Mapping party competition on gender-related issues in Europe

Abstract

Gender-related issues have become increasingly contentious in European politics, but comparative research on how parties address them remains limited. Existing studies have generally focused on particular countries or issue areas, while systematic evidence on the attention parties devote to gender-related issues and the positions they express across countries and over time remains scarce. This article examines the prominence of and party stances on gender-related issues in European national election manifestos. We use manually annotated manifesto data from Denmark, Germany, Hungary, Spain, Switzerland, and the United Kingdom between 2003 and 2021 to develop a fine-grained coding scheme and train computational models. We then combine supervised transformer-based topic classification with zero-shot stance classification to extend the analysis to 511 parties in 29 European countries between 1961 and 2025. The findings show that gender-related issues have become more prominent over time, although this development varies considerably across topics, party families, and periods. The increase is particularly pronounced for sexual and gender-based violence, LGBTQ+ issues, immigration and citizenship, and gender-related issues in the labour market. Care-related issues have remained consistently salient, whereas attention to gender equality in political and social representation has declined. Supportive statements are generally more prevalent than oppositional ones, although a substantial proportion of manifesto statements are neutral or descriptive. Despite fluctuations in temporal trends, positional differences between party families are consistently observed across topics. Left-wing, social democratic, green, and liberal parties generally express more supportive positions, whereas Christian democratic, conservative, and especially radical-right parties are frequently less supportive or more oppositional. However, the size and direction of these differences vary across specific gender-related topics. These findings illustrate the contextual and ideological multidimensionality of European party competition on gender-related issues and call for greater caution in treating such issues as a single political dimension.

1 Introduction

Over the past decades, gender has emerged as a prominent and often polarizing axis of political contestation in Europe. Historically peripheral to mainstream political agendas, issues such as reproductive rights, LGBTQ+ recognition, and gender parity have progressively been integrated into partisan competition, particularly from the 1960s and 1970s onwards with the advent of second-wave feminism and corresponding legal reforms concerning divorce, abortion, and equal pay. These developments laid the groundwork for the subsequent institutionalization of gender equality through national legislation and European integration frameworks. However, such progress has remained uneven and politically contested. In recent years, opposition has intensified, particularly from radical right and conservative actors.

Political parties have been central actors in this evolving landscape, functioning both as gatekeepers and promoters of gender equality. Despite the widespread recognition of their pivotal role, systematic empirical evidence remains limited. Compared to the wealth of studies on gender differences in party support (Coffé et al., 2023; Spierings and Zaslove, 2017) and the descriptive or substantive representation of women (Gilardi, 2015; Hughes and Paxton, 2019), the longitudinal and comparative analysis of how political parties address gender-related issues in their electoral programmes is notably underdeveloped. This gap stems largely from the absence of appropriate data sources, which has led most prior research to rely on qualitative designs or case studies confined to specific countries or party types—thus limiting the potential for broader generalization. In the few studies that have pursued comparative approaches (Akkerman, 2015; O'Brien, 2018), the scope has remained restricted, and core questions concerning whether, when, and how parties engage with gender-related topics have yet to be systematically addressed.

This article seeks to fill this gap by presenting the first comparative and longitudinal analysis of party salience and stances on gender-related issues across Europe. Specifically, we ask: how has the salience attached by European parties to gender-related issues evolved over time? How do these trends vary across different gender-related topics? And finally, what positions do parties adopt on these issues?

To answer these questions, we introduce a novel coding scheme—Manifesto Gender Analysis (MGA)—designed to enhance the granularity and comprehensiveness of gender-related content analysis in party manifestos. Building on a team of trained annotators, we applied the MGA to 412 electoral manifestos issued between 2001 and 2023 in six European countries: Denmark, Germany, Hungary, Spain, Switzerland, and the United Kingdom. These annotated data were used to fine-tune state-of-the-art multilingual transformer-based language models, which were subsequently employed to classify gender-related issue salience and positions in 1972 documents from the Manifesto Project corpus (Lehmann et al., 2025), covering 511 parties over the period 1961–2025.

Our analyses show that gender-related issues have, on average, become more salient over time and are generally addressed in supportive terms. These patterns, however, vary substantially across issue domains and party families. Salience has increased especially in domains that have become more politically articulated in recent decades, including sexual and gender-based violence, feminist notions and movements, immigration and citizenship, and LGBTQ+ issues, while labor-market issues and care work and welfare remain among the most consistently salient areas. Party-family differences are also topic-specific. Left, social-democratic, green, and liberal parties tend to assign greater salience to labor-market issues, education, social representation and participation, feminist notions and movements, and LGBTQ+ issues. By contrast, Christian-democratic parties attach comparatively high salience to care work and welfare and reproductive and family rights, reflecting the overlap between these domains and long-standing concerns with family policy, care, and welfare provision. Healthcare and sexual and gender-based violence display more limited differences across party families, suggesting that some gender-related topics have diffused more broadly across the party system.

Positional patterns are related but distinct. Once parties address gender-related topics, supportive statements are generally more common than negative ones, but ideological differences remain visible in several domains. Labor-market issues, care work and welfare, healthcare, education, reproductive and family rights, sexual and gender-based violence, immigration and citizenship, and feminist notions and movements are clearly structured by party family, with left, social-democratic, green, and liberal parties often expressing more supportive positions. Radical-right parties are consistently among the least supportive, and in some cases the most negative, across a wide range of gender-related topics.

Methodologically, this article contributes to current research on party competition and text-as-data by introducing a systematic coding scheme for gender-related content in manifestos and integrating it with machine learning tools to enable scalable, multilingual analysis. The proposed pipeline leverages the strengths of human annotation and transformer-based classification, allowing both fine-grained interpretability and high-volume inference across countries and time. Substantively, the article provides a new empirical foundation for understanding how gender-related conflicts are structured in party competition. By mapping salience and positions over time and across ideological divides, we show that gender is not a monolithic issue but a layered and multidimensional field of contestation.

The remainder of the article is structured as follows. The next section situates our contribution within the existing literature on gender and party competition, and outlines the conceptual framework guiding our analysis. We then describe the Manifesto Gender Analysis (MGA) procedure and the data collection strategy. Then we present the design and performance of our measurement exercises using language models for estimating gender issue position and salience. The following section presents the core empirical findings on party salience and position across time, topics, and party families. We conclude by reflecting on the implications of our findings for research on gender, political conflict, and party competition in Europe.

2 State of the art

Gender-related issues are not new to European politics. Interestingly, however, compared to gender differences in party support (Coffé et al., 2023; Spierings and Zaslove, 2017) or the political representation of women (Gilardi, 2015; Hughes and Paxton, 2019), much less is known about how political parties incorporate gender-related issues into their electoral campaigns. This is mainly due to the lack of appropriate data sources in this field. Party competition and political parties' policy preferences are typically studied using either expert surveys or the analysis of political texts, including social media posts, press releases, parliamentary speeches, and electoral programmes, also known as party manifestos. Among these sources, expert surveys and manifesto data occupy a particularly prominent position, especially in the comparative analysis of party competition during electoral campaigns. Nonetheless, for a long time, even these widely used sources did not include questions or categories explicitly dedicated to gender-related issues among the concepts they measured.

As a result, the growing interest in how political parties deal with gender issues takes mostly the form of qualitative case studies dedicated to specific parties (Alonso and Espinosa-Fajardo, 2021; Caravantes, 2021; de Lange and Mügge, 2015; Graff and Korolczuk, 2021; Luxton et al., 2025), specific contexts (Cabeza Pérez et al., 2023), and in a few cases with a comparative approach (Beloshitzkaya, 2021; Akkerman, 2015; O'Brien, 2018).

