Electoral fortunes reverse, mindsets do not

Conservatives and liberals have previously been shown to differ in the propensity to view socially-transmitted information about hazards as more plausible than that concerning benefits. Given differences between conservatives and liberals in threat sensitivity and dangerous-world beliefs, correlations between political orientation and negatively-biased credulity may thus reflect endogenous mindsets. Alternatively, such results may owe to the political hierarchy at the time of previous research, as the tendency to see dark forces at work is thought to be greater among those who are out of political power. Adjudicating between these accounts can inform how societies respond to the challenge of alarmist disinformation campaigns. We exploit the consequences of the 2016 U.S. elections to test these competing explanations of differences in negatively-biased credulity and conspiracism as a function of political orientation. Two studies of Americans reveal continued positive associations between conservatism, negatively-biased credulity, and conspiracism despite changes to the power structure in conservatives’ favor.


Overview
Section 1 of this supplement provides supporting details for main text claims, with a structure paralleling that of the main text.
Section 2 presents other analyses that are interesting but largely tangential from the focus of the main text.
Section 3 describes additional analyses, publically available in raw form, that provide extensive details of all models summarized and/or referenced in earlier sections.

Results
Political Orientation and Credulity. In main text we summarized results of linear models predicting the difference between weighted hazard credulity and weighted benefit credulity (i.e. negatively-biased credulity). The full models for Study 1 and Study 2 are displayed in Tables 1 and 2, respectively.
We claim that substituting party for conservatism score produces similar results, and showed a figure comparing raw conservatism scores by party. Republican affiliation is associated with higher weighted negatively-biased credulity in Study 1 (Table 3) and Study 2 (Table 4). We claim that the relationship between conservatism and negatively-biased credulity is "robust to the removal of any one credulity item, across a suite of models with alternative predictors sets (e.g. conservatism subscales), or using unweighted credulity." We now address these three claims.

Item sensitivity.
Is the relationship between conservatism and credulity reported in the main text dependent on any single credulity item?
To investigate this question of item sensitivity, for each study we systematically calculate 16 different versions of the weighted credulity index, removing a different item for each version, and then fit linear models of the resulting credulity score as a function of conservatism, confidence, and demographic items. We then compare those 16 models based  Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data. AIC comparisons of the 16 models for Study 1 based on all scales with a single dropped item is shown in Table 5. The best (Table 6) and worst models (Table 7) removed the "sharks" and "lightning" items, respectively. Both show a positive relationship between negatively-biased credulity and conservatism. The best (Table 9) and worst models (Table 10) removed the "exit door" and "credit card phone" items, respectively. Conservatism is associated with negatively-biased credulity in both.

Unweighted credulity.
We re-ran the models presented in the main text, except that unweighted negatively-biased credulity scores are the DV instead of weighted credulity. For both studies (Study 1: Table 11, Study 2: Table 12), conservatism remains positively associated with negatively-biased credulity.

Alternative predictor sets.
Main text featured models similar to those used by Fessler, Pisor and Holbrook (2017).
We fit a series of alternative linear models of weighted negatively-biased credulity as functions of various combinations of predictor variables (overall conservatism, conservatism subscales, party, confidence) and/or their interactions; all models include demographics as predictors. We can then use AIC to compare all models for a given outcome.
Here we show that, across a wide variety of reasonable linear models, our data support the core claim that political conservatism is associated with greater hazard-biased credulity. AIC comparisons (Table 13) suggest that the main text featured model ("con_all") is unlikely to be the best model. However the best model (Table 14) all available models (see Section 3) are intepretable as showing that conservatism (whether overall or a subscale) and/or Republican affiliation are associated with higher levels of weighted negatively-biased credulity.

Study 2.
For Study 2, we first fit the same series of models as used for Study 1. After this we will consider models that also include the political confidence measure that was only used in Study 2.
AIC comparisons (Table 15) suggest that the main text featured model ("con_all") is unlikely to be the best model. However the best model (Table 16) all available models (Section 3) are intepretable as showing that conservatism (whether overall or a subscale) and/or Republican affiliation are associated with higher levels of weighted negatively-biased credulity. Study 2: Adding confidence as a predictor.
Among the model set allowing political confidence as a predictor, AIC comparisons (Table 17) suggest that the main text featured model ("con_all") is unlikely to be the best model. However the best model (Table 18) all available models (Section 3) are intepretable as showing that conservatism (whether overall or a subscale) and/or Republican affiliation are associated with higher levels of weighted negatively-biased credulity.
Confidence is included in many of the best models, and is always a negative predictor of credulity, but typically with a borderline-insignificant standarized coefficient roughly 1/3rd the size of conservatism (Section 3). Furthermore no models including an interaction term involving confidence are among the best models. Note. All models also include demographics. Lower AICc implies a more parsimonious model.
AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Political Confidence.
In main text we claim that confidence is a borderline-insignificant negative predictor of negatively-biased credulity in the best-fitting models. This is supported in the section above on alternative models for Study 2.
We also claim that this negative relationship is a result of confidence being a stronger positive predictor of benefit credulity than of hazard credulity. This is supported below in Section 2.

