Now You See Me, Now You Don't: Anticipatory Coalitional Strategies in European Representative Democracies

Abstract

This article approaches the puzzle of how parties can strategically anticipate coalition formation and make themselves more attractive coalition partners in their electoral strategies. It argues that parties can strategically adjust the salience of issues that are secondary to them in pursuit of increased ‘coalitionability’. It tests the argument through analysis of the salience of secondary policy dimensions of up to 232 European parties between 1970 and 2019, finding evidence that parties adjust the levels of salience of their secondary dimension in response to the probability of their being included in a coalition government and their distance from the coalition in which they are most likely to be included.

Do political parties adjust the salience of their secondary issues to improve their likelihood of being included in a government coalition? And if so, under which conditions? A burgeoning literature on party strategies in (and between) elections has drawn attention to differences between parties' positioning and their emphasis on issues, mainly accounting for differences within and between parties. In this article, I include the prize of government as an additional point of information to consider when approaching the topic of party positioning. Drawing on the literature on preference tangentiality (Ecker and Meyer 2017; Falcó-Gimeno 2014; Luebbert 1986), I argue that parties are able to adjust their issue focus, or salience, ahead of elections in order to increase their likelihood of joining a government coalition, while simultaneously minimizing the risk that this will result in vote losses in the current election as well as the next one. With increased party-system fragmentation, often in conjunction with waning electoral fortunes for (previously) large mainstream parties, and the growing importance of multidimensional party competition, coalition governments are a common electoral outcome in most European party systems. In extreme cases, such developments may render the party system unstable, at least temporarily, resulting in more complex government formation (Pellikaan et al. 2018). In particular, scholars have noted the increased salience among parties and voters of non-economic, sociocultural issues, which are often considered as a second dimension of conflict in addition to the left–right economic dimension (e.g. Franklin et al. 1992; Hooghe and Marks 2018; Inglehart 2008; Kriesi et al. 2006, 2008).¹ A robust and growing literature on the behaviour of political parties explains well how parties act strategically to compensate for these developments in voter dispersion and which issues are electorally and politically salient (e.g. Abou-Chadi and Stoetzer 2020; Ezrow et al. 2011; Fagerholm 2016; Meyer and Wagner 2019). Nevertheless, the focus has primarily been on the relationship between parties and voters. Yet parties must also relate to other parties in the party system, as forming governing or legislative coalitions is key to attaining office and implementing policy. In contrast to parties' electoral behaviour, within the coalition theory literature an observed empirical regularity is that parties prefer to form coalitions with parties closer to them in the policy space than with some more distant or ideologically dissimilar alternative (de Swaan 1973; Leiserson 1966; Martin and Stevenson 2001). Hence, we see a shift from a primarily instrumental vote-seeking approach during the electoral phase towards a policy-seeking approach during the government formation phase when the likely outcome is a coalition government. Given that optimal vote-seeking strategies are not necessarily congruent with optimal policy-seeking strategies, the question is how and if parties anticipate coalition bargaining when formulating their strategies for winning votes.

Primarily, adjusting salience applies to parties in a position of intermediate bargaining strength – here operationalized as their coalition inclusion probability (Kayser et al. 2022) – that is, parties that are likely to be a member of some government coalition, but not overwhelmingly so. To test if parties adjust the salience of issues that are less important to them, I estimate a dynamic model (system generalized method of moment, GMM) on a panel of parties from post-electoral formation opportunities in up to 27 European parliamentary and semi-presidential democracies between 1970 and 2019.² In the empirical analysis, I use aggregated issue dimensions rather than individual issues due to greater data availability and potential problems arising from policy spaces with an increasingly large number of dimensions. The results indicate that parties with an intermediate likelihood of joining any coalition adjust the salience of a less important dimension downwards compared to parties that are (very) likely or unlikely to join any coalition, but that this is conditional on their distance from their likeliest coalition.

The rest of the article proceeds as follows. First, a section on the theoretical framework used for the article, including a brief review of previous research on party behaviour and coalition politics, is presented. Second, measurements of coalitional prospects and the empirical strategy are discussed. Third, the data and methods used in the analysis are introduced, followed by a section on the empirical analysis and a presentation of the results. Finally, the article ends with a discussion that covers possible future research avenues considering the preceding results.

Theoretical framework

The literature on parties' long- and short-term goals focuses on three such goals: winning votes, gaining office and implementing policy (Müller and Strøm 1999), put into practice as vote-seeking, office-seeking and policy-seeking behaviours.³ Navigating between these sometimes opposing and sometimes mutually reinforcing goals presents a parsimonious framework through which the strategic action of parties can be fruitfully analysed. Even if vote-seeking behaviour is credibly assumed to be instrumental in character by default, it is nevertheless of crucial importance to all parliamentary parties. Winning votes is a prerequisite for winning seats, which in turn is (often) a prerequisite for gaining office, and – depending on the institutional setting – implementing policy. A key component in trying to attract votes is to present a policy profile that (partly) serves this purpose, and consists of the sum of a party's policy positions.

