A statistical model for Condorcet efficiency based on simulation under spatial model assumptions

This essay evaluates six single-winner, multicandidate electoral systems with respect to their tendency to choose Condorcet candidates. To this end I calibrate a logistic multiple regression model from Monte Carlo simulations, based on a multivariate normal spatial model, in which I vary the number of candidates, number of dimensions, correlation structure, and relative dispersion of candidates and voters. I investigate additional spatial-model variations by comparing further simulation results with predictions of the basic statistical model. The results suggest that for many electoral systems, Condorcet efficiency would increase with perceptual uncertainty of candidates' positions and would be low in a polarized society. Of the voting systems studied, approval voting and the Coombs systems appear least sensitive to variations in assumptions.