Machine Learning Culture: Cultural Membership Classification as an Exploratory Approach to Cross-Cultural Psychology.

Abstract

Research in cultural differences generally follow top-down, theoretical approaches. This has overrepresented theories (such as individualism-collectivism) derived mainly from Western-centric observations of cultural phenomenon. We present an alternative, exploratory approach: machine learning for classifying participants' cultural membership on international surveys. Using Wave 6 of the World Values Survey, we show that these models, paired with interpretable machine learning methods (relative variable importance and partial dependence plots), can represent magnitudes of differences between any two countries while simultaneously identifying strongly differing predictors. Analysis 1 constructs indices of cultural distance centered on USA and China, replicating previous research that used alternative methods of distance computations. Analysis 2 zooms in on USA-China, USA-Japan, and Japan-China differences, demonstrating the effectiveness of the method in both uncovering consistently known areas of cultural difference, and identifying novel dimensions for further research. Accordingly, this approach appears to be particularly effective in cultural comparisons that are traditionally overlooked.