Box 1. Poverty Lines and Measures

Measuring poverty begins with some form of threshold, typically called the poverty line. Units of observation (usually households) falling below the threshold are considered poor; those at the threshold or above it are nonpoor. The poverty line can be defined in many ways, and mostly in terms of some monetary value, such as 50 percent of median income or 65 percent of minimum wage. These result in different counts of the poor. To reduce the arbitrariness, some poverty lines are defined as the monetary value at which certain nutritionally adequate diet requirements are met. Others add a nonfood component by direct estimation, or indirectly from the food ratio. Poverty lines are sometimes further adjusted to account for variations in family size (the most common being the per capita adjustment) or age of members. As geographical locations and regional price differences contribute to variations in the estimation, region-specific poverty lines have been constructed.

In practice, the living standard indicator against which the poverty line is compared is either income or consumption expenditure collected from household income and/or expenditure surveys. This type of survey collects data describing the consumption patterns, incomes, some demographic features, and sometimes housing characteristics of households or families. An exercise estimating the headcount index without a household survey, using instead only readily available aggregate economic and social indicators, yielded at best only a rough idea of the prevalence of poverty, and nothing of its extent (Ravallion 1992).

Concerning poverty measures, the class proposed by Foster, Greer, and Thorbecke (1984) is currently the most popular, and with reasons. To this class belongs the simplest measure, the headcount ratio (P0), which measures the prevalence of poverty. Other measures are the poverty-gap index (P1) related to the depth of poverty, and (P2), the severity of poverty.

In addition to measuring poverty directly using household income or expenditure data, efforts have been made to measure the poverty level of a country in conjunction with a set of social indicators. A number of indicators have been proposed as included in the minimum basic needs indicators (MBN), UN minimum national social data set (MNSDS), and OECD's Strategy 21 indicators. The MNSDS, which was proposed in 1996, is a list of 15 indicators to form a suggested minimal set of social indicators for a particular country. It covers indicators such as life expectancy, mortality (infant, child, and maternal), percentage of underweight infants, average number of years of schooling, gross domestic product (GDP) per capita, household income per capita, unemployment rate, and access to safe water and sanitation. In 1990, the United Nations Development Programme (UNDP) introduced a composite index called the human development index (HDI) (UNDP 1990). HDI, which is an improvement over the physical quality of life index (PQLI), is based on three indicators: longevity, as measured by life expectancy at birth; educational attainment, through a combination of adult literacy and gross enrollment ratio; and standard of living, as measured by real GDP per capita expressed in purchasing power parity (PPP) in constant dollars. UNDP later developed the Human Poverty Index (HPI) (UNDP 1997). This index deals with deprivations in three dimensions of human life as reflected in the HDI: longevity, knowledge, and a decent standard of living. The deprivation in longevity is approximated by the percentage of people expected to die before reaching age four and the deprivation in knowledge, by the percentage of illiterate adults. The level of a decent standard of living is determined through a combination of three variables: the percentage of people without access to health services, percentage without access to safe water, and percentage of underweight children under five years. While the HDI measures progress in a community or country, the HPI indicates the extent of deprivation (UNDP 1998). However, it is to be noted that neither HDI nor HPI could be associated with a proportion, number, or segment of the population in poverty or deprivation.

For most of the DMCs, estimating poverty is a recent phenomenon (Table 2). Poverty has become almost a nonissue in the newly industrialized economies and explains the absence of estimates (except in Taipei,China where the poverty incidence is estimated at 1 percent). The data situation has improved significantly between 1985 and 1998. Targets of priority assistance for developing poverty data sources are countries or subregions where data are still not available, including Afghanistan, Bhutan, Myanmar, Tajikistan, and Uzbekistan in Asia and all the Pacific DMCs except The Republic of Fiji Islands and Papua New Guinea.

Consistency ensures a meaningful comparison of estimates of poverty within a country over time. Inconsistency is the result of various factors, including changes in the instruments for data collection, reference period, poverty variable used, unit of analysis, and estimation procedure for deriving poverty lines.

