Party Positioning Under Populist State Leaders

Abstract

The implications of rising parliamentary representation of populist parties have been thoroughly studied but little is known about the impact of populist state leaders on party positions. In this article, we study mainstream parties' strategic responses when a populist takes over as the leader of a nation. We use content-analytical data and large language modelling to measure positions expressed in manifestos from parties from 51 democracies between 1989 and 2018. Employing methods for causal inference from observational data, we find that right-wing populist state leaders induce mainstream parties to differentiate their positions on multiculturalism, possibly leading to polarization of the party system. Under left-wing populist leaders, mainstream parties adopt more homogenous or differentiated positions, depending on the policy category and other contextual factors. Parties are generally more responsive in emerging than advanced countries and in presidential than parliamentary systems.

Populist forces have been on the rise during the past decades, to the point that many countries are currently governed by a populist state leader, such as Viktor Orbán in Hungary, Narendra Modi in India, and Recep Erdoğan in Turkey. The reasons for the success of populists are diverse, including citizens' worries about globalization, rapid technological changes, and rising inequality. The potential repercussions of populist influence are also diverse, ranging from a diminished quality of democratic institutions over higher corruption to decreased macroeconomic performance (for example, Guriev and Papaioannou 2022). A central question in the literature is how mainstream parties react when populist parties gain strength. In many situations, mainstream parties pursue accommodating strategies, adopting positions that are similar to those of populist parties. For instance, mainstream parties may introduce proposals to restrict immigration in the hope of winning back voters lost to radical right parties (for example, Abou-Chadi and Krause 2018; Han 2015; van Spanje 2010).

The analysis of party manifestos – sometimes called party platforms or election programs – is central in this research, as parties use these documents to communicate what issues they prioritize and what positions they take (Harmel 2018).¹ In turn, the range of issues and positions collectively covered by parties participating in an election determines how well a party system represents voters' diverse preferences. The strategic positioning of parties also has implications for legislative and executive processes; for instance, by affecting the chances that parties coalize or find majorities (for example, Ezrow 2007).

As detailed in the next section, the effects of representation of populists in the legislative branch of government on party positions have been thoroughly studied. However, little is known about the impact of populists with executive power. This is an important gap because the implications of legislative v. executive influence for party competition are fundamentally different, especially when a populist takes over as the leader of a state.² While it is important to understand the behaviour of individual parties, our focus is on the impact of populist state leaders on the positions collectively offered by mainstream parties³ participating in an election. Specifically, we study the distribution of party positions, considering that the diversity of positions in a country has first-order implications for voter representation and collective decision-making. We therefore ask: What are the effects of populist state leaders on the distribution of positions offered by mainstream parties?

To answer this question, we analyze party manifestos on 215 elections in 51 countries between 1989–2018. We focus on two policy categories that are particularly relevant under populism: questions related to equality/welfare and multiculturalism. In the literature, measures of party positions are often constructed by counting statements pertaining to a certain position, based on annotations of election manifestos available in the Manifesto Project Database (Lehmann et al. 2023). The degree of differentiation between two parties is then measured as the difference in position counts. Considering that this ‘traditional’ approach has been criticized for a lack of nuance (for example, Akkerman 2015; Protsyk and Garaz 2011), we propose a novel approach to capture positions based on semantic differences between party manifestos. Relying on methods from computational linguistics, this approach allows us to capture differentiation strategies not only in terms of what positions a party has (as in the ‘traditional’ approach) but also how the party proposes to implement a position. Intuitively, with this ‘semantic’ approach, we assume that two parties propose differentiated strategies if relevant statements in their election manifestos are semantically dissimilar.

Estimating the effect of populist state leaders on mainstream parties' positions is challenging due to reverse causality and confounding factors. We address the issue of reverse causality by modelling the impact of the incumbent state leader on party positions. That is, while past behaviour of mainstream parties may affect the selection of the state leader, the selection of the incumbent cannot be influenced by party positioning in the future. To minimize threats due to confounding factors, we use matching techniques and panel data models with time and country fixed effects. Essentially, our research design facilitates a comparison of elections⁴ where public sentiments and political and economic conditions are similar, but where some elections take place with a populist as the incumbent state leader and others with a non-populist leader. For example, assume that two countries have similar levels of unemployment, inflation, and economic prosperity; that their democratic institutions are of equal quality; and that populist parties enjoy similar levels of parliamentary representation. However, for idiosyncratic reasons – such as different electoral rules or majority-minority constellations – one country selects a populist as the state leader while the other does not. In this setup, the populist treatment can be assumed as if randomly assigned.

Our results indicate that right-wing populist state leaders induce mainstream parties to differentiate their positions on multiculturalism, with a tendency towards polarization of the party system. Responses to left-wing populist leaders are context-dependent and may lead party systems to become more homogenous or more differentiated. Overall, the effects are stronger in emerging rather than advanced countries (likely because political structures are less rigid in the former) and more pronounced in presidential than parliamentary systems (likely because the former type allows for more concentration of power).

We contribute to the literature in several ways. First, we complement studies investigating the reactions of mainstream parties to populist forces in parliament, as detailed in the next section. This article differs in that we examine the impact of populist state leaders on parties' differentiation strategies, which allows us to address an important gap in the literature. In addition, while much of the literature focuses on the reactions of individual parties, we study the impact on the party systems as a whole, and hence the diversity of positions available to citizens. Finally, we contribute to the methodology of measuring party positions by using state-of-the-art models for multilingual text analysis to construct and validate nuanced measures of differentiation of positions between parties. These ‘semantic’ measures capture how parties propose to implement a position and, therefore, add another dimension to characterizing positions.

