Replication Data for: Bilinguals in the lab: Why Does Randomising Language Affect Cooperation?

Several experiments find that randomly assigning language to bilingual subjects causes large differences in behaviour, but these results could reflect publication bias. We replicate the largest such effect, from a public goods game in Uganda. Those playing in Luganda (the national language) contribute 28.9% more than those playing in Lugisu (the tribal language). This is within 0.1 percentage points of the original finding, providing strong evidence of a genuine effect of language on choices. We explore mechanisms by eliciting injunctive norms and empirical expectations for every possible action, finding neither differs by language, which rules out norms and conditional cooperation as explanations. Further, our setting excludes confounds including differences in grammar or comprehension. Rich anthropological evidence describes the Gisu as having a self-sufficient and non-cooperative culture. Speaking the associated language appears to activate this low-cooperation cultural frame, whereas speaking Luganda leads to higher cooperation. These results show a large and robust language effect on behaviour, which is consistent with language-dependent preferences for cooperation.