Risk Literacy Promotes Representative Understanding: Numerate People are Less Biased, More Knowledgeable, and More Concerned about Climate Change


 Risk literacy skills, as measured by numeracy tests, are robust predictors of objective knowledge and risk understanding. However, for some people with extreme cultural worldviews, research suggests numeracy might slightly increase polarization of subjective perceptions of climate change risks. Here, we report the first integrated tests linking skills, worldviews, objective knowledge, beliefs, and subjective perceptions among diverse adults. Compared to less numerate people, regardless of cultural worldviews, highly numerate people were 5-8 times more likely to have accurate knowledge and beliefs (52% vs. 24% incorrect), and 3 times more likely to have above average risk perceptions. Structural modeling suggests numeracy may typically promote acquisition of accurate climate change knowledge, which then robustly informs beliefs and perceptions (e.g., up to 40 times stronger influence than worldviews). Even among people with extreme worldviews, rather than amplifying polarization, numeracy was associated with more representative understanding of climate change risks (e.g., well-informed and coherent).

those who do not). Moreover, the study did not assess or document any differences in objective 94 non-normative biases or polarization related to knowledge or personal beliefs. 95 For these and other reasons, we conducted the first two studies to address neglected roles 96 of objective knowledge and beliefs, testing an integrated cognitive model of the relations 97 between numeracy skills and (objective and subjective) climate change judgments. Using a 98 probabilistically representative sample of the U.S. adult population, Study 1 tested a structural 99 equation model mapping the influence of worldviews and numeracy on accurate knowledge, 100 beliefs, and climate change risk perceptions. Using a convenience sample of diverse U.S. adults, 101 Study 2 provided an out-of-sample test of the model from Study 1, adding a novel test of the role 102 of specific, general, and relative risk perceptions (i.e., risk of climate change compared to risk 103 perceptions about other risks to society in general, such as nuclear power and vaccines). were estimated using a bootstrapping method (5,000 bootstraps). Tests of relevant and plausible 120 potential interaction effects were all found to be statistically trivial and unreliable. 121 122

Knowledge Predicts Climate Change Beliefs and Attitudes 123
To provide more context on the relative predictive power of knowledge on belief in 124 anthropogenic global warming, we constructed a binary logistic regression model (see 125 Supplementary predicted probabilities for each of the dichotomized outcome variables across the entire range of 146 numeracy (0-7) was also computed and plotted across numeracy scores for average and extreme 147 cultural worldview groups, based on bootstrapped binary logistic regression model estimates (see 148

Numeracy Predicts Beliefs and Risk Perceptions 151
A binary logistic regression predicting belief in anthropogenic global warming indicated that 152 individuals with the highest numeracy score were nearly 5 times more likely to personally agree 153 with anthropogenic global warming, as compared to those with the lowest numeracy score (see 154

Numeracy Predicts Relative Risk Attitudes 163
All key analyses and models from Study 1 were replicated in Study 2, providing out-of-sample  Table 9-13 for details). However, in Study 2, in addition to all measures from Study 1, we also 170 measured differences in general risk perception (e.g., how risky is climate change and how risky 171 are other general risks such as vaccines, nuclear power, etc.). Accordingly, we constructed and 172 suggest that while numerate people may appear to be no more worried about the overall risks of 180 climate change as compared to less numerate people (e.g., specific climate change risk 181 perceptions), numerate people perceive the risk of climate change to be much more concerning 182 as compared to other risks faced by society (e.g., less numerate people reported that climate 183 change risks were similar to those of most other risks to society).
To provide more context on the relationship between numeracy and general risk 185 perception, a binary logistic regression was constructed (see Supplementary Table 14)  demonstrate that numerate people may typically be less worried about most risks to society in 215 general, as compared to less numerate people. While they were less worried in general, numerate 216 people were significantly more worried about the relative risks of climate change compared to 217 other risks faced by society (see Figure 5). In theory, differences in general risk perceptions of 218 numerate people may at least partially reflect the fact that numerate people tend to know more 219 about many risks in general. To the extent this finding generalizes, other studies that fail to 220 measure specific, general, and relative risk perceptions (i.e., specific minus general perceptions) 221 are more likely to derive distorted estimates of people's judgment and reasoning biases (e.g., 222 interpreting differences in the perceived risk of a $5,000 repair, without considering whether the 223 person is a college student or a millionaire). The tendency to neglect these base-line priors may 224 help explain emerging findings showing numeracy consistently predicts risk understanding even 225 when it appears inconsistently related to specific risk perceptions (e.g., numeracy is the strongest 226 predictor of COVID-19 mis/understanding, yet is not always related to specific COVID-19 risk Ironically, it seems some scientific studies investigating potential biases of people who 252 might ignore climate change science have ignored differences in people's knowledge of climate 253 change science, resulting in biased scientific estimates of people's climate-change biases. This is 254 problematic for many reasons. For example, it would be rational for a well-informed, numerate 255 public to reject the views of experts if the experts were found to be demonstrably wrong or 256 biased. The current findings serve as a powerful example of the risks and interpretive errors that 257 can result from neglecting differences in knowledge, skills, values, and base-line perceptions, 258 which in-turn may potentially confound theory, justifiably threaten credibility, and misdirect 259 practical applications and critically-needed investments. 260

