This article at CNet is risible. This breathless quote at the end, “Score one for the quants, especially the most famous one of them of all, a statistician who is now, unquestionably not a one term celebrity, but a political prediction machine to be taken very, very seriously.”
Given that this was an either or decision, state by state coin tosses with weighted coin for past predilections would have been equally effective in forecasting the final result. The aggregation of state wide poll data will be effective until it no longer is. This is the problem with “big data” it can’t see the missing information; it suffers from the problem of induction. When shifts occur and it makes a big miss, it will garner press and pretty soon everyone will forget the model. They will jump on the next model and next ad nauseum. Who was the best economic prognosticator of the late 90’s. I guarantee he or she is not even in the top ten today. Journalists are story tellers and most of them have no command of history. They appear to live in an eternal present spinning theories and explanations like spiders building webs. But unlike spiders they do not eat what they spin, that is reserved for the people who believe them.