Public Choice

pp 1–19

Strategic gerontocracy: why nondemocratic systems produce older leaders


DOI: 10.1007/s11127-017-0449-5

Cite this article as:
Magni Berton, R. & Panel, S. Public Choice (2017). doi:10.1007/s11127-017-0449-5


One characteristic of nondemocratic regimes is that leaders cannot be removed from office by legal means: in most authoritarian regimes, no institutional way of dismissing incompetent rulers is available, and overthrowing them is costly. Anticipating this, people who have a say in the selection of the leader are likely to resort to alternative strategies to limit his tenure. In this paper, we examine empirically the “strategic gerontocracy” hypothesis: Because selecting aging leaders is a convenient way of reducing their expected time in office, gerontocracy will become a likely outcome whenever leaders are expected to rule for life. We test this hypothesis using data on political leaders for the period from 1960 to 2008, and find that dictators have shorter life expectancies than democrats at the time they take office. We also observe variations in the life expectancies of dictators: those who are selected by consent are on average closer to death than those who seize power in an irregular manner. This finding suggests that gerontocracy is a consequence of the choice process, since it disappears when dictators self-select into leadership positions.


Gerontocracy Democracy Dictatorship Leadership selection 

1 Introduction

This article aims to provide an answer to the question of why age, in some political systems, is relevant in selecting the leader. Sometimes, gerontocracy is an explicit rule. In many cases, it is only an informal rule. We provide here an argument to explain why leaders are chosen regularly from among the oldest potential candidates, and offer some clues suggesting that explicit age-related rules are the consequence of this regularity.

Historically, elderhood often was associated with high status in primitive and agrarian societies (Flanagan 1989; Posner 1995) and, in many tribes, the chief was chosen from among the oldest men (Simmons 1945; Read 1959; Spencer 1965). In early modern societies—in Greece or in Rome—gerontocracy also was explicitly acknowledged, and associated with non-removability (Von Kondratowitz 2009, p. 114). While gerontocracy declines in the contemporary world, it still survives in several regions of the globe, like in Africa (Fortes and Evans-Pritchard 2015).

Gerontocracy is classically explained by two main sets of factors, based either on social ties acquired with age or on personal characteristics, such as wisdom or knowledge (Werner 1981). However, both explanations fail to explain why, in modern democracies, gerontocratic rules tend to disappear. First, in democracies, leaders also need extensive social ties to be elected. Second, as recent evidence shows, aging leaders are not necessarily wiser: they produce less economic growth (Jong-A-Pin and Mierau 2011) and are involved more often in interstate wars (Horowitz et al. 2005; Potter 2007; Bak and Palmer 2010).

We put forward an explanation we call the “strategic gerontocracy” hypothesis. Non-removability reduces the power of people who select the leader (the “selectorate”),1 because they can decide only once. The selectorate, however, can increase its power by choosing a leader with a short life expectancy. Furthermore, when no institutional mechanism to remove the leader is available, the selectorate faces a strong incentive to choose a candidate who is likely to give up power after a few years, in order to minimize possible losses and to avoid the worst-case scenario (i.e., the scenario in which a bad ruler stays in office for decades). Since future leaders may be loved or hated, the minimax strategy consists of shortening the durations of their reigns.

The crucial implication of this argument is that introducing formal limitations of incumbency (through, for example, regular elections) should automatically reduce the average ages of elected leaders. We likewise can expect political systems in which leaders are elected, but cannot be removed, to produce older leaders than systems in which leaders self-select into leadership positions. A quick glance at history lends some credence to the intuition that leaders selected for life tend to be fairly old: For example, in Venice, most elected officials were selected for fixed terms of office except for one who was elected for life, namely the Doge (Tullock 2002). As a consequence, while European kings started their reigns at age 33, on average, the Doge’s mean age at the beginning of his reign was about 68 (Eisner 2011).2 In the contemporary period, the Vatican is an instructive case of a political system in which the leader is elected for life. While the pope is chosen by voting in the College of Cardinals, no mechanisms other than his death are available to dismiss him. Therefore, since 1503, the average age of popes on the day of their election is 64. In particular, the last two popes prior to Francis were more than 75 years old at the time of their elections, which reflects the increase in average life expectancy. No elective democracy is so gerontocratic.3 Electing old popes limits incumbency (the average length of pontificate service is 10 years), which approaches that of elected democratic leaders.

The case of the Vatican is emblematic because elections are the official procedure for selecting the leader of the Catholic Church. As noted above, systems in which individuals self-select into leadership positions—as is the case in many contemporary dictatorships—cannot be expected to be as gerontocratic, since there are no ways of preventing a young dictator from taking over power.4 But many examples still approximate the Vatican: For example, Bialer (1982) and Tullock (1987, p. 159) pointed to a similar tendency to move toward gerontocracy in the former USSR, which political system provided a fairly routinized procedure for selecting the General Secretary of the Party, but none to dismiss him. Indeed, while Josef Stalin was fairly young (44 years) when he accessed office, the age of the following General Secretaries rose continuously and peaked with Chernenko (aged 73 as he entered office). The average age of Stalin’s successors at the time they took office was 62 years. By contrast, Mexican presidents in the PRI era (from 1933 to 2000) took office at the mean age of 48 years. The USSR and the PRI regime were both typical party regimes (see Geddes 1999), but the PRI dictatorship displayed an interesting feature: A constitutional reform in 1933 introduced a fixed 6-year presidential term without the possibility of reelection. The truncated expected tenures of Mexican leaders reduced the party’s incentive to select aging leaders.

