Introduction
European Parliament (EP) elections are the second-largest electoral event worldwide, topped only by India’s general elections in terms of voter turnout. However, the fragmented nature of the contest – characterized by parallel national campaigns and a weak transnational public sphere (Meyer and Gattermann Reference Meyer and Gattermann2022) – renders its analysis particularly challenging.
When investigating electoral behavior in EP elections, students of European politics are constrained by the limited availability of transnational datasets. Flagship resources include the panel-based European Electoral Study (EES) (European Election Studies Executive Board 2024) and Schraff et al.’s European Nomenclature of Territorial Statistical Units (NUTS)-level election dataset (EU-NED) (Schraff, Vergioglou and Demirci Reference Schraff, Vergioglou and Demirci2023). The EES collects data through about 1,000 interviews in most member states, while EU-NED reports parties’ vote shares at the lowest available regional tier since the 1983 election. To ensure spatial consistency with Eurostat’s regional statistics, Schraff et al. report the results for regional units listed in the EU’s official NUTS. The most fine-grained of the three NUTS regional levels, NUTS 3, divides the EU27 into 1,165 territorial units, with the population of individual NUTS 3 regions varying between 11,000 and 6.7 million as of 2021 (Eurostat 2023). In contrast, the official election page of the European Parliament reports only nationally aggregated results (European Parliament 2024a), with more fine-grained data being released only by national statistical offices. Earlier datasets such as Comprehensive European Parliament Electoral Data (COMEPELDA) (Däubler, Chiru and Hermansen Reference Däubler, Chiru and Hermansen2022) and the EP entries in the Constituency-Level Elections Archive (CLEA) (Kollman, Hicken, Caramani et al. Reference Kollman, Hicken, Caramani, Backer and Lublin2024) provided only national or district-level results.
The ability to analyze aggregate voting patterns below the regional tier has proved crucial in studies focusing on the effect of urbanization (de Dominicis, Dijkstra and Pontarollo Reference de Dominicis, Dijkstra and Pontarollo2020; Huijsmans and Rodden Reference Huijsmans and Rodden2025), residential segregation (Weaver and Bagchi-Sen Reference Weaver and Bagchi-Sen2015), local government performance (Weitz-Shapiro Reference Weitz-Shapiro2008), unemployment (Park and Reeves Reference Park and Reeves2020), ethnicity (Hersh and Nall Reference Hersh and Nall2016), or income (Leigh Reference Leigh2005) on political behavior. Variables such as urbanization, income, or migration status are often very heterogeneously distributed within regional administrative units, making statistical analyses at the regional tier much more susceptible to ecological fallacies (Wong Reference Wong, Janelle, Warf and Hansen2004). In a landmark study published in 2020, Dijkstra et al. relied on CLEA data to show that voting for parties opposing European integration in national elections correlates with economic decline, lower employment, and lower average education levels at the local level (Dijkstra, Poelman and Rodrguez-Pose Reference Dijkstra, Poelman and Rodrguez-Pose2020). After decades of scarcity, Europe-wide datasets reporting such variables are increasingly available at the local level, paving the way to more precise analyses of voting behavior in EP elections. As one such example, in 2023, the European statistical office Eurostat released GEOSTAT,Footnote 1 an EU-wide dataset of 13 census variables covering gender, age, employment, nationality, and population mobility on a 1 km2 grid. Yet, thus far, a corresponding local-level dataset of EP election results had been missing, restricting the research community’s ability to conduct fine-grained ecological analyses.
Our dataset. In this note, we bridge this gap by introducing BLUE_EP, the first dataset compiling all municipality-level results of the 2019 and 2024 EP elections. BLUE_EP was assembled by combining and harmonizing, for each year, the 27 official result datasets published by member states’ electoral authorities. We use Eurostat’s nomenclature of Local Administrative Units (LAU) as our primary spatial unit, addressing special cases such as STV constituencies, postal voting, and diaspora polling stations through straightforward extensions of the official typology. This allows for easily integrating BLUE_EP with European and national data sources. Furthermore, every party in BLUE_EP is classified according to its EP group or European party affiliation at the start of the legislative term, if any, and linked to a Party Facts (Döring and Regel Reference Döring and Regel2019) identifier. All competing parties, including non-affiliated parties, are further classified into four broad political families (‘left’, ‘center-right’, ‘populist radical right’, and ‘others’). In total, our dataset covers 90,000 spatial units and 500 parties for each of the two elections, making it one of the largest datasets ever released for single electoral events.
For the first time, BLUE_EP allows for seamless transnational analyses of the results of the 2019 and 2024 elections at the lowest administrative level, including analyses of the effects of various characteristics of municipalities on voting behavior. Additionally, our dataset provides a unified resource to access EP election results. BLUE_EP subsumes all existing national and (regional) European datasets for 2019 and 2024 while providing additional party classification, thereby significantly simplifying the extraction of descriptive statistics at the European level. Through the development of BLUE_EP, we also seek to set a standard for the construction of comprehensive transnational datasets of local-level election results, including for future EP elections.
This note is organized as follows. First, we recall the main characteristics of European Parliament elections and the European Union’s party system. Then, we briefly present the available data sources and our data collection process. Next, we introduce our extended typology of spatial units based on Eurostat’s LAU and review our party classification. Finally, to showcase the relevance of our dataset for electoral studies, we present a case study that combines our dataset with Eurostat urbanization data. Our case study provides empirical evidence for a growing urban-rural gap in voting behavior between the 2019 and 2024 EP elections, leveraging the fine granularity of BLUE_EP’s data.
The BLUE_EP dataset was created as part of our work on the Electoral Bulletins of the European Union (BLUE), a pro-bono project that publishes analyses and electoral data for all elections in the European Union. The dataset is open data and available on Zotero.
