Power to the People? Psychological Mechanisms of Disengagement From Direct Democracy

Abstract

The goal of direct democracy is to bring power to change laws to ordinary citizens. However, it may alienate citizens because policy language is often complex, perhaps impacting citizens’ voting likelihood and support for policies. We invoke theory on processing fluency and compensatory control motivations to explain voting likelihood and policy attitude formation. Using experiments and mediational analyses, we tested theorized links between policy language complexity and these outcomes. Findings suggest that policy language complexity motivates compensatory trust in policy institutions but this does not likely explain decreased voting likelihood. We also found that low processing fluency associated with reading a complexly worded policy or a policy presented in a disfluent font led to lower voting likelihood and less positive policy attitudes, consistent with predictions. Thus, the form direct democracy often takes manipulates the amount of support garnered for policies and ironically encourages citizens to outsource legislation to institutional elites.

Keywords

political psychology fluency compensatory control voting attitudes

Democracy is a system of government in which citizens participate equally in the legislative process. Specific criteria for a democratic process have been proposed, including the participation of most adult permanent residents as citizens, the ability for these citizens to gain an “enlightened understanding” about policies, equal and effective opportunities for participation toward the adoption of preferred policies, voting equality in terms of opportunity and weight, and citizens’ control of the political agenda which should always be open to change (Dahl, 1998, p. 37). The goal of representative democracy is to provide citizens power to self-govern indirectly through elected representatives who form policies that citizens want.

Representatives, of course, do not always legislate in accordance with citizens’ preferences. We see this today; a Gallup poll conducted November 7–10, 2013, revealed a record low of 9% of Americans approving “of the way Congress is handling its job.” An alternative form of democracy—direct democracy—is intended to bring power to the people and away from elites. This form of democracy in the United States—citizens voting directly on legislation as opposed to electing legislators—was expected to lead to more informed, engaged, and efficacious citizens (Barnett, 1915; Bryce, 1910; Cree, 1892; Garner, 1907; Haynes, 1907; Key & Crouch, 1939; Munro, 1912; Sullivan, 1892).

In reality, citizens are less likely to participate in direct democracy than they are to vote for representatives (Cronin, 1989; Dubois & Feeney, 1998; Everson, 1981). They also demonstrate more confusion in the context of voting on legislation compared with voting for candidates (Dubois & Feeney, 1998). These consequences are highlighted by the literature on “negative voting” which documents voters’ tendency to vote “no” when they are unsure about a policy (Bowler & Donovan, 1998; Donovan, Bowler, & McCuan, 2001; Gerber, Lupia, McCubins, & Kiewiet, 2001; Jacob, 2001; Magleby, 1984).

Some researchers argue these outcomes stem from the prohibitively confusing language used to describe policies (Dubois & Feeney, 1998; Magleby, 1984). They detail how direct democracy initiatives—statutes, amendments, or ordinances put to a popular vote—are written as complicated proposals riddled with inaccessible language. While some states mail “voter guides” with descriptions of upcoming ballot initiatives, these are sometimes written just as complexly as the policies (LaPalombara, 1950; Magleby, 1984). Indeed, a majority of surveyed voters in four states agreed ballot initiatives are so complicated that one cannot understand what is going on (Cronin, 1989). Further, the most common criticism of direct democracy is that it is confusing (Bowler & Donovan, 1998). Longer ballots are associated with more abstention from voting on legislation (Cronin, 1989; Darcy & Schneider, 1989; Dubin & Kalsow, 1994; Dubois & Feeney, 1998; Magleby, 1984) and constituents vote “no” more often on longer ballot initiatives and those appearing farther down the ballot (Dubin & Kalsow, 1994, 1996). Altogether, the complexity typical of policies’ language may explain high rates of abstention and degree of support for policies.

Notwithstanding these important past findings, psychological processes remain unclear. We enlist literature that speak to potential cognitive and motivational consequences of reading complex policies, namely, we invoke theory on processing fluency (e.g., Reber, Schwarz, & Winkielman, 2004; Reber, Winkielman, & Schwarz, 1998; Schwarz, 2004; Winkielman & Cacioppo, 2001) and compensatory control (Kay, Gaucher, Napier, Callan, & Laurin, 2008). Following, we review mechanisms of the impact of policy language complexity upon voting likelihood and policy attitudes, describe our experimental approach, and discuss the implications of our findings.

