The Half-life of Leaders and the Half-life of Regimes

Thinking back on the last couple of posts, a couple of questions arise naturally. First, there is the question of the survival of regimes in general, not just democracy: if most democracies die within 15 years or so, what is the median duration (the “half-life,” if you will: the time it takes for half of them to be gone) of other regimes? And second, there is the question of the relationship between the half-life of regimes and the half-life of leaders: do regimes whose leaders tend to have longer half-lives also have longer half-lives? My interest in these questions stems from my current research on the question of legitimacy: my sense is that legitimacy matters much less than people usually think to the survival of large-scale patterns of political power and authority, so I’m interested in trying to figure out if there are systematic differences in survival between more and less “legitimate” regimes and other political structures. So this is another exploratory post, with lots of graphs.

How do we measure the duration of non-democratic regimes relative to democratic regimes? Though democratic regimes are not always straightforward to identify, non-democratic regimes come in a much wider variety of forms – from hereditary, absolute monarchies to single party regimes and multiparty hybrids, and some of these forms shade gradually into one another over the course of many years. (For a sense of this variety, consider the differences between Mexico before the 1990s under the PRI, whose presidents succeeded each other with clockwork regularity every six years and a lively opposition existed but could never win the presidency, North Korea today, where opposition is non-existent and succession is controlled by a tiny clique, and Mubarak’s Egypt.) To get a handle on this question, I’m going to use the Polity IV dataset, which codes “authority characteristics” in all independent countries (with population greater than 500,000 people) from 1800 to 2010. (I’ve been convinced by Jay Ulfelder’s work that the DD dataset I used in my earlier post is not appropriate to study comparative regime survival due to the way it codes certain democracies where alternation in power has not occurred as dictatorships, which systematically biases the survival estimates of democracies upwards).

The Polity dataset is fairly rich. Most researchers seem to use only the composite indexes of democracy and dictatorship it offers, but these indexes, while useful, do not have a strong theoretical motivation, as Cheibub, Gandhi, and Vreeland argue here. For my purposes, it is best to use the dataset to extract those authority characteristics of political regimes it purports to measure: the mechanisms of executive recruitment, the type of political competition, and the degree of executive constraint. Mechanisms of executive recruitment include hereditary selection, hybrid forms combining hereditary and electoral mechanisms, selection by small elites, rigged elections, irregular forms of seizing power, and competitive elections; types of political competition range from the repressed (all opposition banned, as in North Korea) to the open (typical of thriving democracies); and executive constraints range from unlimited to “parity” with the legislature. (See the Polity IV codebook for a full discussion). In theory, the dataset distinguishes eight kinds of executive recruitment mechanisms, ten types of political competition, and seven degrees of executive constraint, plus three different kinds of “interruption” (including breakdowns of state authority, loss of independence, and foreign invasion and occupation), leading to a possible 563 possible patterns of political authority, but these dimensions are all highly correlated (over .99); indeed, only 212 combinations of executive recruitment, political competition, and executive constraint actually appear in the date, most of them only once and for short periods of time, and it is obvious that some combinations do not even make sense. (And those that do make sense do not always capture all the information we would normally want about a political regime: Polity has no good measure for the extent of suffrage in competitive regimes, for example). But the dataset helpfully indicates how long each of these patterns last, so we can attempt a first cut at the question of the half life of regimes using a Kaplan-Meier graph:

The half-life of an “authority pattern” – a combination of an executive recruitment mechanism, a type of political competition, and a specific form of executive constraint – is 6.6 years, though the tail of the distribution is very long: some of them have lasted for upwards of a century. Switzerland, for example, has had the same authority pattern for 162 years, and Afghanistan retained the same authority pattern from 1800 to 1935 (a hereditary monarchy). As it happens, social and political life comes to be mostly structured in most places by the long-lasting patterns, but most patterns of authority do not last that long. Incidentally, at this level of abstraction there are no great regional differences in the half-lives of authority patterns, though it does seem as if authority patterns last slightly longer in Europe and the Americas than in Africa and Asia:

Yet an “authority pattern” is too amorphous a unit of analysis. We might get a better handle on the question of comparative regime survival by looking specifically at the mechanism of executive selection, since the manner in which the chief power in the state is selected is normally thought to be quite important and to have far-reaching consequences: whether supreme power is achievable by hereditary succession only or through designation within a closed elite or via competitive elections or some other means seems to have important consequences.

