Reassessing Direct Democracy and Civic Engagement: A Panel Study of the 2008 Election

Reassessing Direct Democracy and Civic Engagement: A Panel Study of the 2008 Election Daniel A. Smith University of Florida Caroline J. Tolbert University of Iowa Amanda Keller University of Iowa Abstract Recent studies have found that ballot initiatives have an educative effect, increasing individuals political interest, knowledge, efficacy, and turnout, although some of these findings have recently come under fire. In this paper we test if there are individual-level educative effects of ballot measures over the course of a single campaign season. We theorize that the existence of state-specific issue publics may be the causal mechanism underlying the educative process of direct democracy during an election year. We overcome previous data limitations by drawing on a six-wave panel survey of 20,000 respondents where the same individuals are re-interviewed over the course of the campaign. The large sample size allows for sufficient cases within states for multilevel analysis. In addition, a unique survey question designed by the authors asks respondents how interested they were in their statewide ballot measures. We model changes in levels of political interest and political participation over the course of the nine-month election (from December 2007-Octber 2008) and in the last month before the general election (September-October 2008) as a function of several state contextual factors, in order to isolate the effects state ballot measures and the amount spent on initiatives in 2008. We find that direct exposure to ballot initiatives, which creates state-specific issue publics, leads to an increase in an individual s political interest and participation levels over the course of a campaign season. Paper presented at the Annual Meeting of the American Political Science Association Meeting, September 3-6, 2010, Washington, DC

1 Reassessing Direct Democracy and Civic Engagement: A Panel Study of the 2008 Election Might direct exposure to ballot measures increase individuals general political interest and political participation? Is it possible that living in a state with initiatives on the ballot can change a person s political interest and behavior during a single campaign season? If so, this would constitute clear evidence that ballot measure campaigns have secondary effects, even over the short term. Over the past decade, numerous studies using both individual and aggregate level data have shown that ballot initiatives may have an educative effect (Smith and Tolbert 2004). These studies find that being exposed to questions on the statewide ballot can increase political interest, political knowledge, political efficacy, and perhaps most importantly, turnout and candidate vote choice (Smith 2001; Smith and Tolbert 2004, 2010; Tolbert, Bowen and Donovan 2009; Bowler and Donovan 2002; Mendelsohn and Cutler 2000; Donovan, Tolbert, and Smith 2008, 2009; Tolbert, Grummell and Smith, 2001; Benz and Stutzer 2004; Tolbert and Smith 2005; Nicholson 2005; Campbell and Monson 2008; Smith, DeSantis, and Kassel 2006; Bowler, Segura and Nicholson 2006; Cain, Donovan, and Tolbert 2008; Dyck and Seabrook 2010). Although the magnitude of some of these findings has recently come under attack (Schlozman and Yohai 2009; Dyck 2009; Dyck and Lascher 2009), the findings of these critical studies have been marginal, as most of the results have reinforced the educative effects paradigm. Yet, due to a lack of experimental data, a limitation of all the research examining the broad effects of direct democracy on political behavior is that it does not measure individuals propensity to change their attitudes or behavior over the course of an election. Scholars studying the policy effects of successful ballot measures (Marschall and Rubil 2005) have raised the possibility that the impact of direct democracy might be endogenous (Tolbert, Bowen and Donovan 2009). By extension, then, there is a legitimate concern that the educative effects of ballot measures too may be endogenous; that is, that individuals residing in states permitting ballot initiatives, a priori, have a greater affinity for politics and political participation due to cultural reasons related to the process of direct democracy itself. Furthermore, critics of the educative effects of direct democracy have raised concerns regarding ecological inference, that is, that shifts in individual political attitudes or participation levels across the

2 states with and without measures on the ballot may obfuscate differential changes occurring at the individual level that are unrelated to exposure to the mechanisms of direct democracy. In this paper, we offer several empirical tests to determine whether ballot initiatives have educative effects by measuring change in individuals political interest and participation where the same individuals are repeatedly interviewed over the course of the 2008 presidential campaign. In doing so, we are able to address lingering concerns over the possibility that these effects are endogenous. To date, scholars have conducted no research on the purported effects of ballot measures on political behavior using panel data. Previous research assessing the educative effects of ballot measures has leveraged several reputable cross-sectional surveys: American National Election Study (ANES), Pew, Current Population Surveys (CPS) conducted by the U.S. Census Bureau, and the 2006 Cooperative Comparative Election Study (CCES) or aggregate state data. None of these studies, however, have drawn on panel or experimental data, which can be used to track a particular individual s exposure to ballot measures (treatment) and associated change in levels of interest, civic engagement, and political participation, before and after an election season, compared to individuals not exposed to ballot measures (control group). As a result, studies both supportive and critical of the effects of direct democracy are methodologically based on non-experimental state-level inference. Because they do not use panel (or dynamic) survey data, these earlier studies are unable to precisely measure the campaign effects (if any) of ballot measures in changing citizens political attitudes and participation, and are thus open to criticisms that their findings are endogenous or plagued by ecological inference problems. We overcome previous data limitations and concerns over latent endogeneity and ecological inference by drawing on the 2008 Cooperative Comparative Analysis Project (CCAP), a six-wave panel survey of 20,000 respondents. In many ways, panel data may offer the definitive empirical test of the impact of ballot measures on political behavior at the individual level. Drawing on these data, we are able to model changes in individuals political interest and participation over the course of the ninemonth election (from December 2007-October 2008) and last month before the general election (September-October 2008) as a function of state contextual factors, specifically the number of statewide

