Whom Does Voter ID Bar from Voting? Evidence from Texas*

Michael G. Miller Assistant Professor Barnard College, Columbia University Department of Political Science *Prepared for the Election Science, Reform, and Administration Summer Conference, Portland, OR, July 27, 2017 Working paper. Please do not cite or utilize as evidence without author permission.

Acknowledgments and Disclosures: This work is funded solely by Barnard College. The author received no assistance in terms of either intellectual property or financial support from any other organization. Samuel Ackerman and Angela Beam provided crucial research assistance. All mistakes and oversights rest with the author. Abstract In 2013, Texas implemented one of the strictest photo identification requirements for voters in the United States. I exploit a federal court decision mandating that Texas create individual records for each voter who arrived at its polling places without proper identification in 2016, as well as Texas election laws that maintain election statistics separately for Spanish-surnamed voters. I find that more than 16,000 Texans turned out to vote in 2016 without appropriate identification, and would have been disenfranchised under the Texas law. County-level analysis indicates that Democratic vote share, but not the percentage of black, Hispanic, or below-poverty residents, is predictive of the percentage of voters in a given county who arrived without ID, nor does the number of Hispanic voters lacking ID exceed expectations given their participation in the election. Furthermore, Hispanic voters were no more likely to cite either economic hardships or relocation as the reason they lacked ID. That said, an analysis of turnout yields a negative effect of about ten percentage points among Spanish-surnamed voters in 2016. These results suggest that rather than arriving at the poll without ID, many Hispanic voters may have simply stayed home.

1

1

Introduction

The states have long exercised a great deal of control over their voting laws. The past decade has seen a flurry of activity with respect to election regulations, particularly since the Supreme Court’s decision in Shelby County v. Holder (570 US ___ (2013)), which negated the coverage formula that determined which jurisdictions required federal pre-clearance of election law changes under the Voting Rights Act (VRA). None of the new policies has been more controversial than so called “voter ID” laws requesting photo identification (ID) from voters. The recent trend in voter ID laws has been toward “strict photo ID” policies mandating that in-person voters cannot cast a ballot without first presenting a photo ID from a pre-defined list. The laws are controversial because not all would-be voters are likely to have an acceptable ID, and may therefore be disenfranchised. Furthermore, the groups often assumed to be least likely to possess correct identification, such as black, Hispanic, or less affluent voters, also tend to be more likely to support the Democratic Party. As such, the debate over voter ID has taken on a partisan hue and has attracted attention from policymakers and scholars seeking to determine whether the laws do suppress turnout and if so, whether minority voters are more burdened. This paper examines the effects of a strict voter ID law in Texas during the 2016 General Election. I exploit a federal court decision mandating that Texas create individual records for each voter who arrived at its polling places without proper identification, as well as Texas’ policy of maintaining election statistics separately for Spanish-surnamed voters. I find that more than 16,000 Texans (roughly one-fifth of one percent of all voters) turned out to vote in 2016 without appropriate identification. County-level analysis indicates that Democratic vote share, but not the percentage of black, Hispanic, or below-poverty residents, is predictive of the percentage of voters who arrived without ID, nor does the number of Hispanic voters lacking ID exceed expectations given their participation in the election. Furthermore, Hispanic voters were no more likely to cite either economic hardships or relocation as the reason they lacked ID. That said, an analysis of turnout yields a negative effect of about ten percentage points among Spanish-surnamed voters in 2016. These results suggest that rather than arriving at the poll without ID, many Hispanic voters may have simply stayed home. In examining individual-level data in the context of an actual election, this paper advances our understanding of how voter ID laws affect voting behavior.

2

2

Background

“Voter ID” laws are not a new idea; they date to as early as 1950, when South Carolina enacted an (non-photo) identification requirement, and at least four other states followed suit during the next thirty years (Biggers and Hanmer, 2017). In the wake of the controversial 2000 presidential election, several states began implementing voter ID laws with the goal of securing the integrity of their elections. Most of the states that enhanced identification requirements during this time began requiring non-photo identification, but Indiana and Georgia went further. For instance, Indiana’s 2005 law mandated that voters present an Indiana or U.S. photo ID, though it provided a mechanism for voters to cast ballots if they attested to a financial difficulty that impeded their acquisition of acceptable identification. While opponents argued that the law would create an undue burden on voting for some people, especially poorer, older voters, the United States Supreme Court upheld Indiana’s law in Crawford v. Marion County Election Board (553 U.S. 181 (2008)), ruling that Indiana’s interest in preventing electoral malfeasance outweighed those burdens. The Crawford decision paved the way for other states to implement strict voter ID laws, and Texas was among them. Texas had passed its first non-photo ID legislation in 1966 (Biggers and Hanmer, 2017), but until the early 2010s, the requirements to cast a ballot there were typical of many states. In 2011 however, Texas enacted Senate Bill 14 (Election Code §63.001 et seq.), mandating that voters present photo identification before being allowed to cast a ballot. The law was blocked under Section 5 of the VRA, but implemented shortly after the U.S. Supreme Court struck down the VRA’s coverage formula in Shelby County, which also effectively nullified Section 5. At the time of its passage, SB 14 made Texas’ voter identification requirements among the strictest in the nation, designating only three types each of acceptable federal or state IDs.1 Indeed, by some estimates, as many as 600,000 Texans lacked adequate identification under the law (Malewitz, 2016), though voters could obtain an “election certificate” if they possessed the correct documents to do so. One day after the law’s enactment in June of 2013, it was challenged in federal court as discriminatory under Section 2 of the VRA. The first ruling in the case, commonly known as Veasey 1

Acceptable identification includes: U.S. military ID, U.S. passport, U.S. citizenship certificate, Texas election identification certificate, Texas ID or driver’s license, Texas license to carry a concealed handgun. Under the law, the voter’s identification must include a photo, and cannot be expired for more than sixty days.

