Ballot Initiative Knowledge and Voter Turnout:

Evidence from Field Experiments and National Surveys

Jason Barabas1 Charles Barrilleaux2 Daniel Scheller3

DRAFT

Abstract

Policy issues play an important role in explanations of vote choice, but their effect on political participation is less clear. We employ randomized field experiments to determine whether reminding voters about the presence of issue amendments on the ballot influenced turnout in Florida during the 2006 general election. Contrary to findings in some observational studies, providing information regarding initiatives failed to stimulate turnout on all but the least publicly salient ballot amendments placed before voters in that election. In particular, a mail postcard message reminding citizens to vote on ballot measures that had largely escaped media scrutiny increased turnout by a few percentage points. Traditional get-out-the-vote civic duty messages proved to be ineffective as were attempts to remind citizens to vote on several more well-known initiatives. In an attempt to replicate the patterns nationally, we documented strong associations between knowledge of ballot initiatives and intended turnout in two cross-sectional surveys from 2006. Taken together, the empirical results suggest that increasing awareness of ballot initiatives can stimulate voter turnout, especially on relatively obscure issue amendments.

1 Associate Professor, Department of Political Science, Stony Brook University, jason.barabas @ stonybrook.edu. 2 LeRoy Collins Professor, Department of Political Science, Florida State University, cbarrilleaux @ fsu.edu. 3 Assistant Professor, College of Liberal Arts, University of Texas-El Paso, dsscheller @ utep.edu. Issues play an important role in elections, but scholars usually focus on the degree to

which voters use policy preferences to select one candidate over another. Yet in many electoral settings citizens vote directly on issue initiatives in addition to selecting candidates. We have theories of issue voting (e.g., Carmines and Stimson 1980; Nie, Verba, and Petrocik 1979;

Rabinowitz and Macdonald 1989), but surprisingly no comparable concept of “issue turnout” has emerged. The closest one comes to a theory of issue-based turnout is the rational choice explanation of voting (Downs 1957) in which an individual’s decision to vote depends on the probability of affecting the election times the utility gained from having the most preferred candidate win minus any costs. Smith (2001) adopts this perspective in his study of how salient ballot measures stimulate turnout. Information-based explanations of turnout also underlie work by Tolbert and her colleagues (Tolbert, Grummel, and Smith 2001; Tolbert, McNeal, and Smith

2003) as well as other studies that employ formal theoretical perspectives (Grosser and Schram

2006; Lassen 2005).

The idea that issues, and ballot initiatives in particular, can help reverse years of declining voter turnout has great normative appeal (Teixeira 1992). Activists and progressive

reformers claim that direct democracy can help create engaged citizens (Schmidt 1989;

Zimmerman 1999). Scholars are also enamored with the notion of participatory democracy

(Barber 1985; Pateman 1976). We take up the question of whether providing information about

the presence of ballot initiatives affects voter participation.

We examine this question using a methodology—the field experiment—which is the gold

standard of get-out-the-vote evaluations (e.g., Gerber and Green 2000a; 2000b; 2001; 2002;

Green, Gerber, and Nickerson 2003; Nickerson 2006). To date field experiments have not been

used in a large scale study of the effects of get out the vote efforts on voting in elections with 2

ballot initiatives. Our results suggest that providing information on ballot amendments can

stimulate turnout, but mostly in situations when the initiatives are relatively obscure.

The Effects of Ballot Initiatives on Voter Turnout

Ballot propositions have become a popular way for citizens to influence the laws and

policies of their state. Twenty-four states in America practice some form of direct democracy

and usage is on the rise. For instance, David Magleby (1994) found that from 1898 to 1992 over

1,700 initiatives were placed before U.S. voters. That number would have been higher had

hundreds more qualified for inclusion on the ballots. However, Magleby notes that only 38% of

initiatives passed (p. 231), which suggests that voters are fairly discerning. Furthermore, some

states like Oregon and California offer voters many initiatives in elections while others like

Illinois use it sparingly and only for the purpose of altering the legislative process (Tolbert,

Lowenstein, and Donovan 1998).

Past research has often concentrated on how initiatives are placed on the ballot, the role

of special interests, and the effects on public policy outcomes or enfranchisement (e.g., Boehmke

2002; Bowler, Donovan, and Tolbert 1998; Gerber 1999; Smith 2004). However, two rationales are typically advanced for the use of ballot initiatives: (1) preventing the legislature from becoming unrepresentative, and (2) educating the voters on issues and civic skills like voting.

While there is an active literature on the first question (e.g., Gerber 1996; Matsusaka 1995), it is

“educative” effects of initiatives that are of concern here (Tolbert and Smith 2005). Most of the

early studies that ask whether the presence of ballot initiatives increases turnout were not encouraging. In particular, work in the 1980s found no statistically significant relationship 3

between direct democracy and electoral participation (Everson 1981; Gilliam 1985; Magelby

1984).

However, by the early-2000s evidence started to accumulate to suggest that ballot initiatives increase turnout, particularly in midterm elections (Smith 2001), but also in

presidential contests. For example, using a pooled times of data for the 50 states over a 26-year

period (from 1970 to 1996), Tolbert, Grummel, and Smith (2001) find that the presence and

usage of citizen inspired initiatives boosts voter turnout from 3 to 4.5% in presidential races and from 7 to 9% in midterm elections. More recent work by Tolbert and Smith (2005) estimated the

effects at nearly 1 to 2 percentage points for presidential and midterm elections respectively.

They also document variation in effects within types of elections, observing a 4 point turnout

effect in 1992 but a smaller 0.5 percentage point effect in 1996 (Smith and Tolbert 2004, 51).

Methodological Advances and Limitations

Three types of methodological improvements helped move the field from the null

findings of years ago (Everson 1981) to the largely positive effects of today (Tolbert et al. 2001;

2003; 2005; Smith 2001). First, scholars like Tolbert, Grummel, and Smith (2001) obtained more

precise statistical estimates by controlling for potentially confounding factors such as race,

income, and the distinctiveness of the South. Also, they did so while using panel-corrected

standard errors to acknowledge correlations across the repeated state-level observations. Finally,

and most recently, they refined their analyses to consider distinctions between the voter-age

versus voter-eligible populations (Tolbert and Smith 2005).

However, these studies use aggregate-level data. A second methodological innovation has

been to employ individual-level data. For example, “To avoid the ecological fallacies to which 4

aggregate-level analyses are prone…,” Tolbert, McNeal, and Smith (2003, 27) use individual

level American National Election Studies (ANES) data collected during 1996, 1998, and 2000 to

show that ballot initiatives increase turnout in midterm elections, but not necessarily during the

presidential years. Lacey (2005) obtained similar results using a slightly different set of ANES

surveys in 1990, 1992, 1994 and 1996.

Finally, an important methodological advance with theoretical relevance concerns the

salience of the initiative. Mark Smith (2001) finds that ballot initiatives increase turnout in

midterm elections to the degree that they are salient. In other words, Smith argues turnout should

be and indeed was found to be highest when a state’s largest newspaper devoted a lot of front

page coverage to particular amendments on the day after the election (also see Lacey 2005).

However, Tolbert and her colleagues (Tolbert, Grummel, and Smith. 2001) question the validity

of Smith’s measure. They note that “A more direct and simple measure of saliency is the actual

number of initiatives on the ballot each election…” (p. 632). That is, instead of a simple dichotomous measure of whether a state employs the initiative process, they use the number of initiatives on the ballot at any given moment as a way of capturing the amount of information available regarding ballot initiatives (also see Tolbert, McNeal, and Smith 2003; Tolbert and

Smith 2005).

Yet no matter what measure of salience one adopts—front-page media stories, initiative counts, or otherwise—room for improvements remains even with these advances. In particular, there is a tremendous amount of heterogeneity in ballot initiatives rules across the states. Some require the signatures of 10% of the total votes cast in the previous gubernatorial election (e.g.,

Utah) while for others the bar is 5% (Nebraska, Montana), 4% (Arkansas), or just one signature from voters residing in at least two-thirds of a state’s election districts (Alaska). The states also 5 use different time requirements—Idaho, Oregon, and Utah do not limit the amount of time to gather signatures for an initiative while Wyoming has an 18 month limit, South Dakota and

North Dakota each set it at 1 year, and it drops to 289 days in Nevada, 180 days in Michigan, and

90 days in Oklahoma. In addition to these factors, states also vary considerably with respect to requirements for the geographic distribution of signatures, the title and other summary language, signature witnesses, and how signatures are verified (for details, see Tolbert, Lowenstein, and

Donovan 1998).

All of this is to say that cross-state comparisons (e.g., Lacey 2005; Smith 2001; Tolbert,

McNeal, and Smith 2003) are likely to mask a number of important but unmeasured factors beyond the salience or number of initiatives on the ballot. In other words, even if we control for the institutional variations mentioned above, are states with unequal numbers of initiatives the same in all other respects? As one example, in his study of ballot driven agenda setting,

Nicholson (2005) suggests an issue’s importance does not influence whether it appears on the ballot and thus “…endogeneity is not a problem” (p. 41). However, Tolbert, Grummel, and

Smith (2001) seem to dispute this when they consciously choose to concentrate on citizen- inspired initiatives because “groups outside the legislature can propose new legislation [which means] the subject matter of citizen initiatives tends to be more controversial than policy referred by state legislatures” (p. 627).

By the same token, do states that use initiatives heavily, such as Oregon, have more interested citizens or perhaps a press corps that has evolved to cover initiatives differently than in other states? These questions have not escaped the notice of other scholars working in the area.

Tolbert et al. (2001) question the validity of Smith’s (2001) newspaper operationalization of ballot salience. They ask readers, “…is front page news coverage measuring what it is intended 6

to measure (i.e., the saliency of initiative contests), or is it measuring the opinions of newspaper

editors who decide what issues get front-page billing?” (p. 645).

Ideally, to study the effects of initiatives on statewide turnout, we would randomly pick a

state(s), randomly place some number of amendments on the ballot, and then compare the

turnout rates of randomly selected groups of voters in the states with the experimental ballot

initiative treatments to the participation rates of voters in states without the ballot amendments to assess the effects. Or we might place initiatives on ballots in randomly assigned parts of the state but not others, thereby eliminating variations in state institutions as an explanation for their effects. Although desirable from a social science standpoint, such manipulation is highly impractical because it would amount to hijacking a statewide election. Even if it were possible to

convince citizens and elected officials of the merits, getting a study like this through an

institutional review board could prove to be even more difficult.

