The Pure Effects of Voter Registration on Voter Turnout

Transcription

1 The Pure Effects of Voter Registration on Voter Turnout Barry C. Burden Jacob R. Neiheisel University of Wisconsin-Madison Department of Political Science 1050 Bascom Mall 110 North Hall Madison, WI Paper prepared for presentation at the October 27, 2009 meeting of the Political Behavior Research Group, University of Wisconsin-Madison Department of Political Science The authors would like to thank Stéphane Lavertu, Phil Wells of the Wisconsin Department of Administration, and Nathan Judnic at the Wisconsin Government Accountability Board for providing some of the data used in this paper.

2 INTRODUCTION In spite of decades of research on the effects of legal barriers on voter turnout and the composition of the electorate, we believe that this study is the first to measure the impact of the registration requirement itself. 1 Numerous previous studies, employing a wide range of data and methods, have concluded that early closing dates, residency requirements, and other legal hurdles discourage citizens from going to the polls. Scholars have carefully distinguished the effects of each type of restriction independent from the others. Surprisingly, even with the most concerted of data collection efforts scholar have not yet observed the impact of the registration requirement itself. It is challenging enough to distinguish the effects of each election regulation (Kim, Petrocik, and Enokson 1975; Hayduk 2005) and to separate those from cultural and other preexisting factors that would lead to a particular pattern of election laws across the states (Hanmer 2009; King 1994; Knack 1995). Even studies that looked at the introduction of the first personal registration requirements in the 1890s and early part of the 20 th century (Kleppner 1982; Kleppner and Baker 1980) were unable to estimate the effects of voter registration requirements absent additional legal barriers such as long closing dates, not to mention the poll taxes and literacy tests that were in place in the South (Filer, Kenny, and Morton 1991; Key 1949). But even if we believe that the effects of each aspect of registration can be isolated from one another and divorced from the political environment that gave rise to such restrictions on the franchise, estimating the effects of registration itself is another matter. The vast body of research on voter registration s effects analyzes a host of important statelevel regulations such as residency requirements, closing dates, felon disenfranchisement laws, policies for purging the rolls, and voter identification requirements. These important sources of inter-state variation appeared to have significant influence on turnout. Researchers were able to 1 The literature on the effect of administrative barriers to voting dates back at least as far as 1924 with the publication of Merriam and Gosnell s landmark study of the 1923 mayoral election in Chicago (see Hayduk 2005). 1

3 consider the counterfactual of how high turnout would be if every state adopted the most lenient of policies. Early estimates were substantial: Wolfinger and Rosenstone s (1980) seminal analysis put it at nine points and Powell s (1986) cross-national study estimated it to be 14 points. 2 These sizable estimates have framed, and one might say ossified, subsequent research. But recent work has challenged their assumption that registration is a major cause of voter abstention. Timpone s (1998, 155) careful analysis of registration and turnout concludes that Distinterested chronic nonparticipants are not likely to flood the polls simply because registration barriers diminish. Highton (2004, 508) similarly notes that it is a mistake to argue that if registration barriers were removed, non-registered citizens would vote at the same high rate as those who are registered. He reminds the reader that North Dakota, which lacks registration entirely, actually has lower turnout than among registrants in other states. It is vitally important to understand how aspects of registration influence turnout across the states, but we should also be careful to identify the effect of registration itself and not confuse it with other facets of voter registration. Such difficulties have long plagued work in this area, as Kleppner and Baker (1982) note in their critique of the legal-institutional theory as formulated by Rusk and others. The key methodological problem is asking a counterfactual question in which unmeasured differences across people or jurisdictions could be partly responsible for the observed effects. As Hanmer (2009) convincingly argues, the social and political character of a state is often related to the kinds of registration reforms that it adopts. That is to say that the relationship between registration laws and turnout is frequently endogenous, as states that already had high rates of voter turnout and strong participatory cultures to begin with were among the first to begin eliminating legal barriers to voting. A recent study by Ansolabehere and Konisky (2006, 84) also documents this concern nicely. As they explain, cross-sectional studies may suffer from an omitted variables problem because they fail to 2 As late as 1995 one cross-sectional study estimates a total liberalizing effect of 7.6 points (Mitchell and Wlezien 1995). 2

4 control for unobserved state-level characteristics that may affect voter participation rates. This flaw would exaggerate the effects of registration, suggesting that early estimates are inflated. To calculate more realistic estimates, they recommend exploiting longitudinal variation within a state to avoid these biases. Although perhaps the most sophisticated treatment of the topic to date, their study is not without its own share of shortcomings, as the states they examine not only gradually expanded voter registration to every county within the state during the time frame under consideration, but also added closing dates and other restrictions simultaneously). 3 As they note, it is therefore impossible to determine whether the estimates that they provide for the effect of registration requirements on voter turnout reflect the impact of the costs associated with making a separate trip to register ahead of a particular closing date or those incurred simply from having to register before being able to vote. What is needed is a quasi-experimental design in which a single state would implement a voter registration requirement exogenously and where turnout data could be collected in many jurisdictions over time. By holding state-level factors such as political culture and party competition constant, such an approach would rule out unmeasured state characteristics and provide ample statistical power. Further, an implementation that is unadorned by the various facets of registration such as closing dates and identification requirements would permit a clear view of the impact of the registration requirement itself (Nicholson-Crotty and Meier 2002). It is for this reason that we turn to the state of Wisconsin. Prior to 2006, more than 80% of the state s municipalities were exempt from voter registration. It was not until after the state implemented the Statewide Voter Registration System (SVRS) in response to the requirements of the federal Help America Vote Act (HAVA) that it was necessary for voters in all 1,851 of Wisconsin s municipalities to register before casting a ballot (Hanmer 2009; Huefner et al. 2007). We use this 3 See for instance Table 4.1 in Wolfinger and Rosenstone (1980). 3