In particular, research on the intersection of populism, gender and right-wing politics has gained traction in recent years (Alonso and Espinosa-Fajardo, 2021; Akkerman, 2015; Farris, 2017; Graff and Korolczuk, 2021; Kantola and Lombardo, 2017, 2021; Volk, 2025), highlighting the centrality of gender in the political strategies of RRPs—such as the mix of modern and neo-traditionalist positions, as well as the demonization of gender and feminist politics as mobilization strategies. While all these findings are revealing, the comparative study of gender in party politics is still in its infancy, largely due to the already mentioned lack of reliable, valid and scalable data for analyzing gender issues in party platforms and political communication in general. The following section will describe the research findings that are relevant to the context of our paper, i.e., the investigation of party issue salience and the positions of political parties.

This measurement problem is especially visible in studies that examine whether parties respond to women voters or compete over so-called “women's issues.” In this literature, the substantive conclusions depend heavily on how gender-related preferences and issues are defined in the first place (Homola, 2019; Dingler et al., 2019). However, because much of this literature is based on single-country case studies, its findings are often difficult to compare across national contexts, as the following overview shows. Campbell and Childs (2015) show for the 2015 British election campaign that party competition over women's votes has gradually brought the parties closer together on “women's issues.” Their empirical analysis of party manifestos was based on six types of such issues: women and paid work; mothers and childcare; carers and the “cared for;” violence against women and girls; public life; human rights, development and immigration.

Again, for the case of three British general elections (2015, 2017, and 2019), Sanders et al. (2021) build on these categories and provide an even more differentiated coding scheme and study the spread and reach of parties' policy appeals to women voters. They identify ten categories of women (women by age: older women and younger women; mothers/women with caring responsibilities; ethnic minority women; women with disabilities; women on low incomes; LGBTQ+ women; vulnerable women; women abroad; all women) as well as three types of gender equality policies (women's status as women; class differentials; blueprint policies providing an overarching framework for gender equality). Their findings illustrate that the salience of women's issues has increased over time (at least for Labor and the Liberal Democrats, but not for the Conservatives). Moreover, the diversity of women addressed in the studied party manifestos has increased for many, but not all groups of women (not for women with disabilities, women of color, and low-income women).

A recent Spanish case study run by Cabeza Pérez et al. (2023) draws on another fine-grained categorization to study party positions on gender issues. Based on a coding scheme with four fundamental dimensions (welfare and labor market; violence; representation; values; and identity) with a positive and negative framing for each of these dimensions, they clearly illustrate why it is so important to differentiate between different types of gender-related issues. Their measurement via these four dimensions allows for a more nuanced picture on the left-right divide: First, left parties are more supportive for protection against gender violence, equal representation, and above all for cultural values and identity than right parties. Second, party ideology does not matter for welfare and labor market. Third, party ideology still matters for the framing of gender (in)equality.

While the above-mentioned studies focus country case studies, other scholars started to provide some more comparative findings. But it is important to note that this strand of research has a strong focus on radical-right and conservative parties (see also related studies such as Campbell and Erzeel, 2018; Chueri and Damerow, 2023), neglecting other party families. Based on CMP data, but a more fine-grained coding scheme, (Akkerman 2015) analyses the positions of six populist most successful radical-right parties in Western Europe.¹ Her findings show that although

“principles like gender equality and freedom of choice are emphasized in the immigration and integration domain […] almost all the parties are conservative when they address issues related to the family, such as opportunities of women on the labor market, childcare, abortion, or the status of marriage” Akkerman, (2015, p. 56).

O'Brien (2018) uses an even broader country selection (Austria, Belgium, Denmark, France, Germany, Great Britain, Ireland, the Netherlands, Norway, Portugal, Spain, and Sweden) as well as a much more quantitative approach than done before in this field.² She explores women's inclusion on the election platforms of 56 parties between 1980 and 2008 based on a systematic dictionary approach (containing almost 100 words signaling feminine issue framing). Her findings show that conservative parties are less likely to reference women on their policy statements than other parties. Moreover, “right parties generally (and Christian democrats in particular) focus on women in the context of family life, while more transformative issues—such as lesbianism and feminism—remain untouchable” (O'Brien, 2018, p. 49) and the left-right division is too simple for understanding women's representation.

Finally, two papers on the topic appeared recently. Weeks et al. (2025) investigate party attention and party positions in three Southern European countries (Greece, Portugal, and Spain) from 1995 to 2022. They define five broad dimensions of gender-related political interests (gender equality, violence against women, work-family issues, reproductive issues, and sexuality). Based on these categories, their findings reveal two important findings for Southern European countries. First, left-wing parties put more emphasis on gender-related interests than other parties. Second, far-right parties also highlight some gender-related interests standing out for traditional positions. For a larger country selection, i.e., for the case of 25 European countries between 1984 and 2022, Meguid et al. (2025) construct their own data set including five topics: gender rights, gender-based violence, care work and welfare, reproductive rights, and sexuality and gender identity. Their findings show that populist radical right parties emphasize gender-related rights in phases where they electorally vulnerable; however, they instead of adopting more gender-egalitarian positions, they highlight femonationalist stances.

Taken together, this literature review points to two central implications for the present study. First, because existing evidence remains fragmented across countries, parties, and issue domains, we still know relatively little about how systematically parties have incorporated gender-related issues into electoral competition over time (see also Meguid et al., 2025). Second, because gender-related politics covers a wide range of substantively different topics—including labor market equality, welfare and care, reproductive and family rights, gender-based violence, political and social representation, LGBTQ+ issues, feminism, and the gendered dimensions of immigration and citizenship—party engagement with gender cannot be captured by a single generic category. It is therefore necessary to examine which gender-related topics parties emphasize, and how they position themselves on those topics. These considerations structure the theoretical expectations developed below.

3 Theoretical expectations

The preceding discussion highlights both the need and the difficulty of studying party competition on gender-related issues comparatively. Existing research suggests that parties differ in the extent to which they address gender-related topics and in the positions they adopt on them, but the available evidence remains fragmented across countries, party families, and issue domains. Against this background, the main substantive contribution of this paper is to map, comparatively and longitudinally, the importance parties attach to gender-related issues and the positions they adopt on these topics. At the same time, the article also makes a methodological contribution through the MGA measurement exercise, which develops and applies a fine-grained strategy for identifying gender-related content in party manifestos.

This dual contribution shapes the scope of the theoretical expectations advanced here. Since part of the article is devoted to presenting, validating, and discussing the measurement strategy, there is limited space to develop detailed hypotheses for each possible combination of time period, country, party family, and gender-related topic. This is reinforced by the descriptive aim of the analysis, which is to provide evidence across all the gender-related topics included in the coding scheme, rather than focusing on a single issue area. Finally, as discussed in the methods section, the models perform well enough to support aggregate comparisons across time, space, and party families, but they still leave room for improvement in more granular applications. For these reasons, the study formulates broad theoretical expectations rather than highly specific hypotheses.

A first broad expectation concerns long-term changes in the salience and framing of gender-related issues in party competition. Socio-structural transformations over the past decades—such as rising levels of education, labor market participation, and changing family structures—are likely to have reshaped citizens' priorities and, indirectly, parties' issue agendas. A prominent account of these developments is provided by neo-modernization theory (Inglehart and Welzel, 2005), which links cultural change to processes of economic development and human development. As societies move from agrarian to industrial and post-industrial stages, traditional, survival-oriented values tend to give way to more secular and self-expression values, particularly among younger cohorts. Within this framework, gender equality becomes increasingly legitimate as a political concern, and attitudes toward women's roles in the family, the labor market, and politics become more egalitarian (Inglehart and Norris, 2003). From this perspective, one would generally expect a gradual increase in the visibility of gender-related issues in party manifestos over time, alongside a broad shift toward more supportive positions on gender equality, albeit with variation across countries and party families.

