Predicting priority of environmental protection over economic growth using macroeconomic and individual-level predictors: Evidence from machine learning

Individual priority of environmental protection over economic growth is predicted by macroeconomic contexts, demographics, and psychological processes. Corresponding supportive evidence exists for each category of predictors, but limited work has synthesized them. The present study attempted to build a parsimonious model to properly predict this priority by simultaneously considering various predictors. According to prior knowledge, we selected seven macroeconomic, six demographical, and 20 psychological predictors from the World Values Survey (wave 6, 51,348 participants from 47 countries/regions). Using the machine learning approach, we first screened these predictors using a series of variable selection algorithms; then, we trained random forest models to predict the individual environment protection priority. Eight predictors were retained as the most parsimonious predictor set, and the corresponding model achieved an estimated Out-Of-Bag error rate of 34.15%. The retained predictors were Gross Domestic Product growth rate, Gross Domestic Product per capita, carbon dioxide productivity, biospheric value, post-materialistic value, sense of responsibility, self-expansion identity, and attitude toward science and technology, listed in descending order of importance. Partial dependence plots revealed that most of these predictors were non-linearly related to the outcome. These findings highlight the importance of considering macroeconomic and psychological predictors and their non-linear effects for better understanding the antecedents of environmental concern.