Political scientists often wish to classify documents based on their content to measure variables, such as the ideology of political speeches or whether documents describe a Militarized Interstate Dispute. Simple classifiers often serve well in these tasks. However, if words occurring early in a document alter the meaning of words occurring later in the document, using a more complicated model that can incorporate these time-dependent relationships can increase classification accuracy. Long short-term memory (LSTM) models are a type of neural network model designed to work with data that contains time dependencies. We investigate the conditions under which these models are useful for political science text classification tasks with applications to Chinese social media posts as well as US newspaper articles. We also provide guidance for the use of LSTM models.

No abstract

No abstract

No abstract

We theorize and measure a situational self-censorship that varies across spatial-temporal political contexts. Schelling’s insight that distinctive times and places function as focal points has generated a literature explaining how activists coordinate for protest in authoritarian states. Our population of interest is not activists but ordinary citizens, who, we assume, are risk-averse and prefer to avoid trouble. Focal points rally activists for political expression. By contrast, we theorize, ordinary citizens exercise greater than usual political self-censorship at focal points, to avoid punishment as troublemakers. We test our theory by leveraging geotagged smartphone posts of Beijing netizens on Weibo, China’s version of Twitter, to estimate precisely if, when, where, and how citizens engage in political talk. We use a difference-in-differences strategy that compares smartphone political talk at and away from focal places before and after focal times. We find netizens self-censor political talk significantly more at potentially troublesome spatial-temporal focal points.