README.md

To replicate this analysis fully, you will need to do the following:

1. Acquire imagery and extract RCF features. Acquire imagery following Head et al. 2017 and place into data/raw/imagery/head_rep/[countryname]. When collected in this, run 1_create_rcf_features.py.

2. Acquire and process DHS labels. This can be done in two steps. First, users must get written consent from the DHS program for us to share this data. Details on how to access DHS data can be found here: https://dhsprogram.com/data/Access-Instructions.cfm. Second, once consent has been acquired, contact us with proof and we will share the data. Alternatively, you can download the data directly from DHS and process it as described in Head et al. 2017. Once acquired, place the data in the data directory under: raw/head_rep/All_DHS/. Each processed CSV file in this directory should correspond to one country-outcome pair and each row should correspond to a cluster. The files should have the following columns: id, xcoord, ycoord, cluster, name_of_outcome. id and cluster should be equivalent.

3. Create nightime lights features:

Rscript make_nl_features.R

4. Run Ridge regressions using nighttime lights, transfer-learning features (from Head et al.), and RCF features

papermill head_replication.ipynb path/to/output/notebook.ipynb