Risks

RESEARCH ARTICLE

I’ll See It When I Believe It: Motivated Numeracy in Perceptions of Climate Change Risk* Matthew S. Nurse and Will J. Grant

Australian National Centre for the Public Awareness of Science, Australian National University, Canberra, Australia

ABSTRACT People’s attitudes about Anthropogenic Climate Change (ACC) risks are not only influenced by scientific data, such as the likelihood of harm, the consequences of failing to act and the cost and effectiveness of mitigation. Instead, when people receive information about controversial topics of decision-relevant science like ACC they often defer to their political attitudes. Recent research has shown that more numerate people can be more polarized about these topics despite their better ability to interpret the scientific data. In this study, we investigated whether the motivated numeracy effect originally found by Kahan, Peters, Dawson, and Slovic [2017. Motivated numeracy and enlightened self-government. Behavioural Public Policy, 1(1)] on the controversial topic of gun control laws in the United States also applies to people when assessing ACC risks. This randomized controlled experiment (N = 504) of Australian adults extends the motivated reasoning thesis by finding evidence that highly numerate people who receive scientific data about ACC use motivated numeracy to rationalize their interpretations in line with their attitudes.

ARTICLE HISTORY Received 6 November 2018 Accepted 7 May 2019

KEYWORDS Climate change communication; motivated numeracy; motivated reasoning; identity-protective cognition; rejection of science

Introduction

If people, organizations or societies correctly identify risks they can effectively mitigate them. How- ever, groups of people don’t always perceive risks objectively. Indeed, on many issues where there is a clear scientific consensus society’s risk perceptions can be highly polarized. For example, researchers have found instances where groups are polarized about the risk of Anthropogenic Climate Change (ACC) (Lee, Markowitz, Howe, Ko, & Leiserowitz, 2015; Pew Research Center, 2016, p. 22), geneti- cally modified foods (Desaint & Varbanova, 2013; Lamberts, 2017; Pew Research Center, 2015) and the theory of evolution (Swift, 2017; Miller, Scott, & Okamoto, 2006) despite there being scientific agreement about these topics.

Therefore, it is important to understand why people reject scientific evidence. Many people assume that providing facts to resolve information deficits in people’s knowledge would cause audi- ences to form attitudes in line with the evidence (Sturgis & Allum, 2004). Following this line of think- ing, the “science comprehension thesis” (SCT), speculates that those who are better able to correctly interpret data – for example, those with high levels of science literacy, numeracy and education levels – should correctly perceive risks expressed in a data format better than others (Kahan, Peters, Wit- tlin, & Slovic, 2012). If this theory were universally correct, groups of people with the highest levels of science literacy, numeracy and education should, if they are given enough information, be much less

© 2019 Informa UK Limited, trading as Taylor & Francis Group

CONTACT Matthew S. Nurse [email protected], [email protected] *The data that support the findings of this study are openly available in figshare at https://doi.org/10.6084/m9.figshare.7300430 This article has been republished with minor changes. These changes do not impact the academic content of the article.

ENVIRONMENTAL COMMUNICATION 2020, VOL. 14, NO. 2, 184–201 https://doi.org/10.1080/17524032.2019.1618364

polarized about issues on which there is scientific consensus. However, this isn’t always the case, especially in topics of decision-relevant science that have become publicly controversial. While greater education may increase the awareness of ACC it is not a consistent predictor of accurate ACC risk perceptions (Lee et al., 2015). In a meta-analysis of studies from around the world Hornsey, Harris, Bain, and Fielding (2016) found instead that the factor that most strongly predicts this polar- ization is a person’s political affiliation, with people with an affiliation to liberal political parties more likely to believe in ACC risks than those who identify with conservative political parties. There is also evidence that highly educated and science-literate people are more polarized about ACC risks (Drummond & Fischhoff, 2017; Hamilton, 2011; Kahan, 2017; Kahan et al., 2012; McCright & Dun- lap, 2011; McCright, Marquart-Pyatt, Shwom, Brechin, & Allen, 2016).

Researchers have observed similar levels of polarization among more educated or more science- literate people in other publicly controversial topics of science including support for stem cell research (Drummond & Fischhoff, 2017; M. C. Nisbet & Markowitz, 2014), the well-established the- ory of human evolution (Drummond & Fischhoff, 2017; Ho, Brossard, & Scheufele, 2008) and whether scientists are a trusted source of information about vaccines (Hamilton, Hartter, & Saito, 2015). This observational data suggest that SCT does not fully explain people’s attitudes to ACC risk.

Political polarization and motivated numeracy

Kahan, Peters, Dawson, and Slovic (2017)1 conducted an experiment to determine whether what they called the “identity-protective cognition thesis” (ICT) could explain why the ability to under- stand scientific data is often associated with political polarization in observational studies. In contrast to SCT, ICT suggests that when people are asked to interpret facts about publicly controversial topics they will often provide an answer that complies with the attitudes of the social groups that they ident- ify with, rather than the objectively correct answer (Kahan et al., 2017). If ICT holds true, polariz- ation between ideologically motivated groups would not always reduce at higher levels of numeracy.

To test these theories against each other, Kahan et al. (2017) asked participants to interpret the results of a two-by-two contingency table. Participants in the control group were told the data in the table measured the overall effectiveness of a newly developed skin rash cream. Meanwhile, partici- pants in the experimental conditionwere told the same data tablemeasured the effect of gun control on crime rates – a highly politically polarizing topic in the United States, where they conducted the study. To correctly answer this deliberately difficult “covariance detection problem” (Kahan et al., 2017, p. 84) participants would need to consider the ratio of positive to negative outcomes. While people with higher levels of numeracy tended to give the objectively correct answer when they were told it measured the effectiveness of a new rash cream, when others were told the data related to gun control, they tended not to provide an objectively correct answer, but an answer aligned to the attitude of their preferred political party. Therefore Kahan et al. (2017)’s resultswere consistent with ICT and they con- cluded that “numeracymagnified political polarization amonghigh numeracy partisans” (Kahan et al., 2017, p. 75). This finding is somewhat counterintuitive and is worthy of further investigation.

Theory

Previous research explains the mechanisms that may help to understand these results. Psychologists such as K. E. Stanovich and West (2001) and Kahneman (2012) suggest that human thinking pro- cesses can be categorized into “system one” thinking (which is the default mode and is an intuitive, heuristic type of thinking which requires little effort but is more prone to biases), and “system two” thinking (which is more systematic and logical). System two can be used to override system one for more demanding cognitive tasks but only for a short period of time, as it requires effort. Importantly for this research, different people have different capacities for system two thinking. This includes people who have higher levels of numeracy. More numerate people can more readily engage their system two mode for difficult calculations, while less numerate people must rely more on using

ENVIRONMENTAL COMMUNICATION 185

system one (Chaiken, 1980; Shelly Chaiken & Trope, 1999; Kahneman, 2012, pp. 19–30; Petty & Cacioppo, 1986; K. E. Stanovich & West, 2001). This suggests that more numerate people would be more likely to correctly interpret a contingency table, however Kahan et al. (2017) suggests that this is conditional on whether the topic of that table affirms or threatens a person’s attitudes.