Of all the mechanisms of executive selection identified in the Polity IV dataset, only one, “Competitive Elections,” is unambiguously democratic by most people’s lights. Though within the dataset the fact that a regime has competitive elections is no guarantee that it will also have universal suffrage, for the most part “competitive elections” identifies most countries that most people think are democratic. We can thus calculate the duration of all periods of “competitive elections” and compare them to the duration of all “non-democratic” periods – those periods where executive selection happened through some other means. The details are somewhat tricky (see the code), but here are the results:

Some notes. As we might have expected from the discussion in the previous post, full hereditary monarchies (Russia under the Tsars, Saudi Arabia, Iran under the Shah, Portugal and Romania in the 19^th century, Nepal in the 19^th century, among others; there are 65 episodes in 40 countries in the dataset) have the longest half-lives (nearly 32 years; this increases if we collapse the two hereditary monarchy categories. Note these are not “constitutional” monarchies like the British one). But competitive electoral regimes are no slouches, with a half-life of about 17 years (in keeping with Jay’s numbers in this post, though he uses a different dataset), and as time goes on their survival rates seem to converge with those of monarchies. Similarly, “limited elite selection regimes” (e.g., single party-communist regimes, where a narrow clique selects the leader without open competition) have a half-life comparable to that of democracies, but as time goes on they tend to break down more; their survival rates seem to diverge from those of competitive electoral and monarchical regimes. Low survival rates are found especially among political forms that appear to have internal tensions, such as competitive authoritarian regimes, where elections exist and are contested by an opposition, but it is very hard for the opposition to attain real power (e.g., Zimbabwe today). I confess I don’t really understand Polity’s “Executive-guided transition” category, but it’s obviously a regime that is turning into something else (the Pinochet regime in Chile after the 1980 referendum but before the return of competitive elections counts, for example), and “ascription plus election” includes regimes where the monarch retains some real power but the legislature and other executive offices are no longer under its thumb  (only a few are recorded in the data, including Belgium in the late 19^th century and Nepal in the 1980s and 90s); it makes sense that such regimes, halfway between “real” monarchies and purely constitutional monarchies like the British, should have short half-lives as the conflict plays out and either turn into competitive electoral regimes or into more absolute monarchies.

It is also interesting to compare the relative survival rates of competitive electoral patterns of authority vis a vis periods where selection happens by non-competitive electoral means (regardless of whether the selection means stay the same):

Though the difference seems to narrow as time passes, the half-life of non-democracy since the 19^th century has been a bit longer than the half-life of competitive electoral regimes (23 vs. 17 years). In sum, political regimes do not last much more than a generation.

(For those still following, the regional breakdown indicates that competitive electoral periods have had the longest half-lives in Europe and the Americas, whereas non-democracy has had the longest half-lives in Africa and Asia; no special surprises there, though I am not sure about the reason).  

How does this relate to the half-life of leaders? For that, we turn to the ARCHIGOS dataset by Goemans, Gleditsch, and Chiozza, which contains information about the entry and exit date of almost all political leaders of independent countries in the period 1840-2010. It’s a fantastic resource – more than 3000 leader episodes, and information on their manner of exit and entry. And the conclusion one must draw from examining it is that power is extremely hard to hold on to; a ruler’s hold on power seems to decay in an exponential manner (note I haven’t checked that the decay really is exponential in the technical sense, though I'm thinking of doing that). Over this vast span of time, covering all kinds of political regimes, the half-life of leaders is only about 2 years, or a third of the median authority pattern, as we might have expected from the previous post (though the half-life of leaders is even smaller here):

Yet of course it is the people who beat the odds – those who last much longer than the average leader – the ones who shape social and political life. (There’s an endless parade of mediocrities in the dataset, two-bit prime ministers gone after a few months of ineffectual dabbling and the like).

(But don’t some leaders come back to power after losing it? In fact, the vast majority of leaders only attain power once, and never return to power, though about 100 did manage the feat three or more times. In fact, practice does not help; survival in power only appears to decrease the more previous times the leader had been in power, though note that the uncertainty of the estimates also increases, and one might expect that age would take its toll too).

We are now in a position to extend the analysis in the post below by merging the Archigos and the Polity dataset to calculate the survival curves for leaders conditional on the pattern of executive recruitment. Though I would take these curves with a grain of salt, here are the results:

As expected from the previous post, it’s good to be king – the half-life of absolute kings is about 12 years (and it’s almost always king: there are only 41 female leaders in a 3000 case dataset). Interestingly, a similar result for the half-lives of Chinese emperors is reported here (10 years: Khmaladze,  Brownrigg, and Haywood 2010, ungated) as well as for the half-lives of Roman emperors (11 years: Khmaladze,  Brownrigg, and Haywood 2007, ungated). There is something about the deep structure of monarchies in many different periods and societies, it seems, that points to a half-life in power of about 10-13 years for monarchs. 

