Time-Series–Cross-Section Methods

Time-series–cross-section (TSCS) data consist of comparable time series data observed on a variety of units. The paradigmatic applications are to the study of comparative political economy, where the units are countries (often the advanced industrial democracies) and where for each country we observe annual data on a variety of political and economic variables. A standard question for such studies relates to the political determinants of economic outcomes and policies. There have now been hundreds of such studies published TSCS data resemble “panel” data, where a large number of units, who are almost invariably survey respondents, are observed over a small number of “waves” (interviews). Any procedure that works well as the number of units gets large should work well for panel data; however, any procedure which depends on a large number of time points will not necessarily work well for panel data. TSCS data typically have the opposite structure of panel data: a relatively small number of units observed for some reasonable length of time. Thus methods that are appropriate for the analysis of panel data are not necessarily appropriate for TSCS data and vice versa. All of these types of data are a particular form of “multilevel” (or “hierarchical”) data. Multilevel data are organized by grouping a lower level observations in some meaningful way. Thus in the study of education, we may observe students, but students can be thought of as members of a class, and classes can be thought of as members of a school. It is often helpful to think of TSCS data as multilevel data, but with much additional structure. Political scientists have been familiar with panel data in the guise of election studies for well over half a century. TSCS data has become popular only more recently. Adolph, Butler and Wilson (2005) examined political science articles in journals indexed in JSTOR. They found relatively few uses of TSCS terminology before 1975, with an explosion of articles using terms related to TSCS analysis starting in the late 1980’s. They report that 5% of all political science articles from 1996 to 2000 made reference to TSCS or panel terminology and found approximately 200 empirical analyses using TSCS data during that period. Beck and Katz’s (1995) methodological discussion of TSCS data is the most heavily cited American Political Sciences Review article published since 1985.