A Dyadic Method to Investigate Voting Behavior in the Council of the European Union

Abstracts

Using a dyadic approach to explore voting behavior of European Union (EU) member states in the Council of the EU, we investigate the similarity in voting behavior of governments on three policy dimensions: left-right, authoritarian-libertarian, and pro-/anti EU. These policy dimensions are of interest also in other contexts, such as decision-making in international or regional organizations other than the EU. Our dependent variable is based on a new data collection covering the time 2010–2021. Our analysis, employing random effects binary and ordinal logistic regression applied to relations between EU states, confirms that larger distances of governments represented in the Council on these policy dimensions combined are related to larger distances of their vote choices. Larger distances affecting different vote choices within dyads are also confirmed for the pro-/anti EU policy dimension. Moreover, we find that closeness of member states in terms of their budget positions and of domestic public opinion on the EU are related to voting similarities. Member state differences in terms of voting power, by comparison, lead to less divergence of vote choice, for decisions based on the qualified majority voting rule. The length of EU membership, however, does not affect vote choice similarities.

En este artículo utilizamos un enfoque diádico que nos permite estudiar el comportamiento electoral de los Estados miembros de la Unión Europea (UE) en el Consejo de la UE, con el fin de investigar las similitudes entre el comportamiento electoral de los Gobiernos en tres dimensiones políticas: izquierda-derecha, autoritario-libertario y pro-/anti-UE. Estas dimensiones políticas también resultan de interés en otros contextos, como en la toma de decisiones en organizaciones internacionales o regionales distintas de la UE. Nuestra variable dependiente se basa en una nueva recogida de datos que abarca el período entre 2010 y 2021. Nuestro análisis utiliza efectos aleatorios binarios y regresión logística ordinal aplicada a las relaciones entre los Estados de la UE, lo que nos permite confirmar que las mayores distancias entre los Gobiernos representados en el Consejo con relación a estas dimensiones políticas combinadas están relacionadas con unas mayores distancias de sus opciones de voto. Estas mayores distancias, las cuales afectan a las diferentes opciones de voto dentro de las díadas, también se confirman para la dimensión política pro-/anti-UE. Además, concluimos que la cercanía entre los Estados miembros con respecto a sus posiciones presupuestarias y a la opinión pública interna sobre la UE está relacionada con las similitudes de voto. En comparación, las diferencias entre los Estados miembros en términos de poder de voto provocan una menor divergencia en la elección del voto, en el caso de las decisiones basadas en la regla de la mayoría cualificada. Sin embargo, la duración de la pertenencia a la UE no tiene ningún efecto sobre las similitudes en las opciones de voto.

À l'aide d'une approche dyadique pour l'analyse du comportement électoral des États membres de l'Union européenne (UE) au sein du Conseil de l'Europe, nous nous intéressons aux similarités de comportement électoral des gouvernements s'agissant de trois dimensions politiques : gauche/droite, autoritaire/libertaire et pro-/anti-UE. Ces dimensions politiques nous intéressent aussi dans d'autres contextes, notamment dans la prise de décisions d'autres organisations internationales et régionales que l'UE. Notre variable dépendante se fonde sur un nouveau recueil de données couvrant la période de 2010 à 2021. En employant une régression logistique binaire et ordinale à effets aléatoires appliquée aux relations entre les États membres de l'UE, notre analyse confirme que les distances plus importantes entre les gouvernements représentés dans le Conseil sur ces trois dimensions politiques combinées sont associées à des distances plus importantes dans leurs choix de vote. L'effet de ces dernières distances au sein des dyades se voit également confirmé pour la dimension de politique pro-/anti-UE. De plus, nous observons des similarités de choix de vote lorsque les États se rapprochent concernant leur position budgétaire et leur opinion publique nationale vis-à-vis de l'UE. Par comparaison, les différences de droit de vote des États membres conduisent à moins de divergences par rapport aux choix de vote, quand il s'agit de décisions fondées sur la règle de vote à la majorité qualifiée (VMQ). Néanmoins, la durée de l'appartenance à l'UE n'a pas d'effet sur les similarités de choix de vote.

