PPPs for the Measurement of Global and Regional Poverty:

Issues and Options

D.S. Prasada Rao

School of Economics

University of Queensland

Brisbane, Australia

July 2003

This is a prepared for the Development Economics Data Group of the World Bank. The empirical analysis section is based on the work undertaken jointly by D.S. Prasada Rao and Chris O’Donnell of the School of Economics, University of Queensland. The authors wish to thank Ms. Jinsook Lee of DECDG of the World Bank for her efforts in securing unit record data from LSMS/HES, and Mr. Yonas Biru who has always been available for consultation on matters relating to this work.


I. General Background

The current practice of producing global poverty estimates makes use of $1/day and $2/day poverty lines. The use of these poverty lines is justified on the grounds that they closely correspond to the poverty lines used in some of the poorest countries, thus their application is likely to produce a conservative estimate of poverty incidence in the world. The approach also recognises the diversity of methodologies for setting poverty lines practiced in different countries. Using a single poverty line makes it possible to assess poverty incidence across countries and regions that is free from the influence of different country practices in setting poverty lines.

A crucial step in the process of compiling poverty estimates is the conversion of the $/day poverty line into respective national currency units. This conversion is made using PPPs from the International Comparison Program (ICP) for the Private Consumption Expenditure aggregate for the benchmark years. The Penn World Table PPPs for Consumption are also used for this purpose for the non-benchmark years. The national currency equivalent $/day poverty lines are used in conjunction with income distribution information to arrive at global and regional estimates of poverty incidence.

Over the last few years, the use of PPP data in the derivation of global and regional poverty estimates have attracted considerable attention and various limitations of the current approach have been identified. Deaton (2000) and Reddy and Pogge (2002) provide a comprehensive summary of some of the relevant issues.

Some of the principal issues are listed below.

· The ICP-PPPs are based on prices of commodities that are not representative of the consumption baskets of the poor.

· The ICP PPPs weight commodity-specific relative prices with weights that do not adequately represent the consumption patterns of the poor.

· The aggregation methodology used does not offer a direct comparison of a fixed basket of goods and services consumed;

· The PPPs used are not consistent, in their temporal movements, between benchmarks;

· The international poverty lines, $1/day or $2/day are not directly related to any conception of poverty and therefore lack an analytical framework.

· The process of updating $1/day poverty lines over time is not clear.

It may be stressed here that the World Bank explicitly recognises some of these issues. Ravallion (2000 and 2002), in his replies to Deaton and Reddy and Pogge, defends the use of $1/day poverty lines and emphasises that “the vast bulk of Bank’s analytic and operational work on poverty, the ‘$/day” line is ignored, and with good reason. When one works on poverty in a given country, or region, one naturally tries to use a definition of poverty appropriate to that setting. Most of the time, the Bank’s poverty analysts don’t need to know what the local poverty line is worth in international currency at purchasing power parity…..” (Ravallion, 2002). The global and regional poverty estimates reported can be used only as a general guide to levels and broad trends in poverty, but cannot be considered as substitutes for the country-specific poverty estimates derived at national levels.

The current round of the International Comparison Program, scheduled for 2004, presents an excellent opportunity to examine the possibility of compiling poverty PPPs. Given the importance attached to global and regional poverty estimates within the context of the Millennium Development Goals (MDGs) with respect to poverty reduction, it is therefore necessary to integrate the work of poverty-PPP compilation with the regular ICP activities at the World Bank as well as at the national levels.

II. Poverty Specific PPPs: Basic Elements

An important starting point in the construction of poverty specific PPPs is to recognise that the tasks of setting international poverty lines and the construction of poverty-specific PPPs are two distinct components. The choice of the international poverty line is a task best undertaken by the Poverty Section of the World Bank’s Development Research Group, and should, therefore, be considered as being outside the domain of the current efforts to construct poverty-PPP. However, compiling poverty PPPs will involve a detailed examination of the methods and practices involved in setting poverty lines at the country level. Given the complex nature of the tasks involved, it is necessary to develop a close collaborative approach between the DECDG and the Development Research Group. Close consultation and a strong collaboration between those who coordinate the ICP and those who manage the World Bank poverty measurement and monitoring work are essential for successful completion of this work.