3

v. Abbott, came in October of 2014. The U.S. District Court ruling held that the law was racially discriminatory. The 5th Circuit Court of Appeals agreed, and remanded the case to the District Court, whose job it was to find an acceptable interim solution. In August of 2016, the District Court ordered that all voters who possessed an ID required by SB 14 must produce it for pollworkers in order to vote in the November election. But the court also ordered that voters who lacked identification that would satisfy SB 14’s requirements should be allowed to vote if they completed and signed a “reasonable impediment declaration” (RID) attesting that they did not possess a valid photo ID, as well as the reason why they could not obtain one in time for the election. The form contained eight check-boxes (described below) allowing voters to report their impediment(s).2 Texas was also ordered to provide voter and pollworker education, and to provide the RID form in multiple languages. Thus, while SB 14 was law for the 2016 Election, the federal courts had weakened it substantially by allowing voters lacking adequate identification to vote–so long as they first swore to the reason why they could not obtain identification. Importantly, the court’s order also has implications for the study of voter ID laws’ effects on voting behavior.

3

Previous Research

The policy debate over voter ID has taken on a partisan tinge in the past decade. A growing number of studies have found that voter ID laws are likely to be enacted by Republican state governments (Biggers and Hanmer, 2017; Hicks et al., 2015; Hicks, McKee and Smith, 2016; McKee, 2015; Rocha and Matsubayashi, 2014), and conservative voters are generally more supportive of voter ID laws (Wilson and Brewer, 2013). In justifying new voter ID regulations, Texas and other states often cite the need to prevent voter impersonation fraud, possibly as a response to a conservative electorate: A recent Gallup poll suggests that Republicans are roughly twice as likely as Democrats to feel that election fraud is a “major problem” McCarthy (2016). Indeed, in his response to the 5th Circuit’s Ruling that the law was discriminatory, Texas Governor Greg Abbott said, “Voter fraud is real, and it undermines the integrity of the election process.”3 Thus, one possible explanation for 2

Before obtaining an RID form, voters had to produce some other proof of identity from a much wider list of sources. These include “a valid voter registration certificate, a certified birth certificate, a current utility bill, a bank statement, a government check, a paycheck, or any other government document that displays the voter’s name.” 3 State of Texas. https://gov.texas.gov/news/post/governor_abbott_statement_on_texas_voter_id_law1

4

the partisan nature of the debate is that Republicans are focused more on limiting fraud, which pits their willingness to increase identification requirements against Democratic concerns about disenfranchising (presumably Democratic) voters (Minnite, 2010; Hicks et al., 2015). Alternatively, it might be the case that both Republicans and Democrats believe that voter ID laws suppress the Democratic vote, and the fight over voter ID is but one front in a broader conflict as the parties jockey for advantage via election laws (Hasen, 2012). Both parties might therefore believe that Democratic-leaning constituencies such as African-Americans, Hispanics, or less affluent voters are simply less likely to have the kind of identification that a voter ID law might require (Hood and Bullock, 2008). Or perhaps the expectation is more generalized; voter ID laws might be expected to reduce turnout overall, and Democrats perform worse when turnout is low (Hansford and Gomez, 2010). This logic rests on classic models of voting behavior (e.g., Downs 1957; Riker and Ordeshook 1968) which posit that an individual’s turnout propensity results from a consideration of the costs and benefits of voting. It is unclear whether voter ID laws impart a benefit in this calculus; indeed, such laws do not appear to foster increased confidence in elections among voters (Ansolabehere, 2009; Bowler et al., 2015). That said, most strict photo ID laws do impart tangible costs on many voters, which were recognized by both sides in the Crawford argument and in most subsequent litigation surrounding voter ID laws. The logic here is straightforward. Under strict laws, voters who lack sufficient identification must obtain it or risk being disenfranchised. For many, this might mean incurring economic costs in the form of state fees for the ID itself or the necessary documents (such as a birth certificate) to obtain it, as well as less direct costs such as the time and expense of traveling to a state agency. Much previous work has therefore focused on whether voter ID laws depress turnout. Despite the theoretical expectation for reduced turnout, results have been mixed at best; most work has found little or no link between the implementation of voter ID and overall turnout (e.g., Ansolabehere 2009; Hood and Bullock 2012; Mycoff, Wagner and Wilson 2009). There are at least three possible reasons for the null relationship. One could be that the act of passing such a law compels a “backlash effect” as Democrats mobilize in response to what they perceive as an unjust law (Valentino and Neuner, 2017). Second, it is possible that states are able to educate voters before the first election that requires them to produce identification, and voters respond by obtaining the necessary ID. 5

