An Alternative Statistical Measure for Racially Polarized Voting

Measurements of the existence and extent of racially polarized voting are often at the forefront of the evidence presented in vote dilution litigation. Previous models used in the measurement of racially polarized voting have been inadequate for a broad range of cases. The source of this inadequacy is that these models were not based upon assumptions about individual behavior. A new model, which is based on reasonable assumptions about individual behavior and can be approximated by a varying parameters model, is derived. By contrasting the new model with the other models, it is shown that both the correlation coefficient and linear regression can lead to inaccurate, misleading or incorrect conclusions about the state of racial polarization in the electorate. After providing a straightforward statistical test for identifying misspecification and showing how to obtain estimates for this model by the method of maximum likelihood, the model is compared with an individual level data set of an election where racially polarized voting occurred (1988 California presidential primacy). A new measure for the estimation of racially polarized voting is then proposed.