Serious Conflicts with Benign Outcomes? The Electoral Consequences of Conflictual Cabinet Terminations

Abstract

Conflictual cabinet terminations are seismic events in democracies, but their consequences are understudied. I argue that the electoral impacts of conflictual cabinet terminations depend on voters’ perceptions of them. Terminations following non-policy conflicts are electorally costly. They signal parties’ deteriorating governing competence, which reduces parties’ vote shares. In contrast, terminations following policy conflicts signal parties’ unwillingness to compromise their policy positions and clarify parties’ policy profiles, thus allowing them to evade voter punishment and junior coalition parties to reap electoral reward, particularly for those terminations preceded by interparty policy conflicts. Statistical analyses using the Party Government in Europe Database dataset support my argument on policy terminations and reveal more nuanced electoral effects of non-policy conflict terminations. These findings are robust to various alternative explanations, as well as multiple cabinet terminations and time passed from termination to election. The findings have large implications on electoral accountability of intra-cabinet conflicts and the quality of governance.

Terminating a cabinet is one of the most seismic events in parliamentary democracies. “Critical events” that are conflictual in nature (Diermeier and Stevenson 1999; 2000; Lupia and Strøm 1995), such as policy failure, scandals, and policy disagreements, heighten the risk of termination. But what are the electoral consequences of conflictual cabinet terminations? The answers have so far focused on the spatial distribution of the party system and institutional contexts (Mershon 2002), the electoral fates of leavers versus remainers of a cabinet (Warwick 2012), and the moderating effect of policy responsibility concentration (Narud and Valen 2008). Yet, parties’ accountability for conflict terminations may also depend on the circumstances behind them. A party that leaves a cabinet may be laden with blame if it was embroiled with scandals, but may be perceived as blameless if it leaves in order to remain loyal to its supporters. Yet, there is very little research on how the types of conflicts influence the electoral impacts of conflict terminations. This is surprising given their implications on governance.

I address this lacuna in this article. I contend that the electoral impacts of conflict terminations depend on whether these terminations reflect governing competence shocks or parties’ adherence to their policy profiles. Cabinet terminations arising from non-policy intra-cabinet conflicts (termed non-policy conflict terminations), such as scandals, personnel infighting, and poor policy performance, signal deteriorating governing competence and are, therefore, electorally damaging for the parties responsible for the conflicts. In contrast, terminations following intra-cabinet policy conflicts (termed policy terminations) do not harm the responsible parties’ electoral fortunes. These types of terminations reveal to voters that the ruling parties are unwilling to sacrifice their policy positions for the sake of retaining office. Since voters disapprove of compromises and appreciate parties with clear policy positions while in office, ¹ policy terminations should not be evaluated negatively by voters. Thus, ruling parties responsible for the policy conflict behind a termination should not lose votes. And, since intra-coalition conflicts clarify coalition parties’ policy positions (Spoon and Klüver 2017), policy terminations should clarify the responsible junior partners’ policy positions and signal their unwillingness to compromise them. Since both are electorally beneficial to junior partners, policy terminations should be electorally rewarding for them.

I employ data from the European Representative Democracy Data Archive’s (ERDDA) Party Government in Europe Database (PAGED) dataset (Bäck, Debus, Dumont 2011; Hellström, Bergman, and Bäck 2021) to test my arguments. Results from multilevel linear regression models show that policy terminations electorally benefit junior coalition partners, particularly if the policy conflicts are between ruling parties, and do not harm the electoral prospects of prime ministers’ parties. Meanwhile, non-policy terminations damage prime ministers’ parties’ electoral performances, but not those of junior partners. These findings have important implications on voters’ political knowledge, coalition governance, and electoral accountability. First, they suggest that voters are politically sophisticated enough to discern the circumstances behind cabinet terminations. Second, they raise questions on why some non-policy conflicts evolve into government breakdown, particularly in light of the corrective effect that ministerial replacement can have on the government’s electoral image (Dewan and Dowding 2005). Third, my findings complement existing works on coalition governance. They offer additional insights on the policy-office trade-offs that junior partners face and on governing parties’ optimal strategies when facing policy conflicts. Instead of compromising their policy positions for the sake of governance, ruling parties can stick to them by abandoning office without risking vote loss. While this is good news for policy representation, the potential disruption in governance may lead to political instability and inefficient policymaking, particularly given the long cabinet formation process in some democracies. These findings suggest a need to further examine how intra-cabinet conflicts affect electoral and governing accountability.

THEORETICAL EXPECTATIONS

Cabinet terminations during the constitutional inter-election period (CIEP) are on average electorally costly in Western Europe (Narud and Valen 2008). Yet, not all terminations are evaluated equally. Voters would not raise their eyebrows for terminations that accompany the end of an election cycle. In contrast, they reward governments that call early elections while riding the crest of popularity, but remain wary of those that do so in attempt to evade potential future voter punishment (Schleiter and Tavits 2016). Meanwhile, terminations that result from intra-cabinet conflicts may be electorally costly (Mershon 2002). Scandals, political deadlock, or poor handling of natural disasters hinder government’s effectiveness in policymaking. Parties that choose to leave a government may even experience vote loss at the next legislative election (Warwick 2012), perhaps due to voter punishment for the disruption in governance.

Although conflict terminations stem from intra-cabinet crises, some ruling parties seem to evade voter punishment. For instance, in the 2003 Dutch general election, despite dissolving the government only 5 months after its formation, the Christian Democrats (CDA) gained one seat and were allowed to govern for another 7 years. This suggests that voters do not evaluate all conflict terminations the same way. In particular, non-policy conflicts may be perceived differently than policy conflicts. As such, even if both types of conflicts end up breaking the government, they may result in divergent electoral outcomes.