Introduction

Politicization in the European Union (EU) observed since the 1980s (Schmitter 2009) has influenced both intergovernmental EU institutions such as the Council of the EU (hereafter the Council) and supranational, more technocratic ones such as the European Central Bank (Schmidt 2019). This politicization has arguably transformed the EU from an institution of policy without politics to an institution of policy with politics (Schmidt 2019). Bottom-up politicization in the EU (Bressanelli, Koop, and Reh 2020) has been defined as “controversies driven by public debate, political parties and elections” (Kauppi and Wiesner 2018, 227) in which the key aspect is “the polarization of conflict among political actors” (Grande and Hutter 2016, 9). Politicization in the EU, shown by empirical findings (Grande and Hutter 2016), leads us to assessments of European integration in which political party preferences and partisanship affect member states’ behavior in European institutions (Hooghe and Marks 2009). This partisanship can be based on ideological orientations (Grande and Hutter 2016; Hooghe and Marks 2009; Kauppi and Wiesner 2018; Naurin 2018), economic positions (Hooghe and Marks 2009), and Euroscepticism and identity (Grande and Hutter 2016; Hooghe and Marks 2009; Naurin 2018). Politicization in the EU is reflected in decision-making within European institutions, including the Council (Grande and Hutter 2016) and the European Parliament (Guinaudeau and Costa 2022).

Existing studies on voting behavior in the EU Council primarily focus on oppositional voting, coalition formation, and comparing member states' preferences with their voting behavior. However, they overlook the dyadic approach, leaving a gap in understanding council dynamics. To address this, we employ a dyadic approach using council roll call data and official statements. We explore how the distance between two member states' policy positions within each dyad correlates with their voting behavior similarity. This approach offers insight into voting alignment and potential sources of conflict or cooperation. While our results are relevant to the EU, the insights provided may be of interest more generally to studies assessing regional integration schemes. Unlike existing studies focused on the EU, which analyze behavior relative to the winning coalition, our dyadic approach directly examines interactions between pairs of member states, revealing nuanced factors that are associated with voting behavior and enhancing understanding of multilateral decision-making. We measure voting alignment within each dyad using three dependent variables for each legislative act: binary variables indicating if both member states support or oppose the act, if their voting decisions match, and the distance between their decisions. Our dataset spans 1,154 legislative acts in the EU Council from 2010 to 2021, encompassing 423,313 paired decisions. We removed the non-participating voting decisions, resulting in 415,977 paired decisions, forming panel data with dyads as units and legislative acts as observations over time.

We examine how the difference in policy positions between two member states within each dyad relates to their voting behavior alignment. While our analysis pertains to the EU, the approach can also be applicable to assess how differences in policy positions within dyads influence voting outcomes within other regional or international organization, including the United Nations General Assembly (UNGA). Our analysis covers three policy dimensions: ideology, political culture, and EU integration. We calculate government policy positions as weighted averages of their parties' positions based on the ParlGov dataset. Using Euclidean distance, we measure the overall policy position difference between governments in each dyad, along with differences on each dimension separately. Employing random effects logistic models, we find that greater policy position differences correspond to lower probabilities of voting alignment and greater distances between voting decisions. We control for differences in domestic public opinion on the EU, a priori voting power, EU budgetary status, and EU accession timing. Results show that larger differences in EU integration positions decrease the likelihood of voting alignment.

The rest of the article is organized as follows. Section 2 discusses the debate in the literature on voting behavior and policy positions in the Council. We introduce our dyadic dataset in Section 3. Sections 4 and 5 discuss the operationalization of our variables of interest and controls. Section 6 provides the analysis and results. Concluding remarks are in section 7.

