In their article “Don’t Read Too Much Into Early Voting,” Five-Thirty-Eight demonstrates that early voting is a moderately good predictor of final voting.
This is the relationship between early voting and final voting in 13 states for which Five-Thirty-Eight was able to find data:
The correlation is 0.4, which is a medium strength effect size by conventional standards. It implies that a 1 point increase in early vote share predicts a roughly .45 point increase in final vote share. (I re-ran the data to get the unstandardized effect size).
Five-Thirty-Eight calls this relationship “very weak” because they assume that their audience doesn’t know anything about statistics. They also list various theoretical reasons that might lead us to expect that early voting doesn’t predict final voting. Yet, it does.
(The relationship is also called “messy” possibly because it’s p-value is .18, but what can you expect with a sample size so small?)
Five-Thirty-Eight’s analysis is pretty weak due to its small sample size and the fact that it looks at voting in a single year instead of how changes in early voting across elections predict changes in final voting. Still, it’s the best Five-Thirty-Eight could do.
The Alternative Hypothesis will probably release a more rigorous analysis in the future, but that won’t be done in time for the election, so, for now, this is the best we’ve got.
Currently, early voting has Trump winning in Florida.
Trump is losing in Ohio in terms of early voting. He’s doing better than Romney did in Ohio at this time, but he is getting less than half of the early votes. This state was not included in Five-Thirty-Eight’s data, but the data does exist, and it shows that Ohio’s early voting was unusually unpredictive of final voting in 2012. Obama got 36.7% of the early votes (excluding non two-party voters) and won the state. Real Clear Politic’s average of pollsters has Trump winning Ohio, and his lead is gaining with time. Trump will probably win Ohio.
(Note: I am dealing with the absolute voting totals, whereas the article showing positive results in early voting for Trump is looking at CHANGE in turnout.)
Increasingly, it seems like the election will come down to Pennsylvania. Trump is losing Pennsylvania in the polls right now. However, in the Pennsylvania primary, which occurred before Cruz dropped out, RCP’s polling average underpredicted Trump by a whopping 14 points.
There is reason to think that general election polls in Pennsylvania may be dramatically off as well. Pennsylvania saw the 8th largest increase in GOP primary voters (+88%), and 26th largest decrease in Democrat turnout (-31%) despite being a closed primary. For the GOP, the increase in primary turnout in 2016 was 3.27 times larger for open than for closed primaries. So for Pennsylvania to have such a large turnout increase despite being a closed primary is especially promising.
Contrary to another wrong article from Five-Thirty-Eight, The Alternative Hypothesis has shown that these changes in turnout do predict general election turnout. This may be why the polls were so off in Pennsylvania: they are not accounting for this change in turnout which is abnormally huge in Pennsylvania.
There are two takeaways from all this: 1) this election could be very close, and 2) Five-Thirty-Eight isn’t all it’s cracked up to be.