Different Slopes for Different Folks: ß heterogeneity in TSCS Data

• In Event: Poster Session: Political Methodology

Abstract

Time series cross-section (TSCS) data vary over time and space, but many TSCS methods assume homogenous data generating processes. The ß homogeneity assumption requires that the relationship between the independent and dependent variables be the same for all cross-sectional units over all time periods. Common TSCS models produce inconsistent estimates in the presence of ß heterogeneity. This paper considers the range of empirical issues surrounding ß heterogeneity. How can analysts identify ß heterogeneity? Tests for ß heterogeneity exist but are rarely used in applied political science. How can analysts model ß heterogeneity? Varying slope and intercept models can accommodate some types of cross-sectional heterogeneity but make important assumptions about the heterogeneity that complicate dynamic model specifications. Mixture models offer a feasible alternative to varying coefficient models but have important limitations analysts must consider. We conclude by proposing a series of best practices for TSCS analysts. Like the phenomena analysts are interested in studying, the approaches used to analyze these phenomena should vary depending on the features of the data.

Authors

• Clayton McLaughlin Webb, University of Kansas

• Joe Weinberg, University of Southern Mississippi