Policy Positions and Voting Behavior in the Council

The debate on the relationship between member states’ voting behavior in the Council of the EU and their policy positions has not been settled. Earlier studies that use roll call data show left-wing governments are less likely to contest decisions in the Council than right-wing governments (Hosli, Mattila, and Uriot 2011; Mattila 2004). The left-right division in the Council, however, is more prominent before the 2004 enlargement (Hagemann 2008) as the distance of government policy positions from the Council average is correlated with voting behavior in the Council only among pre-enlargement EU member states (Hosli, Mattila, and Uriot 2011). Similarly, on the EU integration policy dimension, while pro-integration governments are less likely to contest decisions, when all member states after the enlargement are considered, there is no correlation between the distance from the Council's average position and the decision to contest (Hosli, Mattila, and Uriot 2011). More recent studies, however, show a clearer association between policy positions and voting behavior in the Council: more extreme governments, both on the right and left sides of the ideological spectrum, and governments further away from the winning coalition's policy position, on both the left-right and EU-integration policy positions, are more likely to contest decisions ( Giuliani 2023;Pircher and Farjam 2021; Pourebrahimi, Hosli, and Roozendaal 2023). Similar studies on the European Parliament show that ideological positions of Members of the European Parliament (MEPs) are correlated with their voting behavior (Hix, Noury, and Roland 2009) and speeches on topics such as monetary policy (Fraccaroli et al. 2022).

We use a new approach to contribute to this discussion. While most studies on voting behavior in the Council focus on oppositional voting and the distance of a member state from the average or the winning coalition on a policy position, in this article, we investigate member states’ voting behavior in relation to each other. Accordingly, we apply a dyadic method to analyze voting behavior of EU states in the Council. The dyadic method allows us to examine the relationship between member states and see the extent to which differences between policy positions of two member states are related to differences in their voting behavior. Though the dyadic method has been used extensively in international organization studies,¹ it has been left out so far from studies examining the relationship between member states’ policy positions and their Council voting behavior with a few exceptions such as Thomson (2011), which applied a dyadic approach to investigate the ideological differences between pairs of member states and their distance in terms of policy positions. Other examples of dyadic methods in EU studies include the voting behavior of MEPs (Proksch and Slapin 2011; Van der Veer and Otjes 2021), or an exploration of trust between EU nationalities using Eurobarometer survey data (Delhey 2007).

Examining member states’ voting behavior in the Council in relation to each other puts our article closer to studies on coalition formation in the Council. In these studies, there is no consensus on the alignment of preferences and policy positions. Some studies show the existence of an ideological coalition formation process in the Council, with a notable emphasis on center-left governments (Hagemann and Hoyland 2008) and the alignment of policy positions, preferences, and voting behavior ( Aspinwall 2002; Hellström 2008; Høyland and Hansen 2014). Other studies, however, do not find evidence for the alignment of policy positions and preferences in the Council (Thomson 2011, 152; Zimmer, Schneider, and Dobbins 2005). This policy drift—the difference between policy positions and predicted preferences of the cabinet—can be related to the structure of a coalition government (Kostadinova and Kreppel 2022). Building on such earlier analyses but providing new insights, in this article, we investigate the alignment between policy positions and revealed preferences through voting and statements, as we examine the association between policy position and voting decision similarities between each pair of EU member state governments.

A Dyadic Dataset on Council Decisions

We use council roll call data to build our dataset.² The Council of the EU publishes voting results on the Consilium website.³ Instead of downloading the data directly from this website, we used the EURLEX R package (Ovádek 2021) to obtain a website query. We removed repeated observations from the dataset as retrieved with the EURLEX R package, and complemented it with other information about decisions on legislative acts, such as from other primary sources, including the EUR-Lex website⁴ or the EU Legislative Observatory (OEIL)⁵ as well as monthly summaries of Council acts published by the Council itself.

Using this method of data collection, the primary dataset used in this article, based on Pourebrahimi (2025), encompasses 1,154 legislative acts approved by the council between 2010 and 2021. These acts include a variety of act types and policy areas, and different legislative procedures and voting rules applied. For each legislative act, the voting decision of each member state was recorded. In essence, member states can vote in favor, abstain, or vote against a proposal. They can also issue an official statement along with their vote. In some cases, a member state does not participate in the voting process; these cases are recorded as “not participating” (notably in cases of decisions concerning Eurozone members only).

