Using Bayesian modeling to better prioritize and access impact evaluations.
This set of files examines the scenario: imagine you have I cities in which you can implement a program, and you have priors on how well each city will perform. If you can do pilot studies to update those priors on a limited subset K of the possible cities, which cities should you perform the pilot in?
To decide which cities to pilot, we use a bayesian hierarchical model coded in R and Stan. In the simplest (1D) case, the cities are not grouped in any way, and we use a one-layer random effects model. In the more relevant (2D) case, the cities are grouped together through some predetermined underlying structure, and we use a two-layer random effects model.
The repo contains the following files/folders:
- CitiesMainCode.R files, which run the full analysis from data generation to the question "how frequently do we change our minds if we run the pilot on this city subset"
- RandomEffectsModel.stan files, which contain the underlying random effects model bayesian hierarchical code called in CitiesMainCode.R files.
- EvaluatingConvergence.R files, which help visualize the results of the Stan REM models and determine if the models are converging properly.
- explanatory_code contains R markdown files and their corresponding pdfs walking through the code and math behind some of the main files.
- single_layer_model_results contains R markdown files with graphs of results using the simpler 1D model.
- presentations contains slides for any presentations done on this work.
- old_code contains initial code written for this project but no longer used.