We obtain rich measurements of risk preferences for 2939 subjects across 30 countries, and use the data to paint a picture of the distribution of risk preferences across the globe using structural equation models. Reference-dependence and likelihood-dependence are found to be important...