role of the discursive context, the question then becomes whether political parties can ‘make austerity popular’ (Barnes and Hicks, Reference Barnes and Hicks2018)?
Disagreement over austerity’s popularity has generated a distinct line of research in which we can roughly distinguish two camps. On the one hand, there is a strand in the literature that does not find that governments that implement austerity are systematically punished for doing so. Expenditure cuts are not associated with increased more frequent leader turnover over the past 140 years (Arias and Stasavage, Reference Arias and Stasavage2019). Alesina et al. (Reference Alesina, Perotti and Tavares1998) show that governments that implemented fiscal austerity do not tend to be replaced in office, neither do they lose popularity. In fact, governments going ‘cold turkey’ and opting for spending cuts (as opposed to tax increases) may be rewarded. Indeed, Giger and Nelson (Reference Giger and Nelson2011) even show that some political parties can ‘claim credit’ for cutting social policy. On the other hand, much in line with Pierson (Reference Pierson1996)’s thesis that the politics of retrenchment pose extensive electoral risks to governments, scholars argue that austerity has caused many unpopular outcomes such as ‘class politics, riots, political instability, more rather than less debt, assassinations, and war’ (Blyth, Reference Blyth2013, p. 229). The absence of electoral punishment is explained, for example, by the strategic planning of austerity packages (Hübscher et al., Reference Hübscher, Sattler and Wagner2021; König and Wenzelburger, Reference König and Wenzelburger2017). Instead, when considering government approval rates, which can be measured more directly after fiscal consolidation, citizens indeed disapprove of austerity (Jacques and Haffert, Reference Jacques and Haffert2021). And despite its strategic planning, austerity is still correlated to political polarization and political instability (Hübscher, Sattler, and Wagner, Reference Hübscher, Sattler and Wagner2023).
This paper contributes to this research agenda by bringing in two literatures that have not been extensively included in this discussion. First, is the study of heterogeneous fiscal preferences by income groups (Page et al., Reference Page, Bartels and Seawright2013), and second, the study of mass preference formation (Zaller, Reference Zaller1992). The study of austerity’s popularity has not yet taken into account enough that fiscal preferences might be systematically structured by material self-interest and hence that some voters are more or less supportive of fiscal retrenchment. Those towards the bottom of the income distribution are typically more dependent on government services and might therefore be naturally further away from supporting austerity. Conversely, if parties support austerity, higher-income voters might already be closer to this position. Second, by including party positions on austerity we can see whether popularity is systematically shaped by the discursive political context underlying fiscal policy. This context is shaped by the cross-party consensus and under consensus there are no alternative party positions voters can select.
I argue that those that are the furthest away from the austerity consensus adapt their preferences the most. This argument crucially assumes parties to be first movers. Instead of thinking of political parties as simply aggregating voter preferences (Key and Heard, Reference Key and Heard1949; Schattschneider, Reference Schattschneider1960), parties are conceptualized as active mobilizers of support for the fiscal strategy of a country (Baccaro et al., Reference Baccaro, Blyth and Pontusson2022). This shifts the focus from the influence of voters on parties to the influence of parties on voters (or the adaptation of citizens to the positions of parties).
Empirically, I test these expectations by matching citizen-level data on fiscal preferences from the International Social Survey Program (ISSP) with data on party positions toward economic orthodoxy measured by the Manifesto Data Project (MPD). Based on a sample of 60 country years, I argue that because low- and middle-income voters are naturally furthest away from supporting austerity based on their socioeconomic position, these groups adapt their preferences most strongly to austerity consensus. I argue that while austerity may be unpopular from a material self-interested position, cross-party consensus on austerity can make austerity more popular.
In the following, I situate my research in the literature on mass preference formation. I proceed to discuss the empirical data and research methods. Then, I discuss the empirical results pertaining to my main hypotheses, concluding with a discussion of the results and an attempt to situate my findings with previous research.
Theory and hypotheses
For its main hypotheses, the theoretical framework draws on two prominent strands within the literature on mass (fiscal) preference formation. First, one line of research argues that opinion formation is mainly driven by the discursive context provided by elites. Second, another line of research argues that material self-interest drives citizens’ attitudes. Finally, combining the two, the argument that elite-driven preference formation is conditioned by material self-interest will be explored.
First, the elite-driven argument is rooted in the notion that responsiveness between citizens and elites goes both ways. Political leaders do not ‘merely respond to the preferences of constituents; leaders also shape preferences’ (Dahl, Reference Dahl1961, p. 164). Indeed, in classic studies of power it is noted that ‘A may exercise power over B by getting him to do what he does not want to do, but he also exercises power over him by influencing, shaping or determining his very wants’ (Lukes, Reference Lukes1974, p. 27). There exist various mechanisms that suggest that preference formation is supply-side-driven. One prominent line of research suggests that this entails a close link between voters and their preferred parties (Campbell et al., Reference Campbell, Converse, Miller and Stokes1980; Lenz, Reference Lenz2013). Parties actively ‘mobilize support’ and ‘secure consent’ (Disch, Reference Disch2011), for example by shaping a narrative that is compatible with the macroeconomic agenda of a country (Hopkin and Voss, Reference Hopkin, Voss, Baccaro, Blyth and Pontusson2022, pp. 381–382). Parties actively explain their actions to voters and translate policies into stories (Esaiasson and Wlezien, Reference Esaiasson and Wlezien2017, p. 702).