For our analysis, each dyad includes two member states’ decisions on a legislative act. As there are 28 member states at maximum, each legislative act can lead to 378 different dyads. Since our dataset includes legislative decisions made by the Council between 2010 and 2021, it covers acts before Croatia joined the EU in July 2013 and after Brexit in February 2020. Accordingly, for some legislative acts, there are only 27 member states in our dataset. In total, our 1,154 legislative acts result in 423,313 dyads. We exclude the “not participating” decisions as they do not necessarily reflect contestation or discontent. Thus, all dyads in which at least one member state did not participate in the voting process (1.7 percent of all cases) were removed. Based on this procedure, the total number of observations is 415,977.⁶

We examine the alignment of voting decisions among EU member states in the Council to understand their voting patterns (“yes,” abstain, or “no”). Table 1 summarizes the voting decisions for each pair of member states across legislative acts from 2010 to 2021. Similarity rates between states are calculated based on equal voting decisions within each dyad. For example, Germany and the United Kingdom (UK) show the lowest similarity at 82 percent. Conversely, some member state pairs, such as Finland and France, have a similarity rate of 99 percent.⁷  Table 1 reveals that in more than 95 percent of the cases analyzed, both member states in a dyad vote in favor of a legislative act. Tables 2 and 3 present the distribution of decision pairs in the Council of the EU between 2010 and 2021. Table 2 reports whether member states adopted the same position (support or opposition), whereas table 3 accounts for the full range of voting options (in favor, abstention, or against), showing that over 95% of decision pairs involved identical voting behavior.

To analyze member states' decisions on legislative acts in each dyad, we employ three dependent variables (two binary and one measuring ordinal distance). The first binary variable indicates whether both countries in a dyad make the same decision regarding a legislative act, either supporting (“in favour” votes) or opposing (“no” votes and abstentions). It takes the value of one if both member states decide the same way and zero otherwise.

The second binary variable compares the voting decisions of each pair of member states within each dyad. It takes the value of one if the decisions match and zero otherwise. Even if both countries do not support a legislative act (e.g., if one votes against and the other abstains), their decisions are recorded as being different.

Finally, we use a dependent variable reflecting the (ordinal) distance between EU member states’ voting decisions. Accordingly, we first sort the decisions based on levels of support, from most to least. At this stage, we also consider the statements member states may have published alongside casting their votes. Formal negative statements published along with votes can be considered to be expressions of discontent (Hagemann, Hobolt, and Wratil 2017; Van Gruisen and Crombez 2019). Therefore, we distinguish between votes in favor without negative statements and votes in favor accompanied by negative statements, as those with negative statements likely reflect less support. To find negative statements, all statements published by member states along with their votes were examined.⁸ A total of 30 statements (five percent of the total) was then selected to build a dictionary of negative keywords and terms.⁹ Given the unique structure of formal statements and the diplomatic language used, creating a new dictionary was more applicable than using already available sentiment analysis dictionaries. The dictionary created in this way was then validated¹⁰ and used to analyze the “sentiment” in all statements. In practice, any statement that included at least one of the keywords or terms in the dictionary was categorized as being a negative statement. In 18,111 out of the total of 415,977 observations, at least one member state issued a formal negative statement.

The distance between the decisions of two member states in each dyad varies from zero to three, based on the rank-ordering of decisions: (1) “yes (in favour) without a negative statement”, (2) “yes (in favour) with a negative statement”, (3) abstention, and (4) “no.” The frequency of each distance is given in table 4, with over 90 percent of all cases showing zero distance (i.e., indicating identical voting decisions).

Differences on Policy Dimensions and Member State Voting Behavior

We now explore correlations between differences in member state government compositions across various policy dimensions and their alignment in terms of vote choices within the Council. We consider three policy dimensions: ideology, political culture (reflecting a libertarian-authoritarian scale), and pro-/anti EU preferences.

To assess government policy positions, we utilize the Parliament and Government Composition (ParlGov) database (Döring, Huber, and Manow 2022), which presents the policy position of each political party in each EU member state based on expert surveys.¹¹ In the ParlGov database, time-invariant party positions are assessed on a scale ranging from zero to ten, with five representing the midpoint. For example, on the ideology dimension, zero and ten represent extreme left and extreme right policy positions, respectively. Lower numbers on the political culture dimension, by comparison, indicate more libertarian attitudes, while higher numbers capture more authoritarian positions. Regarding the pro-/anti EU dimension, zero reflects highly Eurosceptic parties, while ten represents those highly supportive of the EU. Table 5 provides an overview of the policy dimensions based on this categorization.

