Voting turnout is affected by many things; or why journalists need to learn multivariate statistics

downloadMy good friend Tova Wang sent me this headline from the Columbus Post Dispatch:

Early Voting Hasn’t Boosted Ohio Turnout

In support of this headline,  reporter compares turnout in only three elections, only statewide, and only in presidential contests.  This is analysis is as unrevealing–and potentially misleading–as imaginable.

The key to understanding a complex process like voter turnout is to try to maximize, to the degree feasible, variation and covariation among all the important causes (variables).  Political scientists often consider dozens or more different influences on turnout and estimate highly sophisticated multivariate models.

But even a relatively simple exploration can be done far better than the one conducted by the Dispatch.

Let’s start with the presidency.  There are obvious reasons that the nation, and the world, focuses on the American presidential election.  It is almost always the most consequential election held in this country for the most powerful and influential political leader in the world.

But all these reasons are why the presidential contest may be the worst election in order to discern the turnout effects of something like early voting.  In the face of a billion or more dollars in campaign spending, blanket media coverage, and organizational mobilization, the impact of early voting is going to be small.  We may be able to uncover turnout effects, but the context makes it difficult.

At a bare minimum, compare midterm and presidential contests, and if at all possible, include off-cycle elections.

Next, even if limited to a study within one state, there is no good reason not to compare trends across counties.  In a large, heterogeneous state like Ohio, not only do the conditions for voting change across the state, but the voters change as well.

The reporter seems to recognize that African Americans in 2008 responded differently to the Obama/McCain contest in 2008 than they did to the Kerry/Bush campaign in 2004.   And the reporter notes that, due to legal uncertainties, the hours and days of early voting varied across counties in 2012.

So why not compare turnout effects across counties?  By not doing so, the reporter–whether realizing it or not–assumes that the all voting rules and procedures in the state of Ohio are identical and more importantly, that all Ohioans are identical insofar as they responded in different years to different candidates and to different election laws and procedures.

Two esteemed political scientists are quoted in the article and seemed to try to educate the reporter on these points.

Paul Beck’s quote starts with a general point which I think does not accurately reflect the state of the literature on early voting at this stage, but more important is the end of Beck’s quote, where he highlights the most consequential reasons that turnout may be higher or lower:

“People who vote early are people who are typically going to vote anyway,” said Paul Beck, a political science professor at Ohio State University. “So, early voting hasn’t really succeeded in turning out more people to vote. We’ve made it a lot easier to vote, but on the other hand, some people are very discouraged about politics and might not care how easy it is to vote.”

John Green’s quote, on the other hand, is exactly on point in my view:

“If all things are equal, early voting would increase voter turnout, but all things aren’t equal,” said John Green, a political science professor at the University of Akron. “But there are many factors in each election: the closeness of the race, the excitement to vote for a candidate or the degree of anger in the electorate.”

I could not have put it better.  Early voting may not have increased turnout in Ohio, but without at least considering these other factors, the title and thrust of the story are not accurate.