Most DMCs use a kind of household survey as a source of basic data to derive the poverty incidence estimates. Examples are the suvey sosial ekonomi nasional (SUSENAS or socioeconomic survey) in Indonesia and the family income and expenditure survey (FIES) in the Philippines that are conducted at three-year intervals. The use of one type of survey tends to reduce variation between measures. However, changes in the contents (e.g., including details) or manner of collection (e.g., use of the vernacular) from one survey to the next to enhance responses do occur and, in the name of improvement, are regarded as permissible. The common criticism in using a household survey is income underreporting and/or expenditure overreporting that is determined through cross validation with national income accounting. Moreover, because this is a household survey, it leaves out a growing segment of society typically associated with poverty: the homeless. It also leaves out the institutional population that includes those in hostels, nursing homes, military barracks, and other nonhousehold institutions.

The quality of data is hard to assess. One form of manifestation is through sampling errors or, conversely, the precision of estimates from sample surveys. For sample surveys from large populations, what matters is not the sampling rate but the sample size. And the sample size determinations for a desired level of precision are, or should be, made for the smallest areas (where separate estimates are required)-not the whole country or population. This is why total sample size for Indonesia's SUSENAS, for example, is 200,000 households.

Precision is not synonymous with accuracy of estimates from sample surveys. Accuracy includes both sampling error (precision) and nonsampling error. The latter, sometimes referred to as bias, is a catch-all term for all errors other than those due to having observed only a sample of the whole population, such as memory bias, nonresponse, systematic errors in instruments used, and processing errors. One reason often cited why expenditure is preferred over income in assessing poverty incidence is that the former can be more accurately reported by households or individuals than the latter. Unlike sampling error, which is inversely proportional to the square root of the sample size, nonsampling error is usually unaffected by sample size. It is not difficult to find situations where increasing the sample size can result in such things as looser control of the sample survey operation, and lower average ability of interviewees, leading to a higher nonsampling error.

Other statistical truisms need to be considered in planning the data collection for poverty monitoring and analysis. These will be taken up in more detail in the section outlining a statistical information strategy.

Differences in methodology are more evident when comparisons are made across countries than between measurement periods in the same country. The differences can be as clear-cut as in the choice of variable (e.g., income in the Philippines versus expenditure in Indonesia), or unit of analysis (e.g., household in the Philippines versus individuals in Indonesia). However, often the differences and their causes are difficult to discern, much less to measure. One serious impediment in trying to carry out such investigations is the inadequacy of the details in the documentation by the countries of their methods and databases.

It would be hard to find two countries that have the same perception of poverty and the same data specification and methodology for estimating it. One way to spot noncomparability is by juxtaposing countries' headcount ratios against some known or assumed correlates, such as per capita gross national product (GNP) converted to US dollars using the World Bank's Atlas method (Figure 1). The procedure is rough because the per capita GNPs are known to suffer from distortions so that the deviations from the trend reflect both these distortions and the noncomparability of the headcount ratios. However, as a preliminary screening procedure, the graph could show any obvious outliers. Thus, the PRC and Indonesia appear to have very low headcount ratios while those of Kazakhstan and the Philippines look higher than the norm for the majority of DMCs. Moreover, the vertical distance between these two groups of DMCs seems disproportionately long relative to the difference in their per capita GNPs.

Comparability is a major weakness of individual country-defined-and-derived poverty incidence estimates. The weakness is of little consequence to an individual country unless it wants to compare itself with other countries. Comparability is of greater concern to multilateral agencies and analysts as it permits additivity of the estimates and, therefore, easier regional and international assessment, monitoring, and analysis. Thus, it could be argued that the burden of conferring comparability to countries' estimates falls largely on multilateral agencies.