Research on the behaviour of political parties as strategic actors regarding their policy positioning has predominantly been concerned with parties' reactions to environmental developments, such as changes in public opinion and vote losses. A key question is how parties respond in their positioning to shifts in public opinion on the one hand and realized vote losses on the other under conditions of incomplete information. In a recent review article, Andreas Fagerholm (2016) notes that the research on party policy change provides ample, if varied, evidence that parties respond to changes in their surrounding political environment, including the positions of other parties, by adjusting their policy positions. Notably absent, however, is how parties relate to coalitions of other parties.

The literature on coalition politics stresses that more ideologically cohesive coalition governments are likelier to form than those that are less ideologically cohesive (e.g. Bäck and Dumont 2008; Martin and Stevenson 2001, 2010), and that ideological proximity is also a factor in bargaining duration (Martin and Vanberg 2003) and government survival (Warwick 1994). Tendencies towards greater party-system fragmentation and increasingly salient second (and possibly third) dimensions complicate matters for government formation, as they potentially open up the party system for new parties that exploit previously neglected issues to win votes (Bakker et al. 2020b). The presence of more parties in parliament typically results in a more fragmented and complex bargaining environment, with consequences such as increased bargaining duration (De Winter and Dumont 2008; Diermeier and Van Roozendaal 1998; Golder 2006, 2010; Martin and Vanberg 2003). Introducing additional dimensions of conflict might also introduce conflict among previous coalition partners, as there is no guarantee that they are ex ante aligned on all now-salient dimensions. Given that parties appear to prefer ideologically more compact and connected coalitions than the alternatives, when possible (see e.g. Pellikaan et al. 2018 for an example where this preference breaks down), it follows that adjusting policy positions should affect parties' ‘coalitionability’. Such adjustment can primarily occur through adopting specific policy stances, or through focusing on specific policies while avoiding others, or a combination of both approaches.

Spatial theory (e.g. Downs 1957, as an early and particularly influential account) provides an approach focusing on parties' strategic positioning in policy space; salience theory (Budge and Farlie 1983; Robertson 1976) instead focuses on how parties emphasize issues in a strategic manner. A more recent development is to consider the positional and salience approaches as complementary rather than competitive (Elias et al. 2015). Parties do not necessarily value all dimensions and their constituent issues as equally important, and it is therefore likely that they will approach different dimensions based on party-specific factors, and use different strategies as a consequence. These strategies are, though, primarily concerned with parties' relation to voters. From the perspective of coalition bargaining, their implications are different. For example, a so-called blurring strategy is deployed to capture a broader spectrum of voters on a dimension than if the party's position was clear (Rovny 2013). However, this may also have adverse consequences on coalition formation, as it would also be unclear to potential coalition partners what policies the party pursued.

A blurring strategy can be augmented by simultaneously decreasing the focus on a ‘blurred’ dimension, thereby signalling that it is of lesser importance to the party. This can be considered a secondary, coalition-facing strategy. This kind of strategy can be pursued through parties either adjusting their position on specific issues, or adjusting the salience accorded to different issues – what Thomas Meyer and Markus Wagner (2019) have referred to as opinion-based policy change and emphasis-based policy change, respectively. However, these two options are not qualitatively equivalent. Staking out positions means that these positions can at any subsequent point be compared to the policy record of any government the party is or has been a part of. A party that adjusts its positions in anticipation of joining a government coalition could thus try to mitigate this potential threat of drifting away from their policy brand – its unique combination of policy proposals – as perceived by the electorate, essentially anticipating areas of compromise in advance of joining the coalition. The alternative is for parties to focus selectively on issues and issue dimensions in a way that facilitates greater coalition compatibility, either by focusing on the same issues as those favoured by the coalition or by having issue foci that are tangential (Ecker and Meyer 2017; Falcó-Gimeno 2014; Luebbert 1986), while avoiding issues that are likely to lead to friction (and compromise) in a coalition. This is simpler than adjusting positions, and there is already evidence that parties avoid issues that are contentious within the party in election campaigns (see e.g. Steiner and Mader 2019), and between coalition partners outside of the pre-electoral period (Sagarzazu and Klüver 2017; see also Martin and Vanberg 2008).

Against the argument that parties take account of the coalitional implications of their issue focus is the fact that the parliamentary situation facing parties ahead of coalition negotiations is ultimately decided by their ability to attract voters. Moreover, when parties attempt to realize their policy goals through a coalition government, compromise between the constituent parties becomes necessary, which can prove risky for the party's individual policy brand come the next election (Fortunato 2019, 2021; Martin and Vanberg 2011; Plescia et al. 2022). To a certain extent, parties are required to consider voters ahead of coalitions. Failing to win enough votes precludes parliamentary membership altogether, and having too few seats means a weaker bargaining position during coalition negotiations. Both should make joining a government coalition less likely, all else being equal.

