Multilevel predictors of climate change beliefs in Africa

models: they predict responses for each level of the predictor while holding constant the rest of the variables. Despite its advantages, the random forest algorithm has some limitations. For instance, the stochastic processes that provide high flexibility and predictive accuracy can also introduce some variability in the output. Also, this output is somewhat more obscure than the output from other algorithms such as conditional inference trees. The code and data used for this analysis is available online. Data Dependent variables. Data for our dependent variables are largely drawn from the Afro- barometer, which uses a clustered, stratified (by administrative units and by urban or rural location), multi-stage and area probability sampling design [5]. We particularly use data from the seventh round of surveys (R7), conducted between 2016 and 2018 and was the first time the Afrobarometer included climate change-related questions. The dataset is comprised of more than 45,000 observations from 33 African countries. This data is georeferenced at sec- ond-order political boundaries (see S1 File for a detailed variable speciation). This allows us to overlay our climate variables to the same areas where the Afrobarometer surveys were conducted. We include five different dichotomous specifications for climate change beliefs. Our first specification, climate change awareness, measures whether survey respondents have heard about the existence or climate change or not. Second, we account for the respon-

dents’ belief that human activity is the main cause behind the current changes befalling the world’s climate. Third, we are interested on whether respondents developed a risk percep- tion from a changing climate that is making life worse in their country. Our final two speci- fication examine respondent’s beliefs on whether climate change needs to be stopped and whether ordinary citizens can contribute to mitigation efforts—a variable known in the lit- erature as self-efficacy. Independent variables. We include 67 independent variables from the Afrobarometer. We include 7 variables to account for access to information, 33 for political ideology and par- ticipation, 4 for religion, 10 for economic conditions, 9 for demographics, and 4 for percep- tions about the impact of climate change on agriculture. All variables are grounded on previous findings within the literature on climate change beliefs. Descriptive statistics, their corresponding Afrobarometer questions and coding scheme are included in the S1 File. Additionally, we include four climate variables to assess the impact of local climate change on CCBs. Our first three variables are precipitation anomalies, and maximum and mean tem- perature anomalies. Data is from the Climate Research Unit (CRU TS) 4.0 dataset [56]. From this dataset we first obtain the raw 0.5˚ x 0.5˚ gridded data of precipitation, maximum temper- ature, and mean temperature, for each month. We then overlay the 0.5˚ x 0.5˚ grid over the second-level political boundaries where the Afrobarometer survey was conducted. We keep

the mean values of the grids each administrative unit intersects with, weighted by the intersec- tion area. Once we have monthly precipitation, maximum and mean temperature data for each administrative unit, we calculate their anomalies as the monthly deviation from the long- term (1961–1990) mean for that month, divided by the panel’s standard deviation [57]. We then annualize these anomalies by averaging the monthly anomalies of the 12 months before each respondent was surveyed. In the end, we obtain three main climate variables at second- order political boundaries: long-term anomalies of precipitation, maximum and mean temper- atures for the year before the survey was conducted. These variables capture how precipitation and temperature anomalies differ in each administrative region (compared to the 1961–1990 baseline) during the previous 12 months to the survey being conducted. PLOS ONE Multilevel predictors of climate change beliefs in Africa PLOS ONE | https://doi.org/10.1371/journal.pone.0266387 April 5, 2022 5 / 14