3 initiatives on the ballot and the amount spent on initiatives in 2008. We use the panel data to establish a baseline, and then track over time the associated change in levels of political interest and political participation in the months preceding the general election. Because we draw on a natural experiment (states with and without initiatives on the ballot) and individual-level panel data, we are able to pinpoint how much (if at all) individuals actually altered their political interest and participation due to the presence or absence of concurrent ballot measure campaigns in 2008. Beyond the dynamic aspect, the large CCAP sample size allows for sufficient cases within states for multilevel analysis to determine whether in the short and long-term citizens in the two-dozen states permitting ballot initiatives increased their political interest and participation levels during the 2008 campaign. In addition, a unique survey question designed by the authors and embedded in the survey asks respondents how interested they are in their statewide ballot measures before the election (October). This allows us to tie our key exogeneous factors (exposure to and spending on ballot measures) directly to a specific individual s attention to the issues and general political participation. Direct Democracy and Change in Political Interest and Participation Our study builds theoretically on previous research on the educative effects of ballot measures (Smith and Tolbert 2004) and the ability of direct democracy campaigns to set the agenda in candidate races and prime vote choice (Nicholson 2005; Donovan, Tolbert, and Smith 2008). Since the Populist and Progressive Eras, scholars have theorized that ballot measures could increase voters interest and factual political knowledge by making them lawmakers for a day (Smith and Fridkin 2008). Citizen lawmakers, the argument goes, become more engaged in the politics because their voice is heard directly by lawmakers who might otherwise be captured by special interests or party bosses. As a result of their direct involvement in governing their state, citizen lawmakers in states permitting ballot measures have higher levels of political efficacy, specifically confidence in government responsiveness. More importantly, exposure to ballot measures leads to increased citizen interest in politics and participation levels. In short, being asked to vote on issues salient or otherwise can spur civic engagement, and

4 even political participation. Indeed, the most robust empirical finding of the educative effects of ballot measures is the positive impact of direct democracy on voter turnout. Scholars have found that the mere presence of statewide initiatives on the ballot increases turnout, especially in low profile, midterm elections (Smith 2001; Lacey 2005; Tolbert, Grummel, and Smith 2001), but also in higher profile presidential elections (Tolbert and Smith 2005; Tolbert, Bowen, and Donovan 2009) as well as in other cross-national contexts (Lassen 2005). But theoretically, what is the mechanism that causes ballot measures to enhance individuals interest in politics and participation? We argue that by offering citizen lawmakers binary choices some complex, others very simple on issues ranging from morality to economics, ballot measures can engage individuals who typically shun traditional candidate campaigns or who otherwise might not be interested in or want to participate in political life. As we know, not all citizens have the same propensity to be civically engaged in the public sphere. Above and beyond candidate races, statewide ballot issues can stimulate citizens, sparking their interest in politics more generally. Building on Zaller s (1992: 58) argument that the greater a person s level of cognitive engagement with an issue, the more likely he or she is to be exposed to and comprehend in a word, to receive political messages concerning that issue, we suggest that immediate exposure to statewide ballot measures encourages individuals to make top of the head binary choices. Because ballot issues confront citizens with two clear-cut options, ballot measure campaigns provide individuals with immediate and salient information that has the potential of augmenting overall levels of political interest and participation. This might especially be true when there is a large amount of money spent on ballot measures, and associated media attention and communication, in a state. The campaign effects of ballot measures may be similar to residing in a battleground state in the presidential election (Shaw 2006; Panagopoulos 2009). Although individuals often do not have in-depth or encyclopedic knowledge when deciding how to vote on statewide issues (Lupia 1994; Bowler and Donovan 1998), the attention given to ballot measures during the campaign season by elites and the media may nevertheless shape the electoral environment by piquing citizen s interest, leading to higher levels of participation (Donovan, Tolbert, and