Fluency, Compensatory Trust, and Voting Likelihood

People are motivated to perceive sources of control in their lives. When individuals feel a lack of personal control, agents outside the self are more likely to be perceived as in control. For instance, feeling one is capable of understanding political issues and able to exert control over policy allows for the perception of control within one’s life. When one feels a lack of such understanding and control, one is motivated to perceive control from external sources. This is a hypothesis derived from compensatory control theory (e.g., Kay et al., 2008; Kay, Shepherd, Blatz, Chua, & Galinsky, 2010; Shepherd & Kay, 2012). Of particular relevance to the current research is the finding that confusion regarding social issues leads individuals to trust institutions to address social issues and to avoid further information about the issues (Shepherd & Kay, 2012). These findings suggest that when people read about a policy in confusing language, they may resolve a low sense of control by trusting that institutional elites are successfully in control of the policy issue. This may ultimately lead to abstention, functionally outsourcing legislation to institutional elites.

Fluency and Policy Attitude

Psychological theory also provides an account for why policy language complexity may impact policy support. Namely, there is an association between the ease with which information is cognitively processed and attitude formation (e.g., Reber et al., 1998, 2004; Winkielman & Cacioppo, 2001). Greater fluency experienced when mentally processing a given stimulus results in a more positive attitude toward the stimulus (Schwarz, 2004). Thus, to the extent complex language typifies policies, this may lead individuals to support them less. Oppenheimer (2006) found that augmenting the complexity of personal statements for English graduate admissions, translations of Descartes’ writing, and sociology dissertation abstracts resulted in reduced judgments of authors’ intelligence—revealing negative attitudinal consequences of textual complexity. Parallel processes may operate in the context of direct democracy, with opponents of ballot initiatives appearing to capitalize on this bias. This is illustrated by negative campaigns like the following from Salt Lake City: “CONFUSED? Many are. Play it Safe—When in Doubt, VOTE NO!” (Magleby, 1984, p. 142).

However, it is not obvious that this outcome should arise in the context policy language. The policy domain may be one in which complexity indicates a high quality, well-developed policy. While complexity within humanities or philosophical texts may be interpreted as an attempt to obscure a text’s low quality, policy may be a context in which disfluency is interpreted as indicating positively valenced expertise. Thus, testing the association between fluency and policy attitudes sheds light not only on direct democracy, but also potential boundary conditions for disfluency effects.

Current Research

If policies and their descriptions were written more fluently, would this reduce voter confusion, abstention, and negative voting that characterize direct democracy? Our two experiments addressed these questions. Our fluency–compensatory trust–voting hypothesis was that an increase in a policy’s complexity motivates compensatory trust in institutions, in turn lowering voting likelihood. Our fluency–policy attitude hypothesis was that greater policy complexity leads to less positive policy attitudes. Study 1 addresses both of these hypotheses and investigated any mediating roles of processing ease. We presented participants with a potential policy ostensibly under consideration by their state government. The information about the policy was adapted from Shepherd and Kay (2012) and presented to participants in either simple or complex language.

Study 1

Method

Participants

Individuals were recruited in two waves. Wave 1 (N = 83) left our sample small despite considerable effort to recruit participants. A post hoc analysis revealed N = 83 offered .43 in power for a (plausible) effect size of d = .40. Thus, we opted for a second wave, aiming for a final sample of N = 200 (achieving N = 198)¹ for .80 in power for two-tailed tests of significance.