Thus, we argue that selecting aging leaders is not necessarily a consequence of peoples’ genuine preferences or expectations about the competencies of their rulers, but is simply a way for the selectorate to keep some control on the system. Other empirical consequences can be derived from this hypothesis: For example, high-ranked civil servants in democracies are likely to be considerably older than elected officials.5 In general, nonremovable offices are likely to be held by old people. The main testable consequence of this hypothesis, however, is that, in political regimes deprived of an institutional way of dismissing leaders, the age of leaders will be closer to—or even exceed—the general population’s life expectancy.

Thus far, the impact of political rules on leaders’ ages never has been investigated. While leaders are considered to be one of the main units of analysis in some fields of political science, such as political economy and international relations, current research focuses almost exclusively on the impact of leaders’ individual traits on their own behaviors in office, including studies centered on leaders’ ages (Horowitz et al. 2005; Potter 2007; Bak and Palmer 2010; Walter and Scheibe 2013). The argument presented herein departs from prior research on age and leadership to the extent that it investigates the impact of leaders’ ages on the expectations and strategic behavior of the people ruled by those leaders, rather than on the leaders’ own behaviors. The personal characteristics of leaders indubitably influence their activities while in office—like the initiation of international conflicts, for example—but they also influence the choices of people involved in their selection.

In the next section, we show how this research contributes to current scholarship on leaders’ personal characteristics in comparative politics and international relations. We then develop our main argument and derive some testable implications of those hypotheses. Section 4 provides an overview on the data and methods. Section 5 presents the results. Final section concludes.

2 Prior research: political selection, leaders’ characteristics and policy performances

Our article builds on two findings that have emerged in the recent empirical literature. First, leaders matter: The personalities and socio-demographic profiles of decision makers are likely to affect their policy choices and, hence, a wide range of outcomes in the political and economic spheres. Second, the type of political system, and especially the procedures for choosing or removing the leader, influence the kinds of individuals who are selected for office.

The idea that leaders’ personality traits and ruling styles are partly shaped by the rules that brought them to power is far from new. Most of the eighteenth century thought on representative government builds on the assumption that elections are the best way of selecting wiser and more capable leaders (for a discussion, see Manin 1996). Few democracy theorists contest this conventional wisdom. Schumpeter (1942) is one exception: Elections, he argues, tend to confer advantages on professional politicians—who do not necessarily make good statesmen—to the detriment of individuals with more specialized knowledge and experience in the running of state affairs. Yet, Schumpeter still acknowledges that elections are effective ways of preventing completely inept leaders from accessing power: “There are after all many rocks in the stream that carries politicians to national office which are not entirely ineffective in barring the progress of the moron or the windbag,” he concludes. Curiously, however, Tullock (1987, p. 18) applies the same argument to autocrats: The ability to grasp and maintain a position of leadership in unstable systems plagued by political violence constitutes by itself an indication of good tactical skills.

Which institutional procedures allow for the selection of the “best” leaders remains to date an open question, but contemporary scholarship lends some credence to the idea that different political systems tend to select different types of leaders, which, in turn, affects policy outcomes. Selection effects and their consequences have been gauged predominantly in terms of economic performance, beginning with Jones and Olken’s (2005) pioneering study of the effects of leadership on growth. Besley et al. (2011) provide further evidence that the quality of leadership affects economic performance: More educated leaders achieve higher growth rates (however, see Carnes and Lupu 2016). Relatedly, a growing body of work examines how the social and professional backgrounds of decision makers in the economic realm—like, for example, working-class background, training in economics, or prior experience in the business sector—influences their policy performances (Dreher et al. 2009; Hayo and Neumayer 2014, 2016; Joachimsen and Thomasius 2014). Another field in which leaders have gained attention is in the realm of international relations (IR), with recent studies showing that leaders’ socio-demographic characteristics and professional backgrounds exert a nonnegligible influence on their foreign policy choices: leaders who are more risk-averse (Gallagher and Allen 2014) and more experienced (Potter 2007; Bak and Palmer 2010; Calin and Prins 2015) are less likely to get involved in interstate disputes. With regard to leaders’ backgrounds, ample evidence now exists that prior experience in the military influences the way decision makers think about national security affairs (Gelpi and Feaver 2004; Horowitz and Stam 2014; Stadelmann et al. 2015).

Empirical research also confirms the idea that the qualities of politicians, in turn, depend on institutions that regulate access to political office. Besley and Reynal-Querol (2011) find that democracies tend to select more highly educated leaders. Subsequent research shows that education levels of politicians are also affected by electoral rules such as the size of districts (Beath et al. 2016) or the existence of gender quotas (Baltrunaite et al. 2014). With regard to dictatorships, Baturo (2016) finds that autocrats are more likely to have military backgrounds than democrats, have shorter political careers prior to assuming office, are less likely to stem from middle-class families and have, on average, lower levels of educational attainment. Research also finds that leaders who came to power in the aftermath of a coup, a civil war or a revolution are more likely to display high tolerance for risk and a certain proclivity for violence, which in turns impact their propensity to use force against opponents or other states (Colgan 2013; Horowitz and Stam 2014; Kim 2017).

