Party campaign statements and portfolio allocation in coalition governments

Abstract Which party controls which cabinet posts is an important determinant of how multi-party governments work. Existing research shows that parties’ attention to policy domains in election manifestos is a key predictor of portfolio allocation. However, election manifestos are broad documents and typically published months before an election. This research note argues that policy emphasis in the last few weeks before the election matters for portfolio allocation, because parties can focus their message, react to exogenous events and use campaign communication as a commitment device. A test of this argument makes use of a novel dataset on party representatives’ campaign statements. The findings show that the policy focus of campaign statements, especially those stating positions rather than referring to valence, predicts who will control a ministerial portfolio associated with the respective policy domain.

In the weeks before the 2021 German election, future coalition negotiations between the Social Democratic Party (SPD), the Free Democratic Party (FDP) and the Greens -parties with very different views on the appropriate scope of public spending -had become a realistic scenario (and such a government eventually also formed).During the campaign, the liberal FDP emphasised fiscal restraint and a tax freeze as important policy goals.At the same time, the party expressed a strong interest in winning control over the finance ministry.Indeed, the question of who would obtain this important portfolio spurred conflict during the negotiations.Because of the strong signals the FDP had sent with regard to both fiscal conservatism and their interest in that particular cabinet post, it was clear that a coalition among the three parties would hardly see the light of the day if the Free Democrats did not receive this portfolio.
This example illustrates how political parties can use campaign communication to signal their policy commitments to voters and competitors alike.It also echoes empirical results showing that patterns of portfolio allocation influence how multi-party governments work during the legislative period (e.g.Bergman et al. 2021).Moreover, the control over key cabinet posts matters for the degree to which parties can implement their policy pledges during the legislative term (e.g.Bäck et al. 2017).
In this contribution, we argue that the programmatic signals parties send in the final weeks of an election campaign explain -over and beyond the content of the parties' election manifestos (Bäck et al. 2011) -the outcomes of portfolio allocation.The more parties highlight a certain policy issue in the last weeks before the election, the more likely they are to obtain the related cabinet post.To evaluate this argument empirically, we start from existing data on the issue salience of parties as measured in their election manifestos (MARPOR; Volkens et al. 2017) and data from the Coalition Governance in Western Europe project.We then add an updated version of the Comparative Campaign Dynamics Dataset (CCDD; Debus et al. 2016) that provides information on the statements that party representatives made during the final weeks of an election campaign, covering seven European countries between 2007 and 2019.The findings indicate support for our argument: The more parties emphasise a policy domain not only in their election manifesto but also in the last weeks before the election, the more likely it is that the respective party wins control over the cabinet post associated with the respective policy domain.These findings also speak in favour of the real-world relevance of election campaigns.