Populism, Parliamentary Influence, and Executive Power

Common definitions of populism rely on three elements. Populists claim to represent the will of the people, they position themselves as outsiders who challenge the establishment, and they aim for a restoration of popular sovereignty (for example, Mudde 2004). Some scholars emphasize that populists have a distinct approach to communicating with the public, where negativity, emotional appeals, exaggeration, and confrontational language play a central role (for example, Nai 2021; Widmann 2021). Populists do not have a common programmatic direction but can be found on the left and right sides of the political spectrum (Stavrakakis et al. 2017). On the left side, populists typically emphasize issues related to economic inequality and redistribution through taxes and public spending. Their agendas tend to focus on economic dimensions, often featuring anti-capitalist and anti-globalization views, finding a main enemy in financial institutions and financial elites. By contrast, right-wing populists often identify ethnic, cultural, and religious minorities as threats to a nation's identity. Consequently, their agendas usually emphasize anti-immigration appeals and demands for cultural homogeneity and integration (Vachudova 2021).

When populists gain parliamentary representation, they can influence legislation, policy debates, and the formation and stability of coalitions. However, they lack the executive authority to implement their positions directly. Populists in the legislature have a small degree of agenda-setting power by bringing issues to the forefront that may have been ignored by mainstream parties. Parliamentary debates may become more emotional (for example, Valentim and Widmann 2023) and more polarized (for example, Atzpodien 2022). In line with the theory of spatial political competition (Budge 1994; Downs 1957), the electoral success of populists may induce mainstream parties to change their policy positions – to regain lost vote shares (for example, Abou-Chadi and Stoetzer 2020; Breyer 2023; Schumacher et al. 2016), to adjust to changed voter preferences (for example, Adams et al. 2004; Ezrow et al. 2010; Klüver and Spoon 2016), or to strategically respond to the behaviour and positions of the populists (for example, Adams and Somer-Topcu 2009; Williams 2015). For example, the emergence of right-wing populist or radical parties may cause mainstream parties to adopt more restrictive positions on immigration (Abou-Chadi and Krause 2018; Chou et al. 2021; Han 2015; Schumacher and van Kersbergen 2016; Van Spanje 2010) or to reposition themselves in matters of welfare policy (Krause and Giebler 2020). In some cases, mainstream parties may decide to simply ignore emerging populists; for instance, when populists have a small vote share and are not perceived as a threat (Meguid 2005).

The chances that mainstream parties adjust their policy positions are arguably much higher when populists gain executive power, especially when a populist takes over as the state leader. There are at least three elements of why the effects of populist state leaders v. populists in parliament on mainstream parties differ: executive power, agenda-setting power, and the threat of institutional deterioration. First, populist state leaders may implement their positions directly; for example, via presidential orders; through their right to initiate legislation; or by re-structuring, closing, or creating governmental agencies. Second, compared to parliamentary influence, state leaders have considerably more resources and opportunities to address the public and set the agenda. For instance, the media tends to devote more attention to the government – and especially the state leader – rather than the opposition (for example, Garz and Sörensen 2021). Hence, while it may be an option for mainstream parties to ignore populists in parliament, the agenda-setting power of the state leader could invalidate this option when a populist takes over this office. Third, many populist state leaders act in an authoritarian fashion to preserve their power, which may involve the suppression of opposition and civil society actors, the exclusion of independent experts, the restriction of free media (Huber and Schimpf 2017; Juon and Bochsler 2020), and erosion of central bank independence (Gavin and Manger 2023). Consequently, the overall quality of democratic institutions and processes declines (for example, Bellodi, Morelli, and Vannoni 2024; Peffley and Rohrschneider 2024; Waldner and Lust 2018), while corruption and nepotism increase (Sasso and Morelli 2021; Szeidl and Szucs 2021). While populists may also erode democratic institutions through parliamentary influence, the institutional deterioration under populist state leaders poses a particularly strong threat to democracy. Considering that mainstream parties usually value democracy, they face additional pressure to adapt under populist state leadership.

In summary, with populist state leaders, the pressure for mainstream parties to reposition themselves is arguably much higher when populists are merely represented in parliament, due to the state leader's executive power, agenda-setting power, and the threat of institutional deterioration. This pressure is likely highest in areas that are at the core of populists' agenda, including equality/welfare in the case of left-wing populists and multiculturalism in the case of right-wing populists. These areas could be decisive in the upcoming elections, due to the salience of related topics to the electorate when a populist is in power. Hence, repositioning themselves on issues related to equality/welfare and multiculturalism may be the key for mainstream parties to remain relevant in the media and win back voters. We discuss the options to adjust their positions below.

Strategic Responses to Populism and Implications for Party Systems

Understanding the system-wide impact of populist state leaders requires a discussion of the strategy options that individual mainstream parties have. They may decide to ignore the agenda of the populist (dismissive strategy), converge towards this agenda (accommodating strategy), or differentiate from it (adversarial strategy).⁵ Collectively, these strategies could lead to a homogenization or differentiation of choices available to voters in a party system.⁶ Party competition, inherently, is a dynamic process (Budge, Ezrow, and McDonald 2010; Laver and Sergenti 2012), implying that parties may change their positions before a populist takes over as the state leader; for example, when populists gain vote shares in parliament. However, we limit the theoretical discussion to possible adjustments once a populist acts as the incumbent state leader to keep the argument simple and in line with our empirical design, and because adjustments of party positions to populists gaining strength prior to entering office have been previously discussed in the literature (for example, Krause and Giebler 2020; Schumacher and van Kersbergen 2016).

Panel A of Fig. 1 illustrates the case where the mainstream parties of a country implement dismissive strategies to maximize their chances in the upcoming election, implying no change in positions and hence no change in the degree of diversity of choices they collectively provide. As argued above, this strategy is unlikely to be applied in practice due to the strong pressure that mainstream parties experience when a populist takes over as state leader.

Figure 1. Theoretical effects of populist state leaders on party positioning.

