Replication Data for: Do International Students Learn Foreign Preferences? The Interplay of Language, Identity and Assimilation

This contains the do file and dataset to replicate all output for Paul Clist & Ying-yi Hong's Do International Students Learn Foreign Preferences? The Interplay of Language, Identity and Assimilation, forthcoming at the Journal of Economic Psychology. It also contains the experimental script.
Abstract: Every year millions of students study at foreign universities, swapping one set of cultural surroundings for another. This may reveal whether measured preferences are fixed or flexible, whether they can be altered in the short-run by moving country, or learning a new language. We disentangle these influences by measuring international students' preferences. For Chinese students in the UK (who arrived up to five years previously) we randomise a survey's language. We add reference groups in each country, doing the survey in the relevant language. Simple comparisons provide a causal estimate of language's effect, and observational estimates of differences by country, location and nationality. We find language has a large causal effect on a range of survey responses. The effect size is similar to differences by country or nationality (at 0.4 standard deviations), and larger than differences by location (at 0.1 standard deviations). Assimilation theories predict any movement in measured preferences for Chinese students in the UK would be towards those of UK students, even if they may be small. We do not find this. In Mandarin, Chinese students hardly differ from those in Beijing. Yet in English, they are not close to either Chinese students in Beijing nor British students in the UK. This can be explained by an identity-priming model with monocultural subjects. For Chinese students in the UK, speaking English reduces the pull of a Chinese frame without increasing the pull of a British one. International students do not so much learn foreign preferences as learn to ignore old ones. Our reliance on mostly stated preferences enables a rich dataset covering many domains; future work is needed to see if such large effects are also found across a wide range of revealed preferences.