5 intra-state variation in the application of voter registration requirements to overcome many of the methodological shortcomings found in previous studies. This design is especially useful in Wisconsin because the state has very few legal barriers to voting. Indeed, with the exception of North Dakota, a state which has no system of voter registration, Wisconsin has among the most liberal registration laws in America, having implemented election day registration (EDR) and registration by mail in 1976 (Smolka 1977). Absent the closing dates and other restrictions that are still common in other states we believe that it is possible to estimate the pure effects of registration on voter turnout. REGISTRATION REQUIREMENTS AND VOTER TURNOUT IN AMERICA Study after study has found that barriers to voting in the form of legal restrictions reduce turnout (Gilliam 1985; Highton 1997; Kelly, Ayres, and Bowen 1967; Mitchell and Wlezien 1995; Rosenstone and Hansen 2003; Wolfinger and Rosenstone 1980; cf. Fitzgerald 2005). But while numerous studies have attempted to determine which reforms are the most effective in stimulating voter turnout and which types of legal restrictions are most to blame for keeping voters away from the polls, scholars are rarely able to observe the effects of a single requirement or reform. Those who have attempted to do so have often come up with widely divergent estimates of the effects of one type of change. Perhaps the two most important and frequently studied aspects of registration are the closing date and EDR. Both are central to understanding how registration requirements dampen turnout. While scholars generally agreed that EDR lowered the cost of voting for many Americans (Knack and White 2000), thereby increasing overall turnout, the precise magnitude of the effect was a subject of much dispute in the literature (Brians and Grofman 1999). At the low end is Smolka s (1977) case-study approach to investigating voter turnout in Minnesota and Wisconsin after the introduction of EDR in the mid-1970s; in which he estimated the effect at one to two percentage points. At the high end, some studies have projected that implementing 4

6 EDR in an average state could increase turnout by as much as seven to ten percentage points (Brians and Grofman 2001; Burden et al. 2009; Rhine 1996). Numerous studies have provided slightly more conservative estimates, ranging from three to six percentage points on average (Fenster 1994; Hanmer 2009; Knack 1995, 2001; Rhine 1995). These differences may be attributable, at least in part, to the idiosyncrasies of particular datasets and the different methods and measurement strategies employed in each study. Aside from the manner in which they have operationalized EDR, previous studies have also differed in terms of the methodology and data that they employed. Brians and Grofman (2001) summarize the different ways in which scholars have attempted to measure the effects of EDR on turnout, dividing the previous literature into several distinct camps, pointing out the limitations of each approach in turn. Some studies of voter turnout used cross-sectional designs that likely overstated the effects of liberalizing registration requirements on voter turnout, as these studies often overlooked important differences among states (Highton 1997; Highton and Wolfinger 1998; Squire, Wolfinger, and Glass 1987; Wolfinger and Rosenstone 1980). The main problem with such studies is that the first states that introduced EDR in the 1970s already had very high levels of voter turnout (Brians and Grofman 2001; Hanmer 2009), thereby suggesting that there are a number of other factors affecting rates of turnout that vary between states. Inevitably, these factors go unmeasured (Ansolabehere and Konisky 2006; though see King 1994). Even studies using pooled cross-sectional designs (Mitchell and Wlezien 1995; Rosenstone and Hansen 1993) suffer from such shortcomings (Brians and Grofman 2001). Longitudinal designs (Ansolabehere and Konisky 2006; Brians and Grofman 1999; Knack 1995, 2001; Knack and White 2000; Fenster 1994) alleviate many of these concerns by setting aside the omitted variables problem (Ansolabehere and Konisky 2006). These studies employ a longitudinal component to control for variance in space and across time (Brians and Grofman 5

