Social Issue Ballot Measures and Their Impact on Turnout [WORKING DRAFT] Daniel R. Biggers

Social Issue Ballot Measures and Their Impact on Turnout [WORKING DRAFT] Daniel R. Biggers Biggers 1

Abstract Over the past few elections there has been significant discussion among scholars, politicians, and media pundits about the ability of the presence of social issue ballot measures, such as those relating to gay marriage or abortion, to increase turnout. Recent research on the subject, however, has often treated these ballot measures as separate, individual measures, ignoring a possible cumulative effect. I investigate this possibility by examining the relationship between the total number of social issue measures on the ballot and voter turnout. Employing the Current Population Survey November Voting and Registration Supplement from the past five elections, a unique data set which identifies the nature of ballot measures, and hierarchical generalized linear modeling to account for the multilevel nature of the data, I find that the number of social issue measures present on the ballot had a positive impact on turnout during the past three midterm elections and one of the past two presidential elections. I discuss the implications of these findings in the conclusion. [This is clearly a working draft. I finished the results section and conclusion this weekend, so they have not been revised (I apologize if these sections is painful to read). Some pieces are missing (notably parts of the appendix and the figures for my predicted probabilities), but all of the text is there. I welcome any and all feedback, suggestions, criticisms, praise, or condemnation as I get this paper ready to take to Midwest. Thanks in advance for reading my paper.] Biggers 2

Proponents of direct democracy have long championed the theorized benefits it bestows upon its citizens, most notably an increase in their likelihood of turning out on Election Day. Scholars, however, have found mixed empirical evidence to support these claims. Most of the literature generally agrees that the ability to legislate from the ballot does increase turnout during midterm elections, but that it has little or no effect during presidential elections. The lack of a consensus, however, about how to best measure the process has led to the employment of differing measures of the initiative process, as well as the clouding of our theoretical expectations regarding the process. Highlighting the problematic nature of this debate is the fact that there has been a significant focus in recent elections about the impact of ballot measures on the political process. Discussion about whether or not individual propositions are responsible for electoral outcomes has become a recurring event, at least within the media, during each election. Most of this debate has focused exclusively on measures that deal with social issues, such as abortion, gay marriage, or stem cell research, which are often highly salient and have the potential to arise both the passion and interests of the average voter. The increasing focus on the mobilization effect of ballot measures in general, and of social issue ballot propositions in particular, warrants both a further investigation and a better theoretical understanding of the relationship hypothesized by direct democracy proponents. I seek to elucidate the relationship between ballot measures and turnout, contending that, for both theoretical and empirical reasons, the issue content of the proposition is a superior measure of direct democracy than previously employed measures. Taking the issue content into account, social issue ballot propositions, because Biggers 3

of their high salience and controversial nature, are likely to generate the expected positive effect on turnout. To test this hypothesis, I employ the Current Population Study November Supplements for the election years 1998 to 2006 and hierarchical generalized linear modeling to deal with the multilevel nature of the data. I find that an increase in the number of social issue ballot measures boosts turnout during all the midterm and some of the presidential elections over the period under investigation. I discuss the implications of these findings, which enhance our understanding of the effect of direct democracy on turnout, in the conclusion. Ballot Measures and Voter Turnout Normative theorists have long posited that direct democracy can motivate citizens to participate in the electoral process (Barber 1984; Pateman 1970). By employing institutions of direct democracy, such as initiatives, proponents of the process believe that a sense of civic duty and participatory responsibility can be instilled among the citizenry. Supporters argue that the ability to legislate from the ballot box can both inspire and educate the citizen. Their theory posits that a byproduct of this opportunity will be higher levels of political participation, most notably increased levels of turnout on Election Day (Smith and Tolbert 2004). Given that turnout in the United States is low in comparison to similar western democracies (Powell 1986; Teixeira 1992), there are normative reasons to investigate institutions that facilitate an increase in turnout. If a higher participation rate is desirable, and if a process can be identified to help increase this rate, then one must heavily consider their universal adoption (regardless of the citizenry s ability to handle this responsibility) (Piven and Cloward 2000). Biggers 4

The increasing use of ballot initiatives and the rise in media coverage they are afforded would please proponents. Initiative use during the 1990s was at its highest level ever (Tolbert and Smith 2005). During the first years of the new century (through 2004), its use has held steady (Tolbert and Smith 2005), and it appears as though its employment will keep pace with the explosion of its use during the 1990s (Smith and Tolbert 2007). More than half of the states provide for some form of direct democracy, and as of 2004 more than seventy percent of the country s population resides in either a state or city that permits legislating from the ballot (Matsusaka 2004). Empirical research investigating the relationship between citizen legislation and voter turnout, however, has yielded ambiguous results, often due to the varying measures of the initiative process employed. Because of the inherent differences between the two, the literature usually differentiates between presidential and congressional elections. Presidential elections are high information and motivation events, during which the resources necessary to motivate citizens on the margin of voting to turn out, such as information and partisan or nonpartisan mobilization, are readily available. Any impact on turnout that direct democracy might have is likely to be overshadowed by these factors. During midterm elections, however, both voter and media interests are likely to focus on state level candidates and issues (Smith 2001). Mobilization resources are less abundant as well, meaning that marginal voters are more susceptible to information and mobilization from initiative campaigns (Bowler and Donovan 1998; Tolbert et al. 2003). The empirical evidence largely supports these contentions. Some aggregate level analyses suggest that turnout is higher in presidential elections in states that have more initiatives on the ballot (Tolbert and Smith 2005; Tolbert et al. 2001), but not in states Biggers 5