Kahan et al. (2017, p. 66) suggest thatmore numerate peoplemay therefore be better at rationalizing an interpretation based on their political attitudes. This motivated numeracy effect suggests that more numerate people can not only use system one thinking for relatively easy calculations and use system two to arrive at an objectively correct answer to a difficult calculation, but they can also use system two to rationalize their beliefs in the face of contradictory evidence, in a defense mechanism to minimize cognitive dissonance (Kahan et al., 2017, p. 75).Using this path is a formofmotivated reasoning, where a person’s desire to reach a conclusion consistent with their attitudes influences their interpretation of information (Ahern, Connolly-Ahern, & Hoewe, 2016; Kunda, 1990; Xue, 2015). This can happen either by being more likely to accept information that affirms existing beliefs (Nickerson, 1998) or by dismissing information that threatens it (Lord, Ross, & Lepper, 1979).

Previous studies have demonstrated that a person’s attitudes can distort the empirical perceptions of risk in areas of decision-relevant science (Kuhn, 2001; Lewandowsky, Gignac, & Oberauer, 2013; M. C. Nisbet, 2005; Visschers, Visschers, Shi, Siegrist, & Arvai, 2017; Whitfield, Whitfield, Rosa, Dan, & Dietz, 2009). However a person’s ability to arrive at their preferred conclusion is limited by “their ability to construct seemingly reasonable justifications for these conclusions”, such as their system two processing ability (Kunda, 1990, p. 480). However, it does not appear that higher levels of intel- ligence reduce the use of motivated reasoning (Stanovich, West, & Toplak, 2013).

There has been some debate about ICT by Van der Linden (2016) and how science communi- cation practitioners should use Kahan’s findings when communicating about ACC (Van der Linden et al., 2017). In particular Van der Linden et al. (2017) suggests there may not be a simple dichotomy between ICT and SCT, calling for “a more nuanced perspective that carefully integrates – rather than polarizes – well-established theories to evaluate what works, when, for whom, and in what context will lead to a more accurate and informed view of human responses…” (Van der Linden et al., 2017, p. 457). Van der Linden (2016, p. 132) also notes that more research should be done to determine the extent to which ICT may motivate residents of countries other than the United States to process data about decision-relevant science that is considered controversial to them. This is important to the cur- rent investigation because there is some evidence to suggest that identity may play a greater role in attitudes to ACC in the United States than in other countries (Hornsey, Harris, & Fielding, 2018).

Therefore, better understanding the role identity plays among populations and audience segments when processing scientific information such as ACC is a priority for science communication research. The National Academy of Sciences (2017) has called for more research into knowledge gaps in this developing field, including understanding more about how to better communicate topics of publicly controversial science, why some people ignore numerical data when processing decision- relevant science, and the role that social networks and norms play in science communication (National Academy of Sciences, 2017).

Previous research into motivated numeracy

At the time we conducted our research two other researchers had sought to replicate Kahan et al. (2017)’s findings.

Fuller (2015) conducted an experiment with 2,269 Australian participants. This study explored a range of different experimental questions in addition to gun control, as this topic was not considered to be as polarizing in Australia as it is in the United States. These additional experimental questions were the effect of living near a nuclear power station on population cancer diagnosis rates, the effect of refugee settlement on crime and the economic effects of CO2 emissions reduction targets. Polar- ization tended to occur among participants with moderate levels of numeracy, however, contrary to what ICT predicts, at the highest numeracy levels polarization did not occur. However, there were

186 M. S. NURSE AND W. J. GRANT

some notable limitations to this experiment. Firstly, the survey was distributed through Science Alert, a popular science news Facebook page and website. Due to these dissemination methods there was a significant skew in the sample (Fuller, 2015, p. 13) which was much more numerate than the original experiment (mean = 7.3 out of 9, SD = 1.8, compared to mean = 3.7, SD = 2.1).

The only published replication study, conducted by Ballarini and Sloman (2017), concluded that they could not find evidence of motivated numeracy. However, they acknowledged their sample of 55 experiment participants “…may have been too small to detect the effect of motivated numeracy those researchers found.” As Kahan and Peters (2017) noted, the different experiment design employed, the statistical analysis technique used and the relatively homogenous sample collected may also have hampered Ballarini and Sloman (2017)’s ability to replicate the results. Kahan and Peters (2017) responded by successfully conducting a direct replication of the original experiment with 800 participants. After our research was conducted but before its publication we became aware of a study currently in the preprint stage by Sumner, Scofield, Buchanan, Evans, and Shearing (2019) which analyzed 11,225 British voters during the 2016 United Kingdom European Union membership referendum, and found evidence of ideologically motivated numeracy when highly- numerate participants considered the same contingency table presented by Kahan et al. (2017) with the introduction of the table changed to be about the effect of immigration levels on crime rates.

Methods

To refine the question for this study, we conducted three small (n = 50) trials each with an Australian sample via SurveyMonkey. We used the following criteria to select a suitable ACC research question to attempt to replicate Kahan et al. (2017) as closely as possible:

1. It is likely to be politically polarizing. 2. The independent variable is numeric. 3. The dependent variable is numeric. 4. There is a plausible relationship between the independent and dependent variables. 5. It is related to a realistic risk.

Using the data from the trial surveys we were not able to find an ACC topic that adequately polarized supporters of Australia’s two major political parties, the Australian Labor Party and the Liberal/National Coalition, however these trials surveys suggested asking about the forced clo- sure of coal-fired power plants would polarize supporters of two of Australia’s minor political par- ties, the Australian Greens party from the ideological far left and the One Nation party from the ideological far right. Polling data indicated that 10 per cent of Australian voters would vote for the Australian Greens and eight per cent would vote for the One Nation party in the 2019 Austra- lian election (Essential Research, 2018a). This investigation’s research topic was whether there would be a significant (more than 30 per cent) reduction in carbon dioxide (CO2) emissions in cities where governments forcibly closed nearby coal-fired power plants.

