Party balancing and voting for congress in the 1996 national election

We extend Fiorina's balancing theory by developing and testing inferences from a simple and general spatial model of partisan preference in two institutional dimensions. This model contains a number of theoretical and methodological insights that have not yet been articulated in the growing literature on the origins of divided government. We focus here on the model's most basic implications, emphasizing first that the act of party balancing requires that voters have nonseparable partisan preferences across the electoral arenas of U.S. government. We show how it follows that the institution-specific partisan preferences of balancers are not defined on the metric that contains the partisan preferences of nonbalancers. We examine the empirical relevance of this result by estimating a variety of statistical models of congressional voting preference. Analysis proceeds in two directions, first with illustrations of a changepoint (and a resultant heteroskedastic error variance) problem in simple models of voter choice and then with results from more elaborate model specifications. The evidence weighs heavily in favor of the empirical relevance of party balancing in the 1996 congressional election. We conclude by pointing to a number of theoretical and methodological issues that merit further attention in future research on balancing.