Wave 1 was composed of students and staff at a university in Illinois. The sample was 41% female, 18–55 years old (M = 25.21, SD = 6.99), 53% White (one participant also Latino), 23% Asian (one participant also Latino), 12% Latino-only, 7% Black, 2% Pacific Islander (one participant also Asian-White), and 2% Native American-White (one participant also Latino). Party identification (ranging from 1 = strong Democrat to 7 = strong Republican) revealed Wave 1 leaned Democrat (M = 2.81, SD = 1.37). Educational background (1 = did not complete high school, 1%; 2 = diploma, 7%; 3 = special training/partial college, 43%; 4 = bachelor’s degree [BA/BS], 13%; 5 = partial graduate school, 7%; 6 = master’s, 20%; and 7 = doctorate, 9%) revealed that on average, participants (M = 4.11, SD = 1.57) and their parents (M = 4.43, SD = 1.73) had completed college. Wave 2 was composed of students (14%) and Amazon Mechanical Turk (MTurk) workers residing in one of the 36 U.S. states. MTurk is a crowd-sourcing Internet marketplace used by some to collect data. The sample was 40% female, 18–72 years old (M = 32.65, SD = 12.62), 70% White (one participant also Latino), 13% Asian (one participant also White), 9% Black (one participant also White and Latino), 4% Latino-only, 3% Pacific Islander, and had one Native American-White participant. Participants leaned slightly Democrat (M = 3.38, SD = 1.61). Thirty-two percent of participants had a BA/BS, 25% were seeking a BA/BS, 26% did not have a BA/BS, 3% were graduate students, and 14% had graduate degrees. Forty-nine percent of participants’ parents did not have a BA/BS, 23% had one parent with a BA/BS, and 28% had two parents with a BA/BS.

Procedure

For Wave 1, individuals were approached on campus and offered a snack for completing a paper questionnaire. Participants were randomly assigned to read about a potential state initiative for adopting energy technologies described in either complex or simple language. The descriptions, adapted from Shepherd and Kay (2012), told participants they would “read about energy technologies the Illinois state government may utilize if there is enough support from Illinois residents leading up to and following an upcoming election,” and researchers want “to learn more about what individuals living in Illinois think of these energy technologies and the state government’s potential use of them.” Next, participants read: “We are going to ask you your thoughts about a proposed ballot initiative designed to help Illinois deal with today’s energy issues. The ballot initiative is about the adoption of certain means of creating energy for Illinois.” Participants subsequently read about three types of energy. For example, an excerpt from the complex description explained that cellulosic biofuel is “produced from lignocellulose, a structural material composed mainly of cellulose, hemicellulose, and lignin. The cellulolysis process consists of hydrolysis on pretreated lignocellulosic materials. After the hydrolysis process, C5 cellulose material can be converted by exposing the C5 cellulose to microbial cultures that secrete anticellulose enzymes, which degrade the C5 cellulose cell walls of the organic material.” In the simple description, it was explained as “a fuel made from almost any organic input material, including feedstock, waste, and plant material. This material is simply fed into a large tank, where it is exposed to micro-organisms/bacteria. These micro-organisms work to breakdown and convert the organic material into simple sugars and then into liquid ethanol, which can be used as a fuel. This process is comparable to the fermentation process that is used to create alcohol for beverages.”

Wave 2 participants completed the questionnaire via Qualtrics online software. University participants received course credit; MTurk participants received payment. The procedure was similar to Wave 1. However, participants indicated their state of residence and were ostensibly forwarded to questions for people in their state. References to their state were generic. As before, participants were randomly assigned to read about the policy in either complex or simple language. For both waves, participants completed all measures’ items on 7-point scales after the manipulation.

Measures

Processing ease was measured with “I can easily understand how these methods of supplying energy work” (1 = strongly disagree to 7 = strongly agree; Shepherd & Kay, 2012; M = 3.69, SD = 1.90). Compensatory trust was measured following this information: “In your state, there are a number of agencies that deal with the energy plan, including The Energy Board, Natural Resources Consortium, and the Department of Energy. These groups are made up of various scientists, politicians, policymakers, etc. The next two questions below pertain to these groups.” The items were: “To what extent do you trust these groups to manage these sources of energy properly?” and “To what extent do you trust these groups to deal with any issues that are associated with these sources of energy?” (1 = not at all to 7 = entirely; Shepherd & Kay, 2012; α = .94, M = 4.12, SD = 1.31). Voting likelihood was measured with “Would you vote on the issue, or would you abstain from voting, meaning that would not vote on the issue at all?” (1 = definitely would not vote to 7 = definitely would vote; M = 5.18, SD = 1.63). Policy attitude was measured with three items: “Do you like or dislike these energy technologies?” “Do you like or dislike the idea of a policy of using these energy technologies?” (1 = strongly dislike to 7 = strongly like), and “Let’s say you were going to vote. Would you then vote in favor of using these technologies, or vote against them?” (1 = definitely would vote against their use to 7 = definitely would vote for their use; α = .88, M = 4.92, SD = 1.09).