Nonetheless, the expected effect of socio-structural changes might interact with institutional changes determined by political agency. Over the course of the twentieth century, and especially in its second half, European democracies witnessed major reforms expanding women's political and social rights, divorce and abortion legislation, from education to labor rights. Once such reforms are implemented, however, their effects on party competition are unlikely to be linear. Building on theories of thermostatic responsiveness (Wlezien, 1995, 2004), policy change may alter not only the direction but also the intensity of public concern. When policy outputs align closely with prevailing public preferences, gender-related issues may lose urgency and become less salient in electoral competition, even if support for the underlying principles remains stable. Conversely, periods of misalignment between public opinion and policy—whether due to restrictive reforms or stalled liberalization—may be associated with renewed attention and politicization. Over time, this interaction between public demand and policy output is likely to generate cycles of heightened and diminished attention to gender-related issues rather than a monotonic trend.

Ideological orientation is also expected to structure how parties engage with gender-related issues, both in terms of salience and positioning (Cabeza Pérez et al., 2023; O'Brien, 2018; Weeks et al., 2025). Parties on the left have traditionally been more closely associated with egalitarian ideals and are therefore more likely to emphasize gender-related topics and to articulate supportive positions across a wide range of domains. Recent evidence is consistent with this expectation: left-wing parties tend to devote more attention to gender-related interests (Weeks et al., 2025), while studies of party positions find that left parties are generally more supportive of gender equality, especially on topics such as gender-based violence, representation, and values and identity (Cabeza Pérez et al., 2023). Moving toward the right of the ideological spectrum, support for gender equality tends to be more conditional and uneven. Conservative and right-wing parties are less likely to address women or gender-related concerns in their policy statements (O'Brien, 2018), and when they do so, they often concentrate on topics related to family, care, and traditional social roles (Akkerman, 2015; O'Brien, 2018).

Importantly, however, ideological divides are unlikely to be uniform across all gender-related topics. Previous research shows that different topics follow different political logics: ideology appears to matter strongly for gender-based violence, representation, sexuality, reproductive rights, and values and identity, but less consistently for welfare and labor-market issues (Cabeza Pérez et al., 2023; Weeks et al., 2025). Issues such as women's labor market participation, childcare, or work–family reconciliation may therefore attract relatively broad consensus, particularly in contexts marked by demographic aging, labor shortages, or welfare-state pressures. By contrast, topics such as reproductive and family rights, LGBTQ+ issues, feminism, or symbolic recognition are more prone to ideological polarization. In terms of agenda-setting, progressive parties can be expected to emphasize gender-related topics more consistently, while right-wing parties may engage with them more selectively, either defensively or instrumentally, depending on the topic and on strategic considerations.

Within this ideological landscape, RRPs warrant particular attention. Existing research consistently portrays these parties as mobilizing and polarizing actors in debates surrounding gender, sexuality, family norms, and feminism (Akkerman, 2015; Farris, 2017; Graff and Korolczuk, 2021; Kantola and Lombardo, 2017, 2021; Volk, 2025). Many RRPs combine modern and neo-traditionalist positions: they may selectively endorse gender equality and women's rights when these topics can be connected to immigration, integration, or national identity, while simultaneously defending traditionalist positions on family, reproduction, labor, and LGBTQ+ issues (Akkerman, 2015; Farris, 2017; Meguid et al., 2025). As a result, one would expect RRPs to display more negative positions on gender-related issues overall, especially on topics linked to bodily autonomy, sexual minorities, feminist claims-making, and challenges to traditional family arrangements. At the same time, their engagement is likely to be selective rather than comprehensive, avoiding open challenges to widely consolidated norms—such as women's access to education or political rights—that enjoy broad societal support. Their approach is therefore better characterized as targeted retrenchment and selective appropriation, rather than wholesale opposition to gender equality.

Finally, parties' engagement with gender-related issues is likely to be shaped by contingent macro-level events, such as economic crises, political shocks, or public health emergencies. Previous research suggests that crises like the Great Recession or the COVID-19 pandemic have disproportionately affected women's economic security and care responsibilities, thereby altering both material conditions and political opportunity structures. Such disruptions may create openings for progressive claims, for instance around investment in care infrastructures or social protection, but they may also reinforce conservative narratives centered on family, order, and social cohesion. In other contexts, crises may crowd out gender-related concerns altogether, as parties prioritize economic stability, security, or public health. Accordingly, one would expect both the salience and the framing of gender-related issues in party manifestos to vary in response to these events, often in reactive and strategically motivated ways, rather than following a uniform trajectory.

4 Research design

4.1 Party politics and manifestos

Estimating party features and behavior represents one of the main research endeavors of political science and lies at the heart of fundamental theoretical and analytical models to understand political phenomena (Braun, 2023). Party competition and political parties' policy preferences are typically examined using expert surveys (Rovny et al., 2025) or text-as-data approaches, commonly applied to social media data, press releases, parliamentary speeches, and in particular party manifestos. The latter data have enabled longitudinal and comparative analyses of electoral competition across multiple levels of governance — European (Carteny et al., 2023), national (Volkens et al., 2013), regional (Alonso et al., 2013), and local (Gross and Jankowski, 2020). Manifestos, transformed into data by experts who classify their parts according to a pre-existing scheme, are used to analyse with statistical methods how much emphasis political parties place on particular political issues (issue salience), and the positioning of parties on policy and ideological dimensions.

This approach has a long history (Robertson, 1976), and the long lasting and widespread application of this method allowed researchers to gauge its merits and its limits over the years. Manifestos represent one of the few data sources allowing long-term analyses and cross-national comparisons. Moreover, composed mostly of official statements of political parties, manifestos allow to map party preferences which are relevant in an election, and that potentially inform the behavior of parties in parliaments and governments. The issues in manifestos align with policies advocated by parties, providing a self-positioning on key policy issues. This, in some instances, represents a crucial advantage as compared to the other two main approaches used for estimating party positions.

Despite these merits, manifesto-based research is not free from important limitations, and in particular is far from being an unbiased approach. One great weakness of manifesto data is its time-consuming and occasionally unreliable manual coding process, complicated by traditional codebooks with outdated categories. Misclassifications are a systemic issue, prompting calls for radical simplification, such as using natural sentences instead of quasi-sentences (Mikhaylov et al., 2012).

In addition to methodological limitations, there are also critiques aimed at its validity. While policymakers and media pay attention to elite communication and the media coverage surrounding it, manifesto data's relevance is questioned as few people read them, and only a minority of party officials are involved in writing. Additionally, official information from smaller parties can be limited, with documents sometimes being informal statements rather than official manifestos. This latter challenge isn't unique to manifesto data, affecting various quantitative sources where smaller parties may lack visibility. However, a transparent approach allows users to verify document types, providing crucial context for empirical analysis – and this is definitely the case when it comes to the most visible manifesto projects (e.g., Lehmann et al., 2025; Carteny et al., 2023).

Overall, the strengths of the party manifesto analysis method outweigh its weaknesses, which again explains its success as a tool for investigating party behavior, features, and related political phenomena across time and space - and why we rely on this data source for our investigation.

4.2 The MGA coding procedure

Any evaluation of political parties' behavior is subject to the historical context in which the inquiry takes place. Although certain aspects of parties and party rivalry may be relatively stable, each historical period has its own distinctive characteristics. Most of the projects engaged with the analysis of party manifestos attempted to - and, often, succeed in - creating coding procedures able to stand the test of time. Yet, in some instances, new coding procedures have been developed because existing ones could not grasp specific topics. It happened for immigration-related issues for national manifestos (Lehmann and Zobel, 2018), and it applies to gender-related issues across different types of electoral programmes (e.g., Beloshitzkaya, 2021; de Lange and Mügge, 2015; O'Brien, 2018).

