Read on Twitter
Nate came up during the Obama elections obviously, elections with pretty certain outcomes by the end. The only state he ever got wrong was Indian 2008, which wasn't THAT MUCH of an upset
However, 2016/2020 were/are much more uncertain elections than 2008/2012. There was an extremely high chance that Nate was going to get at least one state wrong in 2016 regardless of how perfect the model was.
The same in 2020. The "getting each state correct" heuristic treats election forecasting like a game of Yahtzee rather than a continuous probabilistic practice
It treats forecasting in a binary way, and if you listen to Nate talk about why he started blogging, it was to avoid the ternary lens that pundits had (SAFE D, TOSSUP, SAFE R). All elections fell into one of those categories
When you look at this old map and compare PA to CA, your eyes naturally put PA in the same blue bucket as CA. And when Trump wins PA, you feel lied to
Conversely, the new model FORCES you to actually specifically click on PA or CA, and you realize that Biden's statuses in CA and PA are meaningful different
No longer do your eyes automatically forces PA into the same SOLID D bucket as CA. You see the numbers insted
No longer do your eyes automatically forces PA into the same SOLID D bucket as CA. You see the numbers insted
The dot visualization also displays there are worlds in which Trump wins PA, 26 out of 100. When you SEE the number on the face, it seems tiny, but when you see the mass of dots under Trump's detached head, your eyes say "Hey, there are a lot of scenarios whereby Trump wins PA"