To understand the electoral impacts of conflict terminations, it is important to first determine what voters want from governing parties. Voters care about governing competence—that is, how effectively parties govern and handle issues that are important for all voters. They hold the government accountable for its record in office, particularly for “valence” issues that all voters care about, such as the state of the economy (Duch and Stevenson 2008; Fisher and Hobolt 2010). Incumbents’ mishandling of important issues reduces voters’ perceptions of their governing competence (Green and Jennings 2012). Also, when incumbent parties experience “competence shocks” (i.e., a sudden drop in voters’ evaluations of a party’s competence in handling one issue), this loss of competence results in a loss of general competence—that is, competence in handling other issues (Green and Jennings 2017). This, in turn, is met with voter punishment (Greene and Haber 2015). Ruling parties’ electoral performances also in part depend on their ability to prevent scandals and personnel infighting, as these events also damage voters’ perceptions of a party’s governing competence (Clark 2009; Dewan and Dowding 2005; Green and Jennings 2012).

Meanwhile, voters value parties’ faithfulness to their ideological profiles and punish governing parties that dilute their policy positions. Evidence for this is plenty. Parties that stray from their principled positions lose votes (e.g., LGBTQA+ rights), whereas those that shift their positions in a pragmatic manner (e.g., responses to economic crises) do not (Tavits 2007). In the American context, strong partisans prefer their parties to behave in a partisan manner, even as they disavow interparty conflict (Harbridge and Malhotra 2011). In multiparty governance, partisans of cabinet parties as well as the politically disinterested punish coalition parties that are perceived to be compromising (toward the prime minister’s party) (Fortunato 2019; Greene, Henceroth, and Jensen 2021; Klüver and Spoon 2020). Since voters care about governing competence and ruling parties’ adherence to their policy profiles, their reactions to a conflict termination should hinge on whether the termination projects deteriorating governing competence or sharpens voters’ perceptions of the ruling parties’ policy profiles, at least for those involved in the conflicts.

The above implies that the electoral consequences of conflict terminations depend on the policy versus non-policy nature of the conflicts. I first argue that non-policy conflict terminations are electorally costly to the responsible parties. These types of conflicts encompass scandals with individual ministers, mishandling of natural disasters and other exogenous crises, blaming of governing partners for performance output, and so forth. One example of such a government-ending conflict is the 1994 breakdown of the Fianna Fáil-Labour Party coalition government in Ireland. The crisis was precipitated by the potential appointment of Harry Whelahan, then Attorney General, to become the President and Judge of the High Court. Then-Taoiseach Albert Reynolds had committed to appointing him. However, it was revealed that Northern Ireland’s request to extradite Brendan Smyth, a priest accused of child sex abuse, was deliberately delayed, possibly by Whelahan due to his religious views. This led to public outcry and subsequent opposition to Whelahan’s appointment. Yet, in November 1994, Reynolds appointed him as President of the High Court during a cabinet meeting—despite the absence of Labour ministers. This pushed the Labour Party to withdraw from the coalition. Subsequently, both Whelahan and Reynolds resigned from their respective posts, and a replacement government consisting of Fine Gael, Labour, and the Democratic Left Party was formed.

I argue that non-policy conflict terminations send a strong signal of deteriorating governing competence of the responsible parties. The loss of competence resulting from these conflicts may itself be electorally costly. However, there are instances in which the government can remedy the damage. For instance, when a ministry becomes embroiled in a scandal, the prime minister can fire the minister responsible. Doing so has a corrective effect on voters’ perception on the government’s governing competence (Dewan and Dowding 2005). However, if the conflict is severe enough that the only recourse is to end a government, this signals to voters that cabinet members have lost faith in the government’s ability to continue functioning. That is, the non-policy conflict has debilitated the government to the point of breakdown. As such, a non-policy termination unequivocally reveals deteriorating governing competence of the responsible parti(es), thus damaging their electoral prospects:

Non-Policy Conflict Termination Hypothesis (H1): On average, if the termination of a cabinet is preceded by a non-policy conflict, the part(ies) responsible for the conflict lose votes in the next parliamentary election.

I next examine the electoral impacts of policy terminations—that is, those preceded by policy conflicts within or between cabinet parties. The breakdown of the Syse cabinet in Norway in 1990 is one such example. The coalition consisted of the Christian Democrats, the Center Party, and the Conservative Party, with Jan P. Syse from the latter as prime minister. The coalition agreement contained a “suicide paragraph.” As an important part of coalition bargaining in Norway, it stipulates that the coalition will dissolve if an issue that the parties disagree on and remain unresolved at the time of coalition formation becomes politically salient. The issue at hand was European integration; whereas the Conservative Party was pro-EU, the Center Party was strongly Eurosceptic. In 1990, 1 year after coalition formation, negotiations began on the implementation of the European Economic Area Agreement, in which Norway, a non-EU country, was a signatory. Although the two parties previously discussed the general provisions of the agreement, they subsequently disagreed on the agreement’s implementation, particularly on its consequences on the Industrial Concessions Act and the Act on Financial Institutions. This led to the Center Party’s withdrawal from the coalition, which then collapsed and was replaced by a Labour minority government.

At first glance, policy terminations may signal deteriorating governing competence. Not only did the responsible parties fail to resolve intra-cabinet conflicts for the sake of governance, but there may be accusations that these parties lack commitment for co-governance or the internal discipline for keeping their parliamentarians in line. This may suggest that policy terminations would be evaluated in the same manner as non-policy conflict terminations. Yet, this perspective ignores how government participation affects voters’ perception of parties’ policy positions. Governance carries the risk of diluting a party’s policy positions, which is exacerbated in coalitions. Spoon and Klüver (2017) show that voters misperceive coalition parties’ policy positions, whereas Fortunato and Adams (2015) reveal that voters conflate the positions of junior coalition partners with those of the prime minister’s party. Greene and Haber (2017) even find that parties who form pre-electoral coalitions converge their policy positions toward each other, provided that these coalitions result in vote gains. As such, parties that participate in coalitions become less distinct from each other, at least in voters’ eyes. This is electorally costly. As discussed, voters hold ruling parties accountable for drifting from their ideological positions, for yielding too much to their governing partners, or for not being visible enough in the policymaking process. Thus, a ruling party needs to project clear policy stances to reduce vote loss.