Instead, some scholars studying turnout turn to laboratory experiments (Grosser and

Schram 2006) or quasi-experiments in which some municipalities are exposed to turnout relevant

information while others are not due to real world variations (Lassen 2005). However, neither of

these approaches can overcome concerns about internal validity and external validity (that is,

establishing causation and generalizability of the effects); they excel on one dimension or the

other, but not on both. This has led researchers to hybrid designs, like field experiments, which

can overcome these problems and approximate the ideal research situation described earlier.

Reviving a dormant field experiment research model started by Gosnell in the 1920s

(1927; also see Eldersveld 1956; Miller, Bositis, and Baer 1981), Alan Gerber and Donald Green have documented how randomly assigned get-out-the-vote messages delivered in a variety of

different ways (mail, phone, in-person, door hangers, etc.) affect turnout across federal, state, and 7

local elections in dozens of field experiments across as many jurisdictions with thousands of randomly selected subjects (Gerber and Green 2000a; 2000b; 2001; 2002; Green, Gerber, and

Nickerson 2003; Nickerson 2006; 2007). Recent efforts have extended these techniques to studying efforts to mobilize Asians, Indians, Latinos, and other minority groups (Michelson

2003; 2005; 2006a; Ramirez 2005; Trivedi 2005; Wong 2005). However, ballot initiatives have not been the exclusive focus of a large-scale field experiment on turnout (cf. Michelson 2006b).

As discussed in more detail below, we conduct a field experiment to determine whether

providing voters information on initiatives alters turnout rates.

Information and Initiative Salience

Years ago when scholars encountered null empirical findings (e.g., Everson 1981), the

expectation was that initiatives ought to increase voter turnout. That premise is appealing for

several reasons. A proportion of the electorate is predisposed to vote based upon issue-oriented

politics (Carmines and Stimson 1980; Enelow and Hinich 1984). Given that issues have

increased in salience with respect to vote choice since the 1960s (Nie, Verba, and Petrocik 1979;

Gopoian 1993), it might not be unreasonable to think that voters who are reminded about initiatives should be more likely to go to the polls.

Nevertheless, not all initiatives are equal and would-be voters face cognitive limitations

(Berinsky 2005). As Smith (2001) writes, “it is the most salient initiatives and referenda for which citizens will perceive the greatest differences from voting one way over another….Thus, the initiatives and referenda with the highest public salience should show the strongest relationship with turnout because of their impacts upon the benefits of voting” (p. 700). These views are echoed in a slightly different fashion by Tolbert, McNeal, and Smith (2003) who state, 8

“Ballot initiatives can provide information about and generate interest in an election, which may

lead to higher voter participation” (p. 25). Thus the general expectation from past work is that

initiatives should increase turnout to the degree that they are salient, and having more initiatives

on the ballot should, ceteris paribus, increase the amount of information that citizens receive.

However, we differ from past work in this area not only with respect to the main research

approach we will use (i.e., a field experiment), but also in our expectations regarding the effects

of information and initiative salience. From the voting literature we know that citizens are less

likely to vote for candidates when they lack information or are uncertain about where that person

stands on the issues (Alvarez 1997; Franklin 1991; Burden 2003). The same logic can be

extended to turnout decisions based upon issue initiatives. In other words, contrary to past work

(e,g., Smith 2001), our hypothesis is that we expect to see the largest effects of information on

turnout for initiatives where voters have the least information. If voters already know an

initiative is on the ballot, then reminding them of this fact in experiment should not be

particularly effective. Voters will have already been “treated” on the issue (e.g., Gaines,

Kuklinski, and Quirk 2007), which leaves little if any room for additional effects. If, however,

voters know little about an issue on the ballot, then providing an informative message

encouraging them to vote because of that issue should increase the initiative’s importance.1

We will also revisit the measurement of initiative salience. As mentioned briefly and discussed in more detail below, past work has used initiative counts or media coverage after an election as proxies for how much information voters received (and thus how salient the initiatives are). However, we start with voters—who differ in their baseline level of information regarding each initiative—and then provide information in a random fashion. Thus, we can

1 Since the literature (e.g., Smith 2001; Tolbert et al. 2003) uses terms like salience, information, and importance, we also employ those words. However, we use related terms like uncertainty and awareness as well. 9

structure our predictions based upon the nature of the initiatives, particularly the salience of each

initiative for citizens prior to the election. In the next three empirical sections, we outline our

expectations regarding initiative knowledge and turnout.

Study 1: Florida 2006 Statewide Field Experiment

Instead of leveraging cross-state variation in initiative use as others have, in our primary

analysis we go in-depth on one state—Florida—and several amendments that appeared on the

ballot during the fall of 2006. Compared to South Dakota which was the first state to allow

initiatives in 1898, Florida is a relative newcomer to direct democracy. It first allowed ballot

amendments in 1968, making the second most recent state to do so after Mississippi which began to allow initiatives in 1992 (see Matsusaka 2004 or Smith and Tolbert 2004 for histories of the process).

Florida is often categorized as the only state with a purely direct constitutional system in which citizens are allowed to amend the state’s constitution without legislative intervention

(Donovan and Bowler 1998, 4). In reality, however, Florida is more of a hybrid since many ballot measures originate with the legislature, not citizens, so the formal distinctions between direct vs. indirect and referendum vs. initiative become blurred.2 No matter what they are called,

with a few exceptions (Niven 2006; Smith 2004), Florida’s initiative system has not been studied

even though it is the fourth largest state in the U.S. and it is often seen as an important

battleground for both major parties. Moreover, and as discussed earlier, concentrating on one

2 That is true of the election studied here; only the amendment dealing with tobacco originated from a citizen’s group (Hull 2006). The others originated from the legislature, which makes them similar to popular referendums. Throughout the text we obscure subtle differences in terminology, but a direct initiative allows citizens to draft new state laws or amend the state constitution. In the indirect initiative, a group drafts and qualifies the proposition before submitting it to the legislature for consideration. If the legislature passes the measure, it becomes law. Otherwise the policy is placed on the ballot for the voters to decide its fate (Cronin 1989; Tolbert et al. 2001). Like others, we use the term initiative interchangeably with related terms like direct democracy, propositions, or ballot measures, but technically speaking these initiatives are “amendments” as they are described to voters on the ballot. 10

state allows use to hold a lot of factors constant that might otherwise confound statistical

estimates (e.g., the same qualifying requirements, interest group involvement, historical use of

initiatives, rules for passage, party control of the legislature, topic of the amendment).

The Initiatives

The six statewide amendments on the ballot in Florida during the 2006 general election were as follows:

 State planning and budgeting reforms (“state planning and budgeting”)  Eminent domain limitations (“eminent domain”)  Supermajorities of 60% needed for future amendments (“supermajorities”)  Funding for anti-smoking programs from the tobacco settlement (“tobacco settlement”)  A property tax homestead exemption for low income seniors (“homestead exemption”)  A property tax break for disabled veterans (referred to hereafter as “disabled veterans”)

The ordering in the six dot-points above corresponds to our expectations concerning how salient

each initiative was for voters statewide going into the election, from least to most. We arrived at

that determination after considering media-based techniques for determining initiative salience

used in the past literature.3

As we stated earlier, our study considers a single state, so the number of statewide

initiatives is held constant. That means the counting method of initiative intensity (Tolbert,

Grummel, and Smith 2001; Tolbert, McNeal, Smith 2003) cannot be used here. Smith’s (2001)

methodology for assessing salience also breaks down. Smith used a count of “all paragraphs on

3 Appendix A provides more detailed summaries from the Florida Division of Elections on each of the amendments—in the order they appeared on the election ballot—as well as the number of votes for and against each measure. All six amendments passed. The most popular according to percentage of votes in support were the property tax measures for disabled veterans (77.8%) and low income seniors (76.4%). At the other end of the spectrum were the items on the state planning and budget process (59.8%) and the requirement for supermajorities on any future amendments, which ironically did not garner the sixty percent threshold that subsequent initiatives will need (57.8%).

11 the front page of the state’s largest newspaper and all other newspapers with a circulation of at least half that amount” on the day following the election (p. 701). In this election, only one paper, the state’s second largest (the Orlando Sentinel), provided any front page coverage of the amendments, and even then it only indirectly mentioned the supermajority item in passing. Thus, the front page technique is uninformative in this instance. In particular, the 2006 election featured a newsworthy gubernatorial contest to elect a successor to former Governor Jeb Bush.

The papers in Florida also devoted a considerable amount of coverage to the Democratic take- over of the U.S. House and Senate. The six ballot amendments were almost completely pushed off the front pages of the state’s newspapers.

