No, his model 4 years ago that I can remember did not have the "trend" effect. You're confusing house effect with this trend thing. It's the reason why PEC and upshot are so different than 538 right now. Basically, silver's arguing there's a trend in Trump's favor that is hiding the real number of a poll. He's taking pollsters with small house effects and very reliable and still shifting the number down a lot.
I'm sorry, but if 9 out of 10 polls show Hillary winning by +5, there's no fucking way she's losing by 1 to Trump without polling being completely wrong. And if it's going to revert back after the DNC cuz the "trend" goes the other way, then he's simply overfitting his model and should be ashamed.
Maybe I'm wrong because I do not recall this being in there 4 years ago (house effects and reliability always were but not trend). Had it, Romney would have been a lot higher than he ever was, especially after the RNC.
Silver's polls-only model for November is basically 50/50, so yeah, it's way off compared to upshot or Sam Wang.
Just look at the comparison chart here:
http://www.nytimes.com/interactive/...lights&contentPlacement=2&pgtype=sectionfront
His models' vies of Oh, Nv, Pa, Fla, Co, Mi, Minn, are waaaay different than everyone else.
As I said earlier, this is not an issue of wanting to believe something. For the first time, 538 is different than everyone else...so either Silver's latest model has discovered some new thing to measure the election that is more accurate...or he input something making him less reliable and everyone else is still correct.
I'm sorry, but I don't see a valid explanation of how
right now Nate Silver's model can justify the argument that all the swing state polling is wrong...which is literally the argument.
edit: Here's the clearest way to argue against Silver's model. It gives California a 96% chance being Dem. There is no fucking way it's 99% (which everyone else agrees). Now, you might think that 3% is nothing...but on a model this intricate to give Trump even a normal outlier chance in Cali (even after black swan events are taking into consideration) is fucking ridiculous. That alone demostrates his model is weak, IMO. vermont 88%??? He's giving a 1 in 4 chance to Trump in Oregon? Is the model high???