Abstract
Political parties are central to modern democracy and the selection of their leaders is one of the most crucial decisions for any political party to make. Yet, the analysis of party leadership survival is still in its infancy. The pioneering research has been confined to few countries and decades and has focused exclusively on performance-related explanations. While performance is an obvious determinant of party leader survival, generations of research on party organizations suggest that intra-party factors should matter, too. We argue that, while the political performance of a party leader (winning elections, securing government participation) is important, intra-party support and the rules of leadership selection add substantively to our understanding of why party leaders survive or fall. We test these expectations on a new dataset covering all leaders of Austrian parties between 1945 and 2011. The results of our statistical analysis support our claim and show that intra-party factors have a considerable impact on party leader survival.
Introduction
In modern democracies, party leaders – the ‘ones’ of their respective parties – are particularly important office-holders. As their party’s single most important decision-maker, the leader exercises major influence on the policy >positions the party advances, the party’s behaviour vis-à-vis other parties, and typically the most important government position available to the party when in office is reserved for the party leader. Likewise, the party leader has a great say in advancing or hindering the careers of fellow partisans. At the same time the party leaders constitute an important reason why voters vote for (or against) a party (Aarts et al., 2011; Bittner, 2011). A leader holding on to power tends to be a constant in the calculations of other actors and hence contributes to political stability. Any change in party leadership can be consequential in terms of public policy, inter-party cooperation, the recruitment to other elite positions and electoral competition. For instance, the party change literature argues that leadership change ‘is a sufficient, though not necessary, condition for party change’ (Harmel et al., 1995: 5; see also Harmel and Janda, 1994; Janda, 1990; Wilson, 1980, 1994). A change in party leadership thus sends important signals to the voters, to the other parties, to the economy, to the party activists, and may trigger reactions of considerable consequence. Continuity or change in the position of party leader is thus a relevant political phenomenon worth studying.
It is therefore all the more striking that research on party leader survival has not yet attracted a great deal of scholarly attention. While duration in public political office is a frequently studied subject, political parties have largely been spared from both descriptive and analytical attempts to come to grips with the phenomenon. In contrast to the numerous accounts of the survival of heads of states or governments (e.g. Bienen and van de Walle, 1991, 1992; Blondel, 1980; Bueno de Mesquita et al., 2003), analyses of government durations (Diermeier and Stevenson, 1999, 2000; Strøm and Swindle, 2002; Warwick, 1994), or studies of ministerial tenure (Berlinski et al., 2007, 2010; Huber and Martinez-Gallardo, 2008; Indridason and Kam, 2008), research on party leader survival is very much in its infancy. The pioneering study of Andrews and Jackman (2008) showed that a party’s electoral fortunes and holding government office are the (substantively relevant) factors that drive leadership change. Indeed, being successful at the polls and winning or holding government office are core goals of political parties (Müller and Strøm, 1999; Strøm, 1990). It is only natural that party performance with regard to these goals should determine the fate of the one most responsible for that performance.
Yet, performance-related aspects may not be the only determinants of party leader survival. Drawing on the party organization literature we posit that intra-party factors should also be taken into account. More specifically, we claim that organizational and behavioural characteristics of the leadership selection and removal process impact on the odds of party leader change. While selection and de-selection may be separate processes and follow their own rules (Cross and Blais, 2012a, b), we do not systematically distinguish them in the present article. Such differentiation is irrelevant for our empirical case, where removing a party leader requires electing a successor. Refining our hypotheses for a larger sample may thus well lead to expectations that occasionally differ for the two processes.
There are at least two reasons why we expect that organizational and behavioural characteristics of the leadership selection process matter. First, the rules governing the leadership selection process place themselves on a continuum between open and closed. At the open end they invite challengers to come forward while shielding incumbents from internal competitors at the closed end of the spectrum. (see De Winter, 1993; Kenig, 2008; Punnett, 1993; Strøm, 1993; see also Hazan and Rahat, 2010). Second, the actual selection process reveals information about the intra-party standing of the leaders. Overwhelming support may make the leader appear as the one and only person for the job and being unassailable. On the opposite side of the scale of intra-party standing, a leader who cannot rally the party behind him- or herself may look weak and already doomed. Selection processes thus reveal information on the intra-party standing of the leader. By encouraging or discouraging potential competitors, such information by itself contributes to the dynamics in leadership selection.