The prominent counterargument to the idea of a direct link between individual voters and parties is that citizens might opt for different parties initially aligned with their fiscal preferences. Rather than adjusting their policy preferences within the same party, voters may switch to new parties while maintaining the same policy stance. Addressing the risk of endogeneity is difficult to deal with, especially as the data utilized in this paper’s empirical analysis is not well-suited to parse between selection and cueing effects.
To circumvent the issue of selection, this paper shifts focus from individual party positions to the cross-party consensus on fiscal stance within a given country year. Since voters cannot select a different consensus, any observed shifts in policy preferences cannot be attributed to party switching but must be explained by position switching. In addition to the practical advantages of studying cross-party consensus over the individual party-voter link, there are theoretical reasons to believe that consensus plays a significant role in driving mass opinion change.
Linking elite consensus to mass preference formation requires exploring the role of the discursive context within a national political economy. The premise underlying this inquiry is that what is said in the political sphere is consequential for political outcomes (Hopkin and Rosamond, Reference Hopkin and Rosamond2018). Party elites have the capacity to define the ‘realm of what is politically possible, feasible, and desirable’ (Hay, Reference Hay2004, p. 505). In the following, I will delve into how political parties influence the framing of debates and how citizens might adapt their preferences accordingly.
Defining the limits of what is politically feasible is different from directly shaping voters’ preferences. Indeed, while A could exercise power by trying to change B’s preferences, ‘the most effective and insidious use of power is to prevent such conflict from arising in the first place’ (Lukes, Reference Lukes1974, p. 27). Understanding conflicts among citizens hinges on the availability of political alternatives, making the structuring of these alternatives essential to shaping public opinion (Mansbridge, Reference Mansbridge2003, p. 519). If there is no supply of political alternatives, voters do not have much to base their updating on other than the consensus. In other words: ‘if all political persuasive messages in political media were favorable to one side, and there are no cueing messages to warn people that something runs counter to their values, we can expect voters to adapt their attitudes to this narrative’ (Zaller, Reference Zaller1992, pp. 97–117). O’Grady (Reference O’Grady2022) shows that public opinion in the UK sharply shifted to an anti-welfare stance after a bipartisan, sustained and one-sided anti-welfare discourse. Indeed, bipartisan convergence, in two-party systems, makes mass opinion much more likely to change (Amsalem and Zoizner, Reference Amsalem and Zoizner2022; Druckman et al., Reference Druckman, Peterson and Slothuus2013). The boundaries set by elites determine the scope of conflicts, with narrower boundaries resulting in fewer conflicts.
Examining how this dynamic manifests at the individual level involves theorizing about opinion change within different political contexts, drawing from classic models articulated by Zaller and Feldman (Reference Zaller and Feldman1992). These models are based on the fundamental assumption that individuals carry around multiple attitudes in their heads that might conflict with one another. Opinion change is driven by changes to the sampling from this mix of ambivalent considerations. Two pathways exist for these changes. They can arise either from a shift in the predominant ideas individuals have at the forefront of their minds when responding to survey questions or from a modification in the array of considerations present. If in the discursive context, citizens are not confronted with policy alternatives, we would anticipate to see changes in both the overall mix of factors they consider and the priorities among these factors.
Fiscal policy provides an excellent case for studying whether and to what extent the discursive context drives mass preferences. In fact, in many cases, defining budgetary limits relates directly to what is politically feasible. It has previously been argued that there is congruence between voter attitudes towards fiscal trade-offs and the distinct macroeconomic growth strategies of a country (Hübscher, Sattler, and Truchlewski, Reference Hübscher, Sattler and Truchlewski2023). However, in studying the political and societal consensus underpinning macroeconomic strategies and fiscal policies, the stated goal of the authors is not to examine what led to the formation of these preferences (ibid p. 980). In a similar vein, Stanley (Reference Stanley2014) studies the widespread public acquiescence to the idea of fiscal austerity, despite materialist expectations suggesting resistance to austerity measures. Stanley convincingly argues that elite-driven narratives have to resonate with the ‘mood of the times’, yet he does not offer an explanation of these preferences.
Other studies did seek to explain the directionality in the relation between elites and mass fiscal preferences. Barnes and Hicks (Reference Barnes and Hicks2018) demonstrate how narratives, like the portrayal of the frugal ‘Swabian housewife’, have been instrumental in framing discussions around fiscal austerity, debts, and deficits. Similarly, Barnes and Hicks (Reference Barnes and Hicks2021) and Ferrara et al. (Reference Ferrara, Haas, Peterson and Sattler2022) show that different ways to present the (non)issue of government debt in media significantly impacts public attitudes towards austerity. These results hint at the possibility that austerity can be ‘made popular’. The media serve as a potential mechanism through which elite narratives lead to mass preference alignment. If there is cross-party consensus, the media regard this as true and reproduces these ideas (Hopkin and Rosamond, Reference Hopkin and Rosamond2018).
Though media can act as a conduit for narratives to reach the public, political parties are crucial in shaping narratives around government debt. We know from earlier research that people develop preferences by adapting them from parties (Zaller, Reference Zaller1992; Zaller and Feldman, Reference Zaller and Feldman1992) and that individual fiscal preferences are highly sensitive to the specific political parties that back austerity packages (Bansak et al., Reference Bansak, Bechtel and Margalit2021). Altogether, this has led to the first hypothesis:
H1: All other things constant, voters adapt their fiscal preferences to the cross-party fiscal consensus on austerity.