5 Smith 2009). Indeed, due to the stark contrast between a yes and a no vote on a ballot measure (Lupia 2001) there is no iterative opportunity for compromise or negotiation the inherently controversial nature of ballot measures may effectively dampen accountability pressures that typically cross-cut social and political networks (Mutz 2002). In addition, the increasingly strong partisan and elite cues tied to ballot issues (Nicholson 2005; Smith and Tolbert 2001) may create what amounts to an echo chamber (Key 1966); as elites align on either side of a polarizing ballot issue, the mass public s engagement is likely to become more intense, leading to higher levels of political interest and participation. It is important to note that the particular timing and sequence of ballot measure campaigns may also be important. Because qualification deadlines for statewide measures are typically only 90 days before a general election, statewide ballot measure campaigns often begin later in the campaign season than presidential or competitive congressional races. Most of the intense spending and media coverage occurs in the month immediately preceding election day (Smith 2010). If Zaller s Reception, Attention, Stimulus model is correct (1992: 42-45), and information that is the most salient is that which is most recently received by individuals, then issue elections may be particularly salient in generating general attention to politics and participation. That is, statewide ballot measures may be as likely as other salient campaign topics covered by the media to be at the top of an individual s head, in Zaller s parlance. This argument is consistent with research finding that intense media campaigns in battleground states lead to increased political interest and turnout, even among those least likely to participate such as the poor (Gimpel, et al. 2007), as well as with research showing that more spending in ballot initiative campaigns can lead to higher turnout of lower-educated voters (Tolbert, Bowen and Donovan 2009). It is also consistent with studies finding that negative campaign ads can generate information that augment political interest and knowledge (Freedman, Franz, and Goldstein 2004), heighten political learning, and influence how people think about broader issues and candidates in part due to an emotional response to the issues (Redlawsk, Civettini, and Emmerson 2010). A ballot measure, thus, can elicit powerful

6 emotional cues, especially for those citizens who have minimal information about the issue, which can lead to greater political interest and participation levels. As such, judging by the widespread interest in issues on the statewide ballot roughly three out of four Americans in states permitting direct democracy claim they are somewhat or very interested in ballot measures (Donovan, Tolbert, and Smith 2009) as well as by how much news coverage is devoted to ballot measures by the media (Smith 2001; Donovan, Tolbert, and Smith 2008), we argue that the campaign effects of ballot measures likely generate considerable general interest in politics and political participation. When there is high issue salience and considerable spending on ballot issues, the campaign effects of ballot measures may even be comparable to the heat of an intense, high motivation presidential campaign. If ballot measures have pronounced campaign effects, then, it provides further evidence that campaign effects (Holbrook 1996; Brady and Johnston 2006) are even more contingent than previously thought. Not only do citizens who are exposed to competitive candidate elections have increased likelihood of being interested and participating in politics, those who are exposed to ballot initiatives might likely have increased civic engagement and participation. Going beyond Smith and Tolbert (2010), who extrapolate from Converse (1964) and suggest that statewide ballot measures can stimulate the creation of state-specific issue publics, we aim to measure the educative effects of ballot measures on individual political behavior over the course of a campaign season. Examining the priming effect of minimum wage increases on the statewide ballot of six states in 2006, the authors argue minimum wage state-specific issue publics emerged, which led to perceptible shifts in public opinion on the specific issue of the minimum wage and the broader issue of the economy among those living in the states with the initiatives. Controlling for other factors, Smith and Tolbert (2010) find that Democrats in the six states who were directly exposed to the ballot initiative campaigns were more likely to say they supported raising the minimum wage, as well as say that the economy was the most important issue when casting a ballot, compared to fellow Democrats who did not have the issue on their statewide ballot. Although we do not have panel survey data,

7 Smith and Tolbert (2010: 105) admit, this suggests that exposure to minimum-wage ballot measure campaigns modified support for the policy among partisan subsamples. The creation of state-specific issue publics may be the causal mechanism underlying the educative effects of ballot measures on individuals who are exposed to issue campaigns. The process and politics of direct democracy may have contingent campaign effects across the American states, with ballot issue campaigns having the potential to shape citizens attitudes and behavior who are directly exposed to them during an election season. Since policy preferences for many citizens are often nonattitudes, weakly held, or even contradictory (Converse 1964), they can be susceptible to manipulation by political operatives during the course of a statewide issue campaign. Indeed, Converse finds that only a small minority of the American public possess clear, coherent, ideologically-based ideas on political issues, and that very few individuals possess fixed ideological preferences. Like the contingency of public opinion, the exposure to ballot measures may also shape individuals interest in politics and create incentives to participate that are otherwise absent in other state contexts during an election year. As such, just as public opinion across the American states might be unstable in the short term due to the presence and effects of ballot issues, we argue that the information environment created by numerous or expensive ballot measure campaigns even late in the campaign season, such as in October prior a November general election may have significant effects on an individual s level of political interest and participation. Research Methods: Using Panel Survey Data In assessing the educative effects of ballot measures, scholars have leveraged several reputable cross-sectional national survey (ANES), Pew surveys, CPS, and the CCES, as discussed above. Yet there is no research on the effects of ballot measures using panel data. Panel surveys interviews the same individuals repeatedly over the course of the election. Panel data is a within groups experiment, where the individual at wave one is the control group compared to the same individual at wave two. While it is not necessary to use statistical controls with experimental research, such controls are employed here to

