Repetition increases belief in climate-skeptical claims, even for climate science endorsers

Statistical methods All statistical analyses were conducted using SPSS version 28.0 [43]. To investigate the effect of claim type on the size of illusory truth, in each experiment, we conducted a linear mixed effects model analysis with repetition (repeated vs. non-repeated) and claim type (scientist vs. skeptic) as fixed effects (using numerical effects coding), participant as a random effect, and perceived truth as the outcome variable. Further, in Experiment 2, we added participants’ general endorsement of climate change science as a covariate to the above model. The analyses were preregistered. These linear mixed models are further supplemented by t-tests to examine differences between claim type (scientist/skeptic) and/or repetition (repeated/non-repeated), as a direc- tional follow-up of main and interaction effects. We also present analyses of the main research question across various subsets of the full sample as categorized by climate change concern using SASSY [44], to look at the stronger endorsers of climate science. The linear mixed effects model is the overarching analysis for each experiment (and is consistent with pairwise comparisons). All t-tests are within-subjects, reporting significance for the two-tailed test. We used paired-samples Cohen’s d using a corrected standard deviation of the difference to estimate effect size and corresponding confidence intervals across all t-tests, and report Hedges’ gav which corrects for bias in d. We calculated effects sizes [45] in the linear mixed models using

estimated means and SDs. For all tests, the threshold for significance was p = .05. For all bar graphs, error bars show 95% confidence intervals of each estimated cell mean. To ensure that our data meet the assumptions for linear mixed models, we plotted residuals against predicted values to check for homogeneity of variance and created a histogram for residuals to check for normality of error term. We further examined assumptions for t-tests through inspecting a histogram of differences between pairs to check for normality of differ- ences and extreme outliers. All assumptions were met for both experiments. Experiment 1 Participants. Of 100 HITS posted between the recruitment period from 28/09/2020 to 29/ 09/2020, we received 99 completed responses (65% identified as male, 35% identified as female, 0% other; aged 22–68; 95% American) with no missing data. As preregistered (https:// aspredicted.org/blind.php?x=3jn4m8), we excluded 20 people who looked up the answers and 27 who failed the open-ended bot-check question. We retained 52 participants for analysis (59% identified as male, 41% identified as female, 0% identified as other/non-disclosed; aged 25–68; 95% American), exceeding the requirements of the power analysis. Design, materials, and procedure. We used a fully within-subjects 2 (repetition: repeated, non-repeated) x 2 (claim type: scientist, skeptic) repeated-measures design. For Experiment 1, we piloted an initial set of 42 climate-related claims that people may encounter in everyday life, asking people to categorize each claim as being consistent with

views of a climate scientist, climate skeptic, or unknown. From these pilot data we selected 8 target claims that the majority of participants had classified as being consistent with views of a climate scientist and 8 target claims classified as consistent with a climate skeptic; 4 claims of each type were objectively true and 4 objectively false. We also piloted 32 weather claims with similar perceived validity (16 true and 16 false claims) to act as filler items. All claims and scales across both experiments are listed in S1 File. To implement the standard ITE paradigm, we created two counterbalances of claims. Each counterbalance contained 4 scientist claims, 4 skeptic claims, and 8 filler weather claims, PLOS ONE Illusory truth effect for climate-skeptical claims PLOS ONE | https://doi.org/10.1371/journal.pone.0307294 August 7, 2024 4 / 19