that simply have the process (Tolbert et al. 2001). Individual level results are mixed, suggesting that the number of initiatives is associated with higher turnouts for certain presidential elections (Tolbert et al. 2009), only sometimes (Tolbert et al. 2003), or not at all (Scholzman and Yohai 2008). In addition, the number of initiatives on the ballot over a twenty year time period (Scholzman and Yohai 2008), the cumulative number of initiatives ever to appear on a state s ballot (Scholzman and Yohai 2008), and the presence of salient initiatives (Lacey 2005; Scholzman and Yohai 2008; Smith 2001) do not have a positive impact on turnout during presidential elections, while per capita spending on initiatives does have the hypothesized effect (Tolbert et al. 2009). In contrast, there is a more of a consensus among scholars that initiatives can increase turnout in midterm elections, when, as discussed above, levels of available information, mobilization efforts, and turnout are lower than in presidential elections. Aggregate level analyses suggest that turnout is higher during midterm elections in those states that have higher numbers of initiatives on the ballot (Tolbert and Smith 2005; Tolbert et al. 2001) or those that simply have the process (Tolbert et. al 2001). Individual level examinations have determined that the number of initiatives on the current ballot (Scholzman and Yohai 2008; Tolbert et al. 2003; Tolbert et al. 2009), the number of initiatives to have appeared on the ballot over the past twenty years (Scholzman and Yohai 2008), the total number of initiatives to have ever appeared on a state s ballot (Scholzman and Yohai 2008), the number of salient initiatives on the current ballot (Lacey 2005; Scholzman and Yohai 2008; Smith 2001), and per capita spending by initiative campaigns (Tolbert et al. 2009) are all associated with higher levels of turnout Biggers 6

for the vast majority, if not all, of the elections years investigated (but see Keele (2008), who finds significantly less support for the mobilization effect of initiatives). Why is salience important? (Is a gay marriage ban the same as a bond bill?) In evaluating these findings, we should first identify the exact mechanism that we suspect is contributing to a rise in turnout. There are essentially two possibilities: the initiative process itself, or some aspect of the initiatives that appear on the ballot (such as their number, issue content, campaign spending, or media coverage). If we theorize that it is the presence of the initiative process that stimulates turnout, then we must account for the nonrandom assignment of the initiative process across states. In other words, there are serious endogeneity concerns about determining the relationship between institutional factors and their effects in general (March and Olsen 1984), and calculating the treatment effect of initiatives in this manner in specific. For instance, states that employ the initiative process may have a different political culture which places more emphasis on voter participation. If this is the case, then these states, even in the absence of the initiative process, would likely still have higher turnout rates. We should thus not expect to see an increase in turnout simply because a state has the initiative process. Using a synthetic case control which takes into account the nonrandom assignment of the initiative process across states, Keele (2008) demonstrates this fact, finding that overall there is little evidence that the introduction of the initiative process boosts a state s turnout level. In contrast, if we theorize that it is something about the initiatives on the ballot that motivates people to turn out, we at least in theory acknowledge the importance of the salience of individual ballot propositions. If we contend that direct democracy increases Biggers 7

an individual s probability of showing up to the polls on Election Day, then we must also posit that that individual has some basic awareness of at least one initiative on the ballot. Simply put, if initiatives are the catalyst behind an individual s decision to vote, then they must be aware of them; one cannot argue that some factor the individual is unaware of caused them to vote. These citizens need not necessarily be informed about the measure, but they must at least know about it. As Nicholson (2003) observes, Absent a basic awareness, ballot propositions can do little to promote a democratic citizenship (403). We should thus expect any effects of the direct democracy process to be reliant upon the individual s level of awareness (Dyck and Lascher n.d.). The importance of awareness suggests that the mechanism that is increasing turnout is likely some factor related to each individual measure. While awareness in general is often considered to be based on personal factors, such as an individual s ability and motivation, it is also a product of opportunity (Delli Carpini and Keeter 1996), which reflects factors that deal specifically with the nature of each individual initiative. These factors include issue content (Nicholson 2003), campaign spending (Bowler and Donovan 1998; Nicholson 2003), and media coverage (Nicholson 2003; Pelika 2008). In formulating a proper measure of direct democracy, we should remain conscious of these factors that are identified as influencing awareness of ballot measures. Given that we often control for the individual level factors in our empirical analyses (i.e. ability and motivation), we should focus exclusively on the specific factors of individual propositions (i.e. opportunity). When we acknowledge the importance of awareness, it is immediately apparent that not all initiatives are equal, and we should expect that the ability of the initiative to Biggers 8