Results and Discussion

We analyzed the dependent variables (DVs)—processing ease, compensatory trust (reverse coding ²), voting likelihood, and policy attitude—as a within-participants effect and complexity manipulation and wave as between-participants effects, resulting in a 4 (DV) × 2 (complex/simple) × 2 (wave) mixed model using R multilevel modeling packages (Bates & Maechler, 2009; Tremblay & Ransijn, 2013). We found inequality of means between DVs, F(3, 588) = 58.01, p < .001, but no effect of sample wave, F(1, 588) = .08, p = .78.³ Complexity negatively predicted the DVs, d = −.85, F(1, 588) = 35.93, p < .001, consistent with hypotheses.

However, the effect of the manipulation was qualified by an interaction with DV factor, F(3, 581) = 9.03, p < .001. Analyzing the DVs separately, we found effect sizes varied. Participants in the complex condition reported less processing ease (M = 2.90, SD = 1.86, standard error [SE] = .19) than those in the simple condition (M = 4.46, SD = 1.61, SE = .16), d = −.90, t(196) = −6.32, p < .001, greater compensatory trust (original coding) in the agencies (M = 4.32, SD = 1.32, SE = .13) than those in the simple condition (M = 3.93, SD = 1.28, SE = .13), d = .30, t(196) = 2.09, p ≤ .04, lower voting likelihood (M = 4.83, SD = 1.81, SE = .18) than those in the simple condition (M = 5.52, SD = 1.34, SE = .13), d = −.43, t(196) = −3.06, p = .002 and less positive policy attitudes (M = 4.77, SD = .99, SE = .10) than those in the simple condition (M = 5.08, SD = 1.17, SE = .12), d = −.29, t(196) = −2.03, p = .04. We found an interaction between DV factor and wave, F(3, 581) = 7.43, p < .001. Thus, waves differed in means across some DVs. Complexity interacted with neither wave, F(1, 581) = .1.75, p = .19, nor wave × DV factor, F(3, 578) = .30, p = .82.

Fluency, Compensatory Trust, and Voting Likelihood

Processing ease and trust in the agencies were not associated (B = −.04, SE = .05), t(196) = −.81, p = .42. Thus, trust was not directly explained by low processing ease, suggesting that compensatory control processes may be separable from disfluency effects. Trust was also not a mediator of the effect of complexity on voting likelihood because it did not predict likelihood (B = −.10, SE = .09), t(196) = −1.11, p ≤ .27. Trusting the agencies may have palliated aversion associated with the complex condition, but any such process does not explain voting likelihood.

Analyses revealed processing ease was a good candidate as a mediator of the impact of complexity upon voting likelihood because it predicted likelihood (B = .29, SE = .06), t(196) = 5.00, p < .001 (Baron & Kenny, 1986). When voting likelihood was regressed on both processing ease and condition (simple = 0, complex = 1), the effect of complexity was reduced to nonsignificance (see Figure 1). We investigated this mediational pathway with bootstrapping (Imai, Keele, Tingley, & Yamamoto, 2010) using 10,000 simulations. The mediation effect was B = −.40 with a 95% quasi-Bayesian confidence interval (CI) of [−.65, .19], corroborating that processing ease mediates the impact of policy complexity on voting likelihood. Thus, the fluency–compensatory trust–voting hypothesis that an increase in a policy’s complexity would motivate a compensatory increase in institutional trust and in turn lower voting likelihood was partially borne out. Policy complexity appears to motivate trust in policy institutions, and the low processing ease of a complexly worded policy attenuates voting likelihood, but these outcomes appear separable.

Figure 1.
Low processing ease mediates the negative impact of complexity on voting likelihood. Unstandardized regression coefficients are reported. p < .01; *p < .001.

Fluency and Policy Attitude

Analyses revealed processing ease was a good candidate as a mediator of the impact of complexity upon policy attitudes because it predicted attitudes (B = .23, SE = .04), t(196) = 6.25, p < .001. When attitude was regressed on both processing ease and condition, the effect of complexity was reduced to nonsignificance (see Figure 2). Bootstrapping revealed the mediation effect of fluency was B = −.38, CI = [−.56, −.22], corroborating that reading about the policy in complex language lead to less processing ease and in turn less positive policy attitudes. Altogether, these results imply that when complex language typifies policies brought to a popular vote, citizens’ attitudes toward them are less positive than they would be otherwise. The negative impact of textual complexity on valenced judgments of scholarly texts (Oppenheimer, 2006) appears to generalize to policy language, revealing another context in which disfluency attenuates positive evaluations. If policy complexity interferes with policy support, some citizens may fail to vote in accordance with their latent preferences—if they vote at all.

