In the 2012 cycle, Princeton Election Consortium of Prof Wang and Nate Silver were my go to sites. They both predicted steady clear win for Obama over Romney. The drop in blue you see in the picture above is the tightening or the Romney Surge following the first debate by Obama.
In this cycle, PEC is still predicting a relatively stable race, and it too reports tightening of the race in the recent weeks. The predicted electoral votes for Hillary dropping from 335-340 range to 300-310 range. The meta margin (Dr Wang’s metric) fell to as low as 1.4% and now seems to be regressing to the mean.
But Nate Silver, the apostle of data journalist, the one who preached pure unadulterated data, the one who argued punditry is pollution as far data is concerned, is doing all kinds of models. Adding fundamentals, national-poll to state-poll correlations, pollster ratings etc etc and coming up with not one, not two, but three predictions. The polls-only model is the original 538 model, it looks like. May be updated with trust worthiness coefficients to pollsters to prevent gaming the model. The polls+plus model, (plus what? plus punditry pollution?) swings wildly. Today it briefly showed Trump winning odds at 51% for a few hours before sending Hillary on top.
But this is what gets me, the plot displayed does not show the brief moment when it predicted Donald going over 50%. It edits history. It changes past predictions. For a scientist, once you made a prediction it is carved in stone. You don't get to take it back, and replace it with a different history of predictions. There is no <tt>git --force-push</tt> option here. You can’t edit history.
The whole purpose of now-cast seems to be to garner click baits and nothing more. The Republican candidate seems to be the click-bait candidate, as mentioned in another diary today. Well, Dr Wang is a legitimate prof and does not need click baits to put food on the table. PEC is the only thing to click. You wanted clicks Nate, that is what I ‘m going to deny you.