Introduction

Wim Kok was an electoral disappointment. In the first two elections under his leadership, the vote share of the Dutch Labour Party (PvdA) shrunk by almost 10 percentage points. Why, then, did Kok stay on as party leader for more than 15 years, becoming one of the most iconic post‐war politicians in Dutch post‐war history?Footnote ¹ The answer is that he brought his party into government, first as the junior coalition partner (1989–1994), and later as the leading cabinet party (1994–1998). Only in 1998, 9 years into his leadership stint, did Kok secure his first electoral victory. This anecdote illustrates that a party may not punish its leader for electoral losses if he or she can satisfy the party's ambitions for office at the same time. This paper will show that the case of Wim Kok was no exception: vote gains or losses have no effect on party leader turnover when at the same time parties are promoted to the government or relegated to the opposition. Reversely, electoral outcomes only influence the survival of party leaders if the party's office pay‐offs remain unchanged (i.e., the party stays in opposition or in government).

In this paper, we focus on how the performance of parties in the electoral and governmental arenas affects leader survival. We identify two shortcomings in the existing literature: First, how do parties respond if they achieve one goal but not the other (e.g., losing votes while winning office)? To answer this question, we will explore the interaction between the two. Second, the literature assumes that opposition party leaders are permanently more at risk of deselection than leaders of government parties. We will argue that the effect of pay‐offs works relative to a party's current endowments. Therefore, changes in office status (i.e., moving from government to opposition or vice versa), rather than office status per se, should be the more important determinant of leadership change. Starting with a review of the party goals literature, we will develop these two ideas more in the next sections. Finally, we will use data on over 500 party leaders in 14 parliamentary democracies between 1965 and 2012 (Pilet & Cross, Reference Pilet and Cross2014) to test our hypotheses. We find that electoral performance only matters for party leader survival when parties do not experience changes in their office status at the same time.

Policy, office and votes as drivers of party behaviour

Parties are motivated – to varying degrees – by the goals of office, policy and votes (Harmel & Janda, Reference Harmel and Janda1994; Müller & Strøm, Reference Müller and Strøm1999; Strøm, Reference Strøm1990). The pursuit of office typically entails taking control of (parts of) the executive branch. An office‐oriented party will try to maximise its share of high‐ranking executive offices, such as ministerial posts. One important feature of office pay‐offs is that they are private goods and therefore restricted to a small number of party elites (unless we also consider more abundant patronage jobs; see Ennser‐Jedenastik, Reference Ennser‐Jedenastik2014). Since the direct benefits of office accrue only to a selected few, the desire for office is shared unevenly across the party hierarchy (Strøm, Reference Strøm1990: 576–577), with members and activists typically less willing than party leaders to sacrifice on policy for the sake of office.

Policy‐oriented parties will behave differently from their purely office‐seeking competitors, in that they are willing to forgo (some) office gains in exchange for achieving their policy goals. Since policy pay‐offs are typically understood as public goods (with the exception of pork barrel legislation; see Shepsle & Weingast, Reference Shepsle and Weingast1981), they accrue to all members of a party (and beyond) – although the utility gains may vary with individuals’ ideal points (May, Reference May1973). Policy‐seeking may be understood as setting government policy as close as possible to one's ideal point, or as only echoing policy preferences (i.e., policy purity).

Vote‐seeking implies that parties try to maximise their support at the ballot box. As Harmel and Janda (Reference Harmel and Janda1994: 270) point out, it is impossible to distinguish vote‐seeking from office‐seeking in pure two‐party systems. Yet in multiparty systems, winning elections and controlling the government are much less closely related. Therefore, parties will often face trade‐offs between the maximisation of office benefits and the pursuit of votes.

