New Analysis Shows Voter Identification Laws Do Not Reduce Turnout

OVERVIEW

The 2000 presidential election sparked a fire­storm of debate relating to election reform in the United States. Since then, academics, the media, and elected officials have proffered opinions and implemented policies related to this important political issue. Topics that have been addressed in recent years range from modernizing voting machines and updating voter registration rolls to implementing stricter identification requirements for voting.

In 2002, Congress passed the Help America Vote Act (HAVA).[1] HAVA affects only federal elec­tions and, among other things, requires that the states provide for provisional voting; create a computerized, centralized list of registered voters; and ensure that new voters who register by mail present identification before being allowed to vote in person. HAVA established the Election Assis­tance Commission (EAC) to serve as "a national clearinghouse and resource for information and review of procedures with respect to the adminis­tration of federal elections."[2] Additionally, many state legislatures have enacted their own election reform legislation.[3]

Of the many election reforms currently being considered, one that has incited some of the most cantankerous debate is that of voter identification at the polls. For many, the idea of requiring voters to present identification in order to vote is anath­ema, tantamount to the poll taxes that were once used to prevent African-Americans from voting.[4] They contend that requiring identification at the polls will lead to lower voter turnout, especially among the poor, certain minorities, and the elderly. For others, such as the Protect Arizona Now orga­nization that lobbied in favor of identification requirements for Arizona voters, the problem of voter fraud makes voter identification require­ments a common-sense solution.[5] The standard argument goes that if a person has to show identi­fication to board a plane or cash a check, why shouldn't he have to do the same in order to vote? Additionally, the proponents of stricter voter iden­tification requirements argue that such a policy would bolster the public's faith in the legitimacy of elections and lead to greater voter turnout, not less.

Both sides raise valid concerns. However, even a cursory glance at the literature on voter identifica­tion requirements shows that there is a dearth of empirical research on this issue. While there have been a few studies to address the effect of voter identification requirements using election data,[6] more research is needed in order to appropriately assess the legitimacy of either side's claims.

In response to this debate, the EAC awarded a grant to Rutgers University's Eagleton Institute of Politics and the Moritz College of Law at Ohio State University to study voter identification require­ment laws. The resulting study, Report to the U.S. Election Assistance Commission on Best Practices to Improve Voter Identification Requirements Pursuant to the Help America Vote Act of 2002,[7] included a sta­tistical analysis of the effect of voter identification requirements on voter turnout during the 2004 election by Professor Timothy Vercellotti of the Eagleton Institute.[8] A new version of the analysis with Timothy Vercellotti and David Anderson as authors was presented to the 2006 American Polit­ical Science Association conference.[9] Hereinafter, this study will be referred to as the "Eagleton Insti­tute study."

The Eagleton Institute study found that more stringent voter identification requirements appeared to reduce voter turnout in 2004.[10] In the media, their study has been cited as demonstrating that the strengthening of voter identification requirements to reduce fraud has the side effect of suppressing minority voter turnout.[11]

This Center for Data Analysis report attempts to replicate the part of the Eagleton Institute study that used the publicly available November 2004 Current Population Survey (CPS).[12] This analysis was done because several aspects of the Eagleton Institute study cast doubt on the validity of its findings:

• The Eagleton Institute used one-tailed hypothe­sis tests instead of the more commonly accepted two-tailed tests. The one-tailed test allows researchers to double their chances of finding statistically significant results.
• The 2004 voter identification laws of certain states were misclassified. For example, Arizona and Illinois were incorrectly classified as requir­ing voters to provide identification and state their name for authentication, respectively. However, in 2004 Arizona only required voters at polling stations to sign their name for authen­tication, while Illinois required poll workers to match the signatures of voters.
• Some of the variables used to predict the deci­sion to vote were used inappropriately. For example, the Eagleton Institute study used the November 2004 CPS family income variable, which is an ordinal variable of unequal income ranges, as an interval-ratio variable. Using cate­gorical variables as interval-ratio variables can lead to estimation problems.

After addressing these issues, our reanalysis finds that some of the original findings of the Eagleton Institute study are unfounded. Controlling for fac­tors that influence voter turnout, voter identifica­tion laws largely do not have the negative impact on voter turnout that the Eagleton Institute suggests. When statistically significant and negative relation­ships are found, the effects are so small that the findings offer little policy significance. For example, our analysis indicates that:

• White survey respondents in photo identifica­tion states are 0.002 percent less likely to report voting than white respondents from states that only required voters to state their name.
• African-American respondents in non-photo identification states are 0.012 percent less likely to report voting than African-American respon­dents from states that only required voters to state their name.

In other cases, no effect was found.

• In general, respondents in photo identification and non-photo identification states are just as likely to report voting compared to respondents from states that only required voters to state their name.
• African-American respondents in photo identi­fication states are just as likely to report voting compared to African-American respondents from states that only required voters to state their name.
• Hispanic respondents in photo identification states are just as likely to report voting compared to Hispanic respondents from states that only required voters to state their name.

BACKGROUND

When discussing voting behavior, it is important to consider the factors that influence whether an individual votes or not. According to the "Calculus of Voting" model, an individual will vote when the rewards from voting are positive and will abstain when they are not. The equation for the Calculus of Voting model is as follows:

R = PB - C + D.

The rewards (R) from voting are determined by multiplying the benefits (B) an individual receives when his preferred candidate wins over a less pre­ferred candidate by the probability (P) that his vote will make a difference plus the benefits one receives from voting as an act of fulfilling one's duty or civic obligation (D) minus the costs of voting (C).[13] This is the standard, rational model of voting and will be used to inform the following discussion of voter identification requirements and their effect on voter turnout.

The voter identification issue is often framed as being torn between the opposing aims of "access and integrity."[14] By this we mean that it is com­monly perceived that while voter identification laws may be effective at preventing ineligible individuals from voting (integrity), they may have an adverse effect on the ability of every eligible voter to vote (access). There have been only a few empirical stud­ies on the impact of voter identification require­ments,[15] but this does not translate into a lack of opinions on this topic.

Advocates for more stringent voter identification laws contend that this reform is vital to prevent voter fraud.[16] As more and more elections are won by slim margins, proponents of identification requirements argue that the chances are greater that voter fraud could affect election outcomes.[17] The potential for a small number of voters to have a sig­nificant impact on the outcome of an election became all too evident in the 2000 presidential elec­tion. Given that George W. Bush was declared the winner in Florida (and the next President) by a mar­gin of 537 votes, it follows that even a small number of fraudulent votes (537+1) would matter a great deal.[18] In 2004, there were allegations of voter fraud in the Washington gubernatorial election in which Christine Gregoire won by a margin of 129 votes.[19] Certainly the potential of voter fraud is a matter of concern.

Broadly defined, voter fraud is "the intentional corruption of the electoral process by voters."[20] While voter fraud manifests itself in different forms, examples include individuals who vote but are ineligible (such as non-citizens and felons), individuals who vote multiple times in various pre­cincts, and individuals who vote using someone else's name. Because of the lack of research and the difficulty of collecting data on voter fraud, the extent to which these kinds of voter fraud occur is unknown. Additionally, for similar reasons, we are unaware of the extent to which voter identification laws would curb the type of voter fraud they are intended to prevent.

However, there are some examples of recorded voter fraud. The Department of Justice asserts that since the inception of the Attorney General's Ballot Access and Voting Integrity Initiative in 2002, 120 people have been charged with election fraud, of which 86 have been convicted.[21] Additionally, the Milwaukee Journal Sentinel reports that prosecutors in Milwaukee filed charges against 14 individuals for voter fraud in the 2004 election.[22] Of the 14, 10 were felons accused of voting and four were accused of double voting. Prosecutors obtained five convic­tions. For proponents of strict voter identification requirements, the knowledge that any voter fraud occurs is sufficient to argue that more needs to be done to curb this problem.[23]

The most prevalent critique of the voter fraud argument is that "voter-fraud anecdotes are often misleading, incomplete, and unrepresentative."[24] Proponents of this view contend that upon closer examination of claims of voter fraud, such charges turn out to be either nonexistent or infrequent. For instance, the Brennan Center for Justice at the New York University School of Law found that in 2004, voter fraud occurred 0.0009 percent of the time in the gubernatorial election in Washington and 0.00004 percent of the time in Ohio. They report that these percentages are akin to the likelihood of an American's being killed by lightning.[25]

Opponents of voter identification requirements also argue that the few instances of voter fraud that may be prevented by identification laws do not out­weigh the thousands of legitimate voters who would be disenfranchised because they lacked the neces­sary identification.[26] These critics argue that identi­fication laws will have a negative impact on the ability of certain minorities, the elderly, the dis­abled, and the poor to vote.[27] It is presumed, and some studies have found, that people from these groups are less likely to possess drivers' licenses or other government-issued identification.[28] It is also assumed that many from these groups would be unable or unwilling acquire the necessary docu­mentation. Critics of strict identification laws fur­ther argue that the costs (in both time and money) of obtaining such documentation would be a deter­rent to voting and would likely result in lower voter turnout among poor voters and those who do not have easy access to government offices.[29] It is for this reason that "ID requirements are compared to modern poll taxes."[30]

While it is difficult to accurately assess the num­ber of eligible voters who would be rendered unable to vote because they lack proper identification, some studies have attempted to estimate such fig­ures by looking at the percentage of the population who do not have driver's licenses. For instance, a Wisconsin study found that when considering the entire state, 80 percent of men and 81 percent of women had valid driver's licenses. In contrast, only 45 percent of African-American men and 51 per­cent of African-American women had valid driver's licenses. The percentages for Latinos were also lower (54 percent for men and 41 percent for women).[31] Similarly, a Georgia study found that among registered voters, non-whites, women, and the elderly were less likely to have government-issued photo identification (either a driver's license or state identification).[32]

Although these figures shed light on the types of people who are less likely to have driver's licenses, it is unadvisable to focus on this statistic alone. First, the data still cannot tell us whether those individu­als without driver's licenses have some other form of identification, such as an employee ID, student ID, social security card, or any other form of identifica­tion currently accepted in many states. Second, it cannot tell us about future behavior. Do voters in photo identification states who lack the necessary identification obtain the required identification (such as a driver's license) when the state law is changed? Take for instance the previous study con­ducted in Wisconsin, which currently does not require identification before voting (except for those requirements set forth in HAVA for new voters). Although approximately half of African-Americans in the state are currently without driver's licenses, we do not know if those individuals will get driver's licenses or state IDs if Wisconsin were to require voters to show identification before voting.