Merely reminding voters about new voter ID requirements can increase turnout (Citrin, Green and Levy, 2014), and voter ID laws often include an appropriation to inform the public about the change. Indeed, Texas spent $2.5 million on voter education prior to the 2016 election. Third, voters who lack IDs might be mostly disengaged from politics; voter ID laws may in other words deprive them of a right they never intended to claim. For instance, despite Texas’ efforts at voter education, Jones, Cross and Granato (2017) found that only 20% of non-voters could correctly list allowable pieces of identification, which suggests that they may simply not engage with civic information. Finally, the people who do not have identification might exist in numbers too small to affect the turnout rate much; Jones, Cross and Granato (2017) found that more than 97% of Texas non-voters in 2016 possessed at least one valid piece of identification. Other recent work has argued that voter ID laws are particularly burdensome on racial minorities and/or people in lower socioeconomic strata, and inquiry should be focused there (Barreto, Nuno and Sanchez, 2009; Sobel and Smith, 2009). In a study of Indiana shortly after the Crawford decision, Barreto, Nuno and Sanchez (2009) found that racial minorities (and Democrats) were less likely to possess state-issued photo ID, and a survey of one Texas county after the 2016 election found that black and Hispanic non-voters were more than six points more likely than Anglo nonvoters to report that a lack of proper identification may have contributed to their failure to vote (Jones, Cross and Granato, 2017). The potentially disparate racial impact of voter ID laws is underscored in Hajnal, Lajevardi and Nielson (2017), who use survey data to demonstrate that strict photo ID laws are more likely to reduce turnout propensity among minority voters (and Democrats, more generally). That said, there is some doubt as to the validity of these findings (Grimmer et al., 2017), and Rocha and Matsubayashi (2014) found no evidence that strict photo ID laws disproportionately depress either black or Hispanic turnout. Still, race appears to be an important factor shaping both the passage and execution of voter ID laws. Among whites, increasing levels of racial resentment are correlated with stronger support for identification laws Wilson and Brewer (2013). Moreover, state governments are particularly likely to pass voter ID legislation when African-American and/or Hispanic populations expand (Biggers and Hanmer 2017; Hicks, McKee and Smith 2016; McKee 2015; but see: Rocha and Matsubayashi 2014). Once the laws are implemented, election officials are less likely to respond to requests for help from voters with Hispanic names (White, Nathan and Faller, 2015), and minority voters are 6

also more likely to be asked for identification at the polls (Atkeson et al., 2014; Cobb, Greiner and Quinn, 2012). Thus, race has clear theoretical implications for any assessment of voter identification laws.

4

Expectations

In tandem with Texas’ policies regarding its election records, the District Court’s ruling in Veasey yields an opportunity to engage questions that have to date been largely unanswerable in studies of voter ID laws, particularly with respect to Hispanic voters. There are two specific elements that are especially important. First, like all states, Texas maintains county-level records of voter registration and turnout. However, Texas is unique in that it separately tabulates these figures for voters with a “Spanish surname.”4 This facilitates a design in which turnout among Hispanic and Anglo voters can be separately examined both before and after the implementation of SB 14. Second (and more significant), the court’s requirement that voters without identification file an RID creates an individual-level record of voters who arrived at their polling place and who presumably would have been disenfranchised in the absence of a judicial order. Crucially, the RIDs are tabulated by county and include both voters’ names and the reason they cited for not having appropriate identification on Election Day. A simple tabulation of the RIDs can answer basic questions that have heretofore been elusive: How many people would SB 14 have prevented from voting, and why? Moreover, the existing research described above suggests additional questions that can be engaged with both county and individual (RID)-level data. For instance, if SB 14 had disparate effects on Democrats and/or minorities, we might expect counties with more Democratic or minority voters to display higher rates of voters filing RIDs. We might also expect the RID rate for Hispanic voters to exceed their composition of the electorate, and/or Hispanics to report higher instances of enduring hardships– such as a lack of access to necessary documents–that represent significant obstacles to obtaining appropriate identification. Finally, Texas’ county-level voting records can shed light on whether SB 14 depressed turnout among Hispanics compared to Anglo voters. I engage these questions in the sections below. 4

At this writing, it is unclear how Texas defines “Spanish surname.” State employees have not been able to provide a definitive rule, despite repeated requests.

7

5

Data

In December of 2016 I submitted formal requests for the Reasonable Impediment Declarations to all 254 Texas counties, as well as to the Texas Secretary of State, under the Texas Open Records Act. I requested from each county the number of RIDs that voters completed during the 2016 election, as well as copies of the petitions themselves. I also requested information about registered voters, ballots, and vote totals tabulated by whether the voter had a “Spanish surname,” as defined in the Texas voter file. I successfully obtained all requested information from all Texas counties.

Figure 1: Example Reasonable Impediment Declaration

Each RID contains the voter’s name, as well as a series of check boxes allowing the voter to claim one of eight reasons for not having proper identification. Listed reasons include: lacking necessary documents, having a disability, family or work obligations, lack of transportation, having a lost or stolen ID, having applied for an ID that was not yet received, or some other reason. Voters marking the “other” category were invited to write their own justification. From each RID, I therefore recorded each voter’s name, and his/her marked (or stated) reason for failure to produce 8

appropriate identification. Figure 1 depicts an actual petition filed by a Texas voter in the 2016 election; the format was the same for all counties. The RID total in a given county represents the number of people who turned out to vote without statute-compliant identification, and who would presumably have been turned away from the polling place on Election Day. However, it should be taken as a lower-bound estimate of the law’s effects. The RID total cannot capture the number of people who were deterred from voting upon learning that Texas had passed a voter ID law, nor can it account for people who lacked both photo ID required by SB 14 and the wider range of identification mandated by the district court. This said, the RIDs allow unprecedented, individual-level insight into the reasons voters cited for not having necessary identification at the time of their intent to vote. To the RID and election data, I merged county-level data from the U.S. Census 2015 ACS five-year average. I also used the 2010 Census name dictionary to designate voters filing RIDs as “Hispanic” based on Spanish-origin surnames. I coded as “Hispanic” all names with a pre-defined probability of Hispanic origin; in the analysis below I separately apply a rule of .5 and .3. Key descriptives for all Texas counties, as well as for the counties that reported receiving at least one RID and for those that submitted no RIDs, can be located in the Appendix. Unsurprisingly, the counties that logged no RIDs are considerably smaller in terms of population than those that recorded at least one, but the two groups are comparable with respect to Hispanic population, which is an important aspect of the analysis below. In the following section, I present both countylevel and mixed-level models, as well as descriptive analysis of the RIDs themselves. Given the range of questions I consider, I detail each model or method as it is presented.