All in all, these findings generate a crucial implication: Not only are the socio-demographic traits of leaders related to their policy choices, voters and party cadres (and whoever is involved in the selection of the ruler) are also perfectly aware of this relationship and use these characteristics as proxies for assessing their future performances in office. Of course, voters can err: for example, the empirical record so far suggests that female politicians find it harder to access power, although they tend to perform better while in office (Anzia and Berry 2011; Jayasuriya and Burke 2013; Volden et al. 2013; Brollo and Troiano 2016). This strand of research furthermore shows that decision-makers can be chosen according to their socio-demographic profiles, but for reasons unrelated to their expected performance (see, e.g., Mansbridge 1999)—a consideration that will be given particular attention here.

Age, as a generic socio-demographic characteristic of leaders, has received comparatively scant attention in political science and IR scholarship, but some recent works point toward a clear trend: Generally, elderly leaders produce bad outcomes. First, the age of the leader correlates negatively with economic growth (Jong-A-Pin and Mierau 2011). Second, older leaders are more likely to initiate and escalate interstate disputes (Horowitz et al. 2005). In political science scholarship, the adverse effects of ageing generally are attributed to the shorter time horizons of elder leaders, although alternative, psychological explanations are widespread in leaderships studies (see, e.g., Walter and Scheibe 2013).

Thus, given the generally poor performances of aging leaders, a clear incentive to select relatively young individuals for office should be observed. This, however, begs the question of why gerontocracy was so widespread among traditional societies and why many contemporary polities keep producing elder leaders. Several possible ways of making sense of this paradox are possible. A first explanation is that older leaders mistakenly are considered to be more effective or trustworthy by voters; however, research tends to show that voters’ preferences for older leaders relative to younger ones varies across contexts (Spisak 2012; Spisak et al. 2014). A second possible explanation for the persistence of gerontocracy is that most elder leaders are not chosen by the voters (or by any other “selectorate”), but self-select into office instead (i.e., seize power by force). Yet, this would imply that older individuals are more likely to attempt coups or rebellions, which is somewhat at odds with the fact that risk aversion tends to increase with age (Deakin et al. 2004; Albert and Duffy 2012). Finally, gerontocracy might simply reflect a deliberate attempt on the part of the leader’s ruling coalition to reduce his/her6 expected tenure in office when he/she is elected for life. This might, according to some authors, explain why the age of church leaders varies across religions (Wittman 2014) and why durable autocracies—such as the People’s Republic of China or the former USSR—display certain tendencies to move toward gerontocracy in the long run (Tullock 1987, p. 159).

This paper offers a systematic test of the latter hypothesis, which is not provided by the literature. Specifically, we will show that the ages of leaders depend on two institutional factors: whether they are chosen by consent, and whether they can be removed by the individuals who elevated them to power.

3 Strategic gerontocracy: theory and hypotheses

Strategic gerontocracy is defined as a selection of aging leaders based on premeditated considerations. It is opposed to “taste gerontocracy”, which is based on a genuine preference for old leaders, and to “skills gerontocracy” based on specific skills acquired with age—such as social ties or personal experience—which favor elder leaders’ accession to power.

The main difference between strategic and skills gerontocracy is that the former appears only when the leader is chosen by a selectorate. Under a strategic selectorate, gerontocracy should develop in peaceful situations in which several candidates compete for leadership, only one of whom is chosen. In contrast, skills gerontocracy should also be observed when leaders seize power unilaterally, for example, during wars, coups or revolutions.

Like taste gerontocracy (but unlike skills gerontocracy), strategic gerontocracy assumes that leaders’ socio-demographic traits affect their policies and that people involved in the selection of the ruler are aware of this effect. The difference between both mechanisms is that, whereas taste gerontocracy is unconditional, strategic gerontocracy depends on whether selected leaders are removable through regular and transparent procedures. Indeed, absent such institutional mechanisms, young leaders have the obvious drawback that they can stay in power for decades, which can have disastrous consequences if they happen to be incompetent.

Strategic gerontocracy should thus be relatively common in authoritarian regimes, because most of those regimes do not provide established procedures allowing the elites to get rid of an unwanted leader: in nondemocratic regimes, most authoritarian rulers either die in office or have to be forced out of power by the military, rebel groups, or the regime’s selectorate (Svolik 2012). Importantly, irregular leadership transitions are often costly for the regime’s elites themselves: according to Kendall-Taylor and Frantz (2016), coups bring about regime collapses within a 5-years span in almost 70% of the cases. Thus, from the perspective of the ruling coalition, coups are counterproductive ways of limiting incumbency. Expecting that the dictator ultimately will retire is no solution either: autocrats who step down voluntarily are rare enough that their allies do not want to count on this possibility. The only remaining option is waiting that the leader eventually dies. That option also is the safest one, since the natural death of a dictator is one of the least destabilizing forms of leadership transition (Kendall-Taylor and Frantz 2016). Therefore, nondemocratic ruling coalitions, knowing that they have no reliable way of gauging the leader’s future behavior while in office and that they might not be able to dismiss him, have an incentive to adopt a minimax strategy and to limit the leader’s tenure ex ante: that is, when given the choice, they are more likely to pick a leader who can be expected to die relatively soon.

