Voters and competent decisions in direct democracy

Scholars have routinely observed that voters know very little about politics and have a limited (and often non-existent) understanding about the structure of the political world (e.g. Converse Reference Converse1964; for a review of what voters know, see Delli Carpini and Keeter Reference Delli Carpini and Keeter1996). In response to Converse's finding that voters did not organise their political beliefs according to an academic ideal, Brady and Sniderman (Reference Brady and Sniderman1985) and Sniderman et al. (Reference Sniderman, Brody and Tetlock1991) hypothesise that voters assess political choices by relying on their evaluations of groups. They argue that individuals employ “likability heuristics” – simple affective evaluations of groups such as liberals and conservatives – to arrive at reasonably informed decisions. In essence, voters are able to overcome some (but not all) of their informational deficiencies by using these simple affective assessments to inform their own choices about candidates and issues of the day. Since Brady and Sniderman's (Reference Brady and Sniderman1985) original argument, the belief that voters can make reasonable choices on political matters has gained substantial support in the political behaviour literature (e.g. Carmines and Kuklinski Reference Carmines and Kuklinski1990; Mondak Reference Mondak1993; Lupia Reference Lupia1994; Popkin Reference Popkin1994; Mutz Reference Mutz1998; Lupia et al. Reference Lupia, McCubbins and Popkin2000; Nicholson Reference Nicholson2011).

While much of the above research explores ways in which voters can use heuristics, the agency-theory framework that Lupia and McCubbins (Reference Lupia and McCubbins1998, Chapters 2–5) discusses advances a set of standards against which we can evaluate when and under what conditions heuristics will influence an individual's decisions. In particular, Lupia and McCubbins argue that voters (whom they call “principals”) can use information from cue-givers (whom they call “speakers”) when the cue-giver meets the following common-knowledge conditions: (1) the voter believes that the cue-giver shares a common interest over policy outcomes, or, lacking that, the cue-giver must undertake an observable and costly action to communicate a voting cue to the voter, or some external force or forces are strong enough to substitute for common interest (there is a penalty for lying or a threat of verification, thereby making the cue-giver trustworthy); and (2) the voter perceives the cue-giver to be knowledgeable about the subject matter (thus, the cue-giver is believed to have the knowledge the voter desires).Footnote ³ In other words, Lupia and McCubbins move beyond the blanket assumption that individuals can use information shortcuts freely, and instead propose that the individual's assessment of the cue-giver's credibility will determine whether the individual will incorporate the cue-giver's advice into a decision.

While information shortcuts have great potential to help voters make decisions, the realities of political campaigns place limits the ability of cue-givers to persuade voters. For instance, Garrett and McCubbins (Reference Garrett and McCubbins2008) find that many information shortcuts fail to meet the two basic conditions set forth by Lupia and McCubbins, with or without the existence of external persuasive forces, and are therefore not useful. On the other hand, Boudreau (Reference Boudreau2009a), in work that parallels the work of Lupia and McCubbins (Reference Lupia and McCubbins1998), Gigerenzer (Reference Gigerenzer2000, Reference Gigerenzer2007, Reference Gigerenzer2008), Gigerenzer and Selten (Reference Gigerenzer and Selten2001) and Gigerenzer et al. (Reference Gigerenzer and Todd1999), uses experimental evidence to show that unsophisticated voters can use cues so successfully that they can, in some instances, outperform more knowledgeable (sophisticated) voters. In a related study, Boudreau (Reference Boudreau2009b) validates the Lupia and McCubbins (Reference Lupia and McCubbins1998, 55) hypothesis that an individual may ignore credible cues if she thinks she is capable of making an independent decision (see also, Zaller Reference Zaller1992). Boudreau et al. (Reference Boudreau, McCubbins, Rodriguez and Weller2010) show that for individuals to become informed, cues must be cheap to acquire and, the problems have to be easy enough for individuals to understand, thereby allowing individuals to use their knowledge of cues to make an informed decision. In sum, information shortcuts can work, but it is unclear how often elections meet the conditions for voter persuasion.

