We used machine learning to analyze a diverse dataset with participants from 55 countries to uncover the most important individual and nation-level predictors of climate change beliefs & behaviors. We tested 19 predictors on four outcomes: climate change belief, policy support, willingness to share climate information on social media and engagement with a pro-environmental behavior task in an international sample of 4,635 individuals.

Only four out of 19 predictors showed consistent effects for all outcomes. Two of them were environmentalist identity and trust in climate science. This suggests that safeguarding trust in science and fostering environmental identities should be high-priority goals for scientists and policymakers aiming to promote climate action across diverse contexts.

On the national level, we found that the Human Development Index (HDI) consistently predicted all climate-related outcomes. Interestingly, people from countries with lower HDI exhibited stronger climate change beliefs & behaviors. This is consistent with the precarity hypothesis suggesting that countries with lower affluence and thus not the same ability to buffer against the negative effects of climate change are more sensitive to the need for action. 

Not surprisingly, internal environmental motivation was related positively to all outcomes. However, we found a different pattern of results for external environmental motivation (e.g. “Because of today's politically correct standards, I try to appear pro-environmental.”). External motivation was the second best predictor for willingness to share information online but it had a negative effect on the effortful pro-environmental behavior. So, while external motivation may drive public actions, it can backfire for private, effortful behavior.

Importantly, the majority of the predictors showed divergent effects, predicting some but not all outcomes or even having opposite effects. This highlights the complexity of our responses to climate change. 

We also found a notable difference in explained variance per outcome: 57% for climate change beliefs, but only 10% for the effortful pro-environmental behavior. This aligns with other studies showing that pro-environmental tendencies unfortunately often don’t translate into actual behavior.

To better understand and predict climate actions, future models must integrate factors at different levels (individual and nation-level) and consider the different context (public vs private) in which behaviour occurs.

Todorova, B., Steyrl, D., Hornsey, M.J., Pearson, S., Brick, C., Lange, F., Vlasceanu, M., Van Bavel, J.J., Lamm, C., & Doell, K.C. Machine learning identifies key individual and nation-level factors predicting climate-relevant beliefs and behaviors (2025) npj Climate Action, 4, 46. doi.org/10.1038/s44168-025-00251-4