Thus, we should generally expect autocrats to be on average older than democrats, because the latter can be removed from power without violence (Przeworski 1999). Yet, whereas democracy by definition precludes the possibility of lifelong tenure (see Dahl 1989; Cheibub et al. 2010), nondemocratic regimes are more diverse with respect to the rules under which leaders enter and leave office.

On the one hand, nondemocratic leaders are not always chosen by consent, which makes the strategic gerontocracy hypothesis inapplicable. First, many authoritarian leaders seize power by force: in such cases, their age at the time they rise to office should be essentially random. It is noteworthy, however, that if the takeover is collective and no clear leader emerges within the group of plotters, strategic gerontocracy might still appear. For instance, General Augusto Pinochet was the oldest member of the four-member junta that ruled Chile from 1973 to 1990, which might explain why he was chosen by his co-conspirators to lead the new regime although his contribution to the coup was minimal.7 Second, in many (but not all) monarchies, institutional procedures, such as passing the crown to the monarch’s oldest child, automatically select the new leader: in those cases, strategic gerontocracy does not appear because no one has a say in the selection of the new ruler. Note, however, that although hereditary succession can allow very young leaders to claim power, the gradual institutionalization of primogeniture during the Middle Age (see Kokkonen and Sundell 2014) can be viewed as an attempt to implement strategic gerontocracy within the more general framework of hereditary succession. Indeed, while hereditary succession is known to have stabilizing functions (Tullock 1987; Kurrild-Klitgaard 2000; Kokkonen and Sundell 2014), it is less obvious why the specific institution of primogeniture was so widespread among European monarchies. The theory we propose might explain why priority frequently was given to the oldest son and virtually never to the monarch’s youngest child: whereas the second option would have had the advantage of delivering more stability by delaying the incumbent monarch’s death and the possible succession crises it could bring about, that solution had the drawback of lengthening his expected tenure in office.

On the other hand, the strategic gerontocracy hypothesis does not apply to nondemocratic regimes that display formal or informal rules limiting incumbency and preventing lifelong tenure. These cases are relatively rare, as formal term limits in nondemocratic regimes often are not enforced—when they exist at all. Historical evidence indeed suggests that dictators rarely leave office as a consequence of formal term limits.8 But some well-documented exceptions exist, like the Mexican case discussed in the introduction. Likewise, the People’s Republic of China recently introduced formal limitations on incumbency, and might turn into another possible exception to the strategic gerontocracy hypothesis (Svolik 2012; Koellner 2013). Interestingly, the growing institutionalization of the regime makes the Chinese case a perfect illustration of the mechanisms we propose. Whereas Mao Zedong and his successor Hua Guofeng were in their fifties as they seized power, China’s leaders became markedly older as the succession process became more regulated: Deng Xiaoping and his successor, Jiang Zemin, where both over 70 when they took office. But two rules, implemented during the 2000s, ended the possibility of lifelong tenure: the enforcement of the age limits for party leaders and the two-term limit for presidents (Koellner 2013). In consequence, the mean age of Chinese leaders began to decline again, reaching 61 for Hu Jintao and 59 for Xi Jinping.

To summarize our argument, we should expect systems in which the leader is selected by consent, but cannot be removed from office, to produce older leaders than systems in which leaders self-select into positions of power; but introducing a fixed term in office and regular elections in such systems should reduce the average age of elected leaders. Our main hypothesis is therefore:


In regimes in which (1) no institutional mechanism for removing a leader is available and (2) leaders are chosen by consent, leaders are on average older than in other regimes (democracies or regimes in which the leader takes power irregularly).

According to condition (1), we expect dictators to be older on average than democrats, who can be dismissed after their elected terms in office. Of course, democratic leaders need time to be well known by a large electorate and, for this reason, the difference could be small. However, in spite of this democratic constraint, we could expect that:


Autocrats should be older than democrats.

According to condition (2), we could expect that if no one has a say in the selection of the leader—or, to put it in more concrete terms, if the dictator has seized power by force—his age at the time he accesses office should essentially be random. However, reasons exist to believe that “illegal” dictators should generally be relatively young, since staging a coup or mounting an insurgency requires a certain amount of risk tolerance. Whatever the reason, we expect that:


Legal autocrats should be older than illegal autocrats.

Finally, we do not have expectations on the average age of accession to power of democratic leaders compared to illegal autocrats. In both cases, gerontocracy confers no strategic advantages, but for different reasons: democrats because they can be dismissed, illegal autocrat because they are not selected. The skills-based argument for gerontocracy discussed above might suggest that democracies are more gerontocratic than other regimes. But that possibility goes beyond our argument.

4 Research design and variable measurement

4.1 The dataset

We test our hypotheses on a cross-sectional time-series dataset with 165 countries for the period from 1960 to 2008. Our unit of analysis is the leader: we thus have a total of 1582 cases, but the number of observations in each model varies according to the availability of control variables. Each leader is recorded in the dataset as of the year he accesses office. Most data on leaders—such as their ages or the years they assumed power—are taken from Archigos (Goemans et al. 2009).

