Using Machine Learning to Generate Novel Insights: Culture, Creativity, and Innovation

What are the most important cultural values that shape individuals’ creativity and countries’ innovation? To answer this question in a theory-blind manner, we trained a deep learning model to predict 86 countries’ innovation scores from 846 attitudes, values, and beliefs measured in the World Values Survey (Study 1). When presented with new respondents to whom the model was never exposed to (i.e., an unseen test), the correlation between the model’s predicted and the actual innovation score was r = .95. Feature importance analyses revealed novel predictors of country-level innovation that have not been discussed in the literature, the most important of which was pride in the country’s long history. Two lab experiments provided causal support for the hypothesis generated by the deep learning model—making people feel more proud of their country’s long history increased their creativity. This finding held even when we controlled for general pride, indicating a unique relationship between our novel predictor and creative performance. We thus document that machine learning analyses can be used to generate novel hypotheses while bridging multiple levels of analysis.