European Parliament elections and the EU’s party system
The European Parliament (EP), based in Strasbourg and Brussels, is the parliamentary assembly of the European Union. As a co-legislator and the sole European Union (EU) institution whose members are elected by popular vote, the EP plays a central democratic role in the EU’s political system (Hix and Høyland Reference Hix and Høyland2022). Since 1979, the EP has been elected by direct universal suffrage on the basis of national and regional lists (Viola Reference Viola2015). Over the course of four decades, the mechanics of EP elections have seen little change. Every five years in late May or early June, all EU citizens are called to the polls to elect the members of the next European legislature. Depending on national electoral traditions, the exact voting day may differ in each member state, but all voting operations take place within the same week. National political parties file national or, in some member states, regional lists competing for the votes of the respective country’s residents. Since 1992, EU citizens holding the citizenship of one member state but residing in another are free to register to vote and stand for election in either (Viola Reference Viola2015). The introduction of transnational lists has long been in discussion (Viola Reference Viola2015; Hoffmeister Reference Hoffmeister2020), but such lists have so far never been used. Instead, all of the parliament’s seats (720 as of 2025) are distributed among the member states according to their population, using a method of degressive proportionality that favors smaller states.
The voting systems used for apportioning national seats differ. All but two member states use various flavors of party-list proportional representation (PR) with or without preference votes, in one or multiple constituencies. In these member states, each voter selects one list of candidates in their constituency or, in some member states, an independent candidate. The constituencies’ seats are then distributed to the lists using an apportionment algorithm that varies by member state (Oelbermann and Pukelsheim Reference Oelbermann and Pukelsheim2020). Only two member states, Malta and Ireland, use Single Transferable Vote (STV) in multi-member constituencies. In STV, voters rank candidates by preference in multi-seat constituencies. Candidates reaching the electoral quota via first preferences or transfers are elected; otherwise, the lowest-ranking candidates are eliminated iteratively until all seats fill. In this system, the count of first-preference votes aggregated by candidates’ partisan affiliations comes closest to a party-level preference, even though these aggregate party-level results are not directly reflected in the seat apportionment due to the candidate-level counts. Ireland, in particular, has a history of electing many independent members of the European Parliament (MEPs) (Murphy and O’Brennan Reference Murphy, O’Brennan, Coakley, Gallagher, O’Malley and Reidy2023).
Unlike general elections in large federal polities, EP elections are mostly contested at the member-state level (Meyer and Gattermann Reference Meyer and Gattermann2022). Transnational political campaigning exists, but its visibility and impact are limited compared to more traditional national campaigns (Stier, Froio and Schünemann Reference Stier, Froio and Schünemann2021). For this reason, European elections are frequently analyzed as second-order elections (Ehin and Talving Reference Ehin and Talving2021; Stier, Froio and Schünemann Reference Stier, Froio and Schünemann2021), or more recently, as intermediate between first-order and second-order elections, with both transnational positions regarding EU integration and national concerns playing a major role (Wilhelm, Kritzinger, Plescia et al. Reference Wilhelm, Kritzinger, Plescia, Raube, Wouters, Kritzinger, Plescia, Raube, Wilhelm and Wouters2020; Beaudonnet, Belot, Le Gall et al. Reference Beaudonnet, Belot, Le Gall and Van Ingelgom2024). Despite these national biases, the fact that all European citizens are called to vote during the same week makes them a useful proxy to capture transnational political trends. Moreover, opinion polling in recent European elections has provided some evidence for the emergence of common political themes (climate, migration, international security) (European Parliament 2024b) playing a central role in the campaign, thus suggesting the slow emergence of a form of shared public and political sphere (Braun Reference Braun2021; Della Porta Reference Della Porta2022).
Prior to the constitutive session of each legislature, the elected MEPs regroup into transnational political groups (Bressanelli Reference Bressanelli, Ahrens, Elomäki and Kantola2022; Hix and Høyland Reference Hix and Høyland2022). Political groups in the EP function similarly to political groups in national or regional parliaments. Additionally, they serve as European party coalitions, uniting a large number of heterogenous national parties into a small set (five to eight) of European political groups that play a key role in parliamentary procedure (Hix and Høyland Reference Hix and Høyland2022), maintain an identifiable ideological profile and, at least in the case of centrist groups, display a high degree of cohesion in their voting behavior (Hix, Noury and Roland Reference Hix, Noury and Roland2005; Bowler and McElroy Reference Bowler and McElroy2015; Hublet, Lanoë and Schleyer Reference Hublet, Lanoë and Schleyer2023). Since 1984, all EP legislatures have had at least one radical-left group; one Social Democratic group; one Green group; one center-right liberal group; one Christian Democratic group; one right-wing conservative group; and one euroskeptic or populist radical right (PRR) group (Kaiser and Mittag Reference Kaiser and Mittag2023). The two largest groups, the Christian Democrats and Social Democrats, have provided the core of centrist coalitions in the EP since its early years (Crum Reference Crum, Kritzinger, Plescia, Raube, Wilhelm and Wouters2020). As of 2024, all European groups are structured around one or more so-called European parties that federate most of their MEPs’ national parties. However, European groups still welcome in their ranks MEPs whose national parties are not members of their core European parties, as well as some non-affiliated MEPs. As in most national parliaments, MEPs who fail to become members of any European group (mostly members of far-left, left-wing nationalist, far-right, or satirical parties) are given a non-attached status known in the EP by its French name of ‘non-inscrits’.
Data sources and data collection
Municipality-level election results for the 2019 and 2024 EP elections are released online by national statistical offices and electoral commissions. For 22 of 27 member states (all except Belgium, Greece, Slovenia, Ireland, and Malta), the full breakdown of voter registration, turnout, and party votes is available at the level of Eurostat’s LAU. In the five remaining states, some or all of that data is available only in larger administrative units. The BLUE_EP dataset collects data in all 27 member states at the lowest available level of administrative units. We first review the procedure followed to constitute our dataset in the 22 member states with LAU-level data, and then discuss the modified approach followed in the remaining 5.
