Widening the Lens on Voter Suppression

The Challenges of Knowing Voter Suppression: An Appraisal of Recent Studies

The Challenges of Knowing Voter Suppression: An Appraisal of Recent Studies

SINCE THE 2016 presidential election, public interest in voting misconduct has surged, just as researchers receive their largest batch of voting data since most new voting restrictions took effect. The result is that several new studies on these laws’ impacts have made headlines and captured considerable public attention.

In this section, I review the three most significant of these, examining both the studies’ findings and the methodological debates they inspired. The terms of these debates circulate far less widely than the headlines, but I argue that they have more to teach us. In particular, the methodological challenge of how to distinguish restrictive voting laws’ effects on turnout from the effects of other factors influencing voting behavior demands that we reconsider the nature of voter suppression. The reason this challenge is so ubiquitous is that restrictive voting laws are usually not singular, determinate causes of lost votes. They suppress the vote precisely by combining together with other voting deterrents. In most cases, it is only cumulatively that these multiple causes keep eligible voters from voting. This is what is most important, I argue, in the studies discussed below: They help us to recognize the co-causes of suppressed votes, and to name them as voting-rights issues and necessary targets in the fight against voter suppression.

The Civis AnalyticsPriorities USA study

The first notable study to evaluate the impact of voter suppression laws in 2016 made a major splash. The ensuing scrutiny of its methods, however, made plain the pitfalls of analyzing voter suppression from the proverbial 30,000 feet.

Conducted by Civis Analytics for the super PAC Priorities USA, the study offered a stunning headline: Wisconsin’s strict voter ID law suppressed 200,000 votes—in a state Donald The Challenges of Knowing Voter Suppression: An Appraisal of Recent Studies Trump won by less than 23,000. Wisconsin Senator Tammy Baldwin repeated the figures in a tweet. Hillary Clinton cites them in a chapter called “Why” in her election memoir.15 Yet the study’s model was swiftly and decisively rejected by most voting rights scholars and experts—and with good reason.16

The Civis Analytics-Priorities USA study compares statewide turnout figures from the 2012 and 2016 elections. It finds that, on average, states that adopted strict voter ID laws saw a drop in turnout, while those that made no changes to voting laws had increased turnout in 2016. Already this is misleading, however, in that half of the ten states with new strict ID laws saw turnout go up, while a handful—Mississippi, Wisconsin, and Ohio—accounted for most of the overall drop for the entire group. From there, Civis Analytics treats the overall turnout increase of +1.3 percent in “no-change” states as the national norm—the amount by which each state should have increased its turnout rate in 2016. The study then imputes all change falling short of that level of growth to voter ID laws.

The report says that Wisconsin’s turnout rate dropped by 3.3 percent, and that, if it “had instead increased by the national no-change average, we estimate that over 200,000 more voters would have voted in Wisconsin in 2016.”17 At the national level, “If states where voter ID laws became stricter between 2012 and 2016 had increased turnout by the same rate as that of states where there were no voter ID law changes, we estimate that over 400,000 more voters… would have cast their ballots.”18 Fair enough; these statements are about math. But it is something different to attribute all divergence from the average to voter ID laws. Doing so is particularly problematic when half of the states with new laws actually saw turnout increases.

A final contentious analytical tack in the Civis Analytics-Priorities USA study is its use of a comparison between Wisconsin and Minnesota— which has no voter ID law—to gauge the impact of Wisconsin’s law. Here the study looks at county-level turnout figures, with a particular interest in counties with larger African American populations. On one hand, the comparison is compelling, at once affirming that drop off in counties with more African Americans took place across the board, but also that it was steeper in Wisconsin than in Minnesota. But on the other, we must wonder what the outcome would be were the same analysis run for another neighboring state with no ID law: Michigan. Scholars who have analyzed voter file data based on official turnout records find that African American participation in that state declined at essentially the exact same rate as in Wisconsin.19 This by no means implies that African Americans’ votes were not suppressed in Wisconsin. But it certainly raises questions about Civis Analytics-Priorities USA’s model for assessing the impact of the voter ID laws.