For the second hypothesis, I add to the previous discussion the proposition that fiscal attitudes are also informed by material self-interest and argue that mass opinion change occurs at the intersection of material self-interest and the elite-driven discursive context (Cavaillé and Neundorf, Reference Cavaillé and Neundorf2022). The examination of how income, in particular, systematically shapes individual fiscal preferences is warranted by the understanding that income is not only unevenly distributed in society but, crucially, it influences attitudes towards fiscal matters such as taxation, social policy, unemployment benefits, debts and deficits (Gilens, Reference Gilens2009; Van der Brug and Van Spanje, Reference Van der Brug and Van Spanje2009). Moreover, income is argued to unequally impact political influence (see, for example, Elsässer et al., Reference Elsässer, Hense and Schäfer2021; Gilens, Reference Gilens2012; Schakel, Reference Schakel2021), and studying preference formation along these lines informs the debate on whether this could partly be explained by the rich anticipating elite positions (Joosten, Reference Joosten2023b). I start by examining how income systematically structures individual preferences, followed by a discussion of how this conditions opinion change driven by elites.
Perhaps unsurprisingly, the general consensus in the literature suggests that the wealthy are much more fiscally conservative than the poor. One study shows that compared to the general public, the affluent perceive budget deficits and excessive government spending as a much more important problem facing the government. Moreover, to deal with deficits the rich tend to favor spending cuts rather than tax hikes – again much more than the general public (Page et al., Reference Page, Bartels and Seawright2013). Such distinctive priorities are likely to be informed by material self-interest. Stix (Reference Stix2013), for example, provides empirical support for the hypothesis that those who expect to be affected by contractionary measures in the short run express significantly lower support for these policies than others, while those who expect to benefit from successful consolidation in the medium run show greater support.
The impact of fiscally contractionary measures theoretically varies depending on the composition of fiscal packages. Austerity packages can be designed with very few distributional consequences, for example without including cuts in spending that poor households benefit from and alternatively take the form of tax hikes or ‘neutral’ cuts in spending. However, in practice, some groups are much more affected by contractionary measures than others. In fact, some austerity-induced cuts to welfare spending have had highly concentrated and narrow costs (Fetzer, Reference Fetzer2019). Austerity has, in particular, affected those toward the bottom of the income distribution that are reliant on indirect (tax breaks and subsidies) and direct government services (transfers, health care) (Blyth, Reference Blyth2013, p. 28). If we assume that tax-based austerity mostly impacts affluent citizens and spending-based austerity primarily affects poor households, austerity’s track record across the Organisation for Economic Co-operation and Development’s (OECD) member countries confirms that austerity had mostly unequal material effects to the detriment of the poor. Devries et al. (Reference Devries, Guajardo, Leigh and Pescatori2011) show that across OECD countries between 1978 and 2007 fiscal consolidation by governments has on average been composed of roughly 73% spending-based and 27% tax-based. If the above holds true, we do not have to rely merely on the perception of being impacted by austerity. Instead, we can infer that income is directly connected to being impacted by austerity, and consequently, income is negatively associated with support for fiscal consolidation. Indeed, Hayo and Neumeier (Reference Hayo and Neumeier2017) show that subjective economic well-being negatively correlates with support for fiscal consolidation.
If we assume that there is variation in the fiscal attitudes of citizens along income lines (see also Ferragina and Zola, Reference Ferragina and Zola2022) prior to being exposed to the discursive context, then we should logically expect to see differences in the extent to which these groups align to elite narratives. In the following, I will argue that the intersection of these two determinants of mass opinion formation is key to understanding general support for fiscal policy. The argument is conditional in the sense that it assumes that individuals’ initial materialist orientation renders them more or less prone to be swayed by the discursive context.
To illustrate the mechanisms of such heterogeneous preference adaptation across income groups we can look at an extreme example. Page et al. (Reference Page, Bartels and Seawright2013, p. 55) show that in the United States after the Great Recession, 87% of very wealthy respondents said deficits were a ‘very important problem’ facing the country, while in the general public, only 7% ranks budget deficits as the ‘most important problem’. There are at least two ways in which such starting points inform opinion change.
First, there is an obvious ceiling effect that shapes the extent to which there is room to maneuver in the direction of the cross-party consensus. If, in the given example, party positions collectively emphasize budget deficits as an important issue, then there is limited space for the affluent to push even more in the direction favored by the parties. Using Zaller (Reference Zaller1992)’s conceptual understanding of mass opinion change, there is only so much more to change in the mix of considerations in the wealthy people’s heads or in the sampling of these considerations. Conversely, for the general public, there is a lot of room to move in the fiscally conservative direction. They have hardly any worries about government debt being a salient issue and so elite messages will most likely have some effect on the preferences distribution on debt-aversion in the general public’s heads.
The second mechanism relates to the novelty of information that parties present. This proposes that voters that have not yet considered a certain set of arguments or have not been strongly exposed to it, are more likely to adapt their preferences to this new information (Barnes and Hicks, Reference Barnes and Hicks2022). In terms of income and fiscally conservative messages, Elkjær (Reference Elkjær2020, p. 10) argues that low-income citizens are less exposed to the analogical reasoning of fiscally conservative policies and are therefore more likely to be susceptible to familiar and easily understandable objects, such as the personal household budget.