We now calculate a government policy position as the weighted average of the parties in government, with the weights being determined based on the number of members each party has in the national parliament. To assess the location of each member state government for each policy dimension included in our analysis (for the time of the decision on a legislative act), we base calculations on equation one:

Where

• dimension_ij is member state i’s government policy position at the time of a decision on legislative act j.

• w_kij is the number of MPs affiliated with party k in member state i at the time of the decision on legislative act j.

• dimension_k is the policy position of party k.

• -

  $${{d}_{kij}} = \left\{ {\begin{array}{@{}*{1}{c}@{}} {1,\,{\rm if\,party}\,k\,{\rm is\,in\,the\, government\,in\,member}}\,\\{{\rm state}\,i\,{\rm at\,the\,time\,of\,decision\,on\,legislative\,act}\,j}\\{0,\,{\rm if\,party}\,k\,{\rm is\, in\, not\, the\, government\, in\, member}}\\{{\rm ,state}\,i\,{\rm at\, the\, time\,of\,decision\,on\,legislative}\ act\ j} \end{array}}\right.$$

• n is the number of political parties.

• i = 1, … , 28

• j = 1, … , 1154

Accordingly, the policy position distance between the two member states in each dyad at the time of decision on each legislative act is assessed as the absolute difference between the policy positions of those two member states.

Theoretically, the distance between policy positions can vary between zero and 10. In practice, however, the policy positions of member states are much closer to each other. Table 6 shows distances on each policy dimension according to this assessment.

We now proceed to compare the distribution of distances across different policy dimensions (figure 1). Notably, the integration dimension exhibits the lowest interquartile range, partially due to missing observations, whereas the left-right positioning (ideology) displays the highest interquartile range. While table 6 suggests that the means of distances for two policy dimensions (ideology and political culture) are relatively similar, two-sample t-tests reveal that we can reject the hypothesis of equal means across all dimensions (at a 1 percent level of statistical significance). We conduct two-sample t-tests for all three possible pairs derived from the three dimensions.

Multicollinearity¹² is now essential to check distances on each dimension. The highest correlation (r = 0.35) exists between distances on the ideology (left-right) and political culture (libertarian-authoritarian) dimensions. All correlations are positive, indicating that higher distance on each dimension positively correlates with higher distances on other dimensions. Figure 2 provides a scatterplot for each pair of two policy dimensions.

In addition to the distance between the governments within each dyad on each policy dimension as included in our analysis, we now measure three-dimensional Euclidean distances between the governments for each pair of member states. The Euclidean distance is measured based on equation (2):

Where:

• |$Distanc{{e}_{ii^{\prime}j}}$| is the Euclidean distance between member state i and member state i’ at the time of voting on legislative act j.

Figure 3 shows the distribution of the Euclidean distances between EU governments assessed on a monthly basis. The lowest Euclidean distance (0.11) found is between Slovenia and Sweden in July 2015, while the highest (8.67) is between Italy and Spain in November 2011. Note that, theoretically, Euclidean distance can be up to 17.32. We will now discuss control variables integrated into our estimates.

Other Cleavages in the Council

In our analysis, we control for a range of variables relevant to potentially explaining the similarity in voting behavior of EU member states in the Council. The first control variable is domestic public opinion, which reflects public sentiment towards the EU. This variable is important, for example, in post-functionalist EU studies (Hooghe and Marks 2009). Domestic public opinion has been found to shape member states’ positions in the Council (Franchino, Kayser, and Wratil 2022) and to affect their voting decisions (Hosli, Mattila, and Uriot 2011): Member states use their positions and votes in the Council as a signal to their constituencies (Van Gruisen and Crombez 2019; Wratil 2018). Initially, we explore the role of domestic public opinion toward the EU for each member state across all years of analysis. Subsequently, we evaluate the disparity in public opinion on the EU between each pair of member states within each dyad, again for every legislative act between 2010 and 2021. Utilizing Eurobarometer survey data and the question: “In general, does the EU conjure up for you a very positive, fairly positive, neutral, fairly negative, or very negative image?” for each dyad, we calculate the difference between the percentages of respondents in the two countries with (very or fairly) positive attitudes toward the EU. Since government positions on EU integration may be influenced by domestic public opinion, we examine potential multicollinearity between the distance of public opinion and the distance on the EU integration positions. A Pearson correlation of 0.01 reveals, however, that including these variables in the same model will not induce multicollinearity.

