Bayesian climate shock-to-action targeting analysis using SIHBS 2022 to identify high-risk unreached households and compare humanitarian targeting strategies.
# Who Would We Miss?
## A Bayesian Climate Shock-to-Action Targeting Analysis Using SIHBS 2022
This repository contains a reproducible mini-project using Somalia Integrated Household Budget Survey (SIHBS) 2022 microdata to ask a decision-facing question:
> If emergency support can reach only a limited share of households, which targeting rule misses the fewest high-risk unreached households?
The analysis defines **high-risk unreached households** as households that:
1. experienced a climate/livelihood shock,
2. experienced moderate/severe food insecurity or erosive coping, and
3. reported no formal support from government, local NGO, or international organization.
The model is a **Bayesian hierarchical logistic regression** with region-level random intercepts, fitted with variational Bayes.
## Headline findings
- Analysis sample: **7,212 households** across **17 regions**.
- Weighted climate/livelihood shock exposure: **71.9%**.
- Weighted moderate/severe food insecurity proxy: **44.1%**.
- Weighted formal support after shock: **3.8%**.
- Weighted high-risk unreached outcome: **40.1%**.
- At **20% support coverage**, Bayesian risk targeting reached **34.0%** of high-risk unreached households, compared with **27.8%** under climate/livelihood shock-only targeting and **19.5%** under poverty-only targeting.
## Repository contents
```text
scripts/ Reproducible analysis scripts
figures/ LinkedIn/GitHub-ready figures
outputs/ Aggregated public-safe results
data/metadata/ SIHBS module inventory
data/processed/ Schema only; no household microdata included
reports/ Mini-project report
docs/ Methodology and LinkedIn post drafts
```
## Data note
Raw SIHBS microdata are **not included** in this public repository. To reproduce the analysis, place the official `.dta` files in `data/raw/` with the original filenames, then run: …