Assessing Coup Risk in 2012

Which countries around the world are most likely to see coup activity in 2012?

This question popped back into my mind this morning when I read a new post on Daniel Solomon’s Securing Rights blog about widening schisms in Sudan’s armed forces that could lead to a coup attempt. There’s also been a lot of talk in early 2012 about the likelihood of a coup in Syria, where the financial and social costs of repression, sanctions, and now civil war continue to mount. Meanwhile, Pakistan seems to have dodged a coup bullet early this year after a tense showdown between its elected civilian government and military leaders. I even saw one story–unsubstantiated, but from a reputable source–about a possible foiled coup plot in China around New Year’s Day. These are all countries where a coup d’etat would shake up regional politics, and coups in some of those countries could substantially alter the direction of armed conflicts in which government forces are committing mass atrocities, to name just two of the possible repercussions.

To give a statistical answer to the question of coup risk in 2012, I’ve decided to dust off a couple of coup-forecasting algorithms I developed in early 2011 and gin up some numbers. Both of these algorithms…

  1. Take the values of numerous indicators identified by statistical modeling as useful predictors of coup activity (see the end of this post for details);
  2. Apply weights derived from that modeling to those indicators; and then
  3. Sum and transform the results to spit out a score we can interpret as an estimate of the probability that a coup event will occur some time in 2011.

Both algorithms are products of Bayesian model averaging (BMA) applied to logistic regression models of annual coup activity (any vs. none) in countries worldwide over the past few decades. One of the modeling exercises, done for a private-sector client, looked only at successful coups using data compiled by the Center for Systemic Peace. The other modeling exercise was done for a workshop at the Council on Foreign Relations on forecasting political instability; this one looked at all coup attempts, successful or failed, using data compiled by Jonathan Powell and Clayton Thyne. For the 2012 coup risk assessments, I’ve simply averaged the output from the two.

The dot plot below shows the estimated coup risk in 2012 for the 40 countries with the highest values (i.e., greatest risk). The horizontal axis is scaled for probabilities ranging from zero to 1; if you’re more comfortable thinking in percentages, just multiply the number by 100. As usual with all statistical forecasts of rare events, the estimates are mostly close to zero. (On average, only a handful of coup attempts occur worldwide each year, and they’ve become even rarer since the end of the Cold War; see this earlier post for details). For a variety of reasons, the estimates are also less precise than those dots might make them seem, so small differences should be taken with a grain of salt. Even so, these results of this exercise should offer plausible estimates of the chances that we’ll see coup activity in these countries some time in 2012.

Here are a few of things that stand out for me in those results.

  • My forecast supports Daniel’s analysis that the risk of a coup attempt in Sudan in 2012 is relatively high. It ranks 11th on the global list, making it one of the most likely candidates for coup activity this year.
  • Surprising to me, Pakistan barely cracks into the top 40, landing at 38th in the company of Iraq, Cambodia, and Senegal. Those countries all rank higher than 120 others, but the distance between their estimated risk and the risk in most other countries is within the realm of statistical noise. Off the top of my head, I would have identified Pakistan and Iraq as relatively vulnerable countries, and I would not have thought of Cambodia or Senegal as particularly coup-prone cases.
  • Unsurprising to me, China doesn’t even make the top 40. Perhaps there has been some erosion in civilian control in recent years, as Gordon Chang discusses, but it still doesn’t much resemble the countries that have seen full-blown coup attempts in the past few decades.
  • Interestingly, Syria doesn’t show up in the top 40, either. To make sense of this forecast, it’s important to note that assigning a low probability to the occurrence of a coup attempt in Syria in 2012 isn’t the same thing as a prediction that President Bashar al-Assad or his regime will survive the year. It might seem like semantic hair-splitting, but the definitions of coups used to construct the data on which these forecasts are based do not include cases where national leaders resign under pressure or are toppled by rebel groups. So the Syria forecast suggests only that Assad is unlikely to be overthrown by his own security forces. As it happens, my analysis of countries most likely to see democratic transitions in 2012 put Syria in the top 10 on that list.
  • Two of the countries near the top of that list–Guinea and Democratic Republic of Congo–are the ones where the Center for Systemic Peace’s Monty Marshall tells me he saw coup activity meeting his definition in 2011. Those recent coup attempts are influencing the 2012 forecasts, but both countries were also near the top of the 2011 risk list. This boosts my confidence in the reliability of these assessments.

I hope there’s a lot more on (or off) that list that interests readers, and I’d be happy to hear your thoughts on the results in the Comments section. For now, though, I’m going to wrap up this post by providing more information on what those forecasts take into account. The algorithm for successful coups uses just four risk factors, one of which is really just an adjustment to the intercept.

  • Infant mortality rate (relative to annual global median, logged): higher risk in countries with higher rates.
  • Degree of democracy (Polity score, quadratic): higher risk for countries in the mid-range of the 21-point scale.
  • Recent coup activity (yes or no): higher risk if any activity in the past five years.
  • Post-Cold War period: lower risk since 1989.

