INTRODUCTION: PROGRAMMATIC REACTIONS TO INEQUALITY
Economic inequality is one of the greatest challenges for contemporary democracies with tremendous implications for people’s living conditions, political representation, and political fairness (Jensen and Jespersen Reference Jensen and Jespersen2017; Polacko et al. Reference Polacko, Heath, Lewis-Beck and Dassonneville2021; Solt Reference Solt2008; Reference Solt2010; Trump Reference Trump2018). Despite its relevance and ongoing debates over the inability or unwillingness of political actors to tackle growing levels of economic disparities, research remains inconclusive regarding parties’ reactions to economic inequality.
On the one hand, a large, long-standing literature on the postindustrial transformation of electorates and parties and the weakening of class ties (Häusermann and Bornschier Reference Häusermann and Bornschier2023; Marks et al. Reference Marks, Attewell, Rovny, Hooghe and Riddervold2021) suggest that parties, even left parties, have become unlikely to address high and growing economic disparities with a redistributive agenda that aims at achieving economic equality. An important theme in this literature is that the realignment of the electorate and the rise of a new middle class as a core left support group have made the traditional contrast between a redistributive Left and a noninterventionist, pro-business Right less plausible. From this perspective, it seems unlikely that (left) parties address persistent economic inequalities.
On the other hand, political science studies specifically devoted to the programmatic responses of parties (Pontusson and Rueda Reference Pontusson and Rueda2010; Rueda Reference Rueda2008; Tavits and Potter Reference Tavits and Potter2015) are positive regarding responses to inequality. These more specific studies (Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015) suggest that party responses follow classic left–right patterns. They expect—and find—left responses, as high inequality is assumed to translate to more demand for redistribution and a larger constituency for left parties. Conversely, center–right parties—in lieu of a viable economic response—avoid rather than address inequality (Tavits and Potter Reference Tavits and Potter2015). This idea that right parties deflect from inequality is in line with research on the strategies of center–right parties (Gidron and Ziblatt Reference Gidron and Ziblatt2019; Hacker and Pierson Reference Hacker and Pierson2020). Few would deny the validity of this diagnosis. By contrast, the positive results for the Left are difficult to reconcile with the broader literature, widespread public and scholarly critiques of the Left’s alleged lack of political will to address high and rising inequality (Piketty Reference Piketty2020), and the empirical reality that inequality has been increasing since the 1980s (Atkinson Reference Atkinson2015).
Until now, the lack of comparable time series data on parties’ economic equality appeals has prevented us from directly addressing supply-side questions about how (left) parties position themselves on economic equality and redistribution and whether they respond to inequality. Using new time series data that, for the first time, capture economic equality emphasis in party platforms across 12 countries, we analyze parties’ (non)responses and offer insights into why high inequalities persist.
Theoretical Contribution
We adopt the basic assumption that left rather than right parties respond to inequality because inequality affects and expands their less affluent core constituency (Anderson and Beramendi Reference Anderson and Beramendi2012; Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015). Yet, to reconcile affirmative and skeptical perspectives toward left responses, we further argue that it is necessary to differentiate whether high inequality levels or increases in inequality lead to programmatic reactions. We develop and combine three arguments to explain why (even left) parties, at best, respond to increases in inequality but rarely to high levels. We focus on how a lack of visibility, barriers to mobilization, and status quo bias affect party incentives and thereby undermine level effects or induce change effects. While these are not formal mechanisms that we can directly test here (as is also the case in the reference studies), they jointly provide a credible rationale for why inequality levels fail to trigger responses, whereas rising inequality does.
First, for economic inequality to directly influence voters and party platforms, it must be noticed. However, public perceptions of inequality do not adequately reflect empirical levels of inequality (Kenworthy and McCall Reference Kenworthy and McCall2008; McCall Reference McCall2013). For inequality, as with comparable indicators, established theories of newsworthiness and gatekeeping (Galtung and Ruge Reference Galtung and Ruge1965; Shoemaker and Vos Reference Shoemaker and Vos2009) posit that media reporting prioritizes novel developments and changes over levels of inequality. Thus, persistently high inequality remains invisible or attracts only little visibility, and should thus not incentivize parties to respond via increased emphasis on economic equality. By contrast, rising inequality draws public and political attention—incentivizing parties to respond with increased emphasis.
Second, electoral mobilization presents another obstacle to party responsiveness to inequality. Economic inequality has a negative effect on electoral turnout (Goodin and Dryzek Reference Goodin and Dryzek1980; Solt Reference Solt2008), as it suppresses participation among low-income and—more generally—below median-income voters (Fenzl Reference Fenzl2018; Gallego Reference Gallego2015; Mahler Reference Mahler2008). In line with Anderson and Beramendi (Reference Anderson and Beramendi2012), we argue that, if inequality is high, it becomes harder for left parties to mobilize their poorer (below median) constituencies. This, in turn, reduces the incentives for strong redistributive left platforms and pushes left parties toward more moderate strategies.
Finally, inequality does not necessarily generate political responses because cognitive heuristics and ideological commitments contribute to its self-stabilization (Jost, Banaji, and Nosek Reference Jost, Banaji and Nosek2004; Mijs Reference Mijs2021; Trump Reference Trump2018). Individuals legitimize existing disparities through system justification or meritocratic beliefs, perceiving inequality as either fair or inevitable. While these processes create a status quo bias that dampens redistributive egalitarian demands, experiments show that awareness of rising inequality—rather than awareness of levels—can increase support for redistribution. This, again, reinforces the assertion that parties are less inclined to emphasize economic equality when inequality is high but more likely to do so amid rising inequality.
As a whole, these arguments support our expectation that parties, including the Left, do not challenge high inequality, helping to explain its persistence. While left parties should respond to rising inequality (with the theoretically plausible potential exception of much less visible gains for the top 1%), they are unlikely to address entrenched economic disparities.
Empirical Contribution
Yet, without the new data introduced in this article, it would not be possible to assess party responses to inequality in a way that captures redistributive appeals. Thus, the main empirical contribution of this article is the introduction and analysis of a new, extensively validated, crowd-coded dataset that allows us to track how much parties emphasize economic equality and redistribution, vis-à-vis appeals to equal rights and equal opportunities.
Existing studies that examine party responses rely on too broad measures of appeals to equality and redistribution derived from party manifestos. For instance, statements on equal rights and anti-discrimination—topics that have become increasingly prominent in left-party platforms since the 1970s—are often misclassified as redistributive appeals, despite being conceptually distinct from economic equality (with a correlation of 0.19). As we show, correlations between actual appeals to economic equality and the broad measures commonly used hover around 0.3. We are able to address these limitations based on a new data collection pipeline that combines machine learning as a filtering mechanism, expert judgment for extensive validation, and online crowdcoding for large-scale classification of party programs in 12 OECD countries between 1970 and 2020. This approach enables us to precisely differentiate positive references to economic equality from other equality-related statements. This is a prerequisite for explaining why parties respond—or, more often, fail to respond—to economic inequality, offering crucial insights into why high inequality levels persist.
WHY PARTIES REACT TO RISING INEQUALITY, NOT INEQUALITY LEVELS
We study whether parties respond to economic inequality with economic egalitarianism. While studies that specifically examine programmatic responses to inequality are relatively new and scarce, political scientists and political sociologists have researched the broader question of party responsiveness to the working class for decades. Summarizing this broad literature, we conclude that it yields negative expectations regarding party responses. Yet, none of this work has systematically examined the links between party platforms and inequality. Thus, we turn to research that specifically examines party platforms and programmatic responses to inequality. This scholarship yields the opposite conclusion that left parties do, in fact, respond to inequality. In a third step, we expand on earlier arguments about visibility, mobilization, and status quo biases to explain why we expect party responses to rising inequality but not to high inequality levels.
Why Parties Do Not Respond
In a more general sense, political scientists and political sociologists have examined the question of whether parties respond to economic inequality for a long time. Several bodies of literature suggest that party responses are limited at best, both theoretically and empirically.
First, the considerable literatures on the transformation of party systems, cleavages, and the “death of class” (Häusermann and Bornschier Reference Häusermann and Bornschier2023; Marks et al. Reference Marks, Attewell, Rovny, Hooghe and Riddervold2021) suggest that even left parties have only limited incentives to continue responding to (class) inequality. A key motive in this research is that, as a result of societal transformation (e.g., tertiarization), matters of redistribution, leveling, and welfare state consumption have become less central for the new electoral core groups of the Left (e.g., the sociocultural professionals; Beramendi et al. Reference Beramendi, Häusermann, Kitschelt and Kriesi2015)—which are, on average, more affluent, more educated, and more invested in sociocultural rather than economic issues. Yet, the empirical focus in classic (Kitschelt Reference Kitschelt1994) and contemporary (Häusermann and Kitschelt Reference Häusermann and Kitschelt2024) works in this tradition is less on inequality and redistribution (per se) and more on welfare state recalibration.
