Do Voter ID Laws Change Turnout? Why the Evidence Is Contested

Few election policies provoke stronger feelings than voter identification laws. One side says they prevent fraud and build confidence. The other says they keep eligible people from voting. Both sides make claims about turnout, which can be tested, and testing them is much harder than either side usually admits. This article takes no side. It asks what the data can and cannot show.

The basic problem

States are not randomly assigned to adopt laws. Those that do differ in demographics, politics and how competitive their elections are, and those differences affect turnout by themselves. A simple comparison of turnout between states with and without strict ID rules is therefore badly confounded.

Researchers use a difference-in-differences design instead. It compares how turnout changed in states that adopted a law with how it changed in states that did not.

A large study finds no negative effect

Cantoni and Pons used a panel of 1.6 billion observations from 2008 to 2018 and found that strict ID laws had no negative effect on registration or turnout, overall or for groups defined by race, gender, age or party (NBER). Their most demanding specification gave a point estimate of about −0.1 percentage points for turnout, with a 95% confidence interval of roughly −3.0 to +2.8 points.

Read that interval carefully. "No significant effect" is not the same as "exactly zero." The interval still allows effects of a few points in either direction, and in a close race a gap of one to three points of turnout is not trivial. The two-proportion confidence interval calculator and the difference of two proportions calculator show how such intervals shrink as samples grow.

No significant effect does not mean no effect. It means the data cannot rule out a small one.

Another study shows how fragile estimates can be

Not every study agrees. An earlier paper reported that strict ID laws reduced turnout among minorities. Grimmer and colleagues replicated it with validated turnout data and showed that the result was a product of data inaccuracies, and that when errors were corrected one could recover positive, negative or null estimates, which precluded firm conclusions (Journal of Politics).

Measuring turnout is hard

Turnout itself is tricky to measure. Surveys overstate voting, as our article on turnout denominators explains, and administrative records can be incomplete. A small measurement error can swamp the very effect researchers hope to detect. Estimating turnout within one point alone takes roughly 9,400 respondents, as the sample size calculator shows.

Turnout is not the only claim

Supporters of ID laws point to fraud prevention and public confidence, which are different questions requiring different data. In the same study, strict ID requirements had no measurable effect on fraud, actual or perceived, but fraud is rare and hard to measure, so that finding carries its own uncertainty. Whether a law is worth its costs also involves values, not only statistics.

A careful reading: the best evidence on turnout points to small effects if any, estimates vary with data and design, and reasonable people can still disagree about the policy. When you see a claim, ask whether it concerns turnout, fraud or confidence, how the data were measured, and whether the interval is shown.