Party leaders risk removal if they fail to achieve these goals. Previous research has found that electoral defeat (i.e., losing votes in elections) and being stuck in opposition independently increase the probability of leader replacement, especially for female party leaders (Andrews & Jackman, Reference Andrews and Jackman2008; Ennser‐Jedenastik & Müller, Reference Ennser‐Jedenastik and Müller2015; O'Brien, Reference O'Brien2015) and in Westminster electoral systems (Bynander & ’t Hart, Reference Bynander and ’t Hart2007; So, Reference So2018). Actual performance thus clearly matters, yet expected performance is also important: leaders who succeed a high‐performing predecessor risk early departure, because expectations about party performance may be unrealistically high (Horiuchi et al., Reference Horiuchi, Laing and ’t Hart2013). Finally intra‐party institutions also affect leaders, in that individuals chosen by more inclusive methods face greater risks of deselection (Ennser‐Jedenastik & Müller, Reference Ennser‐Jedenastik and Müller2015; Ennser‐Jedenastik & Schumacher, Reference Ennser‐Jedenastik, Schumacher, Cross and Pilet2015). Similarly, intra‐party conflict over leader succession also stymies leader survival (Bynander & t'Hart, Reference Bynander and t'Hart2006).

There are two ways to think about the effects of office achievement and electoral performance. One assumes that party behaviour responds to absolute utility levels. A higher vote share implies a higher degree of goal achievement and should therefore make leader replacement less likely. The larger a party's vote share, the less likely it is to remove its leader. Similarly, being in office should reduce the odds of replacement, whereas being in opposition puts leaders at greater risk. This approach is the one taken by Andrews and Jackman (Reference Andrews and Jackman2008) and, with regard to office, by Ennser‐Jedenastik and Müller (Reference Ennser‐Jedenastik and Müller2015).

The approach we prefer is to assume that parties respond to changes in their utility, thus judging their leaders’ performance relative to the (most recent) status quo (Mercer, Reference Mercer2005: 4–5).Footnote ² A small party that adds votes should therefore be less likely to replace its leader than a large party that loses votes. In other words, vote share changes, not absolute vote shares, affect party behaviour. Similarly, rather than assuming static differences between government and opposition parties, we conjecture that entering and exiting government office will trigger behavioural responses from parties. Our first two hypotheses capture these predictions:

• H1: Party leaders’ risk of replacement is highest after their party exits government, lowest after it enters government, and intermediate when there is no change in a party's office status (in government or opposition).

• H2: Party leaders are more (less) likely to be replaced after their party's vote share decreases (increases).

The trade‐offs between party goals: Office‐seeking versus vote‐seeking

A substantial part of the literature on party behaviour focuses on the necessary trade‐offs between policy, office, and votes (Strøm, Reference Strøm1994; Strøm & Leipart, Reference Strøm and Leipart1993) and the conflict over prioritising the three goals inside party organisations (Bawn et al., Reference Bawn, Cohen, Karol, Masket, Noel and Zaller2012; May, Reference May1973; Norris, Reference Norris1995).

Sometimes the goals of policy, office, and votes are not in conflict with one another. For example, electoral success increases a party's chance to attain its office and policy goals – although the reverse may not necessarily be true. There are even cases when party leaders were able, for a while at least, to jointly maximise all three goals (Poguntke, Reference Poguntke, Müller and Strøm1999).

Typically, however, parties do face sharp trade‐offs. The risk of electoral defeat after entering executive office has increased markedly over the past decades (Narud & Valen, Reference Narud, Valen, Strøm, Müller and Bergman2008; Van Spanje, Reference Van Spanje2011). In candidate‐centred systems, the trade‐off between office and votes is made even more acute by the fact that ministers receive an electoral premium and thus do not bear the costs of governing (Martin, Reference Martin2016). Curiously, in such situations, the costs of governing fall disproportionately on those people excluded from executive office. Policy goals are in conflict with vote and office goals, if the policy is unpopular (vote), and if potential coalition partners refuse to form a government because of the party's policy stance (office).

While the trade‐offs between seeking votes, office and policy have thus been amply documented, there is very little research to date (aside from the studies of supporting parties in minority governments)Footnote ³ that analyses which goal is prioritised. The literature on party leader survival has hitherto neither theorised nor empirically analysed trade‐offs between party goals. The contribution of this paper is therefore twofold: First, to document that electoral success (winning or gaining votes in elections) and office achievement (entering or exiting government) are distinct dimensions of party performance with independent effects. Second, to theorise and analyse for the first time the trade‐off between electoral performance and office achievement and their joint impact on party leader survival.

