![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
Abstract
Under what conditions can political elites influence elections to favor their preferred policy outcomes by strategically crafting the language printed on the ballot? Drawing on psychological and political theories of voter cognition, we design a survey experiment to assess the degree to which ballot text can influence voter behavior in direct democracy elections and identify factors that may moderate such effects. We show that the language used to describe a ballot measure does indeed have the potential to affect election outcomes, including measures dealing with contentious social issues affecting individual rights. We also find, however, that exposing individuals to basic campaign information—in our case, endorsements from prominent interest groups—greatly attenuates the framing effects of ballot text. Our results suggest that the extent to which ballot text matters depends on the vibrancy of the campaign environment and other information available to voters.
Notes
1. See http://ballotpedia.org/wiki/index.php/List_of_ballot_measure_lawsuits_in_2012#tab=By_topic (accessed October 26, 2012).
2. It is worth noting that every state except Delaware requires the state legislature to receive voter approval via a statewide referendum to pass a constitutional amendment. In these scenarios, the legislature is usually responsible for creating the ballot title and summary.
3. In some states, the political actors responsible for writing a title and summary may delegate this responsibility to civil servants. The ultimate power of assigning a title and summary, however, remains with the constitutionally assigned party. While political actors may not use their influence to craft the description of every ballot measure to favor their preferred outcome, we expect that they will intervene when they care about the policy and when the stakes are particularly high.
4. Three of the remaining states (Colorado, Michigan, and South Dakota) use a commission and only Arizona splits the responsibilities between the proponents and elected officials.
5. Our conception of framing is based on the expectancy model of individual attitudes: , where vi represents a respondent’s evaluation of the issue on attribute i, and wi is the weight that the respondent puts on that attribute. The expressed opinion is thus a weighed sum of the attributes. Framing may influence opinions by affecting both the set of attributes (vi) that voters consider in making their decisions, and how much weight they place on already accessible attributes (wi). See Chong and Druckman (Citation2007), and Nelson, Oxley, and Clawson (Citation1997).
6. Opponents of the measure, including the American Civil Liberties Union and the secretary of state, sued over the ballot wording but did not succeed in their efforts to change the language.
7. In a series of experiments, Lupia and McCubbins (Citation1998) demonstrated that knowledge of endorsement enables individuals to make reasoned choices at least as often as individuals who possess expert-level knowledge.
8. TESS surveys are fielded using Knowledge Networks’ (now GfK) online panel. The data were collected from June 23, 2010, through July 7, 2010. Knowledge Networks asked 9,213 respondents to complete the survey and 6,101 complied, yielding a completion rate of 66.2%. It is important to note that Knowledge Networks relies on a probability-based online panel. While there exists a significant debate about the representativeness of online panels—and nonprobability samples in particular (see, e.g., AAPOR, Citation2010; Ansolabehere & Schaffner, Citation2014; Chang & Krosnick, Citation2009; Malhotra & Krosnick, Citation2007; Yeager et al., 2010)—one potentially relevant question is whether online respondents are more interested in politics than respondents that complete surveys using other models. As Chang and Krosnick (Citation2009, pp. 660–661) find, Knowledge Network’s online panel is statistically indistinguishable from random digit dialing samples across most politically relevant covariates, except political knowledge (Knowledge Network’s respondents are more knowledgeable).
9. The law, Proposition 22, was itself passed by voters via initiative in 2000. Unlike Proposition 8, however, Proposition 22 was a statutory rather than a constitutional initiative.
10. The subjects’ assignment into the cue and no-cue groups did not change during the course of the experiment. If subjects were assigned into the cue group, they saw the endorsements for both ballot measures. As a check on Knowledge Network’s random assignment mechanism, we ran a multinomial logit where the dependent variable was the group assignment and the independent variables were important demographic variables (party identification, ideology, age, education, income, and state of residence). The multinomial logit results (available from the authors) show that Knowledge Networks achieved successful random assignment, with balance on all covariates.
11. The randomization was independent for each measure, so a subject’s assignment to see one description for the first ballot measure did not determine which version of the other measure she would see as the survey progressed.
12. While voters reading different ballot summaries of the same measure may, erroneously, conclude that these represent two different policies, this is precisely the type of deception we wish to measure and the type of confusion strategic elites may try to sow.
13. The growing use of absentee voting makes this type of deliberation much more feasible. We know of no empirical research showing that voters who fill out their ballots in the comfort and privacy of their own home invest substantially more time into the task or read the full text of proposed measures at greater rates than those who vote in person, however. A study that investigates such a research question would be a valuable extension of the existing literature.
14. In a second smaller survey experiment carried out in 2012, we presented a national sample of respondents with identical questions but also provided a link allowing them to read the full text of the actual measure before their consideration rather than only supplying the short title and summary that appears on the ballot. Although we placed this link immediately below the ballot summary and we highlighted the link in a separate font to make it clearly visible and distinguishable, only 6% of subjects chose to click on the link to see the full text, confirming the observation made by Matsusaka (Citation2005) and Magleby (Citation1984) that voters rarely go beyond the ballot text to read the full legal language of the measures they consider.
15. The two versions of the measure also differed in their projected fiscal impact. While the original version estimated no fiscal impact on state or local governments, the 2008 version predicted costs for local governments from the elimination of the tourism business from out-of-state same-sex couples who were coming to California to marry. To make the measures as comparable as possible, both of our versions stated there would be no fiscal impact from the amendment.
16. The Washington measure asked, “Shall public funding of abortions be prohibited except to prevent the death of the pregnant woman or her unborn child?” Since our measure sought to overturn such a ban, we tweaked the Washington language slightly in our application.
17. More than two-thirds of states currently limit public funding to cases where the woman’s life is in danger or in the case of rape and incest. The rest generally provide public funding only for medically necessary abortions. The status quo thus ensures that our measure is one that might reasonably appear on the ballot for the vast majority of survey respondents.
18. Note that the inclusion of a simple interaction term between the frame and cue variables in a regression simultaneously tests two hypotheses: that the presence of cues affects the magnitude of the framing effects and that the effect of the cues varies across different wording used to describe of the same measure. We have little theoretical basis for the latter, so we instead use difference-in-proportions tests to examine more precisely our second hypothesis.
19. As with all TESS studies, inclusion of additional survey elements requires a substantial reduction in subjects. As such, we chose to focus on garnering a larger sample to test our two hypotheses.
20. About one of out every seven respondents of our sample was classified as sophisticates under this definition.
21. Wood and Oliver (Citation2012) focus in particular on the interaction between education and self-reported ideology.
22. It is also important to note that the electorate in our sample differed significantly from the electorate that voted on Proposition 8 and Amendment 7. While our sample of respondents includes voters from across the country, only voters in California and Colorado actually voted on those two measures, respectively.
23. This logic is confirmed empirically in de Figueiredo, Ji, and Kousser (Citation2011).
24. Similarly, Iyengar, Lowenstein, and Masket (Citation2001), who examine slate mailers, have found strong campaign effects for just a single piece of information.
Additional information
Notes on contributors
Craig M. Burnett
Craig M. Burnett is an Assistant Professor in the Department of Public and International Affairs, University of North Carolina Wilmington.
Vladimir Kogan
Vladimir Kogan is an Assistant Professor in the Department of Political Science, Ohio State University.