mugabe’s long shadow in zimbabwe: some statistical evidence

One of the central questions of contemporary Zimbabwean politics is whether the current regime led by President Emmerson Mnangagwa represents a clean departure from the violent 37-year rule of Robert Mugabe. Mnangagwa, a close Mugabe adviser and former security chief, came to power in November 2017 via military coup. In the subsequent 15 months, Mnagagwa and his inner circle of now-civilian advisers have embarked on a public diplomacy campaign to loosen international sanctions against the Zimbabwean economy and its leaders. The government’s message, “Zimbabwe is open for business,” implies a sharp departure from the instability and uncertainty that of Mugabe’s time in power. In the following post, I use evidence from a statistical test of trends in violence against civilians to show that this sharp departure has not occurred. The evidence indicates that repression under Mnangagwa mirrors levels of violence under Mugabe.

There are multiple indications that change has not occurred as advertised. First, the twin sources of Mugabe’s autocratic rule, the ruling Zimbabwe African National Union-Patriotic Front (ZANU-PF) party and the Zimbabwean military, continue to support Mnangagwa’s claim to power. Second, the economic, humanitarian, and political crises that typified the Mugabe government have continued under Mnangagwa’s rule. Anti-government protests in response to a sharp increase in government-directed gas prices this January preceded a period of harassment and extrajudicial imprisonment, hallmarks of repression under Mugabe. Repression by security forces during this period resulted in the deaths of at least 17 people and injuries of dozens of others.

These factors are important, but they do not provide systematic evidence that Mnangagwa’s rule is similar to Mugabe’s. Sticky ruling parties and security forces are a common feature of autocratic transitions. However, they do not guarantee that the political dynamics of the pre-coup regime will persist. The contemporary case of Ethiopian President Abiy Ahmed, for example, demonstrates that individual leaders are able to make political reforms despite structural constraints that the institutional pillars of their regimes impose. Additionally, events like the violence in January might not be representative of the general pattern of violence under Mnangagwa. It is theoretically possible that the average level of violence or repression under Mnangagwa is less than under Mugabe, and that episodes of repression like the violence against protesters in January are simply exceptions to that rule.

To address these problems and test whether Mnangagwa’s repression differs from Mugabe’s, I use a method of statistical analysis called “regression discontinuity.” A regression discontinuity design tests whether some important event accounts for a meaningful difference in an observable trend. If the regression discontinuity model indicates a relationship between the cutoff–in this case, the coup–that is statistically distinguishable from zero, there is evidence of a meaningful difference between the periods before and after the discontinuity. If it does not, there is no evidence of a meaningful difference between the two periods.

In this case, my measure of repression in Zimbabwe is instances of violence against civilians. I use week-aggregated data about violent events from the Armed Conflict Location and Event (ACLED) Project, which compiles event data based on a combination of media sources and reports by non-governmental organizations. The dataset includes observations from the 68 weeks prior to the coup (beginning August 1, 2016) to 68 weeks after the coup (ending March 9, 2019). Others interested in replicating my findings can use ACLED’s Data Export Tool to download the same dataset.

Although I am confident that ACLED’s information on violence against civilians provides an accurate estimate of repression in Zimbabwe, it is important to be transparent about potential shortcomings of the measure and the data I use to observe it. First, observable instances of violence against civilians are not the only political or humanitarian consequence of repression. Not all repression results in observable violence, and observable violence is not the only form of physical nor psychological harm. Indeed, much of the Zimbabwean government’s violence takes the form of subterfuge and harassment that are visible to neither local nor international media observers. Second, some scholarship on media reporting about political violence suggests that datasets that rely on media reporting undercount violent events in locations where international media organizations are less present, like rural areas. For each of these factors, I expect that they will underestimate the scale and scope of repression over time. However, I expect these estimates to be constant during the period under study, as Mnangagwa and Mugabe have relied on “invisible” violence to a similar degree.

There are two important assumptions of a discontinuity design: (1) that the discontinuity is “exogenous,” or external to the trend; and (2) that other explanatory factors were constant through the period of discontinuity. I address each of these assumptions in turn. The primary cause of the coup, infighting within the ruling ZANU-PF party, was unrelated to the primary cause of violence against civilians for much of the period, regime repression of anti-government activity. The coup therefore fulfills the first assumption of the discontinuity design.

The following graphic displays the direct effect of Mnangagwa’s rise to power on violence against civilians. The solid blue line to the left of the dotted “November 2017 coup” cutoff represents the average level of violence against civilians under Mugabe. The dotted line to the right represents the pattern of violence that might have taken place under a post-November 2017 Mugabe regime had the coup not occurred. The red line represents the actual level of violence against civilians under Mnangagwa.


A simple model of violence against civilians in Zimbabwe over time probably fails the second assumption of a regression discontinuity, that other potential explanatory factors do not also change as a result of the change. Regime responses to protest activity might account for some of the new violence: a November 2016 report that I co-authored with Otto Saki and Lawrence Woocher found that regime repression of opposition protests was one of two scenarios under which violence against civilians in Zimbabwe could escalate. Indeed, a high level of protest activity has characterized Mnangagwa’s first 15 months in power, including protests after the coup, protests against the ZANU-PF party’s disputed victory in the July 2018 elections, and the gas-price protests. The larger number of protests might explain some of the varying patterns of violence that we observe in Mnangagwa-era Zimbabwe.

The following graphic displays the effect of the coup on violence against civilians, controlling for–or subtracting–the effect of protest and riot activity during the time period. The grey dots represent the number of observed instances of violence against civilians over time, as displayed in the first graph. The solid black dots represent the number of instances of violence over time, minus the number of protests and riots during the same week. The graphic suggests that the average level of violence against civilians in Zimbabwe persisted after the coup.


These graphics display how much violence against civilians changes as time moves farther from Mnangagwa’s rise–the slope of the relationship–but we should be also be interested in whether this relationship is statistically distinguishable from zero. The graphs suggest a consistent trend of violence against civilians before and after the coup, but a statistical test can describe the coup’s effect (or non-effect) with greater precision. In the “dot-and-whisker” plot below, I show the coefficients and the confidence intervals for the main variables in the discontinuity model. The effect of any variable for which the blue line crosses the dotted black line is not statistically distinguishable from zero. 


I also include the regression table for the discontinuity analysis, for quantitatively-minded readers.


As the “dot-and-whisker” graphic and the regression results demonstrate, the effect of the coup is not statistically distinguishable from zero. Repression under Mnangagwa is more of the same.

The continuation of violence against civilians in Mnangagwa’s Zimbabwe has important implications for both the study of Zimbabwean politics and international policy responses to events there. International donors are tip-toeing towards a policy of re-engagement with the Zimbabwean government, after multiple decades of sanctions. These potential policy changes rest on the premise that Mnangagwa represents a qualitatively different style of leadership from Mugabe. Where violence against civilians is concerned, the empirical record points in the other direction.

Thanks to Patrick McSweeney for his comments on an earlier version of this research design.

<edited to correct image upload>