Intrinsic versus instrumental goals

To theorise about how parties prioritise their goals, we need to distinguish between intrinsic and instrumental goals. Intrinsic goals are those that are valued in and of themselves. While intrinsic goals are ends, instrumental goals are means to other ends. Their value lies in helping a party achieve its other goals. As alluded to above, all three of the objectives that parties pursue may serve instrumental purposes. Electoral success helps parties achieve office and policy goals. Holding office is often a precondition for implementing one's preferred policies (and may, under certain circumstances, even be electorally beneficial; see Martin, Reference Martin2016), and policies are sometimes pursued for electoral reasons (and thus office achievement), not because politicians deeply care about them.

However, only policy and office are intrinsic motivations. Achieving certain policy outcomes is one major reason why people join parties and then decide to run for office in the first place (Bruter & Harrison, Reference Bruter and Harrison2009). The benefits of office are, by definition, private goods (a salary, a car with a driver) and are thus clearly of intrinsic value – as may be the social prestige and public attention that come from taking ministerial office.

Votes, by contrast, are an instrumental goal. As Strøm and Müller (Reference Strøm, Müller, Müller and Strøm1999: 10–11) argue, it is implausible to assume that politicians seek votes for ‘the pure thrill of winning’. Rather, votes are valuable because they translate into better chances of obtaining executive office and implementing one's policy agenda. After achieving these goals, there is no utility in winning even more votes (Riker, Reference Riker1962: 33). Electoral success is thus never an end in itself, but always a means to other ends.

Recognising that parties often face sharp trade‐offs between office‐seeking and vote‐seeking, the crucial question is how they will behave when electoral and office ambitions are in conflict. Only by examining the interaction between electoral performance and office achievement on party leader survival can we deduce anything about the hierarchy of party goals.

In terms of expectations, the pure instrumentality of electoral success logically implies that votes are subordinate to the goals of office and policy. After all, ends are more important than means. Goal achievement in the office domain should therefore have a larger influence on party behaviour. We thus expect the office motivation to dominate when vote‐seeking and office‐seeking motivations have opposite implications for party leader survival. Therefore, we conjecture that electoral performance has no effect on the survival of the leadership, if at the same time there is a change in office status. In other words, if a party enters (or leaves) government, the party leader's political survival should not be affected by the party's electoral performance (H3a). If there is no change in office status, then electoral performance does affect leadership survival. That is, if parties stay in opposition or stay in government, electoral losses (gains) have a negative (positive) impact on survival (H3b). Also, we expect this effect of electoral performance to be weaker when office status changes, than when office status is stable (H3c). In general, we thus expect that the impact of electoral performance is conditional on office achievement.

• H3a: There is no effect of electoral performance if parties change their office status.

• H3b: Electoral performance affects the risk of leadership termination if parties keep their office status.

• H3c: The effect of electoral performance on leader survival is smaller if parties change their office status compared to when they keep their office status.

Policy goals also conflict with vote and office goals, as already discussed. In contrast to vote and office goals, it is difficult to identify when a politicians meets a policy goal. First of all, we do not know whether politicians support policies intrinsically, or for some vote or office gain. Therefore, if policies are implemented we cannot be sure whether a politician achieved his or her policy goals. Second, when a policy is implemented it is often hard to tell whether it fully matches with the stated policy preference of the politician. In sum, policy goals cannot be evaluated in a good way, and therefore our paper focuses strictly on office and vote goals.

Empirical strategy

The data to test our hypotheses come from the COSPALFootnote ⁴ project – a collaboration of country experts that produced detailed data on party leaders in 14 parliamentary democracies between 1965 and 2012 (Pilet & Cross, Reference Pilet and Cross2014). The countries included in the data are Australia, Austria, Belgium, Canada, Denmark, Germany, Hungary, Israel, Italy, Norway, Portugal, Romania, Spain, and the United Kingdom. The data cover over 500 leaders of about 100 parties. To be sure, leadership duration varies enormously, from a handful of leaders who last only a few months, to the most durable ones who spend most of their political life at the helm of their party. The most long‐lasting party leaders in the data set are Helmut Kohl, leader of the German Christian Democratic Union (more than 25 years), Pia Kjærsgaard, leader of the Danish Progress Party and later the Danish People's Party (more than 27 years), and Carl Ivar Hagen, leader of the Norwegian Progress Party (more than 29 years). These outliers are even more notable considering that the median party leader stays in office for just under 4.5 years.