increase voter turnout is conditional on the nature of the initiative. This is because not every proposition is going to meet the opportunity factors discussed above in the same manner. Some ballot measures will simply be more salient than others, due to the fact that their issue content is more compelling or controversial, campaign spending is higher, the campaign is more intense, or they garner more media attention. Certain ballot issues, because they are controversial or highly visible, are likely to stir up more interest than others (Cronin 1989). In addition, because all initiatives are not equal, we should not expect each initiative to impact turnout in the same manner. This means that the total number (or a count) of initiatives on the ballot should not necessarily be viewed as an indicator of increased turnout. For example, it is highly plausible that five ballot measures about obscure bond bills would not have as great of a mobilization effect as that of a single, highly salient gay marriage ban (Keele (2008) and Smith (2001) make similar arguments about how not all initiatives are equally salient). Thus, a count of the propositions on the ballot, which is designed to differentiate between states that make heavy use of the initiative process and those that allow for it but do not employ it often, may not be a proper measure since it may overestimate the salience of some initiatives. Returning to the example above, employing a count would lead us to conclude that the five bond bills have five times the impact on turnout as one gay marriage ban. Because we should expect a heterogeneous effect of different ballot initiatives, employing a simple count is contradictory to our expectations (Keele 2008). Many scholars have at least identified the importance of salience in facilitating higher turnout rates. Besides those scholars who have considered a count of the number of initiatives as a measure of salience (Tolbert et al. 2001), other scholars have looked at Biggers 9

newspaper coverage of initiatives, measuring salience as the amount of front page coverage that ballot measures receive the day after the vote (Lacey 2005; Scholzman and Yohai 2008; Smith 2001). This measure has been objected to by some scholars (Tolbert et al. 2001), who question whether front page newspaper coverage is really a measure of salience, or whether it is a measure of the opinions of editors who decide what is placed on the front page. Lacey (2005) counters this point by arguing that if newspapers wish to stay in business, they must run stories that reflect their market s interests, not the interests of their editors. Thus, market constraints on editors, the scholar argues, add validity to this measure of salience. Avoiding the debate as to what newspaper coverage is really measuring, this measure does appear to capture at least part of what we would deem as salience. The problem, however, is that this measure is based on an analysis done after the election. While such a measure undoubtedly helps us to understand what initiatives did motivate citizens to vote, it does not allow us to understand what initiatives will (or do) facilitate turnout. In other words, this measure of salience does not allow us to know before the election which initiatives will have a positive impact on turnout; we instead must wait until after the election to determine the impact of individual initiatives. While such information undoubtedly aids our understanding of what has happened, we should also be interested in both what is happening and what will happen. While the role of the political scientist may not always be to predict what is going to happen, we should be able to understand what is currently taking place, as well as the roles that individual initiatives may play in the process. While this measure of salience certainly helps scholars to understand previous elections, we should develop a measure of salience that allows us to Biggers 10

understand how individual initiatives are likely to impact the campaign, before the campaign is finished. In essence, we need a measurement that reflects the exact process in which we theorize that propositions increase turnout, and doing so likely requires us to focus on the issue content of the propositions. Given this focus, one measurement of initiatives appears to fit the bill. Why social issues? Ballot propositions that address social issues may be ones that are particularly salient. Such issues include, among others, policies towards abortion, the death penalty, euthanasia, stem cell research, gun control, gay marriage or rights, and civil rights. Many of these issues fit into a framework of morality politics, which involves policies that either attempt to regulate social norms or generate a strong moral response from citizens for some reason (Mooney and Lee 1995). This notion is somewhat akin to Carmines and Stimson s (1980) concept of easy issues, in that they trigger a gut response from the individual and do not require a heightened level of sophistication. Such issues may be particularly salient because they tap core values that reflect deeply held beliefs (Carmines and Stimson 1980) and are often seen as more meaningful to citizens than other, more complex issues (Mooney 2001). Scholarly research suggests that social issue initiatives are highly salient. In his investigation of California ballot propositions from 1956 to 2000, Nicholson (2003) finds that eighty percent or more of respondents claimed to be familiar with initiatives that dealt with issues such as the death penalty and homosexuality (both social issues). In his statistical analysis, the scholar confirms that citizens are in fact more likely to be aware of those initiatives dealing with morality (by thirteen percent) or civil liberties and rights Biggers 11