7 1999: 166) with the introduction of state (or county) and year-level fixed effects yielding some of the most convincing estimates of the effect of EDR on turnout. Until recently, however, studies that have relied upon quasi-experimental designs or fixed effects specifications to contend with both the observable and unobservable differences across states have been sharply limited in terms of the number of observations that they can bring to bear in their analyses. The debate about voter identification requirements, for example, has been unable to reach firm conclusions because of the small effect sizes and limitations of the data (Erikson and Minnite 2009). As Ansolabehere and Konisky note, Without more observations, these studies cannot easily rule out false positive results that is, that there is actually no effect of registration on turnout despite evidence seemingly to the contrary (2006, 85). By exploiting within-state variation in the application of registration laws, however, Ansolabehere and Konisky (2006) are able to overcome most of the shortcomings apparent in previous studies. But even though they examine two states New York and Ohio that are arguably less exceptional than the states with election day registration often studied to generate estimates of the effects of registration laws on turnout (Ansolabehere and Konisky 2006, 88), both states had closing dates well in advance of the election. It is therefore unclear as to whether their estimates of the effects of voter registration requirements on turnout reflect the costs associated with forcing voters to register before they can vote or those associated with the formidable costs imposed by the closing date (Gilliam 1985; Rosenstone and Hansen 2003; Wolfinger and Rosenstone 1980). There are two mechanisms driving their findings. First, an early closing date screens out peripheral voters who are not yet paying attention to the campaign. Second, having separate registration and balloting increases the costs of voting by requiring more information and effort by the individual. EDR eliminates both of these important concerns, but the registration hurdle itself remains. 6

8 When registration requirements were introduced statewide in Wisconsin in 2006, individuals registering to vote for the first time were not forced to navigate many bureaucratic hurdles beyond registration itself. Because Wisconsin offered EDR, new voters were able to register at the polls. Identification requirements in Wisconsin are also fairly light by comparison, as voters are able to use utility bills, bank statements, pay stubs, and other documents as proof of residence. Wisconsin even permits voters who do not have written proof of residency to bring a friend who has identification and lives in the same municipality to vouch for their residency status. At least in theory, then, EDR allows voters to register and vote in one essentially continuous act (Wolfinger, Highton, and Mullin 2005, 3; see also Brians and Grofman 2001; Highton 1997; Highton and Wolfinger 1998; Knack 1995; Teixeira 1992). Some have even gone so far as to argue that same-day registration eliminates all barriers to voting that are associated with registration (Mitchell and Wlezien 1995, 191). But there are reasons to challenge this assumption. EDR is not the absence of registration. As others have noted, EDR is actually more restrictive than the system that existed in many Wisconsin municipalities that lacked registration prior to HAVA (Ansolabehere and Konisky 2006; Hanmer 2009; Smolka 1977). EDR eliminates the closing date, which is the most consequential dimension of registration, but it does not eliminate registration itself, as one still needs to bring proper identification, complete paperwork, and find the correct polling place (Hanmer 2009). Indeed, nearly all of the extensive research conducted on the effects of registration on turnout focus on aspects of registration closing dates, locations, residency requirements, and the like rather than registration itself. This allows for estimates of which dimensions are most important, but does not identify the effect of registration alone. Yet registration is thought to be consequential in its own right (Piven and Cloward 1988), resulting in lower levels of voter participation in the U.S. than in other nations with automatic registration (Powell 1986). 7

9 In this study we therefore examine the pure effect of registration on voter turnout in Wisconsin where, in 2006, the state shifted from a limited system of voter registration to a statewide registration system. Before going on to describe the data and methods we employ, however, we provide a brief history of voter registration in Wisconsin, highlighting the aspects of the state s system that make it perhaps the ideal test case with which to examine the effects of registration requirements on voter turnout. A BRIEF HISTORY OF VOTER REGISTRATION IN WISCONSIN The first voter registration law in Wisconsin, passed in 1864, permitted municipalities to compile lists of registered voters before each general election. This periodic registration system was changed in 1912, though, when the first permanent voter registration system went into effect in the state and even then only in the city of Milwaukee (Smolka 1977). Although voter registration was ushered into practice in Milwaukee during a time when the general movement in election administration in other states was toward more restrictive practices, often involving closing dates for registration well in advance of the election, potential voters in Milwaukee could register at the election office until the last Tuesday night before an election (Harris 1929). Additionally, during the last two weeks of registration the office remain[ed] open until nine o clock in the evening to accommodate voters (Harris 1929, 276). Then, in 1927, a bill sponsored by the Wisconsin branch of the National League of Women Voters was passed by the state legislature that provided for a permanent voter registration system in municipalities of over 5,000 persons (Harris 1929; White 1928). 4 The state s voter registration system remained essentially unchanged until 1976 when, in response to a study by a task force on voter registration and elections, the state introduced 4 According to figures from the 1920 Census, there were only 41 municipalities in the state of Wisconsin that boasted of over 5,000 individuals. In 1930 that figure was little changed, with 45 municipalities having greater than 5,000 individuals. Individual municipalities below this threshold were permitted, however, to adopt voter registration if they so chose (Harris 1929). 8

