Research note: Explicit voter fraud conspiracy cues increase belief among co-partisans but have broader spillover effects on confidence in elections

---

Implications

Research on conspiracy beliefs has documented a number of worldviews and other psychological predispositions with which these beliefs are associated (e.g., Miller et al., 2016; Raab et al., 2013; Uscinski et al., 2016), as well as situational factors that may trigger them, such as crises and the uncertainty they bring, negative economic conditions, or electoral losses (Douglas et al., 2019). Some work has further documented correlations between social media exposure and these beliefs (Enders et al., 2021). Still, the actual content and style of conspiracy cues are likely important in the spread of conspiracy beliefs (Lyons et al., 2019; Nyhan et al., 2016; Uscinski et al., 2016), and at the same time, elites (e.g., elected officials) have become increasingly bold in both implicitly and explicitly endorsing conspiracy theories (Berlinksi et al., 2020; Enders & Smallpage, 2018; McCright & Dunlap, 2017), which may degrade trust in democratic institutions (Berlinski et al., 2020; Clayton et al., 2021; Einstein & Glick, 2015). In this context, the potential differential effect of conspiracy cues remains an important open question.

In this study, we specifically look at the effect of conspiracy cues regarding voter fraud. Perceptions of voter fraud have been studied extensively. Several studies have shown a “winner effect,” in which voting for the winning candidate is associated with greater trust in elections, while those on the losing side are more likely to doubt the legitimacy of the process (e.g., Sinclair et al., 2018). Group-centric perceptions of electoral integrity are another consistent finding (Appleby & Federico, 2017; Edelson et al., 2017; Sances & Stewart, 2015; Wilson & Brewer, 2013). For example, anti-immigration attitudes (Udani & Kimball, 2018) and Republican identification (Bowler & Donovan, 2016) have been linked to perceptions of voter fraud and other beliefs about electoral integrity in the United States. Finally, recent work has further shown the negative consequences of elite rhetoric on the issue. Exposure to actual (though unsubstantiated) claims of fraud made by prominent elected Republicans and conservative media personalities (e.g., Donald Trump, Rick Scott, Marco Rubio, Lindsey Graham, etc.) in 2018 and 2020 appear to have had deleterious effects among co-partisans (Berlinski et al. 2020, Clayton et al., 2021). Still, we know less about the effects of similar claims that articulate or infer an actual conspiracy (with reference to the actors, motives, and methods that one entails). Further, these studies employ real-world messages from Republicans as stimuli. This improves external validity but has limited our ability to make inferences about how members of both parties respond to in- and out-party fraud claims. This is especially important if such cues operate in a similar fashion to more mundane cues from elites.

Top-down models of attitude formation are fundamental to political behavior and public opinion research (Campbell et al., 1960; Zaller, 1992). Campbell et al. (1960), for instance, argue that the political party is “an opinion-forming agency of great importance,” and a party is “a supplier of cues by which the individual may evaluate the elements of politics” (p. 128). In this sense, elite partisan cues serve as an information shortcut (Popkin, 1991), particularly for issues about which an individual may have limited knowledge or familiarity (Li & Wagner, 2021). Thus, partisan cues can induce processing in which individuals evaluate information based on its concordance with their political identity (Chaiken, 1980; Kunda, 1990). Much of this work has traditionally focused on policy preferences (Goren, 2005; Lenz, 2012). As others have indicated though, these cuing effects likely carry over beyond policy attitudes to the reception of misinformation (Flynn et al., 2017; Veenstra et al., 2014) and democratic norm violations especially (Clayton et al., 2021). Some have posited that such norm violations may be treated as unacceptable and rejected or even punished (Carey et al., 2020). Other research has suggested that personal views do serve to constrain the influence of elites espousing unpopular opinions to some degree, as well (Mummolo et al., 2021; Peterson, 2018). In sum, partisan conspiracy cues may operate much like more mundane elite cues, but it is possible that explicitly distasteful messages may be rejected even by co-partisans (see Valentino et al., 2018).

Finally, it is important to consider that a great deal of conspiracy-centric communication is vague, subtle, or perhaps even unintentional, rather than fully articulated (Lyons et al., 2019; McCright & Dunlap, 2017; Starbird, 2016; Starbird, 2017), and the reception of such cues has not been fully explored in the elite cue model discussed above. Presenting suspicious coincidences may spread conspiracy beliefs in the place of a robust theory detailing the responsible parties, their incentives, and their methods (Lyons et al. 2019; Prooijen et al., 2017; Raab et al., 2013; Rich & Zaragoza 2016). Notably, too, the rhetorical strategy of “just asking questions” has been used by commentators to allow for conspiracy ideation while maintaining plausible deniability (Byford, 2014; Novak, 2017; Seay, 2017). We refer to this more subtle version as an implicit cue, while a fully articulated conspiracy theory is referred to as an explicit cue. One recent study compared the effect of explicit and implicit conspiracy cues on the adoption of novel public health conspiracy beliefs (Lyons et al., 2019) concerning Zika, GM mosquitoes, and vaccines, finding that both explicit and implicit conspiracy cues increased conspiracy beliefs. However, questions remain. In particular, what are the effects of implicit and explicit cues in a partisan domain?