This dynamic is important for understanding how voters perceive a policy termination. This event occurs because the ruling part(ies) involved in the policy dispute cannot reach a compromise within or between themselves. That is, it is not the conflict itself that signals the responsible parties’ adherence to their policy positions, but it is the act of terminating the government because of it that does so. Thus, following a policy conflict, cabinet termination relays to voters that the cabinet part(ies) involved in the conflict refuse to give up their policy positions for the sake of office, which helps these parties clarify their policy profiles. Thus, instead of projecting governing incompetence, a policy termination should reveal the responsible parties’ policy positions clearly to voters.

Following this logic, policy terminations should be electorally consequential for the responsible ruling parties. I first discuss the consequence for prime ministers’ parties. In coalitions, the prime minister’s party possesses significantly more policymaking power than its partners (Klüver and Spoon 2020; Warwick and Druckman 2006). Since voters disapprove of compromise, they should punish prime ministers’ parties that are seen as compromising their policy positions to those of junior coalition partners. A prime minister’s party that dissolves a coalition over an interparty policy disagreement sends the signal that it upholds its policy profile. Subsequently, voters should view these parties’ electoral pledges and policy positions in a more credible manner.

The same can be said for terminations preceded by intra-party policy conflicts, including those within single-party governments. Policy conflicts within the prime minister’s party would reveal tension between the desire to adhere to the party’s policy stances, or at least those advocated by some parts of the party, and the need to compromise for the sake of governance. Although intra-party disagreements project deteriorating governing competence, the fact that the conflict is government-ending implies that the party is unwilling to shortchange its core beliefs in order to remain in office. Thus, despite the disruption in governance caused by the government-ending conflict, neither intra- nor inter-party-based policy terminations should signal deteriorating governing competence. Rather, they highlight the prime minister’s party’s policy profile and its unwillingness to sacrifice policy for the sake of office. Consequently, policy terminations should not result in voter punishment for the party:

Policy Termination and Prime Ministers’ Parties Hypothesis (H2): On average, a cabinet termination that was preceded by a policy conflict is not electorally costly for the prime minister’s party responsible for the conflict.

I also contend that since policy termination improves voters’ accuracy in placing parties’ policy positions, it should be particularly beneficial for junior coalition partners. Policy visibility matters for these partners; those that take charge of their own policy profiles fare better at the polls. ² Also, Spoon and Klüver (2017) show that although voters often inaccurately perceive coalition parties’ policy positions, policy conflicts between these parties correct this misperception. Publicizing a policy conflict by way of cabinet termination reveals the nature and depth of disagreement on the policy issue in focus. This provides voters with clarity on what the junior partner(s) in conflict stand for and that they are unwilling to compromise. This benefits junior partners, as voters become less likely to map these parties’ policy positions onto those of the prime minister’s party. As such, given responsibility, policy terminations enhance the visibility and clarity of a junior partner’s policy profile, more so than it would for a prime minister’s party.

This is electorally consequential. As discussed, junior coalition partners suffer from greater policy dilution than the prime minister’s party, which makes them lose comparatively more votes in a parliamentary election. Projecting a clear policy profile would provide an electoral boost for these parties. Since policy terminations reduce voters’ uncertainty on what these parties stand for, they clarify junior partners’ policy positions and distinguish junior partners’ policy profiles from that of the prime minister’s party. Both raise their electoral relevance. Thus, I expect that, assuming responsibility for the conflict, policy termination is more electorally beneficial for junior coalition parties than for the prime minister’s party:

Policy Termination and Junior Partner Hypothesis (H3a): On average, given responsibility for the policy conflict, the electoral impact of a policy termination is more positive for junior coalition parties than for the prime minister’s party.

This is consistent with the aftermath of the breakdown of the Syse cabinet. The Center Party held fast its Eurosceptic position and saw their vote share increase by a whopping 10.27 percentage points in the 1993 Norwegian parliamentary election. The Center Party’s leader, Anne Enger, even became the face of the “No to EU” campaign for the 1994 Norwegian referendum on EU membership (in which the majority of voters voted against). While the party’s electoral performance may not be solely attributed to its withdrawal from the coalition, the direction and magnitude of vote share change, as well as its subsequent role in the referendum, suggest that at the very least, the Center Party stood to gain from the coalition termination, as it allowed the party to clarify its position on European integration.

I further argue that junior coalition partners enjoy this electoral benefit only if the policy termination stems from interparty policy conflicts. A coalition termination preceded by policy conflict within a junior partner does not project clear policy stance, at least not to the same degree as one that is preceded by an inter-party conflict. The latter signals to voters that the junior partner would rather sacrifice office than compromise its policy positions, which would enhance the distinctiveness of the party’s policy profile in relation to that of the prime minister’s party. The former also signals to voters that the party refuses to compromise its core positions for the sake of governance. However, since the policy discord lies within the party, voters’ perceptions of the party’s policy positions may remain muddled, even as they are perceived as policy-oriented. Thus, while prioritizing policy over office benefits should not be electorally damaging for junior partners, voters should be less likely to reward them for policy terminations:

Inter-Party Policy Termination and Junior Partner Hypothesis (H3b): On average, given responsibility for the policy conflict, policy termination electorally benefits junior partners only if the conflict spans between coalition parties.

EMPIRICAL TESTING

I focus on the electoral impacts of conflict terminations. As such, I need to identify cabinet terminations that are the result of intra-cabinet conflicts. To do so, I employ the PAGED dataset (Bergman, Bäck, and Hellström 2021). This dataset contains information on all post-WWII conflict terminations in Western Europe. It provides the relevant data on the circumstances behind each government dissolution and important characteristics of each cabinet. To the best of my knowledge, it is the only dataset that provides systematic coding for conflictual cabinet terminations, the part(ies) responsible for the conflict, and if the termination follows public opinion shocks; other election outcomes; inter-personal discord; or policy disagreements. These are central for allowing me to identify both types of conflict terminations. The data were hand-coded by looking through at least one of the three sources—the Political Data Yearbook, Keesings Data Archive, and various country reports. Each country’s data have been verified and corrected by country experts. ³

My unit of analysis is an individual party that has participated in a non-caretaker government. I include the following parliamentary democracies in my analyses—Austria, Belgium, Denmark, Finland, Germany, Greece, Iceland, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, and the United Kingdom. ⁴ This set of countries has been under stable democratic rule since at least 1980, and none of them has experienced communism. Restricting my analyses to the aforementioned countries offers cleaner tests of my hypotheses. The observations range from 1950 to 2019. ⁵ I exclude all observations in which a party was officially split into three or more parties, ⁶ or multiple parties merging into one party, before the parliamentary election in question. These criteria yield a sample size of 591 ruling parties participating in 262 parliamentary elections.