I argue that the prevailing logic guiding coalition-anticipating strategies is one of mitigating the potential costs of joining a coalition government while simultaneously improving coalitionability at the margins. The central problem facing parties entering coalition governments is how to maximize policy influence in the coalition while maintaining their policy brand. Individual parties can mitigate deviations from their policy brand and streamline compromise processes through strategically focusing on issues that are simultaneously central to their policy brand and compatible with potential coalitions. By narrowly focusing on such a set of issues, parties can avoid failing to implement (central) policy pledges and moreover are more likely to implement what policy pledges they have made. This, however, obviously entails that the set of issues that is downplayed is both sufficiently important to other parties in the party system, while also not being of crucial importance to the party and its voters in the first place. The expectation is therefore that parties will adjust the salience of issues belonging to their secondary dimension, while maintaining a similar focus on their primary dimension, all else being equal. While this is most likely an ongoing process during the entire electoral cycle, it should be most pronounced in what can roughly be considered the election campaign period, especially so for government incumbents (Sagarzazu and Klüver 2017), finalized in the publication of an election manifesto or its equivalent.

The proposed strategy has two desirable properties for a would-be coalition member: it makes it more likely that the party will be included in a coalition due to perceived compatibility, and it also decreases the number of issues that require compromise on publicly stated positions. What ultimately determines whether a party adopts a salience adjustment strategy or not is based on the potential benefits such a strategy can yield in terms of office and policy gains while keeping vote losses at a minimum. While office gains are inextricably linked to joining a government coalition, policy gains can be achieved outside the government coalition, depending on institutional arrangements (Strøm 1990). Hence, a party's self-estimated likelihood of joining a given coalition (or any coalition) is of crucial importance.

Measuring coalitional prospects

A key component in the preceding argument is that parties adjust their issue salience strategically to improve the likelihood that they will join a particular government coalition. Hence, the likelihood of a given party being included in a coalition must be measured.

Mark Kayser et al. (2022) have developed what they term a measure of coalition inclusion probability, or CIP. Building on earlier models of government formation, most notably Lanny Martin and Randolph Stevenson (2001, 2010), they estimate a conditional logit model predicting the likeliest governments to form. Parties that are included in more and/or likelier potential governments have a larger CIP, while parties that are included in fewer and/or less likely potential governments have a smaller CIP. There are two components of the CIP approach that are useful for the present article. First, the authors calculate what they refer to as the gross coalition inclusion probability, or gross CIP, which corresponds to the sum of predicted probabilities for cabinets including a given party and is a measure of a party's bargaining strength. Second, the predicted probability for each potential government can be used to identify which government is most likely to form that is either both winning and includes the party in question, or is almost-winning if it includes the party in question. In subsequent sections, I will refer to these two classes of potential coalitions as the likeliest coalition. How distant a given party is from its likeliest coalition can then also be calculated.

As has been discussed briefly in the first part of this article, parties that find themselves in an intermediate bargaining position – that is, with neither an especially low nor especially high CIP – are expected to be those that adopt the theorized strategy. This, however, relies on two assumptions holding. First, that parties can with some degree of accuracy assess the likelihood that they will be included in a coalition prior to choosing a strategy. Second, if the first assumption holds, the CIP measure needs to serve as a reasonable proxy for the assessment process used by parties. The first is a reasonable assumption to make as long as the party system is not too dimensionally complex – that is, there are fewer than three salient dimensions. As it is already assumed that parties can reasonably predict their and other parties' post-election seat share and infer their position relative to other parties based on pre-election statements and materials (e.g. election manifestos), the first assumption holds. Concerning the second assumption, if the CIP measure is decomposed, there is little – if anything – that deviates from the information readily available to parties themselves.⁴

Coalition inclusion and strategic action

While one component of a party's likelihood of being a member of a coalition is its choice of strategy, it is not the only one. The ideological character of a potential cabinet contributes to its likelihood of being formed as well (Kayser et al. 2022; Martin and Stevenson 2001, 2010). It would therefore be reasonable that a potential cabinet's ideological position is also of relevance to the choice of strategy by its potential members. Specifically, I assume that parties will make strategic choices with respect to the likeliest coalition. If the party is close to this coalition in policy space, there is little to be gained in terms of coalitionability from adjusting salience. Inversely, if the party is distant, it is expected that the salience reduction strategy is more fruitful, as a reduction in salience of an issue could give the impression of a closer proximity in policy space than under previous levels of salience. However, both are conditional on the likelihood of joining any coalition. It is of little use to adjust the salience of a policy if it is either unlikely to make coalition membership sufficiently more likely to the point where losses elsewhere are offset, or if the party is already in a sufficiently strong bargaining position. In brief, then, parties in a (very) weak or (very) strong bargaining position have little need to adjust their issue salience with respect to improving coalition opportunities. However, as discussed previously, parties that are members of a coalition also need to maintain their own policy brand. Parties in a very strong bargaining position can be expected to have a greater leeway in seeking to differentiate themselves from their coalition partners in this respect. Even if they focus on issues that may be contentious within the coalition, their strong bargaining position makes it difficult to exclude them, and they therefore enjoy a strategically privileged position. While this is not likely to affect the behaviour of parties in a strong bargaining position when they are distant from their likeliest coalition (possibly already a consequence of a diverging policy agenda), those in close proximity can adjust the salience of their secondary dimension upwards in an attempt to strengthen their policy brand relative to their closest peers.

