日時：7月8日（木）17:15-18:45
報告者：則友 雄磨氏（東京大学）
場所：Online Seminar in Zoom
幹事：長谷部 拓也
言語：英語
タイトル：Constrained Data and External Validity: Evidence from the Peruvian Amazon

要旨：Constraints arise for researchers when collecting data. Such constraints may result in a sample differing from the target population. This study examines data collection and tests for the external validity of constrained data. A large-scale community census covering 919 communities in the Peruvian Amazon and a systematic review of previous research were performed. It was found that past research focused on systematically different communities instead of unresearched communities in the same basins. The generalizability of empirical findings using a unique representative household survey (nearly 4000 households) was assessed. Based on estimates in previously researched communities, the average treatment effects of idiosyncratic and covariate shocks on poverty were assessed using a traditional extrapolation method. Predictions were tested against the current data gathered on unresearched communities. The results show that the predictions mostly fail, both systematically and meaningfully. To explore the mechanisms involved, researcher constraints and a simple model of site selection bias were combined, characterizing the potential concerns for consistent estimates in a target population, given the constraints. In this context, NGOs and phone coverage are relevant for the impreciseness of the predictions, though there are various associated factors. Overall, more effort is needed from various stakeholders in the data generation process.

*オンラインセミナーへ参加をご希望の方は、幹事にメールなどでご相談ください。