Broadly speaking, the problem with this model is that it neither takes into account, nor controls for, any of the other myriad factors that influence whether voters will or will not go to the polls. It can attribute variation only to the factor it is trying to investigate—voter ID laws. What the Wisconsin-Minnesota comparison shows in particular is that the study treats voters as essentially a series of context-less, isolated units; it assumes they should behave in mostly consistent patterns irrespective of place insofar as they look the same in crude demographic terms. This is what allows for the conclusion that differences in Black voting patterns in Minneapolis-Saint Paul and Milwaukee cannot be explained otherwise than with reference to voter suppression laws.

In bringing this criticism out, we are reminded to think of the ways voters are networked in local communities that influence their behavior. They talk, share information, and influence one another’s opinions, dispositions, and practices. No matter how ubiquitous social media has become, we cannot ignore the significance of place in all of this—that these networks and communication are seated in distinct localities. This is essentially the difference between analyzing voters as members of populations versus communities.

Distinct civic dispositions form and spread in the context of communities through the dialogue of their members. These might include ideas that our votes don’t really matter, that the candidates aren’t so different, or that withholding our support is the only way to get their attention.20 They could also be that new ID laws make voting more trouble than it’s worth or that the people running things will just turn away “people like us” anyway. The collective political knowledge and sense-making that these conversations develop can be just as crucial to securing the exercise of voting rights as any piece of litigation. The critique of the Civis Analytics model’s indifference to this—or any—context surrounding voters reminds us that community networks and knowledge circuits are infrastructure that give those rights effect.

The Journal of Politics Debates

Important debates over how to quantify restrictive voting laws’ impacts also took place in the field of political science in 2017, most notably around a study published in The Journal of Politics. The professional researchers involved in this work avoided clear missteps like those in the Civis Analytics study, careful to control for outside variables as they investigated voter ID laws. Still, methodological disagreements persist among the researchers’ peers. These disagreements spotlight pending challenges not only to studying voter suppression, but also to combatting it.

The study appearing in The Journal of Politics was carried out by Zoltan Hajnal, Nazita Lajevardi, and Lindsay Nielson.21 Though these researchers’ work does not look at the 2016 election, it is groundbreaking as a peer-reviewed assessment of the impact of the post-2011 generation of strict voter ID laws.22 Hajnal et al. use data from the large-sample Cooperative Congressional Election Studies (CCES) of the five election years from 2006 to 2014, allowing them to focus on turnout effects across different voter subgroups—particularly as defined by race/ethnicity.

The Hajnal et al. study provides a strong and robust empirical case that recent voting restrictions are not only discriminatory in potential, but also in actual impact. It begins by comparing turnout rates of racial/ethnic subsets of the CCES sample for states with and without voter ID laws across multiple election cycles. In its straight comparison, the study shows that in general elections Hispanic turnout rates are 7.1 percentage points lower and Asian American rates are 5.4 points lower in states with strict voter ID laws in place. African American turnout is not much different in voter ID states during general elections, but is 4.6 points lower in primaries. 

Hajnal et al. then use regression models to control for a range of possible confounding (“third”) variables that might be influencing the apparent relationship between turnout in communities of color and voter ID laws. But here too they conclude that these laws have an effect on Hispanics more than any other racial or ethnic group. An average Hispanic voter’s likelihood of casting a ballot drops by 10 percentage points if the individual lives in a state with a strict voter ID law, “all else equal.”23 Hajnal et al.’s regression analyses also find that such laws decrease African American and Asian American voters’ likelihood of voting in primary elections. Because they have no noticeable effect on whites, their overall consequence is to significantly increase the white-nonwhite turnout gap.

Despite the numerous factors for which it controls,24 Hajnal et al.’s study quickly drew criticism for having not considered all possible confounding variables that could influence turnout. In a response piece to be published in The Journal of Politics, political scientists Justin Grimmer, Eitan Hersh, Marc Meredith, Jonathan Mummolo, and Clayton Nall make the case that there are likely “unobserved baseline differences between states with and without [voter ID] laws.”25 That is, there are other factors that are both causal contributors to depressed turnout and correlated with the adoption of voter ID laws. Grimmer et al. ground this claim in the results of a “placebo test,” through which they find voter turnout effects in the states with ID laws even before those laws were in place.26

This rebuttal to Hajnal et al.’s claims reinforces a point that should be a lesson for opponents of voter suppression: Many or most suppressed votes cannot have their causes neatly separated, but instead lie at the intersections of multiple factors. Notably, Grimmer et al. cannot say what the additional factor (or factors) whose presence they identify might be. If the original study’s researchers already account for known unknowns, the critique points to unknown unknowns—all those other other variables. Surely the stories of votes not getting cast are diverse; but they must also have patterns. More research and organizational capacity on the ground would enable us to better understand and contend with this multi-causality of vote suppression.