Parties in close proximity to their likeliest coalition and with intermediate bargaining strength find themselves in a precarious situation. Attempting to reduce salience of an issue in order to join a coalition is likely to be damaging to their policy brand, to the extent that the secondary dimension is already sufficiently salient. Parties need to maintain some form of distinction from their closest peers, or potential voters may find little reason to vote for one party over the other based on their policy proposals. Lowering the salience of a dimension that provides such a distinction could therefore be costly. Hence, we should expect these parties to increase the salience of their secondary dimension to maintain policy brand integrity, which may help retain previous voters and potentially attract new ones. While this may induce some degree of risk due to the potential for unsightly compromises, again, the close proximity means that such compromises are more likely to be minor than if the party were distant. Moreover, given the parties' already close proximity, they are unlikely to improve their chances of joining the coalition by reducing the salience of their secondary dimension.

Parties that are distant from their likeliest coalition face a different opportunity structure. Given that they are distant, reducing the salience of their secondary dimension may, as already mentioned, give the impression that they are closer in policy space to the other potential coalition members, or make them more compatible if the distance is due to their secondary dimension. Either way, we should therefore expect parties with an intermediate bargaining strength to reduce the salience of their secondary dimension when they are distant.

The expected behaviour from parties according to their proximity to their likeliest coalition and their bargaining strength is summarized in Table 1. Status quo here refers only to coalition-anticipating salience adjustment strategies, and not to any strategies that predate the coalition-anticipating strategies. For example, a party that chooses to focus more on its secondary issues to attract voters with no subsequent adjustments to improve coalitionability is considered to pursue a status quo strategy.

Table 1. Expected Strategies Based on Coalition Proximity and Coalition Inclusion Probability

Following from the preceding section, the central hypothesis of the article can be stated as follows, with the caveat that it is expected to apply primarily to mainstream parties:

Hypothesis 1: As the distance from the likeliest coalition increases/decreases, a given party with intermediate bargaining strength is more likely to follow a salience reduction/increase strategy for secondary dimensions/issues.

The hypothesis reflects the expected effect of distance in concert with the conditional effect of bargaining strength. Parties that are distant from their likeliest potential coalition can improve their chances of joining it by strategically reducing salience of their secondary dimension, while parties that are proximate to their likeliest potential coalition can stave off potential vote losses by attenuating their particular policy brand in comparison to their proximate peers by increasing the salience of their secondary dimension. Both are conditional on the associated expected costs of adjusting secondary-dimension salience – primarily vote losses, albeit through different mechanisms depending on whether salience is reduced (fewer votes attracted due to positions on the secondary dimension) or increased (loss of votes due to policy compromises reached relating to secondary dimension issues) – which are only sufficiently offset by office and/or policy gains under conditions of intermediate bargaining strength.

Data and methods

To investigate the hypothesis I use panel data, drawing on three different data sources that in concert contain data on historical government formations and coalitions, post-electoral parliamentary parties, policy positions and salience.

Data on governments and their coalitions, the number of seats, pre-electoral coalitions and parties in parliament are retrieved from the Party Government in Europe Database (Bergman et al. 2019, 2021; Hellström et al. 2021), henceforth referred to as PAGED. In total, these data contain information on government formations between 1944 and 2019, with varying start and stop dates on a per-country basis. The data are restructured to be stacked on parties and nested in government formations. Only parties that have been elected to parliament for at least two consecutive elections are included in all the analyses.⁵

I focus my analysis on the election-proximate period – that is, the pre-electoral and post-electoral periods. Parties produce their election manifestos or some equivalent during the pre-electoral period in concert with election campaigns, and parties that are already members of government coalitions have their strongest incentive to differentiate themselves from their coalition partners during this period as well (Sagarzazu and Klüver 2017). Additionally, parties in government can be expected to be punished for defecting from their prior election pledges while still in government (Matthieß 2020). I use contemporaneous data on party positions and salience as we can reasonably assume that parties will make their adjustments based on prospective coalitions, rather than their historical analogues. That parties can reasonably infer the current policy stances and foci of other parties, alongside voters, is a necessary condition for election campaigning to function in the first place. During the immediate post-electoral period, parties then enter some coalition bargaining process if no single party has gained a majority.

