Download presentation
Presentation is loading. Please wait.
Published byBernard Simon Modified over 9 years ago
1
Peter Lanjouw, DECPI PREM Knowledge and Learning Weeks “Exploring the Intersections between Poverty and Gender” World Bank, May 8, 2012
2
The principal problem Scouring the data for insights A second problem The sensitivity of conclusions Can we estimate θ? ◦ Engel-estimates ◦ Estimates from subjective welfare models Remaining caveats
3
Conventional poverty analysis is based on a measure of household per capita consumption (or income) ◦ Household consumption aggregate is built up from multiple components Food Basic Non-food items Education (and health) expenditures Consumer durables Housing Household consumption is divided through by household size to yield per capita consumption ◦ Our best estimate of individual welfare This approach side-steps whole issue of intra- household distribution Huge and growing literature to study within- household allocation and distribution, but as yet no established procedure for capturing differential welfare levels at the individual level.
4
What if we focus on poverty of female headed households? Of widows? Rural India, 1986/7 NSS data (Dreze and Srinivasan, 1997) Family TypeIncidence of Poverty All Households63.4% Male-Headed63.8% Female-Headed57.7% Widow-Headed58.3% Extended;male-headed68.2%
5
The use of a per capita measure of consumption imposes an assumption of no economies of scale in consumption. Where might such economies come from? ◦ Consumption of public goods within the household (radio, water pump) ◦ Bulk purchase discounts on perishable food items ◦ Economies in food preparation (fuel, time)
6
Suppose money metric of consumer ’ s welfare has an elasticity of θ with respect to household size. Then welfare measure of a typical member of any household is measured in monetary terms by:
7
Suppose that ρ is the proportion of household expenditure on purely private goods, and 1- ρ is allocated to public goods. Then the correct monetary measure of per-capita welfare is: Solving for θ yields:
8
In India (in 1986/7) average household size is 5.35. ◦ If ρ =0.9 then θ=0.79 ◦ If ρ =0.7 then θ=0.50 ◦ If ρ =0.5 then θ=0.31 Are conclusions sensitive?
9
The head-count ratio and economies of scale Household Type Mean size Economies of scale parameter θ 10.80.60.4 All households 5.3563.459.654.549.5 Male- headed 5.5663.859.453.948.6 Female- headed 3.6057.761.662.062.6 Widow- headed 3.3258.363.865.166.2 Extended; male- headed 6.7868.260.351.043.5 Source: Dr è ze and Srinivasan (1997), Table 4.
11
The per capita assumption is not innocuous. Conclusions as to the relative poverty of widows versus others, or large households (many children) versus small (elderly), are usually quite sensitive. ◦ Big issue in regions (ECA) where there are big debates regarding public spending priorities (pensions versus child benefits) ◦ Note, over time, economies of scale parameters could evolve (Lanjouw, et al, 2004)
12
Econometric analysis of Engel curves with cross section data offers one entry point. Regress food share on the log of expenditure per person, including household composition as well as household size in the specification (Lanjouw and Ravallion, 1995) Lanjouw and Ravallion (1995) estimate a value for θ of around 0.6 for rural Pakistan. Lanjouw and Marra (2012) obtain an estimate of 0.68-0.69 for rural and urban Vietnam. Lokshin and Ravallion (2002) obtain an estimate of around 0.4 in Russia. Subjective welfare data provide an alternative entry point to gauge presence of economies of scale (Ravallion and Lokshin, 2002, Pradhan and Ravallion, 2000, Ravallion, 2012) Lanjouw and Marra (2012) obtain an estimate for θ of around 0.53 for Vietnam.
13
Engel-curve analysis is prey to a fundamental identification problem ◦ Problem first pointed out by Pollack and Wales (1979) Deaton and Paxson (1998) find further puzzles with this line of enquiry. ◦ Holding per capita income constant larger households report spending a smaller share of their budget on food. Ravallion (2012) points to concerns with the interpretation of the subjective welfare-based estimates of θ ◦ It is not clear that persons with different personalities respond in the same way to subjective welfare questions. ◦ Controlling for latent personality effects with panel data results in non-robust estimates of θ (Lokshin and Ravallion, 2001).
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.