Causal Evidence for Theories of Contagious Civil Unrest

Many types of civil unrest, including protest, violent conflict, and rebellion, have been found to be subject to both inter- and intra-state contagion. These spillover effects are conventionally tested through the application of parametric structural models that are estimated using observational data. Drawing on research in methods for network analysis, we note important challenges in conducting causal inference on contagion effects in observational data. We review a recently developed non-parametric test - the "split-halves test"- that is robust to confounding and apply the test to replication data from several recent studies in which researchers tested for contagion in civil unrest. We find that about half the time findings in the published literature fail to replicate with the split-halves test. Across ten total replications, we do not see strong patterns in terms of which results do and do not replicate. We do, however, find evidence for general contagion in six of the replications, indicating that contagion is a prevalent phenomenon in civil unrest. As such, we recommend that researchers (1) use the split-halves test as a general-purpose robustness check for parametric models of contagion in the study of civil unrest, and (2) consider modeling contagion in research on civil unrest.