More generally, authoritarianism pays in terms of leader tenure, despite the fact that non-competitive regimes do not always last longer than competitive ones. The highest half-lives of leaders beyond monarchs are found in limited elite selection regimes, executive-guided transitions (where non-democratic leaders are changing the rules), and competitive authoritarian regimes; but democracies are more lasting than most of these regimes (except for monarchies; see above).

Another way of looking at this is to calculate what we might call the “personalization quotient” of a regime: divide the half-life of the leader (for a given regime) by the half-life of the regime to get an idea of the percentage of the regime half-life that a leader is expected to last. So a monarch is expected to last about 37% of the half-life his regime (31.86 / 12); this is the most intensely personalized of regimes, as one might have expected given that it is devoted to the maintenance of a family line. The next most personalized regimes are competitive authoritarian regimes (28%), “self-selection” regimes (15%), limited elite selection regimes (16%), and “executive-guided transitions” (40%; this is pretty much by definition, however, so I don't make much of them). A competitive electoral regime has a personalization quotient of 8% - an expected leader half-life of about 2, divided by an expected regime duration of about 17 years. From the point of view of such a leader, it pays to try to move towards a competitive authoritarian regime, and it pays for the leader of a limited elite selection regime to move towards a formal hereditary monarchy (as is happening, in a sense, in North Korea right now, and almost happened in Egypt and Libya). 

But are authoritarian regimes more risky, so that leaders will try to hang on to power more? We can also look at that using the archigos dataset. Though leaders in non-democratic regimes have a slightly higher risk of leaving office with their heads on pitchforks or hanging from lampposts, the vast majority leave by "regular" procedures.  

More, perhaps, could be said. I’ve been wondering, for example, about whether there is a relationship between the breakdown of particular regimes and the tenure of leaders, though I’m not sure how to go about tackling that question. From the point of view of the study of legitimacy, however, what strikes me is the general fragility of patterns of authority and rule: few patterns of authority are expected have half-lives that exceed a single generation, and most don’t last nearly as long, regardless of their “legitimation formula” – heredity, competitive elections, ideology, whatever. Of course, some beat the odds, especially some competitive regimes and some monarchies, and these shape history. But the historical evidence suggests that they are in a sense the exception rather than the rule.

Code necessary for replicating the graphs in this post, plus further ideas for analysis, here and here. You will need to download the Polity IV and ARCHIGOS datasets directly, and this file of codes from my repository.

[Update: fixed some typos,  9 Feb 2012]

Comparative Political Leader Survival, 1946-2008

After playing around with Jay Ulfelder's data on the survival of democracy in the previous post, it occurred to me that I have not seen survival estimates for leaders in different kinds of regimes like the ones he discusses for democracies. So, in the spirit of exploratory data analysis, here are some graphs using data from the DD dataset of political regimes by Cheibub, Gandhi, and Vreeland, which provides information about regime type, effective heads of government, and leadership tenure for most countries in the world for the period 1946-2008. (Fuller data and methods note at the end of the post).

First, let's look at a simple estimate of leader survival for all (effective) political leaders in all regimes in the post WWII era:

The figure shows an estimate of the proportion of leaders who are expected to still be in power after n years. So, for example, after four years in power, less than half of all leaders are expected to still be in power, and after 20 years less than 10% of all leaders are expected to still be in power; the majority of all leaders last less than 4 years in power, and the vast majority less than 5. [Update: of course, some of these leaders come back to power after a shorter or longer period out of power.] This may be easier to see if we draw the plot on a logarithmic scale:

This looks like a classic "long tail" distribution of a kind often produced by "rich get richer" processes: most leaders don't last in power very long, but those who beat the odds can do very well indeed, as power feeds on itself and leaders become increasingly difficult to dislodge. (I won't say anything about power laws for fear of attracting the ire of the statistical gods). 

Nevertheless, democratic leaders and non-democratic leaders aren't equally successful at hanging on to power:

While the median democratic leader can expect less than 3 years in power, the median autocrat can expect a bit less than 7. And the gap widens with time: less than 8% of all democratic leaders can expect to hang on to power for more than 10 years, but more than 40% of autocrats do, and no democratic leader in the sample has lasted more than 25 years in power (Lynden Pindling of the Bahamas and Eric Williams of Trinidad and Tobago; your mileage may vary as to how democratic you think they were, but that's how they are coded), whereas nearly 20% of autocrats do. This may seem obvious (after all, autocrats typically impose larger barriers to political competition than democratic leaders, and ordinary people face larger obstacles in trying to get rid of them) but it also presents a bit of a puzzle, for democracies are supposed to be more responsive to popular wishes and more legitimate, and dictators are always at risk of being overthrown by their close associates. (For one influential explanation of the observed pattern of survival by Bueno de Mesquita, Smith, Siverson, and Morrow, see here and here). The greater legitimacy of democratic leaders, and their closer connection to popular opinion (to whatever degree: let's not exaggerate, either), does not seem to translate into a surer hold on power. 

