Identifying the poor

Order of presentation

• Introduction

• Preview of main results

• Motivation of the study

• Overview of the data

• Part 1: Accounting for economies of scale

  • Method, Results (global, regional)

• Part 2: Assessing competing measures of poverty

  • Method, Results (country-level)

• Conclusion

Introduction

• The first SDG aims to eliminate poverty everywhere

  • Need to measure poverty and identify the poor accurately

• Current method: No household economies of scale

• Proposed method: Use the square-root of household size to account for economies of scale

• Research questions:

  1. How do poverty profiles change with proposed method?

  2. Which method is more likely to identify the poor?

Result #1: Limited changes in poverty

#2: Proposed method likely better

Motivation

• Current method

  • Is easy to communicate

  • Is assumed to be comparable across countries

• However, it does NOT account for economies of scale

  • Economies of scale are increasing in household size

  • Economies of scale are decreasing in private consumption (e.g. food)

• Hard to justify the comparability of per-capita method…

1. Household size varies by country

Source: Global Monitoring Database (GMD), Luxembourg Income Study (LIS),
United Nations Department of Economic and Social Affairs (DESA)

2. Household size varies by region

Source: Global Monitoring Database (GMD), Luxembourg Income Study (LIS)

3. Food share varies by country

Source: IMF Database, authors' predictions

4. Food share varies by region

Source: IMF Database, authors' predictions

5. Food share varies by income status

Notes: Consumption data are expressed in 2017 PPP dollars. All data are for 2019. 
Sources: Poverty and Inequality Platform (PIP), IMF Database, authors' predictions

6. Falling food shares due to growth

Source: ICP 2005, 2017
Note: The figures represent the shares of food, beverages and tobacco in GDP in 2005 or shares of food and nonalcholic beverages in GDP in 2017.

7. Household size evolving differently

Source: United Nations - Department of Economic and Social Affairs (UN-DESA)

Data

• Global Monitoring Database (GMD) - 153 countries

• Luxembourg Income Study (LIS) - 9 countries

• More than three-quarters of the world’s population

• Over 97% of the world’s population living in extreme poverty

• Poverty and Inequality Platform

  • CPI, PPP, population, national accounts data

• Selected GMD surveys for country-level analysis

  • Nigeria, Mali, India, Pakistan, Colombia, Tajikistan, Indonesia, Yemen

Part 1: Accounting for economies of scale How will global & regional poverty profiles change when one accounts for economies of scale?

Method

The global poverty rate in a reference year (2019) is given as:

\[ F(z) = \int_{0}^{z} f(y(x)) \,dx \qquad(1)\]

where \(f(y(.))\) is the global distribution, expressed in \(x\) (i.e. per-capita or square-root) terms and \(z\) is the poverty line.

Set square-root poverty line using three approaches:

1. Ravallion (2015): keeps poverty status of “pivot household”

2. World Bank: keeps the poverty status of countries

3. Own: keeps constant global poverty rate

Country-specific economies of scale

Household consumption, \(x_h\) is both private and public.

Individual consumption, \(x_i\) can be described as:

\[ x_i = \frac{x_h}{n^h} = pvt\frac{x_h}{n} + (1 - pvt)x_h \qquad(2)\]

where \(n\) is household size, \(pvt\) is the share of total household consumption that is private, and \(h\) is the scale parameter.

Solving for \(h\) yields:

\[ h = \frac{-ln(1 - pvt+ \frac{pvt}{n})}{ln(n)} \qquad(3)\]

Result #1: Scale parameter estimates

Notes: Scale parameter estimates are equally weighted across countries. Includes all 162 countries in the sample.

#2: Poverty lines ($2.15, $3.65, $6.85)

a. Scale parameter is 0.5 for all countries

b. Scale parameter is country-specific

#3: Poverty rates (country-level)

Note: Dotted line is a 45-degree line.

#4: Poverty rates (regional, global)

#5: Re-classification of poverty status

Note: Population figures are in millions.

Part 2: Assessing competing measures of poverty
Is the square-root method more likely to identify the poor?

Method

Goal: Examine the strength of correlation between competing poverty measures and covariates.

• Years of schooling, asset ownership, access to electricity, access to improved sanitation, etc.

\[ P_{h,c}^a = \partial_0 + \partial_1Y_{h,c} + \vartheta_{h,c} \qquad(4)\]

where \(P\) is an indicator of being poor, \(h\) subscripts the household, \(c\) subscripts the country, and \(a\) indicates the resource allocation rule (i.e. per-capita or square-root), \(Y\) is a poverty covariate.

Method (continued)

Household size correlates with poverty status and covariates.

\(E(Y|\vartheta_{i,c}) \neq 0\) \(\implies\) \(\partial_1\) will be biased.

To address the bias, estimate:

\[ P_{h,c}^a = \beta_0 + \beta_1(Y_{h,c}|N_{h,c}) + \epsilon_{h,c} \qquad(5)\]

For \(Y_{h,c}|N_{h,c}\), regress \(Y_{h,c}\) on \(N_{h,c}\) and use the residual.

Compare \(\beta_1\) between per-capita and square-root measures.

Result: Proposed method likely better

Conclusion

• The assumption of no household economies of scale is no longer tenable.

• Accounting for scale economies necessary for SDG 1

• Limited changes in regional & global poverty profiles, but substantial re-classification of the poor

• 264 million people re-classified as poor or nonpoor

• Results are nontrivial, as square-root measure seems to perform better in identifying the poor.

Result #1: Scale parameter estimates

Notes: Scale parameter estimates are population weighted across countries. Includes all 162 countries in the sample.