On a Surprising Result of Two-Candidate Election Forecast Based on the First Leadership Time

Abstract

This is a simple but provocative note. Consider an election with two candidates and suppose that candidate A was the leader until counting n votes. How to use this information in predicting the final results of the election? According to the common belief the final number of votes for the leader should be a strictly increasing function of n. Assuming the votes are counted in random order we derive the Maximum Likelihood predictor of the final number of votes for the future winner and loser based on the first leadership time. It appears that this time has little effect on the predicting.