In Which I Acknowledge Adam Przeworski’s Brilliance and Then Argue with Him in Absentia

A few weeks ago, the blog ABC Democracy posted a video of Adam Przeworski speaking at a Kenyon College conference entitled “Should America Promote Democracy Abroad?” Przeworski is widely and justifiably considered one of the preeminent scholars on comparative democratization, so I was very curious to hear what he had to say on a topic that greatly interests me.

It turned out that I agreed wholeheartedly with Przeworski on the conference’s titular topic, but I disagreed with a few assertions he made along the way about the state of our knowledge on transitions to and from democracy. I thought I would take advantage of my blogger’s platform to engage in a virtual dialogue with Przeworski on those issues and then close on some points of agreement.

Point of Disagreement #1: We Can’t Predict Transitions to Democracy

Here’s what Przeworski said, starting at about the 46-minute mark, with the part to which I’m responding in bold:

In spite of an enormous amount of research over the past 30 years, we don’t have a general understanding of why dictatorships fall. There are [sic] statistical work that introduces every possible factor you can imagine–not just the kitchen sink, the grandmother’s attic. And the results are, one, not robust, and, two, in statistical terms, have very weak predictive power. Which leads me, after many years of this kind of work, to believe that, in fact, dictatorships run many different, idiosyncratic risks and fall for idiosyncratic reasons.

Przeworski is surely correct that there are many pathways to democracy, but that doesn’t mean we can’t use statistical models to forecast where and when democratic transitions will occur. In fact, we’ve got solid evidence that we can.

In a report I wrote for my old job as research director for the Political Instability Task Force, I summarized the results of modeling exercise aimed explicitly at assessing the likelihood of transitions to and from democracy in countries worldwide since the early 1970s. As the report describes (pp. 22-24), a relatively simple statistical model discriminates fairly well between impending transitions and durable autocracies. In an out-of-sample forecasting exercise using a simple decision rule (Top 20), that model correctly flagged 26 of the 29 impending transitions (sensitivity of 90 percent) as “high-risk” cases while producing roughly nine false positives for each of those true positives (specificity of 73 percent).

Those accuracy rates are far from perfect, but they’re also a lot better than chance, which is what I hear in Przeworski’s phrase “very weak predictive power.” The specific causes and catalysts of democratic transitions may vary widely over space and time, but there seem to be enough commonalities across recent cases that we can get a decent read on which ones are “ripest” for this kind of change.

Point of Disagreement #2: Well-to-do Countries Never Backslide

According to Przeworski,

We do understand quite well conditions under which democracies survive…There is a fact, which you probably know because I know that some of you have read it, but which continues to be astonishing, which is that no democracy ever fell in a country with per capita income higher than that of Argentina in 1976.

This fact may not be as, well, factual as Przeworski believes. As I noted in a recent post, using economist Angus Maddison’s estimates of GDP per capita, I can think of at least two breakdowns of democracy in countries richer than Argentina in 1976: Thailand in 2006, and now Hungary in 2011.

To be fair to Przeworski, Thailand in 2006 was not much richer than Argentina in 1976–their per capita incomes were $8,238 and $7,965, respectively–and not everyone would agree that Hungary’s crossed the line into authoritarian rule in 2011.

Still, that there’s some doubt about this “iron rule” of politics has deeper implications for our understanding of democratization, and “development” more generally. In American political science, at least, the prevailing view is that democracy is the best and final form of government attained by countries as they modernize and “mature,” politically and economically. This view seems to find confirmation in a world where democracies that have crossed some developmental threshold never fail. If democracy sometimes does fail even in richer countries, however, then the whole premise of modernity as the end stage of a process of growth and maturation becomes a bit muddled. The strong correlation between wealth and the survival of democracy is still there, but the inference from that correlation that modernity is a package deal looks a bit shakier.

Point of Disagreement #3: The Risk of Democratic Breakdown Falls with Each Passing Election

Around minute 49, Przeworski says:

One thing that’s striking is that elections seem to be a self-institutionalizing mechanism. By this, I mean the following: that once a country holds one decent election, the probability that the democratic regime will be overthrown in the future declines rapidly. I can tell you, without an election is 1 in 8; after one election, 1 in 25; after two elections, 1 in 55; after three elections, 1 in 90. So that first decent election–and not even with alternation that was Sam Huntington’s criterion–just having an honest election in which there’s some competition and somebody wins, the winner occupies the office of government and runs an honest election again, that’s enough.