For these reasons, proponents of voter identifica­tion requirements are convinced that requiring identification at the polls would not be an excessive burden to voters. As previously mentioned, identi­fication is required for many things that are consid­erably less important than voting (flying in a plane, buying alcohol, etc.). As "voting is equally impor­tant," if not more important, the argument goes that it makes sense for someone to be required to show identification in order to cast a ballot.[33] Addition­ally, Senior Research Scientist John Lott at the Uni­versity of Maryland Foundation points out that as "almost 100 countries require photo identifications to vote," the United States would be hardly alone in requiring voters to show some form of identification at the polls.[34]

Those who oppose voter identification at the polls argue that other reforms are better suited to preventing voter fraud. For instance, critics of voter identification point to absentee ballots as "the Achilles heel of election security" because voters are often not required to show identification at all.[35] Yet absentee ballots have been largely left out of the voter identification requirement debate. This apparent discrepancy has been used by opponents of voter identification laws as evidence that sup­porters of such legislation are not interested in real voter fraud reform.[36] Rather, critics argue that voter identification supporters are using such laws as an attempt to suppress voter turnout by increas­ing the costs of voting (the "C" from the Calculus of Voting model).[37]

Another argument proffered by supporters of voter identification requirements is that such laws are necessary to maintain the public's faith in the integrity of elections. The Commission on Federal Election Reform (Carter-Baker Commission) at American University asserts that "the electoral sys­tem cannot inspire public confidence if no safe­guards exist to deter or detect fraud or to confirm the identity of voters."[38] This argument, "the ensur­ing integrity hypothesis," contends that public faith in the honesty of elections actually "encourages additional voter participation."[39] Proponents argue that voter identification laws will bolster the public's faith in the outcome of elections. This will increase, not decrease, turnout because voters will feel a greater pride in voting (increasing the "D" or duty component of voting).

Voter identification laws are exceptionally popular among the general public. In a survey of some 36,000 voters, Professors Stephen Ansolabehere and Elting R. Morison of the Massachusetts Institute of Technology found that 77 percent of respondents supported voter identification requirements.[40] For the most part, the majority of respondents sup­ported such laws regardless of race, location (North­east, Midwest, etc.), and political ideology. While those who identified themselves as conservatives had the highest percentage of agreement with iden­tification requirements (at 95 percent), even those who identified themselves as "very liberal" had 50 percent agreement with voter identification laws.[41] Regarding race, more than 70 percent of whites, African-Americans, and Hispanics supported voter identification laws.[42] Additionally, Ansolabehere found only 23 instances out of 36,000 where an individual reported being unable to vote because he lacked the necessary identification.[43]

These survey data are supported by actual voter behavior. In 2004, when Arizonans voted on Prop­osition 200, which would require voter identifica­tion at the polls as evidence of citizenship, it passed with 56 percent of the vote.[44]

Ultimately, it is not the intent of this paper to debate the merits of either side's arguments. Rather, we want to present the major arguments on either side of this issue as background to our analysis. However, the paper does intend to examine more closely one of the claims of this debate: that stricter voter identification requirements depress voter turnout. In order to do that, it is necessary to discuss the different voter identification re­quirements across the 50 states and the District of Columbia.

Voter identification requirements, if any, differ by state, so there is great variability in the way voters from different parts of the country are required to verify their identity before casting a ballot. Some states rely on the honor system where voters merely have to give their names to the election official.[45] Other states only require a signature,[46] with some states going a step further and actually matching the signature to a previously signed document.[47] States with more stringent requirements ask that voters provide identification[48] or photo identification.[49]

The Eagleton Institute study identified two cate­gories of identification requirements (maximum requested and minimum required) and five types of identification requirements (stating name, signing name, signature match, present ID, and photo ID).[50] It is important to note that in 2004, there were no states that had photo ID as a minimum requirement. All states that had a photo ID require­ment permitted voters who did not have such doc­umentation to present alternative forms of ID or sign an affidavit attesting to their identity.[51]

By the maximum requested, the Eagleton Insti­tute study refers to the most identification that an individual can be asked to present in order to vote using a regular ballot. Conversely, the minimum is the least identification that will be accepted to vote.[52] For example, when voting in Louisiana in 2004, a voter would be asked by poll workers to present photo identification. If the individual was unable to present an acceptable form of ID, he was allowed to vote after signing an affidavit stating he is the person he claims to be.[53] In that case, photo ID would be the maximum requested, and affidavit would be the minimum required.

Within the states that require some form of doc­umentation as proof of identity, there are also signif­icant differences. For instance, some states, like Massachusetts, "may" ask that a voter show identifi­cation, but identification is not automatically requested of all voters.[54] In Alabama and Alaska, two states that request identification, this require­ment can be waived if a poll worker knows the voter and can attest to his identity.[55] This is an important issue to consider because it means that different vot­ers within the same state may be affected by differ­ent identification requirements.

Furthermore, by the 2004 election, many states had become compliant with certain provisions in the Help America Vote Act (HAVA) which required identification at the polls from first-time voters who registered by mail and who did not show iden­tification at the time of registration. One state, Pennsylvania, actually went above and beyond HAVA requirements and mandated that all first-time voters needed to show identification at the polls regardless of whether they showed identifica­tion when they registered to vote.[56] Because of HAVA, many first-time voters had to show identifi­cation at the polls even in states that did not other­wise require identification from all voters.

Even among states that require documentation, there is great variability in the types of documenta­tion that is accepted. Some accept only a govern­ment-issued photo identification, while others accept almost any document that demonstrates a person's identity. For example, in 2004, acceptable documentation in Florida ranged from a driver's license and passport to credit card and buyer's club card to utility bill, bank statement, or paycheck (as long as they contained the name and address of the individual).[57] In contrast, some states that required identification to vote are much more restrictive with respect to acceptable forms of identification. One such state, Virginia, only allowed voters to present a voter registration card, Social Security card, employer-issued identification card (as long as it contained a photo), Virginia driver's license, or other Commonwealth or government-issued identi­fication.[58] Furthermore, in many states, individuals who are unable to provide the appropriate docu­mentation are given an alternative, such as signing an affidavit, in order to vote. Finally, Section 302 of HAVA requires that an individual who fails to meet the identification requirements of voting can still vote using a provisional ballot.[59]

The key aspects of this brief overview of identifi­cation requirements of voting is that there is a lot of variability by states as to what is required, and not all identification requirements are created equal. By that we mean that required identification documen­tation for one state may not meet the identity requirements in another state. This is just one of the reasons that it is particularly difficult to study the effect of such laws on voter turnout.

THE DATA

In order to analyze individual voter turnout, this study uses data from the U.S. Census Bureau's Cur­rent Population Survey, November 2004: Voting and Registration Supplement File.[60] The November 2004 CPS voting supplement contains interviews from about 57,000 households. Based on self-described registered voters, the data allow us to model the decision to vote based on individual and household characteristics.

Dependent Variable. The dependent variable is whether or not the respondent reported that he or she voted in the November 2004 election. Respondents who admitted to not being regis­tered voters were omitted, along with those reporting that they were not United States citi­zens. We also omitted those reported to be voting through absentee ballots.[61]

According to the U.S. Census Bureau's analysis of the November 2004 CPS data, 89 percent of regis­tered voters voted in the November 2004 elec­tion.[62] This estimate is drawn from a sample of respondents reporting to be registered voters and is much higher than estimates based on samples of the voting-age population. However, the EAC estimates that 70.4 percent of registered voters turned out to vote.[63] The CPS estimate of 89 percent may be biased upward because it is based on the reported vote, which may be overstated because survey respondents may be disinclined to admit that they did not vote.[64] When turnout is based on the total population over 18 years old, 55.8 percent of per­sons over age 18 voted.[65]

Voter Identification Requirements. The voter identification requirements included in the analysis capture the degree to which a registered voter has to prove his or her identity at the polling station. Two sets of five dichotomous voter identification vari­ables are used in the analysis. The first set is based on the maximum amount of identification that the voter is required to produce in order to prove his or her identity. The maximum state voter identification requirements are broken down into the following classification: state name, sign name, match signa­ture, provide non-photo identification, and provide photo identification. Table 1 presents the voter identification classifications by state used by the Eagleton Institute and the Moritz College of Law at Ohio State University.

For all but two of the states, Illinois and Arizona, we used the classifications that were provided to us by the Eagleton Institute. We recoded these two states because upon researching state election laws, we discovered that the Eagleton Institute had erro­neously reported the identification requirements for these two states. The Eagleton Institute study has Illinois listed as a "state name" state. In actuality, Illi­nois poll workers match a prospective voter's signa­ture to a signature already on file, making Illinois a "match signature" state.[66]

The Eagleton Institute has Arizona listed as a "provide ID" state although Arizona was a "sign name" state at the time of the 2004 election.[67] Iden­tification laws did not go into effect in Arizona until some time after the 2004 election. Arizona could not have been a "provide ID" state before the November 2004 election because Arizonans voted on and approved Proposition 200 on the November 2004 ballot. This initiative is the impetus for the requirement that voters show identification before voting as proof of citizenship.[68]

The second set of voter identification variables recognizes that some states allow voters without proper identification to vote after demonstrating their identity through other means. This minimum requirement set of variables includes state name, sign name, match signature, provide non-photo identification, and swear affidavit. For the probit regressions, the variable for voters stating their names for identification is omitted for reference purposes.