Main procedure
In the 22 countries for which full data were available, the dataset was constructed through the following steps:
-
1. Official municipality-level national results were collected from the websites of member states’ statistical offices or electoral commissions. When available, a machine-readable format was preferred. Where no such format was available, scraping was used to automate the data collection.
-
2. For each member state, a typology of spatial units was developed, extending the LAU typology as presented in Section Typology of spatial units. Similarly, the list of competing parties and independent lists was extracted and enriched with group membership information and Party Facts identifiers (Döring and Regel Reference Döring and Regel2019) as presented below.
-
3. For each spatial unit, the following information was extracted: the number of registered voters; the number of voters turning out (if the numbers of voters’ signatures and ballots in the ballot box were reported separately and did not match, the latter was used); the number of ‘blank’ or ‘none of the above’ votes, if this option existed; the number of invalid votes, excluded blank votes; and the number of votes for each party or independent list. For Ireland and Malta, which use STV, we report only first preferences, aggregated by party.
-
4. The total number of votes obtained by each party at the national level was checked against the total figures reported at the national level, and possible discrepancies were corrected. A few remaining minor discrepancies are reported in the dataset’s codebook. They correspond to cases of mismatch between municipality-level and national data, mostly due to misreporting at the regional level. In case of such discrepancies, the officially released local data was always preferred, except if explicit local-level corrections were provided by the national election authority.
Special cases
In one member state, Belgium, voter registration data is only released at the level of ‘electoral cantons’, while all other data is available at the municipality level. As a result, in Step 3, we set the voter registration numbers to zero for all Belgian municipalities, instead reporting the number of registered voters in special rows at the lowest available regional level, NUTS 3. The extraction process is otherwise unchanged.
Greece reports detailed results in each of its 392 municipalities (dímoi), but the LAU nomenclature uses the 6,132 smaller koinótites. We use a modified Step 2 where the dímoi are used instead of the koinótites.
Slovenia reports detailed results in 89 electoral districts, with most districts spanning multiple municipalities. Again, we use a modified Step 2 with these 89 electoral districts.
Ireland and Malta are the only two member states using STV rather than party-list PR. As is common in STV elections, both countries report only constituency-level results. In the 2019 and 2024 EP elections, there were three constituencies in Ireland and a single constituency in Malta. In both countries, we use the sum of first-preference votes of all of a party’s candidates as a proxy for party results. A similar approach is followed in COMEPELDA (Däubler, Chiru and Hermansen Reference Däubler, Chiru and Hermansen2022) and in the main file of the CLEA database (Kollman, Hicken, Caramani et al. Reference Kollman, Hicken, Caramani, Backer and Lublin2024).
Typology of spatial units
Our typologies of spatial units are a natural extension of Eurostat’s LAU nomenclature for the corresponding election year. LAU codes are variable-length identifiers of the form CC_
$m$
where CC is the two-letter code of the respective member state and
$m$
is a variable-length municipality identifier that generally corresponds to a national municipality code. Each LAU is included in a single NUTS 3 region that we also report, allowing regional data to be seamlessly extracted via grouping and aggregation.
To accommodate special reporting units used by member states’ authorities, we extended this topology in five ways.
First, for the 2024 election only, the codes of all local units not yet included in the latest release of LAU (2023) are added, using national municipality codes as a reference.
Second, we introduce separate spatial units larger than a single municipality that are used to report postal votes (Austria, Germany, Hungary), voter registration (Belgium), or votes not assigned to any local unit (Portugal, Slovenia). Whenever possible, each special spatial unit is mapped to a single NUTS 3 in which its voters habitually reside. If no single NUTS 3 region can be determined, a special pseudo-NUTS region coded CCYYY is used. For example, the Hungarian postal vote, being only reported at the aggregate level, is mapped to a single pseudo-region coded HUYYY.
Third, we introduce special spatial units for citizens residing abroad in the countries where data about their voting behavior is available. In Bulgaria, Cyprus, France, Croatia, Italy, Poland (2024 only), Portugal, and Romania, fine-grained data by country of residence is available. A pseudo-NUTS code of the form CCZDD is assigned to the corresponding spatial units, where CC is the member state’s code, and DD is the ISO code of the state of the corresponding voters’ residence, if available. For instance, entries reporting the vote of Bulgarian citizens residing in Czechia are mapped to the pseudo-region BGZCZ. This allows for a more convenient analysis of diasporas’ spatial voting patterns. In Belgium, Estonia, Greece, Hungary, Lithuania, Latvia, Poland (2019 only), and the Netherlands (2024 only), votes from abroad are reported without information about the country of residence. A special NUTS code CCZZZ is assigned to the corresponding rows, e.g., EEZZZ for Estonian voters residing abroad. Among these, Spain and Belgium provide a fine-grained breakdown by group of provinces (Belgium) or province (Spain) of last residence within the country. In the remaining countries, the votes of citizens residing abroad are not registered separately; they either vote by post or at diplomatic representations (Austria, Germany, Denmark, Finland, Luxembourg, Sweden, Slovenia), only by traveling to their country of citizenship (Italian citizens living outside of the EU), or sometimes not at all (Czechia, Ireland, Malta, Slovakia; Bulgarian and Danish citizens living outside of the EU).
Fourth, we add codes for all Overseas Countries and Territories (OCTs) of member states for which detailed results are reported. This concerns five territories under French sovereignty (French Polynesia, New Caledonia, Saint-Barthélémy, Saint-Pierre-et-Miquelon, Wallis-et-Futuna) and the three special municipalities of the Netherlands (Bonaire, Sint Eustatius, Saba), none of which is part of the EU. As for municipalities not included in the last LAU release, we use national codes as a basis for our extension.
Fifth and finally, we create simple numerical codes for the electoral districts of the three member states without municipality-level results (Malta, Ireland, Slovenia), assigning them a NUTS 3 region, if possible.