Variable Construction

My unit of analysis is a party that has participated in a government. My outcome of interest is ruling parties’ electoral performances—that is, if conflict terminations reduce their vote shares. Thus, it is appropriate to set the change of a party’s vote shares from the previous parliamentary election as my dependent variable. Vote Share Change is the percentage-point change in a party’s vote share (lower chamber) from the previous parliamentary election. For countries with mixed member districts, I use the vote shares from party lists, which are evaluated more toward the party than votes from single-member constituencies. Parties’ vote shares are collected from the MARPOR/CMP dataset (Volkens et al. 2021), as well as official election websites if the dataset does not provide the vote share for said election.

The construction of the main explanatory variables is more complex. Since parties may have participated in multiple cabinets during the inter-election period, my variables for cabinet termination account for a party’s role in any conflictual cabinet termination during the same CIEP. The PAGED dataset provides information on whether or not each government-ending conflict is within one party or between ruling parties. ⁷ It also indicates whether a party that participated in the terminated government is responsible for the government ending-conflict. ⁸ I code a termination as preceded by conflict if the PAGED dataset lists the termination as either driven by intra- or inter-party conflicts within the cabinet. ⁹

I first distinguish between terminations that stemmed from intra-cabinet policy conflicts versus other types of intra-cabinet conflicts. For government-ending interparty conflicts, the PAGED dataset reveals if they are based on policy. ¹⁰ However, it does not contain information on whether a government-ending intra-party conflict stems from policy discord. For these eight cases of the latter category, I first determine if the PAGED dataset lists the terminations as preceded by the following three non-policy categories—“popular opinion shock,” “non-parliamentary election result,” and “personal event.” ¹¹ Next, I use Keesings Record of World Events to determine if the intra-party conflicts are based on concrete policy disagreements. ¹² In my observations, there are no cases in which an intra-party conflict termination results from policy disagreement and also fits the above three non-policy categories.

Policy Termination is equal to 1 if: (1) during the same CIEP, a party participated in a cabinet that terminated due to conflict within or between cabinet parties, as coded by the PAGED dataset; (2) for inter-party conflict terminations, the PAGED dataset categorizes the government-ending conflict as policy-based; (3) for intra-party conflict terminations, the Keesings Record of World Events describes the government-ending conflicts in terms of concrete policy disagreements; and (4) the PAGED dataset indicates the party as being responsible for the conflict. The variable is 0 otherwise. This variable captures a party’s involvement in any policy termination within the same CIEP and its responsibility for it, even if the termination resulted in a non-caretaker replacement cabinet.

All other conflict terminations are listed as non-policy conflict terminations. As with Policy Termination, a positive value for Non-Policy Conflict Termination indicates responsibility for the conflict. In my sample size, there are no observations in which a party has experienced both policy and non-policy conflict terminations in the same CIEP.

I also disaggregate Policy Termination by separating terminations that result from (1) inter-party versus (2) intra-party policy conflicts, ¹³ and (3) policy conflicts that span both within and between ruling parties. ¹⁴ I then construct three binary variables that pertain to each of these categories. The variables are equal to 1 if the criteria for Policy Termination are satisfied, and the policy conflicts that immediately preceded the terminations are categorized as intra-party (Intraparty Policy Termination), inter-party (Interparty Policy Termination), or both within and between ruling parties (Intra- and Inter-Party Policy Termination). The last variable also includes observations in which the party was responsible for an intra-party policy termination and an inter-party policy termination in the same CIEP.

Next, I code a party’s prime minister’s status. Junior Partner is equal to 1 if a ruling party was part of a cabinet that has terminated due to a policy or non-policy conflict, and the party’s status within the coalition at the time of termination was a junior coalition partner. This ensures that I properly attribute the responsible part(ies)’ prime minister status at the time of the termination. Meanwhile, I account for non-responsibility for conflictual cabinet terminations to ensure that my hypotheses are only applicable to the parties responsible for the government-ending conflicts. Termination Not Responsible is equal to 1 if a party participated in a cabinet that terminated following an intra-cabinet conflict, but was not listed by the PAGED dataset as responsible for it, and 0 otherwise. In my observations, there are no cases in which a party was responsible for one conflict termination but not for another in the same CIEP.

Schleiter and Tavits’s (2016) argue that prime ministers’ parties with assembly dissolution power can call early elections to maximize their vote shares, suggesting that this power may independently affect governing parties’ electoral performances. To control for this, I employ Goplerud and Schleiter’s (2016) index of PM dissolution power, which aggregates the prime minister’s de jure powers to dissolve the parliament for each parliamentary session. It ranges from 0 to 10, with 0 representing cases in which the prime minister has no power to dissolve the parliament, and 10 as cases in which they have unilateral power to do so. I follow Goplerud and Schleiter (2016) and designate parliaments with an index score of 8 or above as having strong PM dissolution power. Strong PM Dissolution Power is equal to 1 if this condition holds, and 0 otherwise. I also employ Goplerud and Schleiter’s (2016) index of government dissolution power (Strong Government Dissolution Power), which is constructed and categorized in the same manner as PM dissolution power, but with respect to the cabinet.