Individual Factors. The individual factors included in the analysis capture differences in the race and ethnicity, age, education, household income, marital status, gender, employment status, citizenship, residential mobility, and home owner­ship of the individual respondents. Controlling for such variables as education and age is important because research indicates that these variables are good predictors of voting turnout.[69] The analysis controls for the effect of the individual's race and ethnicity through a set of mutually exclusive dichotomous variables for the following categories: non-Hispanic white, non-Hispanic African-Ameri­can, Hispanic, non-Hispanic American Indians, non-Hispanic Asians (including Hawaiians/Pacific Islanders), and other races, including those report­ing multiple races and ethnicities. The specification of these variables allows us to compare the voting patterns of minorities to those of whites.

A set of dichotomous variables control for the age of the individual respondents that fall into the fol­lowing categories: 18- to 24-year-olds, 25- to 44-year-olds, 45- to 64-year-olds, and 65 years and older. For education, the respondents were classified as either having less than a high school diploma, high school diploma or equivalent, some college, bachelor's degree, or a graduate school degree.

For family income, the Eagleton Institute study used an ordinal family income variable as an in­terval-ratio variable.[70] The family income variable is coded as 1 through 16 with units containing un­equal income ranges. For the purposes of this analy­sis, the effect of family income is controlled for by the inclusion of a series of income range dichoto­mous variables: under $15,000, $15,000 to $29,999, $30,000 to $49,999, $50,000 to $74,999, $75,000 to $149,999, and $150,000 or more.

To control for the influence of marital status, five dichotomous variables signifying being single, mar­ried, separated, divorced, and widowed are included in the model. Single individuals are the default. A dichotomous variable identifying the gender of the individual as a female is also included in the models.

Two dichotomous variables are included to con­trol for the effect of employment. The first is a dichotomous variable signifying whether or not the individual is employed; the second is a dichoto­mous variable for whether or not the person is in the labor force.

To control for whether native-born citizens are more likely to vote than naturalized citizens, a dichotomous variable identifying native-born citi­zens is included. Two dichotomous variables are included to control for community ties. The models control for whether or not the individual has moved within the last year and whether or not the individ­ual owns or rents his or her home. These two vari­ables are included to help control for social connectedness under the theory that those with stronger community ties will be more likely to vote.

State Political Factors. As with the Eagleton Institute study, two dichotomous variables indicate whether a state is considered a battleground state and a competitive state. A state is designated as a battleground state if the margin of victory for the winning 2004 presidential candidate was 5 percent or less. A state was designated as competitive if the margin of victory for governor and/or U.S. Senate races was 5 percent or less.

FINDINGS

The probit regression analyses that follow exam­ine the effects of voter identification requirements on voter turnout. Table 2 presents the original find­ings of the Eagleton Institute's probit regression analysis. Table 3 presents the descriptive statistics of the data used in Table 4. Based on our analyses, six sets of probit regression models are presented in Tables 4 to 9.

The first set of probit regressions contains our replication of the Eagleton Institute study for their analysis of all voters (Table 4). The second set of probit regressions presents the findings for all voters under a different model specification and the cor­rected classification of state identification require­ments for Arizona and Illinois (Table 5). The sixth through ninth sets of probit regressions present our findings for the different model specification and corrected coding for state identification require­ments for whites, African-Americans, Hispanics, and Asians (Tables 6 through 9).

For all of the models, robust standard errors are estimated to correct for correlated error terms within each state. For tests of statistical significance, the standard two-tailed tests are used. See below for a discussion of one-tailed versus two-tailed tests of statistical significance. The calculations in Tables 3 through 9 use the CPS weight, PWSSWGT, as rec­ommended by the Bureau of the Census.

---

One-Tailed Versus Two-Tailed Tests of Statistical Significance

When doing tests of statistical significance for hypotheses, social scientists generally use two-tailed tests. Two-tailed tests are used to check for a difference while ignoring in which direction the difference lies.

For example, a social scientist would use a two-tailed test to determine whether voters in photo identification and give name states have different probabilities of reporting having voted in the 2004 election, regardless of the direction of the relation­ship. By using a two-tailed test, the 5 percent prob­ability is split between both ends of the bell-shaped curve. (See Figure A in Chart 1.) That is, 2.5 percent of the probability that the difference is due to chance is placed in the side that represents respon­dents in photo identification states being less likely to vote, while 2.5 percent is placed in the side that represents respondents in photo identification states being more likely to vote. If the probit coeffi­cient for photo identification states falls within either of the 2.5 percent shaded regions, this finding is determined to be statistically significant. If the coefficient falls within the left (right) tail, photo identification requirements have a negative (posi­tive) relationship with reported voter turnout. If the coefficient falls between the 2.5 percent shaded regions, photo identification requirements are said have no relationship with voter turnout.

When one-tailed tests are used, social scientists are hypothesizing that the relationship between photo identification requirements and reported voting has a specific direction: for example, voter identification requirements decrease (increase) re­ported voting. As determined by the social scien­tist, all of the 5 percent of chance is placed in one end of the bell-shaped curve. If the direction of the relationship is as hypothesized, placing the entire 5 percent chance in one side makes it is twice as easy to achieve a statistically significant finding with a one-tailed test as with a two-tailed test. Figure B in Chart 1 is an example of a one-tailed test where the researcher believes a negative relationship exists. In the case of photo identification requirements and voter turnout, if the coefficient falls within the 5 percent shaded region of the left tail, photo identi­fication requirements would then be said to have a negative relationship. If the coefficient does not fall within the 5 percent region, then photo identifica­tion requirements are said to have no relationship with voter turnout.

According to norms of the social sciences, researchers generally use two-tailed tests. When they deviate from this norm, social scientists gen­erally provide a justification for why they have done so. Consumers of statistical research should be skeptical of findings based on one-tailed tests, especiallywhen such findings do not hold up under two-tailed testing.

---

Replicating the Eagleton Institute's Findings for All Voters

Table 2 contains the findings from the Eagleton Institute's probit regression for all registered voters as presented in their paper. Table 3 presents the find­ings from our attempt to replicate the Eagleton Insti­tute study findings for all voters. In our attempt at replicating the Eagleton Institute's study, we could not entirely match the same number of respondents. The Eagleton Institute's probit regression of all voters is based on 54,973 respondents.[71] Our best attempt at replicating their analysis produced 54,829 respondents-144 fewer respondents. In addition, the results reported in Table 3 use the more com­monly accepted two-tailed significance tests.

While the Eagleton Institute reported that states with sign name, non-photo identification, and photo identification requirements have lower voter turnout than states with only the state name requirement, only the photo identification coeffi­cient in our attempt at replication (Model 1) is sta­tistically significant at the 95 percent confidence level. Respondents from photo identification states are less likely to have reported voting compared to respondents in states that only required voters to say their names at the polling stations. The magni­tude of the negative relationship between photo identification requirements and voter turnout is dif­ficult to interpret with probit coefficients, so the elasticity was calculated. The elasticity figures used in this analysis represent the percentage change in the probability of reporting to vote given a one-unit change in a particular dichotomous independent variable. The survey respondents in photo identifi­cation states are 0.002 percent less likely to report voting than respondents from states that only required voters to give their name for identification.

Model 2 corrects for the Eagleton Institute study's misclassification of the voter identification require­ments in Arizona and Illinois. With the correction, all of the state voter identification variables are sta­tistically insignificant-meaning that none of these requirements has a statistically measurable relation­ship with voting turnout.

Model 3 attempts to replicate the findings of the Eagleton Institute's examination of the effect of minimum requirements. As seen in Table 2, the Eagleton Institute found that the coefficients for sign name, non-photo identification, and swear affidavit states had statistically significant, negative relationships with voter turnout using one-tailed significant tests. However, our analysis pre­sented in Model 3 using two-tailed statistical significance tests finds only the swear affidavit coefficient to be statistically significant at the 95 per­cent confidence level. The survey respondents in swear affidavit states are 0.002 percent less likely to report voting than respondents from states that only required voters to state their name for identification.

It should be noted that although we ran the minimum identification requirement model using the classifi­cations assigned to the states by the Eagleton Institute study, there are some issues with the states considered to have an affidavit as the minimum requirement. These issues should be addressed in follow-up studies. First, the Eagleton Institute study identified only four states as having a minimum requirement of sign affidavit. They are Florida, Indiana, Louisiana, and North Dakota. All but one of these states, Indiana, require some form of identification as the maximum requested. This puts Indiana in the precarious position of requiring, at a maximum, that a voter sign his name before receiving a ballot; if he is unable to do so, he can sign an affidavit and vote. This does not make sense, because Indiana in 2004 did not require identification before voting (other than for those affected by HAVA requirements).

We believe this to be another classification error on the part of the Eagleton Institute. According to the "2004 Indiana Election Day Handbook," the procedure for signing an affidavit only applies to challenged voters who are then given a provi­sional ballot if they sign the affidavit.[72] This vot­ing method would not fall under the guidelines set forth by the Eagleton Institute because it applies to provisional, and not regular, ballots.[73] For these reasons, we believe Indiana should have a minimum identification requirement of sign name, the same as its maximum.

Additionally, there are five other states (Con­necticut,[74] Delaware,[75] Georgia,[76] South Dakota,[77] and Virginia[78]) that require some form of identifi­cation but make exceptions and allow voters with­out the required documentation to sign an affidavit in order to vote. To be classified correctly, these states should also be considered to have a mini­mum requirement of sign affidavit as they too pro­vide opt outs for voters unable to show appropriate forms of identification.

As for the socioeconomic variables in Models 1 through 3, African-Americans are more likely to have reported voting in the election than a group­ing of non-Hispanic whites, American Indians, Hawaiians/Pacific Islanders, and others. In con­trast, Asians are less likely to report voting. Respon­dents aged 45 and above are more likely to report voting than those 18 to 24 years old. Those with an education at or above a high school diploma are more likely to report voting than those without a high school degree. Family income has a positive relationship with the proba­bility of reporting having voted. Mar­ried and female respondents are more likely to report voting than not married and male respondents, respectively. Respondents residing in battleground states are more likely to vote, while respondents who moved within the last six months are less likely to report voting.

