多维贫困：为何不补全缺失的联合分布数据？

Multidimensional Poverty Why Not Make Up the Missing Joint Distribution Data? Benoit DecerfMery FerrandoBalint Menyhert Development EconomicsDevelopment Research GroupApril 2026 A verified reproducibility package for this paper isavailable athttp://reproducibility.worldbank.org,clickherefor direct access. Policy Research Working Paper11348 Abstract Poverty is inherently multidimensional, encompassing bothmonetary and non-monetary dimensions. However, theseoutcomes are often collected in separate surveys, leaving thejoint distribution partially unobserved. To improve socialpoverty comparisons, this paper proposes a new, simplemethod to address this data constraint: assume a fixed valuefor the missing part of the joint distribution. This approachallows the integration of outcomes collected from differ-ent surveys, unlike the mainstream method currently inuse. Drawing on household surveys from six developing countries where both dimensions are observed, the papershows that the method systematically outperforms tradi-tional single-survey measures and “mash-up” measures.Monte Carlo simulations further confirm the robustness ofthe results across a wide range of data-generating scenarios.The findings highlight the value of the proposed methodfor monitoring multidimensional poverty and suggest thatit may also benefit other social indicators facing similardata limitations. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about developmentissues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry thenames of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely thoseof the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank andits affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. MultidimensionalPoverty:WhynotMakeUptheMissingJointDistributionData?∗ BenoitDecerf,†MeryFerrando,‡andBalintMenyhert§ JEL:I32,I38,C81,O15.Keywords:MultidimensionalPovertyMeasurement;SocialComparisons;DataCon-straints;ImputationMethods. 1Introduction Poverty, like well-being, is widely recognized as a multidimensional phenomenon encom-passing both monetary and non-monetary dimensions (Stiglitz et al., 2009). An individ-ual’s poverty status may thus depend on whether or not she cumulates deprivations acrossseveral dimensions. However, a common data limitation is that outcomes in different di-mensions are often collected through separate surveys.For example, at the global level,outcomes in key non-monetary dimensions such as health are typically not gathered inthe same surveys that measure monetary outcomes.1 As a result, while we can observepartial distributions for each dimension separately, the joint distribution of individual-leveloutcomes is usually missing.This lack of information on dependence across dimensionscan seriously distort poverty comparisons. Intuitively, two societies might display identicalmarginal rates of monetary and non-monetary deprivation but differ substantially in multi-dimensional poverty depending on whether these deprivations fall on the same individualsor on different groups (Decancq, 2023). The methods used to address this data constraintare therefore crucial, as they determine the extent to which poverty measures allow forappropriate social comparisons. This is particularly important because poverty measuresare used by international institutions and national governments to monitor progress andallocate resources across regions. The literature has developed three main approaches to deal with the missing jointdistribution data.2Currently, the mainstream solution is to restrict the set of dimensionsto a subset whose outcomes are all collected in a single survey.Prominent examplesinclude monetary poverty measures as well as the main global multidimensional povertymeasures, such as the Global MPI of UNDP-OPHI. These “single-survey” measures enjoycredibility because all their necessary data is observed, but they ignore the dimensions notcaptured in their survey. The second solution allows combining data from several surveys,but adopts an index whose definition does not depend on the unobserved dependence.Typically, this implies adopting a “mash-up” index that combines statistics from eachpartial distribution, an approach that is easy to compute but widely criticized for its weakconceptual foundations (Fleurbaey, 2009; Ravallion, 2011).Mash-up poverty measures,such as the Human Poverty Index (Watkins, 2006), are rarely used in practice. The thirdsolution is to estimate the missing part of the joint distribution through survey-to-surveyimputation techniques (Christiaensen et al., 2012; Dang and Lanjouw, 2023).Althoughappealing, their complexit