4.2 Outcome variable

Our aim is to predict gerontocracy in contemporary regimes. We operationalize gerontocracy in three ways, each of which has its advantages and drawbacks. First, we simply use the ages of leaders at the time they take office (Leaders’ age). However, even controlling for life expectancy, this measure does not take into account variations in leaders’ life expectancies across contexts.

Second, we use the age of each leader when he acceded to power minus the life expectancy in his country (UN Population Division 2015): we label this variable Relative age. The main advantage of this measure is that it allows us to correct for the problems associated with Leaders’ age; its main drawback is that the whole population’s life expectancy does not perfectly reflect the life expectancies of political elites.

Third, we compute a variable we call Lifespan, which captures the number of years each leader lived after his first year in office (i.e., the difference in years between each leader’s age at the time of his death and his age when he assumed office). This measure allows us to better capture the specific life expectancy of political elites, and to determine whether dictators actually rule longer when they self-select. However, the main issue is that the number of observations shrinks to only 669 over the period from 1960 to 2008. The Archigos dataset contains some missing values, but the reduction in the sample size mostly is because many leaders (or former leaders) are still alive. Moreover, it does not take into account the fact that not all leaders die of natural death: some of them—such as Chile’s Salvador Allende—are assassinated while in office, which might artificially deflate the values of Lifespan.

The correlation between the three output variables is significant at the 1% level. For Relative age and Leaders’ age the Pearson coefficient is r = 0.56. The correlation is −0.34 between Leaders’ age and Lifespan, and −0.18 between Relative age and Lifespan. Summary statistics for the three variables are displayed in Table 1.
Table 1

Summary statistics

Continuous variables

Dummy variables


Mean (SE)



N (%)

Leaders’ age

53.99 (0.24)



751 (42.1%)

Life Expectancy

64.41 (0,29)


Regular entry

1698 (82.4)


17.7 (11.5)


Second + Term

278 (13.5%)

GDP per capita

5765.47 (148.62)



187 (9.1%)

Population (log)

8,98 (0.04)



65 (3.6%)

Political experience

12.22 (0.28)


Regular exit

1379 (73%)

We perform our main analysis using Relative age, for three reasons: first, compared to Lifespan, it better captures what we want to measure, namely the expected tenures of leaders rather than the number of years they actually spend in office. Second, it allows us to maximize the number of valid observations. Third, compared to Leader’s age, it is closer to be normally distributed (see Fig. 1 below). However, we also provide results based on the two other measures to assess the robustness of our findings.
Fig. 1

Distribution of leaders’ relative age

Figure 1 shows the distribution of relative age, slightly skewed to the right. On average, leaders’ age is 10.7 years less than the national population’s life expectancy. The median value is −11.8 years. Yet, some extreme cases are observed in which the age of leaders first entering office exceeds by more than 20 years the national average life expectancy.

Interestingly, on average, democratic leaders can expect to live almost 15 years after their first accession to power, while autocratic leaders can expect to live only 5 years more, which equals the duration of one term in many democratic countries. The tendency is different when we observe the age of the leader at entry, which is slightly higher (55 vs. 51) in democracy than in dictatorship, because in democracies, life expectancy is much longer (+13 years on average). At this stage, it is difficult to conclude that gerontocracy is more widespread in dictatorships.

4.3 Explanatory variables

Two conditions trigger strategic gerontocracy: leaders are selected by consent, but not removable once chosen. The second feature fits well with the distinction between democracy and dictatorship, since democracy entails that leaders are elected temporarily and can be removed from power after their terms in office expire (see Przeworski 1999). We thus use the binary democracy measure from Cheibub et al.’s (2010) Democracy and Dictatorship dataset, which is based on Przeworski’s minimal approach to defining democracy and explicitly takes into account whether an incumbent ever has ceded power after losing an election as an essential criterion for a regime to be classified as democracy. This rule thus automatically precludes the possibility of lifelong tenure in democratic regimes. In turn, as discussed above, dictatorships almost always lack formal mechanisms for dismissing rulers. We acknowledge that dictatorship and the absence of fixed terms in office do not coincide perfectly; however, exceptions are rare enough to validate the distinction between democracy/dictatorship as a trustworthy measure of the presence/absence of a procedure for dismissing the leader; furthermore, potential measurement errors would actually bias estimates against our hypotheses (because they would reduce the mean age of dictators selected by consent). The dummy variable we use is called Dictatorship9 (see Table 1).

The presence of a procedure for selecting a leader is measured by a variable indicating whether the leader has accessed power in a regular or irregular way, taken from the Archigos dataset (Goemans et al. 2009). We assume that regular, peaceful leadership transitions indicate the presence of a selectorate, while irregular leadership changes constitute evidence for self-selection. Note that in democracies, it is very rare that the leader accesses power illegally (2.3% of the cases). Such cases include, for example, situations in which democratization is led by someone who is not elected, like Siles Zuazo in Bolivia. In contrast, in authoritarian regimes, 42% of the leaders have accessed power in an irregular way. Therefore, the conditions in which strategic gerontocracy appears—regular entry and no procedure for dismissing the leader—are fulfilled in around one observation out of four.