Table 1 gives an overview of the spatial units in BLUE_EP. In total, BLUE_EP contains 90,062 spatial units for 2019 and 89,916 for 2024, corresponding to 352,618,920 and 355,147,948 registered voters, respectively. Owing to the member states’ different administrative structures, the average population per spatial unit varies significantly, from only 1,303 voters per unit in Czechia to 43,890 in Denmark in 2024 (excluding Ireland and Malta). Of all administrative units, more than a third are located in France, which has the second-lowest average population by unit at 1,404. BLUE_EP provides fine-grained data on diaspora voting in Bulgaria, Cyprus, France, Croatia, Italy, Poland (2024 only), Portugal, and Romania, for a total of 601 (2019) and 706 (2024) spatialized observations.
Spatial units in BLUE_EP
Table 1. Long description
The table provides an overview of the spatial units in BLUE_EP for the years 2019 and 2024. It includes data on the number of units and registered voters for each region. The table has 30 rows and 7 columns. The columns are labeled as Level, Units, Abroad, OCT, and Special for both 2019 and 2024. The row labels include country codes and the type of municipality or constituency. Each row lists the number of units and their distribution across different categories for the specified years. For example, in 2019, Austria (Muni.) had 2,190 units, 93 of which were special, while in 2024, it had 2,187 units, 94 of which were special. The table also includes data for regions like Belgium, Bulgaria, Czechia, Denmark, Estonia, Greece, Spain, Finland, France, Croatia, Hungary, Ireland, Italy, Lithuania, Luxembourg, Latvia, Malta, Netherlands, Poland, Portugal, Romania, Sweden, Slovenia, and Slovakia. Notable trends include the consistent number of units in some regions and significant changes in others.
*Registered voters at NUTS 3 level only å
Dímoi
$ \bullet $
Results by city/country of residence
Units: number of spatial units. Abroad: number of units corresponding to voters residing abroad.
OCT: number of units corresponding to Overseas Countries and Territories (OCT) of the EU, which are not EU territory. Special: number of special spatial units (see Section Typology of spatial units).
Party classification
We classify all parties and independent lists competing in the 2019 and 2024 EP elections using the following decision procedure:
-
1. If the party belonged to a political group in July 2019 or 2024, respectively, assign it to this group. This excludes suspended members.
-
2. Otherwise, if, at that date, the party belonged to one of the European parties listed in Table 3, assign it to the corresponding group. This includes parties with observer or associate status in a European party, but excludes suspended members.
-
3. Otherwise, if a party was in the past a member of a political group, or a direct ancestor of that political group, was not suspended from that group at the time it lost parliamentary representation, and did not regain parliamentary representation since, assign it to this group.
-
4. Otherwise, if the overall ideological position of the party on a left-right scale can be estimated from their public positions and political program, classify the party as Other (left), Other (center-right), or Other (radical right).
-
5. Else, classify the party as Other.
When, in a given member state, a single list of candidates aggregates parties belonging to different political groups as by the above decision procedure, the corresponding list is labeled as an EP group coalition. Table 2 provides statistics about the different groups.
Party classification in BLUE_EP
Table 2. Long description
The table presents party classifications in the European Parliament (EP) for the years 2019 and 2024. It is organized into four main party families: Left, Center-right, Radical right, and Other. Each family is further divided into specific EP groups or coalitions, with corresponding parties listed for both years. The table has four columns: Party family, EP group or coalition, Parties (2019), and Parties (2024). Row 1: Left, GUE/NGL, 34, 30. Row 2: Left, GUE/NGL-Greens/EFA, 6, 13. Row 3: Left, GUE/NGL-S&D, 1, 1. Row 4: Left, Greens/EFA, 38, 46. Row 5: Left, Greens/EFA-S&D, 1, 1. Row 6: Left, S&D, 38, 29. Row 7: Left, Other (left), 104, 94. Row 8: Center-right, RE, 55, 46. Row 9: Center-right, EPP, 45, 44. Row 10: Center-right, EPP-ECR, 1, 1. Row 11: Center-right, Greens/EFA-S&D-RE-EPP, 1, 1. Row 12: Center-right, Greens/EFA-EPP, 1, 1. Row 13: Center-right, Other (center-right), 57, 59. Row 14: Radical right, ECR, 19, 23. Row 15: Radical right, ID, 14, 16. Row 16: Radical right, PfE, 7, 16. Row 17: Radical right, ESN, 7, 7. Row 18: Radical right, Other (radical right), 103, 98. Row 19: Other, Other, 65, 125. Row 20: Other, Greens/EFA-RE, 1, 1. Row 21: Other, S&D-RE, 1, 1. Row 22: Other, S&D-EPP, 1, 1.
GUE/NGL: The Left in the European Parliament. Greens/EFA: Greens–European Free Alliance.
S&D: Progressive Alliance of Socialists and Democrats. RE: Renew Europe.
EPP: European People’s Party. ECR: European Conservatives and Reformists Group.
ID: Identity and Democracy. PfE: Patriots for Europe. ESN: Europe of Sovereign Nations.
Constitutive parties of EP groups used for classification in BLUE_EP
Table 3. Long description
A table with two columns and ten rows, detailing the constitutive parties of EP groups used for classification. The first column is labeled ‘EP group’ and the second column is labeled ‘European party’. The table lists the following: Row 1: GUE/NGL, European Left Alliance; Party of the European Left; Animal Politics EU. Row 2: Greens/EFA, European Green Party; European Free Alliance; European Pirate Party; Volt Europe. Row 3: S&D, Party of European Socialists. Row 4: RE, Alliance of Liberals and Democrats for Europe; European Democratic Party. Row 5: EPP, European People’s Party. Row 6: ECR, European Conservatives and Reformists. Row 7: ID, Identity and Democracy. Row 8: PfE, Patriots for Europe. Row 9: ESN, Europe of Sovereign Nations.
