Fig. 2: Summary plots for the effects of the predictors on the four different outcomes.

A climate change belief, B policy support, C willingness to share on social media, and D pro-environmental behavior task. Each subplot depicts the predictors, ranked from highest to lowest importance for the model output. Each dot corresponds to the SHAP value of an individual prediction example. Pink color indicates higher predictor values, blue lower ones, and gray corresponds to missing values. SHAP values indicate how much the current output of the model deviates from the average model output due to a specific predictor value. A negative SHAP value (extending to the left of the vertical line) indicates a decrease in the outcome measure (or its log odds), while a positive one (extending to the right) indicates an increase in the outcome measure (or its log odds). The absolute mean SHAP value is indicated next to the predictor names on the left of each plot. Significance marked as follows: *** p < 0.001, ** p < 0.01, * p < 0.05. Significance is compared to null hypothesis data obtained from models trained and tested on shuffled data. No multiple comparison corrections were applied. Note: Gender was coded such that male = 0, female = 1; political orientation high scores = more conservative, low scores = more liberal; + climate risk index high scores indicate lower risk, and low scores higher risk. CC = climate change, ID = identity, HDI = human development index, GINI = income inequality coefficient.