Risk Literacy Promotes Acquisition of Accurate Knowledge. Even among people with 261
conflicting and extreme cultural worldviews, the current set of studies indicate that numeracy 262 tends to be robustly associated with more accurate acquired prior knowledge about climate 263 change, which in turn is associated with reduced biases in downstream beliefs and risk attitudes. 264 While results suggest that numeracy is unlikely to polarize subjective attitudes (directly or 265 indirectly), it is possible that numerate people with extreme views may be more likely to 266 selectively evaluate and acquire evidence in biased ways, consistent with their cultural 267 worldviews. In one of the first and only studies attempting to document these potential effects, 268 Kahan and colleagues (2017b) examined differences in interpretation of fictitious evidence about 269 gun control versus rash treatments. The study suggested that more numerate people with 270 extreme worldviews may have applied different standards for accepting/rejecting evidence in 271 self-serving ways. Unfortunately, the relationship between numeracy and prior knowledge (gun 272 control vs. rash treatment) was again neglected in that study. As such, it is unclear if the different reflected differences in the interpretation of questions about the evidence, which can happen 275 when prior knowledge is neglected (see Gigerenzer et al., 1999;Reyna, 1991). Moreover, the 276 fictional raw pattern depicted a very weak relationship, which may be further complicated by the 277 fact that, rationally, how one should interpret new evidence is at least partially a function of how 278 much one knows about prior knowledge (e.g., the more one knows, the less valuable any new 279 weak and inconsistent evidence is likely to be). 280 While we think it is possible that numeracy may sometimes promote polarization in some 281 kinds of reasoning, we caution that such differences will not necessarily imply errors but could 282 instead simply involve well-reflected differences in fundamental values and subjective 283 preferences (e.g., people, including experts, often legitimately disagree about moral issues such 284 as religion, free will, personal responsibility, intentionality; Feltz & Cokely, 2018, Schulz et al., 285 2011). That said, given the observed presence of a robust relationship between numeracy and 286 accurate knowledge in the current studies, and in many others, we find no compelling evidence 287 indicating that numeracy is generally related to objectively non-normative biases or polarization. 288 Even among those who had extreme worldviews, numeracy was found to be robustly related to 289 accurate knowledge, which was by far the strongest factor influencing judgment accuracy and 290 coherence. The estimated independent association between numeracy and knowledge in the 291 current study was almost 20 times larger than the estimated magnitude of the significant 292 polarization reported in the previous climate change study (Kahan et al., 2012). Consistent with 293 Skilled Decision Theory, these results suggest that even when considering controversial issues 294 and complex conflicts of interest, when there is compelling evidence available (e.g., well-295 founded expert consensus), numeracy and risk literacy skills are likely to promote independent acquisition of more accurate and representative risk understanding, independent of people's 297 diverse worldviews and values. 298 299 Methods 300

Study 1 301
Data Collection. The representative sample of the U.S. population was collected in Spring of 302 2016, using a probability-based sampling panel (KnowledgePanel ® from GfK). A total of 305 303 cases were reported for the analysis (see Supplementary Table 3 for demographic 304 characteristics). 305

Measures. 306
Statistical Numeracy. The Berlin Numeracy Test (Cokely et al., 2012) and a three-item 307 scale created by Schwartz et al. (1997) were used to assess numeracy and risk literacy (e.g., In a 308 forest, 20% of the mushrooms are red, 50% are brown, and 30% are white. A red mushroom is 309 poisonous with a probability of 20%. A mushroom that is not red is poisonous with a probability 310 of 5%. What is the probability that a poisonous mushroom in the forest is red?). Using the two 311 tests together increases sensitivity of the measurement, allowing for a wider range of skill 312 assessment.   Note. Results depict the predicted probability for belief in anthropogenic global warming (a) and having above-average global warming risk perception (b) at different levels of knowledge and cultural worldviews. Individualists (egalitarians) were defined as having individualism (egalitarianism) ratings 1 standard deviation above mean, and egalitarianism (individualism) ratings 1 standard deviation below the mean, while the other worldview indices were held at their mean. The shaded area represents 90% confidence intervals from 1,000 bootstrap iterations.