6 6.1

Results RID Analysis: Descriptives

I begin with an important question: How many Texans would the voter ID law have impeded from voting on Election Day? The number of petitions filed exhibited a wide range across counties, though it was fairly low in most: The mean percentage of RIDs (out of total county voters) was one-tenth of one percent in counties that logged at least one RID. Indeed, in 70 counties, election

9

officials reported receiving no petitions at all, and they logged 2 or fewer in 102 Texas counties.5 Nonetheless, the state of Texas received 16,373 RIDs from Texas counties in the 2016 election out of more than 8.5 million votes cast, a rate of about 0.2%. Figure 2 depicts the reasons that voters cited for failure to produce identification.6 As is clear from the response distribution, the RID form may not have adequately captured the true range of reasons that voters were not able to produce identification. The “other” category wins a clear plurality of the choices, with more than 5,400 respondents choosing it. Roughly 4,360 voters reported lost or stolen IDs, followed by those citing work obligations (1,924), disabilities (1,315), inadequate transportation (1,125), a lack of documents necessary to obtain ID (1,111), having applied for but not received an ID (990), and family obligations (642).

Figure 2: Frequency of Reported Impediments, Statewide

5

It seems safe to assume that in most of the counties that reported no RIDs, all voters were compliant. However, during our collection process, a number of county election officials expressed no knowledge of the RID process, and suggested that they had simply turned away voters who lacked accepted identification under SB 14. 6 Figure 1 includes information from voters who made more than one selection.

10

Table 1: Frequencies of “Other” Categories Category Ignorance of Law Protesting Law Ineligible Voter Does Not Drive Financial Reasons Incorrect/Expired/Non-Compliant ID Other Forgot ID No Reason Given Recent Relocation/Student Total

Frequency

Percent

14 27 28 30 44 256 446 617 649 3,322 5,433

0.3% 0.5% 0.5% 0.6% 0.8% 4.7% 8.2% 11.4% 11.9% 61.1% 100%

The large percentage of voters opting for the “other” category warrants closer scrutiny. Table 1 contains a breakdown of the reasons that voters offered after selecting the “other” option. Since the “other” field contained an open-ended response, I placed petitions in categories based on the core message contained in the voter’s written statement. For instance, voters who produced an unacceptable ID were grouped together as “Incorrect/Expired/Non-Compliant.” These voters might have had an expired driver’s license, one with the incorrect name (possibly due to a recent marriage or divorce), or one that was not allowed under the law, such as a Social Security card. Similarly, voters who reported geographical difficulties were grouped together in the “Relocation/Student” category. Voters commonly wrote responses like “not from here,” “recently moved,” “wrong address,” or “student.” Particularly in counties in which universities are located, it was often difficult to distinguish between voters who had produced a student ID and those who had recently moved to the county from some other place. Thus, I created a single category to represent these voters. Other voters reported financial difficulties, said they were either “protesting” the law or were unaware of it, or they may simply have said they “forgot” their ID or “left it at home.” As is clear in Table 1, the majority of voters in the “other” category reported a recent change in their location, such being a student or a recent move, as the reason they lacked identification. Many of these voters did produce an ID that they no doubt (incorrectly) thought would have

11

allowed them to vote. Similarly, more than 600 voters admitted having an (allegedly legal) ID, but failed to produce it. The most common reason for this was that voters forgot their ID, though 27 voters wrote that they were refusing to produce identification as an act of protest against the law. Nonetheless, it is worth noting that about 650 people who filed RIDs conceivably could have produced valid identification on Election Day.7 I return to this point below.

6.2

County-Level RID Rates

I next consider the county-level determinants of RID quantity. Table 2 contains OLS regression coefficients and robust standard errors for models of the percentage of voters who filed RIDs in a given county.8 I fit models of both the overall RID percentage, as well as an “adjusted” percentage that discards RIDs filed by the approximately 650 (statewide) voters who admitted having a legally acceptable ID–but refused for some reason to show it. Thus, the “adjusted” percentage is likely a better measure of the number of people whom the law would have prevented from voting in the 2016 election. I also fit models separately for all counties, as well as a subset of counties in which at least one RID was filed, in an effort to account for disparate implementation–if any–between counties. Unsurprisingly, considering the small number of RIDs filed in many Texas counties, the effect sizes in absolute terms are not large. However, four trends are worth noting from the models. First, age, education level, or poverty rate in a county do not appear to drive the percentage of RIDs cast. Neither counties that skew toward the younger (eighteen to twenty-four years) or older (seventy-five years or older) ends of the spectrum demonstrate a greater propensity for RID filings. Nor does the percentage of high-school graduates in a county appear to be related to the rate of RIDs. The same is true of the coefficients for the percentage of households in a county below the poverty line. Second, neither the percentage of African-American nor Hispanic residents is associated with a county’s RID percentage. Indeed, contrary to expectations given previous findings (e.g., Hajnal, Lajevardi and Nielson 2017), these coefficients are negatively signed in all models–though they to do not achieve statistical significance. Third, as might be expected, turnout is positively associated 7

Moreover, tallying voters whose identification was lost or stolen, who forgot, and who produced expired or otherwise problematic IDs, another 7,888 people had previously demonstrated some capacity to obtain identification in the past–and presumably could do so again. 8 The dependent variable is a percentage ranging from 0 to 1.