Of interest can also be a North-South policy dimension, for which we resort to government budget positions. Accordingly, some member states are net beneficiaries of the EU budget, while others are net contributors. Member states’ EU budget positions, in fact, have been found to constitute one of the main sources of conflict within the Council (Zimmer, Schneider, and Dobbins 2005) and, accordingly, affect their voting behavior (Bailer, Mattila, and Schneider 2015). For each pair of EU member states and each legislative act, we check whether their status as a net contributor or a net beneficiary coincides. To categorize member states as net contributors or net beneficiaries, we use EU budget spending and revenue data, as published by the European Commission.¹³ We calculate the net contribution for each year as total national contributions¹⁴ minus EU budget expenditure¹⁵ disbursed to each member state. Accordingly, member states with positive net contributions are categorized as net contributors, whereas those with negative net contributions are classified as net beneficiaries.

Another control variable integrated into our analysis is the difference between the a priori voting power member states possess. As an exogenous source of influence, voting power can affect member states’ bargaining success (Bailer 2004) and their voting behavior (Hosli, Mattila, and Uriot 2011). Hence, we examine whether the similarity between the voting power within each dyad is correlated with similarity in the voting behavior within each dyad. Utilizing software package ipmmle (Leech 2012), we calculate the standardized Banzhaf power index as a reflection of the voting power of each EU member state for each month between 2010 and 2021. The program measures the standardized Banzhaf power index in the Council's double majority system, given member state population size¹⁶ and the applicable QMV (double majority) rule, where a proposal must be supported by 55 percent of member states representing at least 65 percent of the total EU population.¹⁷ We assume all member states have the same voting power for legislative acts for which the unanimity voting rule applies (as each can cast a veto). We calculate the difference in power within each dyad based on the member state's voting power at the time of voting on each legislative act.

Voting behavior of older and newer EU member states can differ as well (Hosli, Mattila, and Uriot 2011). In the dataset used for this analysis, we categorize old member states as those that joined the EU before 2004 and those that joined in or after 2004 (despite their relatively long time of membership), as new EU members. In our model, we control for whether both EU member states in a dyad are from the same membership category (i.e., both old or both new) or whether they differ in this category (i.e., one old and one new member state). These control variables help ensure that our outcomes are not affected by factors not accounted for in our main model specifications.

Model Estimates: Analysis and Results

In our analysis, we first examine a binary dependent variable that indicates whether both member states in a dyad share the same voting decision (either both support or both do not support a legislative act). As independent variables, we first employ the Euclidean distance between member states within each dyad and, in a subsequent model, utilize distances on the ideology (left-right), political culture (libertarian-authoritarian), and EU-integration (anti-pro EU) policy dimensions. For each dyad, we control for differences in (positive) public opinion towards the EU, voting power, EU budgetary status, and old versus new EU member state status. Given the binary nature of our dependent variable, we utilize a logistic random effects model. Table 7 presents the odds ratios of our first two models.

Model 1 reveals a negative correlation between the Euclidean distance, implying that one additional unit of Euclidean distance between two EU member states in a dyad is associated with 4.3 percent lower odds of observing the same voting decision within that dyad. Similarly, Model 2 demonstrates that greater disparities in ideological and EU integration positions correspond with reduced probabilities of concordant decisions within dyads. However, the first two models do not show any correlation between voting behavior and distance on the political culture dimension. Regarding control variables, we find a negative correlation between differences in voting power within dyads and the probability of shared decisions. Similarly, having the same budgetary position positively correlates with the likelihood of making the same voting decision. Hence, member states with similar policy positions, voting power, and budgetary stances tend to vote in similar ways on EU legislative acts. Our analysis, however, does not reveal any evidence linking similarities in voting behavior to whether a member state has been in the EU before or only since the 2004 enlargement (although this may be different for the time immediately after the 2004 enlargement, which our dataset does not cover).