The political groups are further aggregated into four broader political families: ‘left’, ‘center-right’, ‘radical right’, and ‘other’. The ‘left’ family contains all parties of the GUE/NGL, Greens/EFA, and S&D groups, together with the Other (left). The ‘center-right’ family contains all parties of the RE and EPP groups together with the Other (center-right). Finally, the ‘radical right’ family contains all parties of the ECR, ID/PfE, and ESN groups together with the Other (radical right). This coarser classification can ease transnational comparison when studying the electoral performance of broad ideological blocks, since national parties with a broadly similar ideological profile may belong to different groups within the same political family, or be non-affiliated. For example, while in 2024 the Alternative for Germany was a member of the ESN, the French National Rally was a member of the PfE, and the Spanish The Party is Over party remained non-affiliated, all three parties would likely need to be captured in the same category for the purpose of a transnational analysis of the PRR vote. When a more fine-grained analysis of ideological differences is needed – e.g., for analyzing the evolution of Green or radical-left votes – a grouping by EP group can be preferable. The BLUE_EP dataset provides both levels of aggregation.
The absence of separate ‘radical left’ and ‘center-left’ categories is asymmetrical vis-à-vis the treatment of right-of-center parties. However, we believe that such treatment is pragmatically warranted for at least three reasons. First, the divide between ‘radical left’ and ‘center-left’ is harder to establish by combining EP group affiliations than the divide between ‘radical right’ and ‘center-right’: historically, the voting patterns of the GUE/NGL, Green, and S&D groups in the EP have been far more similar to each other than the voting behavior of the ECR has been to the EPP’s (McElroy and Benoit Reference McElroy and Benoit2007; Hublet, Lanoë and Schleyer Reference Hublet, Lanoë and Schleyer2023). Second, this three-way classification acknowledges the existence of a structural divide within the European right, reflected by the existence of formal or semi-formal ‘cordon sanitaire’ policies (Servent Reference Servent2019; Kantola and Miller Reference Kantola and Miller2021), with no similarly sharp divide existing on the left. Third, our classification yields three macrofamilies of roughly similar size that support a simpler analytical framework, rather than three large and one very small ‘radical left’ family. Since classifying the Greens/EFA group as uniformly ‘radical’ does not conform to established typologies, only the GUE/NGL – the EP’s smallest group – and some NI members could have been labeled as ‘radical left’, accounting for only 7% of all members.
As an alternative, our primary party classification can be refined by mobilizing classifications from the Chapel Hill Expert Survey (Rovny, Polk, Bakker et al. Reference Rovny, Polk, Bakker, Hooghe, Jolly, Marks, Steenbergen and Vachudova2025), the Manifesto Project (Lehmann, Franzmann, Al-Gaddooa et al. Reference Lehmann, Franzmann, Al-Gaddooa, Burst, Ivanusch, Lewandowski, Regel, Riethmüller and Zehnter2025), or other sources, which explicitly distinguish between left-right, GAL-TAN, and mainstream-populist dimensions.
Coalitions spanning across political family boundaries are assigned to the family of their dominant political ideology, if any, and to ‘other’ otherwise. In the 2024 election, the two main coalitions spanning across political families are the Polish European Coalition (Greens/EFA-S&D-RE-EPP) and the National Coalition for Romania (S&D-EPP). The former is assigned to the ‘center-right’ family, being strongly dominated by Donald Tusk’s Civic Platform (PO, EPP) with minor participation of social-democratic and Green parties. The latter is assigned to ‘other’, since its two constitutive parties, the PSD (S&D) and PNL (EPP), are of comparable strength and belong to different political families. The codebook provides further details about our party classification.
For each party, the BLUE_EP dataset contains the party’s denomination in the original national database; its member state; its abbreviation, if any; its political group and family; and its Party Facts (Döring and Regel Reference Döring and Regel2019) identifier, if any. The Party Facts dataset provides a central platform linking global data on over 45,000 political parties. At the time of writing, the Party Facts platformFootnote 2 lists parties from over 200 countries, featuring data linked from over 60 different datasets. The unique Party Facts identifier allows for seamlessly combining information from all these sources.
Case study: A growing urban-rural divide
To demonstrate the potential uses of our dataset, we now report on an analysis of the evolution of the urban-rural divide in voting behavior between the 2019 and 2024 EP elections.
Research questions
In recent years, the existence of a growing urban-rural divide in Western democracies has been the subject of intense scrutiny in both academic research and general-audience media. The hypothesis of a historically high, and in most cases still growing, urban-rural divide, defined as the difference in voting behavior between urban and rural communities of the same polity, has been empirically backed in the case of Canada (Armstrong, Lucas and Taylor Reference Armstrong, Lucas and Taylor2022; Taylor, Lucas, Armstrong et al. Reference Taylor, Lucas, Armstrong and Bakker2024), the United States (Gimpel, Lovin, Moy et al. Reference Gimpel, Lovin, Moy and Reeves2020; Taylor, Lucas, Armstrong et al. Reference Taylor, Lucas, Armstrong and Bakker2024), Great Britain (Taylor, Lucas, Armstrong et al. Reference Taylor, Lucas, Armstrong and Bakker2024), Germany (Haffert and Mitteregger Reference Haffert and Mitteregger2023), France (Brookes and Cappellina Reference Brookes and Cappellina2023), Poland (Marcinkiewicz Reference Marcinkiewicz2018), Hungary (Collini Reference Collini2021), and Sweden (Rickardsson Reference Rickardsson2021). In all of these polities, urban areas tend to be characterized by a higher level of left, Green, and pro-European votes, while rural areas are characterized by a higher average share of PRR and euroskeptic votes. With some exceptions, liberal and centrist parties tend to be more popular in urban areas, while traditional conservative parties obtain better results in rural communities. As this urban-rural divide may be the result of both compositional and contextual factors (Kenny and Luca Reference Kenny and Luca2021; Garca del Horno, Rico and Hernández Reference Garca del Horno, Rico and Hernández2024), differences between urban and rural voting behavior do not, on their own, demonstrate the existence of a causal effect of urbanization on political preferences. Rather, in order to isolate the contribution of urbanization, other potential explanatory factors such as income, educational attainment, age, and migrant background, which are all unevenly distributed between urban and rural areas, must be controlled for. Overall, the preference for left, center-left, and center-right parties in urban areas appears to correlate with a larger share of better-off, more educated, and younger voters and a lower share of migrant population, while the higher prevalence of PRR votes in rural areas can be explained in part by a larger share of middle-class, less educated, and middle-aged voters and a lower share of migrant population. To avoid ecological fallacies, relying on either survey data (Kenny and Luca Reference Kenny and Luca2021) or local-level election results (de Dominicis, Dijkstra and Pontarollo Reference de Dominicis, Dijkstra and Pontarollo2020) rather than regional-level data is advisable.