Alternative Model Specifications

Concerns regarding some of the variables used in the Eagleton Institute study led us to estimate alternative specifications that use the November 2004 CPS data more appropriately.

First, the Eagleton Institute's race and ethnicity dichotomous variables compare African-Americans, Hispan­ics, and Asians to the default group of whites, American Indians, Alaskan Natives, Hawaiians/Pacific Islanders, and those reporting to be more than one race and/or ethnicity. For exam­ple, the Eagleton Institute found that African-Americans were more likely to report voting compared to whites, American Indians, Alaskan Natives, Hawaiians/Pacific Islanders, and those reporting to be more than one race and/or ethnicity.

The descriptive statistics of the data used for the alternative specifi­cations are presented in Table 4. The analyses in Table 5 control for the effect of the individual's race and ethnicity through a set of mutually exclusive dichotomous variables for the following categories: non-His­panic whites, non-Hispanic African-Americans, Hispanics, non-Hispanic American Indians and Alaskan Natives, non-Hispanic Asians (including Hawaiians/Pacific Islanders), and other races, including those reporting multiple races and eth­nicities. For example, this division of race and eth­nic groups allows us to present clearer estimates of how voter identification laws affect the voting probabilities of minorities compared to whites.

Second, the Eagleton Institute study used an ordi­nal family income variable as an interval-ratio vari­able. Using categorical variables as interval-ratio variables can lead to estimation problems, so for the purposes of this analysis, the effect of family income is controlled for by the inclusion of a series of income range dichotomous variables.

Third, the effect of photo identification variables on voter turnout is very sensitive to how the models control for marriage. In addition to a dichotomous variable for whether or not the respondent reported being married, additional dichotomous variables were added for those reporting to be widowed, sep­arated, and divorced. This minor change in marital control variables has a significant impact on the results for the relationship between voter turnout and some of the voter identification variables.

Fourth, the alternative models control for whether or not the individual has moved within the last year instead of the six-month time period used by the Eagleton Institute.

Fifth, a variable indicating whether or not the respondent owns or rents his or her home was added to the alternative models. The residential mobility and home ownership variables help to control for how connected the respondents are to their communities.

Table 5 presents the findings of the alternative model specification for all respondents. Model 4 contains the revised race/ethnicity and income vari­ables along with the variables for residential mobil­ity and home ownership. Of the four voter identification variables, only the photo identifica­tion variable is statistically significant. Photo identi­fication states have respondents that are less likely to have reported voting compared to respondents in states that only required voters to say their names at the polling stations. However, the difference is very small. The survey respondents in photo identifica­tion states are 0.002 percent less likely to report vot­ing than respondents from states that only required voters to state their name for identification.

A slight change in how marital status is con­trolled for in Model 5 makes the findings in Model 4 for photo identification requirements disappear. The inclusion of dichotomous variables to identify respondents if they are widowed, divorced, or separated, in addition to being married, signifi­cantly changes the results for the photo identifica­tion variable. A photo identification requirement no longer has a statistically significant relation­ship with voter turnout. Thus, the finding that photo identification requirements reduce voter turnout in Model 4 is not robust to an alternative model specification.

In Models 6 and 7, Arizona and Illinois are reclassified correctly as requiring voters at polling stations to sign their name and match signatures, respectively. As with Model 4, Model 6 uses only a married dichotomous variable to control for mar­ital status. Model 7 includes additional marital status variables as used in Model 5. After correctly designating Arizona and Illinois, the different ways to control for marital status have no effect on the outcomes for the voter identification variables. All of the state voter identification variables are statistically insignificant-meaning that none of these requirements has a statistically measurable relationship with voter turnout.

Model 8 uses the minimum requirements for voter identification as used by the Eagleton Insti­tute. The only voter identification coefficient to be statistically significant is the swear affidavit coeffi­cient. The survey respondents in swear affidavit states are 0.002 percent less likely to report voting than respondents from states that only require vot­ers to state their name for identification.

As for the socioeconomic variables in Models 4 through 8, the findings are similar to the previous findings. African-Americans are more likely to have reported voting in the election than non-Hispanic whites, while Asians are less likely to report voting. Older respondents and those with higher incomes and more education are more likely to report voting. Widowed, divorced, and separated respondents are less likely to report voting than singles, while mar­ried respondents are more likely to report voting. Female respondents are more likely to report voting than male respondents. Respondents residing in bat­tleground states are more likely to vote, while respondents who moved within the last twelve months are less likely to have reported voting.

Findings by Race and Ethnicity

The impact of voter identification requirements on minority voters has received much media atten­tion recently.[79] To analyze the relationship between race and ethnicity and voter identification require­ments, Tables 6 through 9 present the findings of the probit analyses.

Non-Hispanic Whites. The probit regression results presented in Table 6 contain data for respondents reporting to be non-Hispanic whites. Models 9 and 10 present the findings for the max­imum requirements with Model 10 including the correct voter identification classifications for Ari­zona and Illinois. Except for the photo identifica­tion coefficient, none of the coefficients for the voter identification variables are statistically differ­ent from zero. In both Models 9 and 10, white respondents in photo identification states are less likely to have reported voting compared to white respondents in states that only required voters to say their names at the polling stations. Under both models, white survey respondents in photo identi­fication states are 0.002 percent less likely to report voting than white respondents from states that only required voters to state their name.

The analysis of minimum voter identification requirements in Model 11 finds that white respon­dents are less likely to vote when the minimum requirement entails a sworn affidavit. White survey respondents in swear affidavit states are 0.002 per­cent less likely to report voting than white respon­dents from states that only required voters to give their name.

Non-Hispanic African-Americans. The probit regression results presented in Table 7 contain data for respondents reporting to be non-Hispanic Afri­can-Americans. Models 12 and 13 present the find­ings for the maximum requirements with Model 13 including the correct voter identification classifica­tions for Arizona and Illinois. Except for the non-photo identification coefficient, none of the coeffi­cients for the voter identification variables are statis­tically different from zero. In both Models 12 and 13, African-American respondents in non-photo identification states are less likely to have reported voting compared to African-American respondents in states that only required voters to say their names at the polling stations. In Model 12, African-Ameri­can respondents in non-photo identification states are 0.019 percent less likely to report voting than African-American respondents from states that only required voters to state their name. For Model 13, the elasticity for non-photo identification states is 0.012 percent.

The analysis of minimum voter identification requirements in Model 14 fails to find any statisti­cally significant relationships between African- American voter turnout and the minimum voting requirements.

Hispanics. The probit regression results pre­sented in Table 8 contain data for respondents reporting to be Hispanic. Models 15 and 16 present the findings for the maximum requirements with Model 16 including the correct voter identification classifications for Arizona and Illinois. Model 17 presents the findings for the minimum voter iden­tification requirements. All three models find that Hispanics reported lower voter turnout rates in states with non-photo identification requirements compared to states that only require voters to state their names at the polling stations. All three of these findings are statistically significant at the 95 percent confidence level. Hispanic respondents in non-photo identification states are 0.035 percent to 0.049 percent less likely to report voting than His­panic respondents from states that only required voters to state their name.

Asian Americans. The probit regression results presented in Table 9 contain data for respondents reporting to be non-Hispanic Asian American (including Hawaiians/Pacific Islanders). Models 18 and 19 present the findings for the maximum requirements with Model 19 including the correct voter identification classifications for Arizona and Illinois. Model 20 presents the findings for the minimum voter identification requirements. All three models find that the various state voter iden­tification requirements do not have a statistically measurable relationship with voter turnout of Asian Americans.

DISCUSSION

The findings of this analysis suggest that voter identification requirements, such as requiring non-photo and photo identification, have virtually no suppressive effect on reported voter turnout.

Caution is needed in interpreting the Eagleton Institute's findings, for at least three reasons.

First, their study used one-tailed significance tests that can be used to double the chances of finding statistically significant findings.

Second, the voter identification laws for two states, Arizona and Illinois, were incorrectly classified. From our modeling, this misclassification leads to a negative and statistically significant relationship between photo identification requirements and voter turnout for all registered voters. When Arizona and Illinois are correctly classified, the relationship in our mod­eling is statistically indistinguishable from zero.

Third, the findings for photo identification re­quirements are sensitive to model specification. Us­ing the Eagleton Institute's state voter identification classifications and controlling for marriage with a married or not dichotomous variable, our analysis of overall voter turnout finds that photo identifica­tion requirements have a negative and statistically significant relationship with overall voter turnout. However, when additional marital status vari­ables-widowed, divorced, separated-are in­cluded, the statistically significant relationship for photo identification requirements disappears.

Controlling for factors that influence voter turn­out, states with stricter voter identification laws largely do not have the claimed negative impact on voter turnout when compared to states with more lenient voter identification laws. Based on the Eagleton Institute's findings, some members of the media have claimed that voter identification law suppress voter turnout, especially among minori­ties.[80] Their conclusion is unfounded. When statis­tically significant and negative relationships are found in our analysis, the effects are so small that the findings offer little policy significance.

More important, minority respondents in states that required photo identification are just as likely to report voting as are minority respon­dents from states that only required voters to say their name.

Nevertheless, using data from the November 2004 CPS to study the impact of voter identification requirements on voter turnout does have its limita­tions. The November 2004 CPS is a cross-sectional data set that does not allow social scientists to esti­mate the effect of changing voter identification requirements within states over time. Studies using the November CPS can only provide information on how voter patterns differed between states with dif­ferent voter identification requirements. These stud­ies cannot provide information on how enac

OVERVIEW

The 2000 presidential election sparked a fire­storm of debate relating to election reform in the United States. Since then, academics, the media, and elected officials have proffered opinions and implemented policies related to this important political issue. Topics that have been addressed in recent years range from modernizing voting machines and updating voter registration rolls to implementing stricter identification requirements for voting.