8 provide a more rigorous statistical test. With the panel survey we focus on measuring who changed their civic engagement and participation over time. We overcome previous data limitations by drawing on the 2008 Cooperative Comparative Analysis Project (CCAP) six-wave panel survey of 20,000 respondents (Jackman and Vavreck 2009). The same individuals answered questions throughout the presidential campaign, allowing us to measure what was causing changes in their interest or participation levels. The CCAP has a common battery of questions comparable to the ANES or General Social Survey (GSS), but with sample sizes many times larger than these other omnibus surveys. The large sample size allows for sufficient cases within states for multilevel analysis to test how individual political behavior responds to change in their state political environments. 1 The national survey sample includes 14,000 of which completed all six waves stratified by battleground and non-battleground states. 2 An initial baseline survey was conducted in December 2007 followed by four other pre-election waves conducted in January, March, September and October. Our key question is whether exposure to ballot measures increased citizens political interest and participation in the election, over and above the effects of residing in a battleground state or exposure to competitive congressional races. Change in political interest and political participation over the course of the nine-month election (from December 07-Ocotber 08) and in the last month before the general election (September to October 08) are modeled as a function of state contextual factors, such as competition in presidential and congressional races in the respondent s states, the numbers of initiatives on state ballots, 1 The Cooperative Campaign Analysis Project (CCAP) (Jackman and Vavreck 2009) is a collaborative effort that brings together over 60 political scientists from 25 institutions to produce a six wave panel study conducted on the internet. This sample is constructed using a technique called sample matching (Vavreck and Rivers 2008). The researchers create a list of all U.S. citizens from the U.S. Census to generate a set of demographic, political and behavioral characteristics that should be mirrored in the survey sample. Then, using a matching algorithm, the researchers select respondents who most closely resemble the Census data from a pool of opt-in participants. The sample is stratified to ensure large samples within states. See Jackman and Vavreck 2010 for a descriptive of the sample. More information regarding sample matching is available at http://web.mit.edu/polisci/portl/cces/material/sample_matching.pdf. The models are estimated using survey weights. Using this same technique, the 2006 Cooperative Congressional Election Survey (CCES) produced more precise estimates than more conventional probability designs such as random digit dialed (RDD) phone surveys (Vavreck and Rivers 2008). See also Malhotra and Krosnick 2007. 2 Nine states that makeup both battleground and early primaries (FL, IA, MN, NV, WI, NH, NM, OH, PA) were oversampled such that they are equal in population with the non-battleground.

9 and spending on ballot initiatives in 2008. 3 The Appendix provides detailed information for the coding of all variables in the analysis, and in which waves the questions were repeated. Hypotheses Drawing on the 2008 CCAP panel data, we are able to test several hypotheses concerning the attitudinal and participatory effects that ballot initiatives may have on an individual over the course of a single campaign. H1 Individuals residing in states with more initiatives on the ballot or higher spending on initiative campaigns will be more interested in politics and more likely to participate politically in the campaign in October 08, but not in December 07, before the presidential election began and before any initiatives qualified for state ballots. [Tested in Table 4 & Appendix Table] H2 Individuals residing in states with more initiatives on the ballot or higher spending on initiative campaigns will be more likely to change their level of political interest during the ninemonth presidential campaign (long-term), as well as in the last month before the general election (short-term). [Tested in Table 6] H3 Individuals residing in states with more initiatives on the ballot or higher spending on initiative campaigns will be more likely to change their level of political participation during the nine-month presidential campaign (long-term), as well as in the last month before the general election (short-term). [Tested in Table 7] H4 Individuals residing in states with more initiatives on the ballot or higher spending on initiative campaigns and who report they are interested in the measures on their state ballots will be more likely to express general interest in the election and more likely to participate politically. [Tested in Table 8] H5 Individuals residing in states with more initiatives on the ballot or higher spending on initiative campaigns and who are interested in the measures on their state ballots will be more likely to change their level of general political interest over the nine-month presidential campaign, as well as in the last month before the general election (short-term). [Tested in Table 9] Outcome Variables We analyze two outcome (or dependent) variables 1) political interest; and 2) political participation. We draw on panel data for both variables from the December 2007 baseline wave, the September 2008 wave, and the October 2008 wave. 3 The per capita state spending on initiatives in 2008 variable (Tolbert, Donovan, and Bowen, 2009), is actually a measure of per capita contributions to ballot issue committees in 2008.