This topic met the selection criteria. Recently published polling data also confirmed that suppor- ters of these two political parties are polarized about whether Australia’s coal-fired power stations should continue operating (YouGov, 2017) and whether Australia should take stronger action to reduce CO2 emissions (Essential Research, 2018b; The Australia Institute, 2018) aligned to the pol- icies of the political party they support (Dickson, 2017; Slezak, 2016) and their ideological views (Guy, Kashima, Walker, & O’Neill, 2014). Put simply, the majority of Greens voters support strong government-led action on climate change, while a majority of One Nation Party voters don’t. Our topic also presents a realistic proposal, as an Australian Senate inquiry report proposed the option of regulating the closure of coal-fired power stations to reduce CO2 emissions (The Senate Environ- ment and Communications References Committee, 2017). There is also a plausible relationship between the variables in this topic, it can be expressed numerically, and it is related to risks

ENVIRONMENTAL COMMUNICATION 187

potentially affecting the participants, being the risk of ACC related harm and the risk of energy bills rising if large generators are forced to cease operating.

Hypotheses

This research seeks to conceptually replicate Kahan et al. (2017)’s key third hypothesis in Australian participants on a climate change topic. Our research question is whether ideological polarization in each ACC experimental condition will be the same among participants with lower levels of numeracy and those with higher levels of numeracy.

Participants

A demographically diverse sample of Australian adults was recruited to participate by YouGov, a professional opinion polling company. YouGov administered the questions through their online sur- vey platform.

Participants were selected if they stated they intend to vote for the Australian Greens or the One Nation Party at the subsequent Australian national election. Those that indicated they would vote for other political parties were excluded, leaving 504 participants in our sample. These participants were financially compensated for their time. The survey was conducted from 16 March to 20 March 2018, at a time when the future use of coal-fired power plants was being publicly debated (Martin, 2018; Massola, 2018; Packham, 2018). There was no missing data.

Ethical approval for this research was granted by the Australian National University Human Research Ethics Committee Protocol 2017/791.

Design and variables

Like Kahan et al. (2017) this experiment used a between-subjects design. The sample consisted of equal numbers of participants who supported two selected political parties, determined by asking them about their voting intention for the subsequent Australian national election (Australian Greens n = 252, One Nation Party n = 252). The independent variables were the subject’s political party voting intention (a binary categorical variable) and their numeracy score (measured from 0 to 9). The dependent variable “correct” is a binary dummy variable recording whether the participant answered the experimental question correctly for their allocated condition (1) group or not (0).

Table 1 identifies the main differences and similarities between the current investigation and (Kahan et al., 2017), informed by the replication guidelines outlined in Brandt, Brandt, Ijzerman, Dijksterhuis, and Farach (2014).

Procedure

Our survey obtained demographic data from the participants. As in Kahan et al. (2017) participants then completed a nine-question numeracy test adapted fromWeller et al. (2013). The mean number of correct responses was 4.75 (SD = 2.17). The distribution of numeracy score is approximately nor- mally distributed (Skewness = 0.71) but with a light-tailed distribution (Kurtosis =−2.03). An inde- pendent t-test found no significant difference in the mean numeracy scores for Greens (M = 4.89, SD = 2.34) and One Nation supporters (M = 4.60, SD = 1.97; t (504) =−1.50, p = 0.134, two-tailed).

Each participant was asked to interpret a two-by-two contingency data table apparently showing the results of scientific studies about the effectiveness of a stimulus. We used the same two “skin rash” control questions as Kahan et al. (2017), and for the two experimental conditions we asked partici- pants to consider data about cities where governments had forcibly closed coal-fired power stations and whether that resulted apparently in a 30 per cent reduction in CO2 emissions or not. Figure 1 shows the questions we presented to our participants. To ensure we tested supporters of each

188 M. S. NURSE AND W. J. GRANT

Table 1. Comparisons of the features of the current investigation and Kahan et al. (2017).

Feature Current investigation Kahan et al. (2017) Same / close /

different

Instructions Online survey Online survey Same Task Interpretation of a two-by-two

contingency table Interpretation of a two-by-two contingency table

Same

Control topic Skin cream Skin cream Same Experimental topic

The effect of coal-fired power stations on CO2 emissions

The effect of gun laws on crime rates Different

Measures Independent variables: . Numeracy (nine-point scale), . Voting intention (binary measure)

Dependent variable: . Correct answer (binary measure)

Independent variables: . Numeracy (nine-point scale), . “Conserv_repub” (aggregate Likert scale of

voting intention and political ideology)

Dependent variable: . Correct answer (binary measure)

Close

Analyses Binary logistic regression model Binary logistic regression model Same Sample Australia, random sample, selected for

one of the two chosen voting intentions USA, nationally representative random sample Different

Sample size 504 1111 Different Instrument delivery

Online poll conducted by YouGov Online poll conducted by YouGov Same

Figure 1. Contingency tables presented to the control and experimental condition groups.

ENVIRONMENTAL COMMUNICATION 189

political party with identity-affirming and identity-threatening scenarios, participants were ran- domly allocated into condition groups where the data either showed the stimulus was effective or not effective, by flipping the order of the headings in the table, as indicated by a red oval, not shown to our participants. The participants were randomly and approximately equally distributed according to their voting intention into these condition groups.

Preliminary analyses

Across all four conditions, participants found the covariance-detection problem challenging, with 48 per cent providing an incorrect answer. Figure 2 indicates that the sample was polarized according to their political identities in the two experimental conditions, with participants tending to choose the answer that conformed to their ideological attitudes, rather than the answer that is most in line with the evidence presented in their respective contingency tables. This indicates that ideologically motiv- ated reasoning, a requirement for detecting motivated numeracy, has occurred.2

Analysis

We analyzed our data using the R Statistics program (R Core Team, 2018) using binary logistic regression models, using the program’s “glm” function (RDocumentation, 2019). This is an appro- priate way to analyze data with a binary dependent variable such as our “correct” variable (McCul- lagh & Nelder, 1989; Pituch & Stevens, 2015). This statistical analysis allows inferences to be drawn about which independent variables and interactions of those variables influenced the participants’ ability to provide the correct answer in the condition groups.

Table 2 shows how we assessed the impact of the variables relevant to our hypothesis. We exam- ined the significance of any interaction effects of numeracy score, voting intention and the condition in which they were placed in through a series of hierarchical regression models. Note, however, that

Figure 2. Percentage of participants who provided the correct and incorrect answers by voting intention in each condition. Error bars represent 95 per cent confidence intervals.

190 M. S. NURSE AND W. J. GRANT

Table 2. Regression analysis of the two control conditions.