We split the data set into monthly spells to allow for the inclusion of time‐varying covariates. Observations are right‐censored in case the leader leaves office due to death, illness, term or age limits, or party splits that result in the leader leaving the party (e.g., Pia Kjærsgaard who in 1995 abandoned the Danish Progress Party to establish the Danish People's Party or Ariel Sharon who broke away from Likud in 2005 to set up Kadima). Of course, leaders who were still in office on 31 December 2012 are also right‐censored.

We code a four‐category variable to operationalise the office‐seeking motivation. It captures at every point in time how a party's status has changed (or not) during the past 12 months (or shorter, if the current leader has been at the helm for a shorter period of time): (1) party leaves government, (2) party remains in opposition, (3) party remains in government and (4) party enters government.

The electoral performance predictor is coded in a similar fashion (i.e., using a window of up to 12 monthsFootnote ⁵ for the incumbent leader): (1) no election taking place during the past 12 months, (2) vote share losses of one percentage point or greater, (3) vote share changes smaller than plus or minus one percentage point and (4) vote share gains of one percentage point or greater.Footnote ⁶ While this may not be the most obvious operationalisation for electoral performance, there are two reasons for this choice: First, there are theoretical reasons to include wins and losses as separate predictors. As prospect theory argues, human behaviour often responds more strongly to losses than to gains (Tversky & Kahneman, Reference Tversky and Kahneman1991) – a phenomenon (‘loss aversion’) that is also relevant in the political arena (Jervis, Reference Jervis1992; Patty, Reference Patty2006).Footnote ⁷ Second, since elections take place only once in a few years, there are many monthly spells with zeros in the direct measure of vote share change (about 75 per cent). As a consequence, the kurtosis of this variable is around 35, multiple times above what is typically considered an acceptable range to estimate an effect for a continuous measure (for instance, an upper bound of 7 has been proposed by West et al., Reference West, Finch, Curran and Hoyle1995). Using categories instead of the continuous measure will thus also eliminate the risk that the effect of electoral performance is determined by a small set of outlier data points.

In terms of control variables, we first enter a set of party characteristics into the analysis. Chief among them is party size (measured as the parliamentary seat share). We also include predictors for the removal body, that is, which group within the party has the power to remove the leader (the party leader him‐ or herself,Footnote ⁸ the party executive, the party congress, the parliamentary caucus or party members). The assumption is that more inclusive deselectorates make it more difficult for leaders to keep their coalition of supporters from defecting to a challenger (Bueno De Mesquita et al., Reference Bueno De Mesquita, Morrow, Siverson and Smith2002, Reference Bueno de Mesquita, Smith, Siverson and Morrow2003). Furthermore, the analysis controls for variation across party families (Radical Left, Green, Social Democrat, Liberal, Conservative/Christian Democrat, Radical Right and Regionalist). In line with Andrews and Jackman (Reference Andrews and Jackman2008), we capture leaders’ grace period (the time before they fight their first election). At the individual level, the analysis includes predictors for a party leader's gender and experience in national political office (as an MP or minister). In addition, the regressions specify shared frailties at the country level.Footnote ⁹ Table 1 presents descriptive statistics for the independent variables.

Table 1.

Descriptive statistics of independent variables

Note: ‘12m’ indicates that the variable captures whether the respective event occurred during the past 12 months.

Table 2 gives us a first impression of the probabilities of leadership turnover depending on electoral performance and office achievement. It reports the proportion of monthly spells (hence the low percentages) that record leadership changes. This excludes all right‐censored leaders, thus bringing the overall number of ‘failures’ from 518 leaders down to 361.

Table 2.