Existing comparative manifesto datasets have only recently begun to include gender-related categories, and to the best of our knowledge, only two coding schemes and related datasets have been implemented and made publicly available for analyzing gender-related issues for quantitative and comparative analyses. The 2019 Euromanifesto codebook (Reinl and Braun, 2023) introduced two categories for gender- and LGBTQ-related arguments, while the Regional Manifesto Project (RMP, Cabeza Pérez et al., 2023) added eight gender-related categories to its most recent coding scheme. These innovations represent important steps toward the quantitative study of gender in party politics, but they remain limited for the purposes of this article. First, their gender-related categories are relatively broad and therefore cannot capture the fine-grained variety of issues addressed by parties. Second, because they are embedded in broader manifesto coding schemes, they often require coders to assign a statement either to a conventional policy field, such as labor or welfare, or to a gender-related category. This is problematic because gender-related claims are often articulated through other policy domains.³

To address these limitations, we developed the Manifesto Gender Analysis (MGA) coding procedure. MGA was designed to identify gender-related content in national party manifestos while preserving both thematic detail and comparability across countries.⁴ Coders first split manifestos into quasi-sentences and then assessed whether each quasi-sentence contained gender-related content. Relevant quasi-sentences were assigned to a broader category and a more specific topic category, and their stance was coded as supportive, neutral, or opposed to gender equality. The present article focuses on these two dimensions: topic classification and stance.

The MGA scheme covers five broad domains: labor; family and welfare; participation and representation; rights, discrimination, and violence; and gender-related notions. Each domain encompasses a set of more specific gender-related topics (see Table 1). These topics include, among others, labor-market inequalities, care responsibilities, political and social representation, reproductive and family rights, sexual and gender-based violence, migration and citizenship, feminism, heteronormativity, and LGBTQ+ issues. The selection of these categories was informed by a typology of gender-based needs based on an extensive collection of gender-based needs discussed in the academic literature in Denmark, Germany, Hungary, Spain, Switzerland, and the United Kingdom between 2007 and 2023 (Amacker et al., 2024). At the same time, the scheme was designed to balance substantive granularity with the reliability of the coding procedure (Braun and Carteny, 2024).

The MGA procedure was applied to 412 documents, encompassing the vast majority of the Euromanifesto and MARPOR corpora from 2003 to 2021, across the six national contexts under examination: Denmark, Germany, Hungary, Spain, Switzerland, and the United Kingdom. The collection of manifestos was completed between April and November 2023, and the coding of quasi-sentences—based on the MGA codebook—was carried out between December 2023 and April 2024. Annotations were conducted by eleven coders; their outputs were subsequently harmonized and integrated into a single data frame.

Reliability was assessed through intra-coder and inter-coder checks in four of the six countries. Agreement was evaluated using Gwet's AC1 (Gwet, 2008) and Krippendorff's alpha (Hayes and Krippendorff, 2007), both of which are chance-corrected measures. Overall, the estimates indicate acceptable to strong reliability, especially when using AC1, with most inter-coder comparisons concentrated around 0.70–0.85 and intra-coder comparisons around 0.75–0.90. Krippendorff's alpha is systematically lower, as expected in a coding task characterized by a large residual category and several rare gender-related categories, but the main reliability concerns are concentrated in a small number of comparisons rather than across the coding exercise as a whole. In particular, coder 4 displayed substantially lower reliability, especially according to Krippendorff's alpha, and was therefore excluded from the subsequent analyses to reduce annotation noise in model training and validation.

4.3 Topic and stance classification

After the manual annotation stage, we used the coded quasi-sentences to fine-tune a supervised topic classifier. The purpose of this step was to extend the MGA coding scheme from the manually annotated material to a larger manifesto corpus, while preserving the distinction between the gender-related topics defined in the codebook. Each observation consisted of a quasi-sentence and its immediate textual context, which was included to help the model interpret short or potentially ambiguous statements.

The topic classification task was formulated as a 25-class classification problem, consisting of one residual category for sentences that do not contain gender-related content and 24 substantive gender-related topic categories. We fine-tuned ManifestoBERTa (Burst et al., 2024), a transformer model pre-trained on manifesto text, because preliminary comparisons showed that it performed better than any other alternative. The manually annotated data were divided into training, validation, and test sets. The training set was used to estimate the model, the validation set to guide model selection, and the held-out test set to assess performance after fine-tuning (Figure 1).

Figure 1

A central challenge in this task is the strong imbalance of the annotated data: the residual category is much larger than any substantive gender-related category, while several specific topics occur only rarely in party manifestos. To address this problem, we used an augmented version of the training data, designed to increase the number and diversity of examples for rare categories while preserving the original topic labels.⁵ Augmentation was also used to improve the model's ability to distinguish explicitly gender-related statements from more general policy claims in overlapping domains such as immigration and citizenship, healthcare, and education.

Model selection proceeded through a grid search over the main fine-tuning parameters, comparing models trained on the original and augmented datasets. The augmented models performed consistently better, especially in terms of macro F1, indicating improved performance on less frequent and conceptually difficult topics. The selected classifier used ManifestoBERTa as the base model, with a hidden layer size of 512, a learning rate of 1e-5, a batch size of 32, dropout set to 0.3, and no unfrozen transformer layers. It was trained for five epochs on the augmented dataset.

The final model reaches a weighted F1 score of 0.70 and a macro F1 score of 0.55 on the internal evaluation data.⁶ These results are acceptable for a highly imbalanced 25-class classification task, but they also indicate uneven performance across categories. The classifier performs better for larger and substantively distinctive topics, while rare or conceptually overlapping categories remain more difficult to identify. Additional validation against an independently coded and manually validated set of sentences using GPT-5.5 produces broadly similar, though slightly lower, estimates. For this reason, the empirical analysis prioritizes aggregate patterns and broader topic groupings where appropriate.

The second stage of the classification pipeline estimates the stance expressed toward each gender-related issue identified in the first stage. This step is applied only to sentences classified as belonging to one of the substantive MGA topic categories. Topic and stance classification are therefore treated as separate tasks: the fine-tuned ManifestoBERTa model first identifies the relevant gender-related issue, while stance is subsequently estimated through a context-aware zero-shot natural language inference (NLI) pipeline based on Political DEBATE (Burnham et al., 2024).⁷

We adopted a zero-shot approach because the manually coded data were not large or balanced enough to support a second supervised fine-tuning stage for stance classification across topics, languages, countries, and stance categories. For each topic, we defined a short hypothesis specifying what it would mean for a sentence to support that issue from a gender-equality perspective. The NLI model then evaluates each manifesto sentence, together with its available textual context, against this hypothesis. Entailment is interpreted as support for the topic, contradiction as opposition, and neutral as the absence of a clear substantive stance. This formulation is preferable to a generic sentiment task because support for gender equality is not reducible to positive or negative tone: a sentence may be affectively neutral while supporting a gender-equality position, or use positive language while endorsing a negative position (Burnham, 2025).

We validated this procedure by comparing the NLI pipeline with several local instruction-tuned language models that represent plausible alternatives for zero-shot stance classification, including Llama, Mistral, Gemma, and Qwen. The comparison shows that the NLI pipeline provides the strongest and most balanced performance across the three stance categories. This is most clearly visible in the F1 scores. On macro F1, which gives equal weight to each stance category and is therefore especially relevant for assessing balanced performance, NLI reaches 0.729, followed by Llama at 0.687 and Mistral at 0.632, while Gemma and Qwen perform substantially worse. The same ranking holds for weighted F1, where NLI reaches 0.766, compared with 0.721 for Llama and 0.673 for Mistral. This indicates that the advantage of the NLI pipeline is not driven only by its handling of less frequent or more difficult stance categories, but also holds under the empirical distribution of stance labels.⁸

The confusion matrices provide further support for the choice of the NLI pipeline. Its main advantage is that it offers a comparatively balanced classification across the three stance categories: it recovers a large share of genuinely supportive and dismissive statements, while also preserving a meaningful neutral category for sentences that do not express a clear substantive position (see Figure 2). Some caution is warranted regarding negative statements: of the genuinely negative cases, 167 are classified as negative, 102 as neutral, and 69 as positive. This might make certain parties appear more supportive than their underlying statements indicate. Nonetheless, these results remain consistent with expected relative party positions, and alternative aggregation methods, such as positive versus non-positive, yield comparable patterns and rankings.

