Conference Paper
Poverty Estimation in Data-Poor Countries: Parametric Approximations Based on Limited Statistics
No.
ERF_31_PO227
Publisher
ERF
Date
March, 2025
Topic
I3. Welfare, Well-Being, and Poverty
Technical issues with income datasets and heterogeneous statistical approaches for addressing them give rise to discrepancies in poverty estimates across different studies. We assess how alternative parametric modeling approaches perform under various modes of data granularity. We use a large worldwide set of household income surveys – including notably conflict-affected and high-income countries in the MENA region – on which we artificially impose one of four alternative degrees of data granularity: individual-level microdata, random extraction from the microdata, grouped data, and a pair of basic distributional statistics. We then correct for the data limitations, and estimate poverty headcount ratio and poverty gap, using several parametric distribution functions advanced in prior studies. We find that, when only basic distributional statistics are available, lognormal and Fisk distributions demonstrate a similar, moderate degree of estimation accuracy, compared to other functional forms. With grouped income data, three- and four-parameter models, particularly the beta and generalized beta functions, perform well. With microdata, three- and four-parameter models again outperform two-parameter models. These findings underscore the accurate fit of four-parameter models in various data environments, particularly compared to two-parameter alternatives. The findings also highlight the challenges in modeling the bottom of income distributions when only basic distributional statistics are available.
Speakers
Hassan Hamie
Economist, Poverty and Inequality Research Team, UN...
Authors
Jinane Jouni
Researcher and Data Scientist, American University of...
Authors
Vladimir Hlasny
Economic Affairs Officer, UN Economic and Social...