Many scholars now incorporate information shortcuts as an essential component of vote choice models. These models assume that voters successfully and routinely overcome their information deficiencies by relying on simple cues (cf. Popkin Reference Popkin1994). In some instances, assuming voters’ widespread use of cues is plausible. Countless empirical studies have shown that party labels – the most pervasive cue in politics that happens to mirror how individuals organise and understand the political landscape (Green et al. Reference Green, Palmquist and Schickler2002) – are extraordinarily effective at informing voters about the positions of candidates. For partisan candidate contests, the cue of party identification works well, because it establishes common interest with voters (Downs Reference Downs1957, Chapter 8). Moreover, these labels are available on the printed ballot for candidate contests, which reduces the transaction costs for voters to access these cues. In other words, party labels are the easiest cue to understand and acquire for voters, and they are credible signals of the candidates’ positions. It comes as no surprise that voters often select their fellow partisan on the ballot at rates that often eclipse 90 per cent in congressional elections (e.g. Jacobson Reference Jacobson2008). Moreover, even for non-partisan elections, other aspects of the candidates – such as gender and occupation (McDermott Reference McDermott1997, Reference McDermott1998, Reference McDermott2006) – can substitute for partisanship, which is in line with what Brady and Sniderman (Reference Brady and Sniderman1985) theorise.

It is problematic to assume that voters’ use of cues in initiative and referendum elections is widespread, however. Unlike candidate contests, ballot measures do not come with the typical voting cues printed on the ballot. What is often available for voters is a short description of the initiative or referendum. California, for example, limits the description to 100 words. These brief descriptions may not be detailed enough for voters to make an informed decision (Burnett et al. Reference Burnett, Garrett and McCubbins2010, 319), or, perhaps worse, the text may be subject to elite manipulation (Burnett and Kogan Reference Burnett and Kogan2010). In other cases, the text that describes ballot measures appears in the form of a question, often providing little information about the actual substance of the legislation. The limited ballot text, predisposition of voters to spend very little effort to learn about politics and the direct impact that ballot measures have on policy heighten the importance of endorsements in direct democracy.

Complicating matters, the cognitive costs associated with using cues are much higher for voters in direct democracy when compared with candidate contests. In order for voters to use cues to make informed choices with respect to ballot measures, the following must be true: (1) the source of the cue and the communication environment satisfies both of Lupia and McCubbins’ conditions for persuasion, (2) voters must have learned about the voting cue before they vote and (3) voters recall the voting cue when they make their choice. Our expectation, then, is that voters will rely on information shortcuts to inform their votes on ballot measures far less often than they rely on information shortcuts to inform their decisions for national or statewide candidates for two reasons. First, voters may not learn of the relevant endorsements, or they may simply forget them. Second, many voters will not perceive the source of the endorsement to be credible (the endorsements will therefore not satisfy the two conditions necessary for persuasion).

For direct democracy, Lupia (Reference Lupia1994) is the first to offer empirical support showing that voters can use information shortcuts to approximate a fully informed vote on ballot measures. Lupia conducted an exit survey of Los Angeles voters during the 1988 general election in California to measure the electorate's awareness of prominent endorsers and opponents of five related ballot measures. In his survey, Lupia finds that voters who knew the information shortcuts (cues presented by endorsers) were able to use the cues to arrive at informed decisions on the five initiatives he surveyed (all of the measures dealt with automobile insurance reform). As Lupia and Matsusaka note about the seminal work, “Such evidence supports the claim that voters with apparently low levels of political information can use information shortcuts to emulate the voting behaviour they would have exhibited if they were as informed as the best-informed persons in the survey” (Lupia and Matsusaka, Reference Lupia and Matsusaka2004, 468).

As Bartels (Reference Bartels1996) observes and Lupia and Matsusaka (Reference Lupia and Matsusaka2004) confirm, however, very little empirical evidence examines whether voters, in practice, actually substitute information shortcuts for extensive factual knowledge. Despite the dearth of empirical evidence, scholars have prematurely designated Lupia's findings the conventional wisdom in the study of direct democracy. Recent empirical work on the topic, however, has begun to challenge the assumption that voters’ use of endorsements is widespread (Burnett et al. Reference Burnett, Garrett and McCubbins2010; Burnett and Parry Reference Burnett and Parry2012; Burnett and McCubbins Reference Burnett and McCubbins2013). This research aims to provide additional evidence concerning whether and how often voters use endorsements and their knowledge of specific facts to evaluate ballot measures. In particular, we establish that the effectiveness of endorsements is conditional on whether voters perceive the source of the endorsement to be credible.

