Predicting Climate Adaptation Intention and Collective Climate Action: A Supervised Machine Learning Approach

<p>Individual’s adaptation to climate hazards (e.g., create an emergency plan) and collective climate action (e.g., participate in a protest) arises from the interplay of numerous variables. Traditional linear models often fail to capture this complexity. This study employed a supervised machine learning (ML) approach to address two questions: first, how accurately ML models – incorporating a theory-transcendent set of 94 variables ranging from trust in relevant actors to climate emotions – predict climate adaptation intention and collective climate action compared to sparse, linear models based on established theories; and second, which variables are most important for these predictions. Data were collected in 2024 from 4,246 participants in the UK, Italy, Germany, and the USA using representative quota sampling. We applied models of varying complexity, including three ML algorithms, to predict climate adaptation intention and collective climate action. The ML algorithms substantially improved predictive performance beyond linear regressions grounded in established theories, particularly due to their capacity to model numerous variables. Climate emotions (particularly enthusiasm, guilt, and isolation), general coping styles and age were key predictors of both climate adaptation intention and collective climate action. Additionally, media exposure to climate change and extreme weather, along with social support for these issues, emerged as important predictors. These findings underscore the complexity of predicting climate adaptation and mitigation efforts due to the large number of interacting variables. We discuss potential leverage points for future research, such as climate emotions and the social amplification of climate change impacts through media and interpersonal channels.</p>