Figure 2

4.4 Model predictions

The trained MGA topic model and the zero-shot NLI stance model were applied to the full MARPOR corpus (1,916,108 sentences) for the selected countries (29) in order to generate sentence-level estimates of both gender-related topic salience and position taken by European parties on these topics. The preprocessing of this prediction corpus was carried out in two stages. First, the MARPOR corpus was retrieved and organized in R using manifestoR (Lewandowski et al., 2015). The selected MARPOR texts were merged with the aggregated MARPOR dataset to attach party- and election-level metadata, including country, election date, party identifier, and party name. The corpus was then divided into two files: one containing texts already available as MARPOR sentence-level observations, and one containing full manifesto documents for which sentence-level annotations were not available.

The second stage was implemented in Google Colab, where the exported files were prepared for sentence-level prediction. Full manifesto documents were segmented into sentence-like units using two multilingual procedures: spaCy, combined with automatic language detection through lingua (Stahl, 2026; Honnibal et al., 2020), and Stanza (Qi et al., 2020). The two outputs were compared as a diagnostic check on the number and length of the resulting segments. The segmented texts were then combined with the original MARPOR sentence-level observations, while preserving metadata on the manifesto, party, election, country, and segmentation source. Before prediction, each text unit was formatted consistently with the fine-tuning data, with the focal sentence and, where available, its surrounding context passed jointly to the tokenizer.

The MGA topic model was first used to identify whether each text unit belonged to the residual category or to one of the substantive gender-related topic classes. For text units assigned to a gender-related topic, the NLI model was then used to estimate whether the statement expressed a supportive, neutral, or negative stance toward that topic. The resulting sentence-level topic and stance predictions were subsequently aggregated to document-level estimates of gender-related salience and position, as explained in the following section.

4.5 Aggregation strategy

After generating sentence-level topic predictions, we converted model probabilities into document-level measures through a two-step aggregation procedure. First, each sentence was assigned to a topic class only if the predicted probability of the best-fitting class exceeded a threshold of 0.70. This threshold was selected because it provided the best balance in the model validation stage. Sentences below the threshold were assigned to the residual category, thereby reducing the risk of treating weak or uncertain predictions as substantive gender-related content.

The thresholding procedure was applied in two ways. In the first specification, it was applied directly to the original MGA topic scale, consisting of the residual category and the fine-grained gender-related classes. In the second specification, used in the main analyses, selected fine-grained classes were first aggregated into broader substantive categories, and the threshold was then applied to these aggregated probabilities (see Table 1). We do not rely on the original domains of the coding scheme for this aggregation, because these domains are too broad for the purposes of the analysis and often combine substantively distinct, although related, topics. The remapping therefore represents an intermediate solution: it reduces sparsity by grouping closely related fine-grained classes, while preserving more analytical granularity than the original domain structure would allow. This procedure reduces the likelihood that a sentence is classified as residual simply because the model distributes probability mass across several adjacent gender-related categories. For example, if a sentence receives moderate probabilities across adjacent labor-market or care-related classes, the original fine-grained specification may fail to assign it to any substantive topic, whereas the remapped specification can still classify it as belonging to the broader category. This procedure increases the average level of estimated gender salience but leaves temporal patterns and party-family trajectories largely unchanged.⁹

Once gender-related sentences had been identified, the zero-shot NLI classifier was applied to estimate the stance expressed toward the assigned topic. For each sentence, the stance label was selected as the NLI output category with the highest predicted probability, distinguishing between supportive, neutral, and negative statements. Unlike the topic-classification step, no additional probability threshold was applied at this stage, since the NLI task was used only after the topic classifier had already identified a sentence as substantively relevant.

The final measures were then constructed by aggregating sentence-level predictions to the document level. Topic salience was measured as the proportion of all sentences in a manifesto assigned to each substantive topic category. Thus, for each document and topic, salience captures the share of the manifesto devoted to that specific gender-related issue. Position estimates were constructed among sentences assigned to a given topic, using the distribution of supportive, neutral, and negative statements. Specifically, we calculated a bounded net-position score by subtracting the number of negative statements from the number of supportive statements and dividing this difference by the total number of supportive, neutral, and negative statements in the topic-document cell. Formally, the measure is defined as [(P – N)/(P + N + U)], where (P), (N), and (U) denote the number of supportive, negative, and neutral statements, respectively. The resulting scale ranges from –1 to +1: higher values indicate that supportive statements predominate, lower values indicate that negative statements predominate, and values close to zero indicate either a balance between supportive and negative statements or a predominance of neutral statements. Neutral statements are therefore retained in the denominator, pulling estimates toward the center of the scale rather than being treated as negative or excluded from the calculation.

4.6 Topic and position analysis

The overall prediction process generated estimates for 1,972 manifestos of 511 parties for the period 1961–2025.

We first examine temporal patterns using LOESS smoothing. They are reported for overall gender-related salience and position, and then disaggregated by topic. We discuss country- and party-family-specific patterns in the text where they are substantively relevant, while the full set of disaggregated LOESS estimates is visualized in the Supplementary material.¹⁰ The LOESS curves provide a descriptive representation of how attention to gender-related issues, and the positions attached to them, vary across elections over time. This approach is useful because the corpus covers a long historical period and because the density of available manifestos varies across countries and years. The smoothed estimates therefore help identify broad trajectories without imposing a strict linear functional form.

We then estimate multivariate regression models to assess systematic differences across party families by topic. These empirical analyses focus on the subset of manifestos covering the period from the 1990s to 2025. This restriction is motivated by data availability: from the 1980s onwards the dataset contains substantially more documents across the countries and party families analyzed. Focusing on this period therefore allows us to base the analyses on a broader and more balanced set of observations, reducing the risk that estimates for earlier decades are driven by a limited number of parties, countries, or manifesto documents.

For salience, we estimate topic-specific OLS models and, as a robustness check, Tobit models. The Tobit specification is included because the salience measures are bounded and highly skewed toward zero, as many manifestos devote little or no attention to specific gender-related topics (see Figure 3). In both specifications, the models include country and election-year intercepts to account for unobserved country-level and period-specific differences. Uncertainty is estimated using a cluster-robust variance-covariance matrix at the country level, which yields cluster-robust standard errors of the regression coefficients.

Figure 3

For position, we estimate topic-specific OLS models using the same general modeling logic. These models are also restricted to manifestos from the 1990s onwards, include country and year intercepts, and estimate uncertainty using a country-level cluster-robust variance-covariance matrix. Unlike the salience models, however, the position models include only manifestos with at least one gender-related observation for the relevant topic, since documents without any such statement do not express an observable position on that issue. The dependent variable is the document-level net-position estimate.

Party were assigned to party families relying on the classification provided by the MARPOR project and combined with the PopuList data (Rooduijn et al., 2023), a expert-based dataset widely used to identify populist, Eurosceptic, far left, and far right actors. In particular, the MARPOR “Nationalist” party family was substituted with the RRP family relying on the PopuList classification of far-right actors.

5 Results

5.1 Temporal trends

To examine how party attention to gender-related issues has evolved over time, we begin with trends in gender-related issue salience across election years, measured as the share of quasi-sentences assigned to any gender-related topic,¹¹ and then disaggregate the analysis by gender-related topic.

Our results suggest an overall growth in the salience of gender-related topics, but also show that this growth is concentrated in specific areas (see Figure 4). More specifically, labor-market issues and care work and welfare are among the most consistently salient topics throughout the period, with labor-market salience rising gradually and care work and welfare remaining comparatively high after a marked increase around the late twentieth century. Other domains remain more marginal, but several display clear late-period increases, especially sexual and gender-based violence, immigration, and citizenship, LGBTQ+ issues and, to a lesser extent, feminist notions and movements. Within some of these categories, further specific trends can be detected. For instance, within social and political representation, the two original MGA categories—political representation and social representation—move differently over time. The same applies to the broader category of reproductive and family rights, where the two original MGA categories also follow somewhat dissimilar trends. Nonetheless, these patterns suggest that party competition on gender-related issues has increased not only in salience, but also in thematic scope.