Second, yet related, one could go further back to more historical and country-specific studies that shed light on the changing links between (left) parties and (the working) class. One strength of this work is that it does not only take societal and electoral change into consideration but also sheds light on the role of agency of collective actors such as parties, unions, and business interests. For example, historical work for the U.S. challenges the view that the Democrats moved to the right and left the New Deal coalition in response to an alleged economic shift to the right in public opinion. Instead, this work highlights the role of the waning influence of unions and the increasing importance of business donations (Eidlin Reference Eidlin2018; Ferguson and Rogers Reference Ferguson and Rogers1986; Hacker and Pierson Reference Hacker and Pierson2010). Similarly, the Labour Party’s abandonment of the British working class (Evans and Tilley Reference Evans and Tilley2017) is another illustration that electoral change and political exclusion of the working class are interrelated.
Overall, this work highlights how changes in core groups and strategic priorities have weakened left parties’ incentives and commitment to addressing material disparities.
Then, there is a third body of work that arrives at equally skeptical conclusions looking at parties in government and their policies: this “unequal democracy” literature on the link between preferences of economic groups and policies suggests an outsized influence of the most affluent groups and little to no influence of poor and lower middle classes. Mounting evidence suggests that governments produce policy outcomes that are biased against the poor and favor the most affluent in the United States (Bartels Reference Bartels2008; Bowman Reference Bowman2020; Epp Reference Epp2018; Gilens Reference Gilens2012; Gilens and Page Reference Gilens and Page2014), Europe (Elsässer, Hense, and Schäfer Reference Elsässer, Hense and Schäfer2021; Mathisen Reference Mathisen2023; Rosset Reference Rosset2013; Schakel Reference Schakel2021), and cross-nationally (Lefkofridi and Giger Reference Lefkofridi and Giger2020; Lupu and Warner Reference Lupu and Warner2022). In light of these different literatures, we would not expect party responses. Yet, while the above work provides crucial context and defines the background against which party responses to inequality are unlikely, it does not specifically assess party responses to economic inequality.
Why Do Parties Respond: Parties’ Programmatic Responses to Inequality
Building on research in political economy, a distinct body of work examines how parties respond programmatically to inequality. A central argument in this literature is that left parties should—and often do—adjust their platforms in response to rising economic inequality. These studies build on but move beyond the median voter theorem (Meltzer and Richard Reference Meltzer and Richard1981), which assumes that the median earner benefits from redistribution and that parties adjust their platforms accordingly. However, more recent research challenges these assumptions, noting that the median voter is rarely a net beneficiary of tax-transfer systems (Milanovic Reference Milanovic2000). Instead, left parties’ incentives to address inequality are shaped by the material conditions of their constituencies—particularly those at or below the median-income threshold (Pontusson and Rueda Reference Pontusson and Rueda2010, 681–2) who may experience downward mobility when inequality is high(er). As these voters “fall into lower socioeconomic categories,” they become more receptive to redistributive appeals (Tavits and Potter Reference Tavits and Potter2015, 744).Footnote 1
Taken together, these studies suggest that higher inequality can increase electoral demand for redistribution in two interconnected ways. First, inequality expands the share of the electorate experiencing economic hardship, thereby broadening the constituency receptive to redistributive policies (Tavits and Potter Reference Tavits and Potter2015). Second, it increases the gap between lower- and middle-income groups and the broader electorate, reinforcing the redistributive orientation of left parties’ core constituencies (Pontusson and Rueda Reference Pontusson and Rueda2010). Both dynamics reinforce incentives for left parties to emphasize debates on economic equality and redistribution in their agenda and in their programs.
However, the extent to which left parties respond to inequality also depends on the mobilization of low-income groups (Pontusson and Rueda Reference Pontusson and Rueda2010). High levels of inequality make this mobilization more difficult and weaken the electoral pressures and incentives to target low-income voters (Anderson and Beramendi Reference Anderson and Beramendi2012). Yet, as we argue below, low turnout is only one of several system-level factors that constrain party responses to inequality.
Why Parties Respond to Changes, But Not to Levels of Inequality
Extending the argument presented in the introduction, we draw on and tie together three arguments (on visibility, mobilization, and status quo bias) that help explain the (lack of) incentives for parties to respond. Together, these arguments provide a credible rationale for why even left parties do not respond to entrenched levels of inequality but only to increases.
Visibility/Awareness
Some studies implicitly assume that (levels of) inequality has (have) imminent and direct effects on voters and party platforms. This requires at least one important precondition: that voters and parties take note. However, perceptions of inequality do not mirror the actual levels of inequality in a given context (Kenworthy and McCall Reference Kenworthy and McCall2008; McCall Reference McCall2013). One important driver of public perception is media attention, which influences audiences’ perceived importance of (McCombs Reference McCombs2005) and knowledge about the object of coverage (Eveland and Scheufele Reference Eveland and Scheufele2000). Yet, media coverage does not track real-world indicators, such as unemployment and inflation (Mosley Reference Mosley1984) or immigration rates (Vliegenthart and Boomgaarden Reference Vliegenthart and Boomgaarden2007). Rather, in line with theories of newsworthiness and gatekeeping (Galtung and Ruge Reference Galtung and Ruge1965; Shoemaker and Vos Reference Shoemaker and Vos2009), the media tends to report on novel events or (drastic) changes and developments. This also holds true for reporting on inequality. The available research shows that media attention to economic inequality does not reflect actual levels of inequality (McCall Reference McCall2013 for the United States; Schröder and Vietze Reference Schröder and Vietze2015 for Germany). Instead, media coverage focuses—by and large—on rising disparities (McCall Reference McCall2013) or emphasizes increases in inequality (Schröder and Vietze Reference Schröder and Vietze2015). These biases in coverage result in lower public visibility and salience of inequality levels. By contrast, public attention and issue salience are more pronounced with regard to increases in inequality, given their greater newsworthiness and the more tangible problems and (political) conflicts they yield. Parties’ (non)responses to inequality should correspond to these visibility dynamics. Since media coverage amplifies rising inequality rather than entrenched levels, (left) parties face strong incentives to respond only when inequality is increasing, not when it is (persistently) high.
Mobilization
A second obstacle to parties’ responsiveness to economic inequality is electoral mobilization. While one might expect higher inequality to spur greater voter participation—since a growing share of economically disadvantaged citizens would demand more redistribution through voting (Meltzer and Richard Reference Meltzer and Richard1981)—an influential body of research suggests the opposite. These theories argue that higher inequality depresses turnout among low-income voters, as the wealthy leverage their resources to dominate the political process, thereby demobilizing the less advantaged (Goodin and Dryzek Reference Goodin and Dryzek1980). The idea that high levels of inequality reflect unequal power balances that undermine rather than trigger egalitarian responses is also central to power resource theory (Korpi Reference Korpi1983), according to which the welfare state and inequality function not only as outcomes but as drivers of redistribution. In highly unequal contexts, the poor have particularly limited resources to advance their political interests and the rich tend to control the political agenda (Gilens and Page Reference Gilens and Page2014; Schattschneider Reference Schattschneider1960). In such settings, low-income voters are more likely to “conclude that there is little point to being engaged in politics” and thus disengage from electoral participation (Solt Reference Solt2008, 49). Studies support a negative effect of inequality on turnout (Fenzl Reference Fenzl2018; Gallego Reference Gallego2015; Mahler Reference Mahler2008; Solt Reference Solt2008).