Below, we test whether voters use voting cues to compensate for their lack of factual knowledge about ballot measures for two controversial initiatives that proposed broad policy changes that would affect millions of citizens in California. Using empirical data from an actual election, we test the hypothesis proposed by Lupia and McCubbins (Reference Lupia and McCubbins1998):

In essence, our first hypothesis states that we expect voters who have knowledge of a cue and perceive the cue-giver to be knowledgeable and trustworthy to vote differently when compared with voters who lack such knowledge. In our research design, which we outline below, we expect that a voter's ideology will allow the voter to discern the credibility of the source. For these voters, knowledge of an endorsement will significantly influence their decisions.

In addition, we examine the effects of factual knowledge on vote choice. Similar to Lupia (Reference Lupia1994), we expect that knowledge of a cue is as effective at informing voters' decisions as factual knowledge. In particular, our expectation is that knowledge of a fact will lead voters to decisions that are aligned with their preferences (e.g. Bartels Reference Bartels1996):

Similar to our first hypothesis, we anticipate that the ideological orientation of voters will affect how voters use factual information to arrive at a decision. We discuss these expectations fully in the research design and methods section.

Data

We use data collected during California's general election on 4 November 2008 to test our hypotheses. During this election, we assessed voters’ knowledge about two contentious propositions on the ballot, including their knowledge of endorsements from prominent cue-givers. We also asked respondents to report their vote choices and demographic information (e.g. party identification, income, education). We trained student volunteers to take interviews from voters as they exited the polling both. We surveyed 13 polling locations that covered 19 precincts. Our student volunteers collected 1,002 complete interviews and received 1,051 refusals, yielding a cooperation rate of 49 per cent. We instructed our student volunteers to ask every other exiting voter for an interview to randomise our sample.

We asked respondents about two initiatives that would have amended the California state constitution. Proposition 4 was an initiative that required medical officials to notify the parent(s) or legal guardian(s) at least 48 hours before performing an abortion on an unemancipated minor. Proposition 4 did not require parental or guardian consent. The initiative allowed for a number of exceptions: when the minor showed convincing evidence of maturity; when a court deemed that forgoing notification is in the best interest of the minor; when the parent(s) or guardian(s) had given previous consent; or in the case of a medical emergency. Additionally, Proposition 4 would have instituted mandatory reporting requirements and penalties for non-compliance for physicians.

Proposition 4 was a controversial initiative that had many proponents and opponents. Governor Schwarzenegger offered his support of the initiative on “Meet the Press” the June before the election. Despite this, he was a quiet supporter throughout the remainder of the campaign. The most vocal opponent of Proposition 4 was Planned Parenthood. Planned Parenthood's position satisfied the trustworthiness and knowledgeability conditions, because the group is a well-known and outspoken supporter of pro-choice policies. Thus, many voters shared a common interest with Planned Parenthood, and many others had conflicting interests. Both those who favoured and those who opposed pro-choice policies, therefore, could draw inferences about Proposition 4's policy impact from Planned Parenthood's position. Planned Parenthood also satisfied the knowledgeability condition, because the group presumably has expertise about the effects of abortion regulation. Accordingly, we asked voters the following question about Planned Parenthood's endorsement:

Proposition 8, the second measure we asked about, was an initiative to amend California's constitution to limit the definition of marriage to be between one man and one woman. Proposition 8 offered California voters the opportunity to overrule a decision by the California Supreme Court that had struck down Proposition 22 – an earlier, statutory initiative that limited marriage to be between one man and one woman – as unconstitutional. The court's written decision on Proposition 22 also interpreted the California Constitution to mean that same-sex couples had a legal right to marry.

Proposition 8 attracted wealthy and organised proponents and opponents. Both sides spent an incredible amount of money on their campaigns, totalling more than $108 million. The California Republican Party and Democratic Party took opposite positions on Proposition 8 (with the Republican Party in favour and the Democratic Party against the initiative). Both parties broadcasted their endorsements as part of a slate mailer. Voters often use political party endorsements as a cue on how to vote, because their long-established and well-known ideological positions make them trustworthy (relying on a pre-existing common interest with their members) and knowledgeable (political parties are experts on public policy). Thus, we measure our respondents’ use of the parties’ endorsements with the following questions:

For each ballot measure, we asked a factual knowledge question that requested respondents to recall details about the ballot measures, what Lupia (Reference Lupia1994) calls encyclopedic knowledge. For Proposition 4, we asked:

This question assessed whether voters knew that Proposition 4 only required parental notification and not parental consent for a minor to receive an abortion. Voters who had followed the campaign or had paid close attention to the campaign materials should have learned the correct answer to this question.

For Proposition 8, we assessed whether voters understood that constitutional amendments via the initiative process only require a simple majority to pass. Specifically, we asked:

Perhaps an attempt to mobilise their supporters, the opposition campaign emphasised how easy it was to change the constitution throughout their campaign. Astute readers will note that this factual knowledge question is not specific to Proposition 8 and applies to any constitutional amendment proposed via the initiative or referendum process. Given the straightforward nature of Proposition 8, however, there were few factual knowledge questions we could ask voters about. Moreover, the opposing campaign focused on this piece of information, making knowledge of this fact particularly relevant to Proposition 8. Indeed, we expect that voters who knew this fact were likely to have learned it from the broader informational environment created by the campaigns. Thus, voters who followed the campaigns or were familiar with California's laws regarding the initiative process should have learned the correct answer to this question.

We asked two additional factual knowledge questions that we omit from the analysis below. For Proposition 4, we asked “Under Proposition 4, do you know the number of hours before their child receives an abortion that a parent must be notified”? The correct answer is “48 hours”. This measure, however, was never significant in any regression, and we therefore dropped it from the analysis for simplicity. For Proposition 8, we asked voters, “True or False: Proposition 8 would limit marriage, overturning a recent California Supreme Court decision”. The correct answer is “true”. Over 90 per cent of our respondents answered this question correctly. As a result, it did not provide any predictive power in our regression and we dropped it from the analysis.

By asking factual knowledge questions, we can separate voters who have relatively deep knowledge of a ballot measure from voters who only have knowledge of the information shortcut we asked about. Similar to Lupia before us, including these questions allows us to estimate whether factual knowledge or knowledge of a cue had any effect on vote choice. We provide the ballot summary of each proposition in Appendix.

Research design and methods

We use a post-test-only non-equivalent dependent variable design to test our hypothesis. Our design uses responses to the factual knowledge and information shortcut questions to create variables that estimate the effect of information on vote choice. Unlike Lupia (Reference Lupia1994), who assumes that voters perceived the relevant cue-givers to be credible – in particular, he assumes that voters disliked lawyers and insurance companies but trusted a Ralph Nader-led consumer group, and knowledge of these cues would lead to a vote for or against their respective measures – we must define what an informed choice is by adding an expected effect of our information variables (e.g. Rabinowitz and MacDonald Reference Rabinowitz and MacDonald1989). In particular, we anticipate that information – especially elite endorsements – will affect subgroups of voters differently. Indeed, Lupia and McCubbins (Reference Lupia and McCubbins1998) demonstrate that the ability of information to influence decisions is contingent on the perceived credibility of the source.

Our ballot measures deal with abortion (Proposition 4) and same-sex marriage (Proposition 8), two quintessential ideological issues. As such, we predict that voters who are liberal will view the endorsements from Planned Parenthood, the Democratic Party and the Republican Party differently when compared with conservative voters. Put another way, the effect of the endorsements provided by these groups will be conditional on an individual's stated ideology. Therefore, we add a “direction” to our information variables by interacting them with our respondents’ self-reported ideology. Our approach matches the theoretical framework of Lupia and McCubbins (Reference Lupia and McCubbins1998) and the empirical research of Karp (Reference Karp1998) who adds a direction to his cue variable by interacting it with a thermometer score his respondents gave regarding the cue-giver. In essence, we anticipate that information should help liberal voters oppose these two measures and conservative voters support these two measures.

We employ a quasi-experimental test of our hypotheses by matching respondents in the treatment group (knowledge of a cue) to respondents in the control group (no knowledge of a cue) for both ballot measures. We use a simple matching equation where we predict the propensity of receiving the treatment with common demographic variables (age, income, education and gender) and related knowledge variables (factual knowledge of the proposition and general political knowledge). To implement our matching equation, we use the GenMatch package for R (see Diamond and Sekhon Reference Diamond and Sekhon2005) as implemented by the MatchIt package for R (Ho et al. Reference Ho, Imai, King and Stuart2007).

Why use matching in this research? By using matching, we can be confident that our findings are not the result of some covariate imbalance between our groups (those who knew the cue and those who did not). Matching, then, ensures (1) that we have covariate balance between our treatment and control conditions and (2) that the propensity to receive the treatment is about equal between the treatment and control groups. In order to construct a quasi-experimental test of our hypotheses, we must make the two groups comparable. In an experiment, researchers can administer the treatment to one group and establish a control group that does not receive the treatment. In observational studies, such as ours, we cannot and should not assume that the treatment (knowledge of a cue) was randomly assigned among the population. Indeed, voters’ level of interest in elections is not equally distributed among the population. As such, we expect that certain types of individuals in our sample will be predisposed to receive the treatment; that is, their propensity to receive the treatment is higher than other individuals. Genetic matching, then, balances the propensity to receive the treatment across both the treatment (knew a cue) and control (did not know a cue) groups. By balancing the propensity to receive the treatment across treatment and control groups, we can compare the two groups directly, thereby satisfying the requirements of a quasi-experimental test.Footnote ⁴

After matching our respondents, we use the following logit regression equation to model each respondent's vote choice on Propositions 4 and 8:

$$Pr({{y}_{iz}}\, = \,1)\, = \,1/(1\, + \,{{e}^{ - {{n}_{iz}}}} )$$

where

$$no{ {{n}_{iz}}\, = \,{{\beta }_{\rm{0}}}\, + \,{{\beta }_{\rm{1}}}LIBERA{{L}_{iz}}\, + \,{{\beta }_{\rm{2}}}LIBERA{{L}_{iz}}\,\times \,CU{{E}_{iz}}\, \cr & \quad+ \, {{\beta }_{\rm{3}}}LIBERA{{L}_{iz}}\,\times \,FAC{{T}_{iz}}\, + \,{{\beta }_{\rm{4}}}LIBERA{{L}_{iz}}\,\times \,CU{{E}_{iz}}\,\times \,FAC{{T}_{iz}}\, \cr & \quad+ \, {{\beta }_{\rm{5}}}CONSERVATIV{{E}_{iz}}\, + \,{{\beta }_{\rm{6}}}CONSERVATIV{{E}_{iz}}\,\times \,CU{{E}_{iz}}\, \cr & \quad + \, {{\beta }_{\rm{7}}}CONSERVATIV{{E}_{iz}}\,\times \,FAC{{T}_{iz}}\, + \,{{\beta }_{\rm{8}}}CONSERVATIV{{E}_{iz}}\, \cr & \quad\times \, CU{{E}_{iz}}\,\times \,FAC{{T}_{iz}}\, + \,{{\beta }_{\rm{9}}}CU{{E}_{iz}}\, + \,{{\beta }_{_{{{\rm{10}}}} }}FAC{{T}_{iz}}\, \cr & \quad+ \, {{\beta }_{_{{{\rm{11}}}} }}CU{{E}_{iz}}\,\times \,FAC{{T}_{iz}}\, + \,{{\beta }_{_{{{\rm{12}}}} }}{{X}_{iz}} $$

In Equation (1), Pr(y_iz = 1) is respondent i's estimated probability of voting in favour (where “1” indicates a “yes” vote and “0” represents a “no” vote) of Proposition z. The term n_iz defines the model that estimates Pr(y_iz = 1). In the model, LIBERAL_iz is a dichotomous measure of whether respondent i classifies oneself as a liberal (coded as a “1”) or not (coded as a “0”). CUE_iz is a dichotomous measure of respondent i's knowledge of an elite endorsement for Proposition z. For Proposition 4, CUE_iz measures whether a voter knew Planned Parenthood's position, where respondent i receives a “1” if they know the correct answer and a “0” otherwise. For Proposition 8, CUE_iz indicates whether respondent i was able to identify at least one of the political parties’ positions. Respondent i receives a “1” if they knew either that the Democratic Party of California opposed the measure or the Republican Party of California supported the measure; voters who knew both cues also received a “1”. Voters who did not know either endorsement received a “0”. FACT_iz indicates whether respondent i gave a correct answer to the factual knowledge questions we asked for Proposition z.

For Proposition 4, FACT_iz indicates whether respondent i gave a correct answer to Question (4), which asked respondents recall whether Proposition 4 required parental consent (it did not). For Proposition 8, FACT_iz estimates the effect of respondent i's knowledge of the vote percentage required to pass Proposition 8, where a “1” indicates a correct answer and a “0” represents an incorrect answer. CONSERVATIVE_iz measures whether respondent i self-identifies as a conservative (coded as “1”) or not (coded as “0”). The final term, X _iz , is a matrix of covariates that control for Age, Education, Income and general Political Knowledge,Footnote ⁵ factors that have been previously shown to affect vote choice.