Figure 4

Country-specific analyses show that this increase is not confined to a small number of cases: most countries display either a gradual upward trajectory or a late-period acceleration in gender salience. At the same time, the timing and intensity of this increase vary across national contexts. The upward trend is especially clear in several Western and Southern European countries, whereas patterns are more uneven in parts of Central and Eastern Europe, where gender-related issues appear to be politicized in more discontinuous ways. Party-family trends also suggest that rising gender salience is not limited to left, social-democratic, or green parties, which overall display higher average salience on these topics than other party families. These families show some of the clearest increases, but Christian-democratic, conservative, and radical-right parties also pay growing attention to gender-related issues, especially in the most recent period. As the multivariate regression analysis shows, however, party-family differences are highly conditional on the topic considered, despite some common overall patterns.

Party positions on gender-related issues vary across domains (see Figure 5). This variation should be interpreted with some caution, especially for categories that appear only rarely in party manifestos or are concentrated in more recent decades. In these cases, wider uncertainty intervals and shorter time series indicate that the estimated trajectories are based on relatively sparse evidence. With this caveat in mind, the clearest temporal increase concerns sexual and gender-based violence, where position scores move steadily upwards over time. More moderate upward movements are also visible for feminist notions and movements, social representation and participation, and, to a lesser extent, immigration and citizenship. Other domains, including labor market issues, care work and welfare, healthcare, education, and reproductive and family rights, remain closer to the center of the scale and show more limited temporal change. LGBTQ+ issues and the residual rights, discrimination and violence category display comparatively high position scores in the periods for which they are observed, but they follow a weak trend toward less supportive positions.

Figure 5

Country-level trends largely confirm this aggregate pattern. In most countries, position estimates are centered around mildly positive values, and clearly negative trajectories are rare. Some countries show gradual movement toward more supportive positions over time, while others remain comparatively stable once gender-related issues become sufficiently visible in manifesto texts. The main cross-national differences concern the strength, timing, and continuity of these trends, rather than a sharp divergence in the direction of party positions.

Party-family trends follow a similar pattern. Most party families remain on the positive side of the scale, and the main variation concerns the degree of support rather than a clear divide between supportive and negative positions. Left, social-democratic, green, and liberal parties often display somewhat higher position scores, particularly on social representation, feminist notions and movements, reproductive and family rights, and sexual and gender-based violence. Christian-democratic and conservative parties tend to remain closer to the center, although they are not consistently negative. Radical-right parties appear less supportive in some domains, especially LGBTQ+ issues, feminist notions and movements, and social representation and participation, but these patterns are uneven and in some panels based on sparse observations. Agrarian and residual party families are more difficult to interpret because of the smaller number of observations and more discontinuous series.

A final consideration concerns the implications of the position scale, as defined in the methods section. Because the measure includes neutral statements in the denominator, it captures the balance between explicitly supportive and explicitly negative statements while also taking into account the amount of neutral, descriptive, or otherwise non-directional discussion. This choice mainly affects the location of the estimates on the scale: neutral statements pull position scores toward the center, so that the intercept should be interpreted as the average net balance of supportive and negative statements, conditional on the prevalence of neutral statements. Alternative log-odds specifications, closer to established approaches for scaling coded political texts (Lowe et al., 2011), produce very similar temporal patterns, but shift the estimates downwards because supportive statements are then contrasted more directly with neutral and negative statements together. The substantive interpretation is therefore not that manifesto language is uniformly or strongly supportive, but that explicitly supportive statements are, on average, more common than explicitly negative ones, within a larger set of statements that often remains neutral or descriptive.

Read in this way, the descriptive evidence indicates that the increasing salience of gender-related issues is only partly mirrored by a shift toward more explicitly supportive positions.¹² The strongest temporal change is found for sexual and gender-based violence, while several other topics display more limited or weakly patterned movement once neutral and descriptive statements are taken into account. Party-family differences also appear present but generally moderate: most families are located on the supportive side of the scale, but variation emerges in the degree, timing, and thematic concentration of support. The next section examines these party-family differences more systematically through multivariate regression models, assessing whether the descriptive patterns persist once country and year differences are taken into account.

5.2 Regression analysis

The regression analysis,¹³ based on manifestos published between 1990 and 2025, shows a clear but topic-specific structuring of gender-related salience by party family (Figure 6). These differences should be interpreted in light of the fact that the panels are plotted on different y-axis scales for visual clarity. Some contrasts may therefore appear visually similar across panels even though their substantive magnitude differs considerably.

Figure 6

The largest party-family differences are found in domains where expected salience is also comparatively high. Labor-market issues display the clearest pattern: left, social-democratic, green, and liberal parties attach substantially greater salience to this topic than Christian-democratic, conservative, radical-right, and other party families. A broadly similar, although smaller, left–right gradient is visible for education, social representation and participation, feminist notions and movements, and LGBTQ+ issues. In these domains, gender-related claims appear more strongly embedded in the programmes of parties on the left and center-left.

Other domains follow a different pattern. Care work and welfare is most salient among Christian-democratic parties. A similar pattern appears for reproductive and family rights. This is substantively plausible, since these topics overlap with long-standing Christian-democratic concerns with family policy, care, welfare provision, and the organization of family life.

By contrast, party-family differences are limited for healthcare and sexual and gender-based violence. Although some variation is visible in the estimates, the absolute differences are small compared with those observed for labor-market issues or care work and welfare. These topics' salience therefore appears less strongly structured by ideological differences. Some categories also require particular caution because their overall salience is low and the underlying data are relatively sparse. This is especially relevant for residual or less frequent domains such as other gender representation and other rights, discrimination and violence.

Turning to party positions, the regression estimates suggest that, once parties address gender-related topics, predicted positions are generally more supportive than negative across most domains (Figure 7). This pattern should be interpreted carefully, as discussed earlier in the temporal trends analysis.

Figure 7

The results show a clear ideological structuring for several topics. Labor-market issues, care, care work, and welfare, healthcare, education, reproductive and family rights, sexual and gender-based violence, immigration and citizenship, and feminist notions and movements all display visible differences between party families. In most of these domains, left, social-democratic, green, and liberal parties tend to show more supportive expected positions, while radical-right parties are consistently among the least supportive, or the most negative. LGBTQ+ issues show a similar pattern, although the gradient is less pronounced and the uncertainty intervals are wider for some party families.

For other domains, party-family differences are less clear. Social representation and participation, other gender representation, and other rights, discrimination and violence do not display a strong or consistent ideological ordering across most party families. The main exception is again RRPs, which tend to show lower expected positions than the other groups. This suggests that, even where differences among mainstream party families are limited, radical-right parties often remain distinctive in the direction of their gender-related statements.

Some results are more counterintuitive. Conservative parties show relatively positive expected positions in other rights, discrimination and violence, feminist notions and movements, and LGBTQ+ issues.¹⁴ This does not necessarily indicate that conservative parties as a family are generally more supportive on these topics. Rather, it reflects the fact that relatively few conservative parties engage with these domains, and that those which do are often comparatively egalitarian cases, especially among Swedish, Norwegian, and UK conservatives. As in the salience analysis, the magnitude of the differences should be assessed with caution because the panels use different y-axis scales for visual clarity. Moreover, some topics are based on fewer observations than others, which increases uncertainty and makes precise party-family rankings less reliable.

6 Conclusion

This article has examined how European political parties have addressed gender-related issues in their election manifestos between the 1960s and 2025. Using the MGA annotated data, we developed and applied a computational strategy that combines supervised topic classification with zero-shot stance estimation. This allowed us to move beyond aggregate measures of attention to gender equality and to distinguish between different substantive topics, including labor-market inequalities, care work and welfare, reproductive and family rights, sexual and gender-based violence, immigration and citizenship, feminist notions and movements, and LGBTQ+ issues.