To calculate the distance to potential coalitions, there are two primary options available. The first is to rely on data from the Manifesto Project, henceforth referred to as the Manifesto data or MARPOR (Volkens et al. 2020), which gives access to impressive time series. The second option is the Chapel Hill Expert Surveys (CHES) data on party positions (Bakker et al. 2015, 2020a; Polk et al. 2017). While the time series are notably shorter than in the Manifesto data, CHES crucially includes data on multiple scales (dimensions) for the entire period of observation, in full covering the last election before the first survey in 1999 until the last election before the latest survey in 2019. In this article, I test the hypothesis using both sets of data but with the same underlying model. Drawing on the scales used in the CHES data, the first dimension corresponds to the left–right economic scale, while the second dimension corresponds to the so-called Green, Alternative, Libertarian–Traditional, Authoritarian, Nationalist scale, or GAL–TAN. To facilitate comparison with the positions derived from the Manifesto data, I attach categories to the two different scales following the ones suggested by Christopher Prosser (2014) for the comparable left–right economic and sociocultural dimensions. To calculate the positions from the Manifesto data, I use the logit-based scaling method suggested by Will Lowe et al. (2011).

The use of CHES and MARPOR data to calculate positions by necessity introduces limitations on which parties and government formation opportunities from the PAGED data can be used in the analysis, both in terms of temporal coverage due to the time series available, and also which parties have been coded in the respective sets of data. I proceed with three subsets of data in the subsequent analysis. A first subset includes all observations beginning with the first post-electoral government formation opportunity after 1970 and uses MARPOR to calculate party positions. A second subset includes all observations, beginning with the first post-electoral government formation opportunity where CHES data are available. Finally, a third subset uses MARPOR data to calculate party positions but matches the formation opportunities to the same ones as in the CHES subset.

To measure dimensional salience, I again turn to the Manifesto data.⁶ Salience is measured as the proportion of a text that is focused on policy categories associated with the left–right economic and GAL–TAN scales.⁷ To facilitate comparison with the positional data in CHES, categories in the MARPOR data have been attached to each of the two dimensions drawn from the CHES data. I use the log salience measure proposed by Prosser (2014: 96), which leverages the advantages of a log-based approach (see Lowe et al. 2011) while simultaneously producing an easily interpretable measure, ranging between 0 and 1. The measure is calculated as:

$$Salience = \displaystyle{{{\rm log}( {N\;{\rm quasi{\hbox-}}{\rm sentences\;}\;with\;relevant\;codes + 1} ) } \over {{\rm log}( {N\;total\;{\rm quasi}{\hbox-}{\rm sentences} + 1} ) }}$$

The secondary dimension is determined as whichever dimension between the left–right economic and GAL–TAN dimensions has the highest salience score. In the case of ties, whichever is the secondary dimension for most parties in the party system at the present formation opportunity is determined to be the secondary dimension. It is, however, important to acknowledge potential problems with this approach. Given that dimensions are constituted of multiple individual issues, it may very well be the case that parties have a set of core issues belonging to one dimension, but in aggregate give greater emphasis to issues belonging to another dimension. While this may be unlikely given the theoretical rationale behind dimensions – that they are aggregations of bundles of issues where parties tend to hold correlated positions – it is nevertheless something to take into account when analysing the results.

There are two key independent variables (or rather, the interaction between the two is key): distance from the likeliest winning or almost-winning coalition including the party and bargaining strength, the latter represented by the gross CIP for a given party.⁸

The first variable retrieves the likeliest potential coalition, understood as having the highest predicted response rate from the CIP conditional logit model that satisfies one of two different sets of criteria. The first set of criteria is that it is winning, understood as the coalition members controlling 50% + 1 seats, and that it includes the current party. The second set of criteria is that the potential coalition is not winning and does not include the current party but would be winning if it did include the current party. The distance to this coalition's weighted mean position, the weight being coalition seat share, is then measured as the two-dimensional Euclidean distance between the party's position and the weighted mean position of the coalition.

Gross CIP is a measure produced by the conditional logit model discussed above. It is, quite simply, the sum of predicted probabilities for a given party and all possible governments for a given formation opportunity, and as such it has a theoretical range of [0,1]. To facilitate a more straightforward comparison to the different distance measures, the gross CIP measure is linearly transformed to a range of [0,10], corresponding to the same approximate observed range as the distance measurements.⁹ Additionally the gross CIP measure is also squared, as the expected pattern is non-linear with less tendency towards adjusting secondary dimension salience among parties with low or high gross CIP.