Grimmer et al. have other criticisms of the Hajnal et al. study, as well. One is that the CCES data that Hajnal et al. use is not in fact designed to be representative at the level of state subsamples. Another is nonresponse bias: “The kind of person who lacks an ID is unlikely to be accurately represented in the opt-in online CCES study.”27 Neither of these implies that voter ID laws are not discriminatory, nor that Hajnal et al. necessarily exaggerate their impact—only that the scholars do not have the data to make their precise claims.

But for those committed to advancing voting rights, these criticisms too contain a relevant point. Voter suppression laws pick on voter vulnerabilities that manifest across spectrums of economic, social, and civic life. If these exclusions, or forms of disconnection, make a voter both unlikely to have a voter ID and unlikely to participate in an opt-in survey, they no doubt also make her less likely to be networked with organizations and institutions that defend and facilitate the exercise of voting rights. That is, some of the same factors that foster the suppression of votes also make it harder to detect. Here too, increased capacity for voter-outreach initiatives operating within communities targeted by voter suppression laws would likely weaken their impact, and surely make its mechanisms more knowable—and contestable.28

The Milwaukee and Dane County, WI Survey of Non-Voters

Grimmer et al. end their critique of the Hajnal et al. study saying that “custom-sampling surveys” may be a way for future researchers to better gauge the role of ID laws in voter turnout.29 The final study I review here meets that call. Led by University of Wisconsin political scientists working with the Dane County (WI) Clerk, it too attracted considerable attention for its implication that Wisconsin’s voter ID law likely swung the state to Trump.30 But it is the study’s recognition of the multiple causation or multiple responsibility for suppressed votes, and of the prevalence of voter misinformation, that I find most valuable.

A press release dated September 25, 2017, announced, “A Survey of registered voters in Dane and Milwaukee Counties who did not vote in the 2016 presidential election found that 11.2% of eligible nonvoting registrants were deterred by the Wisconsin’s [sic] voter ID law.” It goes on to say that this “corresponds to 16,801 people in the two counties,” but that an estimate “as high as 23,252” votes would be within the confidence interval.31 This latter figure grabbed the attention of journalists and activists in large part due to how close it was to the margin by which Trump carried Wisconsin (22,748)—and it only covers two of the state’s counties.

These findings are the result of a survey mailed to 2,400 registered Milwaukee and Dane County voters who did not cast ballots in 2016. The questionnaire focused on why respondents did not vote, but also asked about their knowledge of Wisconsin’s voter ID law. An unfortunate limitation of the survey was the low response rate. Only 293 (around 12 percent) of the questionnaires were returned—a very small sample. The study’s finding that the “burdens of voter ID fell disproportionately on low-income and minority populations”—though consistent with all research on such laws—was also questioned due to the even smaller size of these subsamples.32

The strength of a survey-based study of course is that it allows the researcher to ask the voter herself why she did not vote. Surveys that pose this question usually allow respondents to mark more than one answer, and many respondents do.33 The Milwaukee-Dane study based its survey question on one that has long been used in Census Bureau voter studies, asking respondents to mark “yes” or “no” for each of a series of 12 possible reasons they did not vote. This question is followed by a separate one eliciting “the primary or main reason” for not voting (see Figure 2). 

Figure 2 is a reproductions of one question from the Milwaukee and Dane County, WI survey of non-voters.