Overall, researchers and media commentators have advanced two types of claims about the urban-rural voting gap: purely descriptive (unconditional) claims that outline a difference in voting behavior between rural and urban areas; and claims of net associations between urbanization and voting behavior, controlling for other confounding factors. While the latter claims are most interesting from the perspective of political sociology, we believe that establishing clear descriptive facts is also useful from the viewpoint of European political geography. Independently, claims may either target a single election (existence of an urban-rural voting gap) or focus on election-to-election change (widening or reduction in that voting gap). Finally, the results of different party families may be considered, with PRR and left-of-center parties being the most frequently scrutinized. To the best of our knowledge, no fully transnational study of the urban-rural divide in EP elections has examined these claims. In the following, we use BLUE_EP to assess the following hypotheses:
(H1) In EP elections, rural areas display higher support for PRR parties and lower support for left-of-center parties, and urban areas display lower support for PRR parties and higher support for left-of-center parties.
(H2) Controlling for confounding factors, urban voters are more likely to vote for left-of-center parties and less likely to vote for right-of-center parties; the opposite associations are observed in rural voters.
(H3) The differences in average support for PRR and left-of-center parties between urban and rural areas have increased in the course of the last European legislative term.
(H4) Controlling for confounding factors, the association between urban or rural status on PRR and left-of-center vote has increased in the course of the last European legislative term.
In the case of European elections, one might ask whether a certain hypothesis is verified in an EU-wide aggregate or, more strongly, consistently in a majority of member states. In turn, the notion of a ‘majority of member states’ can be qualified to either denote a majority of member states or a number of member states being home to a majority of the EU’s population. Here, we will simply report the number of countries for which each hypothesis holds, as well as the share of the population they represent.
Data and methods
In addition to BLUE_EP, we use local-level and regional-level data published by Eurostat to test the above hypotheses. Eurostat’s Degree of Urbanization (DEGURBA) classification assigns every LAU to one of three categories according to its degree of urbanization: cities, towns and suburbs, and rural areas. The classification is homogeneous across member states and relies on an internationally recognized methodological standard (The European Commission and United Nations Human Settlements Programme 2021). Additionally, we use areal interpolation using Eurostat’s geospatial boundaries of municipalities to project Eurostat’s GEOSTAT dataset onto our local administrative units. Finally, Eurostat provides rich regional statistical databases available at the NUTS 2 and NUTS 3 levels.
Figure 1 shows the results of political groups in the 2019 and 2024 EP elections. For the sake of transnational comparison, we focus on the results of political families. Moreover, to be able to use local and regional Eurostat data and take into account the share of non-voters, we exclude the five countries (Belgium, Greece, Ireland, Malta, Slovenia) without full LAU-level data, including turnout. Together, these five member-states account for less than 7% of the EU’s population, and hence we do not expect their exclusion to significantly affect the validity of our results. We also exclude special voting sections and voting sections in OCTs, for which spatialized socio-demographic data are not available. Further, we group non-voting and invalid or blank votes into a single ‘non-vote’ option. We obtain a final dataset of about 87,000 observations, including five vote shares (non-voters, left, center-right, radical right, others), two dummies for cities and rural areas (baseline: towns and suburbs), and 10 control variables.
Political groups’ total vote shares.
To test (H1), we aggregate the municipality-level results of party families by degree of urbanization, as provided by BLUE_EP, and report the results at both EU and national levels. The hypothesis is deemed validated if a party family’s results increase or decrease linearly between rural areas, towns and suburbs, and cities, as stated in the corresponding hypothesis. In particular, it is not considered validated if the results reach a maximum or minimum in towns and suburbs.
To test (H2), we use EU-wide weighted least-square regressions with country fixed effects and weights equal to the number of registered voters by municipality to estimate the effect of urbanization on voting behavior. We chose weighted least-squares models over ordinary least-square models due to the very heterogeneous size of EU municipalities: on average, Dutch municipalities are 25 times more populous than their French counterparts. Hypothesis (H2) is deemed validated if the effects observed on the two dummy variables have the inverse signs stated in the corresponding hypothesis. In particular, it is not considered validated if the signs are equal.
To test (H3), we consider the difference in vote shares of the party families between 2019 and 2024, between cities and rural areas.
Finally, to test (H4), we extend the models for (H2) with year fixed effects and two interaction variables
$year \times cities$
and
$year \times rural$
capturing the change in the effect of urbanization over time.
For (H2) and (H4), we select the following variables as controls: share of population over 65, share of population born outside the EU, share of population born in the EU outside of their country of residency, share of population who did not change residence within the last year (GEOSTAT); residence within 20 km of the border; share of population with tertiary education (NUTS 2); GDP per capita at purchasing-power parity, natural growth rate of the population, and share of industrial and agricultural activities in gross value added (NUTS 3). These variables capture standard explanatory factors of voting behavior, including age, internal and external migration, wealth, education, and economic activity.
Results
(H1). The aggregated EU-wide results of party families in the three categories of municipalities are shown in Table 4. At EU-level, they support (H1) for both PRR and left-of-center parties in both years.