I control for the prime minister’s cabinet powers. In coalitions, PM parties with wide-ranging cabinet powers may be able to draft policies that benefit them at the expense of their coalition partners. This would improve their policy visibility and consequently their electoral images. Thus, the greater the PM party’s cabinet powers, the less vote loss the party may experience. Taken from the PAGED dataset, PM Cabinet Powers is a composite of different types of powers that the prime minister enjoys (e.g., hiring ministers) and ranges from 1 to 7. Meanwhile, a government’s size and composition may influence clarity of responsibility for performance output and thus independently affect its parties’ vote shares. Minority is coded as 1 for governments that occupy less than $ 50\% $ of the seats in the parliament’s lower chamber, and 0 otherwise. Coalition is coded as 1 if the cabinet is staffed by two or more parties, and 0 otherwise.

Policy platform adjustments may independently affect ruling parties’ electoral performances (Adams and Somer-Topcu 2009). Bawn and Somer-Topcu (2012) find that due to voter discounting, governing parties can improve their electoral fortunes by advocating for more extreme positions. As such, I also control for a party’s shift along the left-right ideological dimension. ¹⁵ Change in Extremism measures the change in a party’s left-right position from the previous election, using data from the MARPOR/CMP dataset (Volkens et al. 2021). A positive value signifies that the party has moved closer to one of the extreme ends of the left-right spectrum since the previous election. A negative value indicates the reverse. Finally, voters may cast their votes based on the country’s national economic performance (Duch and Stevenson 2008). To account for this, Per Capita GDP Growth Rate represents the change in the country’s per capita GDP during the election year from the year before. ¹⁶ Appendix A of the Supplementary Material displays all variables’ summary statistics.

Model Construction

I include all parties that have participated in one or more non-caretaker cabinets in the same CIEP. There may be election- and country-specific factors that influence parties’ electoral performances. Parties’ changes in vote shares may be correlated within an election due to election-specific reasons, or correlated within a country as a result of its party system. These factors would need to be accounted for in my statistical analyses. Thus, I construct mixed effects two-level regression models (Steenbergen and Jones 2002) with election and country levels. I set the covariances of these levels to be unstructured, which allows the intercepts to vary between elections and countries. This frees me from making strong assumptions on how within-election and within-country factors affect ruling parties’ vote share changes. Due to the limited number of observations at both levels, and for the sake of cross-model fit comparability, I employ the maximum likelihood method. I also include the lagged dependent variable to take into account possible reversion to the mean. Governing parties that gained votes from the last election for idiosyncratic reasons are likely to lose votes, such that their vote shares revert to their “mean” levels (Wilson and Grofman 2022). To maximize comparability of my key variables, I restrict each set of models to contain the same observations.

Fifteen countries may be too few for a top level. However, many of my level-two variables display intra- and inter-country variations. This reduces the problem of repeated measures for the top level. In all models, the intra-class correlation is either close to 0 or 0. Thus, I also construct alternative ordinary least-squares regression models with random effects around elections, and country-fixed effects for models with only prime ministers’ parties. Results from these models (Appendix B of the Supplementary Material) do not differ substantively from the multilevel models’ results. This diminishes the possibility that my findings are dependent on modeling choice. To examine if variable selection influences my results, I employ stepwise inclusion of control variables. These results, in Appendix C of the Supplementary Material, reveal that it does not.

Main Results

Tables 1 and 2 display the main results. Model 1 tests my hypothesis on non-policy conflict terminations and includes all observations. To see if the electoral effect depends on coalition status, I separate the observations by coalition (model 3) versus single-party governments (model 2). Model 4 tests my hypotheses on policy terminations by interacting junior partner status with each of the termination variables. Lastly, model 5 tests my hypothesis on inter-party policy termination by interacting junior partner status with each of the intra-party, inter-party, and intra- and inter-party-based policy termination variables. For brevity’s sake, the discussion of all control variables is located in Appendix C of the Supplementary Material. The controls behave as expected, thus lending face validity to my modeling approach.

Table 1. Statistical Results for Multilevel Models (with Election and Country Levels) on the Relationship between Conflict Termination and Changes in the Vote Shares of Ruling Parties, Main Variables Only

Note: The dependent variable is Party Vote Share Change. All control variables of these models are located in Table 2. * $ p<0.05 $ .

Table 2. Statistical Results for Multilevel Models (with Election and Country Levels) on the Relationship between Conflict Termination and Changes in the Vote Shares of Ruling Parties, Control Variables Only

Note: The dependent variable is Party Vote Share Change. * $ p<0.05 $ .

I first discuss the electoral impacts of non-policy conflict terminations. In model 1, the coefficient for Non-Policy Termination is negative and statistically significant: responsibility for a non-policy conflict termination is associated with a 2.17 percentage-point drop in a party’s vote share. The coefficient is positive but statistically insignificant in model 2, but negative and statistically significant in model 3. This suggests that non-policy conflict terminations are only electorally harmful to the responsible parties in coalitions. In models 4 and 5, the coefficient for Non-Policy Termination × Junior Partner is positive and statistically significant: responsibility for a non-policy conflict termination is electorally detrimental, but only for the responsible prime minister’s party. The marginal effects plot in Figure 1 (using results from model 4) reveals that when all other variables are held at their means/modes, it lowers the party’s vote share by 5.8 percentage points, but does not alter those of junior partners.

Figure 1. Marginal Effects Plot of Non-Policy Conflict Termination on the Responsible Coalition Parties’ Vote Share Changes, by Prime Minister’s Party Status

Note: The plot is based on results from model 4 of Tables 1 and 2. The horizontal bars indicate the $ 95\% $ confidence interval.

These results may need to be taken with a grain of salt. Since there are much fewer non-policy conflict terminations than policy terminations, the results may suffer from statistical efficiency. Assuming otherwise, one possible reason for these findings is that the electoral impact of non-policy conflict termination is moderated by clarity of responsibility for the government’s overall performance output. Parties in single-party governments may be more harshly evaluated than those in coalitions, as they are solely responsible for the government’s performance output (Hjermitslev 2020). As such, the additional vote loss from a non-policy conflict termination may be negligible when compared to the general electoral punishment that these parties already face. Another reason may be due to voters attributing more responsibility on the prime minister’s party for the government’s performance output (Fisher and Hobolt 2010). Thus, even if a junior partner is responsible for the government-ending non-policy conflict (e.g., poor response in disaster relief by a ministry), voters may still ultimately blame the prime minister’s party for it.