For our first dependent variable, we classify voting decisions into “supporting” and “not supporting” a legislative act, where both negative votes and abstentions fall into the latter category. Recognizing that opposing votes and abstentions may carry different implications depending on the applicable voting rule, we differentiate between them in the framework of our second dependent variable. Hence, we consider a binary outcome, where a value of one indicates matching voting decisions within dyads (i.e., both are in favor, both against, or both abstain), and zero otherwise. We employ the overall Euclidean distance for one model and ideological, political culture, and EU integration policy positions for the other, including the same control variables as above. We utilize a random effects logistic model to explore outcomes for the second dependent variable. Accordingly, table 8 presents the results of models 3 and 4.

The results displayed in table 8 using the second dependent variable mirror those of the models based on the first dependent variable: An additional unit in policy distance in a three-dimensional space between member states within each dyad corresponds to 4.4 percent lower odds of them aligning their vote choices. Similarly, when considering the policy distances separately, distances on ideological and on EU integration positions are inversely correlated with the likelihood of casting the same vote in the Council of the EU. Therefore, greater policy distances on these dimensions between two member states correspond to lower odds (4.4 percent) of them aligning their votes.

Similarly, the findings regarding the control variables in models 3 and 4 closely resemble those based on models one and two: Larger disparities in voting power and in public sentiments towards the EU between two member states decrease the likelihood of them making the same voting decision in the Council. Additionally, when both member states in a dyad are either net beneficiaries or net payers regarding the EU budget, they are more inclined to exhibit similar voting behavior compared to dyads with differing budgetary positions. In contrast, membership status—whether both member states are old EU members, both are new, or one is old and the other new—does not significantly impact the likelihood of casting the same vote.

The first two dependent variables in our analysis are binary and indicate whether member states in a dyad make the same voting decision. In contrast, our third dependent variable measures the distance between the voting decisions within each dyad. This variable ranks voting decisions ordinally from most to least supportive: “yes,” “yes with a negative statement,” abstention, and “no.” Therefore, a higher value for this dependent variable indicates a greater difference in the ranks of voting decisions between two member states within a dyad. Models five and six explore the relationship between this ordinal dependent variable and the same independent variables and controls examined in our previous estimations. For models five and six, we employ ordinal logistic regression analysis (see table 9).

Although the coefficients in models 5 and 6 show opposite directions compared to models 1–4, they still demonstrate a consistent trend: The positive coefficient for the Euclidean distance suggests that a greater distance between two member states in a dyad leads to a larger difference in their voting decisions. This also applies to the distance between two member states on the EU integration dimension. In model 6, however, the coefficient for the ideological position is not statistically significant.

We also conduct robust estimations for our six models:¹⁸ The results reveal that the Euclidean distance between member states’ three-dimensional policy positions is negatively correlated with the likelihood of them taking the same voting decision and positively correlated with a larger distance between their voting decisions. Among the policy dimensions explored, however, distance of policy positions on EU integration is the only one for which the negative correlation with voting alignments is statistically significant (at a 1 percent level of significance). Among the control variables, the same budgetary position within a dyad is positively correlated (at the 1 percent level of significance) with voting alignment. For domestic public opinion on the EU, we observe a statistically significant negative correlation (at the 1 percent level of significance) with voting alignment only when the dependent variable measures the distance between the voting decisions.