Given that there is a conditional expectation for parties to pursue the strategy, namely that parties with an intermediate gross CIP are those expected to pursue a salience adjustment strategy, two interaction terms are included: one between gross CIP and the distance variable, and one between gross CIP squared and the distance variable. The main term of interest is the latter of the two, given the expected relationship between gross CIP and strategy adoption.

Nicheness is here included as a control, as the theorized strategies are primarily expected for mainstream parties. Niche parties are strongly constrained in their choice of strategy as they primarily compete on issues belonging to a secondary dimension (Adams et al. 2006; Meguid 2005, 2008). To measure nicheness, I use the measure developed by Daniel Bischof (2017), which is a joint measurement of a party's market share and specialization on a set of five issues associated with traditional niche party families. I then construct a dummy variable to indicate if a given party is niche based on the distribution for each country, where a party is determined as niche if its nicheness score is greater than one standard deviation above the mean.¹⁰

In addition to nicheness, I also control for membership in a pre-electoral coalition, and if the party is a member of the outgoing government coalition. Pre-electoral coalitions, in which I follow Sona Golder (2006) and include both pre-electoral coalitions and electoral alliances, bring to the fore the one-sided salience adjustments discussed above, as the constituent parties of a pre-electoral coalition publicly commit to some form of policy coordination. We should largely expect similar patterns to those in the salience adjustment strategies, but with respect to the pre-electoral coalition itself. As a final control, I also include the government incumbency status of the party. Incumbent status is expected to have a moderating effect, even though, as already discussed above, the pre-electoral period represents one of the best opportunities for government parties to present a policy programme distinct from its government compatriots. Further controls include the prospective seat share of the party – that is, its post-election seat share¹¹ – and its general left–right position, corresponding to the general left–right scale for the CHES data subset and the general left–right scale proposed by Prosser (2014) for the MARPOR data subsets (Table 2).

Table 2. Descriptive Statistics

Note: The number of observations refers to the total number of combinations of parties and formation attempts.

Empirical analysis

To conduct the analysis, the following (simplified) model is specified:

$$\eqalign{& Multidimensional\;model\colon \;Salience\;of\;secondary\;dimension \cr & \quad = \beta _1[ {Lag( 1 ) \;Salience\;of\;secondary\;dimension} ] + \beta _2[ {Gross\;CIP} ] \cr & \quad + \beta _3[ {Gross\;CIP^2} ] + \beta _4[ {Distance\;from\;weighted\;mean\;coalition\;position} ] \cr & \quad + \beta _5[ {Gross\;CIP\;\times \;Distance\;from\;weighted\;mean\;coalition\;position} ] \cr & \quad + \beta _6[ {Gross\;CIP^2 \times \;Distance\;from\;weighted\;mean\;coalition\;position} ] \cr & \quad + \beta _7[ {Niche\;party} ] + \beta _8[ {Pre \hbox{-} electoral\;coaltion\;member} ] + \beta _9[ {Incumbency} ] \cr & \quad + \beta _{10}[ {Seat\;share} ] + \beta _{11}[ {General\;left{\rm \ndash }right\;position} ] } $$

β ₂ through β ₁₁ correspond to the independent variables introduced in the previous section. β ₁, however, corresponds to a theoretical expectation that parties adjust their issue focus with reference to previous positions (see e.g. Adams 2001; Budge 1994) and/or issue foci. Given this expectation, a once-lagged version of the dependent variable is included in each model. The terms that are most central to the hypotheses are β ₅ and β ₆, as these correspond to the expected conditional relationship between gross CIP (terms β ₂ and β ₃) and distance from the weighted mean coalition (term β ₄). The model is estimated on each of the three data subsets discussed above.

To estimate the model, I use a system generalized method of moment (GMM) estimator (Blundell and Bond 1998) to contravene the issue of so-called Nickell bias (Nickell 1981)¹² that is otherwise present when estimating a dynamic model with fixed effects. When estimating a model using system GMM, determining the relationship between each regressor and the error term is necessary, as it determines which lags can be used as valid instruments (Kiviet 2020: 6–8). Distance from the likeliest coalition is treated as a strictly exogenous variable, as its construction involves a separate process that is unrelated to the salience of the secondary dimension, and hence is expected to be uncorrelated with the error term. The remaining regressors, excepting the lagged dependent variable, which is by necessity endogenous, are treated as predetermined variables.¹³ The model is estimated using the user-created command xtabond2 (Roodman 2009) in Stata. The instrument matrix is collapsed to avoid unnecessarily inflating the number of instruments.