In the following section we outline our theoretical expectations about the determinants of party leader survival. We focus on explanatory variables pertaining to office-seeking and vote-seeking objectives as well as behavioural and institutional features of intra-party politics. We then present our data and discuss the leadership selection mechanisms in place in the parties under study in more detail. Next, the analytical section reports the results from our statistical analysis and examines the effects of the most relevant variables in more detail. We conclude by discussing implications for future research.
Theory and hypotheses: Explaining party leadership survival
Party leaders constitute a special breed, even among politicians. Most of them have worked hard politically and fought their way up for many years. Being party leader and combining this office with the most attractive public office available to their party is their life. If the decision were left to them, many would probably go on in their political offices for ever – and probably longer than anyone else would consider reasonable. In short, when in office, many of these leaders can hardly see any attractive alternative life course. Yet, since party leaders are mortal not only physically but also politically, then what are the factors that determine whether they will stay in the leadership position or be forced out of office? We distinguish between a party leader’s political performance, that is, the achievement of party goals (policy, office and votes), and the institutional and behavioural environment that party leaders face within their own party.
The political performance of party leaders
Parties make those individuals leaders whom they expect to make the most important contributions to their political performance. While in some (hopefully rare) situations internal reconciliation and consolidation may be what the party most urgently needs, most performance expectations of competitive parties are outward-directed.
Theories of rational party behaviour suggest that policy, office and votes are the prime goals of political parties (Müller and Strøm, 1999; Strøm, 1990). Individual parties may give different weights to these goals and these may differ according to the relevance different intra-party actors have in making party decisions. For instance, party activists are generally considered more policy-oriented and professional politicians more interested in office per se. Different modes of making decisions (such as selecting leaders), which give different weights to various groups, thus may shift weights also between party goals. Likewise, party goals may change over time for many reasons. Still, it is highly unlikely that a party fully dismisses any of these objectives, save for extreme cases such as anti-system parties that reject participation in democratically elected governments on principle.
Also, the goals of policy, office and votes are interlinked. While at times a skilful party leader may be able to maximize all three (Poguntke, 1999), often trade-offs exist and one goal can only be realized at the price of abandoning or disregarding another. The interdependence and potential conflicting party goals become visible when major decisions have to be made: Entering government, for instance, may satisfy a party’s office demands and surely helps in promoting its policy objectives. However, such a decision may be electorally harmful. As the single most important decision-maker within a party, the leader bears the bulk of responsibility for these choices and thus the success or failure in achieving the party’s goals, whatever the relative weights given specifically to the pursuit of policy, office or votes.
We can thus expect that parties will hold on to the leaders who are successful in achieving their parties’ goals while replacing those that fail to do so. More specifically, the party selectorate will consider the performance a party takes under its leader when deciding about maintaining or replacing him or her.
If we assume that all parties are vote-seekers to some extent, we can expect that electoral performance is a major determinant of party leader survival. Party leaders who win elections should, ceteris paribus, be more likely to survive in office than those who lose at the polls. Although we are considering parliamentary elections (with many candidates of each of the parties) rather than elections to chief executive office (featuring only one candidate per party), the party leaders exercise great influence on the results. They do so by their direct appeal to the voters (Bittner, 2011; King, 2002) and more indirectly via their influence on the party list, policy programme and campaign (Poguntke and Webb, 2005). While it may not be strictly necessary for leaders to win elections, it is most likely harmful to lose them. This is so because electoral defeats are ‘sticky’ in terms of image and may require a new leader taking over to allow for a successful restart. Along these lines, previous studies have highlighted the significance of electoral defeat for the de-selection of party leaders (e.g. Andrews and Jackman, 2008; Bynander and ’t Hart, 2007: 53 f.). Finally, if we consider that winning votes also has an instrumental value for the achievement of the other party goals (Strøm, 1990: 573), we must assume the link between electoral performance and party leader survival to be especially strong. We thus conjecture:
Another important aspect of political performance pertains to the achievement of government office. While office-seeking behaviour may be instrumental, for instance in serving the purpose of implementing a party’s policy platform, theories of party competition and coalition formation have argued that politicians value office as an intrinsic goal (Downs, 1957: 28; Riker, 1962: 23). Yet, regardless of whether office is valued intrinsically or instrumentally, we can assume that participation in government is positively related to a party leader’s chances of survival. For one thing, getting into government or, conversely, losing government office is by itself a strong signal of success or failure that may be powerful enough to outweigh electoral performance. Moreover, cabinet participation endows party leaders with a multitude of resources, ranging from cabinet portfolios and policy-making capacities to more mundane government goods that can be allocated strategically. It is reasonable to assume that these resources are used, inter alia, to build or maintain intra-party coalitions and thereby reinforce the leader’s position. Indeed, there is substantive evidence to suggest that, all else being equal, holding government office prolongs the tenure of party leaders (Andrews and Jackman, 2008: 670).