The algorithm for any coup attempts, successful or failed, uses the following ten risk factors, including all four of the ones used to forecast successful coups.

  • Infant mortality rate (relative to annual global median, logged): higher risk in countries with higher rates.
  • Recent coup activity (count of past five years with any, plus one and logged): higher risk with more activity.
  • Post-Cold War period: lower risk since 1989.
  • Popular uprisings in region (count of countries with any, plus one and logged): higher risk with more of them.
  • Insurgencies in region (count of countries with any, plus one and logged): higher risk with more of them.
  • Economic growth (year-to-year change in GDP per capita): higher risk with slower growth.
  • Regime durability (time since last abrupt change in Polity score, plus one and logged): lower risk with longer time.
  • Ongoing insurgency (yes or no): higher risk if yes.
  • Ongoing civil resistance campaign (yes or no): higher risk if yes.
  • Signatory to 1st Optional Protocol of the UN’s International Covenant on Civil and Political Rights (yes or no): lower risk if yes.
Leave a comment


  1. Jake

     /  January 30, 2012

    I’m wondering how you came up with the list of risk factors you use in the coup success and coup attempt models. Did you arrive at them theoretically, or did you use a process as in the 2010 AJPS piece where you ran a bunch of different variables to see what maximized the predictive power of the model?

    • Thanks for the question, Jake. For both analyses–any coup attempts and successful coups only–I started with a list of a few dozen candidate risk factors based on prior theory and research. Some of the variables on that list were quickly dropped because there are a lot of holes in the data sets that try to cover them, and multiple imputation was a bridge too far for these projects. The rest were included in the Bayesian model averaging exercise, which systematically explores the model space involving all possible linear combinations of those variables; selects the most plausible models on the basis of relative accuracy (Akake information criterion); and then averages coefficients across those models, weighting those means by the models’ relative plausibility. In other words, I used theory and data availability to pick a short list of likely candidates, and BMA sorted out out the rest.

  2. After posting this, I realized I should clarify that I wasn’t able to calculate scores for a handful of countries because of missing data on one or more of the risk factors. Countries without scores are: Afghanistan, Bosnia and Herzegovina, Haiti, Libya, Somalia, South Sudan, Swaziland, and Tunisia. Most of these are due to special Polity codes of -66 (foreign occupation), -77 (state collapse), or -88 (transition). In any case, their absence from the Top 40 should not be interpreted as an indication of low coup risk; I simply don’t have an estimate.

  3. For those of you keeping score at home, I see there may have been a coup in the Maldives today. As it happens, my analysis excludes countries with populations of fewer than 500,000, so I don’t have a coup risk score for that country in 2012.

  4. Last March, there were press rumors of military coups in both China, and India; neither appears in your top-40 lists. Are rumors precursors of coups, and/or are they to be ignored altogether?

    • Prof. Modelski, I’m honored to have you read and comment on the blog.

      The short answer to your question is: I don’t know. Only one of the two coup data sets I used also records coup rumors as distinct events, and the creator of that data set (Monty Marshall of the Center for Systemic Peace) cautions that those rumor data are not especially reliable. Consequently, I decided not to include rumors in my indicator of recent coup activity, nor did I include a separate indicator of coup rumors to see if it might have some independent association with the occurrence of coup attempts. On further reflection, though, it seems worth trying in future modeling rounds, if only to get some empirical traction on the question you raise.

  5. Rachel

     /  December 1, 2012

    I was searching for interesting problems to solve for my Pattern Recognition class when I stumbled on this blog entry through your link on the Foreign Policy site. I was just wondering, what made you decide to use BMA?

    • Two things. One, prior studies, including this one from my old research program, show that BMA often produces more accurate forecasts than other techniques, including the logistic regression models traditionally applied to this kind of problem. Two, I didn’t have strong priors about which of the couple dozen variables floating around the lit on coups would actually be most useful as predictors, and BMA offers an efficient exit from the trap of churning through models by hand, looking for models that “feel right.”

      That said, my thinking’s moved a bit since last year. When modeling rare events, I’m now more disposed to go ahead and specify a few models on the basis of competing theories and then average the forecasts from those models instead—possibly using ensemble BMA if there are enough events in the data to support it, or just taking the unweighted average if not. It looks like this approach will deliver similar accuracy, and it’s easier to tie the results back to the specific mental models that are driving them.

  6. Nahela

     /  January 23, 2013

    I’m guessing your model does not predict whether an attempted coup would be successful or not. I have a theory that a coup’s probability of success partly depends on the country’s literacy rate, particularly the youth literacy rate. I’d like to know what you think.

    • Interesting idea, and potentially testable. Could you say more about why you think youth literacy would help predict the success or failure of coup attempts?