Predictors

Model one Model two Model three

Statistic Odds Ratios CI Statistic Odds Ratios CI Statistic Odds Ratios CI

(Intercept) −0.64 0.77 0.35–1.70 −2.99 0.12 ** 0.03–0.48 −2.30 0.16 * 0.03–0.76 Numeracy Score 1.58 1.10 0.98–1.25 3.44 1.54 *** 1.20–1.98 2.63 1.45 ** 1.10–1.92 One Nation −1.90 0.62 0.37–1.02 0.46 1.45 0.30–7.14 −0.22 0.77 0.08–7.53 Rash decreases condition −0.51 0.88 0.53–1.45 3.42 15.41*** 3.21–74.04 2.31 9.73* 1.41–67.16 Numeracy score * One Nation −0.86 0.89 0.68–1.16 0.01 1.00 0.67–1.51 Numeracy score * Rash decreases condition −3.74 0.60 *** 0.46–0.78 −2.39 0.65* 0.46–0.93 One Nation * Rash decreases condition −1.21 0.52 0.18–1.51 0.26 1.46 0.08–25.73 Numeracy score * One Nation * Rash decreases condition −0.76 0.81 0.47–1.40 Observations 252 252 252 Cox & Snell’s R2/Nagelkerke’s R2 0.028/0.037 0.089/0.118 0.091 / 0.121 AIC 349.7 339.5 340.9 Likelihood ratio χ2 7.2 23.4*** 24.0***

Note: *p < .05, **p < .01, ***p < .001.

EN V IRO

N M EN

TA L C O M M U N IC A TIO

N 191

an aggregate of “voting intention” and “political views” used in (Kahan et al., 2017) was not used in the final analysis as our “political views” variable was not statistically reliable (Cronbach’s α = 0.53, compared to α = 0.83 in Kahan et al. (2017) and α = 0.78 in Kahan & Peters (2017)).

Therefore, we used the categorical “voting intention” variable alone to determine the left-right political identity of the participants in this stage of the analysis, along with “numeracy score” and “condition”. We analyzed our control and experimental data separately with the control “Rash does not decrease” condition set as the reference category in the control analysis and the experimen- tal “CO2 does not decrease” condition set as the reference category in the experimental analysis.

We used a confirmatory approach to our model building strategy, consistent with our hypothesis testing (Burnham & Anderson, 2004). Table 2 and Table 3 show the outputs of the three models that were conducted with each consecutive model producing a progressively lower Akaike’s Information Criterion statistic in the analysis of the experimental conditions (Burnham& Anderson, 2004; Pituch & Stevens, 2015).3 This model has the lowest AIC and therefore is the best fit for this data (AIC = 323.8), and is statistically significant χ2 (6, N = 252) = 41.3, P < .05.

Our first model was designed to examine whether there were any main effects of the variables under investigation. In this model neither numeracy score, voting intention nor the condition they were placed in were statistically significant.

The second model was designed to examine any two-way interactions between our variables. This model showed a statistically significant interaction between voting intention and the condition they were placed in for the experimental conditions, but not for the control conditions. This provides evi- dence of motivated reasoning among our participants, where they were less likely to provide correct answers when they found the correct answer threatening to their identities. For example, One Nation supporters were only 6 per cent as likely as Greens supporters to provide the correct answer when they were placed in the “CO2 does decrease” condition odds ratio 0.06, P < .01).

The third – and most appropriate – model was designed to examine whether there was a statisti- cally significant three-way interaction between the condition the participants were placed in, their voting intention and their numeracy score. This would show evidence that numeracy was a moder- ating factor of the motivated reasoning effect observed in model two. This showed a significant (odds ratio = 0.57, P < .05) three-way interaction effect of these variables in the experimental conditions, but not in the control conditions. This suggests that numeracy exacerbated motivated reasoning.

We then conducted a simple effects analysis (Aiken & West, 1991; Hayes, 2005) of the previously observed interaction between voting intention and numeracy score in model three, using the “emmeans” R package (Lenth, 2019). This analysis of our model allows us to test our hypothesis that political polarization between the two voting intention groups would be greater at higher levels of numeracy when the participants considered the data table in the experimental conditions. If this analysis showed that the comparative odds ratios between the two groups were greater at higher levels of numeracy in the experimental condition, but not our control condition, that would provide evidence of the motivated numeracy effect. We split our data into the four separate condition groups to isolate whether the effect of the independent variable “numeracy” on our “correct” dependent variable was moderated by the independent “voting intention” variable in any of the experimental conditions but not the “Rash increases” control condition. Our results, outlined in Tables 4–6, showed that this was the case. As we saw previously in Figure 2, One Nation and Greens supporters were generally inclined to disagree about the interpretation of the data in the experimental con- ditions, but not the control “rash increases” condition. This was exacerbated by higher levels of numeracy in both of the experimental conditions.

Table 4 shows no statistically significant difference between the voting intention groups in the “rash increases” control condition at any level of numeracy, and shows similar odds ratios at each numeracy level, suggesting that numeracy had no effect on political polarization in this condition.

This was not the case in the experimental conditions. Table 5 shows, for example, that a One Nation supporter with a numeracy score of three in the identity threatening “CO2 does decrease” condition was 26 per cent as likely to respond with the correct answer (odds ratio 0.26, P < .01) compared to

192 M. S. NURSE AND W. J. GRANT

Table 3. Regression analysis of the two experimental conditions.

Predictors Model one Model two Model three

Statistic Odds Ratios CI Statistic Odds Ratios CI Statistic Odds Ratios CI

(Intercept) −0.69 0.79 0.40–1.56 −2.92 0.16 ** 0.05–0.54 −1.87 0.30 0.08–1.06 Numeracy Score 1.30 1.08 0.96–1.21 2.35 1.31* 1.05–1.65 1.14 1.15 0.90–1.47 One Nation −1.31 0.72 0.44–1.18 2.06 4.48* 1.08–18.61 0.29 1.31 0.21–8.06 CO2 does decrease condition 0.01 1.00 0.61–1.65 3.84 17.73*** 4.09–76.78 2.13 6.27* 1.16–33.98 Numeracy score * One Nation −0.76 0.90 0.70–1.17 0.89 1.18 0.82–1.71 Numeracy score * CO2 does decrease condition −2.46 0.73* 0.56–0.94 −0.65 0.90 0.65–1.24 One nation * CO2 does decrease condition −5.19 0.06*** 0.02–0.17 −0.33 0.65 0.05–8.29 Numeracy score * One Nation * CO2 does decrease condition −2.03 0.57* 0.33–0.98 Observations 252 252 252 Cox & Snell’s R2 / Nagelkerke’s R2 0.014 / 0.019 0.137 / 0.182 0.151 / 0.202 AIC 353.5 326.1 323.8 Likelihood ratio χ2 3.6 37.0*** 41.3***

Note: *p < .05, **p < .01, ***p < .001.

