We won’t know until the ballots are counted whether turnout or persuasion was critical in 2020. But with political scientists Seth Hill and Gregory Huber, I’ve released new research that tries to understand the relative influence of each of these approaches in the presidential elections of 2012 and 2016. In explaining the 2012 to 2016 shift, it turns out that persuasion mattered more in the places that swung most decisively to the GOP.
How we did our research
Determining the relative importance of turnout and persuasion is critical to both campaign strategy and to governing. How much should a campaign or a public official focus on energizing likely supporters vs. winning over possible swing voters? Research on U.S. elections doesn’t answer that definitively: The empirical record has been mixed, with evidence showing that each strategy, turnout and persuasion, shifted election results at different times and places. In the abstract, persuasion has a mathematical advantage, since persuasion means one fewer vote for the other side and one more vote for one’s own side. But over the past few decades, boosting base turnout has seemed to become more important, as there’s been a well-documented decline in swing voters.
But assessing which strategy will be more effective is challenging. Public opinion polls are relatively good at recording which candidates people say they support, but they’re not as useful in accurately measuring who actually shows up to vote. To tackle this, our research uses data from election administrators, combining states’ voter-file records on who voted in each election with precinct-level data on how specific neighborhoods voted. With this, we can calculate how many voters in each precinct fall into turnout categories, such as having been registered to vote in both 2012 and 2016, but voting only in 2016. In states where citizens register with a party — such as Florida, Pennsylvania and Nevada — we can also calculate how many registered Democrats and Republicans voted or failed to in individual elections. For other states, we instead counted how many citizens recently voted in each party’s primary, giving us a sense of which were more eager to show up.
For this analysis we focused on six key states. Four flipped from Obama to Trump: Florida, Michigan, Ohio and Pennsylvania. We also looked at Georgia, which voted for the Republican in both elections, and Nevada, which voted for the Democrat twice.
Let’s look at Pennsylvania
Some analysts see Pennsylvania as most likely to deliver the winning 270th electoral college vote. Obama won it by 5.4 percentage points in 2012; in 2016, Trump won it by 0.7 points. So did regular voters switch from Obama to Trump — or did different people go to the polls?
To find out, we first calculate how stable each precinct’s electorate was across the two elections. In other words, what fraction of the total ballots were cast by the same voters in both years?
Next, we break the precincts into 10 groups, based on what fraction of all the votes in 2012 or 2016 were cast by two-election voters, arranging them from the least to the most stable group of voters. If Trump won Pennsylvania either because of a surge of different voters — or because turnout dropped in Democratic areas — we’d expect the GOP to pick up more votes in less stable precincts.
We find that the average increase in votes for the Republican candidate is very similar in the stable and the volatile precincts. That’s generally what we found in the other states we looked at, too: Trump won more votes than Clinton both by persuading regular voters to join him and also by bringing out different voters.
Looking at all six states to see whether turnout or persuasion mattered more
For more formal estimates of how much turnout contributed to the shifting outcomes, we next used linear regression, a statistical model that allows us to see how well different measures predict an outcome. We continued to work with precinct-level data, fitting statistical models that predict the change in the Republican vote margin across the number of voters in 15 categories that we defined by party alignment, registration status, and voting in just one or both elections. For instance, one such category would include the number of a precinct’s Republicans who were registered in both elections but only voted in 2016. We can then see just how much these various turnout measures jointly predict precincts’ overall swings.
In ethnically and racially diverse states such as Florida, Georgia and Nevada — where the outcome didn’t change much from 2012 to 2016 — what changed the result most was turnout, or who actually cast ballots. In Florida, for example, the GOP gained an average of about 38 votes per precinct in 2016 compared to 2012. Our statistical model finds that changed turnout accounts for 26 votes per precinct, or 69 percent of the total shift.
But our three Northern states — Pennsylvania, Michigan and Ohio — shifted heavily to the GOP. While Obama won Ohio in 2012 by 3 points, for instance, Trump won the state by 8 points in 2016. And in all three states, we found that turnout shifts helped Democrats in 2016. In other words, Trump won because he persuaded Obama voters to cast ballots for him.
In 2016, persuasion made the biggest difference in the states that swung the most. For 2020, will those voters change their minds again?