As an alternative to Smith’s front-page method, we examined news coverage devoted to the amendments after the election in any section of the paper in ten of the state’s largest print outlets (the nine largest based upon 2005 circulation and one which serves the capital of

Tallahassee). We tallied the number of paragraphs devoted to each amendment as well as their order of mention. By either of these measures, the most salient amendment would be the supermajority item. Article after article mentioned this initiative and the millions that were being spent by its supporters and opponents. The least salient by either metric would be the state budget planning proposal. In other words, using the closest available variant of Smith’s (2001) salience measure, others researchers might expect the biggest effects on the supermajority amendment and the smallest for the state planning and budgeting amendment.4 However, our predictions about where we should see the greatest effects differ from what others might expect

4 Not all ten papers covered the amendments. Those that did were the Miami Herald, the Orlando Sentinel, the St. Petersburg Times, the Ft. Myers News-Press, the Times-Union of Jacksonville, the South Florida Sun-Sentinel, the Sarasota Herald-Tribune, and the Tallahassee Democrat. By numbers of paragraphs across the papers, from most to least salient it would be 1) the supermajority, 2) tobacco settlement, 3) a tie between eminent domain and homestead tax cuts for low income seniors, and 4) a tie between state budget planning and property tax breaks for disabled veterans. By order they were mentioned in the articles, the salience rankings from high to low salience would be 1) supermajority, 2) eminent domain, 3) homestead exemption, 4) disabled veterans, 5) tobacco settlement, and 6) state budget planning. The ordinal rankings are the same if the ten paper averages are weighted by circulation. 12

because we argue it is on the least salient initiatives that providing information should have the greatest effects. In particular, we expect to see turnout effects on the state budget and planning amendment because it was substantially less salient in the minds of potential voters.5

While formal statistical tests are not particularly informative with the media measures

given the small samples, it is the case that the state planning and budget amendment differed

significantly in from the other amendments in terms of the willingness of respondents to express their preferences in pre-election polling. More specifically, the percentage of respondents who

“don’t know” or are undecided about a ballot in pre-election polls can serve as a proxy as it has

in studies of voter information and uncertainty (Bartels 1986). In four statewide surveys in the lead up to the 2006 Florida election, the least salient in the polls as measured by the high level of

“don’t know” responses was the state planning and budgeting amendment.6 In particular, the

state planning and budget amendment had the highest percentage of undecided responses in these

pre-election polls on the amendments, nearly twice what was observed on other amendments

(i.e., 32% don’t know versus the next closest average of 16.8%). A binomial test for proportions

confirms responses on the state planning and budgeting amendment were statistically distinct

from the others (p < .01). Equally important, the other initiatives do not differ significantly from

each other when judged by don’t know responses.7

5 Of course, salience could be ordinal instead of dichotomous. We test this in the empirical analyses later. 6 The four polls were as follows: a Mason-Dixon poll of 625 registered voters between 10/16 and 10/17 which was published in the Orlando Sentinel on 10/29/06, and a Mason-Dixon poll of 625 registered voters between 9/20 to 9/22 published in the Orlando Sentinel on 9/30/06, a Zogby poll of 803 likely voters between Sept. 26-28 and published in the Miami Herald on 10/2/06, and a Zogby poll conducted in “late-October” with 800 likely voters published in the Miami Herald on 11/5/06. 7 We hesitate to advance the don’t know poll response method of determining issue initiative salience because these measures are uncommon and may not be available in other situations. Instead, media coverage seems to be something that one might expect in most elections and it likely varies. Another possible measure of initiative salience is voter rolloff. As with the other measures, the state budget and planning amendment was the one with the fewest votes cast. It garnered more than one-hundred thousand fewer votes (at 4.3 million total) than the next closest amendment (eminent domain). In contrast, homestead and disabled veteran tax exemptions were among the most highly voted upon (between 4.5 and 4.6 million votes). Fortunately, all the measures point to the state planning and budgeting item as being the least salient among the six statewide. 13

The variations in initiative salience help guide our expectations for where we should see

the greatest effects of issue turnout. With Smith’s (2001) post-election media salience measure of

paragraph counts (or more technically, our best attempts to simulate it), one might expect the

greatest effects of turnout on the supermajority item. In contrast, the same metric shows that the

state planning and budget item was the least salient (the number of paragraphs it received on

average in state newspapers was among the lowest we recorded at 1.2 paragraphs). Moreover, the planning and budget amendment happened also to be the least salient according to the average order it was mentioned in each story (which was last at 5.2, meaning it was consistently mentioned last in stories about the six amendments). Thus, it is on the least salient amendment

according to all of the indictors—i.e., the state planning and budget item—that we will look for

issue-based turnout mobilization effects because voters have the most to gain from the

information treatments that we provide in the experiment described next.

Statewide Field Experiment

We obtained a list of all active voters in the state of Florida from the Division of

Elections and selected seven thousand at names at random. The subjects were randomly divided

into groups of a thousand corresponding to the six amendments described in the previous section

plus a seventh that would receive a non-issue get-out-the-vote message. The overall goal is to

compare the turnout rates of those who received the messages with potential voters who did not

(i.e., everyone else in the state). However, a number of cards were returned because the voter

moved or because the records kept by the Department of Elections were outdated. Most previous

work has “assumed that all of the households we intended to treat by mail received the treatment,

an assumption implicitly made in all previous mail experiments” (Gerber and Green 2000, 659,

fn. 10). Since the number of returned cards is usually low, attrition is often not a problem. 14

However, to improve our treatment effect estimates, we recorded and refined our analyses to

concentrate on only those voters who were actually treated (i.e., the cards were not returned).

The non-partisan mailers we sent to voters in the respective treatment groups were mailed

approximately one week prior to the election. Recipients received a single, 4 X 6 yellow postcard

produced and sent by a university print shop/post office. The top of each postcard read

“November 7, 2006 is Election Day in Florida.” Directly below this reminder was a neutral one

sentence description of an amendment on the ballot. For example, the treatment on the initiative

process said, “Amendment 3 proposes to require a 60 percent vote rather than a simple majority

to amend the Florida constitution.” This message sought to decrease information costs for the

voter and increase the salience of the initiative. Nevertheless, the cards were relatively sparse

from the standpoint of deciding how to cast a vote, and the main thing they accomplished was to

remind voters of the presence of ballot amendments in the upcoming election.

Although we have concentrated on the information function, the cards reminded voters

that their vote matters. Under each amendment message, a statement appeared which read, “Now

more than ever, your community, state, and nation need to hear what you think,” followed by

“Please remember to vote on November 7.”8 These statements engendered feelings of civic duty

and increased the perceived probability that their vote mattered (Riker and Ordeshook 1968).

However, these messages were a part of every treatment so no amendment is advantaged.

Appendix B provides more details on the mail scripts.

Study 1 Empirical Findings

To recap, our predictions about the effects of providing ballot initiative information are

for increases in turnout contingent upon the level of pre-existing information that voters have; in

8 For the general GOTV cards, there was no text where the amendment description was located. The card simply had the “Now more than ever…” and “Please remember to vote…” messages. 15

other words, the effects of increasing awareness of the ballot amendments should be the greatest

when voters are most uncertain or unaware. That means we will search for effects on particular

initiatives (i.e., state planning and budget). However, before evaluating the empirical support for

our expectations it is important to determine whether the randomization process “worked” in the

sense of balancing the groups.

Table 1 displays descriptive information on common demographic traits available in the

voter file as well as partisanship. The first column reports the averages for more than 9 million

registered voters in the state. Overall our sample registered voters in Florida nearly 50 years old

on average (48.7), 53.06 percent female, 12.77 percent black, and 11.11 percent Hispanic.

Similarly, Democrats account for 41 percent of the voters (40.89 percent) while Republicans

constitute 36 percent of the sample.9

Insert Table 1 here.

Moving from left to right, the remaining columns in Table 1 show the demographic

averages for any treatment (n=5,407), the generic get-out-the-vote (GOTV) message (n=758), as

well as each of the six statewide amendments (an average group size of 775). In each case, the

group sample average is reported along with the deviations of that average as compared to the statewide statistics. Overall the pattern is one of fairly comprehensive randomization. For most

factors, there are no discernible differences between the statewide sample and the specific

treatment groups. However, the top portion of Table 1 contains a few exceptions. The entire

9 The descriptive statistics reported in Table 1 might differ from census demographic figures or other published accounts because they exclude voters in our treatment conditions or any other respondents who were omitted due to voting early prior to our treatments. Like other analysts who have conducted field experiments in Florida (Niven 2006), we exclude absentee voters because they had most likely already received their ballots before the experiment was conducted (and because the records of when the absentee voters voted were not publicly available). Other than reducing the number of cases eligible for analysis, this should not bias the results because absentee voting rates were similar across the treatment and control groups due to randomization. We also purged 41 individuals who were not yet age 18 on the day of the election and thus could not vote. However, we obtained records of when early voters voted (which is distinct from absentee voting), and we used all available cases that occurred after our treatments were in the field. 16

treatment group appears to be about a half point older (.43 difference with a standard error of

.25), the GOTV group is 2.5 points less Hispanic, the state planning and budget group is about a

year older and almost 4 points more Democratic than the state sample. The bottom half of Table

1 shows the same comparisons for the amendments on the supermajority, tobacco, and the two

property tax measures. Only on the homestead exemption were there differences, with the sample being a little more than a year older and 3 points more likely to include women than the statewide. While these differences are not enormous, several are statistically significant. The

upshot is that once treatment effects are established via simple group mean comparisons it will

be important to control for background characteristics to make sure sample composition

differences are not driving the results.

We move to the statistical models with the controls for background characteristics in a

moment, but before doing that it helps to visual the overall patterns with simple mean

comparisons. Table 2 displays mean levels of voter turnout for each postcard treatment group.

The first column reports the number of individuals receiving some form of experimental

treatment and the second column shows their turnout rates in the 2006 Florida election. The

average size of the treatment groups was 772 with a range of 758 (GOTV) to 801 (homestead

exemption). The average turnout was just under 43% (48.83), with a range from 40.54 (tobacco settlement) to 46.77 (state planning and budgeting).

Insert Table 2 here.

One of the first things to notice about the table is what it does not contain: the statewide

average turnout for individuals who were untreated with any message. While other types of

experiments typically compare those who receive a randomized treatment to individuals who do not (i.e., a traditional treatment vs. control comparison), doing so in here could bias the results. 17

In other words, the statewide average turnout among those who did not receive treatment is just under 40 percent (39.1 percent with a standard error of .02). Given the small standard error and the extreme precision of the estimate with the large sample, a naïve comparison between treatment groups to the millions of individuals who went untreated statewide would result in several of the items registering statistically significant differences. However, doing so could

introduce analytical bias. In particular, and even though the groups were randomized, not

everyone will actually receive a postcard if one is sent. Since the subset of people who receive a

postcard likely differs from those who did not, simple comparisons of treated and untreated cases

has the potential to introduce bias.10

As an alternative, we leverage a feature of the research design to make unbiased

comparisons, or at least comparisons that do not stack the deck in favor of finding results. In

particular, respondents in the generic get-out-the-vote (GOTV) condition can be used as a

placebo control group (see Nickerson 2005). In other words, like those who received a ballot

amendment message, individuals in the GOTV condition also received a postcard so that the

chances of reception are similar. This method of analysis has a few advantages. First, it makes

contact irrelevant since everyone received the contact. Second, the GOTV message does not

promote a ballot initiative, which makes it a useful placebo. In some ways it would have been

better had the placebo involved something totally unrelated to politics (i.e., blood donation; see

Gerber and Green 2001). Nevertheless, such a comparison is a conservative one and it permits a

more direct test of the link between initiative awareness and turnout.