EN V IRO

N M EN

TA L C O M M U N IC A TIO

N 193

a Greens supporter in the same numeracy category. However, in this condition, a One Nation suppor- ter with a numeracy score of seven was only 5 per cent as likely to provide the correct answer as a Greens supporter in the same numeracy category (odds ratio 0.05, P < .01). There was no statistically significant difference between Greens supporters and One Nation supporters in the “CO2 does not decrease” condition, when participants had a numeracy score of three. However, One Nation suppor- ters with a numeracy score of seven were more than four times more likely to provide the correct answer than Greens supporters with the same level of numeracy (odds ratio 4.26, P < .05).

Our analyses lead us to reject our null hypothesis, that ideological polarization in each experimen- tal condition will be the same among participants with lower levels of numeracy and those with higher levels of numeracy. We find instead that people with higher levels of numeracy in each of our experimental conditions were more politically polarized than those with lower levels of numer- acy. This contrasts with the control “rash increases” condition where we did not observe any polar- ization at any level of numeracy.

Table 4. Simple effects of voting intention at different levels of numeracy in the control “rash increases” condition.

Contrast

Moderator levels

Numeracy score odds ratio SE z P

One Nation / Greens 0 0.77 0.9 −0.22 0.820 1 0.78 0.75 −0.26 0.790 2 0.78 0.6 −0.32 0.750 3 0.78 0.47 −0.42 0.680 4 0.78 0.36 −0.54 0.590 5 0.78 0.3 −0.65 0.520 6 0.78 0.32 −0.6 0.550 7 0.78 0.41 −0.46 0.640 8 0.79 0.54 −0.35 0.730 9 0.79 0.68 −0.27 0.780

Table 5. Simple effects of voting intention at different levels of numeracy in the experimental “CO2 does decrease” condition.

Contrast

Moderator levels

Numeracy score odds ratio SE z P

One Nation / Greens 0 0.85 0.77 −0.18 0.856 1 0.57 0.42 −0.76 0.445 2 0.38 0.22 −1.66 0.097 3 0.26 0.12 −2.96 0.003 4 0.17 0.07 −4.26 0.000 5 0.12 0.05 −4.66 0.000 6 0.08 0.05 −4.36 0.000 7 0.05 0.04 −3.96 0.000 8 0.04 0.03 −3.62 0.000 9 0.02 0.03 −3.37 0.001

Table 6. Simple effects of voting intention at different levels of numeracy in the experimental “CO2 does not decrease” condition.

Contrast

Moderator levels

Numeracy score odds ratio SE z P

One Nation / Greens 0 1.31 1.21 0.29 0.772 1 1.55 1.18 0.58 0.565 2 1.83 1.11 1 0.316 3 2.17 1.03 1.64 0.101 4 2.57 1 2.41 0.016 5 3.04 1.19 2.83 0.005 6 3.6 1.71 2.69 0.007 7 4.26 2.59 2.38 0.017 8 5.04 3.86 2.12 0.034 9 5.97 5.57 1.91 0.056

194 M. S. NURSE AND W. J. GRANT

Discussion

The current investigation aimed to progress the theory of motivated numeracy to determine whether it can be observed in participants presented with a politically polarizing ACC topic.

This study makes three contributions to the current motivated numeracy literature of interest to environmental and other science communicators. Firstly, we found evidence of the motivated numeracy effect as observed in Kahan et al. (2017) in an ACC-related topic for the first time. To our knowledge these are the first published findings to show this effect in a topic unrelated to gun control legislation in the United States. Media reporting about the motivated numeracy effect had assumed that it would apply to audiences considering the risks of climate change (Kaplan, 2013; The Economist, 2018), however this had not been empirically tested until now.

Secondly, we observed a situation in the “CO2 does decrease” condition where higher levels of numeracy didn’t just lead to higher levels of political polarization, they led to a lower likelihood of providing the correct answer when that answer did not align with the participants’ political atti- tudes (see Appendix, Figure A1 for a graphical representation of this effect). Previously published investigations into motivated numeracy showed that people with higher levels of numeracy were more likely to respond with correct interpretations of data, albeit they were much more likely to do so if that correct interpretation aligned with their attitudes than otherwise. However, these results were not mirrored in the “CO2 does not decrease” condition, which is more in line with the motiv- ated numeracy results found in Kahan et al. (2017), where higher levels of numeracy led to more political polarization, but each voting intention group were more likely to provide the correct answer at progressively higher levels of numeracy.

This may be because an unexpectedly high number of lower-numeracy Greens supporters responded with the correct answer to this question. One possible reason for this is that many lower-numeracy Greens supporters may have had enough motivation to simply ignore the data interpretation stage of the stimulus, and simply provided an answer that aligned with their identity.

In the “CO2 does decrease” condition One Nation supporters with higher levels of numeracy became gradually less likely to provide the correct answer when it didn’t align with their identities. However, this result is not mirrored in the “CO2 does not decrease” condition where highly numer- ate Greens supporters gradually became more likely to provide the correct answer even when it didn’t align with their identities. One possible explanation for this is that highly numerate One Nation supporters may be more experienced in using motivated numeracy or other mechanisms to reject data about ACC that doesn’t align with their attitudes and identities. This is a possibility that warrants further investigation.

Thirdly, this is the first study that has replicated the motivated numeracy effect in Australian par- ticipants, which may prompt more science communicators both in Australia and in other countries to consider whether motivated numeracy may affect their audiences’ interpretations of data, and to carefully consider how to explain and frame potentially identity-threatening environmental and other scientific information.

Implications

Communicators are often advised to know their audience. However, our findings suggest that environmental and other science communicators are better advised to truly understand their audience segments and the motivations that drive their understanding of scien- tific matters.

It is not often easy for communicators to understand how messages will be processed by their audiences. However, those that do not attempt to understand how identity factors may affect the interpretation of scientific information face significant risks, particularly when communicating about publicly controversial topics of science. Therefore, where appropriate, science communica- tors should conduct audience research prior to the release of scientific information to ensure any

ENVIRONMENTAL COMMUNICATION 195

risks related to identity are understood, and, if appropriate, science communicators should con- sider testing their messages and communications approaches with samples of audiences segmen- ted by identity factors to determine how the information they wish to communicate may be processed.

More generally, science communicators could consider using the power of identity-protective cognition (ICT) by helping to create societal norms based on the principle of accepting scientific evi- dence even when it challenges their attitudes. They should also consider communicating that it is acceptable and even safe for people to change identity-based groups, should the empirical justifica- tion arise. This may help to prevent situations where people may recognize the correct answer to a matter of science but feel uncomfortable about offending people who share similar identities. Why is it, for example, that it is considered socially acceptable to support a different football team to mem- bers of a person’s other social groups, but it is often not considered acceptable to have differing views about climate change, nutrition science or vaccinations?

