Political parties often adopt ambiguous and inconsistent positions, a phenomenon that spans various countries, party systems, and time periods. This practice can limit voters’ knowledge of the policies potential leaders intend to implement if elected, thus posing a considerable challenge to democracy.

The primary objective of this study is to delve into the factors driving such electoral behavior, with a specific focus on a relatively intuitive yet insufficiently explored aspect: the degree of extremism or centrism inherent in political party ideologies. We specifically investigate the relationship between ambiguity and the ideological positioning of political parties across the political spectrum.

Our main finding is the identification of a non-monotonic (inverted U-shaped) relationship between party ideology and ideological blurriness (or ambiguity)^Footnote1 within a sample of 202 European political parties. Specifically, our results indicate that political parties with a centrist ideology tend to possess a more blurred ideology. This relationship holds true for each ideological dimension covered in the available data. Furthermore, the relationship remains robust across different econometric specifications, providing strong evidence of its existence not only within countries but also within parties. While this study is descriptive in nature and a perfect identification strategy is challenging given the research question’s nature, the broad range of correlational results, along with some instrumental variable (IV) estimates, provides support for the interpretation that the ideological position influences the degree of blurriness.

We also offer an explanation for this pattern and present additional empirical evidence to support it. Our explanation builds upon Glazer (1990) but expands on it. It is based on the notion that centrist political parties are perceived as less risky by voters compared to extremist parties. One possible reason is that policies proposed by extremist parties are often untested and unconventional, whereas centrist parties often propose more established policies. If the median voter is risk-averse, this suggests that centrist parties enjoy an ex ante advantage over extremist parties. Consequently, if ambiguity facilitates parties in aligning with the median voter, centrist parties have stronger incentives than extremist parties to strategically utilize ambiguity for electoral success.

To strengthen the plausibility of our explanation, we analyze additional data on electoral outcomes and economic indicators from the respective countries of the studied parties. Our analysis reveals that in countries where extremist parties have recently held power and experienced significant economic fluctuations, centrist parties not in government tend to adopt more ambiguous positions. This finding aligns with our theoretical framework, which suggests that incumbent extremist parties in such contexts should be perceived as riskier, thereby reducing the potential benefits of ambiguity for them.

Furthermore, we explore the empirical plausibility of two alternative explanations for the observed non-monotonic inverted U-shaped relationship between ideology and political position. Firstly, we examine whether this relationship could be attributed to the presence of single-issue parties that adopt centrist and ambiguous policies to attract a broader voter base, given that those policies are not their primary focus (e.g., Rovny 2012; Somer-Topcu 2015). Secondly, we investigate whether this relationship can be explained by the greater interest that centrist parties may have in having more flexibility while in office to choose policies that better deal with new information not available during elections (e.g., Aragones and Neeman 2000; Kartik, Van Weelden and Wolton 2017). Our empirical analyses, however, fail to provide support for either of these alternative explanations.

This paper contributes to an extensive body of literature investigating the relationship between ambiguity and the electoral behaviour of political parties. This literature can be categorized into two groups. First, there is a wide range of theoretical studies exploring why political parties may choose to be ambiguous in presenting their electoral programs to voters. These studies consider factors such as voters’ risk aversion or the intensity of their preferences (Shepsle 1972; Alesina and Cukierman 1990; Aragones and Neeman 2000; Aragones and Postlewaite 2002; Laslier 2006), the importance of maintaining flexibility in office (Aragones and Neeman 2000; Kartik, Van Weelden and Wolton 2017), the significance of electoral competition in multi-party systems (Bräuninger and Giger 2018), context-dependent voting patterns (Callander and Wilson, 2008), the influence of policy-motivated donors (Alesina and Holden 2008), and whether political parties possess information about the median voter’s position (Glazer 1990). While the primary contribution of this paper is empirical, it connects with this theoretical literature in two ways. Firstly, it provides a theoretical explanation for the main empirical result by extending Glazer (1990)’s model. Secondly, the paper explores whether certain theoretical models from this literature can account for the primary empirical result.