However, the literature on coalition governments also tells us that government participation may not necessarily impact all parties uniformly (Falcó-Gimeno, 2012). As Narud and Valen (2008: 383) show, the electoral penalty from holding office is considerably smaller for the prime minister’s party than for the cabinet as a whole. Indeed, coalition rule often demands larger policy sacrifices from leaders of junior government parties while at the same time providing them with fewer resources to buy support from their co-partisans. Thus, entering a coalition places a disproportionate burden on those party leaders. We hence expect government participation to be beneficial only for leaders of senior government parties, whereas the impact on leaders of junior parties should be in the opposite direction.
The third aspect of political performance concerns the achievement of policy goals. Indeed, most parties will value policy gains for intrinsic reasons. At the same time, and under ideal circumstances, such policies may also serve the purpose of enhancing a party’s electoral prospects. While we assume that policy performance is a valid determinant of party leader survival, the measurement of policy success is a notoriously difficult undertaking. In contrast to electoral performance or government participation, there are no clear yardsticks by which to evaluate a leader’s performance. A reliable measure for the policy performance of parties and political leaders has yet to be developed. In any case, it is safe to assume that office and policy success are correlated to some degree. This fact is pushed to its extreme in the idea of ministerial government as formulated by Laver and Shepsle (1990, 1996), who equate holding ministerial office to policy dictatorship within the respective jurisdiction. Thus, to some extent, government participation can be regarded as an indicator of the achievement of both office and policy objectives.
While all competitive political parties share the desire for electoral success, government office and policy achievements, they are quite different in terms of history, organizational culture, and the rules that govern the selection process. We expect that such differences between political parties affect the way they treat their leaders.
Intra-party determinants of party leader survival
A huge body of literature argues that leadership selection mechanisms and devices to hold leaders accountable matter by making incumbents more or less vulnerable to replacement (e.g. Bueno de Mesquita et al., 2002, 2003; Konrad and Skaperdas, 2007; Samuels and Shugart, 2010). We cannot see why political parties should be fundamentally different in this respect from the government institutions they strive to control or, indeed, from any other type of formal organization. Based on this premise we put forward systematic expectations about the impact of intra-party support for party leaders and the institutional features of the leadership selection process.
To begin with, the office of party leader is unique, and typically the number of people who are ambitioned to take up that office exceeds 1. Given all the power that comes with this office, it is not hard to understand why. Also, it can be considered as the ticket to even more desirable offices. Hence, the demand will always be greater than can be satisfied. At times, this inevitably results in high levels of intra-party competition for the leadership position, even if such competition is not always visible from the outside. In the bid for the party leadership, however, incumbents possess a significant advantage. Still, Bynander and ’t Hart (2007: 54) report internal rivalry as a cause of leadership change in almost half of the cases in their sample. Also, plenty of case studies have scrutinized some of the most significant instances of intra-party conflict at times of leadership succession (e.g. Alderman and Carter, 1991; Bynander and ’t Hart, 2008; Cowley and Bailey, 2000).
As a general rule, political leaders need to maintain sufficient levels of support from their subordinates in order to remain in power. The most elaborate version of this argument has been put forward by the proponents of selectorate theory (Bueno de Mesquita et al., 2002, 2003). Political leaders are thus more likely to be replaced if their support coalition shrinks in size so that it can no longer prevent the removal of the leader. Indeed, if a party leader is re-elected but has unusually low support (as indicated, for instance, by votes at party conferences) this is a signal to potential rivals that the incumbent is no longer undisputed. This, in turn, may set a process into motion in the course of which the authority of the leader is further undermined. We therefore assume that the degree of approval that party leaders receive from their selectorate – even in the absence of an opponent – is indicative of the extent to which they are in danger of being dismissed in the near future.