In 2002, Congress passed the Help America Vote Act (HAVA).[1] HAVA affects only federal elec­tions and, among other things, requires that the states provide for provisional voting; create a computerized, centralized list of registered voters; and ensure that new voters who register by mail present identification before being allowed to vote in person. HAVA established the Election Assis­tance Commission (EAC) to serve as "a national clearinghouse and resource for information and review of procedures with respect to the adminis­tration of federal elections."[2] Additionally, many state legislatures have enacted their own election reform legislation.[3]

Of the many election reforms currently being considered, one that has incited some of the most cantankerous debate is that of voter identification at the polls. For many, the idea of requiring voters to present identification in order to vote is anath­ema, tantamount to the poll taxes that were once used to prevent African-Americans from voting.[4] They contend that requiring identification at the polls will lead to lower voter turnout, especially among the poor, certain minorities, and the elderly. For others, such as the Protect Arizona Now orga­nization that lobbied in favor of identification requirements for Arizona voters, the problem of voter fraud makes voter identification require­ments a common-sense solution.[5] The standard argument goes that if a person has to show identi­fication to board a plane or cash a check, why shouldn't he have to do the same in order to vote? Additionally, the proponents of stricter voter iden­tification requirements argue that such a policy would bolster the public's faith in the legitimacy of elections and lead to greater voter turnout, not less.

Both sides raise valid concerns. However, even a cursory glance at the literature on voter identifica­tion requirements shows that there is a dearth of empirical research on this issue. While there have been a few studies to address the effect of voter identification requirements using election data,[6] more research is needed in order to appropriately assess the legitimacy of either side's claims.

In response to this debate, the EAC awarded a grant to Rutgers University's Eagleton Institute of Politics and the Moritz College of Law at Ohio State University to study voter identification require­ment laws. The resulting study, Report to the U.S. Election Assistance Commission on Best Practices to Improve Voter Identification Requirements Pursuant to the Help America Vote Act of 2002,[7] included a sta­tistical analysis of the effect of voter identification requirements on voter turnout during the 2004 election by Professor Timothy Vercellotti of the Eagleton Institute.[8] A new version of the analysis with Timothy Vercellotti and David Anderson as authors was presented to the 2006 American Polit­ical Science Association conference.[9] Hereinafter, this study will be referred to as the "Eagleton Insti­tute study."

The Eagleton Institute study found that more stringent voter identification requirements appeared to reduce voter turnout in 2004.[10] In the media, their study has been cited as demonstrating that the strengthening of voter identification requirements to reduce fraud has the side effect of suppressing minority voter turnout.[11]

This Center for Data Analysis report attempts to replicate the part of the Eagleton Institute study that used the publicly available November 2004 Current Population Survey (CPS).[12] This analysis was done because several aspects of the Eagleton Institute study cast doubt on the validity of its findings:

• The Eagleton Institute used one-tailed hypothe­sis tests instead of the more commonly accepted two-tailed tests. The one-tailed test allows researchers to double their chances of finding statistically significant results.
• The 2004 voter identification laws of certain states were misclassified. For example, Arizona and Illinois were incorrectly classified as requir­ing voters to provide identification and state their name for authentication, respectively. However, in 2004 Arizona only required voters at polling stations to sign their name for authen­tication, while Illinois required poll workers to match the signatures of voters.
• Some of the variables used to predict the deci­sion to vote were used inappropriately. For example, the Eagleton Institute study used the November 2004 CPS family income variable, which is an ordinal variable of unequal income ranges, as an interval-ratio variable. Using cate­gorical variables as interval-ratio variables can lead to estimation problems.

After addressing these issues, our reanalysis finds that some of the original findings of the Eagleton Institute study are unfounded. Controlling for fac­tors that influence voter turnout, voter identifica­tion laws largely do not have the negative impact on voter turnout that the Eagleton Institute suggests. When statistically significant and negative relation­ships are found, the effects are so small that the findings offer little policy significance. For example, our analysis indicates that:

• White survey respondents in photo identifica­tion states are 0.002 percent less likely to report voting than white respondents from states that only required voters to state their name.
• African-American respondents in non-photo identification states are 0.012 percent less likely to report voting than African-American respon­dents from states that only required voters to state their name.

In other cases, no effect was found.

• In general, respondents in photo identification and non-photo identification states are just as likely to report voting compared to respondents from states that only required voters to state their name.
• African-American respondents in photo identi­fication states are just as likely to report voting compared to African-American respondents from states that only required voters to state their name.
• Hispanic respondents in photo identification states are just as likely to report voting compared to Hispanic respondents from states that only required voters to state their name.

BACKGROUND

When discussing voting behavior, it is important to consider the factors that influence whether an individual votes or not. According to the "Calculus of Voting" model, an individual will vote when the rewards from voting are positive and will abstain when they are not. The equation for the Calculus of Voting model is as follows:

R = PB - C + D.

The rewards (R) from voting are determined by multiplying the benefits (B) an individual receives when his preferred candidate wins over a less pre­ferred candidate by the probability (P) that his vote will make a difference plus the benefits one receives from voting as an act of fulfilling one's duty or civic obligation (D) minus the costs of voting (C).[13] This is the standard, rational model of voting and will be used to inform the following discussion of voter identification requirements and their effect on voter turnout.

The voter identification issue is often framed as being torn between the opposing aims of "access and integrity."[14] By this we mean that it is com­monly perceived that while voter identification laws may be effective at preventing ineligible individuals from voting (integrity), they may have an adverse effect on the ability of every eligible voter to vote (access). There have been only a few empirical stud­ies on the impact of voter identification require­ments,[15] but this does not translate into a lack of opinions on this topic.

Advocates for more stringent voter identification laws contend that this reform is vital to prevent voter fraud.[16] As more and more elections are won by slim margins, proponents of identification requirements argue that the chances are greater that voter fraud could affect election outcomes.[17] The potential for a small number of voters to have a sig­nificant impact on the outcome of an election became all too evident in the 2000 presidential elec­tion. Given that George W. Bush was declared the winner in Florida (and the next President) by a mar­gin of 537 votes, it follows that even a small number of fraudulent votes (537+1) would matter a great deal.[18] In 2004, there were allegations of voter fraud in the Washington gubernatorial election in which Christine Gregoire won by a margin of 129 votes.[19] Certainly the potential of voter fraud is a matter of concern.

Broadly defined, voter fraud is "the intentional corruption of the electoral process by voters."[20] While voter fraud manifests itself in different forms, examples include individuals who vote but are ineligible (such as non-citizens and felons), individuals who vote multiple times in various pre­cincts, and individuals who vote using someone else's name. Because of the lack of research and the difficulty of collecting data on voter fraud, the extent to which these kinds of voter fraud occur is unknown. Additionally, for similar reasons, we are unaware of the extent to which voter identification laws would curb the type of voter fraud they are intended to prevent.

However, there are some examples of recorded voter fraud. The Department of Justice asserts that since the inception of the Attorney General's Ballot Access and Voting Integrity Initiative in 2002, 120 people have been charged with election fraud, of which 86 have been convicted.[21] Additionally, the Milwaukee Journal Sentinel reports that prosecutors in Milwaukee filed charges against 14 individuals for voter fraud in the 2004 election.[22] Of the 14, 10 were felons accused of voting and four were accused of double voting. Prosecutors obtained five convic­tions. For proponents of strict voter identification requirements, the knowledge that any voter fraud occurs is sufficient to argue that more needs to be done to curb this problem.[23]

The most prevalent critique of the voter fraud argument is that "voter-fraud anecdotes are often misleading, incomplete, and unrepresentative."[24] Proponents of this view contend that upon closer examination of claims of voter fraud, such charges turn out to be either nonexistent or infrequent. For instance, the Brennan Center for Justice at the New York University School of Law found that in 2004, voter fraud occurred 0.0009 percent of the time in the gubernatorial election in Washington and 0.00004 percent of the time in Ohio. They report that these percentages are akin to the likelihood of an American's being killed by lightning.[25]

Opponents of voter identification requirements also argue that the few instances of voter fraud that may be prevented by identification laws do not out­weigh the thousands of legitimate voters who would be disenfranchised because they lacked the neces­sary identification.[26] These critics argue that identi­fication laws will have a negative impact on the ability of certain minorities, the elderly, the dis­abled, and the poor to vote.[27] It is presumed, and some studies have found, that people from these groups are less likely to possess drivers' licenses or other government-issued identification.[28] It is also assumed that many from these groups would be unable or unwilling acquire the necessary docu­mentation. Critics of strict identification laws fur­ther argue that the costs (in both time and money) of obtaining such documentation would be a deter­rent to voting and would likely result in lower voter turnout among poor voters and those who do not have easy access to government offices.[29] It is for this reason that "ID requirements are compared to modern poll taxes."[30]

While it is difficult to accurately assess the num­ber of eligible voters who would be rendered unable to vote because they lack proper identification, some studies have attempted to estimate such fig­ures by looking at the percentage of the population who do not have driver's licenses. For instance, a Wisconsin study found that when considering the entire state, 80 percent of men and 81 percent of women had valid driver's licenses. In contrast, only 45 percent of African-American men and 51 per­cent of African-American women had valid driver's licenses. The percentages for Latinos were also lower (54 percent for men and 41 percent for women).[31] Similarly, a Georgia study found that among registered voters, non-whites, women, and the elderly were less likely to have government-issued photo identification (either a driver's license or state identification).[32]

Although these figures shed light on the types of people who are less likely to have driver's licenses, it is unadvisable to focus on this statistic alone. First, the data still cannot tell us whether those individu­als without driver's licenses have some other form of identification, such as an employee ID, student ID, social security card, or any other form of identifica­tion currently accepted in many states. Second, it cannot tell us about future behavior. Do voters in photo identification states who lack the necessary identification obtain the required identification (such as a driver's license) when the state law is changed? Take for instance the previous study con­ducted in Wisconsin, which currently does not require identification before voting (except for those requirements set forth in HAVA for new voters). Although approximately half of African-Americans in the state are currently without driver's licenses, we do not know if those individuals will get driver's licenses or state IDs if Wisconsin were to require voters to show identification before voting.