The third column of Table 2 displays the treatment effect of each amendment treatment

group relative to the GOTV condition. The only statistically significant treatment effect is the

10 We thank the anonymous reviewers for pointing out this potential problem and for suggesting the solution discussed next. The contact rates were high, around 88 percent, and nearly indistinguishable across the amendment and GOTV groups. 18 first one involving the state planning and budgeting amendment. As expected, receipt of a postcard reminding voters that the planning and budgeting amendment was on the ballot boosted turnout by 4.16 points relative to the GOTV group (i.e., turnout rates of 46.77 vs. 42.61 percent).

This difference attains one-tailed p < .10 significance, which is not overwhelming, but it is the intended direction and it is large considering that mail treatments typically boost turnout by about a half point in other work (Gerber and Green 2000).11 The other amendments register small or even negative effects versus the GOTV group. Since the planning and budgeting amendment was not particularly salient, we expected the largest effects here, but even an average of the three least salient results in an increase in turnout of about a point (1.19, se=2.53) versus a drop of a point (-.68) for the three most salient amendments (i.e., the last three rows of Table 2).

The fourth and fifth columns of Table 2 extend the placebo group logic further. Since only one group differed from the GOTV baseline (state planning and budgeting), all of the remaining groups can be used as their own placebo control condition. Column 4 re-calculates the turnout effects for each group versus all the other amendments except, of course, the state planning and budget item which was shown to be different in the previous column. As an example, the GOTV entry for column 4 shows .52 with a standard error of 1.96. This insignificant difference is the result of comparing a turnout rate of 42.61 for the 758 members of the group who received a GOTV postcard to the 3,875 individuals with a turnout rate of 42.09 percent who received an amendment message other than state planning and budgeting. That half of a point difference is insignificant and the same holds for all of the other entries except the state planning and budgeting amendment, which again looks to be different than the others. This time the difference is estimated to be 4.68 points with a standard error of 1.95 (p < .05, one-

11 Gerber and Green (2000b) write, “Since it is only reasonable to expect either no effect or a positive effect on turnout from the experiment, a one-sided test is clearly justified” (p. 850). Similarly, we adopt one-sided tests in the field experiment analyses. All of the tests in Table 1 are t-tests for differences in means. 19

tailed) based upon the 46.77 turnout rate for planning and budgeting versus a 42.09 pooled

turnout rate for the other amendment groups. The estimates in column 5 are nearly identical since

the baseline is a blend of the GOTV group as well as all other amendments. Once again,

individuals who received the state planning and budgeting amendment message turned out at

rates about five percentage points higher than others who received messages on amendments or

GOTV (46.77 vs. 42.18, for a difference of 4.59 with a standard error of 1.92, p < .05).

While these results support our expectations regarding the largest treatment effects on the

least salient amendment, earlier we noted several deviations from perfect randomization. Since some of the deviations occurred with the state planning and budgeting item, Table 3 reexamines the patterns with background characteristics included to control for differences in sample composition. The first three columns report probit model estimates with a turnout dependent variable as a function of the experimental treatment in the state planning and budget amendment and demographic variables (age, gender, race, or Hispanic ethnicity) as well as partisanship. In each case, receipt of a state planning and budget post card appears to be positive and significantly related to turnout when using the GOTV group as the baseline (column 1, coeff.

=.09, p < .10), all other amendments (column 2, coeff.=.11, p < .05), or all other amendments plus the GOTV condition (column 3, coeff..11, p <.05). Several of the individual level factors are important. Older respondents are more likely to vote as are nonblack or non-Hispanic individuals. Likewise, Democrats and Republicans are more likely to vote relative to the omitted baseline category of Independents and anyone else who does not identify with one of the two major parties.

Insert Table 3 here. 20

The next three columns of Table 3 show the same patterns, but this time controls are

added for prior voting history in the 2006 primary election as well as the general elections in

2004, 2002, and 2000. In each case the state planning and budget amendment treatment remains

statistically significant (p < .10 or p <.05 depending on the comparison group) and in the

expected positive direction. Prior voting history proves to be significantly related to turnout as

well, always positively.12 It is also worth noting that the standard errors for the models have been clustered to account for voters who live at the same address,13 and in other analyses (not shown),

the same type of models with the other issue amendments are insignificant predictors of turnout.

The final set of estimates in Table 3 shows what happens when we use a salience weighted approach to the experimental treatment instead of a simple dummy variable to denote the state planning and budgeting amendment. In other words, until this point we have been treated salience as a dichotomy, but the final set of estimates in table three has a term for experimental treatment that is coded as 1 for the least salient (and hence the most likely to

exhibit effects in our argument), .6 for the next least salient items (which was a tie between

eminent domain and homestead and thus means both are in “third place”), and so on until the

reaching a value of zero for the most salient item (supermajorities) according to our media-based

count of the number of paragraphs devoted to the amendments in major state newspapers. In the

last three models with the three different variants of the comparison group (e.g., GOTV alone,

other amendments, or GOTV plus other amendments), the salience-weighted indicator of the

experimental treatment is positive and significant. Thus, even though we have been looking for

effects on one amendment in particular, models using treatment groups for all amendments

12 Like Gerber and Green, we use dummy variables for voting and abstaining (relative to the omitted baseline of no voting history on file) to control for historical tendencies to vote for each individual. The abstention terms are in the model but the output is suppressed from the table due to space limitations. 13 The results are the same if we included fixed effects for voters who live at the same address or a dummy variable to denote instances (n < 35) when multiple individuals live at the same address. 21

(weighted according to the likely salience) show positive and statistically significant effects (p <

.10 or better) even after controlling for demographics, partisanship, and prior voting history.

To illustrate the effects of the experimental treatment terms in Table 3, we show the predicted probability change from the baseline placebo group versus membership in the state planning and budgeting group (or the weighted salience group for the last three entries). The treatment effects are estimated to be between 3.7 to 4.9 percentage point depending on the particular set of comparisons and control variables included. While the effects appear to be significant as shown by the confidence intervals, they are modest. In fact, some of the demographic predictors would be more strongly associated with turnout differentials.

Nonetheless, we expected to see and indeed witnessed an elevated level of turnout for individuals who received a postcard relating to the least salient amendment on the Florida 2006 statewide ballot.

The empirical results provide confirmation, even if somewhat surprising given past work, that reminding citizens to vote because of issue amendments often fails to increase turnout. The only caveat to this was on the issue that was the least salient in the minds of the voters. On issues like the state planning and budgeting reform that are not particularly salient, it was indeed possible to mobilize citizens to vote. In contrast, citizens are no more likely to vote when they are encouraged to vote on highly salient issues. Thus, scholars might want to reevaluate the role of issues in turnout, and redirect their attention to the levels of information and salience that voters hold on the issue initiatives prior to the election. However, before concluding we wanted to see if the patterns hold up in another field experiment.

22

Study 2: Field Experiment in Leon County, Florida

The second field experiment was conducted during the same 2006 election in Leon

County, Florida, which has a population of roughly 240,000 and contains the capital city,

Tallahassee. The goal was to undertake a smaller scale version of the first experiment (i.e, only

100 to 125 voters per treatment condition) but to do so with separate randomization from the statewide study. All six amendments were targeted for inclusion in addition to the generic get- out-the-vote message and a seventh local initiative on a county-wide tax increase to provide health insurance for the indigent, which was on the ballot only in Leon County.

In October, 2006, we requested a complete list of the registered voters in Leon County from the Leon County Department of Elections. We then purged all inactive voters.14 For the

dataset, there were originally over 187,000 registered voters in the county. After purging the

dataset of inactive voters, roughly 147,000 active voters remained. The data contain information

like party identification, and an individual’s participation in the previous elections. A few

months after the 2006 general election, we obtained a disk from the Leon County Department of

Elections, containing information about individual-level turnout from that election. Again, we

sought to determine the effect of postcards on an individual’s decision to vote.15

The change from the statewide study to a single count is helpful from the perspective of

testing our theoretical argument. In particular, the county we studied (Leon County) typically

votes at rates much higher than the rest of the state, so this represents a harder test of the

experimental treatment since residents will be less susceptible to treatments. Also, the salience

14 According to the Leon County Department of Elections, they send mail to all registered voters in the county. When the mail is returned to the Department as “Return to Sender,” they list these voters as inactive. The voters may still live in Leon County, but may have moved. The assumption is that the mailing address is not correct in the data file and unlike the statewide study, these names have been purged prior to the random assignment of treatments. 15 There were other aspects of the study, but we focus on those that are most relevant for testing the argument here. Once again, we omitted anyone who might have voted early prior to our experimental intervention (see footnote 9). 23 ordering was different in this county than it was statewide. In other words, statewide we expected to see the increased turnout on one amendment (state planning and budgeting), but in

Leon County we expected to see positive treatment effects on a different low salience amendment (tobacco prevention). That is, in news coverage during the two weeks leading up to the election, the main newspaper serving Leon County, The Tallahassee Democrat, covered this amendment the least (it received only brief mention in two stories as opposed to the supermajorities amendment and the county health tax which were each covered more than twice as much as well as in more depth). The tobacco amendment faired poorly by the other standards employed in the statewide analysis as well. While the nature of the issue might seem more straightforward (and thus more salient) than it was with the arcane state planning and budget amendment, in a town with the state capital the budget item was actually more relevant since the state’s budgets often receive front page coverage throughout the year. Thus, it is on the tobacco amendment that we expect to observe the greatest potential effect for providing information about the presence of a ballot amendment. Since the procedures and issues at stake were similar to those employed in Study 1 we move into the analyses quickly without a lengthy background discussion of the case. We can also move quickly past the randomization check since demographic and partisan factors are nearly always insignificant predictors of treatment receipt.16

Table 4 shows the sample sizes for the groups that received each postcard treatment

(column 1), their turnout rates (column 2), and the treatment effects calculated in three different ways using the placebo control methodology (columns 3, 4, and 5). Turnout in the untreated

16 In probit models predicting treatment receipt relative to the sample of more than one-hundred thousand untreated voters in Leon County (n=127,908), the only significant predictor is age on the homestead amendment (p < .05, two- tailed). Individuals in this treatment group are slightly older than the entire sample. Such differences prove inconsequential in later analyses. 24

sample was 53.3 percent (standard error = 1.4), which did not differ dramatically from the

generic GOTV message turnout rate of 55.34 percent (s.e. = 4.92). The level of statistical

uncertainty is higher in this single county study due to the smaller sample sizes, which average

just over 100 registered voters per group. That means the GOTV group does not differ from the

untreated cases, although once again this comparison is potentially biased and will be refined in a

moment. The turnout rates reported in column 2 of Table 4 all hover in the mid-50s with two exceptions: tobacco settlement (turnout = 64.89) and the homestead exemption (63.55). As with

Study 1, the amendments are presented in order from least to most salient. Thus it is on the amendments toward the top of the list that we expected and indeed saw that turnout rates were the highest after receipt of a postcard reminder that these issues were on the ballot. In fact the average turnout rate among three least salient initiatives in Leon County—tobacco, eminent domain, and homestead exemption—was 61.7 percent. The average was about 5-6 percentage points lower among the three least salient initiatives in the lower portion of Table 4.