Yet, the demobilization of low-income voters affects parties differently, depending on their electoral base and ideological profile. Despite changes in party support over time, people in the lower half of the income distribution still constitute a core constituency of left parties (Abou‐Chadi and Hix Reference Abou‐Chadi and Hix2021; Pontusson and Rueda Reference Pontusson and Rueda2010). This indicates that (low-income) turnout will predominantly affect the electoral performance of the Left. Accordingly, Bechtel, Hangartner, and Schmid (Reference Bechtel, Hangartner and Schmid2016) present causal and correlational evidence that higher turnout due to compulsory voting strengthens left-party electoral performance. This suggests that redistributive platforms of the Left are more electorally viable under low levels of inequality, as poorer voters are more likely to vote. Conversely, at high levels of inequality, an important fraction of the Left’s core constituency is hard to mobilize (Pontusson and Rueda Reference Pontusson and Rueda2010). As their electoral base shrinks, left parties have incentives to adapt their electoral platform to compensate for losses by appealing to moderates rather than poor voters (Anderson and Beramendi Reference Anderson and Beramendi2012), a pattern also supported by research on parties’ welfare positions (Barth, Finseraas, and Moene Reference Barth, Finseraas and Moene2015). Hence, a negative effect of inequality on turnout, particularly among the poor, should weaken the incentives of left parties to respond to high inequality.Footnote 2
Status Quo Bias
Even if we presuppose the visibility and awareness of inequality, status quo bias impedes adequate responses. A large yet contested literature on perceptions of inequality that often starts from the observation that unequal societies do not necessarily imply greater demand for redistribution (Breznau and Hommerich Reference Breznau and Hommerich2019; Trump Reference Trump2023) shows how cognitive heuristics partly reduce demand for redistribution and thereby stabilize high levels of inequality. Such processes can be motivated by system justification, which reflects the desire of individuals to believe that they live in just and fair societies (Jost, Banaji, and Nosek Reference Jost, Banaji and Nosek2004; Trump Reference Trump2018) or by meritocratic beliefs, which lead to the perception that high levels of inequality result from a fair allocation process in which success reflects individual talent, ambition, and hard work (Mijs Reference Mijs2021). Despite varying mechanisms, individuals legitimize levels of inequality (Mijs Reference Mijs2021; Trump Reference Trump2018). People adjust expectations for what fair inequality is to actual inequality.Footnote 3
By contrast, experimental work suggests that individuals react differently to rising inequality. Provided with factual information about rising inequality, individuals are more supportive of redistributive policies in experiments (Kuziemko et al. Reference Kuziemko, Norton, Saez and Stantcheva2015; McCall et al. Reference McCall, Burk, Laperrière and Richeson2017). Thus, while levels of inequality partially self-stabilize cognitively, awareness of rising inequality is linked with increased support for redistribution to counter/correct sudden rises.
This should affect parties’ propensity to mobilize over redistribution and equality. Even if we assume awareness to be given, information about high inequality alone will still not suffice to alter equality preferences among voters and politicians, given their accommodation to and justification of existing inequality levels. Conversely, awareness of growing inequality can alter preferences for redistribution among voters and politicians. Thus, status quo bias makes party responses to levels unlikely, whereas responses to increases remain probable.
Expectations
To summarize, while there is no reason to expect egalitarian responses to the inequality of right parties because advocating redistributive policies promises no electoral rewards for right parties (Gidron and Ziblatt Reference Gidron and Ziblatt2019; Hacker and Pierson Reference Hacker and Pierson2020; Tavits and Potter Reference Tavits and Potter2015), left parties respond to inequality because it affects the relative position and the size of its core constituency. Yet, we do not expect left responses to high levels of inequality. In unequal contexts, attention to the topic is modest, low-income voters that should be susceptible to redistributive policy platforms are hard to mobilize, and voters and parties have (partially) accommodated to the status quo. Given limited attention and demobilized low-income voters, strategic left parties have little incentive to emphasize economic equality. Vice versa, there is greater public awareness about rising inequality and neither barriers to mobilization nor the internalization of the status quo undermine the incentives of the Left to respond to increases.
While we do not and cannot test the direct effect of these rationales on party responses (nor did the reference studies), we posit that they jointly shape the (dis)incentives for (left) party responses to levels and changes of inequality. On this basis, we formulate two expectations.
Expectation 1: Neither left nor right parties are responsive to high levels of inequality.
Expectation 2: Left parties emphasize (class-based) economic equality in response to rising inequality. By contrast, right parties remain unresponsive to rising inequality.
Country Heterogeneity
While our focus is on whether parties react to levels and increases in economic inequality, we recognize that various systemic and contextual conditions are at play. Although the set of 12 countries limits our empirical options, the effects we find may still arise from countervailing dynamics at the country (group) level. To motivate and ground our later discussion of heterogeneous effects, particularly in the analysis and robustness sections, we consider the impact of party system competitiveness, electoral systems, and welfare state regimes.
First, dynamics may vary between different welfare state models—for instance, when comparing Anglo-Saxon (liberal) and European systems. Especially in the universal welfare states of Scandinavia, classic work suggests a preventive approach to inequality, as noted by Svallfors (Reference Svallfors1997). This involves politicizing inequality issues even when overall inequality levels are relatively low, reflecting a commitment to maintaining what Esping-Andersen (Reference Esping-Andersen1990, 27) called “equality of the highest standards” even before inequality rises. In contrast, more unequal Anglo-Saxon countries adopt a more market-based and targeted approach to provision. These contrasting approaches may create different feedback loops (Larsen Reference Larsen2008; Rothstein Reference Rothstein1998): universal welfare programs tend to stabilize broad support for maintaining high equality, while liberal models in Anglo-Saxon countries reinforce existing inequalities through limited provisions and fragmented solidarity. This divergence in approaches could lead to varying responses. To assess these differences, we compare country groups, include country group dummies and consider the party system agenda (Green-Pedersen and Mortensen Reference Green-Pedersen and Mortensen2015, 747).
Second, greater competitiveness of the party system may lead to greater incentives for parties to respond to inequality. We assess the role of new party entries, niche parties, and the number of effective parties (which is usually higher under proportional representation). These factors feature prominently in existing research (Tavits and Potter Reference Tavits and Potter2015), suggesting that competitive environments compel parties, particularly those on the left, to be responsive.
Regarding the electoral system itself, proportional representation likely enhances party responsiveness through increased competitive pressure and the threat of new parties, adding to the positive effects that proportional representation has on redistribution because it leads to more left-leaning and broader government coalitions (Iversen and Soskice Reference Iversen and Soskice2006). For left parties, competition—for instance, from multiple left parties—counteracts the negative impact of high inequality and low turnout on incentives to target low-income voters (Anderson and Beramendi Reference Anderson and Beramendi2012). We return to these contextual factors in the (robustness) analysis.
NEW DATA ON PARTIES’ EQUALITY CONCEPTS VERSUS EXISTING MEASURES OF PARTY RESPONSES
An unresolved limitation that potentially contributes to the previously identified tensions in the literature is the use of an overly broad measure of equality in party manifestos to assess responses to inequality. Assessing whether parties respond to economic inequality requires data that distinguish between emphasis on economic inequality and redistribution versus references to other equally important yet distinct concepts of equality.
The two most relevant studies on party responses to inequality, Pontusson and Rueda (Reference Pontusson and Rueda2010) and Tavits and Potter (Reference Tavits and Potter2015), both faced these data constraints: although both articles are concerned with the changing redistributive profiles of parties in response to inequality, the lack of suitable data made it challenging to discern whether parties indeed respond to inequality with increased emphasis on economic equality and redistribution (as Pontusson and Rueda acknowledge). What is captured is neither an emphasis on redistribution and economic equality (which we will refer to as class-based economic inequalities below), nor a more general economic appeal. Most importantly for our article and in light of debates about the “politics of recognition” and the “politics of redistribution” (Fraser and Honneth Reference Fraser and Honneth2003), the indicators used do not discern parties’ positive references to redistribution and economic equality from statements about recognition and equal rights.Footnote 4 We will first illustrate what is at stake using example statements. Then, we contrast existing indicators with our new data.
To give an impression of the conceptual conflation, we quote selective real-world statements from party manifestos that have been used to establish parties’ responses to inequality (in Tavits and Potter Reference Tavits and Potter2015 and Pontusson and Rueda Reference Pontusson and Rueda2010). However, these statements are difficult to interpret as “redistributive economic appeals” (Tavits and Potter Reference Tavits and Potter2015, 744):
We also support the elimination of school and sports mascots that reflect derogatory stereotypes and that perpetuate racism (Democratic Party, United States, 2016).
The Green Party of Canada believes that it is time to treat Canadians with disabilities with dignity (The Green Party of Canada, Canada, 2008).
People’s differences are not excuses to treat them badly or unjustly (Labour Party, United Kingdom, 2015).
We are the party of inclusion and respect differences of perspective and belief (Democratic Party, United States, 2012).
These statements are not economic equality appeals—in the sense that they are not focused on or related to income or wealth inequality or redistribution. In the studies on party responses to economic inequality (Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015), as well as in the 40 studies on party effects on inequality listed in Appendix D in the Supplementary Information (SI), “economic inequality” refers to income inequality and/or redistribution (Haselmayer and Horn Reference Haselmayer and Horn2024). Rather, the above statements focus on equal rights and (anti)discrimination. Compare this with the examples of what we call class-based economic equality provided below. We use the term class-based economic equality to acknowledge that (un)equal rights can have an economic dimension—irrespective of the dominant use of economic equality for matters of (re)distribution.