Notes: The figure refers to a fictional party system with an incumbent populist state leader, where three mainstream parties (a, b, and c) implement different strategies aimed at maximizing votes in the upcoming elections. The dots denote the location of parties' positions on a policy issue in a spectrum of hypothetical choices. A larger distance between two dots indicates a greater level of differentiation between parties. The arrows indicate adjustments in policy positions between the previous and the upcoming elections. The position of the incumbent populist state leader (and their party) is the one observed by the mainstream parties during the previous elections and assumed to be fixed for the sake of simplicity.

Panel B of Fig. 1 depicts a scenario where mainstream parties implement an accommodating strategy, resulting in homogenization of the party system. Accommodation is an often-used response in situations where right-wing parties gain parliamentary influence and could be a relevant strategy in our context. When a populist is the incumbent state leader, mainstream parties may attempt to win back votes by converging towards the positions that characterize the populist's agenda. If that strategy dominates among all parties participating in an election, their positions should, overall, become more similar, leading to a homogenization of choices available to voters.

Panels C and D of Fig. 1 illustrate scenarios in which mainstream parties use adversarial strategies to position themselves for the upcoming election. We argue that parties' incentives to use an adversarial strategy are particularly strong when the incumbent state leader is a populist. Once populists take executive responsibility, they usually face the dilemma that their anti-establishment views are not compatible with effective governance (Adams et al. 2022). That is, populists may have to abandon (parts of) their rhetorical strategy, or they may fail to provide solutions to pressing issues. Either strategy comes at the risk of losing voter support. Given this dilemma, mainstream parties have strong incentives to work on ‘demystifying’ the populist leader by making broken promises salient, pointing out unresolved problems, and emphasizing the inadequacy of the populist's policy approach – all combined with their own proposals for alternative, better policy options.

The implementation of adversarial strategies can have different implications for the availability of choices in the party system. Panel C of Fig. 1 depicts the case where mainstream parties increase the distance from the populists while converging with each other. This scenario could be relevant if parties believe that the chances of having the populist voted out of office are best when they join forces; for example, because they believe that they reach a necessary majority only by combining their vote shares. In that case, the choices available to voters would be more homogenous.

However, adversarial strategies could also lead to greater diversification of choices (Fig. 1, Panel D). That is, mainstream parties may not only seek to distance themselves from the populist, but they may also choose a position that is sufficiently distinct from that of their mainstream competitors. This scenario could be particularly relevant for complex issues that offer plenty of space for differentiation, such as immigration. For instance, if the incumbent populist is in favour of closing the border to all refugees, the mainstream party (a) a may suggest welcoming all refugees if they apply for asylum while still being abroad, party (b) could propose to open the border for political refugees but not economic migrants, and party (c) may offer to prioritize highly educated migrants.

Finally, it is conceivable that a mix of accommodating and adversarial strategies leads to polarization of the party system (Fig. 1, Panel E). In that case, mainstream parties distance themselves from the incumbent populist while adopting positions at the extreme ends of the spectrum. This scenario does not necessarily imply that voters have more choices but that the choices are concentrated in two poles; that is, where the options available in a party system are presented in a binary manner (either/or, for/against).

Overall, the discussion shows that the effect of a populist state leader on party positions is theoretically ambiguous. It is conceivable that a combination of some or all of the possible strategies applies. For instance, some parties may decide to converge to the populist while others distance themselves; some parties may want to collaborate while others prefer to adopt unique positions. Hence, it is necessary to empirically investigate whether and in what situations any of the scenarios shown in Panels A to E dominate. Specifically, it is useful to evaluate whether a party system is more likely to become homogenous or differentiated under a populist leader. The reason is that the level of diversity of choices collectively provided by a party system has implications for the representation of citizens' preferences, political participation, and trust (Campbell and Heath 2021; Ezrow 2007). For example, overly homogenous systems offer little diversity of policy alternatives, in which case there could be a disconnect between parties and voters. Highly polarized systems are not optimal either, as parties may have difficulties finding a consensus, or may not be able to work with each other because of a lack of agreement about fundamental issues. Normatively speaking, it is optimal if the parties collectively offer diverse policy choices (Benson 2023). Collective decision-making may be more complicated in that case due to the difficulties of finding majorities and forming coalitions. Voters may also suffer from choice overload. However, those challenges are secondary, whereas representation and a diversity of choices are of primary importance. Hence, a party system that offers a variety of choices can be considered optimal from a normative perspective (Powell 1982; Sartori 1976).

Panel F illustrates how position differences between parties can be empirically operationalized. For example, it is possible to calculate the absolute difference Y of position statements y between pairs of parties i and j (that is, Y _i,j = |y _i − y _j|). The mean difference over all pairs of mainstream parties n participating in an election would then yield a system-level measure of the dispersion of positions (Andrews and Money 2009):

(1)$$\bar{Y} = \displaystyle{1 \over n}\mathop \sum \limits_{\,j = 1}^n \mathop \sum \limits_{i = 1}^n \vert {y_i-y_j} \vert $$

Low values of $\bar{Y}$ indicate a homogenous distribution of positions in the party system, whereas high values reflect a differentiated distribution.

Data

To evaluate the effects of populist state leaders on party positions, we compile a dyadic panel dataset that compares mainstream parties participating in the same election, with repeated observations from 215 national and presidential elections between 1989 and 2018 in 51 democracies around the world: see Table A1 in the Supplementary Information for a full list. Using party pairs as the unit of observation allows us to evaluate both the impact of populist state leaders on the party system at large (as described in Equation 1) and the distribution of positions between individual parties. The included countries and investigation period are determined by the availability of parties' election manifestos with content-analytical data in the Manifesto Project Database, as well as the availability of complementary data described below. Using observations on as many countries and years as possible gives us the best chance to obtain generalizable results, and it allows us to test for differences between political systems and countries' states of development.