In addition, the size of the voting body and the method of voting can make a difference in the readiness of dissenters to withhold their support (Müller et al., 1992). While smaller voting bodies and open voting will discourage deviant behaviour, secret voting in large bodies provides an opportunity structure that is favourable to it. Such institutional features of the leadership selection process have hitherto been researched with respect to the competitiveness of leadership contests (Kenig, 2008), their openness (Carty and Blake, 1999; Cross and Blais, 2012a, b; Kenig, 2009; LeDuc, 2001), the degree of oligarchy within parties (Müller and Meth-Cohn, 1991; Müller et al., 1992) or their impact on the removal of prime ministers (Weller, 1994). In what way do the specifics of the leadership selection mechanism impact on the survival of party leaders? We assume that more closed methods of leadership selection pose greater obstacles to potential challengers and thus lead to longer durations of leadership tenures. Smaller selectorates and more elitist voting bodies provide the incumbent party leader with potentially important means by which to influence the decision-making process in his or her favour. A party leader acquainted with the art of intrigue may outmanoeuvre his or her rivals by forming alliances and purchasing support in handing out office or policy rewards to electors – a strategy that is much easier to implement if the recipients of such goods are small in number. Viewing such trade-offs from the opposite side, Bueno de Mesquita et al. (2002: 562) stated ‘public goods become more attractive to provide relative to private goods as the size of the winning coalition increases’. More elitist modes of leadership selection should thus favour incumbent leaders who – particularly when in public office – are able to deliver, while challengers are likely to be limited to making promises for the future and hence face a credibility problem. In addition, the party leader may also be in a powerful position to shape the composition of the upper echelons of a party. He or she can use these (mostly informal) powers to staff the leadership selectorate with loyal followers. 1 A small voting body merely provides opportunities and under most circumstances the party leader should be the one who is best placed to exploit them. If he or she is not willing or able to exploit them, this factor might easily help competitors and turn against the incumbent. It is also plausible that party leaders may be more immune to internal opposition when chosen by a truly wide selectorate. As Mair (1994: 16) argues, enfranchising ordinary party members can be a purposeful strategy by party elites to deprive mid-level activists of their influence. Leaders thus become accountable to a larger and more amorphous selectorate that is more difficult to mobilize and ostensibly more prone to be manipulated. This argument implies that the relationship between selectorate size and party leader survival is not necessarily linear. However, since our empirical case does not cover parties with full membership votes for the party leadership, we state our fourth hypothesis based on our initial intuition:
Along similar lines, we argue that the statutory length of the leadership term impacts on a leader’s odds of being de-selected. This assumption follows a very simple intuition: a leader who needs to seek confirmation in office every year is much more prone to being replaced than a leader with a four-year mandate. To be sure, most modern parties have mechanisms in place that allow for the early de-selection of a party leader (Kenig, 2009: 442). Yet, the formal requirements for such an undertaking present an additional hurdle to a potential leadership challenge and the negative externalities such intra-party processes are likely to cause work in favour of the incumbent leader. Also, a longer office term will invariably include temporary ups and downs of the party while short-termed leaders may preside only over one of these alternatives until the next election comes up. The selectorate’s knowledge that down-periods have been followed by up-periods in the past may then stabilize long-termed leaders in the case of a down-period when the election is due. Lacking such knowledge in the case of short-termed leaders, electors may interpret temporary downs as indicating a permanent trend under the leader and act accordingly. Our fifth hypothesis therefore reads:
Clearly, the intra-party features discussed here do not exhaust party organizational characteristics and differences. Yet, they are the ones we can most directly relate to our dependent variable and they allow for clear intuitions about the specific mechanisms and direction of effects.