Insert Table 4 here.

The key part of table 4 concerns the treatment effect calculations in columns 3, 4, and 5.

With each placebo comparison group (GOTV, other amendments, or a combination of the two),

the tobacco settlement treatment conditions registers statistically significant increases.17 The difference between the tobacco settlement amendment condition and the GOTV group is nearly

10 percentage points (9.55, p <.10). The effect is a bit more muted with the other comparison groups, at 7.25 and 7.57 percentage points for the other amendments and the others plus GOTV

17 The GOTV placebo comparisons in column three of Table 4 compare each amendment group to the GOTV condition. The combined other amendment group features all other amendments except the tobacco settlement, which was judged to be different according to the GOTV analysis. Finally, the last column uses the GOTV group plus any non-significant group used in the other amendment calculations. Given the similarity of the Florida statewide field experiment to the one conducted in Leon County, it is possible to pull out respondents in the statewide sample who happen to reside in Leon County. When we do so, the results are the same with respect to sign and significance (i.e., the tobacco amendment group is consistently above the other placebo controls as in Table 4). 25

respectively. However, the standard error also drops as the sample size grows, which makes the

findings slightly more statistically significant (p < .05). The effects for most other amendments

are smaller or even negative. The only consistent exception to this is the homestead exemption. It

was among the top three least salient and the effects are nearly as large as we observed for the

tobacco amendment, but only with the comparison group of all other amendments does the effect

rise to statistical significance (p <. 10). Nonetheless, the average treatment effects for the three

least salient initiatives is between 4.10 and 6.38 for the three placebo comparison methods, while

it varies between .81 to -1.79 for the most salient initiatives. All of these patterns lend support to the argument that it is on the least salient initiatives that providing information can have the greatest effect. The fact that we were able to change several parameters of the analysis (i.e., starting with a higher baseline level of turnout, changing the configuration of salient amendments, adding a new amendment previously unstudied) and still find the expected patterns is reassuring, but in the next section we conduct one last test to evaluate the argument in other settings and with other initiatives at stake.

Study 3: National Surveys on Ballot Initiative Knowledge in 2006

We have argued that providing information about the presence of a ballot initiative in an upcoming election should make individuals more likely to vote. That effect should be magnified when the initiatives in question are not salient in the minds of voters. In the previous studies we supplied information about the presence of ballot initiatives experimentally, but in any given election some citizens possess this information while others do not. In one last attempt to probe for these patterns, we turn to an analysis of national survey data that contain measures relevant for testing our theoretical story. 26

In the fall of 2006, the Pew Research Center commissioned Princeton Survey Research

Associates International to field two public opinion polls on ballot initiatives in the United

States. The first nationally representative telephone survey was in the field from September 21 to

October 4, 2006 with 1,804 adults selected at random while the second included 2,006 respondents from October 17 to October 22, 2006.18 In addition two items relating to intended

turnout that will serve as our dependent variables, the surveys included the following question:

“From what you have heard or read, will voters in your state this November be voting on any

ballot initiatives, referendums, or state constitutional amendments, or not?” While it does not

explicitly test the salience part of our argument, respondents who are aware of initiatives in the upcoming election should more likely to turnout.19

The first and third columns of Table 5 confirm that this is indeed the case. Knowledge of

an upcoming ballot initiative is positively and significantly related to two forms of intended vote.

The first, which we have dubbed “plan to vote,” ask respondents, “Do you yourself plan to vote in the election this November?” Anyone who indicates they plan to vote or who has already voted by the time the survey was in the field was score in the top category of one. Individuals who said “no” or “don’t know” were scored as a zero while anyone who was not registered to vote, and thus was not asked the question, was treated as missing data. The answers to this question were highly skewed, much like the responses to another vote intention question a few items later in the survey asking registered vote to rate their chances of voting on a ten point scale.20 More 96 percent of the respondents claimed they planned to vote (96.3 percent) and the

18 The second survey included an oversample of competitive congressional districts, but sampling weights are employed in the analysis to make the analyses nationally representative. 19 See Donovan, Tolbert, and Smith (2009) for another analysis making use of these data for different purposes. 20 The ten-point scale question was, “I'd like you to rate your chances of voting in November on a scale of 10 to 1. If TEN represents a person who DEFINITELY will vote and ONE represents a person who definitely will NOT vote, where on this scale of ten to one would you place yourself?” This is the second dependent variable called “Chances 27

average person rated their chances of voting at more than 90% (an average of 9.3 on the ten point

scale). Such social desirability is common in surveys on political participation and it only serves

to make the analysis that much more difficult; explaining variation in a variable that does not

vary is challenging. Nonetheless, the primary variable of relevance to our analysis, Initiative

Knowledge, is positive and significant in both models (p < .05).

Insert Table 5 here.

The second and fourth columns of Table 5 show that the patterns hold even when we

control for demographic and political factors—education, income, age, black, Hispanic, female,

employed, married, and Democrat or Republican.21 The coefficient on Initiative Knowledge in

the “plan to vote” model is .44 with a standard error of .12 (p < .05). This corresponds to roughly

a two point difference between those who know whether an amendment is on the ballot versus

those who do not (first difference = 2.2, with an s.e.=.60). Other factors in the model with

significant coefficients are almost as large or larger. The estimated effect in moving from two

standard deviations below the mean to two standard deviations above it for education is 6.4

percentage points (s.e.=2.4), for age it is 10 points (s.e.=2.2), for female it is about 2 points (-

1.5, s.e.=1.0), for Democrats it is 2.9 points (s.e.=.57) and for Republicans it is 4.1 points

(s.e.=.77). For the “chances of voting” outcome, moving from not knowing whether initiatives will be on the ballot to knowing this increases the chances of voting for the typical person by .04,

which is small but it is important to remember the highly skewed nature of the variable. This

of Voting” and it was recoded so range from 0 (definitely will not vote) to 1 (definitely will vote) will all of the ordinal categories in between and don’t know or refusal turned to missing. 21 Each variable is coded on a zero to one scale with the highest categories as follows: education (1=post-graduate education), income (1=$150,000 or more in family income), age (1=age 94 or older), black (1=African American), Hispanic (1=Hispanic), female (1=female), employed (1=employed full time), married (1=married), Democrat (1=Democrat), and Republican (1=Republican). Missing demographic responses for education, income, or age were recovered via multiple imputation with the Amelia II software program (King et al. 2001). 28

effect remains diminishes only slightly to .03 when we control for the same battery of demographic variables and partisanship.22

Thus, knowing whether initiatives will be on the ballot is consistently related to intention

to vote measured two different ways in nationally representative surveys. Naturally, given the

cross-sectional nature of the data, this part of the analysis is admittedly more speculative with

respect to causality. However, it does have the virtue of broadening our analysis to other states to

with ballot initiatives. So while the findings are not definitive, they suggest that there is a

connection between knowing that initiatives are on the ballot and intentions to vote.23

Conclusions

Elections are the primary way that citizens express their policy preferences, but voting for candidates permits only indirect forms of representation. Occasionally, members of the public vote directly on issues. There is work on how ballot measures influence vote choices (Nicholson

2005; Smith, DeSantis, and Kassel 2006) and turnout (Smith 2001; Tolbert, Grummel and Smith

2001; et al.), but we investigated the degree to whether knowledge of ballot initiatives influenced voter turnout.

We have argued and found that information on ballot initiatives increases turnout, particularly on items that are the least salient in the minds of the voter such as the item that addressed the organization of the state budget and planning process in the statewide Florida

22 Since the model with the ten-point chances of voting measure as the dependent variable is estimated with ordinary least squares regression, the values the fourth column of Table 5 can be interpreted directly as the effect of unit changes in the independent variables on the dependent variable. However, we obtain the same pattern of results in an ordered probit version of the model. 23 The initiative knowledge item is scored 1 if the respondent is in a state with a initiative on the ballot and the person knows this or if they are in a state without an initiative and they know there is not one on the ballot. If we reestimate the model with self-rated beliefs of having an initiative on the ballot irrespective of whether there is one in reality, the coefficient for the plan to vote model remains positive and significant but it diminishes in the chances to vote model (p < .10 instead of p < .05). Thus, accuracy matters, but only somewhat. The belief that initiatives are on the ballot appears to matter almost as much. 29 study or the tobacco settlement item in the study conducted in Leon County, Florida. However, our framework is flexible. In another election or in another state the issues that voters care about might be different. Discerning these effects has been a concern in past work. As Lassen puts it,

“The problem is that information acquisition is endogenous and, therefore, both the decision to vote and the decision to obtain an education or become informed about political issues can be caused by some third, unobservable factor” (104). Lassen adopts a natural-experiment while others employ statistical controls (Smith 2001; Tolbert, McNeal, and Smith 2003; Smith and

Tolbert 2004). Yet, precise causal statements concerning the effects of information on initiatives on turnout are not possible, or at the very least they must be heavily qualified and they rely on a host of assumptions. To bypass these potential problems and concerns, we conducted two field experiments. The results were consistent with expectations and we found that knowledge of ballot amendments was also related to intended turnout in cross-sectional surveys.