For these reasons, proponents of voter identifica­tion requirements are convinced that requiring identification at the polls would not be an excessive burden to voters. As previously mentioned, identi­fication is required for many things that are consid­erably less important than voting (flying in a plane, buying alcohol, etc.). As "voting is equally impor­tant," if not more important, the argument goes that it makes sense for someone to be required to show identification in order to cast a ballot.[33] Addition­ally, Senior Research Scientist John Lott at the Uni­versity of Maryland Foundation points out that as "almost 100 countries require photo identifications to vote," the United States would be hardly alone in requiring voters to show some form of identification at the polls.[34]

Those who oppose voter identification at the polls argue that other reforms are better suited to preventing voter fraud. For instance, critics of voter identification point to absentee ballots as "the Achilles heel of election security" because voters are often not required to show identification at all.[35] Yet absentee ballots have been largely left out of the voter identification requirement debate. This apparent discrepancy has been used by opponents of voter identification laws as evidence that sup­porters of such legislation are not interested in real voter fraud reform.[36] Rather, critics argue that voter identification supporters are using such laws as an attempt to suppress voter turnout by increas­ing the costs of voting (the "C" from the Calculus of Voting model).[37]

Another argument proffered by supporters of voter identification requirements is that such laws are necessary to maintain the public's faith in the integrity of elections. The Commission on Federal Election Reform (Carter-Baker Commission) at American University asserts that "the electoral sys­tem cannot inspire public confidence if no safe­guards exist to deter or detect fraud or to confirm the identity of voters."[38] This argument, "the ensur­ing integrity hypothesis," contends that public faith in the honesty of elections actually "encourages additional voter participation."[39] Proponents argue that voter identification laws will bolster the public's faith in the outcome of elections. This will increase, not decrease, turnout because voters will feel a greater pride in voting (increasing the "D" or duty component of voting).

Voter identification laws are exceptionally popular among the general public. In a survey of some 36,000 voters, Professors Stephen Ansolabehere and Elting R. Morison of the Massachusetts Institute of Technology found that 77 percent of respondents supported voter identification requirements.[40] For the most part, the majority of respondents sup­ported such laws regardless of race, location (North­east, Midwest, etc.), and political ideology. While those who identified themselves as conservatives had the highest percentage of agreement with iden­tification requirements (at 95 percent), even those who identified themselves as "very liberal" had 50 percent agreement with voter identification laws.[41] Regarding race, more than 70 percent of whites, African-Americans, and Hispanics supported voter identification laws.[42] Additionally, Ansolabehere found only 23 instances out of 36,000 where an individual reported being unable to vote because he lacked the necessary identification.[43]

These survey data are supported by actual voter behavior. In 2004, when Arizonans voted on Prop­osition 200, which would require voter identifica­tion at the polls as evidence of citizenship, it passed with 56 percent of the vote.[44]

Ultimately, it is not the intent of this paper to debate the merits of either side's arguments. Rather, we want to present the major arguments on either side of this issue as background to our analysis. However, the paper does intend to examine more closely one of the claims of this debate: that stricter voter identification requirements depress voter turnout. In order to do that, it is necessary to discuss the different voter identification re­quirements across the 50 states and the District of Columbia.

Voter identification requirements, if any, differ by state, so there is great variability in the way voters from different parts of the country are required to verify their identity before casting a ballot. Some states rely on the honor system where voters merely have to give their names to the election official.[45] Other states only require a signature,[46] with some states going a step further and actually matching the signature to a previously signed document.[47] States with more stringent requirements ask that voters provide identification[48] or photo identification.[49]

The Eagleton Institute study identified two cate­gories of identification requirements (maximum requested and minimum required) and five types of identification requirements (stating name, signing name, signature match, present ID, and photo ID).[50] It is important to note that in 2004, there were no states that had photo ID as a minimum requirement. All states that had a photo ID require­ment permitted voters who did not have such doc­umentation to present alternative forms of ID or sign an affidavit attesting to their identity.[51]

By the maximum requested, the Eagleton Insti­tute study refers to the most identification that an individual can be asked to present in order to vote using a regular ballot. Conversely, the minimum is the least identification that will be accepted to vote.[52] For example, when voting in Louisiana in 2004, a voter would be asked by poll workers to present photo identification. If the individual was unable to present an acceptable form of ID, he was allowed to vote after signing an affidavit stating he is the person he claims to be.[53] In that case, photo ID would be the maximum requested, and affidavit would be the minimum required.

Within the states that require some form of doc­umentation as proof of identity, there are also signif­icant differences. For instance, some states, like Massachusetts, "may" ask that a voter show identifi­cation, but identification is not automatically requested of all voters.[54] In Alabama and Alaska, two states that request identification, this require­ment can be waived if a poll worker knows the voter and can attest to his identity.[55] This is an important issue to consider because it means that different vot­ers within the same state may be affected by differ­ent identification requirements.

Furthermore, by the 2004 election, many states had become compliant with certain provisions in the Help America Vote Act (HAVA) which required identification at the polls from first-time voters who registered by mail and who did not show iden­tification at the time of registration. One state, Pennsylvania, actually went above and beyond HAVA requirements and mandated that all first-time voters needed to show identification at the polls regardless of whether they showed identifica­tion when they registered to vote.[56] Because of HAVA, many first-time voters had to show identifi­cation at the polls even in states that did not other­wise require identification from all voters.

Even among states that require documentation, there is great variability in the types of documenta­tion that is accepted. Some accept only a govern­ment-issued photo identification, while others accept almost any document that demonstrates a person's identity. For example, in 2004, acceptable documentation in Florida ranged from a driver's license and passport to credit card and buyer's club card to utility bill, bank statement, or paycheck (as long as they contained the name and address of the individual).[57] In contrast, some states that required identification to vote are much more restrictive with respect to acceptable forms of identification. One such state, Virginia, only allowed voters to present a voter registration card, Social Security card, employer-issued identification card (as long as it contained a photo), Virginia driver's license, or other Commonwealth or government-issued identi­fication.[58] Furthermore, in many states, individuals who are unable to provide the appropriate docu­mentation are given an alternative, such as signing an affidavit, in order to vote. Finally, Section 302 of HAVA requires that an individual who fails to meet the identification requirements of voting can still vote using a provisional ballot.[59]

The key aspects of this brief overview of identifi­cation requirements of voting is that there is a lot of variability by states as to what is required, and not all identification requirements are created equal. By that we mean that required identification documen­tation for one state may not meet the identity requirements in another state. This is just one of the reasons that it is particularly difficult to study the effect of such laws on voter turnout.

THE DATA

In order to analyze individual voter turnout, this study uses data from the U.S. Census Bureau's Cur­rent Population Survey, November 2004: Voting and Registration Supplement File.[60] The November 2004 CPS voting supplement contains interviews from about 57,000 households. Based on self-described registered voters, the data allow us to model the decision to vote based on individual and household characteristics.

Dependent Variable. The dependent variable is whether or not the respondent reported that he or she voted in the November 2004 election. Respondents who admitted to not being regis­tered voters were omitted, along with those reporting that they were not United States citi­zens. We also omitted those reported to be voting through absentee ballots.[61]

According to the U.S. Census Bureau's analysis of the November 2004 CPS data, 89 percent of regis­tered voters voted in the November 2004 elec­tion.[62] This estimate is drawn from a sample of respondents reporting to be registered voters and is much higher than estimates based on samples of the voting-age population. However, the EAC estimates that 70.4 percent of registered voters turned out to vote.[63] The CPS estimate of 89 percent may be biased upward because it is based on the reported vote, which may be overstated because survey respondents may be disinclined to admit that they did not vote.[64] When turnout is based on the total population over 18 years old, 55.8 percent of per­sons over age 18 voted.[65]

Voter Identification Requirements. The voter identification requirements included in the analysis capture the degree to which a registered voter has to prove his or her identity at the polling station. Two sets of five dichotomous voter identification vari­ables are used in the analysis. The first set is based on the maximum amount of identification that the voter is required to produce in order to prove his or her identity. The maximum state voter identification requirements are broken down into the following classification: state name, sign name, match signa­ture, provide non-photo identification, and provide photo identification. Table 1 presents the voter identification classifications by state used by the Eagleton Institute and the Moritz College of Law at Ohio State University.

For all but two of the states, Illinois and Arizona, we used the classifications that were provided to us by the Eagleton Institute. We recoded these two states because upon researching state election laws, we discovered that the Eagleton Institute had erro­neously reported the identification requirements for these two states. The Eagleton Institute study has Illinois listed as a "state name" state. In actuality, Illi­nois poll workers match a prospective voter's signa­ture to a signature already on file, making Illinois a "match signature" state.[66]

The Eagleton Institute has Arizona listed as a "provide ID" state although Arizona was a "sign name" state at the time of the 2004 election.[67] Iden­tification laws did not go into effect in Arizona until some time after the 2004 election. Arizona could not have been a "provide ID" state before the November 2004 election because Arizonans voted on and approved Proposition 200 on the November 2004 ballot. This initiative is the impetus for the requirement that voters show identification before voting as proof of citizenship.[68]

The second set of voter identification variables recognizes that some states allow voters without proper identification to vote after demonstrating their identity through other means. This minimum requirement set of variables includes state name, sign name, match signature, provide non-photo identification, and swear affidavit. For the probit regressions, the variable for voters stating their names for identification is omitted for reference purposes.

Individual Factors. The individual factors included in the analysis capture differences in the race and ethnicity, age, education, household income, marital status, gender, employment status, citizenship, residential mobility, and home owner­ship of the individual respondents. Controlling for such variables as education and age is important because research indicates that these variables are good predictors of voting turnout.[69] The analysis controls for the effect of the individual's race and ethnicity through a set of mutually exclusive dichotomous variables for the following categories: non-Hispanic white, non-Hispanic African-Ameri­can, Hispanic, non-Hispanic American Indians, non-Hispanic Asians (including Hawaiians/Pacific Islanders), and other races, including those report­ing multiple races and ethnicities. The specification of these variables allows us to compare the voting patterns of minorities to those of whites.