Labour will introduce a tax on wealth above £100,000 (Labour Party, United Kingdom, 1974).
Only the top 5 per cent of earners will be asked to contribute more in tax to help fund our public services (Labour Party, United Kingdom, 2017).
In contrast to the previous examples, the latter statements are indeed “redistributive economic appeals” (Tavits and Potter Reference Tavits and Potter2015, 744). The problem is that economic equality statements and equal rights statements are both included in the same indices as responses to inequality.
Naturally, these examples are anecdotal and were merely meant to illustrate the main empirical challenge: to test claims about party reactions to inequality, we must consistently and reliably discriminate between economic and noneconomic equality concepts. Otherwise, we cannot distinguish “redistributive economic appeals” that (left) parties (allegedly) make in response to inequality (Tavits and Potter Reference Tavits and Potter2015, 744) from concerns with equal rights, nondiscrimination, and equal chances. We overcome this challenge by using a reproducible and scalable data-gathering process that combines automatic classifiers with online crowdcoding of specific equality concepts. Analyzing 850,000 programmatic statements from parties, we distinguish positive references to (class-based) economic equality from other equality-related statements and thus speak to the question of whether parties respond to inequality by demanding material equality (refer to the section “Crowdcoding of Equality Concepts in 12 Countries” for details). Drawing on this approach, we elicited new, large-n data that distinguish between different concepts of equality at the party level.
Based on these new data, Figure 1 tracks the emphasis on positive references to equality concepts (relative to the entire emphasis in texts measured at each election). Panel (a) shows the patterns for all parties; Panel (b) contrasts left and right parties. On average, positive references to equal rights and class-based economic equality dominate. As the figure suggests, existing studies did not capture material or class-based economic equality. A comparison with the salience of the economic index from Tavits and Potter (Reference Tavits and Potter2015) (gray area in Figure 1) illustrates the aforementioned multidimensionality issue, especially when we zoom in on left parties. While the economic index (gray) in the reference study points to stability in the salience of redistributive appeals followed by a steep rise, our new data on specific concepts, such as (class-based) economic equality, lead to a different interpretation and point to countervailing trends for the different equality concepts.
Development of Parties’ Focus on Equality, 1970–2020
Note: The colored lines represent different concepts of equality (Varieties of Egalitarianism). The gray area shows the development of the economic position index from Tavits and Potter (Reference Tavits and Potter2015). All the lines are smoothed by fitting a first-order local polynomial with a = 0.7.
In stark contrast to broader and multidimensional measures used in previous studies and in line with what we know about the Third Way of the Left, our economic equality indicator (green line) follows a U-shaped trajectory. From the 1980s to 2000, economic egalitarianism declined to an all-time low, before it started to rise again, while positive references to equal rights (purple line) increased more or less steadily over the 50 years we cover (and the rise is particularly steep for left parties). Therefore, mixing these equality concepts gives the wrong impression of relative stability.
A comparison of the trajectories of left and right parties on the economic dimension suggests that convergence trends peak between the 1990s and the financial crisis (2007/2008) and are—in principle—reversible. The descriptive patterns also challenge the premise that left parties increase attention to equal rights at the expense of class-based equality (on trade-offs and crowding out; Horn, Klüser, and Haselmayer Reference Horn, Klüser and Haselmayer2025). After an initial detraditionalization (less economic equality), the renewed emphasis on redistribution that follows (more economic equality emphasis) is accompanied by a sustained upward trend on equal rights.Footnote 5
Weak correlations between different concepts of equality and between economic equality and indicators such as economic indices and left–right measures (as used in Tavits and Potter Reference Tavits and Potter2015, Pontusson and Rueda Reference Pontusson and Rueda2010, and Polacko Reference Polacko2020) further speak against using broad measures as proxies for redistributive or economic equality responses to inequality (cf. Figure SI-B1). The association between broad indicators and our new more specific (class-based) economic equality measure is 0.32 (for the economic index in Tavits and Potter Reference Tavits and Potter2015) and 0.31 (for the Rile, as in Pontusson and Rueda Reference Pontusson and Rueda2010). Similarly, positive references to equal rights and (class-based) economic equality are only very moderately correlated (0.19).Footnote 6 Therefore, none of these three aspects should be used as a measure to approximate parties’ stances on economic equality and redistribution.
CROWDCODING OF EQUALITY CONCEPTS IN 12 COUNTRIES
Crowdsourced data collection refers to the online collection or annotation of data using multiple judgments from lay coders. Crowdcoding applies this approach to text analysis. It has been presented as an approach to scale expert knowledge for tasks that are (still) too complex for fully automated approaches (Benoit et al. Reference Benoit, Conway, Lauderdale, Laver and Mikhaylov2016) and has been applied to collect and annotate data on party positions, issue salience or sentiment (e.g., Benoit et al. Reference Benoit, Conway, Lauderdale, Laver and Mikhaylov2016; Lehmann and Zobel Reference Lehmann and Zobel2018), and parties’ equality concepts (Horn Reference Horn2019). This article uses a similar approach to gather new data on parties’ emphasis on five concepts of equality: economic equality, equal chances, equal rights, general mentions of equality that remain vague, and a residual category that mostly includes regional, generational, or healthcare-related aspects of equality. In order to reduce the workload for and costs of human coders, we built a text analysis pipeline that fuses humans’ deep semantic understanding with computational efficiency gains. The elicitation of equality conceptions relies on party manifestos, which are regularly published in advance of national elections. In those documents, parties state their goals and vision for the years to come, which renders them a suitable foundation to measure their focus on and understanding of equality. Nevertheless, party manifestos come with their own complications. For instance, it is known that manifestos are not always functionally equivalent across countries and that they have grown in length over time (Hansen Reference Hansen2008). Moreover, it is debatable as to what extent manifestos indeed represent true party positions versus simply being window-dressing for campaign purposes. However, given the lack of data that are comparable in both scope and availability, party manifestos remain among the most widely used sources to tap into party preferences. Thus, we followed this tradition and collected 965 party manifestos across 12 OECD countriesFootnote 7 (1970–2020). Of these, 544 were retrieved from the Manifesto Project (MARPOR) database via their API. The rest were collected manually from different sources such as (online) archives and libraries.
Following MARPOR, we zoom in on quasi-sentences as the units bearing semantic meaning. Thus, raw texts for all manifestos published before 1998 had to be segmented using a binary classifier, as they were not available in digitized form (cf. SI Appendix A for more information). This process segmented all party manifestos into 850,000 quasi-sentences. The next step consists of identifying statements that concern equality. In order to lower the workload for and, hence, costs of human coders, we trained a logistic regression classifier on quasi-sentences that refer to the aforementioned item Equality: Positive). The classifier was fine-tuned to reliably identify the vast majority of relevant statements at the expense of more false positives (we set a minimum level of 0.8 for recall). The (potentially) relevant units were subsequently checked by trained research assistants. This process left us with about 55,451 quasi-sentences that speak to equality and social justice.
To classify these statements into different concepts of equality, we rely on Amazon Mechanical Turk (AMT), a platform that has been used successfully for data collection in research (Berinsky, Huber, and Lenz Reference Berinsky, Huber and Lenz2012; Skytte Reference Skytte2022; Sumner, Farris, and Holman Reference Sumner, Farris and Holman2020) and consistently produces the best results in extensive pre-tests using different platforms. Coders were asked to read short instructions (see SI Appendix A4) and judge whether a statement included a positive reference to equality and equal treatment, and if so, which of the outlined specific categories applied. We also provided the previous statement as context to facilitate valid coding decisions. The pre-selected units (quasi-sentences) were split into tasks of 500–1,000 units and distributed on AMT from February to May 2022. On AMT, we restricted access to contributors with IP addresses from countries with similar geographical and/or cultural backgrounds and control coder accuracy against a benchmark of test statements that have been unanimously coded by three experts. We collected five coding decisions per unit and made sure our workers were able to attain local minimum wages.Footnote 8
In total, 293 workers contributed to the full task, and the modal worker contributed 180 units. To assess their quality, we aggregated individual codings for each unit based on a majority vote approach (De Condorcet [1785] Reference De Condorcet2014). For about 1.5% of units, the crowd did not produce a modal category—that is, an equal number of crowdworkers coded a statement as pertaining to distinct categories. These ties were deemed undecidable by human judgment and excluded from further analysis. Adhering to this process, we obtained a Krippendorff’s alpha of 0.72 between the crowd’s judgment and the experts based on 2,904 coded units, which exhibits very good agreement even by conventional standards of quantitative content analysis (the SI shows country-specific levels of validity with unit-level alpha values ranging between 0.6 and 0.85; see Table SI-A3). These results confirm that crowds are able to replicate expert judgments even for our complex tasks—even at the unit level.