Of course, future studies on different initiatives in different states are needed to confirm these results. We also use one treatment mode, the mail, for which past studies have shown is one of the least powerful methods of mobilizing voters (Gerber and Green 2000a; 2000b; 2001;

2003; cf. Gerber, Green, Larimer 2008; Panagopoulous and Green 2008). Mail field experiments typically register effect sizes on the order of a half of a percentage point. That means the sample sizes we employed were low in statistical power to detect expected effects. Only when the effects were several times bigger than a typical mail experiment did we observe statistical significance, albeit even then at levels below conventional standards. Moreover, the comparisons we employed using the placebo controls made it somewhat harder to observe the effects we anticipated. All of this is to say that the findings make it seem as if providing information mostly does not have an effect, but we were somewhat conservative in our approach. Had we used larger 30 samples or different modes of experimental contact (e.g., telephone or in-person), reminding voters about the presence of ballot initiatives might have been more effective.

On that note, there are a lot of other dimensions that were not tested here. For example, there is a debate about whether ballot initiatives only work in midterm contests or in presidential elections too. As the past research on both ballot initiatives and field experiments has found, the visibility of the election contest moderates these effects. One should expect smaller treatment effects in high interest elections when citizens are already mobilized, or as Tolbert and her colleagues write, “Midterm elections are low information elections with very few sources of mobilization, thus making the electorate more sensitive to those sources of mobilization that do exist, such as the initiative process” (Tolbert, Grummel, and Smith 2001, 632). Since the 2006 general election was technically a midterm contest, it holds this factor constant. However, the

Florida election also featured gubernatorial contest for an open seat so it was not as obscure as some midterms contests are.

In that sense, it is important to note that initiative salience could still be influencing turnout because the natural world was providing “treatments” to voters. Indeed, we observed as much in the study of national surveys in 2006 because knowledge of the initiatives was positively related to intended turnout. Across elections and states, however, information about the issues is likely to remain an important determinant of who votes. A candid revelation of this appeared in a random sample survey of 800 Florida voters less than a year before the 2006 election from Dec. 7 to 24, 2005. The survey conducted by the local firm of Susan Schuler &

Associates asked a series of items on the amendment process. Of relevance to the study here,

75% of those polled said the wording of constitutional amendments is “generally confusing.” On another item, 46% of those surveyed said they “did not get enough information,” which was the 31

modal response compared those who said they only heard one side (17%), some of each side

(2%), got enough information (33%) or did not know (2%). In our field experiments we

attempted to provide information about the presence of ballot amendments to make them more

salient in the minds of the voters. As we learned, not only do citizens want more information, but

providing it can increase their levels of civic participation.

Scholars have been fascinated by the initiative process for years and there is evidence that

policy more closely matches public preferences in states with direct democracy (Gerber 1996;

Matsusaka 1995; cf. Lascher, Hagan, and Rochlin 1996). In the end, though, a lot depends on

voters being willing to vote, or at least giving legislators the impression they will. Initiatives not

only help convey citizen preferences on policy issues, but they can lead to high turnout.

However, providing issue information to mobilize citizens to vote works best when voters lack information about initiatives on the ballot.

32

Appendix A Summary Descriptions of Statewide Amendments from the Florida Division of Elections

AMENDMENT 1: STATE PLANNING AND BUDGET PROCESS

Proposing amendments to the State Constitution to limit the amount of nonrecurring general revenue which may be appropriated for recurring purposes in any fiscal year to 3 percent of the total general revenue funds estimated to be available, unless otherwise approved by a three-fifths vote of the

Legislature; to establish a Joint Legislative Budget Commission, which shall issue long-range financial outlooks; to provide for limited adjustments in the state budget without the concurrence of the full

Legislature, as provided by general law; to reduce the number of times trust funds are automatically terminated; to require the preparation and biennial revision of a long-range state planning document; and to establish a Government Efficiency Task Force and specify its duties.

Passed; Votes For: 2,570,436, Votes Against: 1,724,867

AMENDMENT 3: REQUIRING BROADER PUBLIC SUPPORT FOR CONSTITUTIONAL AMENDMENTS OR REVISIONS

Proposes an amendment to Section 5 of Article XI of the State Constitution to require that any proposed amendment to or revision of the State Constitution, whether proposed by the Legislature, by initiative, or by any other method, must be approved by at least 60 percent of the voters of the state voting on the measure, rather than by a simple majority. This proposed amendment would not change the current requirement that a proposed constitutional amendment imposing a new state tax or fee be approved by at least 2/3 of the voters of the state voting in the election in which such an amendment is considered.

Passed; Votes For: 2,600,969, Votes Against: 1,900,359

AMENDMENT 4: PROTECT PEOPLE, ESPECIALLY YOUTH, FROM ADDICTION, DISEASE, AND OTHER HEALTH HAZARDS OF USING TOBACCO

To protect people, especially youth, from addiction, disease, and other health hazards of using tobacco, the Legislature shall use some Tobacco Settlement money annually for a comprehensive statewide tobacco education and prevention program using Centers for Disease Control best practices. It specifies some program components, emphasizing youth, requiring one-third of total annual funding for 33 advertising. Annual funding is 15% of 2005 Tobacco Settlement payments to Florida, adjusted annually for inflation. Provides definitions. Effective immediately.

Passed; Votes For: 2,786,935, Votes Against: 1,787,230

AMENDMENT 6: INCREASED HOMESTEAD EXEMPTION

Proposing amendment of the State Constitution to increase the maximum additional homestead exemption for low-income seniors from $25,000 to $50,000 and to schedule the amendment to take effect

January 1, 2007, if adopted.

Passed; Votes For: 3,533,101, Votes Against: 1,092,128

AMENDMENT 7: PERMANENTLY DISABLED VETERANS' DISCOUNT ON HOMESTEAD AD VALOREM TAX

Proposing an amendment to the State Constitution to provide a discount from the amount of ad valorem tax on the homestead of a partially or totally permanently disabled veteran who is age 65 or older who was a Florida resident at the time of entering military service, whose disability was combat-related, and who was honorably discharged; to specify the percentage of the discount as equal to the percentage of the veteran's permanent service-connected disability; to specify qualification requirements for the discount; to authorize the Legislature to waive the annual application requirement in subsequent years by general law; and to specify that the provision takes effect December 7, 2006, is self-executing, and does not require implementing legislation.

Passed; Votes For: 3,552,441, Votes Against: 1,011,958

AMENDMENT 8: EMINENT DOMAIN

Proposing an amendment to the State Constitution to prohibit the transfer of private property taken by eminent domain to a natural person or private entity; providing that the Legislature may by general law passed by a three-fifths vote of the membership of each house of the Legislature permit exceptions allowing the transfer of such private property; and providing that this prohibition on the transfer of private property taken by eminent domain is applicable if the petition of taking that initiated the condemnation proceeding was filed on or after January 2, 2007. Passed; Votes For: 3,047,420, Votes Against: 1,365,950 34

Appendix B

Sample Mail Message and Mail Treatment Scripts

November 7, 2006 is Election Day in Florida.

Amendment 6 proposes to increase the homestead exemption for low income senior citizens from $25,000 to $50,000

Now more than ever, your community, state, and nation need to hear what you think.

Please remember to vote on November 7.

Project Vote 2006 [UNIVERSITY ADDRESS OMITTED]

Other Messages Used In Place of the Italicized Passage Above

Amendment 1 proposes to alter the State’s planning and budget process in some important ways.

Amendment 3 proposes to require a 60 percent vote rather than a simple majority to amend the Florida constitution.

Amendment 4 proposes to use at least 15% of the money the State receives annually from the 2005 tobacco settlement for programs that are designed to protect people from the dangers of tobacco.

Amendment 7 proposes to provide certain partially or totally disabled combat veterans aged 65 and older discounts on their homestead taxes.

Amendment 8 proposes to limit the state’s ability to transfer property taken by eminent domain to a private person or private entity.

35

References

Alvarez, R. Michael. 1997. Information and Elections. Ann Arbor: University of Michigan Press.

Barber, Benjamin. Strong Democracy: Participatory Politics for a New Age. Berkeley, CA: University of California Press.

Bartels, Larry M. 1986. “Issue Voting Under Uncertainty: An Empirical Test.” American Journal of Political Science 30 (Nov.): 709-28.

Berinsky, Adam J. 2005. “The Perverse Consequences of Electoral Reform in the United States.” American Politics Research 33: 471-91.

Boehmke, Fredrick. 2002. “The Influence of Direct Democracy on the Size and Diversity of State Interest Group Populations.” Journal of Politics 64: 827-44.

Bowler, Shaun, Todd Donovan, and Caroline Tolbert. 1998. Citizens as Legislators: Direct Democracy in the United States. Columbus, OH: Ohio State Press.

Burden, Barry C. 2003. Uncertainty in American Politics. New York: Cambridge.

Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes. 1960. The American Voter. New York: Wiley.

Carmines, Edward G., and James A. Stimson. 1980. “The Two Faces of Issue Voting.” American Political Science Review 74 (Mar.): 78-91.

Cronin, Thomas E. 1989. Direct Democracy: The Politics of Initiative, Referendum, and Recall. Cambridge, MA: Harvard University Press.

Donovan, Todd, and Shaun Bowler. 1998. “An Overview of Direct Democracy in the American States.” In Citizens as Legislators: Direct Democracy in the United States eds. S. Bowler, T. Donovan, and C. Tolbert. Columbus: Ohio State University Press, pp. 1-26.

Donovan, Todd, Caroline J. Tolbert, and Daniel A. Smith. 2009. “Political Engagement, Mobilization, and Direct Democracy.” Public Opinion Quarterly 73 (1): 98-118.

Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper & Row.

Eldersveld, Samuel J. 1956. “Experimental Propaganda Techniques and Voting Behavior.” American Political Science Review 50 (1): 154-65.

Enelow, James M., and Melvin J. Hinich. 1984. The Spatial Theory of Voting: An Introduction. New York: Cambridge University Press.