10 Modeling Change in Political Interest Respondents were asked How interested are you in politics: very much, somewhat, not that much? Table 1 shows the frequency of responses to the political interest question in the October wave; roughly 50 percent of respondents were very interested in the election in the month before the presidential race, and 30 percent were somewhat interested. This is consistent with other national surveys. The remaining respondents said they were not interested in politics or skipped the question. Table 2a provides a cross-tabulation table of the percentage of respondents that changed their level of political interest from the baseline wave (December 2007) to October of 2008. Reading across the rows, among individuals very interested in the election in the baseline wave, 85 percent remained very interest in October, nine months later. These are politicos who always follow politics, regardless of state contextual factors. Of those individuals only somewhat interested in politics in the baseline wave (second row of the table), 17 percent became very interested by the eve of the general election (October). We also see large changes among those claiming to be not interested in politics. Of those saying they were not much interested in politics in the baseline wave, 28 percent became somewhat interested by October, and an additional 2 percent changed to very interested. Of those who said not sure in the baseline wave, more than one third (36 percent) were somewhat interested and an additional 7 percent very interested by October. Overall, 6 percent of the sample changed to very interested over the course of the nine-month campaign, although there were larger changes within categories. These panel data separated by nine months allow us to measure long-term change over an election campaign. Table 2b presents a similar pattern, but instead highlights change in political interest in the last month before the general election. Of those individuals very interested in the election in September, not surprisingly, almost 90 percent remained very interest one month later (October). Of those individuals who reported being somewhat interested in politics in September, 16 percent became very interested by October. Overall, roughly the same percentage of the full sample (6%) changed to high interest by the last month before the general election. These panel data, conducted two months apart, are ideally suited to measure short-term forces and political behavior.

11 Tables 3a and 3b present the same descriptive data, broken down for individuals from states with one initiative or less (none) on the ballot (3a), and individuals from states with more than one initiative on the ballot (3b). In states with one or no initiatives on the general election ballot, of those who responded they were somewhat interested in the election in September, fifteen percent became very much interested in the election one month later. This shift is less than the 18 percent of respondents who became very interested in states with more than one initiative on the ballot, a 3 percent difference. Among those not interested in September, 25 percent of those living in a state with one or less initiatives on the ballot were somewhat interested by October, compared to 32 percent of those living in states with more than one initiative on the ballot, a 7 percent difference. At the descriptive level, then, there is evidence of a differential effect of living in a state with multiple initiatives on the ballot. In what follows, we use either the baseline (December 2007) or the September 2008 wave as a lagged variable to predict levels of political interest in October 2008. To simplify the multivariate models, we focus on modeling high political interest in the October 2008 wave; respondents in all waves were coded 1 for very much interested and 0 for all other valid categories for the multivariate analysis. This provides a conservative test, as individuals also moved from not interested to somewhat interested as discussed above. Because the outcome variable (political interest) is binary, logistic regression models are used. Modeling Change in Political Participation The second dependent variable is an index of thirteen common offline participation actions (see Appendix for details) also asked in the baseline, September, and October waves. Respondents were asked, Thinking about the presidential candidates and their campaigns, did any of the following things happen to you yesterday? The political activities included: saw campaign ad on TV, received a piece of campaign mail, donated money to a candidate, received a pamphlet on my door, wore a button for a candidate, discussed a candidate with someone, received a visit from a campaign worker, heard a radio ad for a candidate, saw a yard sign for a candidate, went to hear a candidate speak, got a phone call from a campaign, and heard about a candidate at religious service. Respondents were coded

12 between zero and thirteen, depending on the number of traditional political activities in which they were engaged. Figure 1 provides a histogram of responses to this count variable in the baseline wave (December 2007) before the presidential primaries began. The majority of respondents had engaged in zero or one political activity. Figure 2 shows a histogram of the same political activities for the same individuals ninemonths later, on the eve of the general election (October). The modal response is five activities, and the graph now approaches a normal distribution. Omitting the zeros, Figure 3 shows the distribution for the change in the number of political activities from the baseline wave to October 2008, with 3 political activities the most common. Figure 4 shows this same distribution of political acts for individuals residing in states without an initiative on the ballot (left side) and for those residing in states with an initiative on the ballot (right side). The pattern shows individuals are more engaged politically when residing in states with at least one initiative on the ballot. Because the dependent variable measures a count of political acts, and doing one is associated with doing more, negative binomial regression is used for the multivariate analysis for this outcome variable. Mirroring the models measuring change in political interest, we use either the baseline measures or September wave as a lagged variable to predict October levels of political participation to model change. Each team of researchers participating in the 2008 CCAP panel survey also designed a module of questions for a sample of approximately 1000 respondents. Our module, which was approximately 20 minutes long, was administered in the final pre-election wave on October 22 through November 3rd. A unique survey question designed by the authors asks respondents how interested they were in their statewide ballot measures before the election (October). Individuals who said there were very interested in the measures on their state ballot were coded 1, and others 0. Similar findings reported here are found when using an ordinal coding for this variable. Interaction terms are used to model if an individual was interested in the measures on his/her state s ballot and actual exposure to ballot initiatives or spending on initiatives to predict general interest in the election and political participation. A conditional model is also used to predict change in levels of political interest and participation over the course of the campaign.