10 registration by mail and EDR in those municipalities that already had voter registration. Conducted in 1972, the task force, under the direction of Governor Patrick J. Lucey, also recommended that the state move toward a system of mandatory statewide voter registration and that a master voter registry list to be compiled and maintained by the state s election officer (Smolka 1977, 35; Wisconsin Legislative Reference Bureau 1975). The task force s recommendations were not adopted wholesale, however, as the state retained voluntary voter registration for municipalities with fewer than 5,000 people (Smolka 1977). 5 When the new law went into effect in 1976, 245 of the state s 1,850 municipalities had voter registration, and of that number 125 had voluntarily decided to establish permanent voter registration lists. As a result, most of the state s population voted without registration requirements and the remainder operated under EDR. This near absence of registration requirements persisted for three decades. The crisis of the 2000 presidential election would change that. In response to the controversies in Florida over registration lists and voting equipment, in 2002 Congress and President Bush enacted the Help America Vote Act (HAVA). Among the mandates of HAVA was that each state would create a single, uniform, official, centralized, interactive computerized voter registration list that contains the name and registration information of every legally registered voter in the State. Wisconsin began creating a statewide database and voter registration essentially from scratch. So it was not until 2006 that the state would see a number of other recommendations from the 1972 study put into practice when the state instituted mandatory permanent voter registration statewide with the adoption of the Statewide Voter Registration System (SVRS) (Huefner et al. 2007). At the time that voter registration was implemented statewide, 172 municipalities were required to register voters and 167 smaller municipalities had voluntarily set up systems of voter registration. The remaining 1,512 municipalities began registering voters for the first time. 5 According to Harris (1929) early study on voter registration, at one time or another four other states Colorado, Iowa, Minnesota, and Ohio had similar systems of permanent registration. 9

11 We should note that studies of EDR s effect on turnout have mistakenly labeled Wisconsin as an EDR state in the period before Indeed, only a handful of studies even acknowledge the fact that states are not perfectly internally homogenous with regard to their registration requirements (Rhine 1996, 182; Hanmer 2009; Keyssar 2000; Knack 2001; Smolka 1977; Wolfinger and Rosenstone 1980). Here we use this within-state heterogeneity to good effect, providing perhaps the first glimpse at the effects of the registration requirement itself in isolation. DATA AND METHODS We evaluate the effect of voter registration on turnout using the exogenous intervention of HAVA and SVRS in The data compiled for this study were obtained from a variety of different government sources. Our key explanatory indicator of whether or not a municipality registered voters prior to 2006 was gleaned from a report released by the state of Wisconsin s Legislative Audit Bureau in Vote totals for each municipality from 2000 to 2008 were gathered from the Wisconsin State Legislature, while totals from 1996 and 1998 were collected from the Wisconsin Blue Book series. From these totals we were able to calculate the numerator for our dependent variable voter turnout which is constructed in the traditional way by dividing the total number of voters in each municipality by the municipality s voting age population (VAP). 6 For presidential election years we use the total number of votes cast for president, while for mid-term years we use the total number of votes cast for either senator or governor. We depart somewhat from previous studies in that, for years in which there was both a senatorial and gubernatorial race, we use the highest vote total from either of those two races. Doing so, we believe, provides a more accurate picture of how many people actually cast ballots. The denominator for our dependent variable is constructed using estimates of the VAP compiled by the Wisconsin Department of Administration s (DOA) Demographic Services Center. 6 A measure of the Voting Eligible Population (VEP) would be preferable, but it is not available at the municipal level at multiple points in time. 10

12 Unlike previous studies that have been forced to use linear interpolation to estimate VAP in noncensus years (Ansolabehere and Konisky 2006), we were able to obtain yearly estimates that adjust for a number of different factors that, since population change is not a uniform, linear process, go unaccounted for using simple interpolation. It is fortunate that Wisconsin has one of the most decentralized systems of election administration in the country. Elections and voter registration are conducted at the municipal level. This provides more variance on the dependent variable across units and more statistical power, allowing for maximum leverage. Moreover, there is tremendous variation across the state s 1,851 municipalities in terms of their size. They range from the largest city of Milwaukee (with a 2000 VAP of 425,990) down to the town of Cedar Rapids (with a 2000 VAP of 24). This population range is important because it is strongly correlated with the resources each municipality had available to implement registration once it became a statewide requirement. Large communities already had large professional staffs administering municipal government, but many of the state s small municipalities had part-time clerks who lack training, staff, and even offices in which to conduct business. A surprisingly large number worked without assistance out of their homes and were compelled to register voters there. While our measure of whether a community had voter registration represents an improvement over that used in previous studies, it is not without a fault of its own, as we cannot be certain that all of the municipalities that we categorize as having voter registration prior to SVRS had it over the entire time period under consideration ( ). The Legislative Audit Bureau report from which we gleaned this variable only indicated whether a municipality had voter registration before It is possible, then, that we are overestimating the number of municipalities with voter registration in some of the models that employ the full dataset. Of the 339 municipalities that are listed as having voter registration, 245 have had registration since the introduction of EDR in