Case selection, data and method
The extant studies of party leadership selection and de-selection cover the entire range of approaches, from the analysis of particular episodes within one party (e.g. Heppell and Hill, 2008; Heppell et al., 2010), single-party studies (e.g. Denham, 2009; Denham and O’Hara, 2008), single-country studies (Courtney, 1995; Müller and Meth-Cohn, 1991; Marsh, 1993; Punnett, 1991), to a few genuinely comparative pieces (Andrews and Jackman, 2008; Cross and Blais, 2011, 2012b; Kenig, 2008, 2009; Laing and ’t Hart, 2011). Each of these approaches has its distinctive advantages and limitations. In the case of a single event, this is often a complex process upon which the analyst can generate high-quality information. Such an analysis can do justice to factors that are unique to the case or cannot be considered with regard to other cases lacking relevant information. These advantages are balanced by the fact that the results, interesting as they may be, face severe limitations in terms of generalization to a universe. At the other end of the scale, comparative studies have the potential to maximize the variation in both the dependent and the independent variables, thereby inviting generalization. This comes at the price that the complexity of the data generation process perhaps requires greater simplification than we would like to see and/or limiting the analysis to shorter time periods (Kenig, 2008) or a few countries which share some features that allow overcoming these problems (Andrews and Jackman, 2008; Cross and Blais, 2012b). More geographically inclusive research is only beginning to emerge (Horiuchi et al., 2013; Laing and ’t Hart, 2011).
In opting for a long-term single-country study we steer some middle way between the extremes of single-event and long-term multi-country studies. While we choose Austria mainly because country expertise allows us to produce high-quality data on intra-party factors over the full post-war period, it also represents an interesting case. It combines a long-lasting democratic regime with some equally long-lasting political parties – we can observe some parties over a period of 66 years. At the same time, we include parties with shorter fates in national politics. Our selection criterion is to include all parties in our analysis that won parliamentary representation after 1945. 2 These parties experienced considerable ups and downs in terms of electoral support and holding government office. Yet, the Austrian case is perhaps most interesting in combining parties with a range of organizational patterns. Most importantly for the present purpose, the centralization of decision-making varies considerably, leading to different formal procedures of selecting (and de-selecting) party leaders.
From the timeline displayed in Figure 1 it becomes clear that parties differ substantially in the frequency of leadership turnover. By the early 1990s, for instance, the ÖVP had had ten party leaders as opposed to the SPÖ’s five. The FPÖ, in turn, had the same number of leaders between its establishment in the mid-1950s and the late 1990s as it had in the five years that followed its entering the government coalition with the ÖVP (Schüssel I) in early 2000. Two smaller parties in our dataset do not display any variation in leadership during their time of parliamentary representation. Whereas the persistence of Johann Koplenig as leader of the KPÖ may be attributed to the distinct nature of Communist leadership, 3 Heide Schmidt’s considerable public appeal and personal identification with the Liberal Forum rendered her undisputed as this party’s leader. In contrast, the Greens, who were traditionally suspicious of the concept of leadership by a single person established a party leadership position only six years after entering parliament.
Austrian party leaders, 1945–2011.
While it is undoubtedly crucial to understand such idiosyncrasies of parties and their leaders, this study attempts to account for the systematic variation in the survival of party leaders. We therefore specify a Cox proportional hazards model with leadership tenure measured in days as the dependent variable. Starting from the assumption that party leaders want to hold on to their position as long as possible, we apply a censoring regime that treats every departure from the party leadership not caused by death as a failure event (Schleinzer in 1975 and Haider in 2008, both killed in car accidents, as well as Reinthaller in 1958), severe illness (Pröll in 2011), promotion to higher office (Schärf, who was elected Federal President in 1957) or credible voluntary retirement (Vranitzky in 1997). 4 In addition, we treat all observations extending beyond the year 2011 as right-censored.
Figure 2 plots the distribution of the dependent variable. The median duration of leadership tenure is 1310 days; the mean is 1955, with a standard deviation of 1743 days. This makes Austria a typical case among post-war democracies (Laing and ’t Hart, 2011: 113).
Histogram of leadership durations.
The independent variables
In order to test our hypotheses concerning the survival of party leaders, we specify a number of covariates.
First, we examine the impact of holding executive office by including two dichotomous variables, indicating whether the leader’s party is represented as the senior or junior government party during each observational period. As theorized above, we assume that government participation boosts leadership duration, but only for leaders of the senior government party (the Chancellors). The Austrian case is peculiar because of the predominance of grand coalition governments and low partisan turnover in cabinet. A good deal of the variance captured by the two government participation variables is therefore due to variation between parties rather than within parties over time.