The results show that gender-related issues have become increasingly salient in European party competition. This expansion is not limited to a small set of parties or countries, but is uneven across topics, party families, and time. Rather than following a simple linear trajectory, the salience of gender-related issues expands in waves and through different substantive channels. Some topics, such as labor-market issues and welfare, have been comparatively stable components of gender-related manifesto politics. Others, including sexual and gender-based violence, feminist notions and movements, immigration and citizenship, and LGBTQ+ issues, have become more visible in recent decades. The politics of gender has therefore expanded not only in volume, but also in thematic scope.

The analysis also shows that salience and position follow related but distinct patterns. Parties differ not only in how much attention they devote to gender-related issues, but also in the kinds of issues they emphasize. Labor-market issues, education, representation, feminist notions and movements, and LGBTQ+ issues are more clearly structured along party-family lines, with left, social-democratic, green, and liberal parties usually attaching greater salience to these domains. By contrast, work–life balance and reproductive and family rights are especially salient among Christian-democratic parties, reflecting the overlap between gender-related issues and long-standing party traditions around family policy, care, and welfare provision. Other domains, such as healthcare and sexual and gender-based violence, appear less strongly differentiated by party family, suggesting that some gender-related issues have become more broadly integrated into party competition.

When parties take positions on gender-related topics, supportive statements are generally more common than negative ones. Yet this does not imply the disappearance of ideological conflict. Party-family differences remain visible across several domains, including labor-market issues, work–life balance, healthcare, education, reproductive and family rights, sexual and gender-based violence, immigration and citizenship, and feminist notions and movements. Radical-right parties are often distinctive for less supportive or more negative positions, even in domains where differences among other party families are less pronounced. At the same time, their engagement with gender-related issues is selective rather than comprehensive. These parties do not simply reject gender equality as a whole; instead, they are more likely to contest, reframe, or appropriate specific topics, especially when these can be connected to family norms, national identity, immigration, or opposition to feminist and LGBTQ+ claims.

These findings have two broader implications. First, gender-related party competition should not be treated as a single-issue dimension. Different gender-related topics enter party manifestos through distinct ideological, institutional, and policy traditions: some are closely linked to egalitarian and progressive agendas, while others can be articulated by parties across the ideological spectrum. Second, the politicization of gender reflects the interplay between the salience of and positioning on specific issues, but much of this transformation occurs through shifts in issue attention and the expansion of the thematic agenda.

The analysis also points to important limitations and directions for future research. In particular, positioning remains difficult to measure using explicit statements in party manifestos. As discussed earlier, our models perform less well in detecting negative statements. At the same time, opposition to gender equality may be expressed through framing and discourse rather than explicit negative sentences. Such opposition may also be more visible in other political texts, such as online political communication or parliamentary debates. Future research could combine these different sources and dimensions to provide a more comprehensive and systematic picture of opposition to gender equality. The next steps of the MGA project will therefore focus on refined modeling strategies, with the goal of releasing a new dataset and dedicated BERT models that will allow other researchers to extend the analysis conducted here. Future research could also build on these estimates by linking party-level measures to news media, parliamentary behavior, policy outputs or voter attitudes and behavior. It could examine how national contexts, policy legacies, policymaking processes, and political shocks shape how parties incorporate gender-related issues into their political communication.

References

• 1

  AkkermanT. (2015). Gender and the radical right in Western Europe: a comparative analysis of policy agendas. Patterns Prejudice49, 37–60. doi: 10.1080/0031322X.2015.1023655

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  AlonsoA.Espinosa-FajardoJ. (2021). Blitzkrieg against democracy: gender equality and the rise of the populist radical right in Spain. Soc. Polit. 28, 656–681. doi: 10.1093/sp/jxab026

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  AlonsoS.GómezB.CabezaL. (2013). Measuring centre–periphery preferences: the regional manifestos project. Reg. Federal Stud. 23, 189–211. doi: 10.1080/13597566.2012.754351

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  AmackerM.BiasL.RothermelA.NerinoV. (2024). Deliverable 1.1 of the UNTWIST Project: Typology. Zenodo, November.

  • Google Scholar
  • View reference in article

• 5

  BeloshitzkayaV. (2021). Affirmative gender equality policies in Central and Eastern Europe: moving beyond the EU requirements. Party Polit. 27, 953–964. doi: 10.1177/1354068820906785

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  BraunD. (2023). “Text analysis of party manifestos,” in The Routledge Handbook of Political Parties, eds. C. Neil, D. Keith, G. M. Sindre, and S. Vasilopoulou (London; New York, NY: Routledge). doi: 10.4324/9780429263859-46

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  BraunD.CartenyG. (2024). UNTWIST Manifesto Gender Analysis Codebook. Zenodo.

  • Google Scholar
  • View reference in article

• 8

  BurnhamM. (2025). Stance detection: a practical guide to classifying political beliefs in text. Polit. Sci. Res. Methods13, 611–628. doi: 10.1017/psrm.2024.35

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  BurnhamM.KahnK.WangR. Y.PengR. X. (2024). Political DEBATE: efficient zero-shot and few-shot classifiers for political text. Polit. Anal. 34, 329–343. doi: 10.1017/pan.2025.10028

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  BurstT.LehmannP.FranzmannS.Al-GaddooaD.IvanuschC.RegelS.et al. (2024). Manifestoberta. Manifesto Project. doi: 10.25522/MANIFESTO.MANIFESTOBERTA.56TOPICS.CONTEXT.2024.1.1

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  Cabeza PérezL.Alonso Sáenz de OgerS.Gómez FortesB. (2023). Ideology and party positions on gender issues in Spain: evidence from a novel data set. Polit. Gender19, 1204–1230. doi: 10.1017/S1743923X23000338

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  CampbellR.ChildsS. (2015). Conservatism, feminisation and the representation of women in UK politics. Br. Polit. 10, 148–168. doi: 10.1057/bp.2015.18

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  CampbellR.ErzeelS. (2018). Exploring gender differences in support for rightist parties: the role of party and gender ideology. Polit. Gender14, 80–105. doi: 10.1017/S1743923X17000599

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  CaravantesP. (2021). Tensions between populist and feminist politics: the case of the Spanish left populist party Podemos. Int. Polit. Sci. Rev. 42, 596–612. doi: 10.1177/0192512120931209

  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  CartenyG.ReinlA. K.BraunD.PopaS. A.SchmittH. (2023). European parliament election study 1979–2019, Euromanifesto study. Cologne3, 10–4232. doi: 10.4232/1.14120

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  ChueriJ.DamerowA. (2023). Closing the gap: how descriptive and substantive representation affect women's vote for populist radical right parties. West Eur. Polit. 46, 928–946. doi: 10.1080/01402382.2022.2113219

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  CofféH.KittilsonM. C.MeguidB. M.WeeksA. C. (2023). “Parties, issues and gender,” in The Routledge Handbook of Political Parties, 312–322. doi: 10.4324/9780429263859-33

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  de LangeS. L.MüggeL. M. (2015). Gender and right-wing populism in the Low Countries: ideological variations across parties and time. Patterns Prejudice49, 61–80. doi: 10.1080/0031322X.2015.1014199

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  DinglerS. C.KroeberC.Fortin-RittbergerJ. (2019). Do parliaments underrepresent women's policy preferences? Exploring gender equality in policy congruence in 21 European democracies. J. Eur. Public Policy26, 302–321. doi: 10.1080/13501763.2017.1423104

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  FarrisS. R. (2017). In the Name of Women's Rights. Durham, NC: Duke University Press. doi: 10.1215/9780822372929

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  GilardiF. (2015). The temporary importance of role models for women's political representation. Am. J. Polit. Sci. 59, 957–970. doi: 10.1111/ajps.12155

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  GraffA.KorolczukE. (2021). Anti-Gender Politics in the Populist Moment. New York: Taylor & Francis. doi: 10.4324/9781003133520

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  GreeneZ.O'BrienD. Z. (2016). Diverse parties, diverse agendas? Female politicians and the parliamentary party's role in platform formation. Eur. J. Polit. Res. 55, 435–453. doi: 10.1111/1475-6765.12141

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  GrossM.JankowskiM. (2020). Dimensions of political conflict and party positions in multi-level democracies: evidence from the Local Manifesto Project. West Eur. Polit. 43, 74–101. doi: 10.1080/01402382.2019.1602816

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  GwetK. L. (2008). Computing inter–rater reliability and its variance in the presence of high agreement. Br. J. Mathem. Statist. Psychol. 61, 29–48. doi: 10.1348/000711006X126600

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  HayesA. F.KrippendorffK. (2007). Answering the call for a standard reliability measure for coding data. Commun. Methods Meas. 1, 77–89. doi: 10.1080/19312450709336664

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  HomolaJ. (2019). Are parties equally responsive to women and men?Br. J. Polit. Sci. 49, 957–975. doi: 10.1017/S0007123417000114

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  HonnibalM.MontaniI.Van LandeghemS.BoydA. (2020). spaCy: Industrial-strength natural language processing in Python. Explosion. Available online at: https://spacy.io (Accessed January 03, 2026).

