Some of you may have noticed that Nate Silver has become an insufferable bad-hot-take machine, obsessed with driving clicks by talking about the hour-by-hour minutiae of his team’s forecast models. We now have published scholarly evidence that this may be bad for democracy.
In a very robust-sounding paper*, which I admittedly have not read in its entirety, researchers suggest that:
an increase of 20% over even odds in this study lowered voting by 3.4%, and an advantage of 40% [around what Clinton had in the 538 2016 forecast] lowered the voting by 6.9%. If as the evidence provided above suggests, Democrats were more affected [than Republicans]…, probabilistic forecasts may have a strong enough effect on turnout to constitute an important factor influencing the election.
Similar to the observer effect in physics, where you change the result by measuring it, the widespread consumption of these forecasts may well undermine the very things they seek to report. Importantly, this effect was not seen in research subjects who were only shown poll results.
Here’s a good thread on the paper!
Unlike polls that show candidates’ expected vote share, prob. election forecasts convey the estimated probability that a candidate will win. Problem: folks don’t understand probabilities. This paper demonstrates severity of this confusion, and its political consequences. (2/n) pic.twitter.com/MEt9GoUhZs
— Jonathan Mummolo (@jonmummolo) February 28, 2020
When a draft of the paper was published in 2018, Nate Silver dedicated an entire episode of his podcast to “rebutting” it. You can see the rebuttal of the rebuttal here. I’m inclined to believe the authors, for Nate Silver has become a terrible hack. Apparently he was more recently urging scholarly journals not to publish this piece, which is super classy.
*This link is to last year’s draft of the paper; the paper itself was published this month, and is behind a paywall.