13 Controlling for Competitive Candidate Races Since this is a presidential election, we also consider the impact of state context on political engagement. In presidential races, residents of battleground states are smothered in attention from candidates and media, while citizens in non-battleground states barely get noticed (Panagopoulos 2009; Shaw 2006). As a result, research has found individuals residing in battleground states have higher interest and political participation, even those with low incomes (Gimpel et al 2007; see also Pacheco 2008; Lipsitz 2009; Cain, Donovan, and Tolbert 2008). Scholars have also shown competitive congressional races increase interest and participation (Kahn and Kenney 1999; Cox and Munger 1989). Three control variables measure the competitiveness of candidate races in the respondent s state, and proxy electoral campaigning, mobilization drives and media attention. These variables measure competition in the respondent s state in terms of the presidential race (1 minus presidential vote margin), US Senate race (1 minus Senate vote margin), and US House member (1 minus House vote margin) using a state or congressional district identifier merged with the survey data. These variables capture how competitive elections were in each state or congressional district. Alternative models [not shown] using campaign spending instead of vote margin produce similar findings. Because we have individuals residing in states, we cluster the standard errors by state in the analysis to control for these multilevel data. Other control variables include standard demographic and partisan controls (see Appendix for full list and coding). Results Table 4 presents the results for our first hypothesis. Confirming previous research based on crosssectional survey data (Smith and Tolbert 2004; Cain, Donovan, and Tolbert 2008), individuals exposed to more initiatives or increased spending on ballot measures campaigns are more interested in the election (October wave) and are more likely to participate in politics. The results for political interest are reported in columns 1 and 2, and for political participation in columns 3 and 4. When a logged or squared term for the number of initiatives on state ballots is used instead, the same positive coefficient is found (results not

14 shown). Consistent with the literature, both the number of initiatives and on state ballots, and initiative campaign spending leads to higher interest and participation (Tolbert, Bowen and Donovan 2009). We do not know, however, if exposure to ballot initiatives is what changed respondent s interest or participation; rather, we only know that direct democracy is positively associated with civic engagement. The Appendix presents identical models as reported in columns 2 and 4, using spending on ballot initiative campaigns per capita to predict interest and participation in the baseline wave (December 2007) before the presidential primaries began and when no initiatives had qualified for state ballots. The coefficient for spending on ballot initiatives in the respondent s state is not statistically significant. Since this is panel data, with the same individuals measured over time, we know something changed over the course of the presidential campaign. Table 5 presents results for our second hypothesis. A lagged term for political interest is included in the model measuring individuals interest in politics in the baseline wave or in September, thus modeling short or long-term change. The outcome variables in these models measure the factors associated with increasing political interest from low/somewhat to high over the nine-month course of the campaign (columns 1 and 2) or the last month before the general election from September to October (columns 3 and 4). Otherwise the covariates are identical to what is reported in Table 1. As expected, the lagged term is statistically significant across the four models in Table 2. Individuals who reported they were interested in politics in December 2007 are more likely to be interested at the height of the campaign. But in every model, more initiatives on the ballot or increased initiative spending per capita are associated with change in political interest, holding constant demographic factors, partisanship, and how competitive the candidate races were in the respondent s state. In fact, exposure to salient ballot measures was more important for increasing political interest than residing in a battleground state, as measured by the presidential vote margin. Table 6 presents simulations or predicted probabilities based on the coefficients reported in Table 5, holding other explanatory variables in the model at mean or modal values. The top row shows that for an individual residing in a state with zero initiatives on the statewide ballot, the probability of change in

