538 predicts a 2020 sized Harris victory, Georgia and North Carolina flip. THQ predicts a tight Harris win, mostly in the Rust Belt & maybe a NC grab? RCP predicts a tight Trump victory via Pennsylvania.
All 3 agree on Georgia going red and Michigan and Wisconsin going blue. Those states have held their colors firm for quite some time.
How much do uou pay for the silver bulletin? I’d consider signing up if I could know the price before I enter my email address.
$10 a month. I plan to cancel after the election.
Personally, Morris’ model at 538 had such a questionable take on Biden vs Trump it shook my faith in it. I still think it’s good as a polling aggregator, and its tools (such as the interactive map) are pretty solid, but I decided that $10 per month was acceptable to get access to a model I think has a more realistic take on the election.
538’s new model is untested, and for all I know it could be accurate. However, it still has some takes I find extremely unlikely.
Nate Silver’s model (in my opinion) paints a more realistic picture of what to expect; despite Harris’s qualifications and Democrats extremely high enthusiasm, it remains an uphill battle because of the electoral college.
Nate went over his issues with the model… And they seemed well founded. Trying to say Biden was a favorite at the end… Just didn’t seem like it had any defendable points. You have to question the methodology at that point. It refused to decline his chances in the face of declining polls and terrible news cycles. L
Nate’s model has a whole host of issues, not the least being his seeming lack of understanding of probability.
But don’t take the word of some random internet hobo for it, Nicholas Taleb has a whole paper responding to the fundamental flaws with the 538/Nate Silver kind of election forecast.
For one simple point, uncertainty in a binary prediction does not mean that week to week win probabilities swing wildly. It would instead mean the win probability converges to 0.5 for both options. Neither Silver’s model nor the new 538 model display this property (arguable the 538 model is closer), so their outputs cannot be interpreted as win probabilities.