Figure 2.
Low processing ease mediates the negative impact of complexity on attitudes. Unstandardized regression coefficients are reported. ***p < .001.

While our results corroborated the fluency–attitude hypothesis and supported a simplified fluency–voting hypothesis, our manipulation of complexity may not have operationalized disfluency in the most direct way. Thus, we conducted a second study. Participants were presented the policy either in complex language, simple language, or simple language in a disfluent font. We predicted that participants exposed to the simple language manipulation would report greater processing ease and this would predict both greater voting likelihood and more positive policy attitudes.

Study 2

Method

Participants

Given Study 1, we predicted directional effects in which the simple condition would generate greater processing ease and thus greater voting likelihood and more positive policy attitudes. Thus, for all relevant contrasts and correlations, we used directional/one-tailed tests of significance (the exception being omnibus F tests). Desiring .80 in power and selecting d = −.27 (just under the smallest effect size observed in Study 1), we aimed for N = 513.

The sample (N = 520)¹ was composed of MTurk workers residing in one of the 44 U.S. states. The sample was 57% female, 18–76 years old (M = 36.53, SD = 13.97), 78% White (14 also Latino), 10% Black (7 also reporting additional ethnicities), 8% Asian (6 participants also White), 2% Latino-only, and had 9 Native Americans (6 also White and/or Latino), 4 Pacific Islanders (3 reporting additional ethnicities), and 2 Middle Easterners (1 also Native American). The sample leaned slightly Democrat (M = 3.38, SD = 1.61). Forty three percent of participants did not have a BA/BS, 11% were seeking a BA/BS, 31% had a BA/BS, 3% were graduate students, and 12% had graduate degrees. Fifty-two percent of participants’ parents did not have a BA/BS, 25% had one parent with at least a BA/BS, and 23% had two parents with at least a BA/BS.

Procedure and Measures

MTurk participants completed the questionnaire via Qualtrics and received payment. Participants were randomly assigned to read about the policy in complex language, simple language, or simple language in a disfluent font. This latter condition was operationalized by making the font smaller, italicized, and grey (see Gervais & Norenzayan, 2012). Following the manipulation, participants completed the measures of processing ease (M = 4.10, SD = 1.80), voting likelihood (M = 5.46, SD = 1.60), and policy attitude (α = .87, M = 4.89, SD = 1.21).

Results and Discussion

We analyzed the DVs as within-participants and condition as between-participants, resulting in a 3 (DV) × 3 (disfluent font/complex/simple) mixed model. We found inequality of means across DVs, F(2, 1035) = 149.39, p < .001. More importantly, the manipulation impacted the DVs, F(2, 1035) = 41.42, p < .001. Consistent with hypotheses, processing ease and voting likelihood were greater and policy attitudes more positive among individuals in the simple condition compared with those in the complex condition, d = −.94, t(344) = −8.84, p < .001, and disfluent font condition, d = −.25, t(346) = −2.35, p ≤ .01.

However, the effect of condition was qualified by an interaction with DV factor, F(4, 1031) = 14.24, p < .001. We analyzed the DVs separately, finding differences in effect size. The manipulation impacted processing ease, F(2, 517) = 56.32, p < .001. Those in the simple condition report greater processing fluency (M = 4.86, SD = 1.46, SE = .11) than those reading the complex text (M = 3.06, SD = 1.74, SE = .13), d = −1.12, t(344) = −10.44, p < .001, and those reading the disfluent font (M = 4.37, SD = 1.68, SE = .13), d = −.31, t(346) = −2.86, p = .002. Thus, the ease of understanding the technologies in simple language was indeed undermined by a disfluent font. The manipulation also impacted voting likelihood, F(2, 517) = 9.63, p < .001. Those in the simple condition reported greater voting likelihood (M = 5.79, SD = 1.53, SE = .12) than those reading the complex text (M = 5.06, SD = 1.66, SE = .13), d = −.46, t(344) = −4.28, p < .001, and those reading the disfluent font (M = 5.51, SD = 1.52, SE = .12), d = −.18, t(346) = −1.72, p = .04.