Both mechanisms indicate that there is greater alignment across income groups after exposure to the discursive context: in the example of a Keynesian consensus, we would anticipate movement among the affluent, whereas, in an austerity consensus, we would expect lower-income households to shift their views. The proposed dynamics relating to the convergence of preferences have been previously explored in an experimental setting. Studying whether citizens’ attitudes towards fiscal spending are sensitive to how the context is framed, Ferragina and Zola (Reference Ferragina and Zola2022, p. 340) demonstrate that ‘although there were some large variations across sociodemographic categories in the control group, all of these gaps were strongly diminished or fully erased in the treatment group.’
Finally, it is relevant to mention that individual party-voter dynamics might also be at play. From the party’s perspective, those voters that are the furthest away from the party’s position would be the groups that you want to convince when mobilizing support for and securing consent with, your (fiscal) policy agenda. Altogether, and following this line of argument, this leads to the second hypothesis:
H2: All other things constant, low-income voters adapt their fiscal preferences the most to the cross-party fiscal consensus on austerity.
Data and methods
To empirically test the expectations as outlined above, I rely on three survey waves from the ISSP, covering the period 1996–2016. These individual-level data are combined with party positions towards austerity as gathered by the MPD. The main analyses cover more than 43,000 individuals across 30 countries. The included countries are Australia, Belgium, Bulgaria, Canada, Switzerland, Czechia, Germany, Denmark, Spain, Finland, France, United Kingdom, Croatia, Hungary, Ireland, Israel, Iceland, Japan, South Korea, Lithuania, Latvia, the Netherlands, Norway, New Zealand, Poland, Portugal, Sweden, Slovenia, Slovakia and the United StatesFootnote 1 . Some countries are included in all three waves, some in two or just one. This resulted in a maximum of 60 country years. The respective regression tables report on the exact number of respondents and country years per analysis. Some country-years miss data on some of the variables and are therefore excluded from the analyses. Table A2 details a list of countries and years used.
Independent variable
To capture the discursive context, I rely for my independent variable on party attention toward different aspects related to austerity and expansionary fiscal policies taken from the Manifesto Project Database (MPD) (Volkens et al., Reference Volkens2021). While there are various elites, such as ‘politicians, higher-level government officials, journalists, some activists, and many kinds of experts and policy specialists’ (Zaller, Reference Zaller1992, p. 6), as well as corporate elites (Disch, Reference Disch2011, p. 100), as outlined in the theoretical framework, political parties are key to studying the elite-driven discursive context in relation to preference formation (Converse, Reference Converse and Apter1964; Key, Reference Key1966; Leeper and Slothuus, Reference Leeper and Slothuus2014; Mansbridge, Reference Mansbridge2003). The emphasis on election manifestoes is driven by the recognition that a significant portion of parties’ communication to citizens occurs within these documents (Däubler, Reference Däubler2012; Dolezal et al., Reference Dolezal, Ennser-Jedenastik, Müller and Katharina Winkler2012). Manifestoes are often not directly written with the expectation that large shares of citizens will read them. Instead, they could be directed at actors with an intermediate position between parties and voters. Target audiences could, for example, encompass journalists, and the broader media landscape, or they could be used more functionally as a document within tools designed to assist voters in identifying their preferred party. In any of these possibilities, most importantly, these documents serve as reliable indicators of parties’ positions.
The MPD uses content analysis to code the percentage of quasi-sentences in party manifestoes devoted to specific issues. To measure a party’s position on austerity, I use the indicator that is closest to this concept: ‘economic orthodoxy’. This indicator focuses on the need for ‘economically healthy government policy-making’ which includes calls for reduction of budget deficits; retrenchment in crisis; thrift and savings in the face of economic hardship; support for traditional economic institutions such as stock market and banking system; and support for strong currency. This variable reflects fiscal prudence and dealing with the government’s budget through spending and taxation policies related to fiscal debt and deficits and therefore captures ‘relevant aspects of an austerity agenda’ (Kraft, Reference Kraft2017, p. 1438). As this variable defines much of the fiscal starting point of a party manifesto, I argue that this gives an important and meaningful impression of the extent to which austerity is seeping through the remainder of the party manifesto.
Operationalizing fiscal consensus for the main analyses, I use two variables based on the individual party’s attention to economic orthodoxy. First, is the average across all parties in a given country year, weighted by seat share in parliament. With a contractionary fiscal policy stance setting the stage for the rest of individual manifestoes, high average party attention to this item should seep through the political environment in a country year (i.e. the remainder of the party platform emphasizes either cuts or investments). However, high average does not necessarily have to indicate that ‘there is no alternative’. In fact, the average can increase when one (significantly large) party shifts its position to austerity, while another offers a political alternative to this agenda. In this hypothetical case, I would expect less opinion change. This brings me to the second variable, which is consensus across parties. This measure is based on Dalton’s measure of polarization (Dalton, Reference Dalton1985). This measure includes (a) the relative position of each party on a scale – in my case the austerity scale -, and (b) the party’s position weighted by party size. This takes into account that a large party at an extreme signifies greater polarization than a niche party in the same position. For my analyses, I reversed the scores, so that higher values represent ‘consensus’ and not ‘polarization’. This measures the supply side of party competition and is only meaningful for my analyses in combination with the average score: high values can both mean consensus on austerity or non-austerity. The interaction between the two then represents my measure of austerity consensus.
