Why Impotus will resign

[ergative]

538 is good- if you check Nate Silver went above and beyond in 2016 to explain why Trump actually had a "good" chance as compared to the networks' silly statements.

My Jean Dixon comment was aimed specifically at single poll probability statements. By way of contrast, the Covid case/death models always provide a confidence interval.

Yes, their podcast has a recurring feature where they talk about specific poll results that garner a lot of news attention, and ask, 'good use of polls? or bad use of polls?' The general consensus is that the best possible use of a single poll would be for political messaging in a volatile news cycle to get your name to the top, or as a particularly salient example of a trend that's also present in lots of other polls. Otherwise, obsessing over a particular surprising poll result---especially if it contradicts lots of others---is almost always a bad use of polls.

In their election forecasts they also always provide not just confidence intervals but what looks like distributions of possible outcomes over many many simulations of their forecasting model. I really enjoy their data-driven approach--not just because it's accurate, but because they talk about how it works, what assumptions they make, and how it should be updated in light of its previous performance accuracy.