There are also some broad political implications. As noted previously we have observed the strongest motivated numeracy effect in the literature to date, in the One Nation voting intention group. This finding lends weight to recent calls to reconsider whether public mobiliz- ation is the most effective way to encourage governments to adopt more effective public policies to mitigate ACC risks (Nisbet, 2018). If even highly numerate people remain stubbornly polar- ized about the mere facts about ACC risk, and, worse, more numerate people become progress- ively less likely to objectively interpret facts that don’t align with their political attitudes, alternate strategies should be considered. This could include presenting the benefits of ACC risk-minimizing public policies in a way that does not threaten the identities of policymakers and their supporters.

Limitations

As this study’s sample was approximately half the size of the original study, it does not have the same statistical power and therefore does not conform to Brandt’s “replication recipe” in this regard (Brandt et al., 2014, p. 220). Statistical power is important in this kind of experiment, so researchers can draw appropriate inferences, particularly about participants with the highest and lowest levels of numeracy.4

Unlike Kahan et al. (2017) the sample used for this investigation was not nationally representative and therefore the data reflects the prevalence of motivated numeracy in the chosen two voting inten- tion groups, not the prevalence among the entire Australian population.

We also note that studies such as the current investigation only measure the ability of the partici- pants to correctly interpret data, not whether this stimulus would change their behavior (such as increase their support for ACC-related policies or political candidates). We also acknowledge that numeracy may be correlated with other variables that affect the likelihood of correctly interpreting the data, such as general intelligence or cognitive reflection ability.

Areas for further research

Correctly interpreting legitimate scientific information is not the only way to form beliefs about scientific topics, and with “fake news” and other types of misinformation and disinformation increasingly being propagated through social media and websites (Vosoughi, Roy, & Aral, 2018, p. 1), it is appropriate to keep examining the motivated numeracy effect in topics such as ACC.

As this sample was chosen from supporters of two minor Australian political parties, rather than a nationally representative sample, it does not predict the prevalence of the motivated numeracy effect in the general Australian population. This may be an opportunity for further research.

196 M. S. NURSE AND W. J. GRANT

It is likely that motivated numeracy may affect people when considering a small number of severely identity-threatening topics. Future research should consider examining other such topics of decision-relevant science, such as the perceived risks relating to vaccinations, genetically modified foods, nuclear energy and nutrition science. Future research could also determine whether variables correlated with numeracy may be responsible for this effect, such as cognitive reflection ability. Further, research could also investigate whether people more versed in rejecting scientific evidence are more likely to do so.

While not relevant to our research question, sex was a statistically significant variable, with female participants less likely than males to provide the correct answer (odds ratio = 0.52, P < 0.05). While this finding was outside the scope of this investigation we note that it is similar funding to Fuller (2015) and, therefore, future studies should take care to avoid confounding their results.

Conclusion

Shortly after Kahan et al. (2017) was originally published some authors presumed political partisans would process information about ACC using motivated numeracy (Kaplan, 2013; Vivid, 2013). We have demonstrated this for the first time, providing more evidence for the identity-protective cogni- tion thesis (ICT).

By measuring 504 Australian adults’ numeracy and obtaining their voting intentions, we were able to determine whether they used their numeracy abilities or their political attitudes to interpret a table of scientific data about a deliberately politically polarizing ACC mitigation policy proposal. Our results suggest that being objectively correct can be less important to even highly numerate audiences in some circumstances than providing an incorrect answer that they wish were correct. Similar to Kahan et al. (2017) and Kahan and Peters (2017) we conclude that greater levels of numer- acy can increase the likelihood that a person will interpret scientific data aligned with their political identity.

As suggested by Van der Linden et al. (2017) these results suggest the importance of obtaining a “nuanced perspective” to understanding audience segments. There are likely to be other publicly controversial topics of decision-relevant science where different audience segments have different levels of motivated reasoning potentially leading to different levels of motivated numeracy based on the perceived threats to their identities. This again suggests there cannot be a one-size-fits-all approach to persuading each audience segment.

This investigation adds to the evidence that communications strategies designed to persuade people about ACC science should not assume that just because people can understand scientific data that this means they will accept it. Further, we have found evidence that segmenting audiences by political identity may provide opportunities for more effectively targeted communications approaches when communicating about publicly controversial topics of science. Finally, consider- ation should be given to whether an audience segment’s political identities may be so threatened that they not only process challenging information in a motivated way, but also by using motivated numeracy.

Notes

1. Note this study was first published in 2013 and re-published in 2017. 2. Note that Figure 2(a) appears to show a motivated reasoning effect among One Nation supporters in the control

“Rash decreases” condition. It is unclear why this may be the case, and may indeed be simply a statistical arte- fact. Alternatively, it may be the case that the skin care control question may not be politically neutral in some Australian populations. Therefore, we have used the “Rash increases” control condition for the rest of our ana- lyses as it most closely represents the “skin cream” control conditions found by other motivated numeracy researchers (Ballarini and Sloman, 2017; Fuller, 2015; Kahan and Peters, 2017; Kahan, et al., 2017; Sumner, et al., 2019).

ENVIRONMENTAL COMMUNICATION 197

3. We note that there is some debate about the choice of model selection criteria. Given the current investigation is considering matters of social science we do not consider the “true model” is in the candidate set, and therefore we have chosen to use the AIC method rather than other methods such as BIC. See Brandt et al. (2004)

4. Washburn and Skitka (2018) estimated that the sample size of at least 1036 is required for statistical power of at least 0.95 when attempting to replicate Kahan, et al. (2017).

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This work was supported by the Australian National Centre for the Public Awareness of Science.

ORCID

Matthew S. Nurse http://orcid.org/0000-0003-1787-5914 Will J. Grant http://orcid.org/0000-0001-9674-6488

References

Ahern, L., Connolly-Ahern, C., & Hoewe, J. (2016). Worldviews, issue knowledge, and the pollution of a local science information environment. Science Communication, 38(2), 228–250. doi:10.1177/1075547016636388

Aiken, L. S., & West, S. G. (1991).Multiple regression: Testing and interpreting interactions. Thousand Oaks, CA: Sage. The Australia Institute. (2018). Briefing note – Emission reductions and coal phase out. Canberra: Author. Ballarini, C., & Sloman, S. (2017). Reasons and the “motivated numeracy effect”. Paper presented at the 39th Annual

Meeing of the Cognitive Science Society, London, UK. Brandt, M. J., Brandt, M. J., Ijzerman, H., Dijksterhuis, A. J., & Farach, F. J. (2014). The replication recipe: What makes

for a convincing replication? Journal of Experimental Social Psychology, 50, 217–224. Burnham, K. P., & Anderson, D. R. (2004). Multimodel inference:understanding AIC and BIC in model selection.

