sindyliqian/proxies_uganda

Predicting Mortality Outcomes in Randomized Trials with Proxies: Preliminary Demonstration using Uganda Data # Predicting Mortality Outcomes in Randomized Trials with Proxies: Preliminary Demonstration using Uganda Data This project applies the estimator developed in Bastani (2019) to predict mortality outcome in a randomized controlled trial in global health. It attempts to use a proxy variable, health clinic utilization, to improve the predictive accuracy on the true outcome of interest, mortality. Note that this is still work in progress, and here we present a preliminary demonstration on data from a study in Uganda. We may apply the technique to different data sets in the future. Write-up on the motivation, background, data, method and results are in `Uganda_health_proxy.ipynb`. We use the data from Björkman Nyqvist, Walque and Svensson (2017), which can be downloaded here. We also include these files in the folder `AEJ_authors_files`. The Stata do files `1-Uganda_organizing_data_sets.do` and `2-Uganda_merging` were used to reorganize data sets provided by the authors. Data sets generated in the process are stored in the folder `Intermediate_output`, and were used to generate the data set used in this exercise, `Uganda_health.csv`.