A critical question for the evaluation of election reform proposals is to assess their potential effect on outcome variables, and in the case of voter registration reform efforts (a subject of much recent debate, especially in the wake of the recent and controversial Carter-Baker report) one of the most important outcome variables is voter turnout. Thus, the first question regarding voter registration reform is how much might reforms increase voter turnout?

Most of the available research on this subject. studying the potential impact of voter registration reforms across states, asserts that making voter registration easier can increase voter turnout by as much as 7 to 10 percent. The previous research, whether based on studies of actual election returns or based on survey studies, generally tends to estimate increases in turnout of about this magnitude from significant efforts to reduce the barriers of voter registration.

In a recent study (published electronically in Political Analysis), MIT colleagues Stephen Ansolabehere and David M. Konisky offer a very different approach to estimating the potential effects of voter registration reform. Instead of looking at the possible effects of registration reform across states, Ansolabhere and Konisky look at the actual effects experienced by the states of New York and Ohio when those states moved to impose uniform voter registration requirements statewide, which occurred in New York in 1965 and Ohio in 1977. The question then becomes estimating how much voter turnout fell in New York and Ohio counties that did not require voter registration, after the imposition of voter registration requirements, controlling for other attributes of the counties that might be expected to influence turnout.

Ansolabehere and Konisky compile a dataset of county-level data, ranging from 1956 through 2000. When they look at the county-level data in both states across this span of time, they find that in presidential election years counties in New York with voter registration had average turnout rates that were 13 percentage points lower than counties without voter registration; in Ohio, the difference was 10 percentage points between counties with and without registration (page 9). Looking at non-presidential election years, the respective differences in New York were 10 percentage points, and in Ohio 5 percentage points. These estimates are in rough agreement with the previous literature.

But when Ansolabehere and Konisky turn to estimate the effect of imposition of voter registration by looking at the change in turnout in counties that did not have voter registration requirements relative to the counties that had voter registration, and then imposing some control variables in a multivariate statistical analysis (a “fixed effects panel regression” for election geeks), they estimate a slightly lower potential effect. They find that the effect of imposing uniform voter registration requirements led to a decline of turnout of about 5 percentage points in Ohio and 3 percentage points in New York. Yes, the effects of voter registration imposition are statistically significant and are in the direction that we expect, but these estimates are, of course, somewhat lower than those that have been found in other previous studies.

This is an interesting study, with a new take on how to examine the possible effects of voter registration reforms. This study raises some questions, though, that we ought to consider. One question is how representative Ohio and New York are, and whether we ought to generalize to the nation (or even other states) from the analysis of only these two states. My take here is that some additional research as to the same sort of effect in other states, especially in other regions of the county, would be very helpful. Second, I also wonder about the timing of these changes in Ohio and New York, and whether they tell us a great deal about the potential for voter registration reform today; after all, these changes in both states occurred a long time ago, and in a vastly different context than exists today. A third question is whether the “fixed effects panel regression” approach is ideal for this type of “natural experiment”; this “natural experiment” that Ansolabehere and Konisky have found might be a excellent place to explore the “propensity score matching” estimation procedures that have been the subject of much discussion recently in social science and statistics (two of our graduate students at Caltech have been working on applying these procedures to similar problems, see Delia Grigg’s poster presentation on voting technology and residual vote rates and Betsy Sinclair’s award-winning poster presentation on the ballot order effect; more on both of these innovative new research projects later.)