A Latent Measure of Political Protest

In recent years, scholars have developed a number of new databases with which to measure protest. Although these databases have distinct coding rules, all attempt to capture incidents of social conflict. We argue, however, that due to a variety of sources of measurement error, subjective coding decisions, and operational specifications, no single indicator of protest adequately measures how much protest exists in a given place at a given time. As a result, empirical studies that employ these measures yield inferences with limited generalizability. To increase the generalizability of the empirical findings, we suggest using an Item Response Theory (IRT) approach to estimate a latent dimension of protest using nine different protest measures that vary in their operational specifications as well as their temporal and spatial coverage. The estimates of the IRT models are used in two ways. First, to demonstrate how existing measures differ, the IRT’s item estimates are used to compare the nine measures of protest based on their degree of difficulty (the quantity of latent protest required to observe a ‘1’ in the data) and their ability to discriminate (the speed with which changes in the latent quantity of protest affect the probability of observing a ‘1’ in the data). Second, the estimated quantity of protest is applied to both monthly and yearly models of authoritarian breakdown. The results demonstrate that the latent protest variable increases the out-of-sample classification of authoritarian breakdown events; and improves in-sample prediction relative to existing global protest variables. Our study illustrates the potential value of modeling a latent dimension of protest rather than solely relying on observed indicators.