Finally, the manipulation impacted policy attitude, F(2, 517) = 9.98, p < .001. Those in the simple condition reported more positive policy attitudes (M = 5.08, SD = 1.26, SE = .10) than those in the complex condition (M = 4.56, SD = 1.10, SE = .08), d = −.44, t(344) = −4.07, p < .001. However, the difference between the simple condition and disfluent font condition (M = 5.03, SD = 1.21, SE = .09), though in the predicted direction, did not achieve significance, d = −.04, t(346) = −.36, p < .36. Nevertheless, this result suggests that if processing ease significantly mediates any impact of the disfluent font on policy attitudes via indirect-only mediation, it is consistent with our fluency–policy attitude hypothesis (Zhao, Lynch, & Chen, 2010).

Fluency and voting likelihood

Analyses revealed processing fluency was a good candidate as a mediator of the impact of the manipulations on voting likelihood because it predicted voting likelihood (B = .26, SE = .04), t(518) = 6.86, p < .001. When regressing voting likelihood on condition (dummy coding the complex language and disfluent font conditions; Hayes & Preacher, 2014) and on processing ease, the effect of the complex language condition was reduced in magnitude and the effect of the disfluent font condition was reduced to nonsignificance (see Figure 3). Bootstrapping revealed the mediation effect of processing ease for the complex condition was B = −.41, CI = [−.58, −.25], corroborating partial mediation.⁴ The mediation effect for the disfluent font condition was also significant, B = −.11, CI = [−.21, −.03]. Thus, the complex language and disfluent font manipulations negatively impacted the ease with which participants understood the energy policy, in turn attenuating voting likelihood.

Figure 3.
Low processing ease mediates the negative impact of font disfluency and partially mediates the negative impact of complexity on voting likelihood. Unstandardized regression coefficients are reported. p < .05; p < .01; p < .001.

Fluency and policy attitude

Analyses revealed that processing ease was a good candidate as a mediator of the impact of the manipulations on policy attitude because it predicted attitude (B = .29, SE = .03), t(518) = 10.74, p < .001. When regressing policy attitude on condition as well as processing ease, the effect of the complex condition was reduced to nonsignificance and the effect of the disfluent font condition remained nonsignificant (see Figure 4). Bootstrapping estimated the mediation effect of processing ease as B = −.51, CI = [−.66, −.38] for the complex condition and B = −.14, CI = [−.24, −.04] for the disfluent font condition. Thus, the disfluent font and complex manipulations decreased processing ease, in turn leading to less positive policy attitudes. While the total effect of the disfluent font condition upon policy attitude was nonsignificant, recently researchers have argued against a need for an “effect to be mediated.” In doing so, Zhao, Lynch, and Chen (2010) have argued that both the pattern of mediation for the complex manipulation and that of indirect-only mediation for the disfluent font manipulation are consistent with our hypothesis.

Figure 4.
Low processing ease mediates the negative impact of complexity and mediates (indirect-only) the negative impact of font disfluency on attitudes. Unstandardized regression coefficients are reported. p < .01; p < .001.

These results suggest once again that the policy domain is not one in which textual complexity—as well as disfluency operationalized with a disfluent font—is interpreted as evidence of a higher quality policy. The decrease in processing ease due to both disfluent experimental conditions mediated negative impacts on policy attitudes, contributing to the processing fluency literature the finding that the policy domain appears not to be an exception with regard to negative associations between disfluency and liking (Schwarz, 2004).

General Discussion

Generally, we found support for our hypotheses. Participants reading a complexly worded policy did appear motivated to trust policy agencies more than those encountering the policy in simple language, consistent with compensatory control theory (Shepherd & Kay, 2012). However, such trust did not appear responsible for lowered voting likelihood and was not itself directly explained by lowered processing ease—somewhat contrary to predictions. What our results did corroborate were disfluency–voting and disfluency–policy attitude hypotheses. Specifically, our results suggest that complexly worded policies may encourage abstention and negatively impact policy support via low processing ease. Given that complex language often characterizes policies brought to a popular vote, some citizens may not vote in accordance with their latent preferences should they vote at all. These findings provide greater understanding of mechanisms behind consequences of statutes, amendments, and ordinances written in technical language. These findings also suggest individuals like a policy less when its text is disfluent. The typically observed negative impact of disfluency on attitudes and judgments (Oppenheimer, 2006; Schwarz, 2004) appears to apply in the policy domain.