36

Everson, David H. 1981. “The Effects of Initiatives on Voter Turnout: A Comparative State Analysis.” Western Political Quarterly 34: 415-25.’

Franklin, Charles H. 1991. “Eschewing Obfuscation? Campaigns and the Perceptions of U.S. Senate Incumbents.” American Political Science Review 85: 1193-1214.

Gaines, Brian J., James H. Kuklinski, and Paul J. Quirk. 2007. “The Logic of the Survey Experiment Reexamined.” Political Analysis 15 (Winter): 1-20.

Gerber, Alan S., and Donald P. Green. 2002. “Partisan Mail and Voter Turnout: Results from Randomized Field Experiments.” Electoral Studies. 22: 563-579.

Gerber, Alan S., and Donald P. Green. 2001. “Do Phone Calls Increase Voter Turnout?: An Experiment.” Public Opinion Quarterly 65 (1): 75-85.

Gerber, Alan S., and Donald P. Green. 2000a. “The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment.” American Political Science Review 94 (3): 653-63.

Gerber, Alan S., and Donald P. Green. 2000b. “The Effect of a Nonpartisan Get-Out-the-Vote Drive: An Experimental Study of Leafletting.” Journal of Politics 62 (3): 846-57.

Gerber, Alan S., Donald P. Green, and Christopher W. Larimer. 2008. “Social Pressure and Voter Turnout: Evidence from a Large-Scale Field Experiment.” American Political Science Review 102 (February): 33-48.

Gerber, Elizabeth. 1999. The Populist Paradox: Interest Group Influence and the Promise of Direct Legislation. Princeton: Princeton University Press.

Gerber, Elizabeth. 1996. “Legislative Response to the Threat of Popular Initiatives.” American Journal of Political Science 40: 99-128.

Gilliam, Franklin D., Jr. 1985. “Influences on Voter Turnout for U.S. House Elections in Non- Presidential Years.” Legislative Studies Quarterly 10: 339-51.

Gopoian, J. David. 1993. “Images and Issues in the 1988 Presidential Election.” Journal of Politics 55 (February): 151-66.

Gosnell, Harold F. 1927. Getting Out the Vote: An Experiment in the Stimulation of Voting. Chicago: University of Chicago Press.

Green, Donald P., Alan S. Gerber, and David W. Nickerson. 2003. “Getting Out the Vote in Local Elections: Results from Six Door-to-Door Canvassing Experiments.” Journal of Politics 65 (4): 1083-96.

37

Grosser, Jens, and Arthur Schram. 2006. “Neighborhood Information Exchange and Voter Participation: An Experimental Study.” American Political Science Review 100 (2):235- 48.

Hafenbrack, Josh. 2006. “Crist Pledges Property Tax Cut; Republican Brings Campaign to Wellington.” South Florida Sun-Sentinel, Sept. 27th, p. 3B.

Hull, Victor. 2006. “6 Constitutional Amendments Pass.” Sarasota Herald-Tribune Section A, p. 16.

King, Gary, James Honaker, Anne Joseph, and Kenneth Scheve. 2001. “Analyzing Incomplete Political Science Data: An Alternative Algorithm for Multiple Imputation.” American Political Science Review 95(March): 49–70.

Lacey, Robert J. 2005. “The Electoral Allure of Direct Democracy: The Effect of Initiative Salience on Voting, 1990-96.” State Politics and Policy Quarterly 5 (Winter): 168-81.

Lascher, Edward, Michael Hagen, and Steven A. Rochlin. 1996. “Gun Behind the Door? Ballot Initiatives, State Policies, and Public Opinion.” Journal of Politics 58 (Aug.): 760-775.

Lassen, David Dreyer. 2005. “The Effect of Information on Voter Turnout: Evidence from a Natural Experiment.” American Journal of Political Science 49 (Jan.): 103-88.

Magleby, David. 1994. “Direct Legislation in the American States.” In Referendums around the World: the Growing Use of Direct Democracy, eds. David Butler and Austin Ranney. Washington, D.C.: AEI Press.

Magleby, David. 1984. Direct Legislation: Voting on Ballot Propositions in the United States. Baltimore: Johns Hopkins University Press.

Matsusaka, John G. 2004. For the Many or the Few: The Initiative, Public Policy, and American Democracy. Chicago: University of Chicago Press.

Matsusaka, John. 1995. “Fiscal Effects of the Voter Initiative: Evidence from the Last 30 Years.” Journal of Political Economy 103: 587-623.

Michelson, Melissa R. 2006. “Mobilizing the Latino Youth Vote: Some Experimental Results.” Social Science Quarterly 87 (Dec.): 1188-1206.

Michelson, Melissa R. 2006b. “Mobilizing Latino Voters for a Ballot Proposition.” Latino(a) Research Review 6 (Summer): 1-2.

Michelson, Melissa R. 2003. “Getting Out the Latino Vote: How Door-to-Door Canvassing Influences Voter Turnout in Rural Central California.” Political Behavior 25 (Sept.): 247-63.

38

Michelson, Melissa R. 2005. “Meeting the Challenge of Latino Voter Mobilization.” The Annals of the American Academy of Political and Social Science 601 (Sept.): 85-101.

Miller, Roy E., David A. Bositis, and Denise L. Baer. 1981. “Stimulating Voter Turnout in a Primary: Field Experiment with a Precinct Committeeman.” International Political Science Review 2 (4): 445-60.

Nicholson, Stephen P. 2005. Voting the Agenda: Candidates, Elections, and Ballot Propositions. Princeton: Princeton University Press.

Nickerson, David W. 2007. “Quality is Job One: Professional and Volunteer Voter Mobilization Calls.” American Journal of Political Science 51 (April): 269-82.

Nickerson, David W. 2006. “Volunteer Phone Calls Can Increase Turnout: Evidence from Eight Field Experiments.” American Politics Research 34 (3): 271-92.

Nickerson, David W. 2005. “Scalable Protocols Offer Efficient Design for Field Experiments.” Political Analysis 13 (3): 233-52.

Nie, Norman, Sidney Verba, and John R. Petrocik. 1979. The Changing American Voter. Cambridge, MA: Harvard University Press.

Niven, David. 2006. “A Field Experiment on the Effects of Negative Campaign Mail on Voter Turnout in a Municipal Election.” Political Research Quarterly 59 (June): 203-10.

Panagopoulous, Costas and Donald P. Green. 2008. “Name Recognition and Incumbency Advantage: Evidence from a Mass Media Field Experiment Targeting Latino Voters.” Paper Presented at the Annual Meeting of the American Political Science Association, Boston, MA.

Pateman, Carole. 1970. Participation and Democratic Theory. New York: Cambridge.

Rabinowitz, George, and Stuart Elaine Macdonald. 1989. “A Directional Theory of Issue Voting.” American Political Science Review 83 (March): 93-121.

Ramirez, Ricardo. 2005. “Giving Voice to Latino Voters: A Field Experiment on the Effectiveness of a National Nonpartisan Mobilization Effort.” Annals of Political and Social Science 601 (Sept.): 66-84.

Riker, William H., and Peter C. Ordeshook. 1968. “A Theory of the Calculus of Voting.” American Political Science Review 62 (1): 25-42.

Schmidt, David D. 1989. Citizen Lawmakers: The Ballot Initiative Revolution. Philadelphia: Temple University Press.

Sears, David, and Jack Citrin. 1982. Tax Revolt. Cambridge, MA: Harvard University Press.

39

Smith, Daniel A. 2004. “Peeling Away the Populist Rhetoric: Toward a Taxonomy of Anti-Tax Ballot Initiatives.” Public Budgeting and Finance Winter 88-110.

Smith, Daniel A. 1999. “Reevaluating the Causes of Proposition 13.” Social Science History 23 (2): 173-210.

Smith, Daniel A. 1998. Tax Crusaders and the Politics of Direct Democracy. New York: Routledge.

Smith, Daniel A., Matthew DeSantis, and Jason Kassel. 2006. “Same-Sex Marriage Ballot Measures and the 2004 Presidential Election.” State and Local Government Review 38 (2): 78-91.

Smith, Daniel A., and Caroline J. Tolbert. 2004. Educated by Initiative: The Effects of Direct Democracy on Citizens and Political Organizations in the American States. Ann Arbor, MI: University of Michigan Press.

Smith, Mark, A. 2001. “The Contingent Effects of Ballot Initiatives and Candidate Races on Turnout.” American Journal of Political Science 45 (Jul.): 700-06.

Teixeira, Ruy. 1992. The Disappearing American Voter. Washington D.C.: Brookings.

Trivedi, Neema. 2005. “The Effect of Identity-Based GOTV Direct Mail Appeals on the Turnout of Indian Americans.” The Annals of the American Academy of Political and Social Science 601 (Sept.): 115-22.

Tolbert, Caroline J., John A. Grummel, and Daniel A. Smith. 2001. “The Effects of Ballot Initiatives on Voter Turnout in the American States.” American Politics Research 29 (Nov.): 625-48.

Tolbert, Caroline J., Daniel H. Lowenstein, and Todd Donovan. 1998. “Election Law and Rules for Using Initiatives.” In Citizens as Legislators: Direct Democracy in the United States eds. S. Bowler, T. Donovan, and C. Tolbert. Columbus: Ohio State University Press, pp. 27-54.

Tolbert, Caroline J., Ramona S. McNeal, and Daniel A. Smith. 2003. “Enhancing Civic Engagement: The Effect of Direct Democracy on Political Participation and Knowledge.” State Politics and Policy Quarterly 3 (Spring): 23-41.

Tolbert, Caroline J., and Daniel A. Smith. 2005. “The Educative Effects of Ballot Initiatives on Voter Turnout.” American Politics Research 33 (March): 283-309.

Wong, Janelle. 2005. “Mobilizing Asian American Voters: A Field Experiment.” The Annals of the American Academy of Political and Social Science 601 (Sept.): 102-14.