ECONOMIC INEQUALITY: CONCEPTUAL CHOICES
Understanding how inequality shapes party responses requires measures that capture the relative position(s) of different groups rather than overall income dispersion. Theoretical arguments here and in previous research (Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015) emphasize that low(er)-income groups play a central role in shaping redistributive politics, particularly in the context of left-party responses to inequality. While previous studies acknowledge this, data availability constraints have often led researchers to rely on measures that do not reflect the extent to which low-income groups lose ground relative to other groups. They use the Gini coefficient, a measure of overall income dispersion (Tavits and Potter Reference Tavits and Potter2015, 750), and top-income shares, which track the concentration of income among the highest earners (Pontusson and Rueda Reference Pontusson and Rueda2010, 687). Both kinds of measures have clear limitations.
The Gini coefficient, while widely used in studies, is broad and highly sensitive to income shifts in the middle of the distribution (Atkinson Reference Atkinson1970, 256–7). This masks key variations between income groups (Blesch, Hauser, and Jachimowicz Reference Blesch, Hauser and Jachimowicz2022) and does not adequately reflect the relative economic standing of lower-income voters. Similarly, top-income shares, due to their focus on elites, do not indicate whether lower-income groups are falling behind relative to the middle class or the top. While useful for understanding economic concentration at the top, 1% shares do not capture the redistributive pressures that shape parties’ incentives.
Since political responses to inequality are expected to be driven by changes in the relative position of lower-income groups, we argue that below-median-income shares provide the most relevant measure. This measure reflects the share of total income accruing to lower-income groups, directly aligning with the idea that left parties are responsive to the economic grievances of low-income voters. This is our core measure. Additionally, because the below-median share encompasses a broad spectrum of voters, we assess the 90–50 and 90–10 ratios, which capture how those at the lower and upper ends of the below-median group fare vis-à-vis the affluent. Although less theoretically relevant and despite their limitations, we see reasons to also include Gini coefficients and top income shares. In light of debates about the “1%” (Piketty Reference Piketty2020) and the political influence of the affluent (Gilens Reference Gilens2012), it is of interest whether parties respond to top-income shares, though we suspect that gains at the top are less tangible. By contrast, the Gini coefficient, as a rough overall measure that is distinct from the relational measures we focus on, is included only to enable direct comparisons with previous Gini-based results (Tavits and Potter Reference Tavits and Potter2015) and because it remains relevant in public debates.
DATA AND METHODS
Our data include 12 OECD countries representing important variations regarding the role of the state, institutions, and the type of market economy (for our replication data, see Horn, Haselmayer, and Klüser Reference Horn, Haselmayer and Klüser2025). Covering five decades (1970–2020) and varying trends in inequality, our results should paint a broad picture of party reactions to economic disparities in democratic market economies (cf. section below on generalizability).
Dependent Variable
We obtain our dependent variable from the crowdcoding of equality concepts as described above. As we focus on responses to economic inequality, the measure reflects the share of a party’s positive references to economic equality relative to its total manifesto statements in an election. To provide an example, in 2019, the Danish Social Democrats (Socialdemokratiet) devoted 8.4% or 238 of their 2,842 manifesto statements to economic equality, the Conservatives (Konservative Folkeparti) referred to economic equality in 2.5% of their manifesto (28/1,131 statements), while the Socialist Party (Socialistisk Folkeparti) was most attentive to the topic with 11.7% or 84/719 statements. On average, parties in the 12 countries for which we have data emphasized economic equality in 3.2% of their manifesto statements.
Independent Variables
We compile data on all relevant indicators from various datasets for each country in a given year. The SI provides detailed information on these indicators. In line with classic studies on redistribution (Meltzer and Richard Reference Meltzer and Richard1981, 917) and the newer research on party responses to inequality (Pontusson and Rueda Reference Pontusson and Rueda2010, 687; Tavits and Potter Reference Tavits and Potter2015, 749) it inspired, we use market inequality—that is, before taxes and transfers. This helps to ensure that our results speak to the existing studies, the data were more easily available for our broad set of indicators, and we are less at risk of endogenizing redistribution in our explanation.Footnote 9 The SI presents more descriptive information on trends in inequality indicators and party emphasis on economic equality (Table SI-B2). We further provide summary statistics of all variables in Table SI-B2.
Measuring Inequality via Income Shares and Income Ratios
For data on income groups, we rely on the World Income Database (WID), which combines different data sources, such as national accounts, surveys, and fiscal data to compile the most comprehensive data on income levels and distributions available (Piketty and Saez Reference Piketty and Saez2014). We use these data to compute the shares of top and bottom income groups (as discussed above, before taxes and transfers). As argued, most arguments stress that low-income voters are key for (left) party responses. Accordingly, we regard the share of below-median-income earners as the most relevant inequality measure and start our figures and discussions by looking at responses to this measure (bottom 50% share). We complement this with ratios of income groups (to capture how far they are apart). Based on the relevance of the literature, we capture high incomes relative to median income (90–50) and high incomes relative to low incomes (90–10). We also show responses to top-level inequality, the share of the top 1%, which has been prominent in recent research and public debate (Piketty Reference Piketty2020). For the Gini coefficient, we use the Standardized World Income Inequality Database (Solt Reference Solt2020), which compiles and standardizes observations from original data sources (OECD, World Bank, Eurostat, and LIS).
Level Versus Change(s)
In the analysis, we emphasize the differences between levels and changes in inequality indicators. Levels are measured as the level of inequality 1 year before an election. By changes, we mean the annual changes in inequality (obtained by calculating the first differences between t 0 and t −1). Thereby, we focus on levels and changes of inequality that are available and prevalent at the time when parties discuss and publish their manifestos.
Partisanship (Left–Right)
We expect different patterns of responses according to partisanship. For a transparent approach, we obtain this variable from the widely used coding of party families of the MARPOR Manifesto Project (Volkens et al. Reference Volkens, Burst, Krause, Lehmann, Matthieß, Regel, Weßels and Zehnter2021). As is customary, left parties include social democratic parties, green parties, and communist parties. Right parties include liberals, Christian democrats, conservatives, and nationalist parties. Fringe parties (agrarian, ethnic/regional, and special issue) are excluded. This standard coding, which we also confirmed via t-tests to test left–right boundaries, yields an almost balanced share of left (43%, N = 379) and right (57%, N = 496) parties. Following the studies we speak to, partisanship equals 1 in the case of right parties and 0 in the case of left parties. Table SI-B3 provides an overview of each individual party.
Turnout
As discussed, higher inequality has been linked with lower voter turnout of poorer voters (Jensen and Jespersen Reference Jensen and Jespersen2017; Polacko et al. Reference Polacko, Heath, Lewis-Beck and Dassonneville2021; Solt Reference Solt2008; Reference Solt2010). Lower turnout can create disincentives for left responses, which could be less electorally promising if low-income voters do not vote (Anderson and Beramendi Reference Anderson and Beramendi2012). As general turnout is driven by electoral participation of low-income voters (Mahler Reference Mahler2008), we follow the literature and approximate low-income mobilization accordingly (Polacko Reference Polacko2020; Polacko et al. Reference Polacko, Heath, Lewis-Beck and Dassonneville2021; Pontusson and Rueda Reference Pontusson and Rueda2010) using turnout data from the Quality of Government dataset (Teorell et al. Reference Teorell, Kumlin, Sundström, Holmberg, Rothstein, Pachon and Dalli2022).
Controls
Several factors related to differences at the level of parties and countries could drive the extent to which parties make positive references to economic equality. Thus, following Tavits and Potter (Reference Tavits and Potter2015), we consider several controls to reduce omitted variables bias and to improve the precision of our coefficient estimates. At the level of parties, we argue that younger and smaller parties will generally show higher emphasis scores on both inequality dimensions because they strive to differentiate themselves from their more established peers (e.g., Ezrow Reference Ezrow2005; Meguid Reference Meguid2005). Thus, we account for party age (measured in years), party size (a party’s vote share at the national level), and whether a party is a niche party (i.e., ecologist, communist, nationalist party, agrarian, ethnic/regional, and special issue). At the country level, we adjust our models for the effective number of parties in an election (ENP), which may positively affect party responsiveness because the centripetal forces are generally lower in party systems with higher proportionality (cf. Cox Reference Cox1997). Lastly, we also control for economic performance (% GDP growth). The SI contains more details on these variables (SI Appendix B).