Portfolio allocation and its link to parties' policy priorities
Portfolio allocation is a central aspect of government in parliamentary democracy, especially in multi-party governments.The cabinet delegates decision-making power to individual ministers (Strøm et al. 2006), which implies a considerable amount of discretion (Bäck et al. 2022;Martin and Vanberg 2020).Ministries play a key role in drafting legislation, and holding a portfolio provides a politician and her party both agenda-setting powers and influence over the content of legislative bills.
Political parties vary in the priorities they attach to policy areas, and hence may also seek different ministerial portfolios.Bäck et al. (2011) use the salience of policy areas in election manifestos as a predictor of parties' portfolio preferences.This is intuitive, since an electoral manifesto represents the 'official' policy package that has typically been ratified by a party convention or other formal party body (Däubler 2012;Dolezal et al. 2012;Eder et al. 2017).Manifestos address both voters and intra-party groups, so they should give a good indication of the overall importance a party attaches to a policy area.Empirically, the relative coverage of the underlying policy area in a party's election manifesto is a strong predictor of who gets which portfolio (Bäck et al. 2011;Ecker et al. 2015).
While party manifestos provide important information about party policies and their relative coverage, we argue here that the parties' campaign focus will add to an explanation of which party obtains which portfolio.Although voters of course have no direct influence in the coalition negotiation process, the pre-electoral campaign can affect post-electoral bargaining for at least three reasons.
First, certain aspects of the context in which manifestos are written limit their quality as indicators of relative priorities.Manifestos are broad and encompassing documents in which parties cover a wide range of issues (Tresch et al. 2018).Unlike authors of manuscripts for academic journals, manifesto writers face comparatively mild space constraints and need not be particularly selective.In addition, contextual party-level factors such as outgoing government status (Greene 2016) or the party's position in the policy space (Bergman and Flatt 2021) can affect the diversity of issue appeals in manifestos.As Helbling and Tresch (2011: 179-181) report, the correlation between the salience in manifestos and that in media reports is weak (and much lower than for positions).
Second, manifestos are typically written months before the election takes place (Däubler 2012;Dolezal et al. 2012;Eder et al. 2017).In the meantime, the political context may have changed considerably.Exogenous events (Dalmus et al. 2017) can shift the agenda, with new political challenges appearing, and voters may adjust their preferences.Vote-seeking parties are likely to be responsive to voters' new priorities (Wagner and Meyer 2014) and may subsequently also aim at controlling portfolios to implement them.Also, expectations regarding electoral performance may have been updated based on opinion polls.This can alter the set of potential majority coalitions, which can result in the need to send out further programmatic signals to claim programmatic territory.
Third, even a purely vote-or office-seeking party faces incentives to deliver in areas it emphasises in the electoral arena -in order to remain credible (Klüver and Spoon 2020).Succinct campaign statements in the media are much more visible to a broader audience than is lengthy manifesto text.This can also help a party in obtaining the portfolios it prefers for policy reasons at the negotiation table, in the form of a commitment device.In negotiations with potential coalition partners, the party can point to the negative electoral consequences it would suffer from a failure to deliver on those campaign-time promises.If negotiating parties are interested in the success of the formation process, they need to balance such demands by their future partners with their own portfolio preferences.
Taken together, we expect that -even when controlling for further important variables that influence the allocation of cabinet posts to coalition parties, such as issue salience in the manifestos and the intra-coalition strength of parties -emphasising a policy domain in the last weeks of the election campaign should have a positive (predictive) effect on the chances that the party will gain control over the relevant ministry in later coalition negotiations.

Data sources
The evaluation of this hypothesis requires data on the policy profile, i.e. issue emphasis, of parties during the last weeks of an election campaign.We make use of an updated version of the CCDD (Debus et al. 2016; see Baumann and Gross 2016 for a description) that provides the required information based on the coding of party statements in media reports.The set of countries covered in the analysis includes two Scandinavian (Denmark and Sweden), two Continental European (Germany and the Netherlands), one Mediterranean (Portugal) and two Central and East European countries (Czech Republic and Poland).For each covered election, the data provide information on parties' campaign statements from the two broadsheet daily newspapers with the highest circulation, during a pre-election period of 30 days.Since the ideological profile of newspapers can influence the chances that party messages make it into the news (e.g.Haselmayer et al. 2017), the dataset covers a left-and a right-leaning newspaper in each country (see Online Appendix 1 for details).
The CCDD covers all front-page articles related to the campaign; if necessary, it has complemented this set using a 5% random sample of other election-related articles until a minimum number of 60 articles per newspaper/election has been reached.In line with the existing literature on issue salience, we operationalise the relative emphasis of each policy area in party messages as the number of policy statements referring to that area divided by the total number of statements by the party.From the CCDD, we utilise the information about statements that parties make about themselves, concerning issue positions and issue-specific valence.We use two different measures of relative emphasis, since we do not see clear theoretical arguments as to why one of them would be preferable: one considering both types of statement and one considering only positional statements.
These data on issue emphasis during an electoral campaign are complemented by information on the political parties involved in the respective government formation processes, the policy area(s) attached to each ministerial portfolio and the key dependent variable, i.e. the outcome of the allocation process indicating which party gets which portfolio.For Western Europe, we obtain these data from the European Journal of Political Research yearbooks and Keesing's World News Archive.For Central and Eastern Europe, we retrieve this information from the corresponding country chapters in Bergman et al. (2019).Overall, our sample includes 518 potential party-portfolio allocations in 18 coalition governments between 2007 and 2019. 1  In this context, one influential coding decision is assigning each ministerial portfolio the corresponding policy profile, i.e. matching the relevant party statements in the CCDD in order to quantify the campaign salience of all government parties for each ministerial portfolio.In line with the approach established by Bäck et al. (2011), we aggregate various issue categories into broader issue areas that are, in turn, allocated to the corresponding portfolios. 2  Since the coverage of the CCDD is limited, the sample size in our analysis is fairly small.We therefore focus on a small set of essential control variables only.More specifically, we include the seat contribution of a coalition party to adjust for the well-established pattern implied by 'Gamson's Law' (Gamson 1961) -parties should be more likely to obtain a post if they provide more legislative seats to the governing coalition.We interact this measure with an indicator variable for the ministry of finance.In many political systems, this ministry is a particularly important portfolio (Druckman and Warwick 2005), and larger parties should have an advantage in securing this precious prize.