A set of dichotomous variables control for the age of the individual respondents that fall into the fol­lowing categories: 18- to 24-year-olds, 25- to 44-year-olds, 45- to 64-year-olds, and 65 years and older. For education, the respondents were classified as either having less than a high school diploma, high school diploma or equivalent, some college, bachelor's degree, or a graduate school degree.

For family income, the Eagleton Institute study used an ordinal family income variable as an in­terval-ratio variable.[70] The family income variable is coded as 1 through 16 with units containing un­equal income ranges. For the purposes of this analy­sis, the effect of family income is controlled for by the inclusion of a series of income range dichoto­mous variables: under $15,000, $15,000 to $29,999, $30,000 to $49,999, $50,000 to $74,999, $75,000 to $149,999, and $150,000 or more.

To control for the influence of marital status, five dichotomous variables signifying being single, mar­ried, separated, divorced, and widowed are included in the model. Single individuals are the default. A dichotomous variable identifying the gender of the individual as a female is also included in the models.

Two dichotomous variables are included to con­trol for the effect of employment. The first is a dichotomous variable signifying whether or not the individual is employed; the second is a dichoto­mous variable for whether or not the person is in the labor force.

To control for whether native-born citizens are more likely to vote than naturalized citizens, a dichotomous variable identifying native-born citi­zens is included. Two dichotomous variables are included to control for community ties. The models control for whether or not the individual has moved within the last year and whether or not the individ­ual owns or rents his or her home. These two vari­ables are included to help control for social connectedness under the theory that those with stronger community ties will be more likely to vote.

State Political Factors. As with the Eagleton Institute study, two dichotomous variables indicate whether a state is considered a battleground state and a competitive state. A state is designated as a battleground state if the margin of victory for the winning 2004 presidential candidate was 5 percent or less. A state was designated as competitive if the margin of victory for governor and/or U.S. Senate races was 5 percent or less.

FINDINGS

The probit regression analyses that follow exam­ine the effects of voter identification requirements on voter turnout. Table 2 presents the original find­ings of the Eagleton Institute's probit regression analysis. Table 3 presents the descriptive statistics of the data used in Table 4. Based on our analyses, six sets of probit regression models are presented in Tables 4 to 9.

The first set of probit regressions contains our replication of the Eagleton Institute study for their analysis of all voters (Table 4). The second set of probit regressions presents the findings for all voters under a different model specification and the cor­rected classification of state identification require­ments for Arizona and Illinois (Table 5). The sixth through ninth sets of probit regressions present our findings for the different model specification and corrected coding for state identification require­ments for whites, African-Americans, Hispanics, and Asians (Tables 6 through 9).

For all of the models, robust standard errors are estimated to correct for correlated error terms within each state. For tests of statistical significance, the standard two-tailed tests are used. See below for a discussion of one-tailed versus two-tailed tests of statistical significance. The calculations in Tables 3 through 9 use the CPS weight, PWSSWGT, as rec­ommended by the Bureau of the Census.

---

One-Tailed Versus Two-Tailed Tests of Statistical Significance

When doing tests of statistical significance for hypotheses, social scientists generally use two-tailed tests. Two-tailed tests are used to check for a difference while ignoring in which direction the difference lies.

For example, a social scientist would use a two-tailed test to determine whether voters in photo identification and give name states have different probabilities of reporting having voted in the 2004 election, regardless of the direction of the relation­ship. By using a two-tailed test, the 5 percent prob­ability is split between both ends of the bell-shaped curve. (See Figure A in Chart 1.) That is, 2.5 percent of the probability that the difference is due to chance is placed in the side that represents respon­dents in photo identification states being less likely to vote, while 2.5 percent is placed in the side that represents respondents in photo identification states being more likely to vote. If the probit coeffi­cient for photo identification states falls within either of the 2.5 percent shaded regions, this finding is determined to be statistically significant. If the coefficient falls within the left (right) tail, photo identification requirements have a negative (posi­tive) relationship with reported voter turnout. If the coefficient falls between the 2.5 percent shaded regions, photo identification requirements are said have no relationship with voter turnout.

When one-tailed tests are used, social scientists are hypothesizing that the relationship between photo identification requirements and reported voting has a specific direction: for example, voter identification requirements decrease (increase) re­ported voting. As determined by the social scien­tist, all of the 5 percent of chance is placed in one end of the bell-shaped curve. If the direction of the relationship is as hypothesized, placing the entire 5 percent chance in one side makes it is twice as easy to achieve a statistically significant finding with a one-tailed test as with a two-tailed test. Figure B in Chart 1 is an example of a one-tailed test where the researcher believes a negative relationship exists. In the case of photo identification requirements and voter turnout, if the coefficient falls within the 5 percent shaded region of the left tail, photo identi­fication requirements would then be said to have a negative relationship. If the coefficient does not fall within the 5 percent region, then photo identifica­tion requirements are said to have no relationship with voter turnout.

According to norms of the social sciences, researchers generally use two-tailed tests. When they deviate from this norm, social scientists gen­erally provide a justification for why they have done so. Consumers of statistical research should be skeptical of findings based on one-tailed tests, especiallywhen such findings do not hold up under two-tailed testing.

---

Replicating the Eagleton Institute's Findings for All Voters

Table 2 contains the findings from the Eagleton Institute's probit regression for all registered voters as presented in their paper. Table 3 presents the find­ings from our attempt to replicate the Eagleton Insti­tute study findings for all voters. In our attempt at replicating the Eagleton Institute's study, we could not entirely match the same number of respondents. The Eagleton Institute's probit regression of all voters is based on 54,973 respondents.[71] Our best attempt at replicating their analysis produced 54,829 respondents-144 fewer respondents. In addition, the results reported in Table 3 use the more com­monly accepted two-tailed significance tests.

While the Eagleton Institute reported that states with sign name, non-photo identification, and photo identification requirements have lower voter turnout than states with only the state name requirement, only the photo identification coeffi­cient in our attempt at replication (Model 1) is sta­tistically significant at the 95 percent confidence level. Respondents from photo identification states are less likely to have reported voting compared to respondents in states that only required voters to say their names at the polling stations. The magni­tude of the negative relationship between photo identification requirements and voter turnout is dif­ficult to interpret with probit coefficients, so the elasticity was calculated. The elasticity figures used in this analysis represent the percentage change in the probability of reporting to vote given a one-unit change in a particular dichotomous independent variable. The survey respondents in photo identifi­cation states are 0.002 percent less likely to report voting than respondents from states that only required voters to give their name for identification.

Model 2 corrects for the Eagleton Institute study's misclassification of the voter identification require­ments in Arizona and Illinois. With the correction, all of the state voter identification variables are sta­tistically insignificant-meaning that none of these requirements has a statistically measurable relation­ship with voting turnout.

Model 3 attempts to replicate the findings of the Eagleton Institute's examination of the effect of minimum requirements. As seen in Table 2, the Eagleton Institute found that the coefficients for sign name, non-photo identification, and swear affidavit states had statistically significant, negative relationships with voter turnout using one-tailed significant tests. However, our analysis pre­sented in Model 3 using two-tailed statistical significance tests finds only the swear affidavit coefficient to be statistically significant at the 95 per­cent confidence level. The survey respondents in swear affidavit states are 0.002 percent less likely to report voting than respondents from states that only required voters to state their name for identification.

It should be noted that although we ran the minimum identification requirement model using the classifi­cations assigned to the states by the Eagleton Institute study, there are some issues with the states considered to have an affidavit as the minimum requirement. These issues should be addressed in follow-up studies. First, the Eagleton Institute study identified only four states as having a minimum requirement of sign affidavit. They are Florida, Indiana, Louisiana, and North Dakota. All but one of these states, Indiana, require some form of identification as the maximum requested. This puts Indiana in the precarious position of requiring, at a maximum, that a voter sign his name before receiving a ballot; if he is unable to do so, he can sign an affidavit and vote. This does not make sense, because Indiana in 2004 did not require identification before voting (other than for those affected by HAVA requirements).

We believe this to be another classification error on the part of the Eagleton Institute. According to the "2004 Indiana Election Day Handbook," the procedure for signing an affidavit only applies to challenged voters who are then given a provi­sional ballot if they sign the affidavit.[72] This vot­ing method would not fall under the guidelines set forth by the Eagleton Institute because it applies to provisional, and not regular, ballots.[73] For these reasons, we believe Indiana should have a minimum identification requirement of sign name, the same as its maximum.

Additionally, there are five other states (Con­necticut,[74] Delaware,[75] Georgia,[76] South Dakota,[77] and Virginia[78]) that require some form of identifi­cation but make exceptions and allow voters with­out the required documentation to sign an affidavit in order to vote. To be classified correctly, these states should also be considered to have a mini­mum requirement of sign affidavit as they too pro­vide opt outs for voters unable to show appropriate forms of identification.

As for the socioeconomic variables in Models 1 through 3, African-Americans are more likely to have reported voting in the election than a group­ing of non-Hispanic whites, American Indians, Hawaiians/Pacific Islanders, and others. In con­trast, Asians are less likely to report voting. Respon­dents aged 45 and above are more likely to report voting than those 18 to 24 years old. Those with an education at or above a high school diploma are more likely to report voting than those without a high school degree. Family income has a positive relationship with the proba­bility of reporting having voted. Mar­ried and female respondents are more likely to report voting than not married and male respondents, respectively. Respondents residing in battleground states are more likely to vote, while respondents who moved within the last six months are less likely to report voting.

Alternative Model Specifications

Concerns regarding some of the variables used in the Eagleton Institute study led us to estimate alternative specifications that use the November 2004 CPS data more appropriately.