As such, we tested the effects of both explicit and implicit cues about voter fraud in the 2020 U.S. election. We showed that in this context, explicitly stated partisan conspiracy theories increased conspiracy beliefs among co-partisans and decreased confidence in elections regardless of their agreement with the respondent’s partisanship. Implicit cues, however, did not influence respondents’ attitudes. The findings regarding explicit cues are cause for concern, especially if partisan fraud claims have spillover effects on out-partisans’ perceptions of democratic legitimacy. Uptake of elite cues—even those which erode democratic norms—may be unfortunate in this circumstance, but ultimately are not surprising (Berlinski et al., 2020; Zaller, 1992). But broader spillover effects may present an especially wicked problem for democracies facing emboldened losing candidates (Hernández-Huerta, 2020). Our results demonstrate that even in a landscape saturated in claims of fraud, as the United States remains in the wake of the 2020 election and subsequent contestation of results, voters still respond to novel explicit cues.

On the other hand, there were no effects of implicit cues. It is worth mentioning that we presented respondents with a subtle treatment. The implicit cue merely quoted an elected official pointing out a large decrease in down-ballot roll-off between elections and questioning why this had occurred—with a potential implication that the opposing party may somehow be responsible. This is a minimalistic approach to an implicit cue, based on prior work examining whether journalists may unintentionally transmit conspiracy beliefs (Lyons et al., 2019), and partisans may need a more forceful use of “just asking questions” rhetoric to pick up on such cues. Indeed, our minimalistic approach may even be considered an additional control condition as we have no evidence that respondents processed the treatment as intended, and we should be cautious in drawing any conclusion from the null. Future work in this area should test a fuller range of treatments across the implicit-explicit spectrum.

In general, though, this study’s findings fit with prior work showing that partisans tend to accept in-group elite cues (Campbell et al., 1960; Zaller, 1992), even when those cues violate democratic norms (Berlisnki et al., 2020; Clayton et al., 2021) and provide experimental evidence that conspiracy endorsement, specifically, can be driven by partisan cues (see Miller et al., 2015; Uscinksi et al., 2016). More importantly, however, they also demonstrate broader spillover effects on out-partisans, and as such, suggest that these cues may be even more harmful than prior work indicated. These findings suggest that conspiracy belief per se need not serve as the mechanism to erosions of trust in democratic institutions, because those in the out-party, who tended not to express belief in the uncongenial conspiracy theory, nevertheless expressed decreased confidence in elections. It is possible that these broader spillover effects arise instead due to increased epistemic cynicism in general (Balmas, 2014; Guess & Lyons, 2020; Paul et al., 2016; Pingree et al., 2013).

Findings 

Finding 1: Explicit (but not implicit) conspiracy cues influence directly related beliefs about the cause of voting patterns. These effects are concentrated among co-partisans.

We find significant effects of explicit cues among those for whom they are politically congenial. Democrats exposed to an explicit conspiracy cue targeting Republicans were more likely to agree that Republicans had rigged voting machines (M = 2.59, SD = 1.16) than those in the control (M = 1.90, SD = 1.01). Independents also expressed greater agreement in the explicit cue condition (M = 2.43, SD = 1.08) than in the control (M = 2.08, SD = 1.08). Republicans exposed to an explicit conspiracy cue targeting Democrats likewise increased agreement that Democrats were responsible for the decrease in down-ballot roll-off (M = 3.51, SD = 1.22, vs. M = 3.26, SD = 1.10 in the control). These cues were generally unable to influence out-partisans, and in some cases even backfired among them. Further, we generally found no effects of implicit cues regardless of congeniality. These outcomes are depicted in Figure 1 (see Table B2 in Appendix B for full regression results). We did not detect an increase in conspiratorial responses using an open-ended question prior to the above-mentioned closed-ended items (see Clifford et al., 2019 and Lyons et al., 2019 for discussions of conspiracy belief measurement approaches), with only 8-10% of respondents providing a conspiratorial response across conditions (see Table C2 in Appendix C).

Finding 2: Explicit conspiracy cues have spillover effects on confidence in elections for both co-partisans and those of the opposing party.

We also found that explicit conspiracy cues reduce confidence in elections. Our measure of confidence includes items concerning the ability of the public to cast votes as entitled, the fair count of ballots, security of ballots from tampering, the accuracy of voting machines, and broader trust in the electoral system. We found a negative main effect of both explicit cues relative to the control (see Table B3 in Appendix B). In other words, the decrease in confidence was not contingent on cue congeniality (Table B4, Appendix B). We present the results when collapsing by explicitness (i.e., regardless of target party) in Figure 2 (see also Table C3 in Appendix C). However, we did not detect spillover effects on satisfaction with democracy itself, and the treatments did not appear to increase support in willingness to donate toward improved election security.