Model Specification
To analyze whether parties respond to economic inequality, we estimate linear, mixed-effects regression models. These models leverage within-country variation over time in inequality levels and changes, combined with cross-party differences in economic equality emphasis. This variation is generated by fluctuations in income distributions (e.g., bottom 50% shares and top 1% shares) and the distinct responses of left and right parties in their electoral manifestos, allowing for the assessment of how parties adjust their emphasis on economic equality in response to dynamic inequality patterns. The precise model specification intentionally mimics prior influential studies (Tavits and Potter Reference Tavits and Potter2015), allowing for direct comparisons with our empirical results. Thus, our baseline models of party responses use an interaction of inequality with left–right partisanship as independent variables. To account for variation across parties and countries, we include random intercepts at the level of parties and countries. As discussed in the (robustness) analysis and SI Appendices C and E, the results are robust irrespective of time lags, country fixed effects, period dummies, and outliers. We estimate the following model:
$$ \mathrm{emphasi}{s}_{\mathrm{cpe}}={\displaystyle \begin{array}{l}{\beta}_0+{\beta}_1\mathrm{INEQ}+{\beta}_2\mathrm{INEQ}\times \mathrm{partisanship}\\ {}+\hskip2px {\beta}_3\mathrm{partisanship}+ Z\gamma +{\delta}_c+{\delta}_p+{\unicode{x025B}}_{\mathrm{cpe}},\end{array}} $$
where
$ \mathrm{INEQ} $
is a placeholder for our various inequality measures,
$ \mathrm{partisanship} $
is a dummy variable denoting right parties, and
$ Z\gamma $
represents a matrix of control variables (such as party age, party size, and niche party status) with their associated coefficients (as with the model, the control variables mimic the selection in Tavits and Potter Reference Tavits and Potter2015).
$ {\delta}_c+{\delta}_p $
are country- and party-level random intercepts across which we estimate the respective measures of variance.
RESULTS
We expect that neither left nor right parties respond to higher levels of inequality per se with more emphasis on economic equality (cf. Expectation 1); however, left parties react to increases in economic inequality (cf. Expectation 2). In general, positive values of an inequality indicator are associated with higher inequality. This applies to all inequality indicators but the below-median shares (the measure most likely to capture the theoretically crucial low-income groups): here, a positive value indicates that persons at the lower end of the income distribution dispose of larger income shares, which means less/lower inequality, not more. This also means that the expected direction of the relevant estimated coefficients varies based on the chosen inequality indicator.
Within the framework of our model,
$ {\beta}_1 $
represents the response of left parties to levels or changes in economic inequality. In contrast, the interaction term
$ {\beta}_2 $
denotes the difference in responses between right parties compared to their left peers. Thus, our expectations can be expressed in terms of the central coefficients of our models as follows. Beginning with levels of inequality, we do not expect a significantly positive (below median: negative) effect of inequality (
$ {\beta}_1 $
) among left parties, nor do we expect any significant differences based on partisanship (
$ {\beta}_2 $
). However, we do expect that left parties respond to changes in inequality. In our model, this should translate to a significantly positive (below median: negative) effect of economic inequality (
$ {\beta}_1 $
). However, since right parties are expected to remain indifferent vis-à-vis rising inequality, we expect a significant interaction effect (
$ {\beta}_2 $
) in the opposite direction.
In the remainder of this section, we present the most relevant coefficients from our models (cf. Table 1). Gauging the substantive effect of our estimated coefficients, we examine the significant effects of our estimated models by evaluating the shifts in parties’ emphasis on economic equality resulting from both high and low levels of inequality and high and low values for inequality changes. To remain realistic regarding the fitted values, we use the 10th percentile of the respective inequality measure to represent contexts of low (below median: high) inequality, and the 90th percentile to represent contexts of high (below median: low) inequality. Regarding changes in inequality, the 10th percentile analogously represents a decrease in inequality (below median: increase), while the 90th percentile means inequality increases (below median: decreases). In addition, Figure 2 plots the predicted shares to provide a more intuitive representation of the interaction effects.
Explaining Party Emphasis on Class-Based Economic Equality: Type(s) of Inequality and Partisanship
Note: Standard errors in parentheses; +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.
(Class-Based) Economic Equality Statements Conditional on Partisanship
Note: Based on Table 1. The x-axes show inequality indicators across their empirical range, and the y-axes give the predicted share of manifesto statements on (class-based) economic equality. The shaded areas indicate 95% confidence intervals. The bars show the variable distribution.
Party Responses to Inequality Levels
We start by inspecting patterns related to our first set of expectations by looking at reactions to levels of inequality (Figure 2, left column). Here, we did not expect (left) party responses. A glance at the regression coefficients pertaining to Inequality (
$ {\beta}_1 $
) in Table 1 indicate that left parties de-emphasize (class-based) economic equality at high levels of inequality (except for the Gini). The pattern is easier to spot once we consider the predicted shares of left and right parties’ emphasis on economic equality as shown in Figure 2. The figure’s left panels show party responses at various levels of inequality. The x-axis presents the empirical variation of inequality indicators; the y-axis returns the predicted share of statements for left and right parties at given values of inequality.
As we expect that left parties should be most concerned about below-median groups, our discussion focuses on this indicator. The positive direction of the relevant coefficient (
$ {\beta}_1=0.15,\mathrm{se}=0.08 $
) shows the tendency of left parties to de-emphasize (class-based) equality at higher levels of inequality—an effect that does not hold for right parties as shown by the negatively signed interaction term (
$ {\beta}_2=-0.21,\mathrm{se}=0.09\Big). $
The first panel of Figure 2 visually shows how parties respond to the income share of the below-median group. As evidenced by the insignificant coefficient pertaining to the dummy variable denoting right parties (
$ {\beta}_2=2.44,\mathrm{n}.\mathrm{s}. $
), there is no baseline difference in emphasis on economic equality of left and right parties at high levels of inequality: in contexts of high inequality, that is, when the income share of below-median voters is low, both left and right parties devote about 3% of their manifestos to (economically) egalitarian statements. However, at lower levels of inequality, that is, when the bottom-50 share is at the 90th percentile value, our models indicate a strong partisan divide with left parties putting a much stronger emphasis on economic equality compared with right parties (4.9% vs. 2.1% of statements). This divergence stems from left parties emphasizing economic equality at low levels of income disparities and de-emphasizing it when the disparity is substantial (we return to that aspect in the robustness section).
Looking at top-income shares, the general pattern is similar, albeit not all estimated coefficients are statistically significant (
$ {\beta}_1=-0.12,\mathrm{n}.\mathrm{s}.;{\beta}_2=0.21,\mathrm{se}=0.11 $
). This finding contrasts earlier evidence, which argued that left responsiveness depends on low-income turnout (Pontusson and Rueda Reference Pontusson and Rueda2010). To test this argument, we present additional analyses in the robustness section, where we examine a possible interaction of turnout and party responses (cf. SI Appendix C).
Turning to the different income ratios (90–50 and 90–10), we find that both exhibit the same pattern of party responses observed for the below-median share (Ratio 90–50:
$ {\beta}_1=-0.97,\mathrm{se}=0.52;{\beta}_2=1.26,\mathrm{se}=0.61 $
). At low levels of inequality, left parties spend about 4.8% of their manifestos talking about economic equality, whereas right parties emphasize this issue only half as frequently (2.3%). Yet again, when inequality is high (e.g., when the 90th percentile roughly earns thrice as much as the 50th percentile), this difference vanishes, and both party groups converge at about 2.3% of statements devoted to economic equality. Again, this alignment of left and right responses to inequality is primarily driven by left parties stressing equality less as the ratio between high and medium/low incomes increases. This finding holds if we exclude outliers from the data (see Table SI-C14 and Figure SI-C4, which limit the analysis to 98% of the observations based on the distribution of inequality).
The last panel shows results for the most commonly used Gini indicator. Based on this measure, previous research has shown that left parties were more responsive to higher levels of inequality than parties at the right of the political spectrum (Tavits and Potter Reference Tavits and Potter2015). By contrast, using data on parties’ equality concepts, we observe no reaction from left parties to higher levels of the Gini: left and right parties show clear (stable) differences in promoting egalitarianism, yet these are not related to levels of the Gini
$ \left({\beta}_1=1.91,\mathrm{n}.\mathrm{s}.;{\beta}_2=-1.00,\mathrm{n}.\mathrm{s}.\right) $
. This non-finding mirrors our expectation that the Gini is rather ill-suited to capture party reactions to inequality, as it mostly reflects changes in the middle of the income distribution (Atkinson Reference Atkinson1970, 256–7).