13 The remaining 94 adopted registration sometime prior to 2005, thereby making our assumption that the number of municipalities with registration prior to SVRS remained constant from 1996 to 2005 somewhat tenuous. This assumption is more plausible in models that use a smaller subset of the data, or only look at one election before and one election after the introduction of voter registration statewide. Fortunately the results are not particularly sensitive to the size of the window we use. As we shall see shortly, we are forced to make other assumptions as well. Table 1 compares the demographic characteristics of Wisconsin municipalities with voter registration with those that did not register voters in 2000, the only year for which these measures were available at the municipal level. As with previous studies, perhaps the most notable difference between the municipalities that had voter registration and those that did not is population. This is hardly surprising, as Wisconsin, along with many other states, implemented voter registration according to population before the statewide voter registration database was put into place. The average population size in municipalities that were forced to register voters is about twenty times as large as the average population in municipalities without voter registration, although this figure is probably somewhat inflated due to the inclusion of Milwaukee, the state s largest city. In addition, municipalities with registration were more educated on average, had higher median family incomes, and more African Americans than those without registration. Differences in the median age were negligible. As Table 1 shows, among those that had registration prior to SVRS there are also some disparities between municipalities that voluntarily implemented a system of registration and those that were forced to do so by virtue of population size. [Insert Table 1 about here] Unfortunately, data on these socioeconomic characteristics are only available at the municipality level in census years. Previous studies have dealt with this problem by using linear interpolation to calculate between-census values. To guard against other potential confounds, 12

14 though, we exploit the panel structure of our data. Specifically, our panel models include fixed effects for each municipality and year in an effort to isolate the effects of introducing registration. The inclusion of fixed effects at the municipal level allow us to remove any time-invariant, unobserved factors that might influence turnout across communities (Angrist and Pischke 2009; Wooldridge 2002). While we might prefer to measure the socioeconomic characteristics of each municipality over time, we believe that it is reasonable to assume that such characteristics are relatively stable over the time period under examination. What is more, we were able to obtain the most accurate estimates of population available from the Wisconsin DOA for each time point a luxury that earlier studies simply were not afforded. Given that we would expect the various socioeconomic characteristics of each municipality to correlate strongly with population, it is not unreasonable to think that the potential confounding effects of these unobserved variables are almost wholly captured in the models by the inclusion of a variable for population. In addition, we also estimate differences-in-differences models that allow us to examine the short-term influence of the treatment (the introduction of voter registration) on turnout. Because these models rely on a shorter time period (one election before to one election after the implementation of SVRS) it is more likely that these potential confounders are relatively invariant. It is in this way that we mitigate the risk of introducing any bias due to omitted variables. Once again, though, it seems reasonable that any potential bias introduced into the models is likely to affect our estimate of the influence of population on turnout rather than our (more important) estimate of the treatment effect of registration. It is difficult to overstate the incredible leverage that these data afford as we attempt to estimate the effect of the registration requirement itself on voter turnout. Not only is the statewide adoption of voter registration in 2006 exogenous, owing to the requirements imposed on the state by HAVA (Hanmer 2009), but the presence of EDR in Wisconsin when voter registration was 13

15 implemented completely does away with the confounding influence of the closing date. Our population estimates are the best available, as we were not forced to assume monotonic, linear increases from one year to the next. In addition, because elections are administered at the municipal level in Wisconsin, our tests have ample statistical power, even with loss in degrees of freedom incurred with the inclusion of fixed effects. RESULTS Our first regression model is a fixed effects specification that exploits the panel structure of the data, making use of both cross-sectional and over-time variation in the application of voter registration requirements. To examine the long-term average difference in turnout between municipalities with registration and those without, we estimated the following model: (1) T it = β 0 + α i + δ t + β 1 R it +ΨX st +ε it where i denotes municipalities and t denotes time and T it is voter turnout in a particular municipality and year. Our key independent variable is R it, which is a dummy variable equal to one if the municipality introduced voter registration in 2006 as a result of the HAVA/SVRS mandate. We include year dummies for all but one year, δ t, to control for year-specific shocks and ballot effects, and municipal-level fixed effects, represented in the model by α i, which allows us to account for any unobserved, time-invariant factors that might influence turnout. The vector X st represents the natural log of the population in each municipality, while β 1 and Ψ are regression coefficients, β 0 is a constant, and ε it is an error term. This setup is merely a pooled cross-sectional model with both space and time fixed effects. Beyond accounting for population differences, the only moving part in the model is the treatment of introducing a new registration requirement for municipalities that did not previously have it. [Insert Table 2 here] 14

16 Table 2 displays the results of this model, which unambiguously support the hypothesis that voter registration requirements have a negative impact on turnout. The coefficient on our key independent variable is.017 with a standard error of.003, suggesting about a two percentage point decline in voter turnout for municipalities moving from no registration to having registration with the implementation of SVRS. Given the exogeneity of the intervention coming as it did from the federal government and motivated by a crisis in another state six years earlier we are confident that the estimate captures the actual causal effect accurately. We regard this estimate as a pure estimate of the effect of registration because it depends on the most minimalist definition of registration. In the presence of EDR, not even closing dates were a hindrance. We estimated several additional models to test the robustness of these results. We ran the same model formalized in the equation above, only this time omitting the state s largest municipality Milwaukee from the dataset. The results remain unchanged. We also ran this model with an AR1 (first order autoregressive) error process, explicitly acknowledging that voter turnout may not be independent from one election to the next (Ansolabehere and Konisky 2006; Bertrand, Duflo, and Mullainathan 2004). Once again, the results are little changed, as the coefficient on our key independent variable indicates that the switch from no registration requirement to having voter registration reduced turnout by two percent on average. This consistency of findings is not surprising because the municipal fixed effects already account for most of the over-time continuity of turnout within communities. The results of these additional models are shown in Table A in the Appendix. A concern is that there might be something distinctive about those municipalities that voluntarily began to register voters before Recall from Table 1 that the median age in these municipalities was higher than that in the other municipalities that were required by statute to have voter registration. Municipalities that voluntarily adopted registration requirements also had a lower 15

