Tests of whether candidate A or B would fare better against C—and other polling enrichments to open up party strategy
Our main setting is a single-winner election contested by two major parties. A goal is to find whether candidate A or B of one party would be the stronger opponent running against candidate C of the other party. A poll (as many polls do) asks the same set of respondents to choose both between A and C and between B and C. The classical McNemar test and two novel extensions thereof can evaluate the difference between A and B regarding their strength against C. The first extension treats the case where some respondents answer one question but not both, a condition that the McNemar test itself does not handle well. The second covers the case where respondents who do not answer a question are probed further to see if they lean toward either candidate. We provide empirical examples related to the 2016 US presidential election. The A-versus-B strength difference vis-à-vis C can be statistically significant even if the difference between A’s showing against C and B’s showing against C is small. We argue finally that other insights helpful to party and candidate strategy can also emerge from novel augmentations to polling practices.