Percentages of monthly spells with leader replacement

Note: Figures in parentheses indicate the number of leadership changes (excluding censored observations, therefore N = 361).

A glance at the marginal distributions suggests that losing office and losing votes both affect party leader turnover. The highest turnover percentage (4.6) is reported for instances when parties lose both, office and votes, the second lowest (0.5%) for cases when parties enter government and win votes.

Analysis

The dependent variable in this analysis is time in office, requiring the specification of event history models (Box‐Steffensmeier & Jones, Reference Box‐Steffensmeier and Jones2004). Over the past decades, the preferred choice in many social science applications has become the semi‐parametric Cox proportional hazards model (Cox, Reference Cox1972). Its main advantage is that it requires no assumptions about the distribution of duration times. The Cox regression model estimates the hazard rate for each individual i as a function of the unspecified baseline hazard h₀(t), a set of covariates x and a vector of regression coefficients β.

$h_i\left(t\right)=h_0\left(t\right)e^{x_i{\beta }_x}$

Table 3 presents the results from two regression models, the first testing only the direct impact of electoral and office performance, the second adding the interaction terms. Since the dependent variable is time‐to‐event and the event is the removal of a party leader, positive coefficients indicate that an increase in the independent variable raises the probability of leader replacement.

Table 3.

The impact of votes and office on party leader survival (Cox regressions)

Note: Figures are raw coefficients and leader‐clustered standard errors from Cox proportional hazard regressions with shared frailties at the country level, *p < 0.05, **p < 0.01, ***p < 0.001, #p < 0.1.

As Model I shows, electoral performance has a significant influence on party leader survival. Within a year of winning a percentage point or more (the median electoral performance in the ‘vote gain’ category is +3.4 percentage points), the probability of party leaders being removed from their position decreases strongly. The coefficient of –0.886 translates into a hazard ratio of 0.41, meaning that leaders with good electoral performance have an almost 60 per cent smaller hazard rate than those presiding over vote losses. The absence of an election has an almost equally strong effect: a 50 per cent reduction in the hazard rate.

The effects reported for the office variables are similar in size. Upon exiting the executive, party leaders’ probability of losing their position increases by 60 per cent (raw coefficient: 0.471, hazard ratio: 1.60) compared with the reference category (leaders of opposition parties with no change in office status). Yet when parties enter government, the probability of leadership turnover declines sharply. The coefficient of –0.733 translates into a hazard ratio of 0.41 and thus implies a cut in the hazard rate by more than half. Changes in office status are therefore strongly correlated with substantial changes in party leaders’ survival probabilities. By contrast, there is no discernable difference between leaders whose party continues in government (without change) and those whose party stays in opposition. The raw coefficient of –0.181 implies slightly higher survival probabilities for party leaders in government than those in opposition, but the difference is not statistically significant (p = 0.210).

The results reported in Model I thus clearly show that electoral performance and office achievement strongly affect party leader survival, thus supporting H1 and H2. More specifically, the analysis establishes that, in line with our expectations, changes not levels are the more important determinant of party leaders’ fate.

The other variables in the model yield few surprising results. Older leaders are more likely to exit than younger ones, yet there are no gender differences.Footnote ¹⁰ Also, leaders are less vulnerable during their grace period and when their parties command a majority of seats in parliament – which is the case for only 6 per cent of observations. For most parties thus, changes in vote share (especially losses) are the more relevant driver of party leader turnover, whereas absolute size plays no significant role.Footnote ¹¹

Another intriguing finding relates to the influence of removal mechanisms. When the power to install a new leader rests with the party membership, survival probabilities are lower on average. Intra‐party democracy thus hurts party leaders – a result that is consistent with the predictions of selectorate theory (Bueno De Mesquita et al., Reference Bueno De Mesquita, Morrow, Siverson and Smith2002, Reference Bueno de Mesquita, Smith, Siverson and Morrow2003) and confirms findings in earlier research (Ennser‐Jedenastik & Müller, Reference Ennser‐Jedenastik and Müller2015; Ennser‐Jedenastik & Schumacher, Reference Ennser‐Jedenastik, Schumacher, Cross and Pilet2015).Footnote ¹²

Also, there is evidence that leaders of radical right and regionalist parties stay in power longer than leaders in other parties. This result resonates with earlier findings that show radical right parties to be more leadership dominated than other parties (Schumacher & Giger, Reference Schumacher and Giger2017). Leaders of radical right parties may thus have more authority over their parties than predicted by institutional and performance variables alone – hence their longer tenures in the leadership position. A similar logic may apply to regionalist parties. To be sure, some of them – the Flemish Nieuwe Vlaamse Alliantie and the Italian Lega Nord, for example – could also have been coded as radical right parties.