4.4 Control variables

We add some control variables whose omission might bias our estimations. First, monarchy might act as a negative confounder, since monarchs access power in a regular way but can still be very young at the time they take office. We thus add one dummy variable taken from Cheibub et al.’s (2010) Democracy and Dictatorship dataset. The variable Monarchy applies to a few cases, but reduces the number of observations in which strategic gerontocracy is likely to appear (1 out of five, after controlling for Monarchy).

Second, we add a dummy variable indicating whether an interstate or intrastate armed conflict is taking place, since the incidence of a war may artificially reduce life expectancy in a country. The variable is built from the UCDP/PRIO Armed Conflict dataset (Gleditsch et al. 2002; Melander et al. 2016). We take only major wars into account, i.e., conflicts that result in at least 1000 battle-related fatalities per year.

Third, some leaders assume office several times in non-consecutive terms (and thus are reported several times in our dataset), which makes them mechanically older when they assume office for the second or third time. To capture this effect, we include a dummy variable, called Second + Term, that takes on the value of 1 if the leader has been in office previously (about 14% of the observations), 0 otherwise. This phenomenon overwhelmingly concerns democracies–when, for example, a former prime minister is elected again in a parliamentary regime—but it also happens in dictatorships: for example, during the 2000s, King Gyanendra of Nepal dismissed three regularly elected prime ministers and either replaced them with an appointed government or assumed power himself, which explains why he is recorded several times as holding non-consecutive terms.

Our last set of controls is meant to rule out an alternative hypothesis: even if the empirical analysis demonstrates that dictators who take office in a regular way are older than coup leaders, this does not necessarily confirm our argument, since regular entry might raise dictators’ ages for reasons unrelated to strategic selection. It is, for example, possible that dictators who are chosen by consent are older because they had to climb the ladders of a bureaucratic, hierarchically structured organization in which leadership positions are assigned according to the seniority principle. Svolik (2012) points to the hierarchical assignment of services and benefits in party dictatorships, but this explanation might as well apply to military dictatorships or, generally, to every regime that relies on a bureaucratic organization. In such regimes, leaders are more likely to be drawn from a pool of older contenders; conversely, coups plotters are more likely to be found among younger members of the organization, since their incentive to seize power by force should be proportional to the time they otherwise would have to wait until they can reach leadership positions in a regular way.

To control for this alternative explanation, we add a variable that records each leader’s past experience in politics, measured in years. If our hypothesis is correct, years of experience should increase the leader’s age, but the effect of regular entry should still be statistically significant. This variable—called Political experience—is drawn from the Cursus Honorum dataset (Baturo 2016). Table 1 displays summary statistics for Political experience. Interestingly, the prior experience of democratic leaders is almost twice that of dictators (15 and 8 years, respectively), which likely is a consequence of the sizes of the selectorate: in democracies, a long career in politics is a de facto requirement for reaching the highest office, which mechanically increases leaders’ ages. This variable is a proxy of the skills gerontocracy we discussed above.

We also include several control variables that allow us to disentangle the effects of seniority principle and strategic gerontocracy: we include the natural logarithm of total population (UN Population Division 2015), because we suspect that political careers may be longer in more populous countries, due to the more vigorous competition for leadership. We also control for countries’ GDPs per capita (World Bank 2016), because we expect a negative relationship between GDP per capita and life expectancy (which is subtracted from leaders’ ages in our output variable): this may artificially reduce our measured dependent variable in wealthy countries.

4.5 Estimations

Given the continuous and quasi-normally distributed outcome variable, we estimate OLS regressions. Including country and year fixed effects is both the most conservative and the most appropriate strategy, because explanatory variables are not time-invariant and because random effects necessitate stronger assumptions about the lack of relationship between the explanatory variables and the specific effects.

We run four models. Models directly testing H1 and H1b include an interaction term in order to capture the joint effect of dictatorship and regular entry, which measure, respectively, the absence of a procedure for dismissing the leader, and the fact that a leader has been selected legally. Two models are estimated: the first—the main model—includes all of our explanatory variables, and the second includes only our independent variables of interest as well as country and year fixed effects. The aim of the second model is to make sure that our estimations do not produce spurious relationships owing to collinearity between explanatory variables: in particular, the correlations between dictatorship, GDP per capita and formal experience are relatively high. Two other models are reported, which estimate the effect of dictatorship on relative age. The first uses the whole sample and allows us to assess how the absence of mechanisms for removing leaders impact their ages. The second is based on the subsample of observations for which leaders have accessed office regularly. This subsample includes 82% of the observations, which we consider sufficient to offer a relatively reliable result.

We then perform several additional analyses. First, we check the robustness of our main model in two ways: we re-run it after excluding outliers that skew the output variable to the right, corresponding to leaders whose ages exceed the population’s life expectancy by 20 years or more (28 observations). We also take into account the fact that our data include many country units, but relatively few time periods (49); thus, we perform an OLS regression that includes only fixed year effects, clustering coefficients’ standard errors by country.

Second, we check whether our results are sensitive to the measurement of gerontocracy. We thus re-run the main model using Leaders’ age as the output variable, with and without life expectancy as a control variable. Finally, we provide tests based on Lifespan. In order to take into account the fact that some leaders are more likely to die prematurely, we control for the way they exited office: we use the variable Regular exit (described in Table 1)10 as a control and a filter variable.