Meanwhile, the coefficient for Policy Termination is statistically insignificant in models 1–3. However, in model 4, the variable’s coefficient is small and statistically insignificant, whereas the coefficient for its interaction with Junior Partner is positive and statistically significant. The marginal effects plot in Figure 2 reveals that when all other variables are held at their means/modes, responsibility for policy terminations does not affect the vote shares of prime ministers’ parties, but is associated with a 3.58 percentage-point increase in junior partners’ vote shares. Considering Narud and Valen’s (2008) finding of an average 2.5 percentage-point vote loss among coalition parties, and that junior coalition parties on average lose more votes than prime ministers’ parties (Hjermitslev 2020; Klüver and Spoon 2020), this marginal effect can be considered large. These patterns support H3a and H2: policy terminations boost the vote shares of junior partners and do not reduce those of prime ministers’ parties.

Figure 2. Marginal Effects Plot of Policy Termination on the Responsible Coalition Parties’ Vote Share Changes, by Prime Minister’s Party Status

Note: The plot is based on results from model 4 of Tables 1 and 2. The horizontal bars indicate the $ 95\% $ confidence interval.

I now unpack policy terminations by the intra- versus inter-party nature of the policy conflicts. In model 5, the coefficients for Intraparty Policy Termination and Intra- Inter-Party Policy Termination, as well as their interactions with Junior Partner, are all positive but statistically insignificant. That is, being responsible for an intra-party conflict termination does not affect prime ministers’ or junior partners’ vote shares. Although intra-party policy conflicts muddle parties’ own policy profiles, these parties would also be perceived as valuing policy over office, resulting in a net null electoral effect. In contrast, the coefficient for Interparty Policy Termination is positive and statistically insignificant, while the coefficient for its interaction with Junior Partner is positive and statistically significant. The marginal effects plot in Figure 3 illustrates that being responsible for a government-ending interparty policy conflict boosts junior partners’ vote shares by an average of 3.77 percentage points, whereas the responsible prime ministers’ parties do not gain votes from it. This supports H3b and bolsters my argument that assuming responsibility, policy terminations exert null electoral impact on prime ministers’ parties and positively impacts junior partners’ vote shares.

Figure 3. Marginal Effects Plot of Interparty Policy Termination on the Responsible Coalition Parties’ Vote Share Changes, by Prime Minister’s Party Status

Note: The plot is based on results from model 5 of Tables 1 and 2. The horizontal bars indicate the $ 95\% $ confidence interval.

It is noteworthy that the coefficient for Termination Not Responsible is positive and statistically significant in all relevant models. Compared to parties whose coalitions served their full parliamentary terms, those that participated in coalitions that terminated due to conflicts, but were cleared of responsibility for them, on average experience a vote share gain of 3.58 percentage points. The coefficient for Junior Partner is negative and statistically significant in models 4 and 5: accounting for junior partner status in conflict terminations, the average vote share for junior partners is 2.41 percentage points less than that of the prime minister’s party. This suggests that voters are able to pinpoint the parties responsible for government-ending conflicts. They also illuminate the corrective effect that policy terminations can have on junior partners.

Potential Endogeneity

My hypotheses and findings may suffer from a potential endogeneity problem. Namely, the null electoral effect of policy terminations may stem from strategic dissolution by the prime minister’s party, guided by a favorable electoral environment. Voters may be inclined to punish a prime minister’s party for being responsible for the government-ending policy conflict. However, if the party has the power to dissolve the parliament, it may choose to replace the cabinet if public opinion trends against the party, thus escaping potential voter punishment, or call early election if the reverse is true to reap electoral reward. It may even engineer a policy conflict as a pretext for calling early election. In other words, it may not be the prime minister’s party’s policy steadfastness that makes a policy termination electorally benign; instead, the null effect may be attributed to the party’s ability to terminate government at an electorally opportune time.

I examine this potential endogeneity problem through the lens of the prime minister’s assembly dissolution power (see Schleiter and Tavits 2016). I offer two hypotheses that would hold empirically if my findings were endogenous to the prime minister’s assembly dissolution power. For conciseness’s sake, the theoretical discussion behind these hypotheses, statistical testing, and detailed analyses of the results are in Appendix D of the Supplementary Material. The results reveal that even if the prime minister’s party cannot call early elections, policy termination does not harm its vote share. In addition, the electoral impacts of policy and non-policy conflict terminations for junior partners do not depend on the government or the prime minister’s ability to call early elections. These results assuage the endogeneity concern.

Robustness Tests

Multiple Conflict Terminations

In constructing my main explanatory variables, I assumed that the electoral impact of being responsible for one conflict termination is equivalent with the impact of responsibility for multiple conflict terminations in the same CIEP. However, responsibility for multiple conflict terminations may alter voters’ perceptions of a party’s governing competence and policy stance. Voters may forgive ruling parties if they are responsible for only one non-policy conflict termination between two parliamentary elections, but may be more inclined to punish those that are responsible for more than one. In the same vein, voters may not pay attention to one policy termination, but may notice a ruling party’s policy steadfastness if it is responsible for more than one government-ending policy conflicts in the same CIEP.

I conduct additional statistical tests to investigate the potential impact of multiple conflict terminations in the same CIEP. Details on variable and model constructions, as well as the results and discussion of them, are in Online Appendix E (in the American Political Science Review Dataverse; see So 2023). The results reveal that, as expected, multiple conflict terminations do not moderate the electoral impacts of policy termination on junior partners. Also uncontroversially, responsibility over multiple non-policy conflict terminations exacerbates their negative electoral effect on prime ministers’ parties. The divergent impacts of multiple policy terminations on prime ministers’ parties are noteworthy. Responsibility over multiple inter-party policy terminations are electorally harmful, but responsibility over multiple intra-party policy terminations is not. The latter may showcase the extent of prime ministers’ parties’ policy steadfastness and loyalty to their principled policy positions. The former may instead reveal a lack of governing competence, as the party is repeatedly unable to resolve policy conflicts with other cabinet parties. Coupled with the null effect on junior partners, this would suggest that while voters value prime ministers’ parties’ policy steadfastness, multiple terminations reveal these parties’ failures as effective managers of governing parties and may, therefore, damage their perceived governing competence.

