Traditional thinking about democratic consolidation tells us that the risk of democratic breakdown declines over time. This expectation is based on the idea that democracy is, at least in part, a matter of culture. Democracy is thought to depend on certain norms and values, so prospects for its survival should improve as those norms and values deepen and spread. The causal mechanisms involved are learning and habituation. In theory, the passage of time produces positive feedback, and this virtuous circle eventually carries societies to a point where a reversion to authoritarian rule is collectively unthinkable.
That pattern, however, is not what I see in the data. The figure below plots the relationship between the age of a democratic regime (on the horizontal axis) and the risk of its breakdown (on the vertical axis) in a data set on episodes of democracy in “developing” countries from the past half-century. (Here, “developing” just means countries that are not members of the Organization for Economic Cooperation and Development.) The estimates were derived with nonparametric smoothing splines, which can be used to identify nonlinear relationships without making prior assumptions about the functional form of the relationship (see Luke Keele’s book). The solid line shows the estimates, and the dotted lines show the upper and lower bounds of the associated 95% confidence interval. In addition to regime duration, the model also includes a country’s infant mortality rate (normalized to the annual global median and then logged), a measure of its civil liberties (Freedom House’s seven-point index), and indicators for the post-Cold War period and whether or not there’s been an alternation of the party in power since the start of the democratic episode.
Duration of Democracy and Risk of Breakdown
As the figure clearly shows, risk actually increases over time, at least for the first 15-20 years of a democracy’s life span. Once a democratic regime has made it that far, the risk of breakdown appears to be unrelated to the further passage of time. Learning and habituation may be occurring, but they clearly aren’t the only thing going on.
This pattern has important practical implications. In the past, international efforts to support democratic transitions have focused heavily on founding elections, and attention has often waned quickly in the years that follow. If this evidence is correct, then the traditional approach may be a risky one.
Dart-Throwing Chimp 
Joe Wright
/ September 24, 2011Hi Jay,
Two thoughts. First, the risk of democratic breakdown in some countries (Costa Rica comes to mind) may be roughly zero, whereas in other countries this risk is positive. You probably accomplish a lot of this separation by excluding OECD countries from the sample. (See Svolik’s 2008 APSR piece on split-population models of democratic breakdown).
Second, there may be non-proportional hazards. There was once a vigorous debate over whether presidential democracies are more fragile (Cheibub shows that a lot of this difference can be explained by prior regime type: new democracies previously ruled by military regimes are more fragile). I always wondered whether these institutional features (or following Cheibub, prior military rule) influence the shape of the hazard. I haven’t kept up with the literature on democratic breakdown, so I don’t know if some one has recently looked into this possibility.
If we’re going to substantively interpret hazards, we should think carefully about the types of countries where there are theoretical expectations for different baseline failure risks. For example, in countries where the military repeatedly intervenes (Honduras, Pakistan, and Thailand come to mind), we might expect the risk of breakdown to initially be increasing. With a history of military intervention, a new democracy may be less likely to survive its first transition crisis than its second transition crisis — if it gets there. In other countries, however, this may not be the case (perhaps India and Malawi). Level of development (GDP per capita or infant mortality) might be another candidate for non-proportional hazards if the risk of democratic breakdown is close to zero and flat for relatively rich (new) democracies.
From a policy perspective, I’d imagine it’d helpful to know if your finding for the shape of the hazard only occurs in new democracies that transitioned from military rule or only for relatively poor democracies. If I had to place a bet, I’d go for the former (prior military) not the latter (poor).
-Joe
dartthrowingchimp
/ September 24, 2011All great points, Joe. (No surprise there, given how much you’ve already thought about these questions.)
In fact, I started to do some stratification to check for difference in baseline hazards across subgroups (before vs. after the first alternation in power was one I tried), but I decided to cut it short for the blog — for now. Maybe I’ll push that exercise further and do a follow-up post.
Also, I tried to be careful not to push interpretation of that baseline hazard too far. I hope I didn’t overstep. I’m with Neal Beck (and, apparently, you) on that one. That estimation of duration dependency is a catch-all for time-dependent variance not explained by other things in the model, and we shouldn’t tell tidy stories after the fact about what it supposedly means. Mostly, I was just trying to point out that the observed pattern doesn’t support the traditional story.