Elections do not match perfectly with the ISSP’s survey waves and, therefore, I use linear interpolation in the missing years to estimate how relative attention changes between elections (following, among others, Schakel and Burgoon (Reference Schakel and Burgoon2022) and Manow et al. (Reference Manow, Schäfer and Zorn2008)). Sure, election manifestoes are written for election campaigns, but as I take manifestoes as indicators of the discursive context more broadly, we can assume that individual party attention gradually shifts from election to election. Indeed, it is unrealistic to expect that the discursive context changes only and abruptly during the drafting of the manifesto and the interpolated measure can adequately capture the gradual shifts that occur over time. Party positions lagged one year before the measure of individual preferences to try to get at the directionality of the relationship. The empirical models control for one’s own party, thereby testing whether people are responsive to the consensus net of their party identity. This is measured by linking voters to their own party through the party vote intention variable available in the ISSP.
Dependent variable
I use the preferences of voters as measured by the ISSP, specifically the Role of Government modules of 1996, 2006 and 2016. These waves all include a question directly asking respondents whether they are in favor of cuts to government spending. Moreover, these waves all include the same set of questions about spending preferences towards eight different issues: environment, health, law and order, education, defence, pensions, unemployment and culture. As will be detailed in the empirical section, for the main analyses I use spending preferences in health, pensions and unemployment as policy domains on which there is significant polarization among income groups. In addition, these are important social policy spending areas that represent large shares of government spending and are therefore likely to be subject to spending cuts. The other domains are analyzed in the online Appendix, like the survey items gauging voters’ tax preferences, which are divided into preferences for taxing low-, middle- and high-income groups.
All individual preferences are recoded from the scale of 1–5 to go from −100 to +100, with 50-point intervals.Footnote 2 The answer categories for support for spending cuts are reversed so that higher values indicate ‘in favour’ of cuts and lower values indicate ‘against’ cuts. Support for spending in the different policy areas is also reversed so that negative values indicate preferences for less spending and positive values represent demands for more spending. To be sure, I operationalize demands for ‘more cuts’ (or ‘more taxation’ and ‘less spending’) as fiscally austere attitudes, while ‘less cuts’ (or ‘less taxation’ and ‘more spending’) represent a respondent’s wish for more expansionary fiscal policy.Footnote 3
These survey questions are all so-called ‘unconstrained’ questions and might therefore not reflect the trade-offs that policymakers face in real life. Moreover, this means that it is possible for respondents to have ‘inconsistent preferences’, wanting ‘more for less’ or a ‘free lunch’. Then again, as is true for the literature using trade-off questions, the survey items do not exclusively cover all possible policy instruments to run a government budget. I acknowledge this limitation of the data but argue that it is still relevant to study citizens’ overall support for spending cuts and attitudes toward spending and taxation and see whether we can observe systematic variation.
Other relevant variables that I use in the analyses are income and party. For income, I rely on thirds of the income distribution as my group structure. The ISSP asks for net household income, which I use per country year to construct three income groups of roughly equal size. Table A3 shows the distribution across income groups per country year. Party affiliation is measured by asking respondents what party they would vote for if there would be a general election this week.Footnote 4 On average, elections – therefore MPD datapoints – are roughly one-and-a-half-years before ISSP waves. The maximum distance in my data between an election and a subsequent ISSP survey wave is four years. As the variables that I use for the empirical models are interpolated between elections, it should be clear that when the difference is larger (four years) a party position relies more heavily on the election after the ISSP. In the supplementary analyses, I replicate the results using only those surveys that were conducted within two years of an election.
As control variables, I include variables that could both affect party positions as well as voters’ preferences. These are, in line with the literature, logged gross domestic product (GDP), annual GDP growth, and the unemployment rate, and are all downloaded automatically from the servers of the World Bank using the ‘wbstats’ package in R (Piburn, Reference Piburn2020). Furthermore, I include dummies for countries and years. As individual-level controls, I include gender and age. Summary statistics for the independent, dependent and control variables can be found in Table A1, while Table A2 shows the averages by country year.
Empirical strategy
This study aims to empirically examine the correlation between the emphasis on economic orthodoxy in political party manifestoes and the backing for government spending across various income brackets. Thus, I assess whether there is higher support for spending cuts (lower support for spending) among voters who are exposed to higher levels of austerity consensus (H1) and whether this effect is substantively larger among lower-income voters (H2). To investigate these relationships, I particularly consider the interaction term between the standardized measures of economic orthodoxy levels and the standardized cross-party consensus on economic orthodoxy within each country year. All my models are estimated with robust standard errors.
I use a series of two-way fixed effects (FEs) models. This choice is based on several theoretical and empirical considerations. First, country and year FEs are used to account for the fact that respondents are nested in countries and to control for time trends across several decades. The average number of individual-level observations in each country year is 1111, and 370 if subset by income groups. Regarding the latter, the smallest groups were high-income voters in Ireland in 2005 (n = 142) and the largest groups were middle-income voters in Australia in 2007 (n = 992). With 60 units in the main analyses, it is preferred to have higher variability and lower bias (Clark and Linzer, Reference Clark and Linzer2015).