In sum, the results for levels corroborate claims that left parties have “abandoned” low-income groups. This is particularly true where low-income groups are already marginalized compared to the better-off. In such contexts, left and right responses are indistinguishable.
Party Responses to Inequality Changes
Our second expectation posits a different pattern of left responsiveness to rising inequality that threatens the core constituency of low-income voters. Evidence in Table 1 and the left columns of Figure 2 aligns with this expectation. The x-axis presents changes in the indicator.
Again, we start with the below-median share, which should be the most relevant trigger for left-party responses to rising inequality. Importantly, negative values of this indicator correspond to a decrease in their income shares. Our results indicate that left parties put a much stronger emphasis on (class-based) economic equality when the Bottom 50% are losing ground
$ \left({\beta}_1=-0.64,\mathrm{se}=0.27\right) $
. In substantive terms, if the below-median groups of society suffer a 0.7-percentage-point annual decrease in their income share (at the 10th percentile of cases), left parties will devote 4.8% more emphasis on economic equality in their electoral manifestos. Compared to the 90th percentile of cases, where the below-median groups see a slight increase in their income share, left parties cut back their equality emphases by about 15% and by over a standard deviation (considering all parties). Hence, in line with our theoretical framework, left parties respond with more emphasis on economic equality when the rise in inequality is linked to the (decreasing) share of below-median groups. This positive relation does not hold for right parties, where the estimated interaction term essentially nullifies the aforementioned effect
$ \left({\beta}_2=0.99,\mathrm{se}=0.34\right) $
.
Next, we inspect whether the same pattern holds regarding changes in the top 1% of the income distribution. As the relevant coefficients are insignificant
$ \left({\beta}_1=-0,11,\mathrm{n}.\mathrm{s}.;{\beta}_2=0.01,\mathrm{n}.\mathrm{s}.\right) $
, this is evidently not the case. This is also apparent from Figure 2, where there is no political reaction to changes in top inequality. The horizontal lines in the bottom-right panel of Figure 2 show that changes in top-income shares do not inspire more economic egalitarianism in left- or right-party manifestos. Previous research using top-income shares reports stark increases in top inequality and assumes that (left) political parties are not able or willing to respond to trends at the top of the distribution (Piketty Reference Piketty2020). Our results align with this reasoning.
We go on to inspect results for changes in income ratios. In line with our theoretical reasoning, left parties respond with more economic egalitarianism when inequality goes up (Ratio 90–10:
$ {\beta}_1=0.42,\mathrm{se}=0.16 $
; Ratio 90–50:
$ {\beta}_1=1.97,\mathrm{se}=1.14 $
). Between the 10th and the 90th percentiles of changes in inequality as measured by the 90–10 income ratio,Footnote 10 left parties increase the share of economic equality statements in their manifestos by 14.1%. Similarly, looking at the 90–50 ratio, left parties’ emphasis on economic equality increases by 0.3 percentage points from the 10th to the 90th percentiles of cases, which equals an increase of 8% relative to the starting value at the 10th percentile. Flat lines for right parties, which are due to the negatively signed interaction effect (Ratio 90–50:
$ {\beta}_2=-2.18,\mathrm{se}=01.54 $
), visualize the expected nonresponse of these parties.
Regarding changes in the Gini, and in line with our reservations about its suitability, two horizontal slopes indicate that they do not affect the (predicted) share of equality statements
$ \left({\beta}_1=-0.14,\mathrm{n}.\mathrm{s}.;{\beta}_2=-0.08,\mathrm{n}.\mathrm{s}.\right) $
. The sluggish nature of the Gini indicator (typical annual changes are about 0.1 units) renders it demanding to expect that parties take note and react to these marginal changes.
Finally, although we have documented the low/modest correlation (0.19) of class-based economic equality and equal rights above, none of this is to suggest that equal rights do not matter for redistribution. We, thus, also investigate whether appeals to equal rights either directly drive parties’ emphasis on economic equality or are themselves influenced by varying levels and changes of economic inequality. When the dependent variable emphasis on economic equality is replaced with equal rights emphasis, the relevant coefficients become statistically indistinguishable from zero (Bottom 50 (D):
$ {\beta}_1=-0.02,\mathrm{n}.\mathrm{s}.;{\beta}_2=0.05,\mathrm{n}.\mathrm{s}. $
; cf. Table SI-E1.1). These results, documented in Tables SI-E1.1 and SI-E1.2, suggest that emphasis on equal rights itself does not depend on levels or changes in economic equality. Moreover, a test in which we add equal rights as an independent variable to our standard analysis does not change the results discussed previously (Bottom 50 (D):
$ {\beta}_1=-0.64,\mathrm{se}=0.26;{\beta}_2=0.98,\mathrm{se}=0.34 $
; cf. Table SI-E1.2).
Summary of the Results and Robustness
In sum, in contrast to previous research based on multidimensional dependent variables that conflate (class-based) economic and other aspects of equality, our results provide at best partial evidence for the optimistic notion that inequality in democracies is self-correcting.
Examining the impact of inequality levels, we find that right parties hardly ever react to economic disparities. The Left devotes more space to economic equality when low-income groups have “more” (according to ratios and bottom shares)—which suggests that left parties have indeed abandoned the most marginalized groups in contexts of high inequality (levels).
Conversely, left parties are (still) reacting to changes in inequality: when inequality is on the rise, they make more positive references to equality. Finally, much recent research has focused on rising top-level inequality. We show that left parties put less emphasis on equality when the very well-off receive a considerable share of income. Equally worrisome, parties do not react to further gains at the top—which helps to explain why the concentration of incomes at the top increases fast.
The robustness section in SI Appendix C reports tests with various model specifications (including fixed effects and lags), controls (electoral system, redistribution, and decade dummies), party categorizations, observations (excluding fringe parties with less than 2%, excluding outliers, and excluding WID imputed data), SE estimations (SE clustered by countries), and nonlinear interaction terms via kernel regularized least squares. Furthermore, we apply the Romano–Wolf procedure to control the error rate of the various statistical tests. In general, these tests corroborate the results. Below, we discuss whether our results could be due to system-level politicization in equal countries, five additional inequality indicators, living standards, the contexts to which our results apply, and reasons for the 1% exception.
Additional Analyses
Could the Level Result Be Due to Welfare States with Low Levels of Inequality Discussing It More?
When levels yield no effect that can be interpreted as self-correcting (higher inequality, more emphasis), this is in line with our arguments and expectations about unequal contexts. Yet, it may also be that inequality is politicized more in low-inequality contexts such as Scandinavia (Svallfors Reference Svallfors1997). Indeed, the global average emphasis on economic equality is somewhat lower than in our (only) two social democratic welfare states. We implement two empirical strategies: first, we include a dummy for social democratic regimes. This does not change the results (Table SI-E8). Since we only have two social democratic countries, this is only tentative evidence. Thus, we also use party systemic salience—what Green-Pedersen and Mortensen (Reference Green-Pedersen and Mortensen2015, 747) call “system agenda.” Using their exact operationalizationFootnote 11 as a proxy for politicization does not suggest a “preventive” politicization in equal contexts. Controlling for system salience does not change the results (Table SI-E9). Yet, this aspect calls for future research.
Different Inequality Indicators
Table SI-C2 shows regression results based on additional inequality indicators, which perform similarly to the main analysis. For instance, the results for the Bottom 40 share shown in the first column are very similar to those for the Bottom 50/Below Median share, and the Top 10 share is similar to the 1% share. A comparison of Figure SI-C1 with Figure 2 shows similar patterns. Yet, the 60:40 ratio is notable, as the party effect in response to growing gaps between the upper and lower middle classes is very clear. This strong political susceptibility to widening gaps between the low(er) and upper middle classes is plausible in light of our theoretical arguments and work on visible “yardsticks of inequality” (Kevins et al. Reference Kevins, Horn, Jensen and van Kersbergen2018 find that, on the individual level, a key yardstick of inequality is the upper middle class).