Once again, that’s not the pattern I see. In the report I mentioned earlier–and blogged here in September–I find that the risk of backsliding actually increases over time until democracies are in their teens or even early 20s. In Przeworski’s terms, the pattern I see implies that democracies have to survive at least a few election cycles before their risk of breakdown starts to decline, other things being equal. At the same time, I also find that alternation in power does make a big difference; other things being equal, democracies that have seen at least one alternation of the party in power are less than half as likely to fail as ones that have not.

Maybe this disagreement is, at least in part, an artifact of differences in the measures of democracy employed by our respective studies. Unsurprisingly, I happen to think my measure is more useful, but plenty of people use the version on which Przeworski’s assertion is based.

Still, that we can’t be sure Przeworski’s pattern is real is a big deal, not the least because it suggests very different strategies for interested parties seeking to support the survival of democracy in cases that have recently established it. In Przeworski’s world, a strategically minded supporter might focus her efforts on the first one or two elections. In my world, that supporter pretty much needs to keep worrying until a democratic alternation in power occurs. If we’re not sure which of those worlds we inhabit but we care deeply about the survival of democracy, then we’ll probably want to err on the safe side and assume the risk persists much longer than Przeworski’s inference about elections as a “self-institutionalizing mechanism” would lead us to do.

Points of Agreement

Alongside those points of disagreement, there were many things Przeworski said with which I agree wholeheartedly. I’ll close with a couple of those bon mots:

Identifying the causal effects of any kind of policy intervention is extremely tricky.

Yes, in a world with no “control” group, a relatively small number of events, and a dense web of causes and interventions, it’s virtually impossible to say anything with confidence about the marginal effects of specific policies and programs on the prospects for democratic transitions and consolidation.

Last, and without comment:

Look at the United States from the point of view of Russians or the Chinese…It’s a country where half of the population doesn’t vote, even in presidential elections; where barriers of entry to politics are enormous; in which practices which in other countries would be considered political corruption are ubiquitous; a country with the highest degree of inequality among the developed countries; a country in which, at least for black American males, being free means only being out of jail; the oldest democracy in the world which has the highest rate of incarceration in the world. I think that, if democracy promotion is to be at all credible and at all effective, it should begin at home.


Will Democracy Survive in Europe? Part 1

In writings on foreign affairs for popular audiences, there’s a whole sub-genre that can be identified by the tagline: “It’s impossible to predict; now let me offer a prediction.” These stories usually drive me crazy, but Jack Snyder–of democracies-can-also-be-belligerent renown, if there’s such a thing as renown in the microscopic world of popularized political science–recently delivered an unusually successful one with an essay on the future of China. In it, he writes:

Predicting China’s future is a seemingly impossible task that people choose to take on anyway…People want to know because they feel their lives—and their actions—depend on it. Having a vision of possible and likely futures of China matters because outsiders think their choices might affect China’s trajectory and because they want to be prepared to respond to China’s outsize presence.

The same could be said of the future of democracy in Europe. Will democracy survive in the European Union’s Cold War-era members in the face of tremendous economic and social pressures, the likes of which the continent arguably hasn’t seen since the last World War? It’s impossible to say for sure; now let me try anyway.

Snyder’s piece on China works better than most because he considers competing predictions; he is explicit about the theories generating those predictions; and he tries to base his adjudication of those competing predictions on evidence from careful studies. That’s the model I’m going to attempt to follow here.

Like China’s trajectory, the survival of democracy in Europe carries historic weight for the whole planet. The failure of democracy in any one of these countries would mark a momentous turn in a world where liberalism is supposedly ascendant, and Europe is liberalism’s geographic center of gravity. At the same time, the survival of these democratic regimes in the face of such extreme pressures would help exorcise the specter of fascism that has haunted European politics since the 1930s.

Available theories suggest different futures. The dominant take on democratic consolidation among American political scientists comes from modernization theory, which posits that rich countries with post-industrial economies have undergone certain social and cultural changes that effectively “lock in” democracy. The eminent scholar Adam Przeworski famously claims that democracy has never failed in a country with a per capita income higher than Argentina’s in 1973, and Ron Inglehart and Christian Welzel muster survey data to show that the political and economic transformations underlying that factoid correlate with changes in citizens’ beliefs and values in more liberal directions.