Furthermore, I include as individual-level control variables, gender and age that seem to determine spending preferences. To isolate the relationship that I am interested in, considering how both parties and voters might be affected by the macroeconomic context, I include country-level controls GDP, unemployment and GDP growth. All analyses show the aggregate models and differentiate between income groups by splitting the samples. Voter preferences for spending (cuts) are regressed on parties’ attention to austerity at t-1. The observations are weighted by the ISSP weights.
I also specify a range of sensitivity tests with alternative independent and dependent variables, different groupings, and alternative model specifications. These results are reported in the main text and more elaborately discussed in the online Appendix.
Analysis
This section presents the results of the empirical analyses testing the hypotheses. Table 1 shows the fiscal preferences by income group averaged across the complete dataset for those that are included in the regression models. I test the hypotheses in terms of voter support for two spending indicators. First, voter support for cuts to government spending, which relates closely to economic orthodoxy fiscal policy. Second, I test the hypotheses for support for spending categories exhibiting significant polarization across income groups, which also relates closely to economic orthodoxy and additionally reveals clear polarization across income groups, pivotal for testing H2. Furthermore, old-age pensions, health care and unemployment are more than any of the other policy domains relevant to fiscal consolidation or fiscal expansion due to their size and relevance to the welfare state.
Table 1. Average fiscal preferences by income group
Table 1 indicates that there is no evident polarization in average support for government spending cuts among income groups across all country years. While higher-income individuals exhibit the greatest support for spending reductions, the variations are not particularly significant. In fact, middle-income voters show the weakest support. Of course, this hides relevant differences across countries and years where polarization might be present. For example in Denmark in 2008, there is nearly a twenty percentage point difference between low- and high-income voters, as shown in Table A2.
Support for spending in pensions, health care and unemployment shows starker differences between income groups, clearly following the anticipated pattern based on a material self-interest perspective. Low-income voters are the biggest supporters of spending in these policy areas, followed by the middle-income groups and, at roughly similar distance, are high-income voters.
To assess whether citizens align their fiscal preferences to the cross-party fiscal consensus, I first show in Table 2 the results of economic orthodoxy and support for cuts to government spending. As a reminder, I am interested in the coefficients of the interaction term as well as in the effect of the average economic orthodoxy. The interaction term holds the most significance as I expect a shift in preferences when a country year is marked by political parties adopting a firm and unified stance on economic orthodoxy. I moreover expect that the weighted average party position is meaningful too, as the share of economically orthodox political messages should be higher in such a context – even when there are political alternatives available. The consensus term, by itself, lacks specificity because it could imply agreement on any degree of economic orthodoxy. In terms of cuts to government spending, I expect mass support to increase by economic orthodoxy.
Table 2. Linear regression models of voters’ preferences for cuts to government spending
Note: *p**p***p < 0.01.
Looking at model (1) in the first column of Table 2, we can see that while the average position in a country year does not lead to any significant change in voter preferences, when this position is consensual among political parties, opinion shifts accordingly. In terms of control variables, one’s own party shows a strong positive effect. As previously outlined, due to the risk of endogeneity using these data, this can be explained both by a cueing as well as a selection effect. Citizens in richer countries as measured by their (logged) GDP want to see more cuts to government spending. The same goes for higher unemployment rates, while economic growth increases support for government spending. Finally, in terms of individual controls, older people are more fiscally conservative and women are less supportive of spending cuts than men. As these control variables are roughly the same across the different models, I will not further mention these when discussing the results of the split-sample models.
Turning to the different income groups shown in models (2)–(4), we can first observe that even though low- and middle-income voters demand more spending cuts under increased average economic orthodoxy, these results do not reach statistical significance. Peculiarly, the rich demand significantly less spending cuts when average economic orthodoxy is high. Looking at the interaction term, interestingly the average results that were shown in model (1) appear to be driven by the adaptation of low- and middle-income voters. High-income voters do not respond differently to austerity under consensus. The empirical finding that the rich want to see less spending cuts under higher levels of austerity, both when politically polarized and consensual, is counterintuitive and requires further scrutiny. Elkjær and Iversen (Reference Elkjær and Iversen2020) argue that affluent citizens become more favorable towards government spending in times of economic downturn, which could partly explain these results if austerity is proposed in such periods.Footnote 5 Nevertheless, it’s prudent to interpret these findings with caution since they have not been corroborated in other analyses.
Figure 1 visualizes the predicted effect of a one-unit increase in economic orthodoxy on support for cuts to government spending for different levels of consensus. This shows that when there is polarization, depicted by the average minus one standard deviation on the left-hand side, economic orthodoxy has no effect on low-income voters and a slightly negative effect on middle-income voters. When there is very little polarization, on the right-hand side shown at the average plus one standard deviation, both low and middle-income voters align with the political consensus. As could also be concluded from looking at the interaction terms in Table 2, high-income voters negatively respond to economic orthodoxy both when it is consensual or polarized.
Figure 1. Predicted support for cuts to government spending for different levels of consensus.
The results in Table 2 provide support for H1, and partly for H2. But, because in the data preferences for spending cuts are not very polarized by income, the results do not allow for a direct test of the suggested mechanism. To further explore the theoretical expectation that those the furthest away from the consensus adapt the most, Table 3 shows the same models, but taking support for spending in pensions, health care and unemployment as the outcome variable. As discussed, the income differences between income groups are much more pronounced for attitudes toward spending in these areas. To be sure, in these models higher values indicate support for more spending, thus in Table 3, I expect the signs of the coefficients to be negative (as opposed to the expected direction of the coefficient in Table 2).
Table 3. Linear regression models of voters’ preferences for spending in pensions, health and unemployment
Note: *p**p***p < 0.01.
Starting again with the aggregate results demonstrated in model (1), both the weighted mean of economic orthodoxy as well as the interaction term show results in the expected direction. Looking at the samples split by income groups, we can first observe that low- and middle-income voters respond to the weighted average position of economic orthodoxy by demanding less spending. Interestingly, and contrary to the counterintuitive results in Table 2, affluent voters shown in model (4) do not adjust their preferences to the average party position. Looking at the interaction term, all groups show significant coefficient signs in the expected direction, meaning that when austerity is consensual the rich also align their preferences. However, the size of the effect in the high-income sample is half the size of those of low- and middle-income voters. Finally, to be specific, the middle-income voters adapt slightly more than low-income voters.
Again, Figure 2 visualizes the interaction effect of economic orthodoxy average and voter support for spending in health care, old-age pensions and unemployment benefits for a situation where there is consensus and one where there is polarization. In particular, it shows the effect of one unit increase in economic orthodoxy on support for government for different consensuses. We can observe that while there is little movement under polarization (except for a slight positive coefficient for the affluent), especially low- and middle-income voters align with austerity if consensual by demanding less spending. As richer voters advocate for reduced spending, on average, individuals with lower- and middle-incomes tend to align with their stance amidst an austerity consensus, leading to a decrease in the initial polarization.
Figure 2. Predicted support for spending in pensions, health and unemployment for different levels of consensus.
In terms of testing the hypotheses, I can confidently confirm empirical support for H1 that voters collectively adapt to the austerity consensus. The findings regarding H2 exhibit a slightly more nuanced pattern. In both models, low-income voters, who are anticipated to be primarily affected by expenditure-based consolidation, converge with the cross-party consensus. Somewhat surprisingly, middle-income voters also demonstrate alignment, to an even greater extent. This would suggest support for Zaller (Reference Zaller1992)’s hypothesis that is mostly based on receiving and accepting elite cues. Those groups in the middle are the most susceptible to elite messages because they are likely to receive the cue, but are not yet capable of resisting it. Finally, the results for high-income voters are more puzzling and seemingly incoherent. The empirical analyses demonstrate that they want both less spending cuts, regardless of whether consensus exists, as well as less spending in the large social policy areas following austerity messages.
Robustness
In the interest of space, I relegated some substantively interesting additional analyses to the Appendix and will discuss the main findings here.
First, I look at the distinct results by individual spending and taxation survey items that are available in the ISSP. The results, shown in Appendix B, are similar to the main results. Low- and middle-income voters want to see significantly less spending when economic orthodoxy is high and this effect is even more pronounced when there is consensus. The adaptation for low- and middle-income voters is widespread across policy domains. The rich do not adapt unless austerity is consensual, but still to a lesser extent and not in all spending areas. In terms of taxation, the results are more mixed. Low-income voters do not adapt at all, while middle- and high-income voters want to see less taxation on middle- and high-incomes under austerity consensus. Voters do not seem to want to ‘rebalance the budget’ – as support for higher taxation would indicate -, but instead they are more likely to demand a smaller government and possibly trickle-down policies by demanding less taxation on the rich. Interpreting austerity not just as a temporary solution to rebalance the government budget, but as a broader paradigm to shrink the size of the government (Blyth, Reference Blyth2013), means that these results make sense. Moreover, these results are in line with the earlier mentioned notion that austerity in the OECD has consisted primarily of spending-based consolidation as compared to tax-based.
The second substantively interesting additional analysis pertains to using party attention to expansionary fiscal policies as opposed to contractionary fiscal policies. I replicate the main results using Keynesian demand management as the independent variable. I expect results in the opposite direction of those when using economic orthodoxy. Moreover, in using this variable the distance to the party position should be reversed, meaning that I expect the rich to be the furthest away, while low-income voters are closer to a Keynesian position. The results show in Appendix C that in terms of support for spending cuts, all income groups adapt their preferences in the expected direction, demanding less spending cuts. Moreover, the results are much more pronounced when the Keynesian position is consensual. The interaction term shows that, indeed, the consensus effect is somewhat stronger for the rich, but the direct effect of average Keynesian position is stronger for low-income voters. In terms of support for spending on the polarized policy domains, there is no direct effect but there is a positive interaction effect for all groups. All income groups want to see more spending when there is party attention to Keynesian demand management. This analysis shows that there are no large differences across groups so that all groups adapt roughly to the same extent.
Then, in Appendix D, I group survey respondents into left and right voters. Left voters tend to see the economy through a Keynesian framework, while right-wing voters are more likely to look at the economy through a lens that promotes limited state intervention (Hübscher et al., Reference Hübscher, Sattler and Wagner2021, pp. A8−A9). Following Schakel and Burgoon (Reference Schakel and Burgoon2022), under left-wing parties I group social-democratic, socialist and green parties and under right-wing parties, I group liberal, conservative and nationalist parties. As the descriptives show, there is significant polarization between these groups. Left-wing voters are much less supportive of spending cuts and much more supportive of spending on health care, old-age pensions and unemployment, compared to right-wing voters. In turn, the results clearly show that left voters, who are further away from the consensus, strongly adapt to party attention to economic orthodoxy, particularly when consensual. For both outcome variables, a standard deviation increase in economic orthodoxy when there is party consensus changes the polarized situation to near consensus.