Second, we operationalize electoral performance as the percentage change in a party’s popular vote-share at the most recent parliamentary elections. While it would be an equally valid approach to use legislative seat-shares, we choose vote-shares in order to eliminate the subtle effects of the changes to the Austrian electoral system in 1971 and 1992.
Likewise, we record intra-party support as the change in the vote-share that a leader received at the party conference or in the party executive. Negative values thus indicate that support for a leader has decreased, whereas positive values record the opposite.
As for the institutional determinants of party leader survival, we take into account the variation across time and parties in the statutory length of the leadership term. This variable takes on values between one and four years. There is considerable variation across parties, but also over time. Most Austrian parties have increased the length of their executive terms over recent decades (Müller et al., 1992: 122), thus requiring the leader to seek the consent of the party on a less frequent basis. While, for instance, the SPÖ’s Adolf Schärf was consecutively re-elected eleven times between 1945 and 1957, Wolfgang Schüssel had to seek confirmation from the party congress only twice in his 12-year period as leader of the ÖVP.
A second institutional variable concerns the mode of leadership selection. We distinguish between elite-centred and agency-centred modes of selection (Kenig, 2009: 435), represented by the values 1 and 0, respectively. The former records all instances of (1) direct election by the executive or (2) nomination by the executive and subsequent confirmation by the party congress, whereas the latter is coded for all instances of direct election by the party congress. Our expectation here is that less open forms of leadership selection result in greater chances of party leader survival.
As control variables we include the age of the party leader and the year of each observation. While we do not put forward explicit theoretical arguments about these two covariates, our intuition is that older leaders will be more vulnerable and that there is an overall tendency towards higher levels of leadership turnover across time.
Table 1 presents the summary statistics for all variables in the dataset and their hypothesized effect on the removal of a leader.
Summary statistics (n = 229 observations from 45 party leaders).
Note: Missing data are due to incomplete information on intra-party support.
Statistical analysis
We test our assumptions about the determinants of party leader survival on an original dataset covering the 45 leaders of the 8 parties represented in the Austrian parliament between 1945 and 2011. Given that we have no a priori expectations about the functional form of the dependent variable, we opt for the semi-parametric Cox proportional hazards model (Cox, 1972). The maximum likelihood estimation (MLE) that this model employs produces asymptotically unbiased and efficient estimates in large samples. Yet, it is possible that the parameters are biased and/or inefficient as the sample size becomes smaller. A biased estimate would lead to wrong inferences about the effect size associated with a specific covariate. Inefficient estimates would lead to false conclusions concerning the significance of the findings. What, then, are the implications of the small sample size of 45?
First, it needs to be recognized that, while there are only 45 leaders in the sample, the presence of time-varying covariates results in a total of 229 observations. For instance, a leader who competes in five general elections is reported in the dataset as five observations with changing values for the electoral performance variable. Likewise, new observations are generated with each change in the government vs. opposition status of a party, and each time a leader is re-elected by his or her selectorate (thus recording variation in the leader’s intra-party support). Although the 229 observations in the dataset are thus clustered on party leaders, the sample contains much more information than if there were only one observation per variable for each of the 45 leaders. Since MLE is a procedure that finds the parameters most likely to have produced the data, it can be argued that the uncertainty attached to these parameters is considerably reduced by the additional information contained in the time-varying covariates.
Second, one can try to assess the likely size of the bias. According to a simulation study conducted by Langner et al. (2003), the extent of the bias in parameters for dichotomous predictors is rather limited (less than 10 percent) if the risk of failure for the ‘exposed’ cases (i.e. cases taking on a value of 1 for the respective covariate) is not too small (≥ 0.15). For the two dummy variables in the data, the ‘exposed’ cases exhibit much greater risk of failure: 18 out of 23 leaders (0.78) for all leaders in government and 5 out of 9 leaders (0.56) that were selected by the party executive. As an important caveat, however, it should be noted that the amount of bias in the parameter estimates of the continuous covariates is less clear.