12

with a county’s RID percentage in all models, and statistically significant in the models of all Texas counties. As more voters are drawn into the process, we might expect a higher percentage of less-habitual voters, who may also be less likely to produce identification.

Table 2: Determinants of Percentage of RIDs Filed out of all Votes Cast in a County, 2016

All Counties

Dem. 2-Party Vote Share, 2012 2016 Turnout Perc. African-American Perc. Hispanic Perc. Households Below Poverty Perc. High School Graduates Perc. Aged 18-24 Perc. Aged 75+ Constant

Observations R2 Root Mean Sq. Error

Counties Filing at Least 1 RID

Perc. RIDs of votes cast

Adj. Perc. RIDs of votes cast

Perc. RIDs of votes cast

Adj. Perc. RIDs of votes cast

0.005* (0.002) 0.004* (0.002) -0.004 (0.002) -0.001 (0.001) 0.006 (0.003) 0.001 (0.003) 0.005 (0.004) -0.005 (0.006) -0.003 (0.003)

0.003* (0.001) 0.002* (0.001) -0.002 (0.002) -0.001 (0.001) 0.000 (0.002) 0.000 (0.002) 0.005 (0.003) -0.003 (0.003) -0.001 (0.002)

0.006* (0.002) 0.004 (0.002) -0.007* (0.003) -0.003 (0.002) 0.002 (0.004) -0.008 (0.004) 0.005 (0.006) 0.007 (0.008) 0.004 (0.004)

0.004* (0.002) 0.002 (0.001) -0.004 (0.003) -0.003 (0.002) -0.005 (0.003) -0.008 (0.005) 0.006 (0.004) 0.005 (0.005) 0.006 (0.005)

246 0.230 0.00143

246 0.180 0.00102

181 0.278 0.00143

170 0.178 0.00109

* p<0.05. Robust standard errors in parentheses. RID percentage calculated as the number of RIDs filed divided by the number of votes cast. The “adjusted” RID percentage sets aside RIDs filed by voters who admitted to having a valid ID.

13

Finally, all four models yield one consistent, positive, and statistically significant predictor of the percentage of voters casting RIDs in Texas counties during the 2016 election: Democratic support. All models indicate that the percentage of the two-party vote that Hillary Clinton received in a given county is positively, meaningfully, and significantly correlated with the percentage of people whom SB 14 would have deterred from voting. Thus, while the models are not consistent with one common argument regarding voter ID–that it will disproportionately suppress the vote from minority groups–they do offer evidence for another scenario: For some reason, SB 14 may be disproportionately burdensome in Democratic-leaning counties.

6.3

Analysis of Spanish Surnames

The petitions can also provide some information about who, at the individual level, SB 14 may have affected most. One of the core claims that plaintiffs made in Veasey was that the law would disproportionately burden voters from minority groups, who might be less likely to possess acceptable forms of identification. While it is nearly impossible to recover African-American race from voter names, the U.S. Census Spanish surname dictionary can offer a reasonably high chance of determining whether a person identifies as Hispanic (e.g., Perkins, 1993). I therefore analyze RIDs using Hispanic status–as determined by the surname on the RID petition–as an independent variable. To classify voters as Hispanic, I utilize data from the U.S. Census, whose 2010 surname directory calculates the probability that a voter with a given surname will self-report as Hispanic. To account for the uncertainty in this method, below I analyze two independent variables designating a voter as as “Hispanic:” one for voters whose probability of being Hispanic is .5 or greater, and another for voters whose probability is greater than .3. That said, the distinction does not appear to be particularly meaningful. There were 2,937 RIDs filed by voters in “.5” condition, and 2,976 in the “.3” condition. I begin with a straightforward question: Did Hispanic voters file RIDs for any reason at rates greater than we would expect, given their county-level participation in the election? We might expect a negative answer, given the number of Hispanic voters in the RID pool. For instance, employing the “.3” rule for identifying Hispanic voters results in their comprising roughly 18.4% of all RIDs filed. This is substantially less than the Hispanic composition of Texas population, which according to the 2010 ACS 5-year average is 38.4%. 14

Figure 3: Expected vs. Actual RID Counts, County Level

To investigate this question in more depth, I determined the expected number of Spanishsurname RIDs for each county, calculated as the proportion of Hispanic voters who participated in the election, multiplied by the number of RIDs recorded in a given county.9 I then compared the expected number of Hispanic RIDs with observed counts. If Hispanic voters are disproportionately burdened, we would expect the observed RID counts to exceed the expected ones. Once again, there is little top-level evidence that this is the case. The mean expected count is about 12.7 in Texas counties, while the observed mean is 11.6 for the “.5” definition and 11.7 for the “.3” definition.10 Figure 3 depicts the actual county-level count of RIDs filed by Hispanic voters employing the two definitions, plotted against the expected count. Figure 3 also features a reference line with a slope of one, so counties in which the observed count exceeds the expected count should appear to the right of the reference line. As is evident in Figure 3, regardless of which rule is employed for 9

I employed figures from the state of Texas on the number of Spanish-surnamed voters who cast ballots in each county. To date, I have been unable to obtain the rule that Texas uses to define what it classifies as a “Spanish-surname,” so there is an element of mismatch in this analysis. 10 These results are consistent with those obtained when the subset of counties in which at least one RID was filed are analyzed.

