Thread by @dhopkins1776

🚨 New article out 🚨

Excited that my new article "Not by Turnout Alone: Measuring the Sources of Electoral Change, 2012-2016" with @seth_j_hill & Greg Huber is out today in @ScienceAdvances:

advances.sciencemag.org/content/7/17/e…

It looks at a question about elections which is both important and overlooked—when we see changes in party support from election to election, are they driven by changes in who turns out (composition) or in which party consistent voters support (conversion)?

2/

The U.S. uses the secret ballot, so this question is deceptively hard. But we compiled individual-level voter file data from 2012 and 2016 for 6 key states (FL, GA, MI, NV, OH, and PA), and merged it with precinct-level election returns. So we can estimate shifts 2012-2016.

3/

(It's amazing, by the way, that many months worth of data cleaning & crunching can get summarized in a single tweet.

And sincere thanks to the many election administrators, colleagues, and research assistants who helped us put together all necessary the data.)

4/

What's the lesson, what's the takeaway?

The relative influence of composition & conversion in explaining Trump's 2016 win varies by state.

But conversion mattered more in states like PA, MI, and OH where the vote swing was largest.

5/

We first examine vote shifts (y-axis) by the proportion of a precinct's voters who voted in both 2012 and 2016 (x-axis).

There's some state-to-state variability, but not much (if any) evidence that Trump does consistently better in 2016 in places with more variable turnout.

6/

Of course, it matters which side is turning out. Here, we show the change in the GOP vote margin (y-axis) by the change in the GOP turnout advantage (measured either via party registration or primary participation).

7/

Turnout clearly matters. Trump does better relative to '12 in places where relative GOP turnout was up.

But there's a ton of variation turnout advantages can't explain. And look at states like OH, where the GOP does better even in lots of precincts where turnout moves D.

8/

So we turn to statistical models to decompose 2012-2016 shifts into conversion and composition.

In states like NV, the pro-Trump shift 2012-2016 was driven by changes in turnout.

Elsewhere, Trump is outperforming Romney primarily through converting consistent voters.

9/

If you are reading this thread and thinking, "I wonder what it looks like in [COUNTRY X]," you are welcome to get it touch. I'm *very* curious how this varies not just across states or elections but across countries and electoral systems.

10/

If you want to explore the data, check it out here: dataverse.harvard.edu/dataset.xhtml?…

("it"? "them"? if I am cancelled for this tweet, nice knowing you.)

11/

Key to add: we are not the only people interested in this problem, and there's a lot of data we don't have access to--so be sure to check out the work of @yghitza and @jon_m_rob detailed here:

https://twitter.com/jon_m_rob/status/1131204654619156480

12/

Here's a write-up of this research for @monkeycageblog from October 30th, 2020: washingtonpost.com/politics/2020/…