On measuring two-party partisan bias in unbalanced states

Abstract: Assuming that partisan fairness and responsiveness are important aspects of redistricting, it is important to measure them. Many measures of partisan bias are satisfactory for states that are balanced with roughly equal proportions of voters for the two major parties. It has been less clear which metrics measure fairness robustly when the proportion of the vote is unbalanced by as little as 60% to 40%. We have addressed this by analyzing past election results for four states with Democratic preferences (CA, IL, MA, and MD), three states with Republican preferences (SC, TN, and TX) and comparing those to results for four nearly balanced states (CO, NC, OH, and PA). We used many past statewide elections in each state to build statistically precise seats for votes and rank for votes graphs to which many measures of partisan bias were applied. In addition to providing values of responsiveness, we find that five of the measures of bias provide mutually consistent values in all states, thereby providing a core of usable measures for unbalanced states. Although all five measures focus on different aspects of partisan bias, normalization of the values across the eleven states provides a suitable way to compare them, and we propose that their average provides a superior measure which we call composite bias. Regarding other measures, we find that the most seemingly plausible symmetry measure fails for unbalanced states. We also consider deviations from the proportionality ideal, but using it is difficult because the political geography of a state can entangle responsiveness with total partisan bias. We do not attempt to separate intentional partisan bias from the implicit bias that results from the interaction of the map drawing rules of a state and its political geography, on the grounds that redistricting should attempt to minimize total partisan bias whatever its provenance.