Statistical model
Our interest lies in explaining which of the parties in a coalition receives a certain cabinet post.In other words, the coalition as a collective actor decides which of the partners obtains each of the portfolios that form part of the current cabinet. 3We use a conditional logit model (McFadden 1973) to explain the outcome of the decision-making process with covariates.The main specification is equivalent to the approach taken by Bäck et al. (2011) and defines each portfolio in each coalition as a separate choice set.The model predicts which of the N parties in that coalition receives that portfolio, taking into account the salience measures and the other covariates.As discussed in Online Appendix 3, using an alternative specification (which seeks to explain both between-party and within-party outcomes) yields similar results.

Results
Before discussing the findings of the regression models, it is worth pointing out that the correlation between manifesto-based and campaign-based salience is very small (Pearson's r is 0.12; see Online Appendix 4).Hence, these measures do indeed capture different aspects of party communication and can incorporate more recent events that could not be considered when the parties prepared the policy platforms.For example, the attention given to defence policy in the German Christian Democrats' (CDU/CSU) manifesto of 2009 was only 4.3%, whereas 25% of their campaign statements focused on this topic.The Kunduz airstrike in Afghanistan, which was called by German military forces and took place just three weeks before the election, killed 90 civilians and brought defence policy to the top of the political agenda.
Table 1 reports the results of the models that consider the allocations of the various portfolios within a cabinet as independent decisions.The models vary with regard to the policy focus variable that is included.Model 1 considers (only) manifesto salience, Model 2 campaign salience based on all statements and Model 3 campaign salience based on positional statements.The importance attached to policy areas in both arenas is included in Models 4 (with campaign salience based on all statements) and 5 (with campaign salience based on positional statements). 4 The seat share that a party contributes is coded as a proportion, so the results for Model 1 suggest that an increase of 10 percentage points increases the baseline odds of obtaining a portfolio by approximately 56% for the ministry of finance and by approximately 18% for others.notes: robust standard errors in parentheses, + p < 0.10, * p < 0.05, ** p < 0.01. to assess model fit, the observation with the highest predicted probability within the choice set is considered to be predicted as allocation, and the others as non-allocation.
The coefficient of the manifesto-based salience measure has the expected positive sign and is statistically significant at the 1% level.Using campaign focus as a predictor (Model 2) yields similar results (significant at the 5% level).Model 3 shows that the predictive effect of campaign focus is stronger if we consider only party statements covering policy positions (and not valence).The inter-quartile range of this measure is close to 10 percentage points; an increase of that size raises the odds of receiving the portfolio by a factor of 1.89.A comparison of the model fit measures underlines the predictive performance of the campaign-based measures.The share of correct predictions and the sensitivity are higher in Models 2 and 3 than in Model 1. Models 4 and 5 include relative emphasis measures from both manifestos and the campaigns.The coefficient for the manifesto-based measure is smaller than in Model 1 but remains statistically significant (at the 5% level) in both models.The campaign-based measure also has positive predictive effect in both Model 4 and Model 5.Even when holding emphasis in the manifesto constant, a 10 percentage-point increase in campaign focus multiplies the odds of obtaining the portfolio by a factor of, respectively, 1.40 (Model 4, positional and valence statements) or 1.62 (Model 5, only positional statements).Therefore, the findings point out that emphasis of especially policy positions during a campaign is linked with subsequently obtaining the respective portfolio.Finally, the fit in Models 4 and 5 using both types of measure is also higher than in Model 1.
In order to illustrate further the patterns implied by the results, we calculate predicted probabilities by simulating draws from the multivariate normal distribution of the regression coefficients.We do this by assuming a two-party coalition with a small party (contributing 15% of the government's seats in parliament) and a large one (85% of seats).Figure 1 shows predicted probabilities of obtaining the portfolio for increasing values of the manifesto-based salience of the smaller party (party 1), for three different values of manifesto-based salience for the larger party (colours/shapes).If the small party paid no attention at all to the underlying policy areas, its predicted probability to obtain the portfolio is approximately 0.14 (with average emphasis by the larger party).If the smaller party allocated a quarter of its manifesto to the policy area (close to the 99th percentile in the sample), we expect it to receive the portfolio with a probability of approximately 0.48.The values are of course somewhat larger (smaller) if the coalition partner spends below (above) average attention to the policy field itself.
Figure 2 proceeds similarly, measuring campaign focus exclusively on the basis of positional statements by parties.The predicted probability of receiving a ministerial post is approximately 0.16 if the smaller party did not make any statements about its policy positions in the underlying area during the campaign.In contrast, if a quarter of its position-related statements referred to this field, the probability increases to a value of approximately 0.45.Overall, even comparatively small changes in attention during the campaign (recall that the inter-quartile range is approximately 0.10) imply considerable differences in predicted probabilities of obtaining the respective ministry -over and beyond relative coverage in the manifesto.

Conclusion
The allocation of cabinet posts is a central stage during coalition negotiation processes and has important implications for patterns of coalition governance.In this contribution, we showed that the parties' campaign focus indeed matters for office allocation in coalition governments.The results indicate that parties adjust their policy profile not only to maximise their vote share in the upcoming elections but also to send signals to future coalition parties in terms of their preferred cabinet posts.In this context, the communication of policy positions appears more important than a discussion of valence aspects.Generally, such signalling seems particularly relevant in times of an increasing share of voters who decide late in the election campaign, because this type of volatility heightens uncertainty regarding which coalitions are viable.The support of alliances that commanded a majority in opinion polls at the time when the parties published their manifestos may have eroded, while other coalition options may have become more realistic.
Of course, what the media reports about the parties and the statements by those parties' representatives may depend on the preferences of the journalists and editors of the particular news outlets.Nevertheless, we think that drawing more attention to statements of parties closer to the election can be a fruitful strategy for analysing more accurately the coalition preferences of parties in general and their preferences for cabinet posts in particular.Given the findings presented here, the campaign statements by parties as reported by the media have slightly better predictive power than the focus inferred from election manifestos when explaining portfolio distribution in European coalition cabinets.This also speaks against the sceptical view that election campaigns are nothing more than superficial exchanges that have little practical relevance.Public Choice, among others. [alejandro.ecker@uni-heidelberg.de]

Figure 1 .
Figure 1.simulated allocation probabilities by change in manifesto focus.notes: the figure is based on 5,000 draws from a multivariate normal distribution using the results from Model 5 (shown are means and the middle 90% of the simulated values).the range of the x-axis corresponds roughly to the 1st and the 99th sample percentiles.other covariates receive the following values: seat contribution for party 1: 0.15; seat contribution for party 2: 0.85; portfolio is not the finance ministry; salience in campaign for both parties at sample mean of 0.079.the salience in manifesto for party 2 as represented by different shapes/colours is set to 0.082 (sample mean), 0.036 (p25) and 0.117 (p75).

Figure 2 .
Figure 2. simulated allocation probabilities by change in campaign focus (positional statements).notes: the figure is based on 5,000 draws from a multivariate normal distribution using the results from Model 5 (shown are means and the middle 90% of the simulated values).the range of the x-axis corresponds roughly to the 1st and the 99th sample percentiles.other covariates receive the following values: seat contribution for party 1: 0.15; seat contribution for party 2: 0.85; portfolio is not the finance ministry; salience in manifesto for both parties at sample mean of 0.082.the campaign salience for party 2 as represented by different shapes/colours is set to 0.079 (sample mean), 0.019 (p25) and 0.118 (p75).
of voters and party representatives, party politics, legislative behaviour and coalition politics.He has published in the European Journal of Political Research, the Journal of Politics and Political Science Research and Methods, among others.[marc.debus@uni-mannheim.de] Alejandro Ecker is Junior Professor of Politics and Communication at Heidelberg University.His research focuses on the effects of political institutions on multi-party governments, political parties and individual politicians and their consequences for citizen behaviour and voter attitudes.He has published in the British Journal of Political Science, the European Journal of Political Research and

Table 1 .
conditional logit-regressions modelling the allocation of specific portfolios between parties.