      In these forecasts and the ones I produced for 2013, I’m using two different models and combining the forecasts from them. One of those models looks at the probability of any coup attempts, successful or failed, but the other looks at the likelihood of a successful coup. The ratio of the predicted probabilities from those two models can be read as an informal estimate of the odds that a coup attempt will succeed if it happens. To get a more formal estimate, we could also treat coup attempts as the unit of analysis for a statistical model and then their success or failure as the outcome of interest. If I were going to test your conjecture about literacy, that’s probably how I’d want to do it.

      • Nahela

         /  January 23, 2013

        You summed up exactly what I had in mind in your second last sentence. Well recently I did a short data analysis paper on gender gap in youth literacy rates in developing nations. Given the fact that protests/marches in recent times (Egypt, Occupy movement) have been organized through the use of social networking sites, blogs etc., it’s logical to assume that participants are somewhat literate. The majority of protesters seem to be young which is not surprising. Actually now that I think about it, it’d probably make more sense to measure the effect of youth literacy rates on coup attempts as opposed to “successful coups” as the latter largely depends on various other factors.

      • Thanks for clarifying. I’ll see if I can take a look at that issue, both in terms of rates and gender gaps. I have used those measures in models of the onset of civil-resistance movements and seen some hints of a connection there. The challenge for causal inference, though, is that literacy is strongly correlated with many other measures of economic and social “development,” so it’s difficult to tease out effects from any particular one of them in observational data.

  7. bruce

     /  February 4, 2014

    I wonder if you’d consider Thailand at higher risk, post-election, of a coup attempt by anti-govt. forces. I gather that you consider elections as precipitating events in the model. I thought I’d ask since the US State Department seems to be hedging against the risk…

    • Good question. The models don’t give us any traction on this, because the fact that it’s an election year is already factored into the forecast. I wrote a post on this topic two months ago (here), and my sense is that the risk of a coup as defined for these forecasts hasn’t changed much since then, for the reasons described in that post. What seems more likely than a military coup is a “judicial coup” in which courts sympathetic to the Democrat Party chip away Pheu Thai’s power or shove them out of office somehow. This seems increasingly likely to me, but just how likely, I don’t know—maybe we’re up to 60% risk by now? Anyway, I’m sympathetic to Pheu Thai in this fight, so I hope it doesn’t happen.

      • bruce

         /  February 10, 2014

        Jay, Hope you’re right. FYI: Global Post echoes your Coup 2.0 notion, which, one analyst says would involve “a barrage of rulings from courts backed by the army-linked old guard that dissolve the ruling party under allegations of misrule. Like almost all parties in Thailand’s history, Pheu Thai is riddled with corruption allegations, and many appear quite legitimate. “They’re going to create a sort of pseudo-legitimacy, with violence, plus verdicts from the courts saying this government needs to come down,” said Verapat Pariyawong, an independent legal scholar who has both advised and criticized the government. In that scenario, the void could be filled by an unelected people’s council. This would allow the army to claim that “we’re not staging a coup. We’re obeying the true powers,” he said. “In my mind, it’s the perfect coup. A new type of coup.” But that outcome carries the same risks: heavy and perhaps violent organized resistance from Thais who feel their electoral choices are squashed by well-heeled elites… ”


  1. Modeling and predicting coups | Economics 350:Quantitative Methods
  2. On (the now shattered) Malian Democracy « Opalo's weblog
  3. Why We Shouldn’t Be Quite So Surprised by the Coup in Mali « Dart-Throwing Chimp
  4. Chroniques maliennes : les marchands de sable « o104
  5. Fortnightly(ish) Review: Hip-hop in Mali, Nepal, Springsteen and Uzbek Elite Pop « World Politics Blues
  6. Rapports & Analyses: The Watch List: Predicting state failure isn’t as hard as you think.  | AfricVisions
  7. It’s Not Just The Math « Dart-Throwing Chimp
  8. The Coup Trap, Mali Edition « Dart-Throwing Chimp
  9. New Coup Throws Off Plans For Intervention In Mali | News Talk One
  10. Coup Forecasts for 2013 « Dart-Throwing Chimp
  11. Did Libya Cause Mali? « Dart-Throwing Chimp
  12. Map: The countries most at risk for a coup in 2013
  13. Is Eritrea due for a revolution?
  14. Is Eritrea due for a revolution? « Time for change
  15. The 3rd world view Coup Predictions In 2013 | The 3rd world View
  16. Mali puts coup forecasting in the spotlight – LSMedia
  17. Is Eritrea due for a revolution? | Zegabi
  18. Most Popular Posts of 2013 | Dart-Throwing Chimp
  19. A Research Note on Updating Coup Forecasts | Dart-Throwing Chimp
  20. Coup Forecasts for 2014 | Dart-Throwing Chimp

Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


Get every new post delivered to your Inbox.

Join 5,742 other followers

%d bloggers like this: