Elsevier

World Development

Volume 28, Issue 12, December 2000, Pages 2123-2155
World Development

Poverty Comparisons Over Time and Across Countries in Africa

https://doi.org/10.1016/S0305-750X(00)00075-9Get rights and content

Abstract

We use Demographic and Health Surveys (DHS) to compare “poverty” at two or more points in time within and between African countries. Our welfare measure is an index resulting from a factor analysis of various household characteristics, durables, and household heads’ education. An advantage of this measure is that for intertemporal and intraregional comparisons, we need not rely on suspect price deflators and currency conversion factors. The wide availability and similarity of questionnaires of the DHS facilitate comparisons over both time and countries. Our results generally show declines in poverty during the previous decade, largely due to improvements in rural areas.

Introduction

The contentious debate on the effectiveness of economic and social policy in Africa over the past decade continues largely unresolved. One reason for the disparate views on the role of reform in alleviating poverty is that we remain largely ignorant about the basic question of what has happened to poverty during the last 10 years. Addressing this issue is a pre-requisite to improving our understanding of the underlying social and economic processes that have contributed to changes in economic well-being.

A new generation of nationally representative household income and expenditure surveys has helped to provide a better understanding of living standards in Africa.1 These surveys have been very useful in our analysis of the level and characteristics of poverty on the continent. They have defined welfare and the corresponding notion of poverty based on the use of consumption expenditures (including the imputed value of home consumption), generally regarded as the preferred money metric of utility.2 Much of the available household survey data that have been used to measure poverty are both recent, done within the past 10 years, and in the form of one-time cross-sections. Thus, while we have learned a great deal about poverty at a particular point in time in many African countries, the view remains a snapshot. In the vast majority of African countries, we remain unable to make intertemporal comparisons of poverty due the unavailability of data. Where survey data are available at more than one point in time, the determination of changes has proven problematic. First, survey designs change. It is now well established that differences in recall periods,3 changes in the survey instrument (e.g., the number and choice of item codes listed),4 and even the nature of interviewer training, can have large systematic effects on the measurement of household expenditures. Compounding this problem, intertemporal comparisons of money-metric welfare are only as precise as the deflators used. Consumer price indices are often suspect in Africa, due to weaknesses in data collection and related analytical procedures. Thus, relying on official CPIs is often precarious, at best.5 Alternatives such as deriving price indexes from unit values, where quantity and expenditure data are collected, also have some serious drawbacks.6

In combination, these factors have limited what we know about changes in poverty, and the reliability of the relatively few estimates that are available. This motivates our use of the Demographic and Health Surveys (DHS) as an alternative instrument for assessing changes in poverty, relying on an asset index as an alternative metric of welfare.

The DHS have been collected in a large number of African countries, and in many cases, at more than one point in time.7 The surveys were not designed for econometric (or even economic) analysis. Instead, the purpose of the surveys was to assist governments and private agencies in developing countries to better evaluate population, health and nutrition programs. Consequently, there are no data on income or expenditures, the standard money metric measures of well-being. Despite this important drawback, the DHS do contain information on household assets that can be employed to represent an alternative to a money metric utility approach to welfare measurement.8 The DHS also have two distinct advantages: they are available at two or more points in time for a large number of countries in Africa, 11 to be precise, and key survey instruments are standardized for all countries. Therefore, we can confidently compare living standards, across time periods, within a given country, and also across countries for many of our poverty measures.

In the absence of income or expenditure measures, we derive a welfare index constructed from the households' asset information available in the survey. This is the outcome of a factor analysis of various household characteristics (water source, toilet facilities, and construction materials) and durables (ownership of radio, television, refrigerator, bicycle, motorcycle and/or car) as well as education of the household head. We assume that there is a common factor, “welfare,” behind the ownership of these assets, and allow the factor analysis to define that factor as a weighted sum of the individual assets.9 One of the advantages of this measure is that for intertemporal and intraregional comparisons, we need not rely on what are often tenuous and suspect price deflators that are used to compare money metric measures of welfare.10

In this paper, we compare “poverty” as measured by our welfare index over time.11 We do this by comparing percentages of families whose welfare falls below a certain level in the asset index distribution. We also compare the distributions of our asset welfare measure at the two (or more) points in time when the DHS data were collected, using standard tests for welfare dominance Ravallion, 1991, Ravallion, 1994, Davidson and Duclos, 1998. That is, we try to identify distributions that will show less poverty regardless of the poverty line or poverty measure used. Our next approach is to decompose poverty measures regionally (as in Ravallion & Huppi, 1991). This allows us to see whether overall changes in poverty are due to changes in one or more particular regions, or movements between regions with different poverty levels. Finally, we use the asset index to make cross-country comparisons of poverty.

Before presenting our results, we discuss in some more detail the methods employed, and the data we use. We conclude with a summary of our findings.

Section snippets

Asset index

To construct an index of the household assets recorded in the DHS survey requires selecting a set of weights for each asset. That is, we want an index of the formAi=γ̂1ai1+⋯+γ̂KaiK,where Ai is the asset index for household i, the aik's are the individual assets, k, recorded in the survey, and the γ's are the weights, which we must estimate. Because neither the quantity nor the quality of all assets is collected, nor are prices available in the data, the natural welfarist choice of prices as

Data

The Demographic and Health Survey (DHS) program has conducted over 70 nationally representative household surveys in more than 50 countries since 1984. With funding from USAID, the program is implemented by Macro International Inc. For our purposes, 11 sub-Saharan African countries have cross-sectional surveys available for two or more periods.22 The DHS surveys are conducted in

Asset index weights

The weights for the asset index from the factor analysis procedure appear in Table 1. The signs are all as expected, with positive weights on all but the assets that are defined relative to left out variables that indicate greater wealth (i.e. surface drinking water, no toilet facilities and low floor quality). The magnitudes across the 12 countries are surprisingly stable. We find large positive weights placed on ownership of a television and a radio, as well as piped drinking water and flush

Conclusions

There remain widely divergent views of the impact of economic and social policy on the objective of poverty alleviation in Africa. This, in part, reflects the fact that there is great uncertainty about a relatively simple question: has poverty been declining in Africa over the past decade. Until more is known about poverty trends, it will be all but impossible to have a serious debate, and to arrive at correct lessons about the role of economic and social performance in African economies in

References (21)

  • S. Appleton

    Problems of measuring changes in poverty over time: the case of Uganda 1989–1992

    Institute of Development Studies Bulletin

    (1996)
  • Bourguignon, F., Fournier, M., & Gurgand, M. (1998). Labor incomes and labor supply in the course of Taiwan’s...
  • S. Chen et al.

    Is poverty increasing in the developing world?

    Review of Income and Wealth

    (1994)
  • Davidson, R., & Duclos, J. -Y. (1998). Statistical inference for stochastic dominance and for the measurement of...
  • Demery, L. (1995). Poverty in Ghana: Understanding the past and anticipating the future: A synthesis of the first phase...
  • Demery, L., & Mehra, K. (1996). Measuring poverty over time: Dealing with uncooperative data in Ghana. Mimeo. Poverty,...
  • Filmer, D., & Pritchett, L. (1998). Estimating wealth effects without expenditure data – or tears: An application of...
  • J. Foster et al.

    A class of decomposable poverty measures

    Econometrica

    (1984)
  • J. Foster et al.

    Poverty orderings

    Econometrica

    (1988)
  • C. Grootaert et al.

    A new regional price index for Côte d’Ivoire using data from the international comparisons project

    Journal of African Economies

    (1994)
There are more references available in the full text version of this article.

Cited by (280)

  • Consolidating and improving the assets indicator in the global Multidimensional Poverty Index

    2022, World Development
    Citation Excerpt :

    Their measurement accuracy varies: durable goods are considered easier to measure than productive assets (Chowa et al., 2010). Asset indices have been used to triangulate or stand in for monetary poverty measurements based on household consumption expenditures (Ngo & Christiaensen, 2018; Wittenberg & Leibbrandt, 2017) – especially when income or expenditure data are missing (Sahn & Stifel, 2000), have substantial measurement errors, or do not reflect permanent income (Ferguson et al., 2003; Filmer & Pritchett, 2001; Maitra, 2016). Asset indices typically aggregate a set of assets using, or partially justified by, methods such as a counting approach, principal component analysis (PCA), factor analysis, and/or multiple correspondence analysis (MCA) – although some authors also utilise anchored regression analysis,3 or so-called ‘asset scores’ that simply count the number of (weighted) items, where weights are assumed to be equal to the inverse of the proportion of households who own that item (Morris et al., 2000; see also Cappellari & Jenkins, 2007 for a related application of a ‘sum score’ to a deprivation scale construction).

View all citing articles on Scopus
View full text