How to predict election winners from a poll

Abstract

Suppose that we have k candidates in an election and that the top m winners will be elected. Assume that the voters can select up to m'(< k) candidates and that the votes are non-transferable. In the election polling, a pollster will normally report on the shares of votes received by the candidates, together with their standard errors. However, such statements cannot answer some basic questions, such as 'who will win the election' or 'how well will the m leaders in the poll perform in the final election'. In this paper, we use subset selection techniques to predict the winners. Furthermore, we construct predictive lower bounds for certain random variables to predict the election result of the leaders in an election poll.