For the presidential election he is right: There will not be enough of them to prove him wrong anytime soon. But there is more data on the individual states and the senate races . And there I guesstimate that his predictions are too correct: According to his numbers he should be wrong more often. That would mean that his assumed distribution is wider than the actual polling error and he should make predictions more aggressively. Yes, this is one of the points of making a numeric model: You can see trends and estimate the impact of events. In theory, a probabilistic model also gives you the chance to evaluate and compare different models. But for the latter, the usual problem of data analysis applies: There is never enough data.