In addition to the theoretical literature, there is a vast body of empirical research examining position blurring by political parties. For instance, Han (2020) used data on party positions and public opinion on major political issues in Western Europe and found that, in polarized environments, political parties present clearer positions on the issues they primarily focus on, but less clear positions on secondary issues. Bräuninger and Giger (2018) estimated ambiguity from electoral manifestos and found evidence that platforms become more ambiguous as the preferences of two key stakeholders, the voting public and the party’s core constituency, diverge. Some studies have also utilized experimental data and found evidence that ambiguity can be a winning strategy (Tomz and Van Houweling 2009) and that it is more popular among non-centrist voters when one of the candidates is a known centrist (Tolvanen, Tremewan and Wagner 2022). While these papers offer valuable evidence regarding the factors influencing position blurring by political parties, they do not directly address how this blurring relates to the ideological position of parties on the political spectrum. Only a few papers, to our knowledge, Rovny (2012, 2013); Lo, Proksch and Slapin (2016), and Praprotnik (2017), have explored this relationship directly, making them closely related to our study.

Rovny (2012, 2013) explores how political parties strategically employ ambiguity in elections where multiple issues are at play. Rovny argues that parties tend to emphasize dimensions where they hold extreme positions while blurring their stance on others. Rovny supports this theory with cross-sectional empirical evidence from over 100 political parties across 14 countries in 2006. In comparison to Rovny, although our paper also examines the relationship between ambiguity and centrism and finds an inverse relationship between the two, it differs on three important points: it expands the analysis to cover various time periods, more parties, and more countries; it proposes much more robust econometric evidence regarding the existence of this relationship, and it presents an alternative theory that does not focus on the notion of single-issue parties, which we argue is consistent with more robust and broader evidence.

Praprotnik (2017) examines ambiguity in electoral competition in Austria and suggests that extremist parties, being niche parties, are penalized for presenting vague and ambiguous programs. In contrast, centrist parties, not being niche parties, have more incentives to be ambiguous. Praprotnik links this hypothesis to the government status of parties, arguing that government parties have incentives to decrease clarity in their campaign strategies compared to parties in opposition. Consequently, extremist parties in government might be equally or even more ambiguous than centrist parties in opposition. In comparison to Praprotnik, our paper also examines the relationship between ambiguity and centrism, but expands the analysis beyond Austrian political parties and niche extremist parties.

Lo, Proksch and Slapin (2016) propose a method to estimate the ideological clarity of political parties based on party manifestos, and apply it to 74 parties in four countries. They find a positive correlation between their estimates of ideological clarity and the level of ideological extremism of political parties in those countries. They suggest that centrist parties may find ambiguous positions advantageous because they can appeal to a larger segment of the electorate located in the center of the political spectrum. This study expands upon Lo et al.’s research by including more countries and a larger sample of over 200 parties, resulting in more robust findings. Additionally, we explore alternative explanations and provide empirical evidence to support a new explanation that complements and extends the previous work.^Footnote2

The paper follows the following structure: Sect. 2 provides a comprehensive overview of the data utilized in the study and outlines our empirical approach. Section 3 presents the key findings of our analysis. In Sect. 4, we put forth our primary explanation for the observed results. Section 5 then explores various alternative explanations. Lastly, Sect. 6 concludes.

2.1 Data

This paper’s analysis primarily relies on the Chapel Hill Expert Survey (CHES) dataset (Jolly, Bakker, Hooghe, Marks, Polk, Rovny, Steenbergen and Vachudova 2022), which provides party position data on ideology for numerous national parties across various European countries.^Footnote3 The CHES dataset incorporates information from multiple surveys conducted in 1999, 2002, 2006, 2010, 2014, 2017 and 2019. These surveys involve a consistent assessment of party positions by a substantial number of experts, encompassing general left-right ideology, economic left-right orientation, and social values (GAL-TAN).^Footnote4 The first set of variables employed in this study is derived from these assessments of party positions. In the more recent surveys (2017 and 2019), the CHES introduced a series of questions to gauge the extent of ambiguity in the establishment of these positions by political parties. The second set of variables used in this study is based on these measures of blurriness.^Footnote5

Given our focus on examining the relationship between ideology and blurriness, our analysis is restricted to the years 2017 and 2019. We merge the available data on party positions in the economic and social values dimensions with information on the associated degree of blurriness. The resulting sample consists of 202 political parties, spanning across two time periods. This comprehensive dataset incorporates assessments from a minimum of 25 experts, covering at least two ideological dimensions.