Traditional Measures of Party Positions

We focus on manifestos to evaluate parties' differentiation strategies rather than parliamentary debates, campaign speeches, or press releases because manifestos are less likely affected by current events. These documents tend to describe the core topics and ideas pertaining to parties' medium- to long-run ideologies (Garz and Szucs 2023; Harmel 2018). Importantly, the manifesto texts provided by the Manifesto Project Database cover a large number of countries over long periods, in line with our goal to draw generalizable conclusions about the impact of populist state leaders on party positions. Our sample includes all manifestos for which human content annotations are available.⁷ After excluding parties with an average vote share of less than 5 per cent (cp. Abou-Chadi and Krause 2018), the data include 416 parties and 1,005 manifestos.

We analyze the content of the manifestos in two policy categories: equality/welfare and multiculturism. We choose these categories because they are of central importance to populists' agenda. Left-wing populists tend to focus on a reduction of (economic) inequality, especially by expanding the welfare state via increased taxation and public spending. By contrast, most right-wing populists emphasize the need for cultural homogeneity and restrictions on immigration due to the perceived problems resulting from multiculturalism. In addition, the categories equality/welfare and multiculturism generally tend to be the ones in the focus of parties' attention – judging from the corresponding shares of statements tagged in the Manifesto Project Database – and they are important categories previously analyzed in the literature on party positions (for example, Abou-Chadi and Krause 2018; Krause and Giebler 2020; Schumacher and van Kersbergen 2016).

In this literature, party positions can be measured by subtracting the number or share of negative statements in a category from the number or share of positive statements in that category. For instance, if a manifesto includes many statements highlighting the importance of schools, hospitals, and unemployment benefits, researchers typically assume that this party is in favour of expanding the welfare state. Similarly, if the annotations indicate many positive references to multiculturalism, researchers may conclude that the party favours a multicultural society.

To capture positions regarding equality/welfare, we subtract the number of quasi-sentences tagged by the Manifesto Project Database as 505 (welfare state limitation) and 507 (education limitation) from the number of quasi-sentences under the codes 503 (equality: positive), 504 (welfare state expansion), and 506 (education expansion).⁸ The resulting position measure covers topics that are at the core of the agendas of left-wing populists. For example, a central aspect of Evo Morales' agenda was the ‘redistribution of wealth and political power’ (Madrid 2019, 165), Cristina Fernández de Kirchner's rhetoric approach focused on defending ‘the interests of the common man against the privileged elite’ (Acemoglu, Egorov, and Sonin 2013, 771), while Jacob Zuma considered himself a ‘champion of economic equalization’ aiming for ‘material redistribution’ (Comaroff 2011, 101).⁹

To construct the position measure on multiculturalism, we subtract the percentage of quasi-sentences tagged as 608 (multiculturalism: negative) from those tagged as 607 (multiculturalism: positive). The former category captures favourable ‘mentions of cultural diversity and cultural plurality’, while the latter category refers to favourable mentions of ‘cultural integration’ and ‘cultural homogeneity’ (Lehmann et al. 2023). Hence, the balance of both categories refers to parties' support for multiculturalism v. cultural protectionism, a core issue on the agendas of right-wing populists (for example, Mudde and Rovira Kaltwasser 2013).

Semantic Measures of Party Positions

The coding scheme of the Manifesto Project does not assess how parties propose to implement their positions or what their specific policy approach looks like. For instance, the annotations may indicate that two parties are in favour of expanding education, but one party proposes to strengthen universities while the other one prefers to hire more elementary school teachers. Similarly, there could be cases where the annotations suggest that two parties are in favour of multiculturalism but one party promises to foster the recruitment of high-skilled workers from abroad while the other party wants to legalize undocumented immigrants. Hence, the annotations in the Manifesto Project Database face limitations when evaluating parties' differentiation strategies (cp. Akkerman 2015; Protsyk and Garaz 2011).

For that reason, we re-analyze the content-analytical data of the Manifesto Project Database by using automated text analysis. The basic idea is to compare the statements made by a party in a given category (for example, equality/welfare) with the statements of another party in the same category. If the content of their statements overlaps, both parties can be assumed to share similar ideas and policy approaches. By contrast, if there is little overlap between statements, the parties offer different views in that category. Automated text analysis has been widely used to compare political texts. For example, a growing literature on political media bias relies on analyses of textual similarities to show that a media outlet and a political party share the same ideology if they use similar language (for example, Garz and Rickardsson 2023; Gentzkow and Shapiro 2010; Martin and Yurukoglu 2017).

We implement the comparison by modelling the semantic similarity of manifesto statements using Bidirectional Encoder Representations from Transformers (BERT), similar to the approach proposed by Ceron, Nikolaev, and Padó (2023). Transformer models tend to outperform previous approaches on most natural language processing tasks, such as text classification, named entity recognition, part-of-speech tagging, and multilingual analysis (Devlin et al. 2018; Licht 2023). We use a multilingual version of Sentence-BERT designed to evaluate the semantic similarity of texts when different languages are involved. The model is trained in more than 100 languages and supports text comparison both within and across languages (Reimers and Gurevych 2020). The ability to make cross-language comparisons is important in our context, considering that parties in some countries compete in more than one language (for example, Belgium). In addition, the model was trained to catch the meaning of a statement independent of phrasing, which implies that our measure of differentiation of positions primarily captures differences in substance rather than rhetorical differences or different writing styles. The advantage of using a transformer model over other methods used for text comparison (for example, word2vec) is that context-specific meanings of words are retained. That is, the transformer model generates individual vectors of embeddings for the same word when used in different contexts (for example, ‘house key’ and ‘key of a song’).

We retain the full text of all statements (‘quasi-sentences’) falling in the policy categories under consideration (that is, equality/welfare and multiculturalism), as tagged by the coders of the Manifesto Project. These statements (225,291 in total) form a text corpus of 1,963,930 words. We use the multilingual Sentence-BERT to extract the word embeddings and create numerical representations of the statements, on which we compute the cosine distance for all possible combinations of statements between all pairs of mainstream parties participating in the same election.¹⁰ For each policy category, we then compute the mean cosine distance between party pairs and rescale the resulting variables to vary between 0 and 1. Values closer to zero indicate that the statements made by a party in a policy category overlap with the statements made by another party in the same category, whereas values closer to 1 reflect little overlap.¹¹

We verify the validity of the approach by qualitatively evaluating randomly selected manifesto pairs from those 19 countries¹² with languages spoken among the two co-authors using the method of close reading (for example, Grimmer and Stewart 2013). We confirm that the cosine distance based on the Sentence-BERT model facilitates an accurate comparison of semantic overlap both for individual pairs of statements and larger manifesto parts. For the sake of brevity, we illustrate the results of this exercise with English-language manifestos only.

Tables 1 and 2 compare random manifesto paragraphs with relatively similar and relatively different content, respectively. For example, Table 1 shows statements about equality and welfare by the Democratic and Republican parties from their manifestos before the 2004 United States elections. Both parties generically talk about the importance of educating children, yielding a relatively low value of the cosine distance (0.44) for these paragraphs.

Table 1. Examples of manifesto paragraphs with relatively similar content

Notes: The cosine distance is based on word embeddings extracted via a pre-trained S-BERT sentence transformer model. The measure is bound between 0 and 1. Larger values indicate low levels of overlap.

Table 2. Examples of manifesto paragraphs with relatively different content

Notes: The cosine distance is based on word embeddings extracted via a pre-trained S-BERT sentence transformer model. The measure is bound between 0 and 1. Larger values indicate low levels of overlap.

As an opposite example, Table 2 shows two paragraphs related to multiculturalism by the same parties in 2012. Here the Democratic Party talks about immigration, its benefits for the country, and the need for policy reform in that area. By contrast, the paragraph from the Republican Party manifesto discusses questions related to Indigenous Americans, thus addressing a completely different topic. Accordingly, the cosine distance between both paragraphs is relatively high (0.78). Tables 1 and 2 provide further examples from other countries and policy categories, all suggesting the same result: If two parties have similar proposals, the cosine distance takes low values, whereas it takes high values when two parties present different proposals.

Table A2 illustrates how the semantic measure of position differentiation differs from the traditional position measure. The table lists all statements in the welfare state category by the New Zealand National Party and Green Party of Aotearoa New Zealand in their 2011 manifestos. According to the traditional measure – which refers to the percentage of positive statements minus the percentage of negative statements – both parties would be assumed to have almost identical positions in that area as they are both in favour of expanding the welfare state. By contrast, with a value of 0.83 (that is, close to 1), the average cosine distance indicates that both parties are fairly differentiated in the matter. The reason is that the New Zealand National Party primarily talks about the pension system, whereas the Green Party of Aotearoa New Zealand discusses various aspects of the welfare state, including health, education, housing, and support of local communities.

As another example, Table A3 lists all statements in the multiculturalism category by Canada's New Democratic Party and the Conservative Party from their 2015 manifestos. The traditional measure again indicates very similar positions, whereas the average cosine distance (0.64) suggests a moderate to high level of semantic differentiation. Looking at the statements confirms that there is little overlap between the parties: while both of them pledge their support for Indigenous languages and communities, the Conservative Party – but not the New Democratic Party – talks about problems related to immigration (for example, honour killings, forced marriages) and requirements for Canadian citizenship.

Populist State Leaders

For each of the 215 elections in our dataset, we check whether the incumbent state leader appears in the database of populist leaders compiled by Funke, Schularick, and Trebesch (2023). This database categorizes state leaders between 1900 and 2018 in 60 countries as right-wing populist, left-wing populist, or non-populist based on a systematic literature review of 770 sources. The definition applied to compile this database identifies state leaders as populist if they claim to be the sole representative of ‘the people’ in a fight against dishonest elites, similar to Mudde's (2004) characterization. We prefer to use this database over other populism databases because the classification criteria are suitable for both presidential and parliamentary systems as well as advanced and emerging countries. Importantly, the data covers more countries and longer periods than alternative sources (for example, Hawkins et al. 2019; Kyle and Meyer 2020), which maximizes the number of observations in our analysis dataset.

Using this database, we identify a total of 11 populists acting as incumbent president or prime minister in 22 elections: Boyko Borisov, Recep Tayyip Erdoğan, Lech Kaczyński, Carlos Menem, Benjamin Netanyahu, Viktor Orbán (all right-wing populists), Robert Fico, Cristina Fernández de Kirchner, Evo Morales, Aléxis Tsípras, and Jacob Zuma (all left-wing populists): see Table 3 for details. We retrieve their party affiliation to match them with the manifesto data and create binary treatment variables that tag those elections where the incumbent state leader is a right- or left-wing populist.

Table 3. Populist leaders in the sample

Notes: The sample excludes those elections and incumbent populist leaders where annotated manifestos are not available.

Political, Economic, and Institutional Conditions

To capture a country's overall political and economic climate at the time of an election, we use data from the World Bank's World Development Indicators Database on GDP per capita (at constant 2015 USD), inflation (annual growth rate of the GDP deflator in per cent), and unemployment (percentage of the total labour force). Following Funke, Schularick, and Trebesch (2023), we use the first principal component of the Varieties of Democracy (V-Dem) indices on judicial independence, election fairness, and media freedom as a proxy for the quality of institutions. We use these macroeconomic and institutional indicators to capture potential drivers of populism (Guriev and Papaioannou 2022). It would be desirable to include variables that capture further potential drivers, such as income and wealth inequality, trade patterns, technological change, migration flows, and geopolitical tensions (for example, Borwein et al. 2024; Milner 2021; Remmer 2012). Unfortunately, data on these and other potentially relevant variables are not consistently available. We therefore calculate the aggregate populist vote share (overall populist parties participating in an election) using data from the Global Populisms Project Database. This variable captures public sentiments towards populism independent of underlying drivers. Finally, we create a binary variable that takes the value 1 for party pairs where one party provides a non-populist incumbent state leader, and 0 otherwise. This variable accounts for the possibility that parties modify their strategy when dealing with the state leader's party (populist or not).

Empirical Strategy

Estimating the effect of populist state leaders on mainstream parties' positions is challenging, due to confounding factors and reverse causality. For instance, the broader patterns in a nation's political and economic landscape as well as voter preferences may simultaneously influence the choice of its state leader and parties' strategies. Similarly, party positioning in earlier elections may affect the choice of the current state leader. For example, populists may emerge if mainstream parties fail to represent their citizens (Kriesi 2014) or if mainstream parties collude and use public resources to collectively survive (Katz and Mair 1995).

We tackle these challenges as follows. First, we address the issue of reverse causality by investigating the impact of the incumbent state leader on party manifestos. While the past behaviour of mainstream parties may affect the selection of the state leader, the selection of the incumbent cannot be influenced by party positioning in the future. Second, we estimate the effect of populist state leaders on positions in a panel model with two-way fixed effects. Including country fixed effects helps to account for the overall tendencies of countries to elect a populist state leader, as well as baseline differences in the degree of homogenization/differentiation of the party system (for example, due to the configuration of electoral rules; see Cox 1990). Time fixed effects capture global trends in populism over time. Third, and most importantly, we use propensity score matching (Rosenbaum and Rubin 1983) to find ‘statistical twins’ for countries with an incumbent populist state leader.¹³ For each treated party pair, we find one or more untreated party pair(s) with a similar propensity score based on observable factors related to populism; that is, economic conditions (GDP, inflation, and unemployment), the quality of institutions, and public sentiment towards populism (proxied by populists' parliamentary vote shares).

The aggregate parliamentary vote share of populist parties is a particularly powerful matching variable because it allows us to compare elections under similar public sentiment conditions towards populism, but where one election is treated with a populist incumbent state leader and the other is not. This approach is similar in spirit to Abou-Chadi and Krause's (2018) approach, who use a regression-discontinuity design to compare parties that have similar vote shares but differ in their representation in parliament due to electoral thresholds. A regression-discontinuity design is not possible in our context though because the selection of the state leader is not always based on a quantifiable assignment rule but the result of negotiations between certain parties, which is why we rely on a matching approach.

We provide a detailed description of the matching procedure in the Supplementary Information. In short, we first use a probit model to estimate the impact of the political, economic, and institutional variables mentioned above on the likelihood of observing an incumbent populist state leader in a given election, yielding propensity scores for each treated and control observation. We then use kernel matching to compute weights that minimize the observed differences between treated and control groups.¹⁴ As Table 4 shows, prior to matching, we observe significant differences in means of most variables between treated and control observations. After the matching, these differences disappear (Table 5), which suggests that the matching yields a dataset that is balanced on observables.

Table 4. Summary statistics (pre-matching)

Notes: Treated observations are those with a populist state leader, while untreated observations are those with a non-populist state leader.

Table 5. Differences between treated and untreated observations after matching

Notes: N = 2,083. Matching based on propensity scores.

We use the matched dataset to estimate the impact of incumbent populist state leaders on the distribution of positions of mainstream parties in a given policy category:¹⁵

(2)$$Y_{i, j, e, c, t} = \alpha _1Pop_{e, c, t}^{left} + \alpha _2Pop_{e, c, t}^{right} + \alpha _3L_{i, j, e, c, t} + \theta _c + \vartheta _t + \varepsilon _{i, j, e, c, t}$$

where Y _i,j,e,c,t is the difference in positions between a pair of parties i and j participating in election e in country c and year t. Pop ^left and Pop ^right are binary treatment indicators. The coefficients α ₁ and α ₂ measure the average treatment effect of left- and right-wing incumbent populist state leaders, respectively, on position differences over treated party pairs. Considering that the treatment takes place at the election level (that is, all parties participating in an election face the same incumbent state leader), the treatment effects refer to the party system as a whole. That is, the coefficients α ₁ and α ₂ quantify the difference in the dispersion of positions between treated and untreated elections, which can be interpreted as changes in an election-level measure of dispersion similar to $\bar{Y}$ in Equation 1 in the theory section. A negative sign of α ₁ and α ₂ implies a more homogenous distribution of positions, whereas a positive sign indicates the opposite. The binary indicator L takes the value 1 for party pairs with a non-populist state leader, and 0 otherwise. θ _c and $\vartheta _t$ are country and year fixed effects.¹⁶ We compute standard errors that are clustered by election – consistent with the level of treatment – to account for the correlation between party pairs participating in the same election (Cameron and Miller 2015).¹⁷

With matched data, the treatment effects from Equation 1 capture reactions by mainstream parties to populist incumbent state leaders in addition to any changes in positions that may have been made prior to or independent of that, such as responses to populists gaining strength in parliament. For example, matching the aggregate parliamentary vote share of populist parties implies that we compare situations where mainstream parties have similar incentives to adjust their positions, but where some observations are treated and others are not.

Results

Main Results

Estimation results are summarized in Table 6. Accordingly, we do not find any significant effects of incumbent populist state leaders on the distribution of mainstream parties' positions regarding equality and welfare (Column 1). These null effects are consistent with dismissive ignorance (Panel A in Fig. 1) or a combination of strategies shown in Panels B to E in Fig. 1, where diverse reactions of parties cancel each other out.

Table 6. Impact of populist leadership on party positions

Notes: Based on matched data on 2,083 party pairs. The dependent variables capture the degree of differentiation between pairs of parties in the policy categories stated in the column headers, measured as the average cosine distance between parties' manifesto statements pertaining to the categories. All models include country fixed effects, year fixed effects, and binary variables indicating party pairs with a state leader (output omitted). Standard errors (in parentheses) are clustered by election.

*p < 0.10, **p < 0.05, ***p < 0.01.

However, we find a marginally significant effect (at the 10 per cent level) of left-wing populist state leaders on positions regarding multiculturalism. According to Column (2), the difference in positions between mainstream parties decreases by 0.249 points, which corresponds to an effect size of 0.249/0.445 = 0.56 standard deviations. This effect is consistent with (accommodating or adversarial) homogenization, as shown in Panels B and C in Fig. 1. Column (2) further indicates that right-wing populist state leaders cause mainstream parties to differentiate their positions on multiculturalism, compared to matched elections with a non-populist incumbent state leader. This effect is significant at the 1 per cent level and amounts to 0.295/0.445 = 0.66 standard deviations. The increase in differentiation is consistent with adversarial diversification (Fig. 1, Panel D) and polarization (Fig. 1, Panel E).

To get a sense of whether party systems become more diversified or polarized under right-wing populist state leaders, we plot the distribution of differences in positions on multiculturalism in Fig. 2. The figure offers visual support that positions tend to be more polarized in elections with right-wing populist incumbents than in matched elections with non-populist leaders. That is, when the incumbent state leader is a right-wing populist, the distribution tends to be bimodal, with one cluster of party pairs with relatively small differences in positions and another cluster where differences are relatively large. This pattern is more compatible with polarization rather than diversification because the distribution of positions in a diversified system would be unimodal and comparatively flat.

Figure 2. Distribution of party positions on multiculturalism (a) Left-wing populists. (b) Right-wing populists.

Notes: The graphs show the distribution of position distances between party pairs, using the residuals from regressions that are identical to Equation 2 but without treatment indicators. Smaller values on the x-axis indicate more similar positions between parties, whereas larger values reflect more differentiated positions. The solid lines are Kernel density plots.

Robustness Checks and Sensitivity Analysis

The results remain robust across a wide range of additional checks that test the sensitivity of our baseline findings, as reported in the appendix. Table A4 summarizes estimation results when using traditional instead of semantic position measures. The estimates in Columns (1) and (2) are based on the same set of observations as used in Table 6, whereas the estimates in Columns (3) and (4) are based on an additional sample where the traditional measures, but not the semantic measures, are available. The estimates from both samples confirm the baseline estimates. However, the effect sizes tend to be smaller when using the traditional position measures. For example, according to Column (2), the negative effect of left-wing populist leaders on the differentiation of positions on multiculturalism amounts to 0.628/4.094 = 0.15 standard deviations (v. 0.56 standard deviations in the case of the semantic position measure), and the positive effect of right-wing populist leaders equals 1.200/4.094 = 0.29 standard deviations (v. 0.66 standard deviations in the case of the semantic position measure).¹⁸ Considering that the traditional position measures are based on counting positive and negative statements, while the semantic measures account for differences in meaning, the smaller effect sizes imply that parties not only differentiate on what positions to take but also more subtly on how to implement a position.

The results are robust to further modifications of the baseline specification. In Table A5, we evaluate the sensitivity of the estimates to our assumption that party pairs have identical positions if neither party makes any statement in the relevant policy category (cosine distance = 0) and that a pair is maximally differentiated if one but not the other party expresses a position on an issue (cosine distance = 1). In Columns (1) and (2), we use semantic position measures where 0s and 1s are replaced with the values of the 1st and 99th percentiles of the measures. In Columns (3) and (4), we replace 0s and 1s with the 10th and 90th percentile values. The resulting coefficient estimates remain qualitatively similar.

Table A6 presents results when we use an alternative text-analysis approach for the construction of the position measures. Rather than using a multilingual version of Sentence-BERT for the computation of cosine distances, it is possible to first obtain machine translations of the non-English manifestos, and then use the monolingual, English Sentence-BERT to create our position measures (cp. Licht and Lind 2023). The estimates do not noticeably change, which helps to cross-validate our semantic position measures.

In Table A7, we present results based on alternative matching approaches; that is, inverse probability weighting and entropy balancing. The estimates are fairly similar to the baseline specification, which indicates that our results are not sensitive to the choice of matching method. Table A8 shows that we obtain similar results when we do not winsorize the matching variables but use their original values. In the baseline specification, we cluster the standard errors by election, at the same level as the treatment assignment. Table A9 reports standard errors clustered by country, clustered by party pair, and not clustered but robust to heteroskedasticity, respectively. The size of the standard errors remains similar when using these alternative computations. Table A10 shows that the results do not substantially change when we add indicators for countries' development status (advanced v. emerging) or political system (parliamentary v. presidential) to the set of matching variables.

In Table A11, we evaluate the results when using an alternative definition of populist state leaders. Specifically, we reclassify any populist as non-populist if at least one of the comparison lists used by Funke, Schularick, and Trebesch (2023) considers them as non-populist (that is, Edwards 2019; Hawkins et al. 2019; Kyle and Meyer 2020; Magud and Spilimbergo 2021). With this approach, Boyko Borisov, Robert Fico, Cristina Fernández de Kirchner, and Aléxis Tsípras are reclassified as non-populist. The results in Table A11 indicate that the impact of right-wing populist state leaders on the distribution of mainstream parties' positions on multiculturalism remains similar. By contrast, the effect of left-wing populist leaders disappears, which is not surprising considering that the reclassification reduces the number of left-wing populist leaders from 5 to 2.

According to Table A12, we obtain qualitatively similar results when we collapse the data to the level of treatment; that is, run the regressions at the election level – with mean differences in positions as dependent variables – rather than at the level of party pairs. Finally, Figure A2 summarizes the results of a placebo exercise that involves 1,000 regressions with randomly created ‘fake’ treatment dummies. For both policy categories, the distribution of coefficients from these placebo regressions is centred around zero, which offers reassurance that the actual treatment effects are no statistical artefact.

Context-Specific Effects

Historically, populism has been primarily considered a phenomenon in emerging economies. More recently, there also have been trends towards populism in advanced economies. Previous research shows that parties' reactions to political and economic conditions may differ in both types of countries (Guriev and Papaioannou 2022). As our sample includes both advanced and emerging economies, it is therefore useful to investigate whether populist state leaders have differential effects on party positions, depending on a country's level of economic development. For that purpose, we categorize the countries in our sample as either advanced or emerging, utilizing the classification provided by the International Monetary Fund.

Figure 3 summarizes the estimation results when using the semantic position measures. According to Panel A, we find that the positions of mainstream parties in emerging countries regarding equality and welfare become more homogeneous when the incumbent state leader is a left-wing populist. As Panel B shows, the effects of populist state leaders on positions regarding multiculturalism tend to be somewhat stronger in emerging rather than in advanced countries. An explanation for the differential effects could be that parties in emerging countries are often younger and more prone to change rather than parties in advanced countries.

Figure 3. Differential effects of populist state leaders, by countries' development status. (a) Equality and welfare. (b) Multiculturalism.

Notes: The graphs show regression results from specifications similar to Equation 2 but where we interact the treatment indicators with binary indicators for advanced v. emerging countries using matched data. All effects are relative to the reference category – elections with non-populist incumbent state leaders. The dependent variables are semantic position measures. The spikes denote 95 per cent confidence intervals, based on standard errors clustered by election.

Another useful split refers to potential differences between parliamentary v. presidential democracies, as the different systems may facilitate different levels of concentration of power for the state leader. In parliamentary systems, the cabinet and legislators tend to work relatively closely with each other, whereas there is more independence in presidential systems. This independence may allow a populist president to dominate the national discourse, which in turn may induce differential responses by mainstream parties. We categorize the countries in our sample as parliamentary or presidential, based on the Database of Political Institutions provided by Scartascini, Cruz, and Keefer (2021).

Figure 4 shows corresponding estimates based on semantic position measures. According to Panel A of the figure, the positions of mainstream parties on equality and welfare become more homogeneous in presidential democracies under right- and especially left-wing populist leaders. Regarding multiculturalism (Panel B), both left- and right-wing populist leaders in presidential systems induce mainstream parties to differentiate their positions. Overall, these findings suggest that parties' differentiation strategies are more responsive to populist leadership in presidential rather than parliamentary systems.

Figure 4. Differential effects of populist state leaders, by countries' political system. (a) Equality and welfare. (b) Multiculturalism.

Notes: The graphs show regression results from specifications similar to Equation 2, but show where we interact the treatment indicators with binary indicators for the parliamentary v. presidential system, using matched data. All effects are relative to the reference category – elections with non-populist incumbent state leaders. The dependent variables are semantic position measures. The spikes denote 95 per cent confidence intervals, based on standard errors clustered by election.

Conclusion

This article analyzes the impact of populist state leaders on the positions of mainstream parties. For that purpose, we compile a comprehensive dataset on elections, party manifestos, state leaders, and political and economic conditions in 51 democracies between 1989 and 2018. Focusing on the policy categories equality/welfare and multiculturalism, we complement the traditional approach of measuring party positions with a novel approach that quantifies nuanced differences in positions based on a semantic analysis of manifestos with large language models.

Using panel data and matching techniques, we find that right-wing populist incumbent state leaders induce mainstream parties to differentiate their positions on multiculturalism, with tendencies towards polarization. This effect occurs across countries and political systems but is more pronounced in emerging rather than advanced countries, and it is stronger in presidential rather than parliamentary systems. The effects of left-wing populist leaders on positions are context-specific. Depending on a country's political system and development status, these state leaders may induce mainstream parties to adopt more homogeneous positions (in questions related to equality/welfare) or more differentiated positions (in questions about multiculturalism). Comparing results between traditional measures of party positions and our semantic measures, we generally find stronger effects when using the latter, which suggests that parties respond more by adjusting how they propose implementing a position rather than adjusting what positions to advocate.

These findings are not without limitations. First, as with any method for causal inference from observational data, our matching approach facilitates a causal interpretation of the results only if the identifying assumptions are met. In our case, identification of the causal effect hinges on including the correct observables for the matching and that no unobserved confounders drive both the selection of state leaders and the content of party manifestos. Unfortunately, it is not possible to test this assumption, nor is it possible to compile complementary evidence based on a randomized controlled trial in our case. Second, our analysis is limited to the policy categories of equality/welfare and multiculturalism. Arguably, these are the most relevant categories in our context, but we encourage future research to expand the investigation to other areas. Third, our results do not speak to the reactions of individual parties but pertain to the distribution of positions in a country's party system as a whole. For example, our empirical design does not support a distinction between adversarial and accommodating homogenization. Further research is necessary to investigate in which areas and under what conditions a populist state leader causes positions of mainstream and populist parties to converge or diverge. Fourth, populist state leaders may cause parties to reposition themselves by modifying their manifestos, but the availability of positions offered to voters may also be affected by the entry and exit of parties. Unfortunately, our data does not support analyzing this channel, as manifestos of parties that participate in an election but fail to receive sufficient votes are not consistently included in the Manifesto Project Database. Future research may leverage alternative data sources to study parties' entry and exit.

While it is important to understand the circumstances under which populist state leaders induce party systems to become more homogenous or more differentiated, it is not always clear what the optimal level or form of differentiation of positions across parties is. More differentiation could be beneficial for voter representation, but too much differentiation could create a choice overload in the electorate and complicate collective decision-making due to the difficulties in finding majorities. Hence, the impact of populist state leaders on the distribution of party positions is not generally ‘good’ or ‘bad’, normatively speaking. An important exception is our finding that positions on multiculturalism are more polarized under populist leaders. There is ample evidence that polarization may harm democratic processes and impose high costs on society (for example, Benson 2023). In this regard, our results add further evidence of the harmful consequences of populism.

In contrast to parliamentary speeches or media interviews, positions expressed in party manifestos do not reflect short-run reactions by mainstream parties but are indicative of lasting decisions to adopt new positions. Most parties go through lengthy internal processes – including discussions among members and functionaries – before approving any changes at a board meeting or party convention. Thus, the impact of populist state leaders described in this article is not short-lived but alters party systems in the long term.