5 Results

5.1 Main results

Table 2 shows our main results, based on Relative age. The explanatory power of our models is high: the explained variance ranges from 62 to 73%. Coefficients are stable across models. Second + Term and political experience have the expected effects, while population, war and monarchy do not have any statistically significant impact. GDP per capita has a (negative) impact on relative age only in Model 4, which uses the subsample of legally appointed leaders.
Table 2

Impact of dictatorship and regular entry on leaders’ relative age (OLS regression)


Model 1

Model 2

Model 3

Model 4











Regular entry









Regular # dictatorship


















Population (log)










Second + Term










Political experience


















































Standard errors in parentheses

*** p < 0.01; ** p < 0.05; * p < 0.1

Model 1 answers the question of whether dictators’ life expectancies are less than those of democrats. The answer is no. The coefficient associated with Dictatorship has the expected sign but is small and insignificant. Note also that Regular entry is insignificant.

Model 2 adds the interaction term, which is statistically significant, has the expected sign and raises the explained variance slightly. Thus, regularly appointed dictators are much closer to death than dictators who seized power illegally, and also closer to death than democrats (by about 2 years). This result also confirms that democrats are older than irregular dictators. All in all, the interaction term shows that the most gerontocratic systems have leaders regularly chosen by a selectorate but non-removable, which is consistent with our strategic gerontocracy hypothesis.

Model 3, which includes only our main predictors and country and year fixed effects, confirms previous results. The impact of the interaction term is smaller and, as expected, the explained variance declines. However, the similarity of the results highlights that the coefficient on the interaction term is not sensitive to the inclusion or exclusion of other independent variables.

Finally, Model 4 uses the subsample of leaders who have taken power in a regular manner. The model confirms that dictators have shorter life expectancies, since they are more than 2 years closer to death than democrats are. Note, however, that the variable is significant only at the 10% level. Interestingly, it loses its significance altogether when we remove political experience from the model (results not reported). Political experience is a way to control for skills gerontocracy, which is more prevalent in democracies. Therefore, we can conclude that the effect of skills gerontocracy compensates for that of the strategic gerontocracy which exists in dictatorships with regularly appointed leaders.

5.2 Robustness checks

We perform four first robustness checks in order to ask whether our results are sensitive to the measurement of the dependent variable (Models 5 and 8), to the inclusion or removal of extreme values (Model 6) and to model specification (Model 7). The coefficients of our main independent variables for Models 5, 6 and 7 are plotted in Fig. 2 (full models can be seen in the online appendix).
Fig. 2

Three alternative models to estimate the impact of regime and regular entry on gerontocracy

First, when we use Leaders’ age as an output variable and control for life expectancy (Model 5), the previous results hold. The interaction term continues to be significant, even though its impact declines slightly. We also re-run Model 5 and remove the country’s average life expectancy from the set of control variables, in order to check the sensitivity of our results (Model 8). The estimated coefficients do not change fundamentally.

Second, when leaders whose age exceeds life expectancy by more than 20 years are excluded (Model 6), the results hold and the coefficient on the interaction term is even larger. Dictators who entered power regularly are now 3 years closer to death than democrats.

Finally, we run a model that includes only year-fixed effects, with standards errors clustered by country (Model 7). Under this less conservative estimation method, the results are more congruent with the strategic gerontocracy hypothesis. While the difference between democrats and illegal dictators is no longer significant, legal dictators now are 7 years closer to death.

To sum up, all models show that gerontocracy increases under two joint conditions: when leaders are selected legally, and when they cannot be removed once in power.

5.3 Further analysis: is strategic gerontocracy effective?

Previous analyses show that ruling coalitions in dictatorships tend to select leaders who display relatively short life expectancies. However, nothing in our results so far indicates whether this strategy is effective, i.e., whether dictators’ actual lifespans from the moment they assume power is shorter when they are selected by consent. The variable Lifespan allows us to verify this.

Two caveats are in order. First, as discussed above, the number of observations lost with this variable is considerable. One of the consequences is that we cannot include country and year fixed effects in our estimates. We thus simply adjust coefficients’ standard errors for country clusters.

Second, this measure is problematic because, for reasons unrelated to strategic selection, some leaders die of non-natural causes and reasons exist for thinking that this risk is not distributed equally across political systems. In particular, leaders who are removed from power by force are more likely either to get killed in the process or to suffer some form of post-tenure punishment that somehow increases their odds of dying prematurely (Goemans et al. 2009). A simple bivariate analysis reveals that leaders who lost power regularly are on average 5 years older than the others at the time of their death, this result being significant at the 1% level. This difference in longevity, while not quite offsetting the effects of strategic gerontocracy, might still be strong enough to close the gap between legally appointed dictators and the other dictators on the values of the new output variable, because the latter are more likely to be removed from power by force.11

Model 9 (see online Appendix) confirms these expectations: no sizeable difference is found between democrats and both categories of dictators. The coefficients associated with the three independent variables are small (although correctly signed) and do not come close to conventional levels of statistical significance. Regular exit, on the other hand, increases leaders’ average lifespan by 6.5 years (p < 0.001). Years of prior political experience and population size, as expected, significantly reduce leaders’ lifespans.