Several basic ingredients are necessary to construct PPPs for poverty measurement.

1. The first and foremost component is the price data. Since PPPs are essentially spatial price index numbers (for comparisons across countries), we need to collect data on prices “paid by the poor”. Several issues arise here. It is necessary to identify the poor in each country before price data are collected. This poses a problem of circularity in the tasks to be conducted. This issue can be resolved by using either country-specific rates of poverty incidence or rates of incidence implicit in previous World Bank publications. These could be considered as a reasonable starting point for the exercise. The second issue is to identify the goods and services that are typical of the consumption of the poor. The task of identifying “food” items may be easier than listing “non-food” items. In addition to identifying the list of items, it is also necessary to identify the type of outlets where poor make their purchases as well as the typical sizes of their purchase. It is also necessary to take into account possible regional and rural-urban variations in the list of consumption items.

2. The second element needed in PPP computation is data on expenditure shares that reflect the consumption patterns of the poor. It is expected that consumption pattern of the poor differs significantly from that of the average consumption pattern observed in a country. Again, there is need to identify the poor in order to be able to study their expenditure patterns.

3. The price and weights data from different countries need to be aggregated using a suitable index number methodology that can result in a set of PPPs.

The ultimate objective is to develop a framework to integrate the task of producing poverty-specific PPPs within the framework of ICP. The following are crucial steps in achieving this objective.

1. Examine the nature and extent of price data necessary for the construction of poverty-specific PPPs.

2. Identify the existing sources of price information within the institutional framework and price surveys conducted by the national statistical organisations.

3. Establish a survey framework for any additional price data required to supplement existing price data.

4. Propose a suitable aggregation methodology using the existing ICP practices and examine the sources of data on weights to be used in the aggregation process.

5. Investigate the problem of updating the international poverty line for temporal comparisons of global poverty estimates. The parallel problem of extrapolating poverty-PPPs for non-benchmark years will also be considered as a part of the ongoing poverty-PPP work.

III. A Pilot Study

A pilot study involving a total of eight countries, covering Cote d’Ivoire, Egypt, Ethiopia, Ghana, India, Indonesia, the Philippines and Thailand, with the following aims was conducted.

The starting point for the work was to examine: (i) country-specific practices in establishing poverty lines with the aim of identifying a basket of goods and services that may be truly termed a “poverty-basket”; (ii) if national statistical organisations regularly collect data on prices paid by the poorer sections of the population; (iii) the household expenditure survey data available in different countries to check if such data could be used in measuring prices paid by the poorer sections; and, finally, (iv) to extract expenditure shares, for commodity groups, that can be used as weights in aggregating price data.

A brief summary of the general findings is provided below.

· There is a common thread in the methodology used in determining poverty lines in these countries. All poverty lines consist of food and non-food components, with the food component essentially being based on a specific energy requirement. Household expenditure surveys are the main source of data for this purpose. However, there are subtle differences in actual translation of caloric needs into monetary values. Non-food poverty line determination varies a great deal. None of these countries actually use or identify a consumption basket associated with the “poor”. Spatial and temporal adjustments of poverty lines are made using Laspeyres-type price index numbers.

· All these countries rely on household expenditure (HES) or living standards measurement surveys (LSMS) as the principal source of data for poverty line determination and subsequent estimation of poverty incidence. Close examination of the survey questionnaires shows that these surveys collect fairly detailed information on total expenditure and quantity for various food and non-food items. The non-food items generally appear in a fairly aggregated form. These surveys often include non-market consumption (from own production and in-kind transactions) and an imputed value. These surveys provide useful unit value information that can be used to supplement price information collected from other sources.

