Quantifying and Contextualizing Violent Collective Action Event Datasets

Abstract How does newspaper-based event data compare to a government data source? While scholars have long recognized the importance of and biases present in newspaper-based event data, few studies have compared newspaper reports with official government data to better understand the severity and impact of such biases. We develop this comparison in the context of riots, a form of violent collective action that represents an important middle ground between peaceful protests and protracted civil conflict. Using newly collected police precinct-level government data from India, we compare these data to a high-quality newspaper source. Though similar at the aggregate level, newspaper riot reports correlate poorly with government data at the local level. We model the frequency of newspaper and government riot reports based on literacy, location, and other demographic characteristics to better understand the discrepancies between these two sources. We conclude that newspaper riot data does partially reflect aggregate riot trends, but the newspaper editorial process also plays an important role. Government data is better for within country comparisons and for analyzing event trends over time. Our findings suggest that using collective action event data from both sources may help ensure that results are not driven by biases in either data source.