Sociological Methods & Research, 33(2), 261–304. doi:10.1177/0049124104268644 Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in

persuasion. Journal of Personality and Social Psychology, 39(5), 752–766. doi:10.1037//0022-3514.39.5.752 Chaiken, S., & Trope, Y. (1999). Dual-process theories in social psychology. New York, NY: Guilford Press. Desaint, N., & Varbanova, M. (2013). The use and value of polling to determine public opinion on GMOs in Europe.

GM Crops & Food, 4(3), 183–194. doi:10.4161/gmcr.26776 Dickson, S. (2017). Confirmation the major parties have lost the plot on power prices – only one nation has a plan for

energy security. Lutwyche: Pauline Hanson’s One Nation Party. Drummond, C., & Fischhoff, B. (2017). Individuals with greater science literacy and education have more polarized

beliefs on controversial science topics. Proceedings of the National Academy of Sciences, 114(36), 9587–9592. doi:10.1073/pnas.1704882114

The Economist. (2018, December 8). What psychology experiments tell you about why people deny facts. The Economist. Retrieved from https://www.economist.com/united-states/2018/12/08/what-psychology-experiments- tell-you-about-why-people-deny-facts

Essential Research. (2018a). The Essential Report 5 June 2018. Carlton, Australia: Author. Essential Research. (2018b). The Essential Report 10 April 2018. Carlton, Australia: Author. Fuller, R. (2015). Controversy in the face of evidence: Investigation of personal identity protection (Masters Science

Communication Thesis). Australian National University, Canberra. Guy, S., Kashima, Y., Walker, I., & O’Neill, S. (2014). Investigating the effects of knowledge and ideology on climate

change beliefs. European Journal of Social Psychology, 44(5), 421–429. doi:10.1002/ejsp.2039 Hamilton, L. C. (2011). Education, politics and opinions about climate change evidence for interaction effects. [journal

article]. Climatic Change, 104(2), 231–242. doi:10.1007/s10584-010-9957-8 Hamilton, L. C., Hartter, J., & Saito, K. (2015). Trust in scientists on climate change and vaccines. SAGE Open, 5(3),

231–242. doi:10.1177/2158244015602752 Hayes, A. F. (2005). Statistical methods for communication science. Mahwah, NJ: Routledge. Ho, S. S., Brossard, D., & Scheufele, D. A. (2008). Effects of value predispositions, mass media use, and knowledge on

public attitudes toward embryonic stem cell research. International Journal of Public Opinion Research, 20(2), 171– 192. doi:10.1093/ijpor/edn017

198 M. S. NURSE AND W. J. GRANT

Hornsey, M. J., Harris, E. A., & Fielding, K. S. (2018). Relationships among conspiratorial beliefs, conservatism and climate scepticism across nations. Nature Climate Change 8(7), 614–620. doi:10.1038/s41558-018-0157-2

Hornsey, M. J., Harris, E. A., Bain, P. G., & Fielding, K. S. (2016). Meta-analyses of the determinants and outcomes of belief in climate change. Nature Climate Change, 6(6), 622–626. doi:10.1038/nclimate2943

Kahan, D. M. (2017). ‘Ordinary science intelligence’: A science-comprehension measure for study of risk and science communication, with notes on evolution and climate change. Journal of Risk Research, 20(8), 995–1016. doi:10. 1080/13669877.2016.1148067

Kahan, D. M., & Peters, E. (2017). Rumors of the ‘nonreplication’ of the ‘motivated numeracy effect’ are greatly exag- gerated. Yale Law & Economics Research Paper No. 584. Retrieved from https://ssrn.com/abstract = 3026941

Kahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2017). Motivated numeracy and enlightened self-government. Behavioural Public Policy, 1(1). Retrieved from https://www.cambridge.org/core/journals/behavioural-public- policy/article/motivated-numeracy-and-enlightened-selfgovernment/EC9F2410D5562EF10B7A5E2539063806

Kahan, D. M., Peters, E., Wittlin, M., & Slovic, P. (2012). The polarizing impact of science literacy and numeracy on perceived climate change risks. Nature Climate Change, 2(10), 732–735. doi:10.1038/nclimate1547

Kahneman, D. (2012). Thinking, fast and slow. London: Penguin. Kaplan, M. (2013). The most depressing discovery about the brain, ever. Retrieved from https://www.alternet.org/

media/most-depressing-discovery-about-brain-ever Kuhn, D. (2001). How do people know? Psychological Science, 12(1), 1–8. Retrieved from http://www.jstor.org.virtual.

anu.edu.au/stable/40063559 Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. doi:10.1037/0033-2909.

108.3.480 Lamberts, R. (2017). The Australian beliefs and attitudes towards science survey – 2017. Canberra: I. a. S. Department of

Industry. Retrieved from https://www.industry.gov.au/data-and-publications/the-australian-beliefs-and-attitudes- towards-science-survey-2017

Lee, T. M., Markowitz, E. M., Howe, P. D., Ko, C.-Y., & Leiserowitz, A. A. (2015). Predictors of public climate change awareness and risk perception around the world. Nature Climate Change, 5(11), 1014–1020. doi:10.1038/ nclimate2728

Lenth, R. (2019). emmeans: Estimated marginal means, aka least-squares means. R package version 1.3.3. Lewandowsky, S., Gignac, G. E., & Oberauer, K. (2013). The role of conspiracist ideation and worldviews in predicting

rejection of science. PLOS ONE, 8(10), e75637. doi:10.1371/journal.pone.0075637 Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories

on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098–2109. doi:10.1037/ 0022-3514.37.11.2098

Martin, P. (2018, January 11). A future without coal-fired power stations is inevitable. Sydney Morning Herald. Retrieved from https://www.smh.com.au/opinion/a-future-without-coalfired-power-stations-is-inevitable- 20180110-h0g47v.html

Massola, J. (2018, January 31). Big surge in opposition to Adani, new polling reveals. Brisbane Times. Retrieved from https://www.brisbanetimes.com.au/politics/federal/big-surge-in-opposition-to-adani-new-polling-reveals- 20180131-p4yz4o.html

McCright, A. M., & Dunlap, R. E. (2011). The politicization of climate change and polarization in the American pub- lic’s views of global warming, 2001-2010. The Sociological Quarterly, 52(2), 155–194. doi:10.1111/j.1533-8525.2011. 01198.x