issues (by eighteen percent). In addition, social issues are consistently the most cited by respondents when asked open-ended questions about which initiative issues are on the ballot, such as in a mid October 2004 Pew Foundation national survey (cited in Donovan et al. 2005). Social issues often appeal to an individual s passions, rational reasons, and emotions. Ballot initiative campaigns over controversial issues, such as gay marriage or abortion, are rarely presented in dispassionate or unemotional terms. Instead, these campaigns are often emotionally rich and provoke diverse responses (Tolbert 2005). Mooney (1999) argues that on many of these issues a sizeable portion of the population holds a strong and unwavering position on such issues, and Magleby (1984) notes that aggregate opinion with regards to such issues when debated as ballot measures shifts very little over the course of the campaign. Not only do many citizens hold strong, unwavering opinions, but they may be more willing to make their voices heard on these issues (Mooney and Lee 1995). The fact that many citizens see their basic values as being threatened, combined with the absence of information barriers that often limit their participation, may provide a greater incentive for them to become involved (Mooney 2001). Because social issues are often symbolic in nature and presented in a simplistic manner, the average citizen requires little knowledge or information to be able to either assert their opinion or participate to some degree in the decision making process (Mooney 1999). This combination of technical simplicity and high salience may lead citizens to be both more likely to participate (Mooney 1999) and facilitate this participation (Mooney and Lee 1995). Biggers 12

Because the nature of social issues mean that they are often highly salient, their campaigns are likely to garner significant media attention and lead to highly contested campaigns, which in turn will decrease the costs of political information. Such campaigns may also generate organized interests on both sides of the issue that can contact potential voters (Bowler et al. 1998). Thus, it is plausible that social issue initiatives, which are often highly salient and appeal to people s passions and emotions, would have a significant impact on interest in the campaign. This interest may translate into an increase in the probability of voting, at least during midterm elections (Tolbert and Bowen 2008). In addition, the strong feelings described above may lead to increased mobilization attempts by those on both sides of the issue, as has been documented with regards to abortion initiatives (Roh and Haider-Markel 2003), leading to higher awareness. Given the highly salient nature of other issues, such as gay marriage (Barclay and Fisher 2003), such attempts may extend to other social issues as well. Regardless of who places the issue on the ballot, such issues often generate their own campaigns with organizations capable of contacting potential voters (Donovan et al. 1998). All of these factors are likely to combine to facilitate and increase turnout. The impact of social issue ballot measures, most notably proposed gay marriage bans, has received significant attention. One need only go back to 2004, when eleven states adopted gay marriage bans, to find an unresolved debate about the effect that such initiatives had on turnout. In the immediate aftermath of the campaigns, some scholars found no evidence that gay marriage ballot measures had a discernible impact on voter turnout (Abramowitz 2004; Burden 2004; Smith et al. 2006). Other scholars suggest that gay marriage bans stimulated turnout only in some states (Donovan et al. 2005) or states Biggers 13

that were not battleground states (McDonald 2004). Campbell and Monson (2008) conclude that gay marriage ballot measures mobilized white evangelical Protestants to show up at the polls but demobilized secularists from showing up to vote, acting at crosspurposes in affecting aggregate turnout. What we should take away from this conversation is that some scholars identified such ballot measures as being able to increase voter turnout even in the atmosphere of high information, mobilization, and salience that characterizes presidential elections. The questions raised about the effects of gay marriage propositions in 2004 largely centered on the mobilization of evangelicals. Scholars have noted that the mobilization of evangelicals is often maximized when social issues are either on the ballot or major factors of the political campaign (Wilcox 2006). Even initiatives that appear to have little chance of passing can mobilize such voters to show up to the polls (Witt and McCrokle 1997). Because such individuals have traditionally been less likely to show up to the polls on Election Day (Wilcox 2006), maximizing their mobilization will likely increase turnout. While such issues may serve to mobilize evangelicals, scholars have noted that such mobilization campaigns are susceptible to countermobilization attempts (Haider-Markel and Meier 1996; Wilcox 2006), which would have an even greater impact on swelling the voter ranks. Grummel (2008), using a dummy variable to distinguish whether or not there is a morality issue on the ballot, finds that turnout is higher when there is a morality issue 1 initiative on the ballot in midterm but not presidential elections. The scholar suggests that turnout might be as much as eight percentage points higher when a state has a 1 Grummel defines morality issues to include gay marriage, gay rights, stem cell research, abortion, and euthanasia. Biggers 14