As in other similar surveys, many respondents in the Milwaukee-Dane study reported multiple reasons for not voting. The researchers code as “prevented from voting” all of those who listed one of the voter ID responses as their primary reason for not voting, or indicated in a separate question that they lack qualifying ID. Those who listed voter ID as among the reasons, but not the primary one, are included in a total of non-voters “deterred” from voting by the ID law.34

The Milwaukee-Dane study finds that almost as many non-voters listed the voter ID as a contributing (non-primary) cause as were identified as having been “prevented” by it from voting. Of course, we must wonder on what basis voters make these distinctions, and arrive at their rank orderings of causes. As political scientist Ryan D. Enos chimed in via Twitter, a person is often unable to name the precise reason she takes (or does not take) an action; requesting an explanation after the fact tends to generate post-hoc rationalizations.35 But this is only one side of the matter, and it is not safe to assume—as Enos’s comment might imply—that non-voters over-report the role of restrictive voting laws. Voters presumably have different ways of thinking about how to weigh the relative import of suppressive factors, and which is “primary.”36 And what about those voters who knew they could not miss work or find childcare that Tuesday, and thus never even came to know the law’s additional hurdles, or that their out-of-state ID or proof of residence would be rejected at the polls?

As the Wisconsin survey also shows, so too are many voters misinformed about the content of voter ID laws. This is consistent with other recent studies, including a similar survey conducted with Texans who did not vote in 2016. There the vast majority of those who named the photo ID requirement as their main reason for not voting actually possessed an acceptable ID.37 Likewise, the Milwaukee-Dane study finds that most of those who believed they did not have qualifying ID in fact did.38 In both cases, the studies’ authors acknowledge that the laws, and legislatures that passed them, are largely responsible for the confusion. But so too do they insist that the confusion—not just the law—must be urgently fought, through voter education, outreach, and empowerment.39

Even better than the debates arising from the two previous studies, the Milwaukee-Dane study illustrates the multiple structural causes that go into voter suppression. These can be so intertwined and cumulative in their impact that it is hard even for non-voters themselves to pick one “main” cause. Sometimes they can. But often their choice is based on misinformation. In other cases, choosing a primary reason is simply arbitrary. What all of this tells those who are committed to voting rights is that the restrictive laws are only one—and likely not the easiest—target that could be removed to bring about a different result. If our goal is the universal effective exercise of voting rights, we are just as arbitrary when we choose to fight voter suppression by focusing only on the laws while ignoring other exclusionary causes.

Epilogue to 2017’s Quantitative Studies

In the final months of 2017, two media events put voter suppression in the public spotlight, if only for a while. The first was the release of a cover article in Mother Jones titled “Rigged.” In it, voting rights expert Ari Berman lays out the full case for saying that voter suppression handed Donald Trump the state of Wisconsin in the 2016 election.40 The second was coverage of the special election for a US Senate seat in Alabama, a state with notorious and endemic voting-rights challenges. 

Berman’s article draws on both the Civis Analytics and the Milwaukee-Dane studies, but is most notable for its long-term, on-the-ground investigation in Wisconsin. It profiles several disenfranchised voters and chronicles their exasperating efforts to obtain voting credentials and cast ballots in 2016. The Washington Post’s Philip Bump was quick to criticize the piece, among other things calling Berman’s reportage “anecdotal examples of people prevented from voting.”41 But this characterization is patently unfair; any qualitative researcher can recognize that the cases Berman highlights are only a selection from a larger corpus of findings.42 This is important to note because qualitative work like Berman’s is—for reasons discussed above— uniquely able to access many of the patterns and processes by which votes get suppressed. It is the knowledge of these—and not the number of suppressed votes—that must guide voting-rights work.

Still, there is room for concern over the selection of cases Berman chose to present in his important piece. The problem is that they essentially illustrate only one type of pattern by which votes are suppressed. It is one in which voters are absolutely determined to vote, exhaust every effort, and still end up being thwarted. 

This is one story, and it is a powerful one. But its depiction of voter suppression is much too narrow. Specifically, it is too narrow in how it represents the disenfranchised voter. As a result, it misses the opportunity to make a more robust voting-rights argument. Such an argument should include voter suppression’s multiple causation, and in doing so, must include more than only the most sympathetic cases of prospective voters. Most of those who are systematically and structurally deterred from exercising the franchise are not the grandmother, the veteran, the survivor of a fire, all extraordinarily committed to voting. For many, no doubt, new voting restrictions are instead a final straw in a larger structural context that already minimizes incentives and reason for faith in the electoral process. If those voters are not included in the analysis, we get an abridged version of what voter suppression is, and a correspondingly abridged understanding of its solutions.43