First, the Eagleton Institute's race and ethnicity dichotomous variables compare African-Americans, Hispan­ics, and Asians to the default group of whites, American Indians, Alaskan Natives, Hawaiians/Pacific Islanders, and those reporting to be more than one race and/or ethnicity. For exam­ple, the Eagleton Institute found that African-Americans were more likely to report voting compared to whites, American Indians, Alaskan Natives, Hawaiians/Pacific Islanders, and those reporting to be more than one race and/or ethnicity.

The descriptive statistics of the data used for the alternative specifi­cations are presented in Table 4. The analyses in Table 5 control for the effect of the individual's race and ethnicity through a set of mutually exclusive dichotomous variables for the following categories: non-His­panic whites, non-Hispanic African-Americans, Hispanics, non-Hispanic American Indians and Alaskan Natives, non-Hispanic Asians (including Hawaiians/Pacific Islanders), and other races, including those reporting multiple races and eth­nicities. For example, this division of race and eth­nic groups allows us to present clearer estimates of how voter identification laws affect the voting probabilities of minorities compared to whites.

Second, the Eagleton Institute study used an ordi­nal family income variable as an interval-ratio vari­able. Using categorical variables as interval-ratio variables can lead to estimation problems, so for the purposes of this analysis, the effect of family income is controlled for by the inclusion of a series of income range dichotomous variables.

Third, the effect of photo identification variables on voter turnout is very sensitive to how the models control for marriage. In addition to a dichotomous variable for whether or not the respondent reported being married, additional dichotomous variables were added for those reporting to be widowed, sep­arated, and divorced. This minor change in marital control variables has a significant impact on the results for the relationship between voter turnout and some of the voter identification variables.

Fourth, the alternative models control for whether or not the individual has moved within the last year instead of the six-month time period used by the Eagleton Institute.

Fifth, a variable indicating whether or not the respondent owns or rents his or her home was added to the alternative models. The residential mobility and home ownership variables help to control for how connected the respondents are to their communities.

Table 5 presents the findings of the alternative model specification for all respondents. Model 4 contains the revised race/ethnicity and income vari­ables along with the variables for residential mobil­ity and home ownership. Of the four voter identification variables, only the photo identifica­tion variable is statistically significant. Photo identi­fication states have respondents that are less likely to have reported voting compared to respondents in states that only required voters to say their names at the polling stations. However, the difference is very small. The survey respondents in photo identifica­tion states are 0.002 percent less likely to report vot­ing than respondents from states that only required voters to state their name for identification.

A slight change in how marital status is con­trolled for in Model 5 makes the findings in Model 4 for photo identification requirements disappear. The inclusion of dichotomous variables to identify respondents if they are widowed, divorced, or separated, in addition to being married, signifi­cantly changes the results for the photo identifica­tion variable. A photo identification requirement no longer has a statistically significant relation­ship with voter turnout. Thus, the finding that photo identification requirements reduce voter turnout in Model 4 is not robust to an alternative model specification.

In Models 6 and 7, Arizona and Illinois are reclassified correctly as requiring voters at polling stations to sign their name and match signatures, respectively. As with Model 4, Model 6 uses only a married dichotomous variable to control for mar­ital status. Model 7 includes additional marital status variables as used in Model 5. After correctly designating Arizona and Illinois, the different ways to control for marital status have no effect on the outcomes for the voter identification variables. All of the state voter identification variables are statistically insignificant-meaning that none of these requirements has a statistically measurable relationship with voter turnout.

Model 8 uses the minimum requirements for voter identification as used by the Eagleton Insti­tute. The only voter identification coefficient to be statistically significant is the swear affidavit coeffi­cient. The survey respondents in swear affidavit states are 0.002 percent less likely to report voting than respondents from states that only require vot­ers to state their name for identification.

As for the socioeconomic variables in Models 4 through 8, the findings are similar to the previous findings. African-Americans are more likely to have reported voting in the election than non-Hispanic whites, while Asians are less likely to report voting. Older respondents and those with higher incomes and more education are more likely to report voting. Widowed, divorced, and separated respondents are less likely to report voting than singles, while mar­ried respondents are more likely to report voting. Female respondents are more likely to report voting than male respondents. Respondents residing in bat­tleground states are more likely to vote, while respondents who moved within the last twelve months are less likely to have reported voting.

Findings by Race and Ethnicity

The impact of voter identification requirements on minority voters has received much media atten­tion recently.[79] To analyze the relationship between race and ethnicity and voter identification require­ments, Tables 6 through 9 present the findings of the probit analyses.

Non-Hispanic Whites. The probit regression results presented in Table 6 contain data for respondents reporting to be non-Hispanic whites. Models 9 and 10 present the findings for the max­imum requirements with Model 10 including the correct voter identification classifications for Ari­zona and Illinois. Except for the photo identifica­tion coefficient, none of the coefficients for the voter identification variables are statistically differ­ent from zero. In both Models 9 and 10, white respondents in photo identification states are less likely to have reported voting compared to white respondents in states that only required voters to say their names at the polling stations. Under both models, white survey respondents in photo identi­fication states are 0.002 percent less likely to report voting than white respondents from states that only required voters to state their name.

The analysis of minimum voter identification requirements in Model 11 finds that white respon­dents are less likely to vote when the minimum requirement entails a sworn affidavit. White survey respondents in swear affidavit states are 0.002 per­cent less likely to report voting than white respon­dents from states that only required voters to give their name.

Non-Hispanic African-Americans. The probit regression results presented in Table 7 contain data for respondents reporting to be non-Hispanic Afri­can-Americans. Models 12 and 13 present the find­ings for the maximum requirements with Model 13 including the correct voter identification classifica­tions for Arizona and Illinois. Except for the non-photo identification coefficient, none of the coeffi­cients for the voter identification variables are statis­tically different from zero. In both Models 12 and 13, African-American respondents in non-photo identification states are less likely to have reported voting compared to African-American respondents in states that only required voters to say their names at the polling stations. In Model 12, African-Ameri­can respondents in non-photo identification states are 0.019 percent less likely to report voting than African-American respondents from states that only required voters to state their name. For Model 13, the elasticity for non-photo identification states is 0.012 percent.

The analysis of minimum voter identification requirements in Model 14 fails to find any statisti­cally significant relationships between African- American voter turnout and the minimum voting requirements.

Hispanics. The probit regression results pre­sented in Table 8 contain data for respondents reporting to be Hispanic. Models 15 and 16 present the findings for the maximum requirements with Model 16 including the correct voter identification classifications for Arizona and Illinois. Model 17 presents the findings for the minimum voter iden­tification requirements. All three models find that Hispanics reported lower voter turnout rates in states with non-photo identification requirements compared to states that only require voters to state their names at the polling stations. All three of these findings are statistically significant at the 95 percent confidence level. Hispanic respondents in non-photo identification states are 0.035 percent to 0.049 percent less likely to report voting than His­panic respondents from states that only required voters to state their name.

Asian Americans. The probit regression results presented in Table 9 contain data for respondents reporting to be non-Hispanic Asian American (including Hawaiians/Pacific Islanders). Models 18 and 19 present the findings for the maximum requirements with Model 19 including the correct voter identification classifications for Arizona and Illinois. Model 20 presents the findings for the minimum voter identification requirements. All three models find that the various state voter iden­tification requirements do not have a statistically measurable relationship with voter turnout of Asian Americans.

DISCUSSION

The findings of this analysis suggest that voter identification requirements, such as requiring non-photo and photo identification, have virtually no suppressive effect on reported voter turnout.

Caution is needed in interpreting the Eagleton Institute's findings, for at least three reasons.

First, their study used one-tailed significance tests that can be used to double the chances of finding statistically significant findings.

Second, the voter identification laws for two states, Arizona and Illinois, were incorrectly classified. From our modeling, this misclassification leads to a negative and statistically significant relationship between photo identification requirements and voter turnout for all registered voters. When Arizona and Illinois are correctly classified, the relationship in our mod­eling is statistically indistinguishable from zero.

Third, the findings for photo identification re­quirements are sensitive to model specification. Us­ing the Eagleton Institute's state voter identification classifications and controlling for marriage with a married or not dichotomous variable, our analysis of overall voter turnout finds that photo identifica­tion requirements have a negative and statistically significant relationship with overall voter turnout. However, when additional marital status vari­ables-widowed, divorced, separated-are in­cluded, the statistically significant relationship for photo identification requirements disappears.

Controlling for factors that influence voter turn­out, states with stricter voter identification laws largely do not have the claimed negative impact on voter turnout when compared to states with more lenient voter identification laws. Based on the Eagleton Institute's findings, some members of the media have claimed that voter identification law suppress voter turnout, especially among minori­ties.[80] Their conclusion is unfounded. When statis­tically significant and negative relationships are found in our analysis, the effects are so small that the findings offer little policy significance.

More important, minority respondents in states that required photo identification are just as likely to report voting as are minority respon­dents from states that only required voters to say their name.

Nevertheless, using data from the November 2004 CPS to study the impact of voter identification requirements on voter turnout does have its limita­tions. The November 2004 CPS is a cross-sectional data set that does not allow social scientists to esti­mate the effect of changing voter identification requirements

CONCLUSION

Controlling for factors that influence voter turn­out, voter identification laws largely do not have the claimed negative impact on voter turnout based on state-to-state comparisons. When statistically signif­icant and negative relationships are found, the effects are so small that the findings offer little policy significance. White survey respondents in photo identification states are 0.002 percent less likely to report voting than white respondents from states that only required voters to state their name. Afri­can-American respondents in non-photo identifica­tion states are 0.012 percent less likely to report voting than African-American respondents from states that only required voters to state their name.

In other cases, no effect was found. In general, respondents in photo identification and non-photo identification states are just as likely to report voting compared to respondents from states that only required voters to state their name. African-Ameri­can respondents in photo identification states are just as likely to report voting compared to African- American respondents from states that only required voters to state their name. Hispanic respondents in photo identification states are just as likely to report voting compared to Hispanic respondents from states that only required voters to state their name.

David B. Muhlhausen, Ph.D., is a Senior Policy Ana­lyst and Keri Weber Sikich is a research assistant in the Center for Data Analysis at The Heritage Foundation.