· The consumer price index compilation in these countries was closely examined. Special focus was placed on price indices for lower income groups and on spatial price comparisons within countries. There are a few countries where consumer price indices for low-income groups (eg the bottom 30% of the population and rural agricultural labourers) are regularly compiled and published. Another interesting observation emerging from discussions with officials at the national statistical offices is that very useful information on item and outlet specifications is collected as a part of CPI compilation, but which is not electronically compiled or disseminated.

IV. Price Level Comparisons using Household Expenditure Survey Data

As a part of this exploratory phase, econometric analysis of household expenditure survey data is being conducted. The aim is to derive unit values or average commodity prices and use them in measuring differences in price levels. The following are the principal aims of the empirical analysis.

1. Measuring price level differences across income groups. As a first step, average prices and quantities are derived for different income groups (eg decile groups) from the household expenditure surveys. Various binary and multilateral index number methods such as the unweighted and weighted Elteto-Koves-Szulc (EKS) and the expenditure share weighted Geary-Khamis (GK) are being used in deriving consistent price index numbers.

2. Measuring price level differences across income groups using unit record data at the household level. In order to be able to make use of the very detailed data available from the household expenditure surveys, it is necessary to apply regression methods. The unweighted country-product-dummy (CPD) and weighted CPD methods are being used to handle household level data.

If price level differences across income groups can be measured unambiguously, it would be possible to adjust the current ICP-PPPs to derive more meaningful poverty-specific PPPs. Lack of quality information attached to consumption data in the household expenditure surveys is a major problem. If the higher income groups are shown to pay higher prices, this may be partly attributable to the higher quality of products included in their consumption.

3. Construction of PPPs for cross-country comparisons. Conceptually, it is possible to make use of the share weighted-GK, EKS or weighted CPD methods to combine data from the four countries to compute PPPs for price comparisons for different income groups. If the bottom three deciles (30%) are considered poor, PPPs associated with this bottom group may be considered more appropriate for poverty related work. Given the differences in product specification and detail in different household expenditure surveys, it may be necessary to apply this approach at a slightly more aggregated level.

4. Spatial Price Comparisons. Given unit record data, with details on the location of households by regions within each country, an attempt is being made to construct spatial/regional price index numbers for different income groups.

The pilot project, within its broad objectives, also examines the sensitivity of the derived PPPs to the aggregation methods used. The project considers a number of new aggregation methods such as the weighted EKS, share-weighted GK and weighted CPD, along with some of the standard methods. A brief description of these aggregation methods is provided in the following section.

IV.a. Aggregation Methods used in the study

The study made use of several binary index number formulae, such as the Paasche, Laspeyres, Fisher and Tornqvist, in deriving price index numbers for computing PPPs for income groups. In addition a number of multilateral index number methods including the Elteto-Koves-Szulc (EKS) and weighted EKS; the country-product-dummy (CPD) methods and number of variants of the methods; and the expenditure-share weighted Geary-Khamis method have also been employed. These aggregation methods are briefly described below.

IV.a.1. Binary Methods

Let pij and qij denote the price and quantity of product i in income group j. The following are the four commonly used binary index number formulae.

Paasche: Pjk =

Laspeyres: Ljk =

Fisher: Fjk =

Tornqvist: Tjk = exp[0.5(vij + vik)log(pij/pik)]

where vij = are value shares.

IV.a.2. Multilateral Methods

The current analysis of household expenditure survey data within the pilot project employs a range of new or modified aggregation methods. This section describes these methods and their properties. Further details of these methods can be found in the papers cited.

Method 1: Weighted EKS Method

The computational form for the Elteto-Koves-Szulc (EKS) index is given by

(1)

where Fjk denotes the Fisher price index number for country k with country j as the base. A simple interpretation of the EKS index is that any binary comparison between countries j and k is an unweighted averaged of all the linked comparisons between j and k using links l = 1, 2, …, M. Rao (2001) showed that the standard EKS is essentially based on the least squares estimation of the parameters of the simple regression equation: