This repo contains replication code for the paper "Fairness and representation in satellite-based poverty maps: Evidence of urban-rural disparities and their impacts on downstream policy" by Emily Aiken, Esther Rolf, and Joshua Blumenstock (to appear in the proceedings of IJCAI '23). The paper is available on arXiv.
This replication repo contains the code used for the paper, but does not include the data. Data can be downloaded from the following sources:
- ACS survey in the US: Download via Folktable
- Census in Mexico: Download via IPUMS
- DHS surveys: Download via the DHS website
- India SECC survey: Download via SHRUG
- Satellite featurizations from MOSAIKS: Download via siml.berkeley.edu
The code in this repo is organized as follows:
- Notebooks 1a - 1d: Cleaning survey data and extracting geometries to match to satellite imagery.
- Notebooks 2a - 2c: Sampling tiles to obtain satellite featurizations from MOSAIKS, aggregating featurization together to a single representation per geometry. Includes replication code for Table S1, Figure 2, Figure S1.
- Notebooks 3a - 3b: Machine learning simulations for predicting poverty and predicting urban/rural indicator.
- Notebook 4: Analysis of ML simulation results. Includes replication code for Tables S2 and S3 and Figures 1, 3, S3, S4, and S5.
- Notebook 5: Recalibration approaches. Includes replication code for Figures 4 and S6.
- Notebook 6: Map visualizations. Includes replication code for Figure S2.