This research brief is republished with permission from CATO Institute.
February 19, 2020
By Mark Hoekstra and Vijetha Koppa
From 2000 to 2018, 20 states enacted voter identification requirements, bringing the total to 34 states. The most pronounced shift has been toward strict voter identification. While no states had such laws in 2000, 10 states had enacted (and sustained) such laws as of 2018. Under strict requirements, a vote is counted only if the voter produces a photo ID (in seven states) or nonphoto ID (in three states) within a specified period. These laws are controversial and have come under immense public and legal scrutiny. Proponents argue that these laws are necessary for protecting the election process from fraud and note that identification is required for other normal life activities. Critics argue that voter impersonation fraud is rare and that the laws are designed to disenfranchise low‐income and minority voters. They also note that as many as 11 percent of American adults lack a valid photo ID required to vote and argue that there are significant costs and impediments associated with acquiring a valid ID. The purpose of this paper is to evaluate the potential effects of these strict voter identification laws on voter turnout and election outcomes
Assessing the causal impact of these laws is difficult for multiple reasons. The infrequency of elections, the recent enactment of these laws, and the fact that the vast majority of prospective voters have IDs make it difficult to evaluate the effects. In addition, ongoing legal challenges muddy the waters. For example, Texas’s strict voter law was enacted in May of 2011, struck down by a federal court in August of 2012, reinstated in June of 2013, struck down again in October of 2014, reinstated five days later, and ultimately struck down by the U.S. Court of Appeals for the Fifth Circuit in July of 2016. In 2017, Texas passed a nonstrict version of the law in which a voter without an ID could cast a provisional ballot and have it counted by signing an affidavit, bringing an end to the seven‐year‐long litigation. While Texas is perhaps an extreme example, the general legal ambiguity of these laws raises questions as to whether voters correctly perceived whether the law was in effect and what the law required.
To overcome these challenges, we take a novel approach toward addressing these laws’ impact on turnout and election outcomes. Rather than attempt to identify effects using a typical policy evaluation methodology, we carefully document how many people vote without IDs in states that do not (yet) have strict voter identification laws. Specifically, we use administrative voting records from Michigan and Florida to identify the size of the voting population that potentially could be either fraudulent (if you believe the laws’ proponents) or disenfranchised (if you believe the critics). Both Michigan and Florida have nonstrict voter identification laws, which means they ask for IDs from voters but have provisions through which votes may be cast and counted without voters actually producing IDs. Michigan counts votes cast by voters without IDs after they sign legal affidavits regarding their identities at the polls. Florida allows voters without IDs to cast provisional ballots and then counts the ballots if their signatures at the polls match those on their voter registration forms. Importantly for this study, because of these provisions, both states track the number of ballots cast by individuals who did not have IDs.
Using these data, we are able to record the number of votes cast by individuals without IDs. In doing so, we identify the maximum number of votes cast that might not have been cast and counted if a strict voter ID law were passed in these states. This enables us to show, under a variety of conservative assumptions, the maximum extent to which a strict voter ID law would reduce turnout or affect election outcomes. The strength of this approach is that it enables us to estimate the impact of a strict voter identification requirement without relying on assumptions regarding the counterfactual. In contrast, we identify clear upper bounds of the effect of the law on both turnout and election outcomes. We view this as the central contribution of our study.
The limitations of this approach are twofold. While we can clearly identify effects for potential laws if they are passed in the (large) states of Michigan and Florida, it is an open question as to whether the results would extend to other states. In particular, our results are most relevant for the more than 20 states that already have some voter identification laws in place; we would expect our results to be less relevant for the minority of states that have no voter identification requirements. In addition, while we expect our approach to overestimate effects, given that some people without an ID would acquire one if a law were passed, we do not account for effects on voters who already have the necessary ID. For example, we will not capture effects on those who falsely perceive that the law affects their ability to vote or those who are more likely to vote because they perceive an improvement in election integrity.
Results indicate that there is little scope for strict voter identification laws to affect voter turnout. This finding stems directly from the extremely small number of votes cast by individuals without IDs, even in settings where such votes are explicitly allowed and counted. Specifically, we show that a voter identification law would reduce turnout by no more than 0.06 percent in Florida and 0.2 percent in Michigan. This suggests that at least in these two states, very few voters without IDs choose to vote even when they can.
Unsurprisingly, the small effects on turnout imply that there are very few elections in our sample that could have been affected by a strict voter ID law. Even under the most extreme assumption — that all votes for the winner (and none for the runner‐up) cast without an ID would be excluded under strict law — we estimate that a strict law could have changed the outcome in fewer than 0.35 percent of local elections and 0.09 percent of state and national elections in Florida. Similarly, we show that fewer than 0.55 percent of state and national elections in Michigan could have been affected. Estimates under more reasonable assumptions result in even smaller (and likely more accurate) potential electoral impacts. In short, the evidence presented here indicates that even if the worst fears of critics or proponents were true — that all those who would have voted without IDs are fraudulent or that all would be disenfranchised — it would have at most a tiny effect on election turnout and outcomes.
To our knowledge, this is the first paper to use administrative data to carefully document the number of voters who voted without IDs and the number of elections that could potentially be affected by strict voter identification laws. In doing so, it complements two other strands of literature on voter identification laws. The first has focused on estimating the number of people in the general population who lack the identification necessary to satisfy strict voter identification laws. Estimates are generally nontrivial, giving rise to concern about these laws’ potential effects. A national survey reports that nearly 7 percent of U.S. citizens did not have ready access to documents providing proof of citizenship and that as many as 11 percent of citizens lacked government‐issued photo identification. Studies have also documented that the lack of identification is concentrated among those who are low‐income, female (often due to name change after marriage), elderly, African American, or Hispanic. Similarly, the American National Elections Studies indicates that 7 percent of citizens lack a government‐issued photo ID. The estimates in this paper do not necessarily imply that those estimates are overstated. Rather, it is possible that the vast majority of individuals without identification do not vote even in the absence of a strict ID requirement. This could be because they have little interest in voting or because they mistakenly believe that their vote will not be counted if they do not have an ID despite efforts by the states to make it clear that the votes count. Regardless, our results indicate that a change from a nonstrict voter ID law to a strict law — the margin over which the most‐serious legal challenges have been raised — is unlikely to have a meaningful effect on voter turnout or election outcomes.
In addition, our paper also contributes to the literature that uses policy evaluation methodologies to identify the effects of these laws. Results from those studies are mixed, with some finding increased turnout and others finding significant declines. The advantage of our approach relative to these studies is that we can assess the prospective effects of the laws without making assumptions about the counterfactual.
Our results suggest that the practical importance of strict voter identification laws is likely overstated. Specifically, our findings indicate that unless voters without identification in other states vote at much higher rates than their counterparts in Michigan and Florida or unless the laws affect the voting of citizens who have IDs, the passage of these laws is unlikely to affect voting behavior and election outcomes.
This research brief is based on Mark Hoekstra and Vijetha Koppa, “Strict Voter Identification Laws, Turnout, and Election Outcomes,” NBER Working Paper no. 26206, August 2019, http:// www.nber.org/papers/w26206.