Real Wages as Proxy for Living Standards
Real worker wages might have an independent effect on party positions, or, alternatively, they could matter as a complementary factor that should be taken into consideration. Building on the idea that inequality precipitates declines in standards of living at and below the median, we assess this potential channel in Table SI-E2. We use the inflation-adjusted (i.e., real) average production worker gross wage (OECD wage data)—after we confirmed with the official statistics from the Bureau of Labor Statistics and other national agencies that the values are characteristic for non-tertiary educated workers below the mean income—to control for the potential effect of fluctuations in real living standards. First, we test whether controlling for real worker wages changes our results, which, we find, it does not (Bottom 50 (D):
$ {\beta}_1=-0.65,\sim \mathrm{se}=0.27;{\beta}_2=0.99,\sim \mathrm{se}=0.34 $
; cf. Table SI-E2.1). Second, we analyze the extent to which real production worker wages by themselves or interacting with partisanship induce parties to stress economic equality. All tests in Table SI-E2 suggest that this is not the case.
Generalizability, Country Groups, and Time Periods
In affluent democracies with similar cleavage structures (e.g., Norway), we see no specific reason to limit the generalizability—meaning that change versus levels pattern in responses should apply elsewhere in principle. As for Eastern Europe and Southern Europe, more research is needed to determine whether similar dynamics are at play despite the different scope conditions. Given the postcommunist legacy in Central and Eastern Europe and the late democratic and welfare transition in parts of Southern Europe, we would be very cautious.
Within the cases we study, questions of heterogeneity concern the distinctiveness of the more unequal English/Anglo-Saxon democracies vis-à-vis Western and Northern Europe and time periods. Probing into potential period effect, we run our models on both halves of our 50-year time series. In both cases, the results remain substantially the same (cf. the new version of Figure 2 pre- and post-1995 in Table SI-E3.1). We also compare (liberal) Anglo-Saxon welfare states and European welfare states (cf. Table SI-E3.2). These figures are based on Table SI-E6. Both groups exhibit similar results, but in Anglo-Saxon countries, left parties react slightly more to the below-median share, whereas parties in European countries are more sensitive to the 90–50 ratio. Finally, redoing our analysis controlling for whether a country is Anglo-Saxon also confirms our results (cf. Table SI-E7 to Table 1).
What Could Account for the 1% Exception from the Response Pattern?
The response inherent to our framework is that large gains of a small group at the top are less visible, presumably because these gains are disproportionately driven by capital income. Then there is research on policy that suggests that parties are mostly responsive to the rich (Gilens and Page Reference Gilens and Page2014), so the nonresponse could reflect this elite bias. Finally, it is plausible that in not responding to gains at the top, left parties reflect lukewarm egalitarianism from winners of the knowledge economy. This skepticism of highly educated voters regarding redistribution is a reoccurring yet disputed finding (Abou‐Chadi and Hix Reference Abou‐Chadi and Hix2021; Bullock Reference Bullock2022; Gelepithis and Giani Reference Gelepithis and Giani2022; Kiss, Polacko, and Graefe Reference Kiss, Polacko and Graefe2023; Scott Reference Scott2022). Their subjective upward mobility is high (Berger and Engzell Reference Berger and Engzell2020), and they prioritize social investment and competitiveness over soak-the-rich positions (Busemeyer Reference Busemeyer, Landwehr, Saalfeld and Schäfer2022). This could help to explain why the rise of the knowledge economy has been associated with lower taxes for the rich. A related education-based reading is that parts of the Left view elites as fellow professionals with similar education credentials, innovators worthy of political support and meritocratic admiration (Sandel Reference Sandel2020, chap. 4).
CONCLUSION: CONSEQUENTIAL RATCHET EFFECT RATHER THAN DEMOCRATIC SELF-CORRECTION
Our point of departure for this article was countervailing perspectives on parties’ responses to economic inequality. How can we reconcile results according to which left parties respond to inequality (Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015), the more pessimistic broader political science literatures on the end of class politics and unequal democracy, the popular critique(s) that “the Left” has abandoned the less well-off, and upward trends in inequality?
Based on new data on parties’ focus on economic equality, we show that these views are not incompatible if we differentiate between levels of inequality and rising inequality. To measure party responses, we compiled new data on parties’ equality concepts, which enabled us to distinguish a focus on economic equality from other equality concepts in 12 OECD countries between 1970 and 2020. We further proposed three rationales with different implications for party responses to levels and changes of inequality. Moreover, we made the qualification that measures capturing inequality with a focus on low-income groups are best suited to capture theories about (left) responses to inequality. Our conclusions are threefold.
First, there is no positive—at times even a negative—programmatic response to high levels of inequality, which seems to corroborate views that the Left has abandoned its core constituency (Piketty Reference Piketty2020; Sandel Reference Sandel2020). This reflects self-reinforcing mechanisms of high (levels of) inequality as observed in studies of public visibility, voter demobilization, or individual-level heuristics that help to justify inequality (McCall Reference McCall2013; Solt Reference Solt2008; Trump Reference Trump2018). It is particularly noteworthy that the Left de-emphasizes rather than emphasizes the topic when the lower-income groups are most in need of help. Overall, our findings support the notion that higher (levels of) inequality undermines (undermine) left egalitarianism.
Second, there is a positive left response to changes in inequality: in line with a more “optimistic” perspective (Pontusson and Rueda Reference Pontusson and Rueda2010; Tavits and Potter Reference Tavits and Potter2015), we show that left parties respond. Yet, contrasting previous evidence, the Left only reacts with positive references to economic equality when income disparities are rising; reflecting expectations that rising inequality is more salient and creates electoral incentives for the Left to respond.
Third, as if these findings did not have enough problematic implications, we observe limits in party responsiveness to rising inequality. There is no reaction to growing top-level inequality. Although rises in the share of top income earners figure prominently in recent research, they are disregarded by parties—helping us to understand why inequality at the top grows particularly fast. This suggests that left responses to rising inequality and the absence of responses to rising levels of top inequality are not a contradiction. Left responsiveness simply does not extend to growing top-level inequality. The fact that this qualification vanishes when the “merely” well-off 90th percentile is pitched against the median or low incomes indicates that party reactions are “denominator-driven”: as expected, left parties are more likely to react to (below) median incomes than to (more) insulated top shares.
Overall, our findings suggest a “ratchet-effect” heuristic: left parties may still push back against rising disparities but have given up on lowering existing levels of inequality. To us, the findings imply lock-in effects that should also be reflected in policy patterns and (pre)distribution profiles. While we can only focus on party manifestos here, the question of whether a ratchet-effect pattern describes policy and inequality outcomes is an important avenue for future research. Finally, while our results are robust to potential confounders, we find no responsiveness for top(-heavy) inequality, leaving little hope for the kinds of (fiscal) egalitarian responses advocated by Piketty (Reference Piketty2020), the Occupy movement, and others.
As for future research, we hope our party data will help us and others specify scope conditions—for instance, regarding the changing education and income profiles of voters or oft-overlooked mediatization—under which parties pursue certain pathways to (in)equality.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://doi.org/10.1017/S0003055425100907.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/DPNEKE.
ACKNOWLEDGEMENTS
We are grateful to Jonas Pontusson, Kris-Stella Trump, Marius Busemeyer, Nadja Wehl, Peter Hall, Kees van Kersbergen, David Rueda, Carsten Jensen, Christian Breunig, Jane Gingrich, Paul Marx, and Silja Häusermann for comments, suggestions, and encouragement during various stages of this project. This article also benefited from feedback at the 2022 APSA Annual Meeting, the 2022 ECPR General Conference, the 2022 Inequality Workshop at the University of Oxford, and the Colloquium of the Inequality Cluster in Konstanz. Felix Bäckstedt, Olivia Debout, Ida Marie Lindskoug, Mathieu Meier, Sophia Reuter, Lukas Röder, Simon Rittershaus, Anna-Lena Scheer, Benjamin Maderstein, Luke Hosford, Jonas Engel, Jana Dolle, and Alexandra Sorge provided fantastic research assistance. Kattalina Berrichoa, Isabelle Guinaudeau, Anthony Kevins, and Ann-Kristin Kölln gave great feedback on the coding instructions, and Simon Persico and Wolfgang Müller shared manifestos from their data collections. We extend our gratitude to the old and new editors and the fair reviewers.
FUNDING STATEMENT
This research was funded by the Emmy Noether Research Group “Varieties of Egalitarianism” (A.H.; Grant No. 428250727, Deutsche Forschungsgemeinschaft).
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors declare the human subjects research in this article was reviewed and approved by the University of Konstanz and and certificate numbers are provided in the appendix. The authors affirm this article adheres to the principles concerning research with human participants as laid out in APSA’s Principles and Guidance on Human Subject Research (2020).
Open access
Comments
No Comments have been published for this article.