,
5 tweets,
3 min read
Read on Twitter
I just undertook a mammoth effort for @TheEconomist's data team to determine what would have happened in the 2016 election had every eligible voter actually turned out. In this Medium post (and the tweets that follow) I explain how I did it. THREAD 1/5 medium.economist.com/would-donald-t…
2. See this thread for more about the piece.
We used a statistical method called multi-level regression and post-stratification (Mr P) to turn national-level US election polling into estimates of state-level voter behavior in the 2016 election.
We used a statistical method called multi-level regression and post-stratification (Mr P) to turn national-level US election polling into estimates of state-level voter behavior in the 2016 election.
3. Mr P enables us to break down the relationship between demographic variables (such race and education) and voting habits at the state level. You can imagine that we could figure out the election for any combination of turnout and partisan loyalty for any groups (like so). 
4. The technique was not easy—but it was rewarding. We chose to explore voter habits in a fully Bayesian context, which taught us a lot about both American political behavior and statistical programming. Here is what some of the computer code looks like (imagine 2000 more lines).
5/5. In the end, we would do it all again. The finished product (partially visualized here) is more rewarding than anything else I have worked on for the newspaper. Again, here's the methodology write-up: medium.economist.com/would-donald-t…