Timing of Termination

The relationship between conflictual cabinet terminations and coalition parties’ electoral performances may be moderated by the timing of terminations. Duvall and Pétry (2019) show that ruling parties’ ability to fulfill their electoral pledges drops dramatically after the first half of their mandate. This would suggest that terminating a government after this period may be less electorally costly, as doing so would relieve ruling parties from the difficulties in fulfilling their electoral pledges. Yet, Stiers, Dassonneville, and Lewis-Beck (2020) show that rather than ignoring the government’s performance early in the election cycle, voters consider them in their evaluations of incumbent parties. Following this logic, voters may also consider cabinet terminations that occur early in the election cycle in their voting calculus. This would imply that the electoral impacts of conflictual cabinet terminations do not depend on their timing. Indeed, an empirical investigation, detailed in Online Appendix F (So 2023), reveals that the timing of conflict terminations does not moderate the electoral impacts of conflict terminations. This lends further credibility to my hypotheses.

Post-Termination Opposition Status

The article’s main results may be an artifact of out of government effects. A party may land in opposition after a policy termination. This opens the possibility that the change in government status, rather than the clarifying effect on policy positions, explains the lack of voter punishment from policy termination. Grynaviski (2010) finds when voters are making their voting decisions, they rely on opposition parties’ rhetoric during electoral campaigns and on ruling parties’ record in office. As such, for a ruling party that exited government, voters may discount the party’s stint in office and the government-ending policy conflict, and instead focus on its policy promises. Results from statistical testing (in Online Appendix G [So 2023]) refute this: a party’s post-termination status does not moderate the electoral consequence of policy termination, neither for prime ministers’ parties nor for their governing partners.

Mainstream versus Niche Parties

Mainstream parties are the top contenders for government. Yet, there are instances where single-issue parties join the government as junior partners (e.g., the Green Party in Germany and the radical right-wing Freedom Party in Austria). This may have consequences on the likelihood of policy termination and its electoral consequences. Greene (2017) finds that the more diverse the issues that coalition parties addressed in their electoral platforms, the more room there is for negotiation during policy conflicts, and consequently the more durable a coalition becomes. Thus, coalitions in which one party is a single-issue party may be at greater risk of breakdown. Meanwhile, voters’ conflation of coalition parties’ policy positions may exert the most electoral harm on single-issue parties, as their policy profiles rests on one issue. Since projecting clear policy positions is likely more important for these parties than for mainstream ones, the clarifying effect of a policy termination may be most electorally beneficial for single-issue parties. Both would suggest that the positive electoral impacts of policy termination on junior partners are driven by single-issue parties.

I examine this possibility via statistical analyses. Since there are very few coalitions in which the junior partners are fielded by single-issue parties, I run same models as those in Tables 1 and 2, but exclude single-issue parties. The results (in Online Appendix H [So 2023]) do not differ substantively than those in Tables 1 and 2. This suggests that my findings are not driven by the presence of single-issue parties.

Intra- versus Inter-Party Conflicts

The electoral impacts of conflictual cabinet terminations may depend their intra- versus inter-party nature. Conflicts within a party signal governing incompetence, as voters perceive the party as internally divided; this is electorally harmful (Greene and Haber 2015; So 2021). Intra-party conflicts also signal high clarity of responsibility: voters can easily identify the party responsible for the conflict. Based on these considerations, one may argue that the electoral impacts of conflict terminations should differ for intra- versus inter-party conflicts. Specifically, cabinet terminations preceded by intra-party conflicts may be electorally damaging to the responsible parties, but terminations preceded by inter-party conflicts may not affect these parties’ vote shares. However, results from statistical testing (in Online Appendix I [So 2023]) suggest otherwise: neither responsibility for intra-party nor inter-party conflict termination reduces the responsible parties’ vote shares. These results assuage the concern that the article’s main findings are masking the intra- versus inter-party nature of government-ending conflicts.

PM Cabinet Powers

Finally, one may contend that my results on policy termination are endogenous to the prime minister’s breadth of power within the cabinet, that is, PM cabinet powers. Prime ministers with sweeping powers can dictate which policy areas receive attention and, therefore, prevent policy conflicts and policy terminations. As such, the electoral impacts of policy termination may be an artifact of low PM cabinet powers. It may be that voters punish prime ministers’ parties for government-ending policy conflicts, but less so if they possess few cabinet powers, as they are in less control of the cabinet’s policy agenda than those with wide-ranging cabinet powers.

Results from statistical testing (Online Appendix J [So 2023]) do not support this argument. PM cabinet powers do not moderate the electoral impact of a policy termination on the junior coalition partner(s) responsible for the conflict. Although greater cabinet powers lessen prime ministers’ parties’ electoral gains from policy termination, on average even prime ministers’ parties with the most cabinet powers do not lose votes for being responsible for a policy termination. That is, voters do not punish prime ministers’ parties for policy terminations, even if they wield great cabinet powers. These findings further support my hypotheses on policy terminations.

DISCUSSION

Do voters punish parties that are responsible for conflictual cabinet terminations? I argued that the answer depends on the nature of the conflicts. Specifically, non-policy conflict terminations should be electorally detrimental to the responsible parties, as they damage these parties’ governing competence. In contrast, policy terminations clarify the responsible parties’ policy profiles and signal their unwillingness to compromise them. Thus, this type of cabinet terminations should not reduce the vote shares of the responsible parties, and in the case of inter-party policy terminations, should be electorally beneficial for junior partners.