Model II reports the coefficients from the interactions between office achievement and electoral performance. Yet, since interaction effects are difficult to interpret from the regression coefficients alone, Figure 1 plots the average marginal effects (AMEs) on the relative hazard, broken down by the office and electoral performance indicators.

Figure 1.

Marginal effects of electoral performance (reference category: vote loss) by (change in) office status.

Note: Average marginal effects on the relative hazard, with 95‐per cent confidence intervals. Reference category is vote loss. All calculations are based on Model II, with all other variables held at their observed values.

Our general expectation in hypotheses 3a–3c was that the impact of electoral performance should be conditional on office performance. We will first look at the effect of electoral performance in the conditions of office status change (leave government, enter government) and office status stability (stay in government, stay in opposition). After that we will analyse the difference between these two effects.

Figure 1 shows the marginal effects of electoral performance for each office status with vote loss as the reference category. Starting from the left, there are no significant effects of electoral performance when parties leave office, as all the AMEs of electoral performance above the ‘leave government’ label include zero within their confidence interval. Winning votes thus does not boost party leaders’ survival chances when their party loses its place in government at the same time. A similar picture emerges for government party leaders whose parties retain their office status. In these cases, we observe no significant differences in survival chances between cases with electoral losses, electoral gains, no election nor or little change. In sum, following H3a, electoral performance has no effect if office status changes.

The story is different for party leaders whose office status is stable. Electoral performance has a negative and significant effect for opposition party leaders who experience no change in office status (‘stay in opposition’). Specifically, winning votes and the absence of elections increases the probability of survival compared to the reference category of losing votes. An electoral performance resulting in little or no vote share change, however, does not differ in its impact from vote losses. This suggests that opposition party leaders are expected to add votes, not just to hold on to the party's prior vote share. Electoral outcomes hence clearly shape leaders’ odds of staying in their position when parties are and remain in opposition.

Similarly, electoral performance has a negative and significant effect for government party leaders who experience no change in office status (‘stay in government’). Vote losses may just end the career of the party leader despite holding on to office. These findings are in line with H3b.

Our last point concerns the difference in the effect of electoral performance between the different office conditions. The effects plotted in Figure 1 indicated whether the effect of electoral performance in one of the four office conditions is significantly different from zero.

However, we also theorised in H3c that the effect of vote gains in the ‘leave government’ and ‘enter government’ conditions would be weaker than the effect of vote gains in the ‘stay in opposition’ and ‘stay in government’ conditions. We find that the effect of vote gains (compared to losses) in the ‘enter government’ condition significantly differs from the effects in the ‘stay in opposition’ (diff = 0.523, p = 0.035) and ‘stay in government’ conditions (diff = 0.657, p = 0.056). Here, indeed, the effect of electoral performance is weaker when office status changes than when it remains stable, thus supporting H3c. However, there is no statistically significant difference in the effect of vote gains (compared to vote losses) between the ‘leave government’ condition and ‘stay in opposition’ (diff = 0.380, p = 0.472) or between the ‘leave government’ and ‘stay in government’ condition (diff = 0.246, p = 0.636). Figure 1 already reveals this: The large confidence intervals of the ‘vote gain’ effect in the ‘leave government’ condition overlap with the ‘vote gain’ effects in the stable office conditions in the middle. In sum, vote gains do not prolong the tenure of a leader in the ‘leave government’ condition, yet this effect is statistically indistinguishable from the (statistically significant) effects of vote gains for leaders in the stable office conditions. This means we have to reject H3c for parties that leave government, while we find it confirmed for parties that enter government.