The case of the December 2017 special election in Alabama drives my point here home. There, major news outlets made room for some belated discussion of the state’s exclusionary voting laws,44 but it was dwarfed by the sex-crime allegations surrounding GOP candidate Roy Moore. When Moore’s opponent, Doug Jones, ultimately emerged victorious, the storyline all but disappeared.45 A decisive factor in the race was that African Americans in Alabama not only turned out at a higher rate than whites; they also voted at an unprecedented level for a non-presidential-year election.46 So what does this say about voter suppression in Alabama—a state that bans early voting and same-day registration, limits absentee voting, and instituted a voter ID requirement at the same time it closed dozens of offices that issue them, disproportionately in counties with large African-American populations?47

We can find the answers we need in what little post-election media attention the issue received. LaTosha Brown of Black Voters Matter Fund was a leader in the effort to mobilize Black voters in Alabama—around, and in spite of, restrictive laws. Surely the restrictions in many cases worked in tandem with voters already being ambivalent about whether the election mattered; but homegrown power-building groups like Brown’s tackle both. Of the restrictions, Brown told the New York Times, “I do think that very committed, focused people will find a way [to vote]. But is that fair?”48 Assuming it is not, then neither would it be fair if only those who work unimaginably hard to vote were treated as deserving of voting-rights energy and resources. The equal and effective exercise of voting rights means creating whatever structures, capacity, and infrastructure are necessary to facilitate the participation of everyone.

  • 15. Hillary Rodham Clinton, What Happened, Simon & Schuster, 2017. See p. 420.
  • 17. See for example, the reactions compiled in Rick Hasen, “Some Words of Caution About New Study Finding Voter ID Cost Clinton WI and Election,” Election Law Blog, May 9, 2017.
  • 18. Civis Analytics and Priorities USA, “Voter Suppression Analysis,” Press Release, May 3, 2017.
  • 19. Bernard L. Fraga, Sean McElwee, Jesse Rhodes, and Brian Schaffner, “Why did Trump win? More whites – and fewer blacks – actually voted,” Monkey Cage/The Washington Post, May 8, 2017. Fraga et al. find that the voter participation rate for African Americans dropped by 12.4 percentage points in Michigan and 12.3 points in Wisconsin. These figures differ from those of the US Census Bureau’s Voting and Registration supplements, which show a smaller drop in African American turnout in Michigan from 2012 to 2016. However, the Census Bureau’s estimates have a number of known methodological problems, and in any case, official ballot counts for Wayne County (Detroit) and Flint, MI support Fraga et al.’s findings.
  • 20. See the findings from focus groups with drop-off voters in Milwaukee: Civic Engagement Fund and Brilliant Corners Research & Strategies, Breaking Away: Exploring the Third Party Millennial “Protest” Vote of 2016, June 2017.
  • 21. Zoltan Hajnal, Nazita Lajevardi, and Lindsay Nielson, “Voter Identification Laws and the Suppression of Minority Votes,” The Journal of Politics 79(2): 363-379 (2017).
  • 22. The authors rightly note that any inconclusiveness in peer-reviewed research on voter ID laws comes from studies that pre-date this wave of strict voting restrictions. See ibid.
  • 23. Ibid., p. 368.
  • 24. Hajnal et al. control for age, education level, family income, nativity, gender, marital status, having children, being a union member, owning a home, being unemployed, and religion, as well as other state-level electoral laws that may encourage or discourage voter participation. Ibid., p. 367.
  • 25. Justin Grimmer, Eitan Hersh, Marc Meredith, Jonathan Mummolo, and Clayton Nall, “Comment on ‘Voter Identification Laws and the Suppression of Minority Votes’,” forthcoming in The Journal of Politics, p. 2. The version of this article referenced herein is dated August 16, 2017.
  • 26. Ibid., pp. 7-8.
  • 27. Ibid., p. 3.
  • 28. Hajnal et al. themselves note this knowledge gap, saying, “It could be that more minorities do not have the requisite ID, that the costs of obtaining an ID are too high for minorities to bear, that passing these laws sends a signal to minorities that they are not wanted at the ballot box, or some combination of the above. We simply do not know.” Hajnal et al., “Voter Identification Laws and the Suppression of Minority Votes,” p. 377.
  • 29. Grimmer et al., “Comment,” p. 12.
  • 30. Kenneth R. Mayer and Michael G. DeCrescenzo, “Supporting Information: Estimating the Effect of Voter ID on Nonvoters in Wisconsin in the 2016 Presidential Election,” September 25, 2017.
  • 31. Kenneth R. Mayer and Scott McDonell, “Voter ID Study Shows Turnout Effects in 2016 Wisconsin Presidential Election,” Press Release, September 25, 2017.
  • 32. The researchers themselves state that their “estimates cannot be extrapolated to the state of Wisconsin as a whole,” making the small sample size somewhat less problematic. Mayer and DeCrescenzo, “Supporting Information,” p. 2.
  • 33. For example, the post-2016 election wave of Pew Research Center’s American Trends panel survey (n=4,183) identified 472 non-voters. Of these, 1/3 selected more than one reason for not voting; 1/6 chose at least three reasons. It is a tiny sample, but 14 out of the 20 who mentioned problems with voter ID, mail-in ballots, or their registration also mentioned another reason for not voting—be it that “something came up,” they didn’t have time, or they thought their vote wouldn’t matter anyway
  • 34. Mayer and DeCrescenzo, “Supporting Information,” p. 5. It is the latter category for which the researchers estimate the 11.2% figure; under the stricter definition, they find that 6% of non-voters were “prevented” from casting ballots.
  • 35. There may be further psychic incentive to give a rational account of behavior that is socially disparaged, as failing to vote may be perceived to be.
  • 36. For example, when responding to a question about the “primary” reason for not voting, some may consider material cost paramount, others might think of the final, deciding factor—“the straw that broke the camel’s back.”
  • 37. Mark Jones, Renée Cross, and Jim Granato, “The Texas Voter ID Law and the 2016 Election: A Study of Harris County and Congressional District 23,” University of Houston Hobby School of Public Affairs, April 2017.
  • 38. Mayer and McDonell, “Voter ID Study Shows Turnout Effects in 2016 Wisconsin Presidential Election.
  • 39. See further Renée Cross, Jim Granato, and Mark Jones, “State should focus on voter education,” Houston Chronicle, April 7, 2017
  • 40. Berman, “Rigged.”
  • 41. Philip Bump, “The case that voter ID laws won Wisconsin for Trump is weaker than it looks,” The Washington Post, October 20, 2017
  • 42. Berman has also published other findings elsewhere, e.g. Ari Berman, “How the GOP Rigs Elections,” Rolling Stone, January 24, 2018. Furthermore, he is not the only journalist to profile disenfranchised Wisconsin voters, who, according to another account, were “not hard to find.” Christina A. Cassidy and Ivan Moreno, “In Wisconsin, ID law proved insurmountable for many voters,” The Associated Press, May 14, 2017.
  • 43. I of course in no way mean to suggest that this is Berman’s intent.
  • 44. For examples, Vann R. Newkirk II, “What’s Missing From Reports on Alabama’s Black Turnout,” The Atlantic, December 7, 2017; Scott Douglas, “The Alabama Senate Race May Have Already Been Decided,” New York Times, December 11, 2017; and David Leonhardt, “Voter Fraud in Alabama,” New York Times, December 12, 2017.
  • 45. This despite that numerous complaints of voter suppression were reported on election day. Pema Levy, “Reports of Voter Suppression Tactics Pour In From Alabama Election,” Mother Jones, December 12, 2017.
  • 46. Official individual voting records showed Black turnout in December 2017 at 44.6 percent, and 48.2 percent among women. For point of reference, Alabama statewide turnout rates for midterm elections in 2014, 2010, and 2006 were 33.2 percent, 43.4 percent, and 37.5 percent, respectively.
  • 47. Kyle Whitmore, “As it turns out… Bentley’s driver’s license closures were racial, after all,” AL.com (Alabama Media Group), January 5, 2017.
  • 48. Alan Blinder and Michael Wines, “Black Turnout in Alabama Complicates Debate on Voting Laws,” New York Times, December 24, 2017. A stellar post-Alabama piece arguing that, “Relying on lawsuits to tackle the problem of voter suppression is a losing strategy,” is: Mikki Kendall, “Want to thank black voters for defeating Roy Moore? Tackle voter suppression,” The Washington Post, December 15, 2017.