Zimmerman, Joesph. 1999. The Initiative: Citizen Law-Making. Westport, CT: Praeger. 40

Table 1. Randomization Check Comparing Statewide Demographic Characteristics to Treatment Conditions Any Treatment Generic Turnout State Planning & Budget Eminent Domain Statewide Difference Difference Difference Difference Average (s.e.) Ave.from State Ave. from State Ave. from State Ave. from State Age 48.70 (.01) 49.14 .43 (.25) * 48.77 .06 (.65) 49.87 1.16 (.65) * 48.81 .11 (.64) Female 53.06 (.02) 53.97 .90 (.68) 54.22 1.16 (1.81) 53.62 .55 (1.79) 53.48 .42 (1.79) Black 12.77 (.01) 12.84 .06 (.46) 12.80 .02 (1.21) 14.60 1.82 (1.27) 12.76 -.02 (1.20) Hispanic 11.11 (.01) 10.69 -.42 (.42) 8.58 -2.54 (1.02) ** 10.08 -1.03 (1.08) 11.34 .23 (1.14) Democrat 40.89 (.02) 41.24 .36 (.67) 38.26 -2.63 (1.77) 44.57 3.69 (1.79) ** 40.08 -.81 (1.76) Republican 36.09 (.02) 35.55 -.55 (.65) 36.94 .85 (1.75) 33.33 -2.76 (1.70) 36.98 .89 (1.73) Cases 9,278,872 5,407 758 774 776

Supermajorities Tobacco Settlement Homestead Exemption Disabled Veterans Statewide Difference Difference Difference Difference Average (s.e.) Ave. from State Ave. from State Ave. from State Ave. from State

Age 48.70 (.01) 48.80 .10 (.67) 48.85 .14 (.63) 50.00 1.30 (.65) ** 48.80 .09 (.66) Female 53.06 (.02) 53.46 .40 (1.80) 52.33 -.73 (1.80) 56.05 2.99 (1.75) * 54.53 1.47 (1.81) Black 12.77 (.01) 11.63 -1.14 (1.16) 12.31 -.46 (1.18) 13.98 1.21 (1.23) 11.70 -1.08 (1.17) Hispanic 11.11 (.01) 10.85 -.26 (1.13) 10.10 -1.01 (1.09) 11.24 .13 (1.12) 12.61 1.50 (1.20) Democrat 40.89 (.02) 41.83 .94 (1.78) 40.67 -.21 (1.77) 42.82 1.94 (1.75) 40.34 -.54 (1.78) Republican 36.09 (.02) 34.51 -1.58 (1.72) 35.36 -.73 (1.72) 35.21 -.89 (1.69) 36.53 .44 (1.75)

Cases 9,278,872 765 772 801 761 ** p < .05, * p < .10, two-tailed.

41

Table 2. Florida Statewide Field Experiment Turnout Rates and Treatment Effects 1 2 3 4 5 Received Turnout Treatment Effect Treatment Effect Treatment Effect Treatment Rate vs. GOTV vs. Other Amends. vs. Others & GOTV Experimental Group n % (s.e.) % (s.e.) % (s.e.) % (s.e.) Get-Out-the-Vote (GOTV) 758 42.61 (1.80) -- -- .52 (1.96) -- -- State Planning & Budgeting 774 46.77 (1.79) 4.16 (2.54) * 4.68 (1.95) ** 4.59 (1.92) ** Eminent Domain 776 41.24 (1.77) -1.38 (2.52) -1.07 (1.98) -1.13 (1.94) Supermajorities 765 43.40 (1.79) .79 (2.54) 1.63 (1.99) 1.46 (1.95) Tobacco Settlement 772 40.54 (1.77) -2.07 (2.52) -1.93 (1.99) -1.96 (1.95) Homestead Exemption 801 42.95 (1.75) .33 (2.51) 1.08 (1.96) .93 (1.92) Disabled Veterans 761 42.31 (1.79) -.30 (2.54) .28 (2.00) .16 (1.95) ** p < .05, * p < .10, one-tailed.

42

Table 3. Issue-based Turnout in the 2006 General Election: Florida Statewide Field Experiment Placebo Controls Prior Voting History Included Salience Weighted Treatments Only Other Others Only Other OthersOnly Other Others GOTV Amend. & GOTV GOTV Amend. & GOTV GOTV Amend. & GOTV Experimental Treatment .09 * .11 ** .11 ** .12 * .12 ** .12 ** .12 * .10 ** .08 ** (.07) (.05) (.05) (.08) (.06) (.06) (.08) (.06) (.05) Age .02 ** .02 ** .02 ** .01 ** .01 ** .01 ** .01 ** .01 ** .01 ** (.00) (.00) (.00) (.00) (.00) (.00) (.00) (.00) (.00) Female -.03 -.04 -.04 -.03 -.09 ** -.10 ** -.03 -.09 ** -.10 ** (.07) (.04) (.04) (.08) (.04) (.04) (.08) (.04) (.04) Black -.26 ** -.21 ** -.20 ** -.22 ** -.22 ** -.19 ** -.22 ** -.22 ** -.19 ** (.10) (.06) (.06) (.12) (.07) (.06) (.12) (.07) (.06) Hispanic -.31 ** -.24 ** -.26 ** -.13 -.16 ** -.16 ** -.13 -.17 ** -.16 ** (.12) (.06) (.06) (.14) (.07)(.07) (.14) (.07) (.07) Democrat .29 ** .28 ** .27 ** .07 .05 .04 .07 .04 .03 (.09) (.05) (.05) (.10) (.06) (.05) (.10) (.06) (.05) Republican .26 ** .35 ** .33 ** -.05 .05 .02 -.05 .04 .02 (.09) (.05) (.05) (.10)(.06) (.05) (.10) (.06) (.05) Voted in 2006 Primary ------.72 ** .99 ** .89 ** .72 ** .98 ** .88 ** (.31) (.21) (.19) (.31) (.21) (.19) Voted in 2004 ------.21 * .24 ** .26 ** .21 * .24 ** .26 ** (.14)(.08) (.07) (.14) (.08) (.07) Voted in 2002 ------.45 ** .42 ** .42 ** .45 ** .42 ** .42 ** (.16) (.09) (.08) (.16) (.09) (.08) Voted in 2000 ------.42 ** .28 ** .31 ** .42 ** .28 ** .31 ** (.14) (.08) (.07) (.14) (.08) (.07) Number of cases 1,532 4,649 5,407 1,532 4,649 5,407 1,532 4,649 5,407 Note : The entries are probit coefficients with robust standard errors in parentheses. The standard errors have been clustered to account for voters who live at the same address. The dependent variable is turnout in the 2006 Florida general election (1=voted; 0=did not vote). The constant and terms for absention in each prior election have been suppressed for presentation purposes. ** p < .05, * p < .10 (one-tailed).

43

Table 4. Leon County Field Experiment Turnout Rates and Treatment Effects 1 2 3 4 5 Received Turnout Treatment Effect Treatment Effect Treatment Effect Treatment Rate vs. GOTV vs. Other Amends. vs. Others & GOTV Experimental Group n % (s.e.) % (s.e.) % (s.e.) % (s.e.) Get-Out-the-Vote (GOTV) 103 55.34 (4.92) -- -- -2.30 (5.26) -- -- Tobacco Settlement 94 64.89 (4.95) 9.55 (6.99) * 7.25 (5.45) * 7.57 (5.40) * Eminent Domain 97 56.70 (5.06) 1.36 (7.06) -1.11 (5.46) -.71 (5.40) Homestead Exemption 107 63.55 (4.67) 8.21 (6.78) 7.12 (5.25) * 5.45 (5.16) Disabled Veterans 100 57.00 (4.98) 1.66 (6.99) .76 (5.30) -.36 (5.33) State Planning & Budgeting 107 57.01 (4.81) 1.67 (6.88) -.76 (5.25) -.36 (5.18) Supermajorities 107 53.27 (4.85) -2.07 (6.90) -5.27 (5.25) -4.74 (5.18) County Tax for Health Care 110 58.18 (4.72) 2.84 (6.82) .65 (5.19) 1.01 (5.12) * p < .10, one-tailed. 44

Table 5. Ballot Initiatives and Intended Turnout: 2006 Pew Survey Data Plan to Vote Chances of Voting Initiative Knowledge .51 ** .44 ** .04 ** .03 ** (.12) (.12) (.01) (.01) Education -- .63 * -- .07 ** (.38) (.02) Income -- .05 -- .02 (.32) (.03) Age -- 1.44 ** -- .18 ** (.27) (.03) Black -- .04 -- .01 (.24) (.02) Hispanic -- -.15 -- -.04 * (.23) (.03) Female -- -.29 ** -- .01 (.15) (.01) Employed -- .12 -- .03 ** (.14) (.01) Married -- -.04 -- .01 (.13) (.01) Democrat -- .30 ** -- .03 ** (.14) (.01) Republican -- .34 * -- .02 (.18) (.02) Constant 1.58 ** .66 ** .90 ** .73 ** (.08) (.25) (.01) (.03) Number of cases 2,521 2,521 1,776 1,776 Source: 2006 Pew Research Center national telephone surveys from. Sept. 21-Oct. 4, 2006 (n =1,804) and Oct. 17-22, 2006 (n =2,006). Note : The entries are probit or ordinary least squares regression coefficients in the first two or last two columns respectively with robust standard errors in parentheses. The standard errors have been clustered to account for voters who live in the same state. The first dependent variable, Plan to Vote, is intention to turnout in the 2006 general election (1=plans to vote; 0=does not plan to vote/don't know). The second dependent variable, Chances of Voting , asks respondents to rate their chances of voting in the November 2006 election on a ten point scale (1=definitely will vote; 0=definitely will not vote). The dependent variables were asked of registered voters only. The number of cases declines in the second model because the dependent variable was asked only of a random half of the sample in the first survey. ** p < .05, * p < .10 (two-tailed).

45

Figure 1. Issue-based Turnout: Experimental Effects 20

15

10 Percentage Point Point Percentage 5 Change in Likelihood of Voting of Likelihood in Change

0 Other Other Other Other Other Other & GOTV & GOTV & GOTV Amendments Amendments Amendments Amendments Amendments Amendments Get-Out-the- Get-Out-the- Get-Out-the- Vote (GOTV) Vote (GOTV) Vote (GOTV) Placebo Controls Prior Voting History Salience Weighted Included Treatments