Methods
Claim: conservatism was associated with major party affiliation. See  Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Section 2: Other Interesting Results
In the first part of this section, we analyze weighted hazard and benefit credulity separately, as a way of breaking down the main text results that focus on difference scores.
We first see that there is a consistent positive association between conservatism and hazard credulity, but this pattern is not consistent for benefits credulity. We then consider more complicated models. For either form of credulity, we find that the best models include an interaction between party and conservatism. Decomposing this interaction reveals significant positive associations between conservatism and (either type of) credulity within Democrats, but not within Republicans. We then compare the role of political confidence in predicting hazard or benefit credulity, finding that it is positively associated with both, but more so with benefit credulity. This pattern results in the negative relationship between political confidence and negatively-biased credulity.
Next we analyze conspiracy mentality, finding that conservatism is strongly positively associated, even when accounting for political confidence.
Finally we look at sex differences in the estimated magnitudes of hazard and benefit claims. The only noteworthy finding is that, in Study 2, men rate hazards as less severe than do women.

Hazard credulity and Benefit credulity
Here we analyze only weighted hazard credulity or weighted benefit credulity, rather than the difference between them, as in the main text. First, note that overall there was a positive relationship between overall conservativism and hazard credulity in Study 1 ( Figure   3) and Study 2 ( Figure 5), whereas conservativism predicted benefit credulity in Study 1 ( Figure 7) but not Study 2 (Figure 9).
To probe predictors of credulity more thoroughly, we fit a series of models with various sets of predictors, then compare models based on AIC, as above.
For both studies and both credulity types (Tables 22 -25 best-fitting models (and all lesser models), and the decomposition of the significant interaction terms see Section 3. In all cases, the interaction decomposes such that Democrats show a significant positive association between overall conservatism and credulity, whereas there is no significant relationship between those variables for Republicans. These patterns are visualized in Figures 4 -10, which show that Republicans as a group are both more conservative and more hazard-credulous than Democrats, and that Democrats who approach Republican-level conservatism also approach Republican-level hazard credulity. There are no party differences in benefit credulity. Within party, Republicans show no association between benefit credulity and overall conservatism; Democrats show a positive relationship. Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Confidence, Hazard, and Benefit
In the main text we claim that confidence is a negative predictor of negatively-biased credulity because confidence predicts benefits credulity more strongly than hazard credulity.
Here we support this claim.
Tables 26 & 27 compare models predicting hazard and benefit credulity, respectively.
In the best model of hazard credulity (Table 28), confidence is a significant positive predictor, with a standarized coefficient of roughly .10. In the best model of benefit credulity (Table 29), confidence predicts benefit credulity with a standardized coefficent of .63, although this is qualified by a 3-way interaction term. In the second best model of benefit Note. All models also include demographics. Lower AICc implies a more parsimonious model.
AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.
credulity (Table 30), which is not much worse than the best model, confidence predicts benefit credulity with a standardized coefficent of .21, and this parameter is at least as large when decomposing the 3-way interaction term in the first model. Thus we conclude that confidence has a stronger relationship to benefit credulity than to hazard credulity, which drives the negative relationship between confidence and negatively-biased credulity. Note. All models also include demographics. Lower AICc implies a more parsimonious model.
AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Conspiracy
Here we analyze conspiracy mentality. First fit a series of models with various sets of predictors, then compare models based on AIC, as above.

Study 1.
Model comparisons for Study 1 are displayed in Table 31. The best two models (Tables 32 & 33) shows that conservatism and hazard-biased credulity are positively associated with conspiracy mentality, as are being non-white, less educated, and older. All models are presented in Section 3. Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.      Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Study 2.
Model comparisons for Study 2 are displayed in Table 34. The best model (Table 35) shows that conservatism and hazard-biased credulity are positively associated with conspiracy mentality, as is being female. All models are presented in Section 3. If confidence is allowed as a predictor of conspiracy mentality, it appears in the best models (Table 36) as a positive predictor but conservatism remains a strong positive predictor (Table 37). Note. All models also include demographics. Lower AICc implies a more parsimonious model. AICcWt is interprettable as the weight of evidence for the model being the best among candidates, given the data.

Sex differences
Models of negatively-biased credulity often show that men are less credulous than women, e.g. the featured model for Study 2 in the main text. This trend is consistent with arguments about sex differences in the general valuation of costs and benefits (Sparks et al, 2017).
To further examine this trend, we created scales using only the magnitude values (i.e. the weights) for hazard items ("costs") or benefit items ("benefits"). In Tables 38 -41, we model those as a function of conservatism and demographics, finding that only cost estimates show a sex difference, and only for Study 2.