The Five Canadas of Climate Change: Using audience segmentation to inform communication on climate policy

consistent Akaike information criterion (CAIC)–where lower values indicate superior fit–did not converge on a single solution, which is often the case in LCA [69]. We thus performed bootstrapped likelihood ratio tests (BLRT) which provided p-values to assess whether moving from 4- to 5- and 6-segments lead to a statistically significant improvement in model fit. P-val- ues for both the 5- and 6-class solutions were significant at p<0.001. Finally, we calculated the Bayes factor (BF) to compare the 5- and 6-class solutions. A BF of greater than 10 provides strong support for the model with fewer classes [70], pointing in our case to the selection of the 5-class model. As is recommended when fit indices provide reasonable support for one or two candidate models [69], we also looked at how the models compare to each other in terms of face validity. From this point of view, the 5-class model clearly offered the most informative and practical results. On the one hand, the four-segment model omitted the distinction between the Doubtful and the Dismissive groups, which masks some nuance, as results from our five-segment model (described below) make evident. On the other hand, the additional group created in the six-segment model is very similar to the Concerned segment we already have in the five-segment model and thus generated confusion while not substantially contrib- uting to a better understanding of how motivations, behaviours and preferred societal responses differ across groups. Survey experiment

We embedded a vignette experiment in the survey, using a 2 X 4 factorial design, in which the first factor manipulated the price level of the policy, while the second manipulated different options for revenue use and messaging. Both levels in this experiment represent emphasis framing. The first factor had two levels–informing respondents that a carbon tax would result in 2 cents per litre (low cost) and 11 cents per litre (high cost) increase in the price at the gaso- line pump. These prices correspond to the equivalent price per litre of gasoline resulting from a $10 carbon tax, and a $50 carbon tax, respectively, which align with the price schedule out- lined by the Pan Canadian Framework on Clean Growth and Climate Change proposed by the Trudeau government in Canada at the time the survey was conducted. The second factor involved testing a number of potential options for revenue use and messaging around such policy design, as outlined in Table 3. Our dependent variable is an additive index of three questions assessing respondents’ over- all attitudes toward carbon taxes. The first and second items constituting the index include measures of the perceived fairness and efficacy of the experimentally manipulated carbon tax description, measured with the question: “On a scale from 0 to 10, where 0 represents very unfair and 10 very fair, how fair do you think this federal carbon price is?” and “On a scale from 0 to 10 where 0 represents very ineffective and 10 very effective, how effective or ineffec-

tive do you think this federal carbon price will be at reducing emissions?” The third item mea- sured people’s willingness to pay the carbon tax: “All things considered, would you be very willing, fairly willing, not very willing, or not at all willing to pay higher taxes in order to Table 2. Model fit statistics. Model LL BIC(LL) SABIC CAIC BLRT p value BF 4 classes -15721 32939 32268 33149 - >15.000 5 classes -15570 33014 32175 33277 0,000 >15.000 6 classes -15482 33212 32205 33529 0,000 - Note: LL = log-likelihood; BIC = Bayesian information criterion; SABIC = sample-size adjusted BIC; CAIC = Consistent Akaike information criterion; BLRT = bootstrapped likelihood ratio test; BF = Bayes factor. Bolded values indicate “best” fit for each statistic. https://doi.org/10.1371/journal.pone.0273977.t002 PLOS ONE The Five Canadas of Climate Change PLOS ONE | https://doi.org/10.1371/journal.pone.0273977 November 23, 2022 7 / 21