15

Hispanic voter identification, the counts display a strong, positive correlation (r>.97) across the count range. Moreover, relatively few counties display observed counts that exceed expected ones. Thus, there is little reason to conclude that Hispanic voters filed RIDs at at rate higher than what would be expected, given their county-level participation in the election. However, perhaps there are important differences among voters who filed RIDs. As noted above, a large number of voters reported having obtained some identification in the past, which suggests that they could conceivably do so again. It is therefore worth examining whether Hispanics are more likely than non-Hispanics to report longer-term issues that would seemingly create a more serious impediment to obtaining identification. To that end, Figure 4 depicts the reasons that Hispanic and non-Hispanic voters cited as a reasonable impediment on their declarations, once again employing the two decision rules. As is evident from the columns in Figure 4, regardless of the rule, Hispanic and Non-Hispanic voters demonstrated nearly identical propensities to select all but two of the categories: Specifically, Hispanic voters who filed RIDs were much more likely than non-Hispanics to report a lost or stolen ID, and less likely to select the “other” category. The distribution of responses evident in Figure 4 therefore offers little reason to conclude that Hispanic voters were more likely to report lacking identification due to economic hardship. That said, I explore the relationship between Hispanic status and reasons cited on RIDs in greater depth using mixed-effects logistic regression models of voters (level 1) in counties (level 2). I fit models of two dichotomous dependent variables. The first is an indicator for whether a voter reported a lack of transportation or necessary documents, family or work obligations, a disability, or financial reasons for not having proper identification. The second indicates that the voter reported a relocation-related impediment, such as an out-of-state ID or a recent move. The models can therefore determine whether Hispanic voters were more likely to report true hardships (compared to more “temporary” impediments such a misplaced ID) that prevented them from obtaining identification, and also whether they were more likely to experience difficulty associated with moving more frequently.11 11

These models employ the .5 probability rule.

16

Figure 4: Distribution of RID Reasons, by Hispanic Status

Table 3 includes coefficients and standard errors derived from the mixed effects logistic regression models. Both models feature random county intercepts, as well as fixed effects at both the county (demographics) and individual (Hispanic indicator) level. Unsurprisingly given the evidence presented so far, the models do not support the notion that Hispanic voters experience difficulties stemming from either hardship or an itinerant lifestyle at greater rates than non-Hispanics. In fact, both models return a significant, negative coefficient for the Hispanic indicator. Thus, the models suggest that if anything, Hispanics were less likely to report an RID due to either an enduring hardship or a recent relocation.

17

Table 3: Mixed Effects Logistic Regression Models: Determinants of RIDs Filed Due to Hardship or Relocation

Hardship

Relocation

-0.192* (0.051)

-0.718* (0.073)

1.061 (0.706) 1.580 (1.537) -0.907 (1.184) -0.484 (0.822) -4.480* (2.016) -3.636 (1.919) 2.564 (1.799) 0.140 (1.667) 1.512 (1.933)

1.888 (1.106) -1.038 (2.353) -2.533 (1.792) -2.439 (1.258) -4.818 (3.267) 0.988 (3.026) 4.382 (2.680) -4.370 (2.581) -0.981 (2.999)

0.500 (0.055)

0.752 (0.089)

Number of RIDS Filed (Level 1) 16,122 Number of Counties (Level 2) 184 Log Likelihood -10,193 Wald Chi-Squared 24.14 * p<0.05. Standard errors in parentheses.

16,122 184 -7,552 141.4

Individual-Level Fixed Effect: Voter is Hispanic County-Level Fixed Effects: Dem. 2-Party Vote Share, 2012 2016 Turnout Perc. African-American Perc. Hispanic Perc. Households Below Poverty Perc. High School Graduates Perc. Aged 18-24 Perc. Aged 65+ Constant Random Effects: S.D. (Counties)

18

6.4

County-Level Turnout

Thus far, the results indicate that while SB 14 would have impeded several thousand voters who turned out in 2016 from voting, there do not appear to be disparate effects of the law for Hispanic voters. Importantly however, since an analysis of RIDs relies on information from voters who turned out, it cannot assess the possibility that news of the law’s passage prevented voters from going to the polls. I therefore conclude by comparing turnout in 2012 and 2016 separately for Hispanic and non-Hispanic voters. The state of Texas identifies Spanish-surnamed voters in its voter file. Taking Spanish surname as a proxy for Hispanic status allows for the calculation of Hispanic and Non-Hispanic voter turnout at the county level. The clear advantage of this approach is that it allows for the trend among non-Hispanics to establish a counterfactual.

Figure 5: Turnout Trends from 2012 to 2016, Spanish-Surnamed and Non-Spanish Surnamed Voters

Figure 5 therefore plots mean county-level turnout for the two groups in each election. NonHispanic turnout is higher in both elections; in 2012, Non-Hispanics turned out at a rate nearly ten points higher than Hispanics (58.3% to 48.7%). However, while turnout among non-Hispanics increased from 2012 to 2016 by more than five points (to 63.7%), Hispanic turnout decreased by a similar margin during that period, dropping to 43.6%. The difference-in-differences therefore 19

implies that some effect in the 2016 election depressed Hispanic turnout (relative to non-Hispanic turnout) by about 10.4 percentage points. Thus, while there is little evidence in the RID data for a disparate impact of SB 14 on Hispanic voters, this result suggests a disproportionate number of Hispanics failed to turn out in 2016. This method cannot definitively identify a cause for such an effect, but it also cannot rule out SB 14 as the reason for this change.

7

Conclusion

The number of states requiring photo ID rose at a steady pace in the decade since Indiana first began requiring them in 2005; at least seven states now require photo ID before allowing people to vote, at least two such “strict ID” policies (including in Texas) have been thwarted by federal court actions, and new bills are pending in a number of others. As an ever-growing number of voters are subject to strict photo ID laws, the debate over them takes on increased importance. It is therefore crucial to consider how voter ID laws affect voters’ ability to cast a ballot, and also whether the laws are discriminatory in burdening some voters more than others. The District Court’s order to require a record of all Texas voters who turned out while lacking identification offers unprecedented insight into questions surrounding voter ID policies. My results indicate that more than 16,000 eligible voters who arrived at the polls with the intention of casting a vote would have been disenfranchised by SB 14; this figure amounts to roughly one-fifth of one percent of voters. It is important to note however that this sum almost certainly represents the lower bound of SB 14’s impact. Presumably, the voters who turned out without proper identification either had heard about the court order allowing them to vote, or were completely unaware that Texas had passed a strict photo ID law. However, it is impossible to tally the number of voters who heard that a voter ID law had taken effect, did not realize that they would still be allowed to vote, and therefore never turned out at all. A simple difference-in-differences calculation suggests that this effect might be meaningful for Hispanics, as the difference in Spanish-surname turnout from 2012 to 2016 lagged that of Anglo turnout by more than ten points. This result is consistent with, but not definitive proof of, SB 14 disenfranchising a disproportionate number of Hispanic voters. That said, my analysis yields no further evidence of SB 14 bearing more heavily on Hispanics. The percentage of RIDs filed by Hispanics did not exceed expectations given their participation in

20

the election, nor were Hispanics more likely to report lacking an ID due to hardship or relocation; Hispanics were in fact less likely to report these difficulties. Rather, the only county-level predictor of a higher percentage of voters filing RIDs was Democratic vote share. This is consistent with much previous work that has noted the partisan and/or ideological character of strict voter ID legislation (Biggers and Hanmer, 2017; Hicks et al., 2015; Hicks, McKee and Smith, 2016; McKee, 2015; Rocha and Matsubayashi, 2014; Wilson and Brewer, 2013), as well as with other work noting that Democratic affiliation is an important predictor of disenfranchisement (Hajnal, Lajevardi and Nielson, 2017). Thus, my results indicate that SB 14 may have effectively targeted Democratic voters. In examining claims made by real voters lacking identification in an actual election across an entire, large state, these findings mark an important contribution to the body of literature on voter ID, most of which relies solely on election or survey data. However, it is also important to point out the limitations of this paper. SB 14 was unique in that it was at the time of its passage the strictest photo ID law in the United States. It is unclear whether that should affect the applicability of these results to other states, but it is still important to stress that these findings are from a single election in a single state. As always, future work should replicate and extend these results as more data become available. Finally, while my findings indicate a higher proportion of RID filings in Democratic-leaning counties, they do not speak to the mechanism by which this may be occurring. Future work should more thoroughly evaluate whether and how the Democratic Party and affiliated groups might mitigate the deleterious effects of laws like SB 14, perhaps through education or focused mobilization efforts.

21

References Ansolabehere, Stephen. 2009. “Effects of identification requirements on voting: Evidence from the experiences of voters on Election Day.” PS: Political Science & Politics 42(1):127–130. Atkeson, Lonna Rae, Yann P Kerevel, R Michael Alvarez and Thad E Hall. 2014. “Who asks for voter identification? Explaining poll-worker discretion.” The Journal of Politics 76(4):944–957. Barreto, Matt A, Stephen A Nuno and Gabriel R Sanchez. 2009. “The disproportionate impact of voter-ID requirements on the electorate: new evidence from Indiana.” PS: Political Science & Politics 42(1):111–116. Biggers, Daniel R and Michael J Hanmer. 2017. “Understanding the adoption of voter identification laws in the American states.” American Politics Research p. 1532673X16687266. Bowler, Shaun, Thomas Brunell, Todd Donovan and Paul Gronke. 2015. “Election administration and perceptions of fair elections.” Electoral Studies 38:1–9. Citrin, Jack, Donald P Green and Morris Levy. 2014. “The effects of voter ID notification on voter turnout: Results from a large-scale field experiment.” Election Law Journal 13(2):228–242. Cobb, Rachael V, D James Greiner and Kevin M Quinn. 2012. “Can voter ID laws be administered in a race-neutral manner? Evidence from the city of Boston in 2008.” Quarterly Journal of Political Science 7(1):1–33. Downs, Anthony. 1957. “An economic theory of political action in a democracy.” Journal of Political Economy 65(2):135–150. Grimmer, Justin, Eitan Hersh, Marc Meredith, Jonathan Mummolo and Clayton ¨ Nall. 2017. “Comment on Voter Identification Laws and the Supression of Minority Votes¨.” Typescript . Hajnal, Zoltan, Nazita Lajevardi and Lindsay Nielson. 2017. “Voter Identification laws and the suppression of minority votes.” The Journal of Politics 79(2):363–379. Hansford, Thomas G and Brad T Gomez. 2010. “Estimating the electoral effects of voter turnout.” American Political Science Review 104(2):268–288. Hasen, Richard L. 2012. The voting wars: From Florida 2000 to the next election meltdown. Yale University Press. Hicks, William D, Seth C McKee and Daniel A Smith. 2016. “The determinants of state legislator support for restrictive voter ID laws.” State Politics & Policy Quarterly 16(4):411–431.

