The World Distribution of Income: Falling Poverty and… Convergence, Period

Abstract

We estimate the world distribution of income by integrating individual income distributions for 139 countries between 1970 and 2000. Country distributions are constructed by combining two widely-used data sets: the PPP-Adjusted National Accounts data of the Penn World Tables is used to anchor the mean and Deininger and Squire (1996) and World Bank microeconomic surveys are used to pin down the dispersion.

The WDI is used to estimate poverty rates and headcounts. The CDF for 1990 stochastically dominates that of 1970, which means that poverty rates declined for all conceivable poverty lines. The 2000 CDF also stochastically dominates the 1970 distribution for all relevant levels of income. The two distributions for levels below $262 cross only because Congo/Zaire is included in the analysis, even though no good National Accounts data is available for this country for the late 1990s.

Poverty rates are reported for four poverty lines. For all lines, poverty rates in 2000 were between one-third and one-half of what they were in 1970. There were between 250 and 500 million less poor people in 2000 than in 1970. The number of people that live on less than one-dollar-a-day in 2000 was about 195 million, an order of magnitude less than the 1.2 billion widely publicized by institutions like the World Bank and the United Nations. We analyze poverty across different regions and countries. Asia is a great success, especially after 1980. Latin America reduced poverty substantially in the 1970s but progress stopped in the 1980s and 1990s. The worst performer was Africa, where poverty rates increased dramatically since 1970.

We estimate nine indexes of income inequality implied by our world distribution of income. All of them show substantial reductions in global income inequality during the 1980s and 1990s.

Finally, we argue that when in 2000, the United Nations established the Millenium Goal of halving the 1990 poverty rate, the world had already gone between 60% and 70% of the way towards achieving it.

We construct an estimate of the WDI for each year from 1970 to 2000. We do so by first estimating a distribution of income for each of 139 countries accounting for 93% percent of the world’s population in 2000. Individual country distributions are constructed using two widely used data sets. First, we use PPP-adjusted GDP per capita data from the Penn World Tables 6.1 (Heston, Summers and Aten (2002)) to anchor the mean of each country’s distribution. Second, the within-country dispersion is estimated using the income and expenditure micro surveys World Bank’s World Development Indicators which expand Deininger and Squire (1996). Since microeconomic surveys are not available annually for every country, we need to make some approximation (discussed in Section 2) to assign a level of income to each quantile for each country and year. We then use a non-parametric approach to estimate a smooth income distribution for each country/year. Finally, these individual distributions are integrated to compute the WDI.

The related literature includes Bourguignon and Morrison (2002) who attempt to estimate the WDI going back to 1820. Like Sala-i-Martin (2002), Bourguignon and Morrison (2002) estimate the WDI directly by assuming that each quintile in each country is made of individuals with identical incomes. Another drawback of Bourguignon and Morrison (2002) is that their analysis comprises only 33 countries or groups of countries and ends in 1993.

Another related paper is Bhalla (2002)⁹. Although the methodology and the data used by Bhalla differ from that of this paper, his main conclusions in terms of the evolution poverty and global income inequality are quite similar. Bhalla (2001) uses a parametric approach called the “Simple Accounting Procedure” (SAP) to approximate the Lorenz Curve for each individual country.¹⁰ As we will discuss in the next section, we use a non-parametric approach to approximate the density function.¹¹ Another difference from Bhalla (2002) is that he uses World Bank PPP data rather than the Penn World Tables data to pin down the mean of the distribution. For most countries, the choice of data set does not matter much. It does, however, for the largest country in the world: the growth rates of PPP-adjusted per capita GDP reported by the World Bank are much larger than those of the PWT.¹²

Download full text paper here