Third, there are good reasons for assuming that inefficiency in MLE estimates does not necessarily lead to completely wrong inferences. As Hart and Clark (1999) demonstrate, small sample sizes do not increase the probability of a true null hypothesis being rejected. While the risk of making type I errors is thus rather limited, type II errors are more likely to occur with small samples (Hart and Clark, 1999: 8–10). This suggests that we should be careful in reverting to the null hypothesis even in the absence of statistical significance, since a small sample may simply not contain enough information to detect a true relationship.
With these caveats in mind, we present a number of Cox proportional hazard regressions in Table 2. In addition to the full model (Model 7), we report six bivariate models, one for each covariate, in order to show that the effects are robust across different specifications. All models are calculated with standard errors clustered on party leaders.
Determinants of party leader survival. 6
Note: Figures are hazard ratios from Cox proportional hazards regression with standard errors clustered on leaders; t-statistics in parentheses; + p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.
The interpretation of the reported hazard ratios is straightforward: a value above 1 indicates a greater likelihood of removal as the value of the independent variable increases, whereas values below 1 mean that leaders are less likely to be replaced.
Taken together, the covariates in the analysis have reasonably good explanatory power. The full model yields a McFadden’s R-square of almost 0.2, thus suggesting that a substantial portion of the variation in leadership duration can be accounted for by the independent variables. Also, an inspection of the baseline hazard (not reported) shows that party leaders’ odds of being removed increase monotonically over the course of their tenure. This suggests that irrespective of the effects of the independent variables, the passage of time works against party leaders retaining their position.
Looking at the individual predictors more closely, Models 1 and 7 very much yield the expected result for the electoral performance variable. Winning elections raises the chances of party leaders staying in office, while electoral defeat is likely to doom their odds of survival. This effect is quite strong: a hazard ratio of 0.793 indicates that a gain in the vote-share of 1 percent decreases a party leader’s chance of being de-selected by over 20 percent. Also, the impact of electoral success (or failure) is almost unchanged when the other covariates are included (see Model 7). This fully confirms our expectations about the importance of winning votes for the survival of party leaders and is in line with the findings reported by Andrews and Jackman (2008: 670). In addition, the pseudo R-squares strongly suggest that electoral success (or failure) has the greatest explanatory power of all covariates in the analysis. More than half of the explained variance in the full model is accounted for by the electoral performance predictor.
By contrast, the two indicators for government participation (one for senior and one for junior coalition parties) fail to reach statistical significance in the full model. The strong divergence in the direction of the effects and the weakly significant hazard ratio in Model 3 suggest that the intuition behind H2a and H2b is not necessarily false. Yet, the hypothesized effects are probably too small to materialize in our limited sample. The statistical models thus yield no conclusive evidence as to the validity of the hypotheses pertaining to government participation. We have also tested for the difference between the leaders of senior and junior government parties by switching reference categories (results not reported). As above, no statistically significant effect emerges.
This is surprising in light of the previous findings (Andrews and Jackman, 2008: 670), yet it may well be explained by the high stability in the composition of Austrian post-war governments. While both traditional parties of government, the SPÖ and ÖVP, have been in power during at least 50 of the past 66 years (see Figure 1), the other parties mostly had to limit themselves to the opposition role. Thus, there is relatively little inner-party variance in the office-related predictors to account for the survival of party leaders.
We test for the theorized party organizational factors individually in Models 4 to 6. Changes in intra-party support clearly emerge as a significant predictor of party leader survival. This corroborates our assumption that declining approval of a leader within the party ranks is a valid indicator of a leader’s shrinking odds of survival. 5 The hazard ratio reported in the full model suggests that a 1 percent change in intra-party support alters the risk of removal by over 4 percent. Given that leadership elections are typically uncontested and therefore often yield voting results similar to those in autocratic regimes, it is remarkable that even small changes in the support levels are indicative of a leader’s chances of staying in office.
Furthermore, we find a significant effect of the selection mechanism. Leaders that were directly elected by a party congress as opposed to an elite-centred process are in much greater danger of being replaced. A hazard ratio of 0.236 in the full model indicates that this effect is substantial. All else being equal, selection by the executive thus brings down the chances of being de-selected by more than 76 percent. More exclusive processes of selection thus correlate with party leaders being less exposed to the risk of removal. Party leadership office periods are defined by the party rules in all Austrian parties. This implies that the process of removing an incumbent leader is identical to that of selecting a new one. Thus our result fully corroborates the findings by Cross and Blais (2012b: 125), who report a substantial (if only barely significant) effect of the regulations guiding the removal of party leaders: ‘the easier it is to remove the leader, the shorter the stay at the helm of the party’. In fact, since the process of removing a leader is identical to the selection of a successor, it could be argued that it is the selectorate’s removal authority that actually matters. Yet, since selectorates and de-selectorates are empirically equivalent, the analysis cannot distinguish between these two effects.