17 percentage of African Americans than did other municipalities with registration. Even though most other differences were negligible, there is still the possibility that these municipalities are somehow different from those that were forced to register voters by virtue of population size. Excluding these municipalities from the analysis reduces the size of the effect that we found in previous models, but the basic relationship remains unchanged. With voluntary registration municipalities dropped from the analysis, the introduction of voter registration requirements statewide in 2006 nevertheless still reduced turnout in municipalities that were not required to registered voters by about one percent on average, although with a slightly larger standard error (see Table A in the Appendix). As mentioned above, measurement error is a possibility, given that we cannot be certain whether the municipalities that we list as having voter registration prior to 2006 had registration over the entire period under examination. It is for this reason that we shorten the time period over which we examine the effects of the registration requirement on voter turnout a move which makes it more likely that our measure of our key independent variable is reasonably accurate. Although this reduces our statistical power, shortening the window is a more conservative test of the main causal effect. We reestimated the model formalized above after excluding the period from 1996 to The results of this model are displayed in the last column in Table A in the Appendix. The coefficient on our key independent variable in this model is similar in magnitude to that in previous models, and suggests that the introduction of voter registration reduced turnout by two percentage points on average in those municipalities that did not register voters prior to SVRS. 7 Taken together, the results of these models strongly suggest that the introduction of voter registration requirements statewide lowered voter turnout in Wisconsin, even absent the closing 7 We also estimated a differences-in-differences model that included a dummy variable for whether a municipality had voter registration prior to 2006, a time dummy that switches on for observations obtained after the change in the law, and an interaction term that multiplies the two, plus a control for population. The results of this model suggest about a 1.3 percentage point decline in turnout with the introduction of voter registration. See Angrist and Pischke (2009) and Meyer (1995) for more on this type of model. 16

18 dates and other bureaucratic hurdles that voters in states without EDR must navigate. While informative, the fixed effects models reported above only afford a look at the average effect of introducing voter registration, and tell us little about the immediate effect of registration requirements on turnout. Our next models therefore build upon the fixed effects specification, estimating a differences-in-differences or first-differences model that allows us to examine the shortterm change in turnout in municipalities with registration requirements in place prior to 2006 as compared to those without (Ansolabehere and Konisky 2006; Konisky 2009). The specification is as follows: (2) T it - T it-1 = β 0 + β 1 (1 - R it-1 ) + Ψ(X it - X st-1 ) + ε it, where, as in equation one above, i indexes municipalities, t indexes years, and our dependent variable, T, is voter turnout at the municipal level. R is a dummy variable coded one if a municipality had voter registration prior to 2006 and zero if it did not have registration. Since all municipalities had registration following the adoption of SVRS in 2006, R equals one in the election immediately after the change. The difference is 1 - R it-1, and so municipalities that had voter registration before the law changed take on a value of zero in the post period. Once again, X it is the natural log of each municipality s population; the difference, X it - X st-1, represents the change in this variable between the pre and post time periods. As before, β 1 and Ψ are regression coefficients, β 0 is a constant, and ε it is an error term. [Insert Table 3 about here] We estimated this model separately for presidential and off-year elections, examining the short-term effects of introducing voter registration in each, comparing 2002 to 2006 in the first model and 2004 to 2008 in the second model. The results of these models are presented in Table 3, and tell a story that comports well with the results of the fixed effects models described above. The coefficient on the registration variable for the first model, comparing turnout in 2002 to turnout in 17

19 2006, is , which suggests that there was an immediate effect of the registration requirement on turnout at the municipal level. This effect appears to be slightly larger in presidential years, but is nevertheless still on par with earlier estimates of about a two point decline in turnout. Our results suggest that voter registration requirements, no matter how minimal, reduce turnout, yet the mechanism underling this connection remains unclear. If decades of research on this topic were our only guide, we would hypothesize that registration is primarily a burden on the less educated and experienced voters. We do not doubt that this is the case. Yet it is also possible that the disturbance caused by the introduction of registration is a function of variations in election administration as well. In particular, it seems plausible that the effect varies by the size of the municipality. That is, while the average effect of introducing registration requirements statewide may be negative, some municipalities might have been better equipped to handle the additional administrative burdens than others. Ansolabehere and Konisky, for instance, suggest that larger cities may have more registration problems than smaller communities (2006, 88). And while they note that the existing literature has not demonstrated such an interaction with population, they nevertheless go on to test the interactive effect of registration laws and population. They do not, however, find a statistically significant interaction. There is reason to believe that we might very well find such an interaction in these data, as municipalities in Wisconsin vary considerably with respect to the amount of resources that they can bring to bear in implementing changes mandated by the state. Contrary to Ansolabehere and Konisky (2006), however, we hypothesize that smaller municipalities, not larger ones, encounter more problems registering voters. Larger municipalities almost always have more professionalized and greater numbers of election workers than do smaller municipalities, and larger tax bases to fund their activities. In contrast, election officials in smaller municipalities often have fewer resources to marshal in meeting state-imposed guidelines, and frequently require assistance from county election 18