There are at least two knocks on this theory, though. First, there are some recent exceptions to Przeworski’s rule of thumb, and they suggest that the innoculative effects of socio-economic modernization may not be as powerful as modernization theorists have supposed. Thailand is one, and Hungary is another. According to Angus Maddison’s widely used estimates, in constant dollars, Argentina’s GDP per capita in 1973 was $7,962. When Thailand’s democratically elected government was swept aside by a military coup in 2006, its GDP per capita was $8,238. Ironically, that figure had stood at $7,886 the year before, meaning Thailand only crossed the Argentina-in-1973 threshold in the same year its democracy was usurped. Hungary didn’t cut it so close. As it has slid back into authoritarian rule in the past year and a half, its GDP per capita was roughly twice as large as Argentina’s in 1973, having crossed Przeworski’s virtual finish line sometime in the mid-1990s.

Second–and, in my view, more significant–all of the data on which this “iron law” of democratic consolidation is based come from a specific historical era that was arguably exceptional in some important regards. Modernization theory was formulated in a period of bipolar world politics in which the threat of apocalyptic war with the Soviet Union compelled an unprecedented degree of political cooperation in Western Europe. That bipolarity coincided with a period of global economic integration and liberalization that was driven in no small part by the deliberate efforts of policy-makers in the polar power with which Europe was aligned.

When we use cross-national data to look back on this era, we see a strong correlation between wealth, liberal values, and democracy. What we lack is the counterfactual history in which these socio-economic trends occurred without the supporting international architecture. Without that “control group” world, we can’t be confident that the observed patterns in democratic consolidation don’t depend on features of the international system that are no longer present.

One major alternative to modernization theory as a lens for looking at present-day Europe comes from theories emphasizing the effects of economic performance on the risk of democratic breakdown. This perspective is nicely summarized by Larry Diamond, who in the late 1990s asserted that, “It is by now a truism that the better the performance of a democratic regime in producing and broadly distributing improvements in living standards, the more likely it is to endure.”

The causal chains in mental models emphasizing the effects of economic performance generally run through popular legitimacy. The trouble starts when economic malaise erodes popular confidence in the ability of democracy to “deliver the goods.” As faith in democracy wanes, support for authoritarian alternatives waxes. Economic slumps can also produce social unrest that may tempt military leaders to seize power in an attempt to preserve or impose order.

Or so the thinking goes. In fact, the empirical evidence is a bit squishy on this point. Some statistical studies have confirmed this claim, but others have not. Importantly, in the statistical models these studies have produced, any increased risk associated with slow growth is overwhelmed by the effects of national wealth, so that rich countries appear to be “cured” of their vulnerability to democratic breakdown in response to poor economic performance. This immunity is implicit in my own research, where I only see increased risk of breakdown when I restrict the analysis to non-OECD countries. This immunity is made more explicit in Milan Svolik’s split-population modeling, which shows that economic recessions do affect the timing of democratic breakdowns, but only in countries where democracy has not yet consolidated. Among consolidated democracies–identified primarily by their high levels of per capita income–the risk of democratic breakdown is virtually nil, however the economy is performing.

As I read it, then, the empirical evidence on economic performance actually ends up supporting the “lock-in” argument from the modernization model. In the period covered by our cross-national data sets, recessions have sometimes helped to undo new democracies, but wealthy, long-established democracies like Europe’s have proved immune to these maladies.

Again, though, what we can’t learn from these theories and the studies that have tested them is the extent to which this pattern depends on the geopolitical peculiarities of the past half-century–or, for that matter, some other system-level feature I’ve failed to consider. That, to my mind, is where we reach the limits of current evidence. Statistical studies of the Cold War and post-Cold War periods can reassure us that we’ve never seen a democracy as rich and long-lived as the ones in the E.U. before enlargement fail before, and they can show that these democracies haven’t been vulnerable to coups and collapse in the way that newer and poorer democracies have. What they can’t tell us is whether this year is now different. Within the system those studies have examined, Western Europe’s democracies have proved to be resilient–but are we still inhabiting that system, or has the world changed in ways that re-open the door to democratic breakdown? To try to answer that question, historically informed speculation is the best we can muster.

This post is the first of a two-parter. I split it up because I know my attention starts to wander after 1,000 words, so I figure yours probably does, too. In Part 2, I’ll apply my own game-theoretic model of democratic consolidation to contemporary Europe to see where it leads.

Economic Growth and the Survival of New Democracies

Last week, a senior official in Egypt’s Muslim Brotherhood warned the U.S. against cutting aid to his country at a time when Egypt is, he suggested, on the brink of economic collapse. In an interview with the Washington Post, Khairat Al-Shater said that reductions in Western aid would exacerbate an economic crisis that could “transform a peaceful revolution into a hunger revolution.”

Al-Shater’s warning reflects a widely held view that new democracies can be made or broken by their economic performance. “It is a cardinal principle of empirical democratic theory,” democratization scholar Larry Diamond writes, “that hard economic times are supposed to mean hard times for democracy.” This principle has been confirmed by a few statistical studies on the survival of democratic regimes: other things being equal, the risk of democratic breakdown does seem to be higher when GDP growth rates are slower. (See here, here, here, and here for affirmative findings and here for a negative one.)

After reading Al-Shater’s warning, I decided to revisit this question with an emphasis on the real-world concerns of the moment. Instead of looking at the entire life course of all democracies, as previous studies generally do, I wondered what economic performance around the time of a democratic transition–like we saw in Tunisia in 2011 and like we might be seeing right now in Egypt and Libya–would tell us about the prospects that a democratic regime will survive well beyond its founding elections. Perhaps these earliest years create impressions and encourage strategies that enable or afflict the ensuing regime during this formative period in ways we can’t see when we lump entire episodes of democracy together.

To test this conjecture, I used a global data set to identify all transitions to democracy that occurred during the period 1955-2008. With that case list in hand, I built a logistic regression model of the relationships between the conditions under which those transitions occurred and the odds that the ensuing regimes would survive for at least five years. (As it happens, surviving for just five years is actually a pretty big deal. Of the 103 transitions to democracy that occurred during that period, only 62 produced regimes that lasted longer.)

From prior research, we know that higher levels of economic development, the absence of political polarization, prior democracy, and the end of the Cold War are all associated with improved prospects for democratic consolidation, so all of those factors were included in the model. To capture the marginal effects of economic performance on prospects for democratic survival–the original point of this exercise–I added measures of annual percent change in GDP per capita for the three years bracketing the transition: the one before, the year of the transition, and the year after. (See the end of the post for more details on the modeling.)

As expected, I found that a new democracy’s survival prospects are indeed better when its economy grows faster around the time of its birth. In contrast to al-Shater’s gloomy prognostication, however, the effects I observed were not large. The marginal effects from GDP growth in the year of the transition are illustrated in the line plot below. As the chart shows, a difference of several percentage points in GDP growth–a large swing in most real-world situations–would produce only a very modest difference in the estimated likelihood of surviving past five years, other things being equal. (I don’t think p-values are as informative as estimates of marginal effects, but for those of you wondering, the p-value in this instance is 0.28; the coefficient is 0.046.) The association with growth on either side of the transition are not captured in that chart, in part because they were even weaker (coefficients of 0.027 and 0.018 and p-values of 0.54 and 0.70, respectively).

We can also see the weakness of this effect in the modest contribution of those growth rates to the statistical model’s ability to accurately assess risk in the historical cases. The figure below plots Receiver Operating Characteristic (ROC) curves for versions of my model with and without the measures of initial GDP growth. ROC curves summarize a model’s ability to discriminate between cases with and without some feature of interest–in this instance, surviving past five years. The better the model does, the farther the line pushes toward the upper left-hand corner, and the larger the area under the curve (AUC). As you can see, adding measures of GDP growth to the model doesn’t improve the accuracy by a whole lot, producing just about a 2% bump in AUC.

On the whole, I’d say these results run counter to the spirit, if not the letter, of prevailing expectations. The marginal effects of economic performance flow in the anticipated direction, but they don’t have anything close to the kind of “make or break” impact that Al-Shater and Diamond’s statements imply. A new democracy’s level of economic development and the occurrence of acute political polarization tell us the most about its survival prospects, and variations in economic performance around the transition don’t seem to move the needle a whole lot beyond that.

I wonder if the prevailing wisdom about the dire consequences of poor economic performance for democratic consolidation isn’t at least in part a case of the availability heuristic at work. Historical cases of economic crisis followed by democratic collapse easily spring to mind (Weimar Germany, anyone?), and it’s not hard to generate a plausible story linking those two events. What those plausible stories seem to overlook, though, is that many of those attempts at democracy probably would have failed anyway, even in the absence of economic crisis, because that’s the fate of most democracies across a wide range of conditions. Meanwhile, there are plenty of countervailing examples of young democracies that survived sharp economic contractions (say, Greece after military rule, or much of post-Communist Europe), but these null cases seem to be more forgettable.

Even if economic growth had a stronger impact on prospects for democratic survival than my analysis indicates it does, I’m skeptical that this information would be as useful to policy-makers seeking to promote democracy as I suspect they think it is. Assume for a moment that a bump of a few percentage points in GDP growth in the transition year would double the odds in favor of democratic survival. Can anyone tell me what policy interventions will reliably pump growth rates that far, that fast? If foreign aid or economic policy could work that kind of magic, wouldn’t the “developing” world already be a lot richer?

I’ll wrap this post up by going back to where we started, namely, the Middle East after the “Arab awakening.” Even though GDP growth doesn’t contribute much to it, the model’s overall performance isn’t bad. After looking at those ROC curves, I wondered what the model would say about the prospects for the survival of new democracies in three Arab countries on the cusp of new tries at democracy: Tunisia, Egypt, and Libya. Of the three, only Tunisia would already qualify as democratic by my definition, but Egypt and Libya are both in the midst of transitions from authoritarian rule that could put them over the threshold soon. So I took the IMF’s latest projections of their growth rates and plugged them into the model, along with recent data on their levels of economic development and my best guess as to whether or not they would qualify as acutely polarized according to the data set I used for that indicator. Here’s what came back as estimates of the  probability that each of those new democracies would make it to their sixth birthday, assuming that, of the three, only Tunisia would not qualify as acutely polarized:

  • Tunisia: 82%
  • Egypt: 48%
  • Libya:  89%

The contrast between Tunisia and Egypt’s survival prospects did not surprise me, but the high estimate for Libya did. Interestingly, expected economic growth seems to be contributing to this result. According to the IMF, Libya’s economy contracted by more than 60% in 2011, but it’s expected to recoup some of those losses in 2012 with an astonishing annual growth rate of nearly 70%. That value is so unusually large that it packs a lot of wallop, even though the weight for GDP growth in the equation is small. Whether that anomalous leap translates into a tremendous boost for democratic consolidation in the real world is another matter. Color me dubious.

Details of the Modeling

The sample for the statistical analysis described here comprises 103 democratic transitions that occurred in countries worldwide during the period 1955-2008.  These transitions were identified using the same data set on episodes of democracy that was summarized in my book. To focus on transitions most like the ones occurring in the Middle East today, cases where new countries were “born” with democratic regimes were excluded from the analysis. I did not use the well-known and widely-used Democracy and Dictatorship Data Set because, as elaborated in this working paper, I have serious concerns about its utility for survival analysis. (That said, I would be very interested to see how sensitive the results reported here are to the choice of measures of democratic transitions and breakdowns. I’d do it myself if this were an academic paper, but, hey, it’s just a blog post.)

Once I’d assembled a roster of relevant cases, I used the ‘glm’ command in R to estimate a logistic regression model that included the covariates listed below (with sources in parentheses). The analysis file includes one record per transition. The dependent variable in this model was a binary one indicating whether or not a democratic episode lasted more than five years beyond its transition year. As noted above, 62 of the 103 cases did.

  • Annual percent change in GDP per capita in years t-1, t, and t+1, where t is the year in which the democratic transition occurred (World Development Indicators)
  • Infant mortality rate, relative to annual global median and logged (U.S. Bureau of the Census)
  • Political polarization (a.k.a. “factionalism,” indicated by a score of 3 on Polity’s PARCOMP variable)
  • Any prior episodes of democracy
  • Post-Cold War period

The ROC curves were created using ROCR.

If you’d like to replicate and tinker with this analysis, please email me to ask for the data set and R script. My address is ulfelder <at> gmail <dot> com.

Update: The code I used for this analysis is now on Github, here. The data set in .csv form can be downloaded from my Google Drive, here.

  • Author

  • Follow me on Twitter

  • Follow Dart-Throwing Chimp on
  • Enter your email address to follow this blog and receive notifications of new posts by email.

    Join 13,646 other followers

  • Archives

  • Advertisements
%d bloggers like this: