World Bank. Why Not Make Up the Missing Joint Distribution Data?
This paper by Benoit Decerf, Mery Ferrando, and Balint Menyhert examines a key limitation in multidimensional poverty measurement: the lack of joint distribution data across different deprivation indicators.
In practice, many datasets only provide marginal distributions (e.g., education, health, income separately), but not how these deprivations overlap at the individual level. The paper questions whether it is valid to statistically “reconstruct” this missing joint distribution.
The core problem
Multidimensional poverty measures—such as the Alkire-Foster method—require knowing who is deprived in multiple dimensions simultaneously.
However, when joint data is missing, researchers often try to simulate or estimate correlations between dimensions. The authors argue that this is a risky shortcut because:
- Different joint distributions can produce the same marginal data
- Assumptions about correlation may be arbitrary
- Results can vary significantly depending on the method used
In short: you can’t reliably infer how deprivations overlap just from averages.
Key findings
The paper demonstrates that “making up” joint distributions can lead to:
- Biased poverty estimates
- Misidentification of who is actually poor
- Incorrect policy targeting
Even small errors in assumed correlations can generate large differences in multidimensional poverty indices.
The authors show that there is no unique solution to reconstructing joint distributions—multiple plausible versions exist, and each leads to different conclusions.
Policy implications
The main takeaway is quite blunt:
- Researchers should avoid fabricating joint distributions unless absolutely necessary
- Governments and institutions should prioritize collecting better data, especially integrated household surveys
- Transparency about assumptions is essential when estimation is unavoidable
The paper ultimately argues that data quality matters more than methodological creativity in this context.
Conclusion
This study challenges a common workaround in poverty analysis and emphasizes the limits of statistical reconstruction.
Its central message is practical and somewhat uncomfortable:
If the data doesn’t exist, you can’t safely invent it without risking misleading conclusions.
Reference
Decerf, B., Ferrando, M., & Menyhert, B. (2024). Multidimensional poverty: Why not make up the missing joint distribution data? World Bank. https://doi.org/10.1596/1813-9450-11348