All the data in this study is available at the expert-party-year level. In our primary analysis, we aggregate this information by averaging across experts for the same party-year. However, all our results remain robust when examined at the individual expert-party-year level.^Footnote6 The information on party positions is measured on a scale ranging from 0 to 10. In the economic dimension, a value of 0 represents an extreme left position, while a value of 10 indicates an extreme right position.^Footnote7 In the social values dimension, a value of 0 corresponds to a pro-libertarian/postmaterialist stance, whereas a value of 10 represents a pro-traditional/authoritarian perspective.^Footnote8 Similarly, the data on blurriness is also measured on a scale ranging from 0 to 10. A value of 0 indicates no blurriness at all, whereas a value of 10 signifies extreme blurriness.^Footnote9

In addition to the main dataset, we incorporate supplementary databases to gain deeper insights into the underlying mechanisms that drive our primary findings. These additional databases comprise country-level data on economic outcomes, such as GDP per capita sourced from the International Monetary Fund (IMF), as well as party-level data on electoral outcomes, including the seat-share of parties in each election obtained from the ParlGov project (Doring, Huber and Manow 2022). A more detailed description of these databases will be presented in subsequent sections of the study.

2.2 Empirical strategy

In our initial econometric specification, we model the outcome , which represents the average expert opinion on the level of blurriness in the political position of party p regarding issue i in country c during year (or wave) t, as

where  is the average opinion of experts on the position of party p regarding issue i in country c during period t.  refers to the square of this position. The term  are country  year fixed effects, and  is the error term.^Footnote10 As we will explain later, the inclusion of  is crucial in our identification strategy, as it allows us to control for various country-related factors that may change over time, such as population, quality of national institutions, characteristics of party systems, and political polarization. We also introduce political party fixed effects in certain specifications to account for party-related factors that remain constant over time. Including these fixed effects greatly strengthens the robustness of our results, indicating that the relationship exists not only within countries but also within parties.^Footnote11

The coefficients of interest are  and , which capture the relationship between the parties’ positions on each issue and the perceived blurriness of these positions by experts.

To facilitate the analysis of mechanisms, we also estimate different versions of the following equation:

where  represents the difference between 5 and the level of extremism of party p on issue i in country c during period t. Extremism is measured as the absolute value of the difference between 5 and the average expert opinion on each party’s position. The coefficient of interest, , captures the relationship between the level of ideological centrism of party positions on each issue and how experts perceive the blurriness of these positions.^Footnote12

This paper is descriptive in nature, and achieving a perfect identification strategy is challenging given the research question’s nature. Specifically, the models presented in Eqs. (1) and (2) may not establish causal effects due to at least two endogeneity concerns.

Firstly, there could be omitted factors driving the association between blurriness and party position. While we control for any time-varying variables at the country level and include party fixed effects, certain time-varying variables at the party level might still influence the results. For instance, the level of internal dissent within each party could be one such variable: centrist parties may experience more internal dissent, and internal dissent may lead to more blurred positions. Additionally, the age of a party might matter, with older parties possibly adopting more centrist positions and having less blurred stances. Government status is another potential factor: centrist parties might be more frequently in government, and governing parties may be rewarded for presenting clear ideological stances. In the next section, we show that the main results remain robust even after controlling these potential time-varying party-level factors, which helps alleviate the first concern.

Secondly, a more critical issue is the potential presence of simultaneity bias in the models of Eqs. (1) and (2). This bias could occur if the perception of a party’s position as blurred (by experts) increases the likelihood of identifying that position as centrist. To address this concern, we adopt an instrumental variables (IV) approach by using lagged values of position (or centrism) as instruments. The effectiveness of this estimation strategy hinges on two primary assumptions. Firstly, the relevance assumption, which we verify with our data. Secondly, and more importantly, the assumption that conditional on fixed effects and controls, lagged values of position (or centrism) affects blurriness only through their contemporaneous values (i.e. the exclusion restriction assumption). Although our IV strategy does not resolve all endogeneity concerns, we think it provides a reliable approach as long as the lagged values of position do not influence blurriness through unaccounted time-varying party-level factors. In addition to providing supplementary evidence that aligns with this assumption, it is important to highlight that we could not identify any such factors that could undermine the validity of this IV strategy (given the substantial number of controls and fixed effects included).

Figure 1 presents the relationship between experts’ opinions on each party’s position on economic issues (Fig. 1a) and social values (Fig. 1b) and the perceived blurriness of these positions. Both figures exhibit a clear non-monotonic inverted U-shaped pattern. Notably, the peak of these curves occurs around the midpoint of the ideological position distribution (approximately 5), which corresponds to centrist political parties. Complementing these results, Figs. 2a and b show a strong positive association between the level of centrism of each party’s position (as defined in the previous section) and the level of blurriness of these positions, consistent with the non-monotonic inverted U-shaped relations found in Figs. 1a and b.

Columns (1) to (6) in Table 1 provide the estimates of Eqs. (1) and (2) for the outcomes analyzed in Fig. 1 (Panel A) and 2 (Panel B). Columns (1) to (4) report the estimates of Eq. (1), showing positive and negative coefficients for  and , respectively. All coefficients are individually and jointly statistically significant at conventional levels. Additionally, the peak points of the curves fall between 4.1 and 5.3, well within the range of ideological positions (from 1 to 10). These results confirm the inverted U-shaped relationships observed in Figs. 1a and b. Columns (5) and (6) present the estimates of Eq. (2), showing positive and statistically significant effects, which is consistent with the strong positive correlations seen in Fig. 2a and b. Finally, columns (7) and (8) in Table 1 report fixed effects OLS estimates of a hypothetical monotonic relationship between a party’s position and blurriness. However, all these estimates are statistically insignificant at conventional levels, providing evidence for the existence of a non-monotonic relation.^Footnote13

As a complement to the previous findings, Fig. 7 and Table 7 in Appendix A.1 show estimates of Eq. (1) for an alternative outcome: the standard deviation in expert assessments relative to party positions on the economic dimension (Fig. 7a and columns (1) to (4) of Table 7) and the social values dimension (Fig. 7a and columns (5) to (8) of Table 7). As discussed earlier (see footnote 5), while we do not view this alternative measure of blurriness as a substitute for the baseline results, it serves as an important robustness check. Notably, the results in Fig. 7 and Table 7 align with our baseline findings, confirming the previously observed inverted U-shaped relationship.

As discussed in the previous section, the estimates in Table 1 may suffer from serious endogeneity bias. To address this concern, I re-estimate the models in Eqs. (1) and (2), while (i) controlling for potential confounding time-varying variables at the party level, and (ii) instrumenting position and centrism with lagged values.

Table 8 in Appendix A.1 presents estimates of the baseline models while incorporating additional controls, namely, the level of internal dissent within each political party, the age of each party, and whether the party is in government. The results shown in Table 8 are statistically indistinguishable from those displayed in Table 1. As previously explained, this supplementary evidence reduces the likelihood of omitted variables bias in our baseline results. Table 9 in Appendix A.1 provides instrumental variable (IV) estimates for the models described in Eqs. (1) and (2), employing lagged position (and centrism) as instruments. The table presents the IV estimates for the second stage, along with the F-statistic pertaining to the first stage. In addition to providing evidence consistent with the relevance assumption (e.g. showing a F-statistic consistently exceeding 10), Table 9 also shows estimates that are statistically equivalent to those presented in Table 1. As previously argued, this supplementary evidence reduces the likelihood of simultaneous bias in our baseline results and offers additional evidence against the existence of omitted variables bias.^Footnote14

While we acknowledge that the findings in Tables 8, 9 and 10 do not entirely eliminate bias in the main results of this section, nor do they establish causality, at the very least, they indicate that any potential bias is likely to be small.