As a thought experiment, I built a little "naive model" that assumes all electoral votes save those in the 7 swing states are decided. Of the 7, assume odds are 50-50 for Harris or Trump. There are 128 possible outcomes. Assume that the outcome in each state is a coin flip and each state outcome is independent of the other. So, in this naive model, each of the 128 possible outcomes is equally likely. Nate Silver suggests such an assumption is bad indeed. He may be right. But in this naive model, Harris has a 55% probability of winning the electoral college, with an average across scenarios of 273 electoral votes. Silver Bulletin, 538, Decision Desk, & the Economist models are all quite close to this. Compared to the others, Silver Bulletin puts in more adjustments to give one more movement (which drives more clicks). But if one treats the naive model of 7 coin flips determining the election as the "null" hypothesis, all the other models seem indistinguishable from it. It's 7 coin flips from now until November. So, do something else (or better yet, help others to vote).
What if Musk has pivotted to making twitter/X a mouthpiece for Trump and a means to get inside the administration to enrich himself, perhaps via SpaceX or new businesses that meet MAGA approval?
A ha ha ha epistemic Bayesianism again. I recall an occasion about a decade and a half ago when you made similar points. Can you guess what it was? I'm thinking of Cosma Shalizi poking a little gentle fun at Rebonato for getting muddled over point estimates and confidence intervals.
As a thought experiment, I built a little "naive model" that assumes all electoral votes save those in the 7 swing states are decided. Of the 7, assume odds are 50-50 for Harris or Trump. There are 128 possible outcomes. Assume that the outcome in each state is a coin flip and each state outcome is independent of the other. So, in this naive model, each of the 128 possible outcomes is equally likely. Nate Silver suggests such an assumption is bad indeed. He may be right. But in this naive model, Harris has a 55% probability of winning the electoral college, with an average across scenarios of 273 electoral votes. Silver Bulletin, 538, Decision Desk, & the Economist models are all quite close to this. Compared to the others, Silver Bulletin puts in more adjustments to give one more movement (which drives more clicks). But if one treats the naive model of 7 coin flips determining the election as the "null" hypothesis, all the other models seem indistinguishable from it. It's 7 coin flips from now until November. So, do something else (or better yet, help others to vote).
Noah Smith's take on Elon Musk buying Twitter has to be his worst take of all time.
What if Musk has pivotted to making twitter/X a mouthpiece for Trump and a means to get inside the administration to enrich himself, perhaps via SpaceX or new businesses that meet MAGA approval?
80%, eh?
I can get there with Laplace's Pollster.
A ha ha ha epistemic Bayesianism again. I recall an occasion about a decade and a half ago when you made similar points. Can you guess what it was? I'm thinking of Cosma Shalizi poking a little gentle fun at Rebonato for getting muddled over point estimates and confidence intervals.