Past research on voter ID laws’ effects
For some time, it has been difficult to identify a clear effect of voter identification laws on turnout. In a comprehensive review published last year, political scientist Benjamin Highton wrote that the best-designed studies have found “modest, if any, turnout effects of voter identification laws.” He wrote that this could reflect the actual small effects of these laws, or perhaps that the strictest forms of voter identification are relatively recent and thus their full effects have not yet emerged.
Hence the interest in a newer article, “Voter Identification Laws and the Suppression of Minority Votes,” written by political scientists Zoltan Hajnal, Nazita Lajevardi and Lindsay Nielson. Their main finding: “Strict identification laws have a differentially negative impact on the turnout of racial and ethnic minorities in primaries and general elections.”
Hajnal and his co-authors relied on a large national survey called the Cooperative Congressional Election Study that surveyed about 50,000 people in federal elections between 2006 and 2014. Each respondent in the survey — whose personal information is known to the survey firm, YouGov — was checked against actual state voter files to try to mitigate the fact that survey respondents often report voting when they did not. The personal identifying information was not provided to researchers.
Based on their analysis, Hajnal et al. estimate that a strict voter identification law could be expected to depress black and Latino turnout by about nine percentage points and Asian American turnout by 12.5 points. Those are large effects.
Questions about the latest research
Recently, though, this finding has been called into question by political scientists Justin Grimmer, Eitan Hersh, Marc Meredith, Jonathan Mummolo and Clayton Nall. They make two key points — one about data and one about methodology.
First, it is actually not always easy to match respondents to voter files. The criteria for doing this match have changed over time and vary across states. The percentage of respondents who could not be matched to voter files increased from 10 percent in 2010 to 30 percent in 2014. Among Hispanics, the increase was even larger: from 15 percent to 42 percent.
As a consequence, in some states in some years, Grimmer and colleagues argue that Hajnal and his co-authors overstate the share of the voters by about 25 percentage points, while in other states they underestimate turnout by about 10 points. In a handful of cases, the challenge of matching respondents to the voter file means that estimated turnout is essentially zero. Thus, Grimmer and his co-authors argue that the CCES data matched to state voter files are just not an accurate way to estimate turnout in the states and thus how turnout may be related to voter ID laws.
A second point focuses on a typical challenge for social science methodology: using correlations in the statistical models of Hajnal, Lajevardi and Nielson to infer causation. Grimmer and his co-authors argue that there is a challenge here. They fitted the same statistical models to voter turnout in years before the voter identification laws were enacted. These models suggest that these laws affected turnout even before the laws were enacted.
This suggests that it’s inappropriate to attribute these differences in turnout to voter identification laws. After discussing other problems that they see in the Hajnal et al. paper, Grimmer and his co-authors concluded: “Using these data and this research design, we can draw no firm conclusions about the turnout effects of strict voter ID laws.”
Hajnal, John Kuk and Lajevardi responded to Grimmer and his co-authors, but their response did not seem convincing to me. Hajnal and his colleagues argue that the article by Grimmer and collaborators “seeks to convince readers that voter ID laws help minorities as much as they help whites,” but they never said that. What Grimmer and his co-authors wrote was that in their attempt to replicate a particular analysis of Hajnal et al., it could appear that voter identification laws increase turnout. But Grimmer and his co-authors did not put much stock in that claim, given all the problems of the data and analysis.
Hajnal and his co-authors also write that the re-analysis by Grimmer et al. “confirms the core finding of our research, which is that strict voter ID laws discriminate.” But the point of Grimmer et al. was that the data used by Hajnal et al. were not sufficient to estimate the effects of voter ID laws. Hajnal et al. also write, “Strict voter ID laws do have a racially disparate impact — we all agree. … This point is not in dispute.”
But this point is in dispute.
In Hajnal and his co-authors’ piece for the Monkey Cage, the headline read: “Do voter identification laws suppress minority voting? Yes.” This claim was premature. Voter turnout varies from election to election, and survey-based estimates of voter turnout among different ethnic groups are noisy. As a result, we cannot make such strong and confident claims about any racially disparate effects of these laws.