20 officials to meet their statutory mandates. One official report on Wisconsin illustrates these concerns nicely, writing that: The highly decentralized nature of election administration in Wisconsin means that elections are actually conducted by clerks working with limited resources and balancing competing demands on their attention. Indeed, even communication and training are challenges, particularly when many clerks do not have accounts and most do not work full time in their clerk capacities (Burden et al. 2009, 11). It is for these reasons that we believe that we are likely to see a significant interaction effect with population, with turnout in larger municipalities suffering far less as a result of the introduction of the registration requirement than in smaller municipalities. [Insert Table 4 about here] To test this conjecture, we added an interaction term multiplying population (in thousands, for the sake of interpretation) with our indicator for whether a municipality had voter registration to our original fixed effects specification and reestimated the model. The results of this model are shown in Table 4, and indicate the presence of a significant positive interaction effect between registration laws and population. Consistent with our expectations, it appears that larger municipalities suffered less of a drop in voter turnout as a result of the change, while turnout in the smallest municipalities in Wisconsin fell even more sharply than might have been expected based on the average effect of having voter registration reported in the initial fixed effects models. The direct effect of population remains negative, with larger municipalities have lower turnout on average, but the interaction suggests that registration was less of a hindrance in larger municipalities than in smaller ones. There appears to be a population tipping point where the depressant effect of registration is offset by the greater ability of the clerk to implement the new requirement is a way that facilitates rather than inhibits voter participation. 19

21 DISCUSSION AND CONCLUSION For over a century registration with the state has been a prerequisite for voting for nearly everyone in the United States. Registration was introduced during the Progressive Era largely to combat fraud, and there is some evidence that it was successful. But all reforms have multiple consequences, some of them unexpected. The introduction of registration lowered turnout and worked to keep turnout at lower levels in the U.S. than in other comparable democracies, or at least that has been the conventional wisdom. We have scrutinized this assumption on two fronts. First, we have made clear that most studies of the effects of registration have not been of registration per se but rather of dimensions, or facets, of registration. Prime among these has been the closing date by which registration must be completed, to be contrasted with EDR where potential voters may both register and vote on the same day. We have shown that, while these two variations of registration laws may be vitally important, neither is registration itself. Using data on the exogenous introduction of the most minimal of registration laws in Wisconsin municipalities in which EDR is available and identification standards are light, we have shown through a variety of modeling approaches that the pure effect of registration is approximately two percentage points. This is well below the figures cited in the literature, which include this effect as well as other elements in the registration process. Although small when compared to earlier estimates of the effects of registration requirements on voter turnout, it is important to note that it is not zero, as previous studies of EDR have suggested (Mitchell and Wlezien 1995; Wolfinger, Highton, and Mullin 2005). Nevertheless, in considering why turnout is so much lower in the U.S. than other industrial democracies, registration itself cannot account for much of the explanation. Second, we have shown that the varying effects of registration are not due solely to the capacities of individuals. Although individual level differences may well be important determinants of how consequential registration requirements are, we have shown that administrative capacity is 20

22 also a central determinant. Based on our understanding of the varying resources available to clerks according to population, we have shown that registration had a much larger negative impact in smaller municipalities where clerks were presumably less able to implement a new registration requirement with ease. Election reforms do not exist in isolation but rather depend on the ability and willingness of election officials to implement them. In the case of registration it seems that administrative capacities made the introduction of registration affect electorates in small communities disproportionately. 21