While models 1–6 reveal a negative correlation between voting power distance and voting alignment, one should be careful in terms of reaching conclusions on correlations between power distances and vote decisions in the Council. After all, for legislative decisions in which the unanimity rule applies, all EU member states hold the same power (all have a veto). Thus, power distances within each dyad for legislative acts requiring unanimity are simply zero. Moreover, opposing votes are less frequent in cases requiring unanimity compared to those decided based on the QMV (double-majority) clause. Accordingly, it can be expected that the coefficient for the control variable “voting power” is affected by the fact that in cases requiring unanimity, power distances between member states are zero; accordingly, taking the same voting decision is more frequent. To further explore this aspect, we also run the models for cases only in which the QMV rule applies.¹⁹ Indeed, based on this alternative estimation, the direction for the power distance coefficient is different from what is observed for models one to six: Power distance, in cases based on the double-majority rule exclusively, is positively related to voting alignments.

To measure the size of the correlations, we also run the standardized versions of models 1–6.²⁰ Among the independent variables of interest, both the Euclidean distance and the distance on the EU integration dimension are correlated with voting alignment, but the size of the correlation is small. This is not surprising, however, as the variation in the dependent variables we include in our models is relatively small.

Conclusions

Based on a novel dataset on EU Council decisions for the 2010–2021 period (Pourebrahimi 2025), we provide a dyadic approach to member state voting decisions. Independent variables capturing locations of EU governments on three policy dimensions—left-right, authoritarian-libertarian and pro-/anti EU—are largely based on the ParlGov dataset.

Using estimates based on random effects binomial and ordinal logistic regression analyses, we demonstrate that the dyadic approach confirms some results presented in earlier studies but also provides new insights. We find that Euclidean distance on the policy dimensions assessed is positively related to differences in vote choices of governments in the Council of the EU.

We cluster voting decisions in two ways—either binary as opposition or support for a legislative act or in an ordinal setting, ranking EU member state vote choice in the Council from high to low support (“yes,” “yes with a negative statement,” abstention, and “no”). Both approaches confirm that higher Euclidean distance on the three policy dimensions combined within a member dyad increases the likelihood of different vote choices. Taking the left-right, authoritarian-libertarian and the pro-/anti-EU dimension as predictors, we again find that a higher distance on the EU integration dimension within a dyad is correlated with higher differences in member state vote choices.

Our control variables include aspects such as member state voting power (a reflection of “size”), EU budget positions and whether a member state joined the EU before or after the 2004 enlargement. Closeness in a dyad on aspects such as a member state's EU budget position—as net beneficiary or net payer—will increase the likelihood that a similar vote is cast, as our estimations demonstrate. This also applies for voting power, but if the analysis focuses on QMV (double-majority) decisions exclusively, member states in a dyad will vote more differently the closer they are in terms of size (voting power). Our estimates show that the length of EU membership, contrary to findings in some earlier studies, does not matter regarding the similarity of vote choice within a dyad.

With an approach focused on member state dyads, we aim to add to existing literature on voting in the Council of the EU and demonstrate the usefulness of this approach to assess voting behavior in regional or international organizations more generally. While our analysis does not capture all potential variables that might matter for member state vote choice (including linguistic closeness, shared cultures or geographical locations), we believe to have included some of the most important potential factors as control variables into our estimates. Future research could base assessments of government left-right locations, political culture or other variables we use to reflect member state locations on given policy dimensions on either different scales or different datasets estimating government locations on such dimensions. Nonetheless, we expect our dyadic approach might enhance insights into voting behavior in the Council of the EU, based on a dataset capturing votes between 2010 and 2021. Moreover, we hope our analysis demonstrates more generally how the policy dimensions resorted to and the tools applied to explore similarities between member state voting behavior might inspire other research focused on voting patterns in regional or international organizations.

Data availability

Data will be shared upon request.

Author Biography

Arash Pourebrahimi is a Lecturer at the Institute of Security and Global Affairs, Leiden University, The Hague, The Netherlands.

Madeleine O. Hosli is a Full Professor of Political Science and a Jean Monnet Chair Ad Personam at the Institute of Security and Global Affairs, Leiden University, The Hague, The Netherlands.

Jaroslaw Kantorowicz is an Associate Professor at the Institute of Security and Global Affairs and Research Associate at the Department of Economics at Leiden University.

Notes

The data underlying this article are available on the ISQ Dataverse, at https://dataverse.harvard.edu/dataverse/isq.

Footnotes

References