15 political interest over the nine-month campaign from low/somewhat to high was almost 17 percent. But if two initiatives (mean) appeared on his/her state s ballot, or $2.62 dollars (mean) were spent per person in initiative campaigns in the state, this probably rose to 18 percent. If an individual lived in a state with 5 initiatives on ballot, or $7.03 spent per capita on initiative campaigns (+1 standard deviation from the mean), the probably increased to 19 percent. Residing in a state with the maximum number of statewide initiatives the 2008 ballot (10), or the maximum amount spent on initiatives ($19 per person), increased the probability of change to twenty percent. When omitting those who are always interested in politics, and those who never got engaged [anyone who didn t change], and focusing only on the individuals who changed from low/somewhat to high interest, we find that residing in a state using direct democracy was consistently associated with increased civic engagement in the long term, even if the change appears to be modest. The panel data provides robust evidence that ballot measures are associated with increased political interest. Table 6 also shows similar substantive effect of changing levels of political interest over the short term, in the month before the general election. One new finding for scholars is that salient issue elections may be able to engage voters even at the tail end of a political campaign. Table 7 reports models parallel to Table 5, but the outcome variable is change in political participation using the 13-point index shown in Figure 3. A lagged term for participation is included measuring individuals political activity in the baseline wave or in September. The outcome variables measure the factors associated with increasing political participation over the nine-month course of the campaign (columns 1 and 2) or the last month before the general election from September to October (columns 3 and 4). Again, residing in a state with more initiatives on the ballot, or increased spending per capita on issue campaigns is associated with more participation in politics. Across the models in Table 7, residing in a battleground state is associated with change in political participation, as is residing in a House district that is highly contested. But over and above battleground state residence, exposure to salient ballot measures increased levels of political participation. Since we are only measuring change in participation across the same individuals at varying points in the campaign, concerns about endogeneity are minimized. It appears that state context is responsible for their changes in political behavior.

16 Table 8 goes one step further to test hypothesis 4. We draw on a random subsample of 1000 respondents who were asked about their level of interest in measures on their state s ballot. We thus can measure if citizens are aware and interested in the issue elections using direct survey question. Interaction terms measure if the respondent was interested in the ballot measure multiplied by the actual number of measures appearing on his or her state ballot (columns 1 and 2) or multiplied by initiative spending (columns 3 and 4). In the first two columns of Table 8, the interaction term is positive and statistically significant; citizens interested in the measures on their state ballot and who reside in a state with more salient issue elections were more likely to report high interest in the election in the October wave. Estimated predicted probabilities indicate the substantive effect of this interaction term is large. Holding other variables in the model at mean/modal values, if an individual is not interested in the ballot measures and lives in a state with zero measures on the ballot, his or her probability of being very interested in politics is.69. If the same individual is interested in the ballot measures and lives in a state with two initiatives on the ballot (mean), he or she has a.86 probability of being very interested in politics, a.17 percentage point increase. Similar substantive effects are found for political participation (Table 8, columns 3 and 4), however the base terms are statistically significant rather than the interaction term. Predicted probabilities allow us to understand the substantive meaning. Holding other variables in the model at mean/modal values, if an individual is not interested in the measures on his or her ballot and lives in a state with zero initiatives, the probability of doing at least one political activity is.70. If the same individual is interested in the ballot measures, and lives in a state with two initiatives on the ballot (mean), he or she has a.85 probability of doing one political activity, a.15 percentage point difference. Varying the number of initiatives on the ballot from minimum (1) to maximum (10) for those who are interested in the measures on their state s ballot results in a.05 percentage point change over and above the.15 change of doing at least one political activity. Finally, Table 9 puts all the pieces of the puzzle together to test hypothesis 5. The models are set up like Table 8 using only the 1000 person sample of respondents who received the question about their

17 interest in ballot measures, but the dependent variable is change in political interest, as in Table 5. Thus, we use a lagged term for the individual s general level of political interest in the baseline wave to model change. Again we see individuals interested in ballot measures in their state and exposed to more initiatives or spending on initiatives are more likely to change their general interest in the election over the short and long term campaign; see the positive interaction terms in columns 1 and 2. They are also more likely to change their participation levels, as seen by the positive interaction terms of in columns 3 and 4. The models in Table 9 provide an individual-level covariate measuring interest and awareness of the measures on one state s ballot combined with an exogeneous measure of exposure to salient issue contents to predict changing levels of interest and participation for the same individuals. It provides the highest level of confidence that salient ballot measure contests do indeed increase political engagement and participation, above competitive candidate races. Competitive House and Senate races are also associated with increases in political interest, providing content validity to the models. Discussion Drawing on dynamic survey data, we reassess direct democracy s educative effects, specifically how ballot initiatives may affect individuals general political interest and participation levels during the course of a single election season. The large sample size of the six-wave 2008 panel survey allows us to examine changes in interest and participation over nine months for individuals who were directly exposed to statewide ballot initiatives, compared with those who did not live in states permitting direct democracy. Expanding on Zaller s (1992) argument that the greater level of cognitive engagement with an issue leads to a higher possibility of receiving political messages (RAS model), and the idea that ballot initiatives can create state-specific issue publics (Smith and Tolbert 2010), we suggest that individuals exposed to direct democracy campaigns are more likely to make top of the head decisions on a broader range of issues than those who do not live in an initiative state, which leads to greater political interest and participation, even in the short-term. We find that individuals exposed to more initiatives on their state s ballot or increased spending on initiatives were more likely to change their overall interest in