Not all autocrats do equally well; absolute monarchs are especially successful at holding on to power:

Though the uncertainty of the survival estimate is larger for monarchs than for other regimes (there are just fewer monarchs in the sample) their advantage is large enough to be noticeable above the noise: nearly 60% of all monarchs can expect to last 20 or more years in power, while only 20% of other autocratic rulers can expect to survive that long, and less than 1% of democratic leaders can hope for such a career. This is another reason to think the Middle Eastern monarchs are probably safer from being overthrown than the leaders of the "republican" regimes, as Victor Menaldo has recently argued. His argument points to specific features of the political culture of these monarchies that enable elites to better monitor and discipline leaders; but other things may be going on as well (monarchs elsewhere in the world also appear to have done well, so whatever enables monarchs in the Middle East to survive appears to also work elsewhere, though admittedly most of the world's absolute monarchs since 1946 have been concentrated in the Middle East). It is also interesting to note that military and civilian dictators do not differ (much) in terms of their survival expectations (the estimates fall within each other's 95% confidence intervals), despite theoretical and empirical work that suggests that military regimes are less stable than civilian dictatorships. (Of course, this could be due to any number of things, including problems with the coding of the data and the fact that the stability of regimes is a different thing from the stability of any given leader's grip on power).

I was also curious to see whether the survival of leaders differs across regions of the world. And at least for non-democratic leaders, that seems to be the case:

There's a lot of uncertainty in these estimates (and I could have made a mistake), but in general it seems to be the case that autocratic leaders have had less success hanging on to power in Latin America, despite the USA's not always benevolent influence in the region. That was surprising to me, so perhaps someone will tell me why this is wrong. By contrast, democratic leaders all have very similar survival expectations all over the world; no evidence of "regional" effects seems evident:

Now that I've mentioned the USA's influence, we might as well look into whether autocrats (or democratic leaders) have had more trouble hanging on to power during or after the cold war. Surprisingly, it seems they have not: leaders in both regimes had the same survival expectations in both periods. But this was tricky to figure out how to calculate, and it is the most likely spot where I might have made a mistake (see sources and methods note below):

Sources and methods. A full description of the DD dataset can be found here, including the criteria it uses for categorizing regimes as democratic or non-democratic and a general defense of its methodological approach. (It used to be possible to download it as well from that page, but the form no longer seems to be working. I've animated the dataset here.) These criteria have been criticized for a variety of reasons, but in general DD does not suffer from worse problems than many of the other common datasets of political regimes (like Polity IV or Freedom House). It is possible that some of the coding decisions they make might influence the estimates of survival presented above, e.g. because they err on the side of classifying some regimes as dictatorships that could have been considered democratic (when there has been no alternation in power). This would tend to bias downward the survival estimates of democratic leaders. At any rate, DD includes information about leaders and their tenure, which is missing in other datasets and makes the data-wrangling easier, though this information is not always complete (there is sometimes more than one leader in a year for a given country, a fact that the dataset must omit, given its country-year resolution) and is not quite in the right format for survival analysis. I thus had to reshape  it (R code and a general description of the process; rank amateurism on display). I created three data files: one for the plots of survival for all leaders and leaders by democracy/non-democracy (ddsurvival.csv); one for the plots of survival by autocratic regime type (ddsurvival2.csv); and one for the plots of survival during and after the cold war (ddcoldwar.csv). (R code for generating all plots is here). These files treat the leader spell as a case; "right censoring" occurs when the leader dies or if the leader is still in power by 2008 (see the DD codebook; the files use a variable called ecens2). Since DD does not distinguish between deaths by natural causes and political assassinations or death in revolution, this introduces a certain amount of bias; in theory, "political" deaths should not result in"censoring" of the data. I should note that the plot of survival by autocratic regime  type does not take into account some cases where "left censoring" occurs (i.e., when a regime starts before 1946), though the number of cases where that is a problem is very small. Finally, there are a small number of repeated cases in ddsurvival.csv and ddcoldwar.csv due to problems guessing the right "entry date" for the leader; these must introduce some small amount of error, though I couldn't possibly say how much or in what direction the bias would work.

[Update, 1/31/2012: Fixed minor typos]
[Update, 1/02/2012: Changed location of code and data files]