Beyond income, which bundles of deprivation cluster together in Ethiopia — and where?
# Multidimensional poverty in Ethiopia
> Beyond income, which bundles of deprivation cluster together in Ethiopia — and where?
## The question
Income poverty is the dominant measure but not the only one — and in lower-resource settings, the income line miscounts in both directions. The Multidimensional Poverty Index (MPI), developed by OPHI and adopted by the UNDP, instead asks: across health, education, and living standards, how many *deprivations* does a household face, and how intense is that bundle? This project applies the OPHI Global MPI methodology to Ethiopia using the most recent DHS wave (2019), decomposes the result by region and by deprivation dimension, and overlays a descriptive ML feature-importance pass (Random Forest / XGBoost + SHAP) to identify which household characteristics most associate with MPI status. We are **not** comparing Ethiopia to other countries (that requires harmonized cross-country MPI — a different exercise), and we are **not** building a poverty-targeting tool (that requires program design, ethics review, and validation this study does not perform).
## Data
| Source | Granularity | Time coverage | Access |
|--------|-------------|---------------|--------|
| DHS Ethiopia 2019 | Household + member records; clusters + strata + weights | 2019 wave (most recent) | **Registered (free); license forbids redistributing raw microdata** |
| OPHI Global MPI methodology | Methodology + indicator definitions | — | Public |
| GADM 4.1 administrative boundaries | Ethiopia admin-1 (region) polygons | Current | Public; cached on first use under `data/external/` |
Real shapes loaded: **HR 8,663 × 29 cols, PR 40,659 × 9 cols, BR 23,007 × 3 cols.** Raw DHS microdata is **not committed**; the repo contains the PII-free derived analytic file (`data/processed/households_mpi.parquet`, 51 KB) and the GADM boundary cache. To reproduce, register with DHS and place the recodes in `data/raw/` per `data/raw/README.md`.
## Method
The OPHI MPI is comp …