Imagine you train an algorithm to influence the voters to win the election. What would be the objective ? Maximize the probability that your candidate wins. This does not imply winning the popular vote, nor maximizing the margin of the electoral vote.