Microeconomics and Macropolitics: A Solution to the Kramer Problem

Abstract

Estimation of economic voting models is complicated by the possibility that voters treat certain economic conditions as "politically irrelevant" and do not attribute responsibility for such conditions to the incumbent party. Kramer (1983) suggested that this phenomena could account for the discrepancy between micro survey and aggregate time-series estimates of the economic voting model. Statistical methods are developed for testing the Kramer hypothesis and applied to presidential voting data from 1956 to 1984. With proper treatment, the estimated individual level income effect based on pooled cross-sectional surveys is as large as that found in aggregate time series data, Ordinary regression estimates are shown to be biased by a factor of approximately seven. It is also shown that ordinary regression estimates tend to overstate "sociotropic" or national level economic effects. Nonetheless, even using consistent estimation techniques, sociotropic effects are still found, though they are slightly smaller than the individual level effects.