Jul 9, 2016

How does U.S. voter turnout vary between election years?

In a previous post, I examined the voter funnel for the 2014 U.S. midterm elections. However, having discovered an 11% difference in registration rates between native and naturalized citizens, I wanted to know whether that difference persists during Presidential elections.

Using the same Census dataset, I computed the funnels for every election between 2004 and 2014 (the only years for which data is available). In both this chart and in all the subsequent analyses in this article, the “funnel” is assumed to be Resident → 18+ → Citizen → Registered → Voted.

Each bar represents the number of people who have met each of the prior criteria. For example, the green bar for “Registered” represents the total number of 18+ U.S. resident citizens who have registered to vote in that election year.

Note that the exact shape of this funnel will vary depending on the order in which you apply the citizenship and age criteria. However, both methods will demonstrate that age is a much stronger mechanism of disenfranchisement than citizenship: in 2014, there were 21.8M non-citizen U.S. residents and 74.4M minors. This particular finding defied my intuition and forced me to reconsider the arguments for child suffrage most memorably argued in The West Wing.

The same general trends persist from year to year, but there are indeed major differences between Presidential and midterm election years. To better understand these differences, I asked the following two questions:

At each stage of the funnel, what is the “conversion success rate”?
For all those people who didn’t vote, what was the most salient reason?

“Turnout” in this case means that you successfully cast a vote on election day. As with the bar chart funnels above, note that each conversion rate is conditional. That is, a 70% registration conversion rate means that 70% of 18+ U.S. resident citizens registered to vote, *not* that 70% of all U.S. residents registered.

Observations based on these conversion rates:

Both registration and turnout rates are significantly higher in Presidential election years than midterm election years.
Turnout rates are much more volatile, suggesting that voters care more about Presidential races, Presidential campaigns are better at generating voter turnout, or both.
If we do a simple multiplication of these conversion rates, the total funnel had a 29.4% success rate in 2014. The Presidential turnout factor yielded a success rate of 46% in 2012.

To compute these metrics, I divided the number of people who dropped off at any given point in the funnel by the total number of people who didn’t vote. For example, in the 2008 election, 15M people registered but failed to show up on election day. Thus, of the 168M U.S. residents who did not vote in 2016, 15M/168M = 9% didn’t vote because the turnout criterion was prohibitive. I recognize that this is an oversimplification because some people may have chosen not to register because they knew the turnout criterion was prohibitive, but this metric is a decent approximation.

Observations based on these numbers:

Across election years, age and registration are the primary reasons that U.S. residents do not cast ballots.
If we examine any individual non-voter, there’s a ~50% chance that they didn’t vote because of circumstances which were at least nominally in their control: registration and turnout. I recognize that for many people, there are harsh realities which make these steps all but impossible, but it’s sobering to realize that the primary barriers to truly universal suffrage are behavior, not identity.

My concluding question of my last post was whether naturalized citizens register at higher rates in Presidential election years. Based on the last few Presidential election cycles, the answer seems to be yes, but still at much lower rates than their native counterparts.

All of the data use to prepare these charts and analyses is available in this publicly comment-able Google Spreadsheet.