Results

The results of the system GMM regressions are reported in Table 3. The first column reports point estimates for the MARPOR-subset data, the second column the CHES-subset data, and the third column the MARPOR-subset data but matching the post-electoral formation opportunities to those in the CHES-subset data. There is a note of caution in directly comparing the point estimates for the distance variables between the CHES and MARPOR data subsets. Due to the logit transformation performed to obtain party positions from the MARPOR data, there are no fixed endpoints, compared to the 0-to-10 scale used for the CHES positions. However, despite this caveat, the direction of each relationship remains identical across models, with all terms remaining significant except for the distance from coalition variable in the matched MARPOR data subset, indicating that both distance measures perform similarly to some extent. Moreover, the main interaction terms remain significant across data subsets. Niche party status is significant in the MARPOR data subsets, and in the wrong direction, while it is not significant in the CHES data subset. Of the other controls, only the general left–right position of the party reaches customary levels of significance, but with a comparatively small coefficient. A point of concern for the CHES and matched MARPOR data subsets is the low p-value of the exogeneity test for the IV level instruments. I will return to this below when discussing the results from several robustness tests. Moreover, the lagged dependent variable, while significant and in the expected direction, sports a comparatively small coefficient as far as lagged dependent variables are concerned, indicating that prior levels of secondary salience influence later levels of secondary salience, but that it is nevertheless limited. One possible explanation may be that there are systemic differences between different elections on which issues are in focus.

Table 3. Estimates of Main Model

Note: Robust standard errors in parentheses, clustered by party. * p < 0.05, ** p < 0.01, *** p < 0.001.

Interpreting the effects as such is, however, a difficult prospect given only point estimates due to the interaction terms involved. To aid interpretation, the marginal effects for the estimated models (Brambor et al. 2006) for each data subset are plotted in Figure 1. For all three data subsets, the expected curvilinear effect of bargaining strength, represented by gross CIP, is present, with secondary salience trending higher as bargaining strength gets lower or higher. Notably, however, the effect is significantly weaker at the high end of bargaining strength, to the point where it is not significantly different from parties with a moderately high bargaining strength in the CHES data subset and the matched MARPOR data subset. The pattern is most pronounced in the 1970s–onwards MARPOR data subset. Moreover, the effect size – while not especially strong overall – is also the greatest in this data subset. Distance from the likeliest coalition also exhibits the expected pattern, with more distant parties placing a greater emphasis on their secondary dimension, while closer parties do not. Care should, though, be taken when considering the results for the matched elections MARPOR data, as the uncertainty surrounding the marginal effects does not support the effect observed in the other two data subsets. On the whole, however, the results as such thus support the hypothesis that parties in a position of moderate bargaining strength are more likely to downplay their secondary issues, compared to parties in weaker or stronger bargaining positions.

Figure 1. Average Marginal Effects

To test the robustness of the results, I have performed several additional tests.¹⁴ First, three iterations of the same model are estimated, but with 10% of parties dropped at random. The main effects remain in the same direction and significant in each iteration. It does not appear that the results are overly affected by which section of the sample is included in the analysis for any of the data subsets. Moreover, the low p-value for the instrument exogeneity test noted for the estimates from the CHES and matched MARPOR data subsets noted above varies considerably, indicating that certain parties may be particularly influential on the exogeneity of the instruments. Considering that the main results remain nearly identical in these iterations, this lends support to there being an effect. Second, a model using an alternative measure for bargaining strength has been used instead of gross CIP, namely the Banzhaf power index. Similar albeit not identical results to the main model are observed in the estimates using the CHES and 1970s–onwards MARPOR data subsets, while the matched MARPOR data subset does not perform well under this alternative specification. Overall, using gross CIP appears to produce more precise estimates while conforming to the same pattern as using the Banzhaf power index. This is not surprising given the additional information used to estimate gross CIP compared to the Banzhaf index. Third, alternative models that are based on single-dimensional distances rather than a joint two-dimensional distance are estimated. Again, differences between the multidimensional models and their single-dimensional alternatives are largely similar for all data subsets. Finally, I include decade dummies to check if there is any time-variant effect, with the 1990s set as the reference category. While the estimates for the 1970s–onwards MARPOR data subset remains near-identical to the main model, the same does not hold true for the CHES data subset, where all terms except for the decade dummies fail to reach significance, and in the matched MARPOR data subset only the general left–right position reaches significance in addition to the decade dummies. In part, the impact of the decade dummies is expected. The party-system developments that are of theoretical interest to this article – increased dimensionality and fragmentation – are assumed to be most pronounced in the last two to three decades.

A potentially complicating factor in arriving at more precise results is the use of aggregative and predetermined dimensions in the analysis and, moreover, the choice to use the same two dimensions for all countries at all times. It is not necessarily the case that all party systems under review here are structured around the economic left–right dimension and the sociocultural dimension, or at the very least not at all times. While this is partly mollified by the modest time series used here, it may be a more pressing concern if more extensive time series are used.