These results reveal an impediment to the political engagement envisioned by direct democracy advocates. However, simply reducing the complexity of policies’ language could lead fewer citizens to abstain from voting and reduce unrepresentative vote choices. Conversely, these results imply that using complex language in policy descriptions may keep policies from passing, as complexly worded policies appear to encourage abstention as well as negative voting.

Limitations and Future Directions

Our experimental designs complement past research by offering greater internal validity. However, caution regarding the interpretation of mediational analyses is necessary because associations between DVs remain correlational. Our results corroborate* our causal framework but they should not be interpreted as confirmatory given the limitations of mediational analyses (Fiedler, Schott, & Meiser, 2011). Nevertheless, taken together, the results of prior research and our experiments converge to suggest the complexity of policies’ language is problematic for the accomplishment of the goals of direct democracy. Field experiments with greater ecological validity represent a promising next step in this research program. Scholars’ future efforts might involve collaborating with political mobilization groups to create materials paralleling policy information while manipulating textual fluency and observing voters’ preferences and actions.

These findings also raise questions about the reach of the compensatory trust motivated by encounters with complex policy language. For instance, perhaps encountering incomprehensible policies motivates preferences for specific traits during candidate elections. Inaccessible policy language might lead individuals to prefer candidates with a more autocratic leadership style because individuals may question the viability and efficacy of participatory democracy. When ballots include both policies and candidates, the presence of the policies may impact who is elected. Policy complexity may even impact citizens’ preferences for appointed versus elected officials, as electing officials ideally involves understanding and competence among voters. Candidates evaluated immediately following encountering a complex policy may also be evaluated less positively—showing once again a possible means through which policy complexity may impact attitudes.

Conclusion

The results of the current research suggest that direct democracy as it often appears seems to diminish the engagement it was created to foster. Born out of a distrust of institutional elites, the design of direct democracy may ironically make people trust them more. Designed to invigorate Americans’ participatory spirit, direct democracy may instead encourage abstention and disconnect policy attitudes from latent policy preferences. However, these outcomes may have less to do with direct democracy itself than with a very specific and modifiable feature of its execution. Given the improved understanding of the mechanisms behind such political disengagement, a path toward bringing more citizens into the political process is visible.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: The University of Chicago Psychology Graduate Student Organization supported data collection costs associated with this research. Additionally, the first author was supported in part during the writing of this manuscript by an Interdisciplinary Postdoctoral Fellowship funded by NSF (SES1228559).

Notes

References

Barnett

(1915). The operation of the initiative and referendum in Oregon. New York, NY: Macmillan.

Baron

R. M.

Kenny

D. A.

(1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182.

Bates

Maechler

(2009). Package ‘lme4’. Retrieved from http://lme4.r-forge.r-project.org

Bowler

Donovan

(1998). Demanding choices: Opinion, voting, and direct democracy. Columbus: Ohio State.

Bryce

L. J.

(1910). The American commonwealth. 3rd ed. New York, NY: MacMillan.

Cree

(1892). Direct legislation by the people. Chicago, IL: McClurg.

Cronin

T. E.

(1989). Direct democracy: The politics of initiative, referendum, and recall. New York, NY: Harvard.

Dahl

R. A.

(1998). On democracy. New Haven, CT: Yale.

Darcy

Schneider

(1989). Confusing ballots, roll-off, and the black vote. Western Political Quarterly, 42, 347–364.

10.

Donovan

Bowler

McCuan

D. S.

(2001). Political consultants and the initiative industrial complex. In Sabato

L. J.

Ernst

H. R.

Larson

B. A.

(Eds.), Dangerous democracy? The battle over ballot initiative in America (pp. 101–134). New York, NY: Rowman and Littlefield.

11.

Dubin

J. A.

Kalsow

G. A.

(1994). Participation in direct legislation: Evidence from the voting booth. Mimeo. Pasadena: California Institute of Technology.

12.

Dubin

J. A.

Kalsow

G. A.

(1996). Comparing absentee and precinct voters: A view over time. Political Behavior, 18, 369–392.

13.

Dubois

P. L.

Feeney

(1998). Lawmaking by initiative: Issues, options, and comparisons. New York, NY: Agathon.

