4 Comments

re: modeling approval w/polarization as a feature - how imbalanced is that metric? I'd think that most of the training data would come from low-polarized cycles (this Washington Post article has a pretty illustrative gif of just how recently house/prez vote shares have become correlated: https://www.washingtonpost.com/politics/2021/02/19/2020-saw-least-split-ticket-house-voting-decades/).

Expand full comment
author

Hey Mark

The polarization variables runs from 11% to 8% from 1950 to 1990, and then from 8 to 4% from 2000 to 2020. One thing I want is a better measure with daily data — from Gallup, I have the daily observations of the partisan difference in POTUS approval, and curious to see if using that instead improves the model. The in-sample R2 right now is _pretty_ good, at 0.75, but, like, for elections we try to get higher juice...

Expand full comment

I think the reason for Biden's low numbers is Propaganda 24/7 on talk radio and Fox. I think it works. I think the reasons for that success are that USians are a nation of people with low critical thinking skills, low public education (deliberate policy choices by GOP), and addicted for decades to media. John Perry Barlow once pointed out that TV is the mass hysteria making machine that takes the place of the medieval church, spouting falsehoods that people must believe - or die.

Expand full comment
author

I certainly think that that's part of the story.

Expand full comment