Protest: Random or Contagious?

This paper tests the argument that outbreaks of protest occur contagiously over time against the alternative expectation that protest occurs randomly over time. Using univariate ARIMA models, it analyzes a time series of protest in the postwar United Kingdom, consisting of the number of day/locales of protest events. The (logged) data are probed for three periods of temporal aggregation: annual, quarterly, and monthly. Five principal findings emerge: (1) The best yearly model was (0, 1, 0); the best quarterly and monthly models (0, 1, 1). Thus, annual protest displays a random walk with a slight upward trend, while quarterly and monthly protest display, after differencing, a first-order moving average process. (2) Nonstationarity in protest, indicated by the differencing of protest, appears in all three temporal aggregations. (3) Cycles appear very weakly with quarterly aggregations and not with monthly aggregations. (4) The 1950s had less protest than the 1970s; the 1960s were a time of increasing protest activity. (5) Forecasts of protest generated by all ARIMA models investigated were poor. We conclude that autocorrelation effects are indeed present in British protest and that, therefore, protest actions at successive time points are not completely random but are interdependent. Finally, this paper explores the implications of these findings for theories of protest.