For some things, there really is a season, and a smattering of memorable events suggests that political instability in Africa might be one of them. In the Central African Republic (1966), Ghana (1981), and Nigeria (1983), putschists famously struck on New Year's Eve or New Year's Day. On Christmas Day in 1989, rebels led by Charles Taylor invaded Liberia. The mutiny that first tipped Cote d'Ivoire into crisis erupted in December 1999; the Central African Republic's most recent collapse started in December 2012; and the rebellion that triggered a civil war in South Sudan began in December 2013.
If the end of each year and start of the next is an abnormally unstable time in Africa, knowing this fact could help us anticipate disruptive events in the region. So, is it?
We analyzed data on three different forms of political instability in Africa and found that, for the most part, it is not. Contrary to the hypothesis suggested by our list of easily-recalled anecdotes, December and January do not produce significantly more coup attempts or battle events on average than other months of the year. The partial but significant exception to this pattern comes from protest and riot events, where we do see evidence of a substantial seasonal spike in January.
Contrary to the hypothesis suggested by our list of easily-recalled anecdotes, December and January do not produce significantly more coup attempts or battle events on average than other months of the year.
Many of the events in our initial list were coups, so we began our analysis there. Taking data on all coup attempts worldwide since 1950, we reduced the data to events in African countries; summarized that reduced list as a time series of monthly event counts; then used a statistical model to look for persistent differences in those counts across the months of the year.
The plot below summarizes the results. The plot shows rate ratios, which compare each month’s estimated baseline rate to an arbitrarily selected reference month’s. Here, we used August as the reference month because its baseline rate fell in the middle of the pack. The vertical bars represent 95% confidence intervals, while the dots near the middle of those bars are the model’s single best estimate. So, according to our model, November, July, and April are the months with the highest generic risk of coup attempts in Africa. December is next in line, but its rate is not that much greater than August’s, and January's rate is actually a bit below August’s.
To check for seasonal cycles in political violence and social unrest, we turned to data produced by the Armed Conflict Location & Event Data Project, a.k.a. ACLED. This dataset only starts in 1997, but it offers careful and continuous coverage of the ensuing 20 years. For these analyses, we aggregated the event data into monthly counts for each country rather than for the continent as a whole. Then we used statistical models to look for evidence of monthly cycles in those counts while controlling for persistent differences across countries in their volatility.
The results are summarized in the two plots that follow. The first plot shows results for battles, the second for protests and riots. In both, the basic layout is the same as the plot for coup attempts, but now the vertical axis represents the typical change from the previous month in the count of events, rather than a rate ratio. For consistency’s sake, we again used August as the point of reference. In the top plot, we see that March is the hottest month for battle events; June and January run warmer than most; and December turns out to be unremarkable. In the second plot, we see that January is clearly the continent's peak month for riots and protests, while December is, if anything, an unusually quiet one.
So, on the whole, we find mixed evidence that the months bracketing New Year’s Day are a particularly volatile time in African politics. January does turn out to be a peak period for social unrest, but December is not, and neither of those months produces unusually high numbers of coup attempts or battle events.
In a broader sense, this exercise reminds us to beware the availability heuristic when looking for seasonal patterns in politics. Even when we can recall what look like clusters of cases around focal points on the calendar, the ready availability of those examples in our memories is not a reliable indicator of cyclical variation in the real world.
More specifically, we used the ‘season’ package in R to fit a Poisson model with month-specific effects that account for differences in the number of days in those months. We selected August as the reference month because its event rate was close to the overall monthly average.
For these event types, we used the ‘lme4’ package in R to fit multilevel models with random effects for each country and fixed effects for each calendar month. Here, we used month-to-month changes in event counts as the target of our models instead of the raw counts.