18 politics or participation over the nine-month election campaign, and also in the last month before the general election. Since we measure change in interest and participation, and not absolute levels, we can be assured the effects we find from direct democracy campaigns on individual political behavior are not because of latent endogeneity tied to an individual s state. Using a unique survey question designed by the authors, the results also show individuals interested in the measures on their state s ballot and exposure to more initiatives or spending on initiatives express a greater general interest in the election, and are more likely to be politically active. Similarly, individuals interested in the measures on the ballot and exposed to more ballot measures are more likely to change their levels of interest and participation over the course of the election campaign. The irony is that the common assumption of political science scholars is that direct democracy has the greatest spillover effects on political behavior and turnout in lower information midterm elections (Smith 2001; Tolbert, Grummel and Smith 2001). This study shows strong effects of issue campaigns and ballot measures in a presidential election. The dynamic panel data allows us to measure changes in civic engagement over the course of an election campaign, and associated state contextual factors, in a way that has not been done before. Our findings have important implications not only for those who study campaign effects, but also those interested in building civic engagement. State level issue publics, fostered during issue campaigns, may increase civic engagement and participation in politics.

19 Figure 1: Histogram of Index of Political Participation (count of 0-13 activities) in December 2007 (Baseline) Figure 2: Histogram of Index of Political Participation (count of 0-13 activities) in October 2008

20 Figure 3: Histogram of Change in Number of Political Participation Activities (0-13) from Dec 07 to October 08 (Zeros omitted) Figure 4: Histogram of Change in Number of Political Participation Activities (0-13) from Dec 07 to October 08 (Zeros omitted) Respondents with no initiatives on the ballot Respondents with 1 or more initiatives on the ballot

21 Table 1: General political interest, October (08) Frequency Percent Very much interested 8,528 49.96 Somewhat interested 5,328 31.21 Not much interested 1,863 10.91 Skipped/Not Asked 7.98 Table 2a: Change in Political Interest from Baseline/Dec 07 (row) to October (08) (column), Row Percents Political Interest in October (08) Very much Somewhat Not much Total Political Interest at Baseline (Dec 07) interested Very much 5,596 interested (85%) Somewhat 654 interested (17%) Not much 31 interested (2%) Not Sure 11 (7%) Change represent upwards movement Change to very interested: 696 (5.83%) interested 949 (14%) 2,741 (70%) 347 (28%) 55 (36%) change to somewhat interested: 402 (3.37%) interested 37 (>1%) 532 (14%) 87 (57%) 11 (38%) Total became more interested: 1098 (9.20%) 6,581 (100%) 3,928 (100%) 154 (100%) 29 (100%) 11,935 Table 2b: Change in Political Interest from September (08) (row) to October (08) (column), Row Percents Political Interest in October (08) Very much Somewhat Not much Total interested interested interested Political Interest in Very much interested 5,824 (91%) 544 (9%) 24 (>1%) 6,392 (100%) September (08) Somewhat interested 658 (16%) 3,077 (76%) 331 (8%) 4,065 (100%) Not much interested 31 (2%) 424 (27%) 1,104 (71%) 11,559 (100%) Change represent upwards movement Change to very interested: 689 (5.73%) Change to somewhat interested: 424 (3.53%) Total became more interested: 113 (9.26%) 12,017

22 Table 3a: Change in interest September (column) to October (row), 1 initiative or less on ballot Political Very much Somewhat Not much Total Interest in October interested interested interested Political Interest at Very much interested 3,851 (91%) 378 (9%) 22 (>1%) 4,251 (100%) Baseline Somewhat interested 409 (15%) 2,077 (77%) 195 (7%) 2,681 (100%) Not much 21 258 762 1,040 interested changes represent upwards movement (2%) change to very interested: 430 (5.39 %) (25%) change to somewhat interested: 258 (3.24%) (73%) Total became more interested: 688 (8.63%) (100%) 7,972 Table 3b: Change in interest September (column) to October (row), more than 1 initiative on ballot Political Very much Somewhat Not much Total Interest in October interested interested interested Political Interest at Very much interested 1,973 (92%) 166 (8%) 2 (>1%) 2,141 (100%) Baseline Somewhat interested 249 (18%) 1,000 (72%) 136 (10%) 1,384 (100%) Not much 10 167 5342 519 interested changes represent upwards movement (2%) change to very interested: 259 (6.40%) (32%) change to somewhat interested: 167 (4.13%) (66%) Total became more interested: 426 (10.53%) (100%) 4,044