Research
Poverty Relapse Monitoring and Governance Model
Built collection, cleaning and modelling workflows for social differentiation data, supporting classification, clustering and identification experiments.
Background and problem
Poverty relapse risk in rural western China is multi-factor and spans multiple periods. The research needed to integrate scattered social differentiation data into a computable indicator system before it could support risk identification and governance decisions.
Constraints
- Data sources are scattered with inconsistent definitions, requiring cleaning and alignment first.
- The research period is long (2024–2027), so the workflow must be reusable and transferable.
- Individual and household information is involved, so analysis and presentation must be de-identified.
Approach and architecture
The work advances through four layers — “identification → mechanism → intervention → management”: the base layer is data collection and cleaning, the middle layer is classification prediction and cluster analysis, and the top layer is the explanation and presentation of governance decisions.
Key decisions
- Indicator system first: determine which fields are obtainable, then design the model, avoiding “the model waiting for data”.
- Scripted workflow: cleaning, feature and evaluation scripts are organised in stages so experiments stay reproducible.
- Explanation first: governance scenarios need interpretable results, so models and features with better interpretability are preferred.
Results and measurement
Produced a reusable workflow for data collection, cleaning and modelling that supports classification prediction, cluster analysis and individual identification experiments, providing a data and methodology foundation for further research.
Specific indicators and conclusions are subject to the project team’s published results; unpublished data is not shown on this site.
Technical details
- Broke social differentiation data down into indicator definitions that can be collected and cleaned.
- Validated risk stratification through classification prediction, cluster analysis and individual identification experiments.
- Used SPSS/Stata for statistical testing and result verification.
Reflection
The biggest cost in a multi-year project is drift in data definitions. If I did it again, I would lock down the data dictionary and version management in the first year to reduce alignment work later.
Scope and security
This is a static case study. The page does not run project code or collect uploads, logins or visitor data. Metrics come from project records and local experiments; reproduce details from the source or report. Read the site security model →