14.

Everson

(1981). The effects of initiatives on voter turnout: A comparative state analysis. Western Political Quarterly, 34, 415–425.

15.

Fiedler

Schott

Meiser

(2011). What mediation analysis can (not) do. Journal of Experimental Social Psychology, 47, 1231–1236.

16.

Garner

J. W.

(1907). Primary vs. representative government. Proceedings of the American Political Science Association, 4, 164–174.

17.

Gerber

E. R.

Lupia

McCubins

M. D.

Kiewiet

D. R.

(2001). Stealing the initiative: How state government responds to direct democracy. Upper Saddle River, NJ: Prentice Hall.

18.

Gervais

W. M.

Norenzayan

(2012). Analytic thinking promotes religious disbelief. Science, 336, 493–496.

19.

Hayes

A. F.

Preacher

K. J.

(2014). Statistical mediation analysis with a multicategorical independent variable. British Journal of Mathematical and Statistical Psychology, 67, 451–470.

20.

Haynes

(1907). The education of voters. Political Science Quarterly, 22, 484–497.

21.

Imai

Keele

Tingley

Yamamoto

(2010). Causal mediation analysis using R. In Vinod

H. D.

(Ed.), Advances in social science research using R (pp. 129–154). New York, NY: Springer.

22.

Jacob

(2001). The initiative process: Where people count. In Sabato

L. J.

Ernst

H. R.

Larson

B. A.

(Eds.), Dangerous democracy? The battle over ballot initiative in America (pp. 94–99). New York, NY: Rowman and Littlefield.

23.

Kay

A. C.

Gaucher

Napier

J. L.

Callan

M. J.

Laurin

(2008). God and the government: Testing a compensatory control mechanism for the support of external systems. Journal of Personality and Social Psychology, 95, 18–35.

24.

Kay

A. C.

Shepherd

Blatz

C. W.

Chua

S. N.

Galinsky

A. D.

(2010). For God (or) country: The hydraulic relation between government instability and belief in religious sources of control. Journal of Personality and Social Psychology, 5, 725–739.

25.

Key

V. O.

Crouch

(1939). The initiative and referendum in California. Berkeley: University of California.

26.

LaPalombara

J. G.

(1950). The initiative and referendum in Oregon: 1938-1948. Corvallis: Oregon State College Press.

27.

Magleby

D. B.

(1984). Direct legislation: Voting on ballot propositions and the United States. Baltimore, MD: Johns Hopkins.

28.

Munro

(Ed.). (1912). The initiative, referendum and recall. New York, NY: D. Appleton.

29.

Oppenheimer

D. M.

(2006). Consequences of erudite vernacular utilized irrespective of necessity: Problems with using long words needlessly. Applied Cognitive Psychology, 20, 139–156.

30.

Reber

Schwarz

Winkielman

(2004). Processing fluency and aesthetic pleasure: Is beauty in the perceiver’s processing experience? Personality and Social Psychology Review, 8, 364–382.

31.

Reber

Winkielman

Schwarz

(1998). Effects of perceptual fluency on affective judgments. Psychological Science, 9, 45–48.

32.

Schwarz

(2004). Metacognitive experiences in consumer judgment and decision making. Journal of Consumer Psychology, 14, 332–348.

33.

Shepherd

Kay

A. C.

(2012). On the perpetuation of ignorance: System dependence, system justification, and the motivated avoidance of sociopolitical information. Journal of Personality and Social Psychology, 102, 264–280.

34.

Sullivan

(1892). Direct legislation by the citizenship through the initiative and referendum. New York, NY: True Nationalist.

35.

Tremblay

Ransijn

(2013). LMERConvenienceFunctions: A suite of functions to back-fit fixed effects and forward-fit random effects, as well as other miscellaneous functions. R package version 2.1. Retrieved from http://CRAN.R-project.org/package=LMERConvenienceFunctions

36.

Winkielman

Cacioppo

J. T.

(2001). Mind at ease puts a smile on the face: Psychophysiological evidence that processing facilitation elicits positive affect. Journal of Personality and Social Psychology, 81, 989–1000.

37.

Zhao

Lynch

J. G.

Jr. Chen

(2010). Reconsidering Baron and Kenny: Myths and truths about mediation analysis. Journal of Consumer Research, 37, 197–206.