Finally, we find no significant effect for the leadership term covariate. Party leaders’ risk of replacement therefore seems to be independent of the statutory lengths of their term. In other words, the mere fact that some party leaders need to seek approval from the party on a more frequent basis than others does not explain variation in the total duration of party leaders’ tenure.
The effects of the control variables are in the expected direction, yet only the time variable is significant at the 10 percent level. Party leader turnover thus increases over the six decades that we observe. A myriad of factors could help to explain this trend, from a greater intensity of political competition to higher degrees of electoral volatility and heightened media scrutiny.
Overall, the performance of the covariates relating to the parties’ internal life support two of our initial expectations: accounting for the intra-party standing of leaders and the institutional rules of leadership selection considerably enhances our understanding of the determinants of party leader survival. More oligarchic mechanisms of selection (and thus removal) help leaders retain their position, while more open selection (and removal) mechanisms increase the risk of being de-selected. Also, there is good support for the notion that intra-party support levels are indicative of a party leader’s chances of survival.
Figures 3 and 4 provide a more intuitive account of the impact of two variables of theoretical interest that are novel to the study of party leadership survival. They depict survival functions contingent on either of the two variables with all other variables held constant at their mean. These estimations are based on the full model (Model 7).
Impact of intra-party support on party leader survival.
Impact of selection mechanism on party leader survival.
As becomes clear from Figure 3, the impact of intra-party support is small but consistent. All else being equal, a loss in support of 10 percent takes the survival function below the 0.5 threshold several hundred days before the zero-percent reference line reaches that point. Party leaders striving for a long tenure should therefore rightly worry about their backing at party conferences (or in other voting bodies).
Figure 4 displays the impact of the selection mechanism on the survival of party leaders. Again, it becomes clear that a more exclusive mode of leadership selection results in higher chances of survival. The degree of oligarchy that becomes manifest in such institutional characteristics is clearly a relevant factor in explaining party leaders’ fate.
In order to sum up the findings of the above analysis, Table 3 reports the hypothesized effects for all independent variables and evaluates them against the results from the statistical models. A good part of our initial expectations are borne out by the data. Yet, while there is substantial evidence for the general importance of electoral performance, there seems to be no uniform impact across parties of the government participation variable. As to the intra-party variables, we find that variation in the level of intra-party support and the nature of the leadership selection mechanism make a difference, whereas the statutory term length does not impact on party leader survival.
Overview of results.
Conclusion
The selection of its leader is one of the most significant decisions for a party to make. Such decisions often have profound consequences for a party’s electoral prospects, the policies it promotes and its performance in public office. A better understanding of the determinants of the selection and de-selection of party leaders is therefore crucial for party researchers and students of democratic politics more generally. We add to this understanding by providing, to the best of our knowledge, the first systematic account of the impact of institutional and behavioural characteristics of intra-party life on party leader survival. Our analysis shows that electoral success and institutional intra-party factors are the most significant determinants of party leader survival. However, we do not find a strong impact of government participation. This may be due to our small sample size, but could also be a finding specific to the Austrian case.
This caveat can be stated more generally. Although the parties included in our analysis show considerable variation with respect to most variables, they are all nested in one country. As always with single-country studies, it remains to be seen to what extent our specific results generalize beyond the case. Yet, we firmly believe that our initial intuition will hold up: organizational factors do have a major impact on the precise definition of party goals, the relative weight given to each of them, on how the performance of the one who leads is assessed and whether he or she will survive or be replaced in the party leadership position.
Footnotes
Acknowledgement
This research was carried out under the auspices of the Austrian National Election Study (AUTNES), a National Research Network (NFN) sponsored by the Austrian Science Fund (FWF) (S10903-G11). We are grateful to the anonymous reviewers for many useful comments.