The results are quite different, however, if we exclude from the sample all leaders whose lifespans are truncated—that is, leaders who were overthrown—which allows us to eliminate the bias discussed above. Model 10 is performed on the subsample of leaders who left office in a regular way, with all controls included. Our main predictors are all correctly signed and statistically significant, although the coefficient on dictatorship is significant only at the 10% level: democrats and dictators who seized power illegally enjoy longer lifespans after assuming office than legally selected dictators. The coefficient on the interaction term is highly significant and indicates that regular access to power in a nondemocratic country reduces leaders’ lifespans by 3 years, on average. These results also hold when we remove all controls, which increases the sample size to 609 observations (Model 11). In that model, regularly appointed dictators are still significantly closer to death at the time they assume office. Thus, some tentative evidence exists that strategic gerontocracy is effective under certain conditions: ruling coalitions in dictatorships not only tend to select older leaders, but also succeed in picking up individuals who actually will die within a shorter time period.

6 Conclusion

While gerontocratic rules for selecting leaders have been largely widespread in human history, surprisingly the impact of political institutions on leaders’ age has never been investigated. This article analyzes why age can be a relevant factor for choosing leaders. We address the question of why people often prefer to be ruled by older leaders. In short, our answer is: because they die quickly.

However, age is relevant only if two conditions are fulfilled: first, a more or less large group is able to choose the leader; second, it is impossible to dismiss the chosen leader. Under these conditions, the younger the leader, the higher the cost of making the wrong choice. We argue that choosing the chief from among the oldest men—which is a rule shared by many tribal societies—is a consequence of the possibility of selecting the leader by consent and the impossibility of dismissing him.

Democratic systems have kept the selection by consent, but they have introduced a term-limiting procedure to dismiss leaders after a predetermined number of years in office. Therefore, the problem of getting rid of bad rulers is solved without having to resort to gerontocracy. Gerontocracy is also absent from many nondemocratic systems because individuals self-select into leadership positions instead of being chosen: in those cases, the eldest candidates have no particular advantages.

This article compares the latter two situations to those in which leaders are selected legally, but cannot be dismissed once in power: these two joint conditions trigger the mechanism we called “strategic gerontocracy.” Based on a statistical analysis on a dataset that covers the period from 1960 to 2008, we find that the life expectancy of leaders is shorter in dictatorships than in democratic systems, except when the dictator takes power in a non-consensual way.

This result shows that gerontocracy is not only a consequence of social ties or personal characteristics acquired with age, because it is more prevalent in specific political systems. We also observed that long political experience is much more useful in democracy to become the leader. That is why the observed difference in life expectancy between democrats and legal dictators is not very large. Based on our results, we argue that, while democracies are still ruled by old leaders, the mechanisms that drive gerontocracy are different from one political system to another. In a democratic competition, candidates should be rather old on average, but age does not influence the probability of winning. In contrast, in dictatorships, peaceful leadership transitions increase the odds of success of the oldest candidate.


We borrow the expression from Bueno de Mesquita et al. (2003).


Means are calculated for the period from AD 600 to 1800.


For example, since 1789, US presidents have been elected, on average, at age 55.


Indeed, coup plotters can be very young, like for example Sierra Leone's Valentine Strasser (who became at age 25 the world's youngest head of state as he took over power in 1992).


If this explanation is correct, it provides a justification for the seniority principle that regulates promotions in most public administrations.


In the remainder of the paper and for the sake of simplicity, we will deliberately refrain from using gender-neutral formulations while referring to leaders: Women remain relatively rare among elected leaders, and dictators almost always are men.


Interestingly, observers have often wondered about the reasons behind this choice. According to Geddes (1999, p. 123), the coup plotters deliberately selected him because he was known for his adherence to rules and lack of charisma, and they mistakenly believed that those traits would prevent him from consolidating his personal power. Another of Pinochet's personal characteristics – his age – is an alternative explanation: the three other members may have anticipated the possibility of an attempt to usurp power and identified a way of limiting the duration of his personal rule.


The most comprehensive source to date (Svolik 2012) records 50 such instances out of approximately 700 cases for the period from 1946 to 2008.


The original variable (“Democracy”) takes the values of 1 for democracies and 0 for dictatorships. To ease the interpretation of the results, we recoded the variable, the value of 1 now indicating dictatorship.


Data are drawn from Archigos’ variable “exit”. That category includes defeat in elections, term limits and voluntary retirement.


The data confirm that legally appointed leaders are significantly less likely to be overthrown. They also live significantly longer, even when life expectancy is controlled for.



The authors would like to thank this journal’s editors and one anonymous reviewer for their excellent suggestions. This paper was presented at the international conference on the “Political Economy of Democracy and Dictatorship” in Münster, Feb. 2017, and greatly benefited from the comments of the participants. The authors also wish to thank Abel François for his advices, and Alexander Baturo for generously sharing his data.

Compliance with ethical standards

Conflict of interest


Supplementary material

11127_2017_449_MOESM1_ESM.docx (18 kb)
Supplementary material 1 (DOCX 17 kb)

Copyright information

© Springer Science+Business Media New York 2017

Authors and Affiliations

  1. 1.Sciences Po Grenoble, PACTEUniv. Grenoble AlpesSaint-Martin-d’HèresFrance
  2. 2.Sciences Po BordeauxCentre Emile DurkheimPessacFrance

Personalised recommendations