22

Hicks, William D, Seth C McKee, Mitchell D Sellers and Daniel A Smith. 2015. “A principle or a strategy? Voter identification laws and partisan competition in the American States.” Political Research Quarterly 68(1):18–33. Hood, MV and Charles S Bullock. 2008. “Worth a thousand words? an analysis of Georgia’s voter identification statute.” American Politics Research 36(4):555–579. Hood, MV and Charles S Bullock. 2012. “Much ado about nothing? An empirical assessment of the Georgia voter identification statute.” State Politics & Policy Quarterly 12(4):394–414. Jones, Mark P., Renee Cross and Jim Granato. 2017. “The Texas Voter ID Law and the 2016 Election: A Study of Harris County and Congressional District 23.” Typescript . Malewitz, Jim. 2016. “Texas Voter ID Law Violates Voting Rights Act, Court Finds.” The Texas Tribune . [Online; accessed 13-July-2017]. McCarthy, Justin. 2016. “Four in Five Americans Support Voter ID Laws, Early Voting.” Gallup . [Online; accessed 11-July-2017]. McKee, Seth C. 2015. “Politics is local: State legislator voting on restrictive voter identification legislation.” Research & Politics 2(3):2053168015589804. Minnite, Lorraine C. 2010. The myth of voter fraud. Ithaca. Mycoff, Jason D, Michael W Wagner and David C Wilson. 2009. “The Empirical Effects of Voter-ID Laws: Present or Absent?” PS: Political Science & Politics 42(1):121–126. Perkins, R Colby. 1993. Evaluating the Passel-Word Spanish surname list: 1990 decennial census post enumeration survey results. US Department of Commerce, Economics and Statistics Administration, Bureau of the Census. Riker, William H and Peter C Ordeshook. 1968. “A Theory of the Calculus of Voting.” American political science review 62(1):25–42. Rocha, Rene R and Tetsuya Matsubayashi. 2014. “The politics of race and voter ID laws in the states: The return of Jim Crow?” Political Research Quarterly 67(3):666–679. Sobel, Richard and Robert Ellis Smith. 2009. “Voter-ID laws discourage participation, particularly among minorities, and trigger a constitutional remedy in lost representation.” PS: Political Science & Politics 42(1):107–110. Valentino, Nicholas A and Fabian G Neuner. 2017. “Why the Sky Didn’t Fall: Mobilizing Anger in Reaction to Voter ID Laws.” Political Psychology 38(2):331–350. White, Ariel R, Noah L Nathan and Julie K Faller. 2015. “What do I need to vote? Bureaucratic discretion and discrimination by local election officials.” American Political Science Review 109(1):129–142.

23

Wilson, David C and Paul R Brewer. 2013. “The foundations of public opinion on voter ID laws: Political predispositions, racial resentment, and information effects.” Public Opinion Quarterly 77(4):962–984.

24

Appendix

Table 4: Characteristics of All Counties, and Counties in Samples

Variable Name

All Counties

Recorded At Least One RID

N=254

N=184

Mean

S.D.

Min.

Max.

Mean

S.D.

Min.

Max.

Population

104,483

375,413

117

4,356,362

140,919

435,834

565

4,356,362

Med. HH Income

$46,746

$10,788

$22,741

$89,152

$47,868

$10,865

$22,741

$89,152

Clinton Vote Perc.

0.260

0.166

0.032

0.807

0.278

0.164

0.070

0.792

2016 Turnout

0.594

0.073

0.368

0.821

0.592

0.066

0.394

0.821

0.064

0.067

0.000

0.336

0.074

0.070

0.000

0.336

Perc. Hispanic

0.338

0.230

0.028

0.987

0.330

0.227

0.039

0.953

Perc. HH Poverty Perc.

0.131

0.056

0.000

0.372

0.133

0.053

0.009

0.372

Perc. HS Grad.

0.793

0.079

0.463

0.966

0.801

0.072

0.591

0.934

Perc. Aged 18-24

0.091

0.032

0.008

0.290

0.096

0.032

0.034

0.290

Perc. Aged 75+

0.073

0.025

0.000

0.171

0.068

0.022

0.029

0.148

Perc. African-American

Recorded no RIDs N=70 Variable Name Population

Mean

S.D.

Min.

Max.

8,708

11,504

117

62,648

Med. HH Income

$43,752

$10,052

$26,172

$73,750

Clinton Vote Perc.

0.213

0.165

0.032

0.807

2016 Turnout

0.596

0.091

0.368

0.788

0.040

0.050

0.000

0.248

Perc. Hispanic

0.358

0.236

0.028

0.987

Perc. HH Poverty Perc.

0.125

0.063

0.000

0.354

Perc. HS Grad.

0.772

0.094

0.463

0.966

Perc. Aged 18-24

0.078

0.026

0.008

0.140

Perc. Aged 75+

0.085

0.029

0.000

0.171

Perc. African-American

County-Level Demographic Data From 2015 ACS 5-Year Averages

25