In Appendix E, I group voters in other different categories. Specifically, I look at education and self-reported being informed. While, in particular, education is argued to be a better predictor of the ‘cultural dimension’, as opposed to the ‘economic dimension’ that is studied in this paper (Schakel and Hakhverdian, Reference Schakel and Hakhverdian2018; Van der Brug and Van Spanje, Reference Van der Brug and Van Spanje2009; Werfhorst and Graaf, Reference Werfhorst and Graaf2004), these are categories that are often used to study whether someone receives and accepts elite cues (Zaller, Reference Zaller1992). I group voters into three education levels and three broad categories of being informed and replicate the main results. In line with the argument that these are not the best predictors of the economic dimension, average support does not show the same pattern by low, middle and high groups. The highly educated are least supportive of cuts to government spending and also of support for higher spending, and the most informed are most supportive of cuts to government spending as well as being most supportive of spending. It is then perhaps unsurprising that indeed the highest groups adapt the most.
Finally, in Appendix F, I consider several technical robustness tests. As a general remark, the results do not hold to all alternative specifications and I contextualize in more detail the individual tests in the Appendix.
Discussion
This paper started from the empirical puzzle that some authors find that austerity triggers popular discontent while others do not find that governments’ re-election chances are systematically harmed after implementing austerity packages. This is a relevant debate not in the last place because scholars have argued that we have been living in an ‘age of permanent austerity’ (Pierson, Reference Pierson1996). I argued that if the latter is true, from a mass preference formation perspective we would expect that voters have adapted to this austere paradigm, muting its unpopularity. This led to the first hypothesis, in which I expected that voters in general adapt to the cross-party austerity consensus. Moreover, as austerity has unequally distributed effects on society, I conceptualized how fiscal preferences are structured by broadly defined income groups. Combining the two, I expected that low-income voters, who are naturally more distant from supporting austerity, would adapt mostly to when party attention to austerity is high.
I find clear support for the first hypothesis. Across all models, on average voters adapt either directly, or only when consensual, to party attention to economic orthodoxy in the expected direction. Perhaps, more interestingly, is the question of whether cross-party consensus leads to consensus among voters. Are low-income voters’ preferences further away from supporting austerity and do they converge under cross-party consensus? I find support for this claim, although the convergence does not exclusively pertain to low-income voters, but also to middle-income voters. These dynamics are most obvious when looking at the polarized spending areas of unemployment, pensions, and health care. The same dynamics are observable for left- and right-wing voter groups, which is another source of stark polarization. I conclude that while austerity might be unpopular among some broad segments of society, it can be ‘made popular’ (Barnes and Hicks, Reference Barnes and Hicks2018) by political elites.
Two contributions emerge from this article. First, the absence of electoral consequences for fiscal austerity can be explained by the fact that voters adapt to austerity as advocated by political parties, thereby refraining from punishing governments upon their implementation of such policies. Particularly when groups less inclined to support austerity measures based on their socioeconomic status align with the political consensus, then it is not surprising that voters do not punish governments. This supports the thesis that the implementation of austerity has to resonate with the public’s sense of debt as a moral obligation (Stanley, Reference Stanley2014). Second, these results could suggest that unequal political representation in favor of the rich by parties (see, for example, Schakel and Burgoon, Reference Schakel and Burgoon2022) generates a consent-making loop whereby the rich influence party positions, which in turn influence the political preferences of low- and middle-income voters. Although Elsässer and Haffert (Reference Elsässer and Haffert2022) did not find that fiscal pressure unequally constrains policy responsiveness, on average, policy responsiveness decreases.
Finally, the conclusion of this article raises the question of where parties’ economic policy positions come from, if not from their voters. Economic policy positions are argued to be the preserve of unelected technocrats such as central bankers (Hopkin and Rosamond, Reference Hopkin and Rosamond2018) or economic experts (Joosten, Reference Joosten2023a). Austerity consensus, such as in the 2010s, might then be ‘attributed to the advice of economic experts, pressures generated by financial markets, and lobbying by export-oriented firms’ (Pontusson, Reference Pontusson2024, p. 9). Studying mutual responsiveness and inequalities therein along the long chain of political representation, then, requires more attention (Burgoon et al., Reference Burgoon, Lupu, Pontusson and Schakel2022).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1755773924000171.
Acknowledgments
Earlier versions of this paper were presented at the SPERI Doc/Post-Doc Colloquium at the University of Geneva on 24 November, the Swiss Political Science Association Annual Congress on 4–5 February 2021 and the European Political Science Association Annual Conference on June 24–25, 2021. I thank participants of these seminars for their constructive feedbacks, and, in particular, Elisa Volpi, Thomas Sattler, Jonas Pontusson as well as the two anonymous reviewers for their constructive feedback.
Note on contributor
Max Joosten is a Postdoctoral Researcher in the Department of Political Science at the University of Amsterdam (UvA). He received his PhD from the University of Geneva in 2023.
Funding
This work was supported by the European Research Council under Grant agreement no. 741538.
Open access