Party families’ vote shares by degree of urbanization
Long description
The table compares party families’ vote shares by degree of urbanization in the European Union for the years 2019 and 2024. It has two main sections, one for each year, with three rows each representing cities, towns and suburbs, and rural areas. Each section contains columns for Left, Center-right, Radical right, Other, and Non-vote. Row 1: Cities, Left 23.0, Center-right 18.3, Radical right 8.9, Other 0.8, Non-vote 49.0. Row 2: Towns and suburbs, Left 20.9, Center-right 18.5, Radical right 13.1, Other 0.5, Non-vote 47.0. Row 3: Rural areas, Left 16.8, Center-right 19.1, Radical right 13.3, Other 0.5, Non-vote 50.3. Row 4: Cities, Left 20.9, Center-right 16.6, Radical right 10.7, Other 1.3, Non-vote 50.5. Row 5: Towns and suburbs, Left 17.4, Center-right 17.4, Radical right 15.3, Other 1.4, Non-vote 48.4. Row 6: Rural areas, Left 12.9, Center-right 16.7, Radical right 17.3, Other 3.1, Non-vote 49.9.
A breakdown by country is shown in Figures 2 and 3. For left-wing parties in 2019, (H1) is supported in 13 countries, accounting for 55% of the sample’s registered voters. In 2024, it is supported in 14 countries, accounting for 71% of registered voters. For PRR parties in 2019, (H1) is supported in only 11 countries, accounting for 62% of registered voters. This increases to 16 countries in 2024, accounting for 79% of registered voters. Overall, (H1) is supported in a majority of countries for both years and party families.
Results of the 2019 EP election.
Results of the 2024 EP election.
Figure 3. Long description
The image contains multiple bar graphs showing party family shares by member-state and municipality size for the 2024 European Parliament election. Each bar graph represents a different member-state, with the horizontal axis indicating the percentage of registered voters and the vertical axis listing the member-states. The graphs are divided into sections for rural areas, towns, and cities. Each bar is color-coded to represent different party families: Left, Center-right, Radical right, Other, and Abstention. The data shows the distribution of voter preferences across these categories for each member-state. The graphs highlight variations in party support based on the size of the municipality, with some member-states showing significant differences between rural areas, towns, and cities.
(H2). The results of our EU-wide models are shown in Table 5. They show statistically significant effects of urbanization on the performance of most political families and the share of non-voters. In both elections, cities vote more frequently for left-wing parties and less frequently for radical-right parties. The coefficients for rural areas have opposite signs. This supports (H2).
Weighted least squares regression (reference category: towns and suburbs)
Table 5. Long description
The table presents the results of weighted least squares regression analysis, comparing political party preferences and non-voting behavior in cities and rural areas for the years 2019 and 2024. The table has 10 rows and 7 columns. The columns are labeled as Left, Center-right, Radical right, Other, Non-vote, and Controls. The rows are labeled as Cities, Rural areas, Controls, Fixed effects, and R-squared. Row 1: Left, Cities, 1.66***, Rural areas, -0.89***, Controls, Yes, Fixed effects, Yes, R-squared, 0.74. Row 2: Left, Cities, 2.45***, Rural areas, -0.91***, Controls, Yes, Fixed effects, Yes, R-squared, 0.72. Row 3: Center-right, Cities, -0.31***, Rural areas, -0.25***, Controls, Yes, Fixed effects, Yes, R-squared, 0.59. Row 4: Center-right, Cities, 0.29***, Rural areas, -0.91***, Controls, Yes, Fixed effects, Yes, R-squared, 0.64. Row 5: Radical right, Cities, -1.64***, Rural areas, 1.11***, Controls, Yes, Fixed effects, Yes, R-squared, 0.72. Row 6: Radical right, Cities, -1.86***, Rural areas, 1.76***, Controls, Yes, Fixed effects, Yes, R-squared, 0.64. Row 7: Other, Cities, 0.20***, Rural areas, -0.05***, Controls, Yes, Fixed effects, Yes, R-squared, 0.77. Row 8: Other, Cities, -0.34***, Rural areas, 0.96***, Controls, Yes, Fixed effects, Yes, R-squared, 0.77. Row 9: Non-vote, Cities, 0.10, Rural areas, 0.08, Controls, Yes, Fixed effects, Yes, R-squared, 0.66. Row 10: Non-vote, Cities, -0.56***, Rural areas, -0.90***, Controls, Yes, Fixed effects, Yes, R-squared, 0.60. The table shows the coefficients for different political parties and non-voting behavior in cities and rural areas, highlighting significant effects of urbanization on political preferences and voting behavior.
*
$p{\rm{\lt }}0.05$
**
$p{\rm{\lt }}0.01$
***
$p{\rm{\lt }}0.001$
.
Next, we run 22 separate state-level models. For the PRR vote in 2024, we observe a significant (
$p{\rm{\lt}}\;0.05$
), positive effect in cities and a significant, negative effect in rural areas in 9 of 22 member states, accounting for 57% of registered voters. For the left-of-center vote in 2024, significant inverse effects are observed in 10 member states, accounting for 60% of registered voters. For both party families in 2024, (H2) is supported in a majority of member states only when taking into account population counts. The urban-rural gap is driven by large member states: out of 10 member states with a population larger than 10 million,Footnote
3
a significant correlation between cities and left-wing vote and urban areas and PRR vote is found in 8, with only Spain and the Netherlands showing non-significant relationships. In contrast, in 2019, (H2) is not supported for the PRR vote: only six countries (46% of registered voters) conform to the tested pattern for PRR. For the left-of-center vote, the pattern is found in only seven member states representing 52% of the sample’s registered voters, a thin majority by population only.
(H3). The EU-wide difference in left-of-center vote between cities and rural areas was 6.2 percentage points (pp) in 2019; it widened to 8 points in 2024 due to the larger losses of left-of-center parties in rural areas (−3.9 pp) than in cities (−1.1 pp). Over the same period, the urban-rural gap in PRR vote increased from 4.4 pp to 6.6 pp. This substantiates the descriptive hypothesis (H3) at EU level. The hypothesis is also supported at the state level: in 15 of 22 member states (90% of registered voters), cities have become relatively more favorable to left-wing parties than rural areas between 2019 and 2024, while rural areas have become comparatively more favorable to PRR parties in 17 of 22 member states (71% of registered voters).