  • Google Scholar
  • View reference in article

• 29

  HughesM. M.PaxtonP. (2019). “The political representation of women over time,” in The Palgrave Handbook of Women's Political Rights (London: Palgrave Macmillan UK), 33–51. doi: 10.1057/978-1-137-59074-9_3

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  InglehartR.NorrisP. (2003). Rising Tide: Gender Equality and Cultural Change Around the World. 1st ed. Cambridge: Cambridge University Press. doi: 10.1017/CBO9780511550362

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  InglehartR.WelzelC. (2005). Modernization, Cultural Change, and Democracy: The Human Development Sequence. Cambridge: Cambridge University Press.

  • Google Scholar

• 32

  KantolaJ.LombardoE. (2017). Gender and the Economic Crisis in Europe. Cham: Springer International Publishing. doi: 10.1007/978-3-319-50778-1

  • CrossRef
  • Google Scholar

• 33

  KantolaJ.LombardoE. (2021). Strategies of right populists in opposing gender equality in a polarized European Parliament. Int. Polit. Sci. Rev. 42, 565–579. doi: 10.1177/0192512120963953

  • CrossRef
  • Google Scholar

• 34

  KittilsonM. C. (2011). Women, parties and platforms in post-industrial democracies. Party Polit. 17, 66–92. doi: 10.1177/1354068809361012

  • CrossRef
  • Google Scholar

• 35

  LehmannP.FranzmannS.Al-GaddooaD.BurstT.IvanuschC.et al. (2025). Manifesto Project Dataset (Version 2025a) [Data set]. Manifesto Project. doi: 10.25522/MANIFESTO.MPDS.2025A

  • CrossRef
  • Google Scholar

• 36

  LehmannP.ZobelM. (2018). Positions and saliency of immigration in party manifestos: a novel dataset using crowd coding. Eur. J. Polit. Res. 57, 1056–1083. doi: 10.1111/1475-6765.12266

  • CrossRef
  • Google Scholar

• 37

  LewandowskiJ.MerzN.RegelS. (2015). manifestoR: Access and Process Data and Documents of the Manifesto Project [Dataset]. CRAN: Contributed Packages. The R Foundation. doi: 10.32614/CRAN.package.manifestoR

  • CrossRef
  • Google Scholar

• 38

  LoweW.BenoitK.MikhaylovS.LaverM. (2011). Scaling policy preferences from coded political texts. Legislat. Stud. Quart. 36, 123–155. doi: 10.1111/j.1939-9162.2010.00006.x

  • CrossRef
  • Google Scholar

• 39

  LuxtonL.JabraD.BraunD. (2025). Anti-feminism in the spotlight: media coverage of German parties' gender and sexuality policies in the 2024 European Parliament election campaign. Eur. J. Commun. 40, 345–359. doi: 10.1177/02673231251349002

  • CrossRef
  • Google Scholar

• 40

  MeguidB. M.CofféH.WeeksA. C.KittilsonM. C. (2025). Strategic inclusion without transformation: how populist radical right parties engage with women's interests. Compar. Polit. Stud. doi: 10.1177/00104140251381760

  • CrossRef
  • Google Scholar

• 41

  MikhaylovS.LaverM.BenoitK. R. (2012). Coder reliability and misclassification in the human coding of party manifestos. Polit. Anal. 20, 78–91. doi: 10.1093/pan/mpr047

  • CrossRef
  • Google Scholar

• 42

  O'BrienD. Z. (2018). “Righting” conventional wisdom: women and right parties in established democracies. Polit. Gender14, 27–55. doi: 10.1017/S1743923X17000514

  • CrossRef
  • Google Scholar

• 43

  QiP.ZhangY.ZhangY.BoltonJ.ManningC. D. (2020). “Stanza: a python natural language processing toolkit for many human languages,” in Proceedings of the 58th annual meeting of the association for computational linguistics: system demonstrations, 101–108. doi: 10.18653/v1/2020.acl-demos.14

  • CrossRef
  • Google Scholar

• 44

  ReinlA. K.BraunD. (2023). European election studies 2019: manifesto project. GESIS, Köln. ZA7891 Datenfile Version.

  • Google Scholar

• 45

  RobertsonD. (1976). A Theory of Party Competition. London: Wiley &Sons.

  • Google Scholar

• 46

  RooduijnM.PirroA.HalikiopoulouD.FroioC.Van KesselS.De LangeS.et al. (2023). The PopuList 3.0. doi: 10.17605/OSF.IO/2EWKQ

  • CrossRef
  • Google Scholar

• 47

  RovnyJ.PolkJ.BakkerR.HoogheL.JollyS.MarksG.et al. (2025). The 2024 Chapel Hill Expert Survey on political party positioning in Europe: twenty-five years of party positional data. Elect. Stud. 97:102981. doi: 10.1016/j.electstud.2025.102981

  • CrossRef
  • Google Scholar

• 48

  SandersA.GainsF.AnnesleyC. (2021). What's on offer: how do parties appeal to women voters in election manifestos?J. Elect. Public Opin. Part. 31, 508–527. doi: 10.1080/17457289.2021.1968411

  • CrossRef
  • Google Scholar

• 49

  SpieringsN.ZasloveA. (2017). Gender, populist attitudes, and voting: explaining the gender gap in voting for populist radical right and populist radical left parties. West Eur. Polit. 40, 821–847. doi: 10.1080/01402382.2017.1287448

  • CrossRef
  • Google Scholar

• 50

  StahlP. M. (2026). Lingua: The Most Accurate Natural Language Detection Library for Python. Available online at: https://github.com/pemistahl/lingua-py (Accessed January 03, 2026).

  • Google Scholar

• 51

  VolkS. (2025). Rallying “round the drag: anti-gender mobilization and the mainstreaming of the far right. Eur. Societ. 27, 59–82. doi: 10.1162/euso_a_00009

  • CrossRef
  • Google Scholar

• 52

  VolkensA.BaraJ.BudgeI.McDonaldM. D.KlingemannH.-D. eds. (2013). Mapping Policy Preferences From Texts: Statistical Solutions for Manifesto Analysts. Oxford: Oxford University Press. doi: 10.1093/acprof:oso/9780199640041.001.0001

  • CrossRef
  • Google Scholar

• 53

  WeeksA. C.CaravantesP.Espírito-SantoA.LombardoE.StratigakiM.GulS. (2025). When is gender on party agendas? Manifestos and (De-)democratisation in Greece, Portugal, and Spain. South Eur. Soc. Polit. 30, 561–593. doi: 10.1080/13608746.2024.2412746

  • CrossRef
  • Google Scholar

• 54

  WlezienC. (1995). The public as thermostat: dynamics of preferences for spending. Am. J. Pol. Sci. 39:981. doi: 10.2307/2111666

  • CrossRef
  • Google Scholar

• 55

  WlezienC. (2004). Patterns of representation: dynamics of public preferences and policy. J. Polit. 66, 1–24. doi: 10.1046/j.1468-2508.2004.00139.x

  • CrossRef
  • Google Scholar