I actually made a linear regression model using the number of registered Republicans, Dems, and Indys in PA in every election since 2000, plotted against the raw vote margin, weighted the variables to achieve a near 1.0 correlation - and the numbers for this year showed about a 250k margin for Trump.

Now, when I tried doing the same thing for Florida, it showed Trump winning very narrowly, which didn’t make sense (FL likely right of PA), probably because it’s tough to do LR with only 5 inputs (5 elections between 2000-2016), but all I’m saying is that this is totally plausible at least.