To summarize, there is a significant and, within the expectation that the phenomenon occurs at the margins, predicted difference in the expected direction for parties with medium-to-low bargaining strength. This also accounts for the majority of parties. Parties that are distant from their likeliest coalition and that have low bargaining strength tend to give more attention to their secondary dimension than parties that have low bargaining strength but are close to their likeliest coalition. As bargaining strength increases, we see the expected pattern of lowering secondary issue salience for parties that are distant and increasing salience for parties that are closer. Hence, the hypothesis being tested finds support, but with a caveat that the paucity of parties with high bargaining strength and high distance means that it is uncertain whether the results hold for this category of parties. Likewise, parties with an intermediate bargaining strength are virtually indistinguishable in terms of secondary salience with regard to their distance to their likeliest coalition.

Conclusions

The aim of this article was to investigate party-issue salience during elections and coalition formation for parties in European party systems. The nature of democratic elections is for parties to compete over votes and thus appeal to voters, while the reality of the majority of European party systems is that joining a coalition is increasingly becoming a necessity if a party is to play a part in government, and it must therefore also appeal to other parties. How do parties strategically anticipate coalition formation? Drawing on the theory of preference tangentiality (Ecker and Meyer 2017; Falcó-Gimeno 2014; Luebbert 1986) and the literature on the risks faced by parties in making coalition compromises (Fortunato 2019, 2021; Martin and Vanberg 2011; Plescia et al. 2022), I advanced two arguments. First, that we should expect what Meyer and Wagner (2019) have called emphasis-based policy change in coalition-anticipating strategies, as adjusting issue emphasis is likely to pose less risk to a party than changing its position on that issue, both during the election and in government. Second, I argued that the deployment of such a strategy is based on the likelihood that a given party will join a government coalition, a measure of its bargaining strength, and the ideological distance from the government coalition that it is likeliest to join. Parties that can meaningfully improve their likelihood of joining a coalition with emphasis-based policy change were expected to decrease emphasis on issues that are of less consequence to them, and increasingly so as they become more ideologically distant from their likeliest coalition. To test these expectations, I leveraged panel data on post-electoral government formations in up to 27 European parliamentary and semi-presidential democracies. Using three different data sources and two different ways to measure party positions, I show that parties by and large do act in accordance with the expectations outlined above. Parties in an intermediate position of bargaining strength tend to decrease their emphasis on their less important issues compared to parties in a (very) weak or (very) strong bargaining position. Likewise, parties that are more distant from their likeliest coalition tend to decrease their emphasis on secondary issues compared to parties that are less distant from their likeliest coalition.

The results have implications for our understanding of party behaviour and coalition formation. While it is already clear that parties consider government coalitions before elections have been held – given the existence of pre-electoral coalitions and electoral alliances – less attention has been paid to less formalized behaviour from parties to anticipate the coalition bargaining process. While the results reported here may not be particularly strong, they nevertheless indicate that (some) parties adjust dimensional salience (and likely issue salience) in a way that is consistent with anticipating the coalition bargaining stage, or, alternatively, that parties who fail to make such adjustments lose out on coalition membership and eventually (electoral) relevance. Regardless, there are substantial implications for how we can interpret party action. First, the results illustrate additional conditions under which we can expect parties to pursue emphasis-based policy change as opposed to position-based policy change. Parties that are uncertain about immanent coalition membership after the upcoming election may be tempted to use an emphasis-based strategy to improve coalitionability or to protect their policy brand. Second, it adds further evidence that it may be fruitful to consider parties in relation to various group memberships, such as being a part of political blocs (Green-Pedersen and Mortensen 2015) or, as in this article, potential coalitions when considering their positioning in policy space. Party leaders are naturally aware of their available credible alternatives and ought to act accordingly.

Another potential approach is to address the topic through an issue-focused approach instead. Using aggregative dimensions (e.g. the left–right policy dimension) may introduce a problem where parties diverge on narrow issues within broader dimensions in a manner contrary to prevailing party-systemic logic – for example, a party that is in favour of both lower taxes and expanded welfare transfers. An analytical framework reliant on dimensions can miss such cases and potentially bias the estimates, which an analytical framework based on issues is less susceptible to. While the potential for unworkable levels of complexity remains, this could be avoided through a more restricted research design, focusing on certain sets of issues and/or countries rather than a particularly broad set of both (see e.g. Otjes and Green-Pedersen 2021 for a recent example of such a restricted design). Given that there is a significant effect even at the high aggregative level of issue dimensions, there may be an even more pronounced effect if each dimension is decomposed into their constituent issues. It also seems plausible that parties choose to avoid issues that may have adverse consequences to their coalition prospects, in an analogue to how parties avoid policy issues that are internally contentious within either their own party or a government coalition.

A further possible avenue for future research is to focus more on particular party systems, rather than the comparative approach employed here. While this article provides limited evidence that parties do employ salience adjustment strategies, it does not account for how these strategies are deployed beyond their presence in election manifestos. This may be suitable for both qualitative and quantitative approaches. Moreover, it may well be the case that parties exclude contentious issues from their election manifesto – a written record – while they may raise them in less formal contexts.