Read on Twitter
Spent the weekend looking at AP "call" times from 2016. Here are three graphs. (I hope the thread stays together.) The first simply shows when the polls closed and when the AP declared a state for a candidate.
The second graph is a scatterplot that shows the time (in hours) it took to declare a state (y-axis) against the ultimate margin of victory by the winning candidate. Intuitive result overall, but still significant variability.
The third graph is a scatterplot that shows the time (in hours) it took to declare a state (y-axis) against the % of ballots cast by mail. Counter-intuitive result: the more mail ballots, the faster the call (with lots of variability).
I'm not suggesting that more ballots _cause_ results to come in faster, or that the same pattern will hold in 2020. I _am_ suggesting that if you want to anticipate when a state's likely winner will be clear, you need to know more than just what % of ballots are cast by mail.