Finding 3: Predispositions are strongly associated with fraud beliefs and related attitudes, but do not consistently moderate cue uptake. 

We also tested whether a number of background characteristics made respondents more or less likely to be affected by conspiracy cues. Specifically, we looked at respondents’ predisposition to conspiracy thinking; feelings toward the media; feelings toward Whites, Blacks, and Hispanics; affective polarization; political interest; and political knowledge as potential moderators. We found that these had strong relationships with our outcomes of interest: predisposition to conspiracy thinking and affective polarization were associated with less confidence in elections, for instance, while media trust, political knowledge, and political interest were associated with greater confidence (Table B5, Appendix B). However, these characteristics tended not to have consistent amplifying or dampening effects on conspiracy cue uptake (Tables B6–B12, Appendix B). Based on the large number of exploratory moderation tests (seven moderators for each of the four cues, applied to seven outcome variables, for a total of 196 significance tests) and potential for false positives, we applied the Benjamini-Hochberg Procedure (a statistical procedure that controls for the false discover rate) with an acceptable false discovery rate set at 20%. As a result, four interaction terms were found to be significant. However, three of these did not appear to increase the given cue’s targeted belief and may still be spurious: 1) media affect X the explicit Republican fraud cue’s effect on belief that Democrats manipulated voting machines; 2) affect toward Blacks X the explicit Republican fraud cue’s effect on belief that Democrats manipulated voting machines; and 3) affect toward Hispanics X the explicit Republican fraud cue’s effect on belief that Democrats manipulated voting machines. The fourth significant interaction was affect toward Hispanics X the explicit Democrat fraud cue’s effect on belief that Democrats manipulated voting machines, b = .01, p < .005. Given that none of the moderators produced a consistent pattern of effects, these exploratory results should be viewed with caution.

Methods

We preregistered our hypotheses, measurements, and analyses via OSF (registration: https://osf.io/aynqd/?view_only=886fe9dc77f642f6a16874a534950419; for data and materials see: https://osf.io/wp4az/?view_only=b67e8f7ba71a4e58b16c2d8b8a20d731). The 2,111 respondents in the United States were recruited via Prolific, with quotas set for equal proportions for sex and partisanship (Republican, Democrat, and independent). The final sample was 34.6% Democrat, 33.2% Republican, and 32.2% independent. Respondents ranged in age from 18 to 83 (M = 35.27, SD = 12.72) and the median education level was a four-year degree (44.6% had less than a four-year degree). 79.4% of respondents claimed white racial/ethnic background. For a comparison to American Community Survey (ACS) demographic estimates, see Table A1 in Appendix A. Respondents were paid $1.30 (mean completion time was 8.9 minutes; median time was 7.4 minutes). The data collection period was from November 3–4, 2021 (one to two days after the off-year 2021 election).

Respondents first completed demographic information and potential moderators, as well as two pre-treatment attention checks. No respondents failed both attention checks. Respondents were then asked to carefully read a brief local news article and were assigned to read one of five articles containing one of the following: 1) an explicit conspiracy cue targeting Democrats, 2) an implicit conspiracy cue targeting Democrats, 3) an explicit conspiracy cue targeting Republicans, 4) an implicit conspiracy cue targeting Republicans, or 5) a placebo article about cooking. An elected official in each of the treatment articles noted a decrease in down-ballot roll-off in their state from 71,498 in 2004 (3.98%) to 550 in 2020 (0.03%). Each also briefly defined roll-off voting. In the explicit conditions, the elected official additionally alleges that voting machines had been programmed to switch votes in favor of the opposing party.

Following treatment, all respondents viewed a brief definition of “roll-off” voting, leading into the open-ended measure of conspiracy belief about its decrease in 2020: “‘Roll-off’ is a political science term for when people cast a ballot in some races but don’t bother voting in others. Some have pointed out decreased roll-off in the 2020 election.” Respondents were then asked: “In your view, what likely caused the decrease in down-ballot roll-off in the 2020 election? It’s ok to say you don’t know.” Responses that mentioned intentional (fraudulent) actions by Republicans [Democrats], intentional (fraudulent) actions by prominent elected officials, or manipulation of voting machines were coded as a conspiracy response. All open-ended responses were coded by a set of two independent coders (99.6% agreement on 2,111 responses). Disagreements (0.4% of cases) were resolved through discussion.

Next, all respondents answered closed-ended questions about the role of the Republican and Democratic Parties in the vote disparities. Respondents then answered questions about confidence in elections, satisfaction with democracy, and willingness to donate toward enhanced election security. Finally, all respondents were debriefed.

For full detail on stimuli and measurements, see Appendix A.