morality issue on the ballot. Grummel, however, does not allow for multiple such issues on the same ballot to have a cumulative effect. In essence, the scholar is stating that having both a gay marriage and an abortion ban on the ballot will have the same mobilizing effect as having just a gay marriage ban. While the two groups that would be motivated by such bans are likely similar, they are probably not exactly the same. In short, Grummel s (2008) research points us in the right direction but provides avenues upon which to expand. I contend that social issues propositions likely have the ability to stimulate turnout in midterm elections, but their potential to increase turnout will be drowned out by the high information, mobilization, and salience atmosphere that dominates presidential elections. I further contend that the total number of propositions will not be associated with an increase in turnout in either type of election. More formally, I hypothesize that: (1) An increase in the number of social issue propositions on the ballot will not increase turnout in presidential elections. (2) An increase in the number of social issue propositions on the ballot will increase turnout in midterm elections. (3) An increase in the number of total propositions on the ballot will not increase turnout in either presidential or midterm elections. Data and Methods To investigate these hypotheses, I examine the previous five national elections for which there is individual level data readily available: the presidential elections of 2000 and 2004, and the midterm elections of 1998, 2002, and 2006. In examining the hypothesized relationship between initiatives and turnout, previous individual level Biggers 15

analyses have employed either the NES or similar data (but see Scholzman and Yohai 2008; Tolbert et al. 2009). These data sources suffer from the fact that they are comprised of a relatively small number of respondents in total and a very small number of respondents from some sparsely populated states. This has led scholars to draw conclusions about states that make heavy use of the initiative process but have small populations, such as South Dakota, from only a few individuals (Scholzman and Yohai 2008). Sometimes such states are not even represented, as is the case with the 2004 NES, which contains no individuals surveyed from South Dakota. In addition, these data sets, while they randomly sample at the national level, are not representative at the state level. Thus, while we are explicitly concerned about the impact of initiatives in each state, scholars have worked with data that is not comprised of a random sample from each state. To address these data concerns, I employ the Current Population Study (CPS) November Supplements. The CPS November Supplement is particularly well suited to answer questions about turnout (Wolfinger and Rosenstone 1980) because it helps to counteract the problems with employing data sets such as the NES: it has an extremely large sample size (at least 70,000 citizens in each sample who responded as to whether or not the voted) and it randomly samples at both the state and national level. In contrast to these other data sets, however, the CPS, while containing information on important demographic information for each individual, is not exhaustive in providing information on the factors that are known to affect turnout. Such factors missing from the CPS include strength of partisanship, attention to campaigns, and media consumption habits. Biggers 16

While the absence of such measures is certainly unfortunate, the use of the large and representative CPS data set outweighs the drawback of not having such variables. 2 The large sample for each year is also crucial in accounting for the clustered (or multilevel) nature of the data. Clustering arises when we have data that has been compiled from multiple levels. In this case, the data set is hierarchical in nature due to the fact that the lower level units of analysis, individuals (i), are clearly nested within the higher level units, the states (j) (Jones 2007). Clustering arises from the fact that attributes associated with the states in which individuals reside do not vary across individuals within each group (Primo et al. 2007). So, for example, every respondent who lives in Colorado will have the same minority party, the same voter registration law, and, most importantly, the same right to place citizen legislation on the ballot (and will be exposed to the same initiatives). Such state attributes are either unique to Colorado or at least are not shared by residents of every other state in the country. Clustering is of concern to scholars for a number of reasons, but the most pressing is that the failure to account for the multilevel nature of the data leads to misestimated standard errors. By ignoring the clustered nature of the data, scholars fail to take into account the dependence among individual responses within the same state. In other words, because these attributes are not randomly assigned to states, and the states that individuals reside in determine their exposure to such attributes, there is a lack of independence among the errors. This lack of independence violates the independent and identically distributed assumption of OLS regression (Primo et al. 2007). In most cases, the correlation between the errors is positive, which will cause the estimated standard errors to be too low, leading us to calculate t-statistics that are too high. Doing so leads 2 All noncitizens and those under the age of eighteen surveyed in the CPS were excluded from the analysis. Biggers 17

to inflated Type 1 error rates, in which predictors appear to have a significant effect on the dependent variable when in reality they do not. Relating this to our situation, the failure to account for clustering may lead us to conclude that initiatives have a positive impact on turnout when this is not in fact the true relationship. The clustered nature of the data necessary to investigate the relationship in question has routinely been ignored by scholars (but see Scholzman and Yohai 2008; Tolbert et al. 2009). To account for the multilevel nature of the data, I model the multilevel nature of the data employing a form of hierarchal linear modeling. Because individual level data is employed, the dependent variable is the respondent s self-report of whether or not they voted in the election. 3 The presence of a dichotomous dependent variable requires the use of hierarchical generalized linear modeling. As individual level (or level-one) independent variables, I include demographic characteristics that are collected by the CPS and known to influence one s propensity to vote. These include the individual s age measured in years, as well as their age squared; income and education measured as ordinal variables; residential mobility (Squire et al. 1987); and dummies for male, African-American, and Hispanic. 4 As state level (or level-two) independent variables, I include the variable of interest, initiatives, which is a count of the number of social issue ballot propositions that 3 It is well known that self-reported turnout rates are habitually higher than actual turnout rates. Reported turnout rates from the CPS are as follows: 50.9% for the 1998 CPS, as opposed to the actual turnout rate of 39.3%; 66.4% for the 2000 CPS, as opposed to the actual turnout rate of 55.3%; 51.5% for the 2002 CPS, as opposed to the actual turnout rate of 40.5%; 71.6% for the 2004 CPS, as opposed to the actual turnout rate of 60.7%; 56.2% for the 2006 CPS, as opposed to the actual turnout rate of 41.3%. While the overall reported rate is certainly inflated, it is (except in 2006) less than the over reported rates associated with the NES over a similar period of 14-18% (Duff et al. 2007). 4 See Appendix A?? p.?? for a complete description of the question wording and coding of each variable. Biggers 18

appear on the state s ballot. 5 Many scholars have only focused on ballot initiatives, or those measures put on the ballot by citizens, when investigating the hypothesized relationship. Doing so, however, ignores a large number of measures that appear on the ballot and through which citizens can legislate from the polls, such as legislation placed on the ballot by the legislature. While initiatives clearly differ from other means of direct democracy, most notably in the fact that they allow citizens to actually draft the legislation, both types of propositions permit the direct participation by citizens in the legislation making process. This difference may undeniably affect both what issues are placed on the ballot and the legislative parameters of these issues. It is unclear, however, why a measure would be more likely to mobilize citizens simply because it was placed on the ballot by a citizen (or interest group), as opposed to, say, the state legislature. This is especially true if we accept that, as discussed above, it is not the actual initiative process itself that works to increase turnout. For example, referendums, like initiatives, allow ordinary citizens to take control of the agenda (Matsuaka 2005:187). With regards to legislative referendums, all but one state requires amendments to the state constitution to be put to a popular vote (Matsuaka 2005). Thus, including only initiatives in the measure captures only a part of the direct democratic action that is taking place during an election. Because it is likely that the average citizen does not know the difference, it is unlikely that such a difference would be meaningful with regards to the hypothesized relationship. While there might be somewhat of a tendency for there to be more of a campaign around citizen induced initiatives, there is nothing to stop groups from organizing around other types of ballot 5 See Appendix B p.?? for a complete list of all social issue ballot propositions; issues include [where are they compiled from?] Biggers 19

measures. Likewise, an initiative, just like a legislative referendum, will have difficulty stimulating turnout, regardless of the support behind it, if the average citizen simply is not concerned about its issue content. Therefore, I contend that the type of direct democracy makes little difference in terms of motivating the average citizen to vote. I code the number of total ballot propositions on an individual s ballot in a similar manner, as well as those that deal with environmental and tax issues for comparison. 6 Other state level variables are also employed in the model: dummies for the presence of senatorial or gubernatorial races are included, as their presence might serve to stimulate turnout; a variable for the number of days between the closing date to register to vote and the actual election is used; and a dummy for southern states. Following the common practice in the literature, I make a distinction between midterm and presidential elections in my analyses. account, is: The resulting level-one HGLM equation, taking the individual level variables into 2 Pr( voting) = β 0 j[ i] + β1 j[ i] ( age) i + β 2 j[ i] ( age ) i + β 3 j[ i] ( education) i + β 4 j[ i[ ( income) + Because the intercept is treated as a function of the level-two factors, the leveltwo model is: β 5 j [ i] ( black) + β 6 j[ i] ( Hispanic) + β 7 j[ i] ( mobility) + β8 j[ i] ( male) + e β 0 j[ i] = γ 00 + γ 01( socialmeasures) + γ 02 ( senate) + γ 03 ( governor) + γ 04 ( registration) + γ 05 ( south) + u0 j This means that the full multilevel model, employing a logit link function, is: 6 To compile these lists, I relied on the classification of each initiative by issue concern carried out by the and the Initiative and Referendum Institute. While the Initiative and Referendum Institute is a pro-initiative organization, there is no reason to believe that their classification of the subject matter of the initiative is in any way biased. Biggers 20

Pr( voting) = Pr( yi ) = log[pr( yi ) /1 Pr( yi )] = γ 00 + γ 01( socialmeasures) + γ 02 ( senate) + 2 γ 03 ( governor) + γ 04 ( registration) + γ 05 ( south) + γ 10 ( age) + γ 20 ( age ) + γ 30 ( education) + γ 40 ( income) + γ 50 ( male) + γ 60 ( black) + γ 70 ( Hispanic) + γ 80 ( mobility) + u0 j Results and discussion Table 1 presents the results of HGLM for the presidential elections of 2000 and 2004 using the social issue measure of direct democracy. The results are mixed. For the 2000 election, the relationship between the number of social issue measures and turnout is actually negative, but the relationship is not statistically significant. This supports the contention in hypothesis one, that there is no relationship between social issue measures and turnout in presidential elections. In 2004, however, there is a positive effect on turnout associated with an increase in the number of social issue ballot measures. This finding goes against hypothesis one and suggests that social issue measures may be able to stimulate turnout even in the highly salient atmosphere of presidential elections. In both models, increases in age, education, income, residential immobility, and being African American are all associated with an increase in the likelihood of voting, while age squared and being male or Hispanic, as well as an increase in the closing date, are all associated with a decrease in the propensity to vote. 7 [Insert Table 1 About Here] There is more concrete evidence for the hypothesized relationship between the total number of propositions and turnout. Table 2 presents the results for the presidential elections of 2000 and 2004, using a count of the number of propositions on the ballot. The results demonstrate the lack of a relationship between this measure of direct 7 See Table C1 in appendix C to compare results from a HGLM model to the same model run in logit. Biggers 21

democracy and the propensity of turning out to vote. In 2004, the relationship is actually negative, though it is not statistically significant. 8 With regards to the rest of the model, the same level one and two variables are significant as in the models discussed above: an increase in age, education, income, or residential immobility, or being African American, is associated with an increased probability of voting, while age squared, being male, being Hispanic, and an increase in the closing date are all associated with a decrease in the likelihood of turning out on Election Day. These findings mirror the majority of those found by other scholars that the initiative process has no positive impact on turnout during presidential elections. As mentioned earlier, the factors that are present in significant quantities during presidential campaigns, such as information and partisan and nonpartisan mobilization, likely serve to trump any mobilizing ability of the propositions on the ballot. 9 [Insert Table 2 About Here] For the midterm elections examined in this study, however, there is a consistent, positive relationship between social issue measures and turnout. Table 3 presents the HGLM results for the midterm elections of 1998, 2002, and 2006 from models that include the social issue proposition measure. In all three models, the coefficients for the variable are positive and significant, indicating that an increase in the number of social issue propositions on the ballot is associated with an increase in an individual s propensity to vote. With regards to the level one variables in the model, an increase in 8 This finding contrasts with Tolbert et al. (2009), who control for the multilevel nature of the data and find that an increase in the number of initiatives on the ballot had a positive impact on turnout in 2004. Scholzman and Yohai (2008), also dealing with the clustered nature of the data, find no increase in turnout in 2004 associated with an increase in the number of initiatives on the ballot. 9 See Table C2 in appendix C to compare results from a HGLM model to the same model run in logit. The table reveals that the failure to account for the clustered nature of the data would have led us to incorrectly conclude that an increase in the number of total ballot propositions is associated with a decrease in turnout in 2004. Biggers 22

one s age, income, education, or residential immobility are all associated with an increase in the likelihood of showing up to the polls, as is being African American. Being Hispanic, on the other hand, is associated with a decrease in voting in 2002 and 2006, as is an increase in age squared in all three models. For the level two variables, an increase in the number of days between the voter registration deadline and the election is associated with a decrease in voting in 1998 and 2002. Having a senate race on the ballot is associated with an increase in voting in 2002, while residing in a southern state is associated with a decrease likelihood of voting in 1998 and 2006. None of the other level two variables in the three models were associated with a change in the likelihood of turning out. The three models thus provide substantial evidence for hypothesis number two, which states that an increase in the number of social issue propositions is associated with an increase in the likelihood of voting in midterm elections. 10 [Insert Table 3 About Here] With regards to the relationship between the total number of ballot measures and turnout, the results are mixed. Table 4 reveals that there is an increase in the likelihood of voting associated with an increase in the number of propositions on the ballot in 2002, 11 but the relationship does not hold for the elections of 1998 and 2006. 12 Like the model for social issue measures, increases in age, income, education, residential immobility, and being African American are all associated with an increase in the propensity to vote for all three elections. An increase in age squared is associated with a 10 See Table C3 in appendix C to compare results from a HGLM model to the same model run in logit. 11 This finding mirrors that of Tolbert et al. (2009), who use the number of initiatives as their measure. Scholzman and Yohai (2008), however, find no relationship between initiatives and turnout in 2002, using the same measure of direct democracy as Tolber et al. (2009). 12 Both Scholzman and Yohai (2008) and Tolbert et al. (2009) find an increase in turnout in 2006 associated with an increase in the number of initiatives on the ballot. Biggers 23

decrease an individual s likelihood of showing up on Election Day in all three elections, while being Hispanic is associated with decreased turnout in 2002 and 2006. For the level two variables, living in a southern state is associated with a decrease in the likelihood of voting in all three elections. An increase in the closing date is associated with a decrease in voting in 2002, and the presence of a Senate race is associated with an increase in the likelihood of voting in 2002. There is thus supporting, but not unequivocal, evidence to support the second part of hypothesis three, that an increase in the number of total ballot propositions is not associated with an increase in the likelihood of voting in midterm elections. 13 [Insert Table 4 About Here] As discussed earlier, the number of tax and environmental propositions on the ballot were also employed as the direct democracy measure in the models presented above. The results (not shown) demonstrate a lack of uniformity in the effect of such propositions on turnout. While an increase in the number of environmental propositions is associated with an increase in voting in 1998, 2002, and 2004, there is no relationship between the two in 2000 or 2006. For tax measures, there is no significant relationship for any of the presidential or midterm elections under investigation. Such findings support the contention that we should be concerned with the issue content of the ballot measure, while also revealing that social issue propositions appear to be the most likely to consistently mobilize voters. Thus, while it appears that ballot measures that deal with other types of issues besides social issues may be able to increase turnout, the other issue 13 See Table C4 in appendix C to compare results from a HGLM model to the same model run in logit. The table reveals that the failure to account for the clustered nature of the data would have led us to incorrectly concluded that an increase in the number of total ballot propositions is associated with an increase in turnout in 1998 and 2006. Biggers 24

areas fail to increase turnout either uniformly across midterm elections (as social issue measures do) or have a positive impact of turnout in as many total elections. The impact of social issue ballot measures on turnout is more clearly demonstrated by calculating the predicted probability of an individual showing up to vote on Election Day as their exposure to these propositions increase. Figure 1-A demonstrates the changes in predicted probability for an individual voting in 1998 as there is an increase in the number of social issue measures on the ballot. On average, the presence of each additional social issue on the ballot increases the likelihood of voting by 2.75%. Moving from having no social issue measures on the ballot to the maximum number that were present in 1998, four, increases a citizen s propensity to vote from 50% to 61.4%, holding all other variables constant at their actual values. This is an increase of 11.4% in the likelihood of voting. The increase in the propensity for an individual voting in 2002 associated with an increase in the number of social issue measures is illustrated in Figure 1-B. In 2002, the addition of a social issue measure on a citizen s ballot increases their likelihood of voting, on average, by 2.7%. This impact is very similar to that in the 1998 election. A change in the number of social issue propositions from the minimum (zero) to the maximum (two) found on ballots in 2002 increases the likelihood of an individual turning out to vote from 53.4% to 58.8%, holding all other variables constant at their actual values. Figure 1-C presents the change in the predicted probability of an individual voting associated with an increase in the number of social issue measures on the ballot for the 2004 presidential election. On average, the increase in the likelihood of voting associated with the presence of each additional social issue measure on the ballot is 2.3%. Holding Biggers 25

all other variables constant at their means, a citizen exposed to no social issue measures has a propensity of voting of 72%, while an individual exposed to two such measures, the maximum number that were on a ballot in 2004, have a likelihood of voting of 76.6%, a difference of 4.3%. For 2006, the change in the predicted probability of an individual voting based on the number of social issue measures is reported in Figure 1-C. On average, the placement of an additional social issue proposition on the ballot increases the propensity to vote by 2.5%. Increasing the number of social issue measures on an individual s ballot from zero to three (the maximum number that was present on a ballot in 2006) changes a citizen s propensity to vote from 55.6% to 63.1%, holding all of the other variables constant at their actual values. This translates into an increase in the likelihood of turning out of 7.5%. Contrary to hypothesis three, there is a positive relationship between the total number of measures on the ballot and the likelihood of voting in 2002. While the relationship is statistically significant, looking at the changes in predicted probabilities reveals that the increase is not substantively significant. On average, each additional measure on the ballot increases the likelihood of voting by 0.43%. Moving from zero measures to the maximum number present on a ballot in 2002, fourteen, increase the propensity to vote from 52.1% to 58.1%, or 6%. Thus, while there is an increase in turnout associated with an increase in the number of total measures on the ballot, the effect is more than five times smaller than the smallest effect discussed above that is attributed to social issue measures. Conclusion Biggers 26

Stressing the importance of awareness, this study finds that social issue ballot measures have a significant impact on increasing the likelihood of turning out for national elections. This is true for both midterm elections and some presidential elections, which are often already high salience events. Such issues, because they appeal to the emotions of citizens, are often of extreme importance to individuals, and are not technically complex, are in a prime position to stimulate turnout in every election. In addition, such measures are likely to facilitate turnout by generating highly contested campaigns and extensive media coverage. While other issues may be able to mobilize turnout in some elections, both theoretical and empirical evidence suggests that social issue propositions habitually possess this potential. This study makes a further contribution to the existing literature by looking at all of the propositions on the ballot, instead of just initiatives (as other studies have done) In doing so, the measure employed in this study captures the entire direct democracy experience, as opposed to only a part of the process. When we consider all of the measures on the ballot, the results demonstrate that an increase in the total number of propositions is only rarely associated with an increase in turnout during midterm elections, and not at all during presidential elections for the period under investigation. These findings contradict much of the empirical evidence presented in this field of research and suggest that the way we measure direct democracy is of crucial importance to understanding its true effects on turnout. The measures we employ warrant further theoretical and empirical justification. This new focus on awareness in general, and on the issue content of individual ballot measures in particular, provides numerous avenues upon which to expand. Besides Biggers 27