Multidimensional poverty measurement with individual preferences Koen Decancq – Marc Fleurbaey – François Maniquet UNDP – March 2014
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Poverty is multidimensional Who is poor?
1. Motivation Multidimensional poverty measurement without paternalism? Let agents aggregate the dimensions themselves “... those with a stake in the outcomes will almost certainly be in a better position to determine what weights to apply than the analyst calibrating a measure of poverty.” (Ravallion, 2011) Acknowledge the heterogeneity in the “opinions on the good life”
2. Multidimensional poverty measure We axiomatically derive the following procedure
2. Multidimensional poverty measure We axiomatically derive the following procedure
2. Multidimensional poverty measure We axiomatically derive the following procedure … and apply it to real-world data (from Russia)
3. Estimating preferences We use RLMS-HSE ( ) We consider four dimensions of life –Equivalized expenditures –Objective (constructed) health index –Constructed house quality index –Unemploment (binary) Deprivation thresholds: 60% of median value in each continuous dimension
3. Estimating preferences Problem: we don’t observe “opinions on the good life” We estimate them based on life satisfaction data We run a simple life satisfaction regression, with some econometric sophistications, –Heterogeneity in β coefficients –Decreasing marginal returns –Control for personality traits (in α) And then plot indifference maps based on β’s
3. Estimating preferences
4. Results: headcounts
4. Results: overlap of bottom 16,1 % 2,9% 3,5% 1,6% 3,6%4,1% 2,4%
5. Conclusion Multidimensional poverty analysis with respect for preferences … … is ethically attractive … is theoretically possible … is empirically implementable
Life satisfaction regression