Grading pollsters purely on their topline performance in recent elections leads to some weird problems under the hood — like giving a black-box firm that said Trump would win 30% of black voters an A- rating. Helpful to note the limitations with @NateSilver538's grade ratings
If you took a test and got like 10 questions wrong, but the teacher picked the 1 you got right to actually grade, are you actually a good student?
The point is not that the ratings are bad, but they're blind to data quality issues -- if you're getting good toplines because sampling and weighting errors are canceling each other out, it's fair to question whether the grade actually reflects the quality
Political journalism will still be better off for Nate weighting his averages by past predictiveness in 2008. It's a good innovation. We have found that it helps in our own aggregating, too. But the letter grades mask a lot of other nuances that I think is rather important.
One final note: Assigning grades for pollsters spreads the false expectation that firms will generate hyper-accurate predictions of elections, often better than theoretically possible! Context on historical sampling/non-sampling error gets completely lost. gelliottmorris.substack.com/p/lower-your-e…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
sorry, but this is literally the ideology of competitive authoritarianism
coincidentally, the people walsh thinks are qualified to vote are disproportionately white elites who think like him
there's a legacy of racism angle here, but for now i'll just share my blog post on how public opinion scholars have thought about the "quality" of a person's attitudes from the last time prominent conservatives proposed restrictions for restriction's sake gelliottmorris.substack.com/p/democracy-is…
#NEW A study of movie theater showings of the racist film Birth of a Nation shows that lynchings rose 5x in the places it was shown, while KKK activity shot up for generations after. Our piece on @ProfDesmondAng’s paper. A long legacy of violent hate. economist.com/graphic-detail…
This graph is pretty terrifying in how it reminds you how easy it is to manipulate people with media
.@ProfDesmondAng’s modeling found that counties that had screened The Birth of a Nation were nearly 70 percentage points more likely to have a Ku Klux Klan chapter (Klavern) 2 decades later (around 1935). Between 2000-2019, the effect was still detectable: an 18 %age point jump.
The optimal strategy when punditing on more than two options is to give a 40% chance to the top scenario and get everything else as close as possible. Summers should have gone with:
In a binary bet, eg between inflation and no inflation, it’s optimal for a pundit to assign a 60-40 split between the options. That way they’re usually right, but if they’re wrong they get to say “I told you so” and are usually less wrong than other people.
This is probably a horrible way to make money, but it’s a reasonably good strategy for relative media credibility if you kinda know what you’re talking about but haven’t developed a good model to approximate reality
Alexander knows what he's talking about here. Our polling with YouGov finds that ~12% of adults say they've received 2doses of the vaccine, which is the same share the CDC reports. Don't see much evidence of response bias, except maybe some expressive hesitancy from Republicans.