McCright, A. M., Marquart-Pyatt, S. T., Shwom, R. L., Brechin, S. R., & Allen, S. (2016). Ideology, capitalism, and c- limate: Explaining public views about climate change in the United States. Energy Research & Social Science, 21, 180–189. doi:10.1016/j.erss.2016.08.003. Retrieved from http://www.sciencedirect.com/science/article/pii/ S2214629616301864

McCullagh, P., & Nelder, J. A. (1989). Generalized linear models. Second edition. London: Taylor & Francis. Miller, J. D., Scott, E. C., & Okamoto, S. (2006). Public acceptance of evolution. Science, 313(5788), 765–766. Retrieved

from http://www.jstor.org.virtual.anu.edu.au/stable/3846903 National Academy of Sciences. (2017). Communicating science effectively: A research agenda. Washington, DC:

T. N. A. Press. Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology,

2(2), 175–220. doi:10.1037/1089-2680.2.2.175 Nisbet, M. (2018, October 11). The IPCC Report is a wake up call for scholars, Advocates, and philanthropists.

Medium. Retrieved from https://medium.com/wealth-of-ideas/the-ipcc-report-is-a-wake-up-call-for-scholars- advocates-and-philanthropists-36415d4882f

Nisbet, M. C. (2005). The competition for worldviews: Values, information, and public support for stem cell research. International Journal of Public Opinion Research, 17(1), 90–112. doi:10.1093/ijpor/edh058

Nisbet, M. C., & Markowitz, E. M. (2014). Understanding public opinion in debates over biomedical research: Looking beyond political partisanship to focus on beliefs about science and society. PLOS ONE, 9(2), 88473. doi:10.1371/ journal.pone.0088473

ENVIRONMENTAL COMMUNICATION 199

Packham, B. (2018, 12 February). Our ‘beautiful coal’ must rival Trump’s: Canavan. The Australian. Retrieved from https://www.theaustralian.com.au/national-affairs/our-beautiful-coal-must-rival-trumps-canavan/news-story/ 08fa1b11775bb2cec8477156a817b2ae

Petty, R., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to persuasion: Theory and research. New York: Springer-Verlag.

Pew Research Center. (2015). An elaboration of AAAS scientists’ views. Washington, DC: Author. Pew Research Center. (2016). The politics of climate. Washington, DC: Author. Pituch, K. A., & Stevens, J. P. (2015). Applied multivariate statistics for the social sciences: Analyses with SAS and IBM’s

SPSS. Sixth edition. New York: Taylor & Francis. R Core Team. (2018). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical

Computing. Retrieved from https://www.R-project.org RDocumentation. (2019). GLM – fitting generalized linear models. Retrieved from https://www.rdocumentation.org/

packages/stats/versions/3.5.2/topics/glm The Senate Environment and Communications References Committee. (2017). Retirement of coal fired power stations

final report. Canberra: Commonwealth of Australia. Slezak, M. (2016, 28/11/2016). Electricity from coal should be phased out in Australia – Senate report. The Guardian

Australia. Retrieved from https://www.theguardian.com/environment/2016/nov/28/electricity-from-coal-should- be-phased-out-within-10-years-senate-report

Stanovich, K., West, R., & Toplak, M. (2013). Myside bias, rational thinking, and intelligence. Current Directions in Psychological Science, 22(4), 259–264.

Stanovich, K. E., & West, R. F. (2001). Individual differences in reasoning: Implications for the rationality debate? Behavioral and Brain Sciences, 23(5), 645–665.

Sturgis, P., & Allum, N. (2004). Science in society: Re-evaluating the deficit model of public attitudes. Public Understanding of Science, 13(1), 55–74. doi:10.1177/0963662504042690

Sumner, C., Scofield, J., Buchanan, E., Evans, M.-R., & Shearing, M. (2019). The role of personality, authoritarianism and cognition in the United Kingdom’s 2016 Referendum on European Union Membership. OSF Preprints. Retrieved from https://doi.org/10.31219/osf.io/n5r67

Swift, A. (2017, May 22). In US, belief in creationist view of humans at new low. Gallup. Retrieved from http://www. gallup.com/poll/210956/belief-creationist-view-humans-new-low.aspx?g_source = position5&g_medium = related&g_campaign = tiles

Van der Linden, S. (2016). A conceptual critique of the cultural cognition thesis. Science Communication, 38(1), 128–138. doi:10.1177/1075547015614970

Van der Linden, S., Maibach, E., Cook, J., Leiserowitz, A., Ranney, M., Lewandowsky, S.,…Weber, E. U. (2017). Culture versus cognition is a false dilemma. Nature Climate Change, 7(7), 457–457. doi:10.1038/nclimate3323

Visschers, V. H. M., Visschers, V. H. M., Shi, J., Siegrist, M., & Arvai, J. (2017). Beliefs and values explain international differences in perception of solar radiation management: Insights from a cross-country survey. Climatic Change, 142(3-4), 531–544. doi:10.1007/s10584-017-1970-8

Vivid, D. (2013). Motivated numeracy. Retrieved from http://berkeleysciencereview.com/motivated-numeracy/ Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146–1151.

doi:10.1126/science.aap9559 Washburn, A. N., & Skitka, L. J. (2018). Science denial across the political divide: Liberals and conservatives are simi-

larly motivated to deny attitude-inconsistent science. Social Psychological and Personality Science, 9(8), 972–980. doi:10.1177/1948550617731500

Weller, J. A., Dieckmann, N. F., Tusler, M., Mertz, C. K., Burns, W. J., & Peters, E. (2013). Development and testing of an abbreviated numeracy scale: A Rasch analysis approach. Journal of Behavioral Decision Making, 26(2), 198–212. doi:10.1002/bdm.1751

Whitfield, S. C., Whitfield, S. C., Rosa, E. A., Dan, A., & Dietz, T. (2009). The future of nuclear power: Value orien- tations and risk perception. Risk Analysis, 29(3), 425–437. doi:10.1111/j.1539-6924.2008.01155.x

Xue, W. H., Donald, W., Marks, A. D., Phillips, W. J., & Zhao, S. (2015). Cultural worldviews and climate change: A view from China Chinese culture worldviews. Asian Journal of Social Psychology, 19(2), 134–144. doi:10.1111/ajsp. 12116

YouGov. (2017). YouGov / Fifty Acres Survey Results. Retrieved from https://au.yougov.com/news/categories/yougov- fifty-acres-poll/

200 M. S. NURSE AND W. J. GRANT

Appendix

Figure A1. Predicted probabilities of responding with the correct answer by condition group with 95 per cent confidence intervals.

ENVIRONMENTAL COMMUNICATION 201

Copyright of Environmental Communication is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.