Statistical analyses using cabinet termination data from the ERDDA’s PAGED dataset support my argument on policy terminations. The findings are not endogenous to the strategic nature of cabinet terminations, at least not as they trigger early elections. The results also reveal that prime ministers’ parties of single-party governments and junior coalition partners do not lose votes from non-policy conflict terminations. Particularly for the latter, one reason may be that voters map junior partners’ responsibilities for non-policy conflicts onto the prime minister’s party. This would have big implications on electoral accountability. Although participation in coalitions may be electorally costly, junior coalition partners may be insulated from deteriorating governing competence, at least relative to the prime minister’s party. Further research on how voters perceive the governing competence of junior partners would no doubt enrich scholars’ understanding of electoral accountability in coalition governments.

This article serves as one of the first steps toward understanding how conflictual cabinet terminations affect parties’ electoral health. As such, it is important to provide an overall picture of how these terminations affect ruling parties’ vote shares and investigate the external validity of my hypotheses. For this purpose, the testing of my hypotheses has been large-N and observational in nature. Although the statistically significant correlations are by no means absolute proof of my hypotheses’ internal validity, results from additional statistical tests regarding endogeneity concerns and robustness tests raise the plausibility of the theoretical mechanisms. Nevertheless, one can extend the above findings by testing the causal mechanisms laid out in this article. One possibility would be to conduct a survey experiment. ¹⁷ Another would be to conduct four detailed case studies of real-world conflictual cabinet terminations, varying on non-policy and policy terminations and on single-party versus coalition governments.

This article’s findings generate several implications on multiple subfields within political science, particularly on voters’ political sophistication, governing competence, and coalition governance. First, based on voter conflation of coalition parties’ policy positions, one may expect voters to be too politically naive to understand the circumstances behind cabinet terminations and assign responsibility for them. My findings of divergent electoral impacts of conflict terminations suggest that they possess the political sophistication to do so, thereby bolstering the findings of Spoon and Klüver (2017). My results also suggest that voters pay attention to political events and can process them in a systematic manner. Thus, they are complementary to Fortunato and Adams (2015): while voters may lack the expertise to disentangle coalition parties’ policy positions, they do understand why cabinets end and incorporate these reasons into their voting calculus. This, in turn, hints at the need for further research on how voters react to media reporting of events within governments, and how and when these reports affect vote choice.

Second, in light of Dewan and Dowding’s (2005) finding that firing ministers responsible for scandals corrects the government’s electoral image, my results raise questions on when and why a non-policy conflict would results in government breakdown. Since non-policy conflict terminations lower the vote shares of prime ministers’ parties in coalitions, one may wonder why some governments end up dissolving instead of only firing the responsible ministers. It may be that some non-policy conflicts are too severe, such that ministerial replacement is insufficient for improving the government’s perceived competence. This hints at the possibility that the types of non-policy conflicts, for example, sexual assault allegations against one minister versus ineffective handling of a natural disaster, may shape the conflicts’ trajectories. And, media’s heightened attention to these conflicts may exacerbate them, to the point where cabinet termination becomes necessary. Or, akin to Green and Jennings’s (2017) finding that perceived incompetence in handling the economy evolves into general incompetence, mishandling of a natural disaster or scandal within one ministry may evolve into perceived mishandling of all issues and mistrust in all ministries, thus making government breakdown unavoidable. More scholarly attention on voters’ reactions to non-policy conflicts within governments would certainly improve our understanding of how political events affect governing competence and electoral accountability of ruling parties.

Finally, this article’s findings illuminate how government-ending policy conflicts can improve the electoral health of the parties responsible, particularly for junior partners. This may seem counterintuitive given previous research on coalition bargaining, which may suggest that policy terminations erode ruling parties’ governing competence. Conflicts within a coalition, when coupled with poor government performance (i.e., poor economic evaluations), can result in voter punishment for all coalition parties (Plescia and Kritzinger 2022). Also, since coalition parties utilize institutional mechanisms to prevent policy dilution ¹⁸ and enforce party discipline (Ceron 2016; Laver 1999), policy terminations may reveal the responsible parties’ lack of competence for resolving policy differences and ensuring collective cabinet responsibility. The findings of this article cast a different light. With voters’ tendencies to conflate coalition parties’ policy positions, which are clarified by policy conflicts (Spoon and Klüver 2017), as the departure point, the conclusion is that serious, government-ending policy conflicts have benign electoral outcomes for the responsible parties.

This is consequential for coalition governance. Research on junior coalition partners reveals that these partners are penalized for coalition participation. ¹⁹ This makes one wonder why they participate in coalitions. One perspective pertains to the trade-off between office and policy goals: the desire for office motivates parties to enter into coalitions as junior partners, despite not being able to control the government’s policy agenda fully and risking vote loss (Müller and Strøm 2000). The results of this article offer a refinement: this trade-off is not necessarily ironclad, particularly if junior partners have the option to withdraw from government to regain their grip on their own policy profiles.

Relatedly, rather than attempting to reach a compromise within or between ruling parties, which would further dilute a ruling party’s policy profile, governing parties may be electorally better off if they start anew. This suggests that ruling parties are incentivized to terminate governments for the sake of maintaining their policy positions. This, in turn, provides contexts for cabinet parties’ utility functions generated by the vast literature on strategic cabinet termination and competing risks, ²⁰ particularly for termination with replacement, termination with early election, and non-termination. Specifically, the clarifying effect of policy terminations may tip the favor toward termination. Although outside the scope of this article, my results raise the question of how voters’ perceptions of junior partners’ policy positions influence the risk of cabinet replacement versus early election. Further research on this will surely contribute to our understanding of when and why coalitions break down and of coalition governance in general.

Last but not least, the findings on policy terminations also have implications for policy representation and voters’ ability to hold their governments accountable for effective policymaking. Sticking to their policy profiles means that ruling parties can faithfully represent their supporters, which raises the credibility of their election pledges. Thus, the lack of voter punishment for policy termination can improve the quality of party representation in parliamentary democracies. At the same time, cabinet termination is not without cost to governance. The time spent on forming a replacement cabinet, or on campaigning in the case of early elections, reduces a government’s efficiency and effectiveness in the policymaking process. More examinations into how intra-cabinet policy conflicts affect the quality of governance would advance the literature on the electoral accountability of parliamentary governance.