What do citizens associate with “climate change”? A machine learning approach to analysing open-ended survey responses in multiple languages

AbstractKnowledge about public perceptions of climate change is needed to ensure the democratic legitimacy of mitigation and adaptation policies, to facilitate their implementation, and for effective communication of policies and scientific research. Yet most survey research is limited by the use primarily of questions with fixed response options, which means that relevant perspectives and nuances may be missed. Furthermore, most studies focus on individual countries. To fill this gap in the literature, this paper presents a method to analyse short textual responses to open-ended questions in multiple languages. We use word frequency analysis and the machine learning tool BERTopic, based on sentence embeddings, to categorize the textual responses to a question about associations with climate change, asked in French, German, Norwegian, and English. Our chosen classification of responses produces topics that may usefully be labelled as Global warming, Controversy, Cars, Water, Ice, Catastrophe, Ozone, Nature, Pollution, Weather and Seasons. Analysing the predictors of topic prevalence, we find that the Cars topic, which also covers industrial emissions, is disproportionately chosen by young respondents and by residents of France and Norway. The Water topic, which includes references to both floods and sea-level rise, is chosen more by older respondents and by those residing in France and Germany. In line with earlier research, education has only minor effects on associations with climate change once other variables are taken into account. Finally, political orientation is a weak predictor of topic prevalence, suggesting a need for further research into the content of ideological divisions over climate change.