Regression in Space and Time: A Statistical Essay

Regressions on data jointly structured in space and time, commonly referred to as the pooling of cross sections of time series, can be formidable both in the strength of their design properties and in the number of special statistical problems encountered with them. This essay deals briefly with the potential applications of pooled design and more extensively with the special statistical problems commonly associated with analysis in space and time together. Four estimators-ordinary least squares, least squares with dummy variables, error components, and an adaptation of Box-Jenkins ARMA models to the pooled estimation problem are reviewed, with an effort to suggest where each may find application in political science research. The four estimators are then illustrated by analysis of the regional dynamics in party issue polarization over issues of racial desegregation in the U.S. House of Representatives.