(H4). The results of our EU-wide model with interaction variables
$year \times cities$
and
$year \times rural$
are shown in Table 6. For the PRR vote, the signs of the coefficients of the interaction variables support (H4). For the left-of-center vote, both signs are positive, with the coefficient of the interaction variable for cities (
$1.29$
) being much larger than the coefficient of the interaction variable for rural areas (
$0.20$
), also supporting (H4).
Weighted least squares regression: EU
Table 6. Long description
The table presents the results of an EU-wide model with interaction variables year times cities and year times rural. It includes five columns labeled Left, Center-right, Radical right, Other, and Non-vote. The rows are labeled Cities, Rural areas, year (2024), year times cities, year times rural, Controls, and R squared. Each cell contains numerical values representing coefficients or indicators. For example, the Left column shows values like 1.41 for Cities, -1.00 for Rural areas, -4.86 for year (2024), 1.29 for year times cities, and 0.20 for year times rural. The table also indicates the presence of controls and fixed effects for each column. Notable trends include positive coefficients for year times cities and year times rural in the Left column, with a significantly larger coefficient for cities compared to rural areas.
*
$p{\rm{\lt }}0.05$
**
$p{\rm{\lt }}0.01$
***
$p{\rm{\lt }}0.001$
.
Running 22 national models with interaction terms, we observe a consistent pattern in nine countries accounting for 51% of registered voters for the PRR vote – a thin majority. The pattern suggested by (H4) is found in only four countries (33% of registered voters) for left-of-center vote, leading us to reject the hypothesis in this case.
Summary. Combining BLUE_EP with GEOSTAT and other regional statistics, we have found support for (H1), (H2), and (H3), as well as (H4) for PRR vote only.
In summary, in both 2019 and 2024, urban areas have indeed shown higher support for left-of-center parties and lower support for PRR parties, while rural areas have displayed an opposite pattern. This trend is observed across member states. When controlling for confounding factors, we still find urban voters to be more likely to vote for left-of-center parties and rural voters to be more likely to vote for PRR parties, although this effect is observed consistently across countries only in 2024. The urban-rural gap increased between 2019 and 2024 in absolute terms across a majority of countries. This increase persists after controls, but is then less consistent across countries.
Conclusion
In this note, we have presented BLUE_EP, the first dataset compiling all municipality-level results of the 2019 and 2024 European Parliament elections. Our dataset provides official voting results for 90,000 spatial units and 500 parties for each of the two elections. In particular, it includes all publicly available results of OCTs of the EU and voters abroad. Through its use of the common European typology of LAUs as well as Party Facts identifiers, BLUE_EP is easily integratable with existing data sources.
To the best of our knowledge, BLUE_EP is one of the largest datasets covering individual electoral events ever published. By number of data points, it likely only comes second to datasets from India’s Lok Sabha elections (Jensenius, Chhibber, Alam et al. Reference Jensenius, Chhibber, Alam, Gupta and Somanathan2025).Footnote 4 BLUE_EP is comparable in size to the MIT Election Lab’s precinct-level datasets of US federal elections,Footnote 5 which feature more spatial units but fewer parties. BLUE_EP maps EU-wide electoral data at the local level in a way consistent with the EU’s own statistical typology, yielding fine-grained data comparable to its US and Indian counterparts. While we think that such data collection and harmonization should eventually be a task for the European Statistical System, our use of 27 separate extraction scripts was enough to overcome the lack of a harmonized framework. We are confident that the same approach can be extended to both other European elections and national elections to constitute broader transnational datasets.
To showcase the potential of BLUE_EP, we have presented a case study on the significance of the urban-rural cleavage in the 2019 and 2024 European elections, providing evidence for a widening urban-rural voting gap. We believe that our dataset can be useful to the political science community to answer many other relevant questions related to the spatial dimensions of electoral competition in the European Union. To cite only a few, studies investigating the spatial and economic patterns of right-wing populism (Schraff and Pontusson Reference Schraff and Pontusson2024), electoral behavior in border regions (Nasr and Rieger Reference Nasr and Rieger2024), or the effect on voting of external shocks such as the Covid-19 pandemic (Fernandez-Navia, Polo-Muro and Tercero-Lucas Reference Fernandez-Navia, Polo-Muro and Tercero-Lucas2021) or environmental disasters (Jusko and Spáč Reference Jusko and Spáč2024) could strongly benefit from BLUE_EP’s local granularity. The key challenge in each of these applications consists in identifying appropriate transnational data sources to combine with the vote counts, which we exemplified through the use of GEOSTAT. Additionally, BLUE_EP also provides a single source from which all national- and regional-level results can be seamlessly extracted, avoiding the manual processing of heterogeneous data formats.
Data availability statement
All of the BLUE_EP data is available online at: https://zenodo.org/records/14569325.
Acknowledgements
We thank Sofia Marini (Aarhus Universitet) for her feedback on a preliminary version of this note.
Funding statement
The author received no funding to conduct this work.
Competing interests
The author states that there is no conflict of interest.
Ethical standards
(if relevant) NA.
Use of AI
(if relevant) AI models (ChatGPT, Claude, Perplexity) were used for spelling and stylistic checks under the author’s supervision.
Author Biography
François Hublet is a graduate of École Polytechnique (Palaiseau) and ETH Zurich. Since 2021, he has been leading the Electoral Bulletins of the European Union (BLUE) within Groupe d’études géopolitiques (GEG). His research at GEG focuses on electoral dynamics, voting behavior, the European political system, and border region policies, with a focus on quantitative methods.
Recent publications include Where the Borders Lie: Mapping Cross-Border Communities in 10 Western European Countries (Transportation Research Record, 2024, co-authored with A. Sallard) and A Tale of Three Cleavages (Revue française de science politique, 2025, co-authored with M. Lanoë).
Open access
