Supporters of Tukish President Tayyip Erdogan celebrate after soldiers involved in the failed coup attempt surrendered on the Bosporus Bridge in Istanbul on July 16, 2016. (Yagiz Karahan/Reuters)

Coup attempts often come as a big surprise — and bring big changes. Purges and imprisonments since Turkey’s July 2016 failed coup, for instance, have some observers worried about the state of democracy in Turkey, while others worry about Turkish rapprochement with Russia.

Anticipating exactly when a coup will occur is generally difficult, but a number of researchers work on forecasting where coup attempts are likely to occur. Jay Ulfelder created annual projections of coups d’état for 2012 to 2015, but his blog has been on hiatus since then.

Our research has developed tools to forecast irregular leadership changes, which include coups. Using some of the tools and similar data, we’ve now created a 2017 forecast for the risk of coups for 161 countries (see Figure 1).

Figure 1 shows the probability of coup attempts in 2017, based on our models. The numbers indicate the percent chance that a country will have a coup attempt this year. We forecast Thailand, Burundi, Turkey, Central African Republic and Chad as countries more likely to see coup attempts in 2017 than, say, Canada or Australia. Source: Andreas Beger and Michael D. Ward

We use a small number of statistical models to create separate forecasts for the risk of a successful coup, like in Thailand in 2014, or failed coup attempt, like in Turkey in 2016. Then we combine them to get a single forecast for the risk of a coup attempt for each country. The complete forecasts are available in a CSV file, and we explain the technical details in a separate post.

The models look for patterns in past coup attempts using data from 1960 to 2017 that we assembled using the Powell and Thyne coup data, Polity Project, World Development Indicators, Armed Conflict Dataset, global food prices (FAO), oil prices (BP) and the Gleditsch and Ward list of states.

Here are some of the things that what we included: how long the current leader has been in power, whether he or she was chosen democratically and the type of government in place. We also consider GDP, economic growth, population and infant mortality; these variables are more difficult to construct in war-torn or highly authoritarian societies, but we use estimates wherever required. We tabulate information about the spread of communication technologies — specifically, Internet access and cellphone ownership.

Thailand, Burundi — we’re looking at you

Our approach also gives us the chances that a coup attempt would succeed or fail. Figure 2, below, lists the 30 places coup attempts are most likely — and the estimated probability of success or failure in each case.

Figure 2 shows the 30 countries most likely to see a coup attempt in 2017, based on our models. The chance of a coup attempt (red) is based on the risk of a successful coup (green) or a failed attempt (blue). Turkey, for example, has a 10 percent chance of a coup attempt, based on 4 percent and 6 percent chances of a successful coup or failed attempt. With these numbers, we can also say that if another coup attempt happens in Turkey, the chance that it would fail is roughly 60 percent. Source: Andreas Beger and Michael D. Ward.

We emphasize two main points. First, the probabilities are on the order of 1 in 10 — or even 1 in 20. Burundi topped the list with a 12 percent chance for a coup attempt. A perfect model would predict near 100 percent for all countries that will have a future coup attempt and near zero percent for all others.

Low probabilities do not mean that coups will — or won’t — occur. Rather, we offer a holistic view of the risk of a coup.  If we have 10 countries with a 10 percent risk, we should still expect to see one attempt on average.

Second, although we cannot say with certainty that there will be another coup attempt in Turkey or Burundi, there is a good chance that any coup attempts in 2017 will be among these 30 countries. The countries at the top of this list may not be surprising — each has shown recent instability or institutional weaknesses. Reality is a low probability event.

We tested our model retrospectively — and it worked

We checked that these forecasts are reasonably accurate by gauging whether our models would have foreseen prior coups. These retrospective forecasts take models that had access to data up to a certain point, and use them to forecast coup risk for the next year, e.g. data up to 2014 to forecast 2015, and so on.

So here’s the big question — would our model have foreseen the Turkish coup attempt in 2016?

Our 2015 forecasts are consistent with Ulfelder’s forecasts, which don’t peg Turkey as a high risk. But our 2016 forecast for Turkey spiked, ranking it the 11th highest risk for a failed coup attempt, and 26th highest overall risk for any coup attempt.

Looking at all 161 countries, then, we’re fairly confident about the ranking of country risk. The anticipated number of coup attempts in 2017 is approximately five — three successful coups and two failed attempts.

These numbers are probably on the high side. The forecasting models rely heavily on data during which there was an annual average total of five to 10 coup attempts. But since 2001, coups have been getting much rarer, with an average two to four attempts per year.

We don’t know what has caused the gradual decline in the number of coup attempts. But it may be related to the spread of the Internet and social media. Coups are partially a game of setting expectations — creating the perception that the coup has already succeeded. That’s why rebel forces often seized the radio and television broadcast stations and proclaimed control.

The Internet may make it easier to mobilize populations, but governments generally can control overall communications, making it difficult for rebel groups to transform perceptions of who is in charge. While coups may be becoming easier to coordinate, they are simultaneously becoming easier to suppress.

In the case of Turkey, President Tayyip Erdogan’s government co-opted social media to urge millions of Turkish citizens to oppose the coup attempt. His first, crucial, public message during the coup attempt was conveyed to CNN Turk via FaceTime.

Our models can’t predict exactly what will cause a coup

High-risk cases all have markers for instability, however. For example, Burundi has been in crisis since May 2015 when President Pierre Nkurunziza sought and obtained a third term.

Thailand has been under martial law, with strong restrictions on civil liberties, since the 2014 coup. The country approved a new constitution in 2016 and scheduled elections for 2017 — but, as some researchers point out, elections often increase the risk of further coup attempts.

Are we likely to see a coup in Russia — or the United States?

Russia, maybe. We estimate that the risk in Russia is about 6 percent, which places it in the top 20 countries risk-wise. If one were to occur, the odds are 2 to 1 that if would fail.

The U.S. presidential transition has also led to Russian claims that a coup may occur in the United States. The same statistical models suggest that the U.S. risk is about 2 percent — and ranks the United States No. 103 out of 161 countries. The risk may seem high but reflects the variable for infant mortality, which is higher in the U.S. than in other developed countries.

There’s no way to include a metric for the idiosyncrasies that happen behind the White House or Kremlin walls, though. Nor do our models tell us what specific events should be warning signs of an impending attempt.

But most of the countries at the top of our 2017 forecasts have undergone coup attempts in recent history, or have existing open political conflicts that test their normal institutions. Despite the fact that all coups may appear surprising, there are no surprises in our list of most coup-prone countries for 2017.

Andreas Beger is as a data scientist at Ward Associates. Follow him on Twitter @andybeega.

Michael D. Ward is a professor at Duke University and founder of Ward Associates. Follow him on Twitter @3876.