Taken together, the evidence presented here supports the notion of a conditional impact of electoral performance. Whenever office status changes, electoral success or failure make little difference for party leader survival. By contrast, if there is no change in office status, winning versus losing votes does matter. In addition, when parties enter government (yet not when they exit from the cabinet), the impact of electoral performance on leadership survival is significantly larger than when they maintain their office status. These results offer, to the best of our knowledge, the most direct evidence to date for the priority of office‐seeking over vote‐seeking in party behaviour.

Conclusion

In their pioneering analysis of party leader survival, Andrews and Jackman conclude that parties ‘are at their core motivated by electoral performance’ (Reference Andrews and Jackman2008: 657). Our conclusion is radically different. As our analysis suggests, success in the electoral arena is just a sideshow – the more important prize to obtain is government. This result is much more in line with the prevailing theory of the utility of office versus votes (Müller & Strøm, Reference Müller and Strøm1999). To be sure, these diverging conclusions may to some extent be driven by case selection: Whereas Andrews and Jackman focused mostly on Westminster democracies where electoral performance strongly determines office achievement, our study includes a much broader set of democracies – many of them coalition systems in which the connection between electoral success and government participation can be quite tenuous.Footnote ¹³ More importantly though, our paper is the first one to explicitly theorise and model the interaction between performance in the electoral and governmental arenas in order to determine which domain has a greater impact on party leader survival.

Our findings show that electoral performance and office achievement have consequences for party leaders, yet there are important implications for party behaviour in general. Whether and when parties choose to depose their leader and select a new one teaches us about the goal orientation of parties. Party goals, in turn, are relevant for theories of party competition, coalition formation and public policy making. For example, party position shifts are typically explained by efforts of parties to increase their vote share (for a review, see Adams, Reference Adams2012). This behaviour arguably produces responsiveness to voters. If office‐seeking trumps vote‐seeking, the expectations and outcomes of party performance in the governmental arena should interfere with – or even impede – the responsiveness of parties to voters (Schumacher et al., Reference Schumacher, de Wardt, Vis and Klitgaard2015; Van Der Velden et al., Reference Van Der Velden, Schumacher and Vis2018). Although theoretically important, empirical evaluations of party goals are rare, and therefore, typically, party goals do not leave the theory sections of papers (Meyer & Wagner, Reference Meyer and Wagner2016; Pedersen, Reference Pedersen2012). Using leadership survival, we demonstrate one way to evaluate party goals.

One party goal that we do not study (directly) is policy. This is primarily for practical reasons. It is not evident how achieving policy goals should be measured, and whether parties strive for merely shifting government policy towards their ideal point (policy‐seeking) or getting their own policy implemented in full (policy purity). In addition to this, there is overlap between achieving policy and office goals, because parties usually need office to implement policy. However, this is not always the case, especially in countries with a tradition of minority governments. In sum, it remains elusive how to evaluate policy goals, and indeed some of the leaders in our analysis might have been deposed by either disappointed office seekers or disappointed policy seekers.

While our analysis may be of most interest to party politics scholars, it also highlights the more general pressures and incentives under which political leaders operate. Because party leadership is the most important stepping stone to high executive office, government leaders will often have internalised their party's ambitions and thus pursue them to the best of their abilities. This, in turn, has implications for political elite behaviour in a variety of areas, from government formation to domestic policy making and international negotiations.

In this article, we establish an effect of performance in recent elections and government formation (and termination) outcomes on party leadership survival. What we leave open is the question whether prospective performance might also motivate parties to remove their leaders. Parties may rationally anticipate electoral defeat or ejection from government and therefore replace their leader in an effort to turn the tide. A recent example of this is when the Austrian People's Party before the 2017 parliamentary election replaced the mildly liked Reinhold Mitterlehner with the much more popular Sebastian Kurz. Within a few weeks, the party had gained 10–12 percentage points in the polls and was thus able to secure first place in the election and the position as senior coalition party in the post‐electoral government formation. Such examples clearly suggest that parties do not only respond to actual, but also to expected electoral performance (i.e., polls). This certainly presents a promising avenue for future research.