23 REFERENCES Angrist, Joshua D., and Jörn-Steffen Pischke Mostly Harmless Econometrics: An Empiricist s Companion. Princeton: Princeton University Press. Ansolabehere, Stephen, and David M. Konisky The Introduction of Voter Registration and Its Effect on Turnout. Political Analysis 14: Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan How Much Should We Trust Differences-In-Differences Estimates? The Quarterly Journal of Economics 119(1): Brians, Craig Leonard, and Bernard Grofman When Registration Barriers Fall, Who Votes? An Empirical Test of a Rational Choice Model. Public Choice 99(1-2): Brians, Craig Leonard, and Bernard Grofman Election Day Registration s Effects on U.S. Voter Turnout. Social Science Quarterly 82: Burden, Barry C., David T. Canon, Stéphane Lavertu, Kenneth R. Mayer, and Donald P. Moynihan Wisconsin Election Data Collection Grant Program Evaluation Report: A Report to the Wisconsin Government Accountability Board. Burden, Barry C., David T. Canon, Kenneth R. Mayer, and Donald P. Moynihan The Effects and Costs of Early Voting, Election Day Registration, and Same-Day Registration in the 2008 Elections. Paper presented at the annual meeting of the American Political Science Association, Toronto. Erikson, Robert S., and Lorraine C. Minnite Modeling Problems in the Voter Identification- Voter Turnout Debate. Election Law Journal 8(2): Fenster, Mark J The Impact of Allowing Day of Election Registration Voting on Turnout in U.S. Elections from 1960 to 1992: A Research Note. American Politics Quarterly 22: Filer, John E., Lawrence W. Kenny, and Rebecca B. Morton Voting Laws, Educational Policies, and Minority Turnout. Journal of Law & Economics 34(2): Fitzgerald, Mary Greater Convenience But Not Greater Turnout: The Impact of Alternative Voting Methods on Electoral Participation in the United States. American Politics Research 33(6): Gilliam, Franklin D., Jr Influences on Voter Turnout for U.S. House Elections in Non- Presidential Years. Legislative Studies Quarterly 10: Hanmer, Michael J Discount Voting: Voter Registration Reforms and their Effects. New York: Cambridge University Press. Harris, Joseph P Registration of Voters in the United States. Washington, D.C.: Brookings Institution. Hayduk, Ronald Gatekeepers to the Franchise: Shaping Election Administration in New York. DeKalb: Northern Illinois University Press. Highton, Benjamin Easy Registration and Voter Turnout. Journal of Politics 59(2): Highton, Benjamin Voter Registration and Turnout in the United States. Perspectives on Politics 2: Highton, Benjamin, and Raymond E. Wolfinger Estimating the Effects of the National Voter Registration Act of Political Behavior 20(2): Huefner, Steven F., Daniel P. Tokaji, and Edward B. Foley, with Nathan A. Cemenska From Registration to Recounts: The Election Ecosystems of Five Midwestern States. The Ohio State University Michael Moritz School of Law. Kelley, Stanley, Jr., Richard E. Ayres, and William G. Bowen Registration and Voting: Putting First Things First. American Political Science Review 61: Key, V.O Southern Politics in State and Nation. New York: Vintage. 22

24 Keyssar, Alexander The Right to Vote: The Contested History of Democracy in the United States. New York: Basic Books. Kim, Jae-On, John R. Petrocik, and Stephen N. Enokson Voter Turnout among American States: Systemic and Individual Components. American Political Science Review 69: King, James D Political Culture, Registration Laws, and Voter Turnout among the American States. Publius: The Journal of Federalism 24: Kleppner, Paul Who Voted? The Dynamics of Electoral Turnout, Santa Barbara, CA: Praeger. Kleppner, Paul, and Stephen C. Baker The Impact of Voter Registration Requirements on Electoral Turnout, Journal of Political and Military Sociology 8: Knack, Stephen Does Motor-Voter Work? Evidence from State-Level Data. Journal of Politics 57: Knack, Stephen Election Day Registration: The Second Wave. American Politics Research 29: Knack, Stephen, and James White Election Day Registration and Turnout Inequality. Political Behavior 22(1): Konisky, David M The Limited Effects of Federal Environmental Justice Policy on State Enforcement. Policy Studies Journal 37(3): McDonald, Michael P Portable Voter Registration. Political Behavior 30(4): Meyer, Bruce D Natural and Quasi-Experiments in Economics. Journal of Business & Economic Statistics 13(2): Mitchell, Glenn E., and Christopher Wlezien The Impact of Legal Restraints on Voter Registration, Turnout, and the Composition of the American Electorate. Political Behavior 17(2). Nicholson-Crotty, Sean, and Kenneth J. Meier Size Doesn t Matter: In Defense of Single- State Studies. State Politics & Policy Quarterly 2(4): Piven, Frances Fox, and Richard A. Cloward Why Americans Don t Vote. New York: Pantheon Books. Powell, G. Bingham, Jr American Voter Turnout in Comparative Perspective. American Political Science Review 80: Rhine, Staci L Registration Reform and Turnout Change in the American States. American Politics Research 23(4): Rhine, Staci L An Analysis of the Impact of Registration Factors on Turnout in Political Behavior 18: Rosenstone, Steven J., and John Mark Hansen Mobilization, Participation, and Democracy in America. New York: Pearson-Longman. Smolka, Richard G Election Day Registration: The Minnesota and Wisconsin Experience in Washington, D.C.: American Enterprise Institute for Public Policy Research. Squire, Peverill, Raymond E. Wolfinger, and David P. Glass Residential Mobility and Voter Turnout. American Political Science Review 81(1): Teixeira, Ruy A The Disappearing American Voter. Washington, D.C.: The Brookings Institution. Timpone, Richard J Structure, Behavior, and Voter Turnout in the United States. American Political Science Review 92(1): White, Leonard D Public Administration, American Political Science Review 22(2): Wisconsin Legislative Audit Bureau An Evaluation: Voter Registration. Report Wisconsin Legislative Reference Bureau Recent Changes in Voter Registration. Informational Bulletin 75-1B-7. 23