To illustrate, consider the methods of calculating the food poverty line and the (total) poverty line of the five Association of Southeast Asian Nations (ASEAN) countries of Indonesia, Malaysia, Philippines, Thailand, and Viet Nam. The approaches followed by the ASEAN countries in setting the food requirements for estimating the food threshold or food poverty line are similar (except for the most recent methodology used by Thailand). A typical food bundle (menu) for the poor is set up that satisfies a specified nutrient requirement or an energy requirement of about 2,000 calories per capita per day. The items in this food bundle, which vary from country to country with respect to both type and number, are then priced to derive the food poverty line. For Thailand, further adjustments to account for variations in household composition by sex and age are made. Thus, this approach can be used for deriving roughly comparable food poverty lines for ASEAN countries. Nevertheless, differences in the food bundles have been argued to result in incomparability of poverty incidence estimates across countries. For instance, the Philippines' food expenditure is argued to be too generous (fairly diverse with grains contributing 66 percent of the daily caloric requirement, which is lower than that in Indonesia and Thailand), resulting in an inflated poverty incidence compared with those of the other countries (World Bank 1996).

An even more significant contributor to the discrepancy is traced in the procedures used to adjust the food poverty line into a (total) poverty line (Box 2). In this regard, the ASEAN countries can be grouped into two. The Philippines and Viet Nam (also Thailand in the past) follow the Orshansky approach by dividing the food poverty line by the food share of the "poor group."^2/ However, this methodology has a serious drawback.

If the food share of the poor group declines (indicating an improvement in income), the food poverty line is divided by a ratio that becomes progressively smaller over time so that the total poverty line would likewise be pushed to a progressively higher level. Thus, the poverty incidence would remain unnecessarily high. Indonesia and Malaysia belong to the second group where the absolute expenditure for a predetermined list of essential nonfood items is added to the food poverty line to arrive at the total poverty line. However, the essential nonfood item specifications differ between the two countries.

Aside from DMCs themselves, the World Bank has done the most work on monitoring and analyzing poverty in Asia and the Pacific. Basic data collection is done through the living standards measurement surveys (LSMSs), which are usually cofinanced by the UNDP and implemented by a country's planning ministry or national statistics office. If implemented consistently, LSMSs ensure the uniformity of concepts, definitions, and methods across countries. LSMSs have been the first or only source of poverty information in some countries. However, there were occasions when the World Bank produced estimates and poverty assessments that did not agree with the countries' own (David et al. 1999).

Box 2. 1984-1988 Poverty Incidence Indonesia and the Philippines

In the 1980s, Indonesia and the Philippines were very similar in many respects based on their socioeconomic indicators. However, their headcount ratios showed an astonishing difference of 37 percentage points, with Indonesian estimates pointing to a much lower poverty incidence. A study that reviewed the methodologies used by the two countries in deriving their poverty estimates revealed that the way the nonfood component was computed in the poverty line contributed largely to the discrepancy (Asra and Virola 1992). In Indonesia, the Central Bureau of Statistics used a certain percentage of the food poverty line to account for the nonfood component. The percentage adjustment was derived as the proportion of the expenditure on a fixed bundle of basic nonfood items to total expenditure of the class where the food poverty line lies. In the Philippines, the total poverty line was calculated by dividing the food poverty line with the ratio of food expenditure to total expenditure. Indonesia's poverty line was 11 percent more than the food poverty line while it was 87 percent for the Philippines. Using the survey records, the study then recalculated Indonesia's poverty incidence by adopting as closely as possible the methodology used by the Philippines. The results showed that only between 4 and 8 percentage points separated the two estimates. This underscored the risks in comparing the countries' poverty estimates, as well as the need for caution in interpreting and using the countries' own estimates.

The World Bank has proposed $1/person/day at 1985 PPP prices as an international poverty line (World Bank 1990, Ravallion et al. 1991). The idea has gained the attention of analysts and international agencies because of its pioneering nature. However, it is not clear whether it has gained acceptance in individual countries. Some familiarity with its beginning and empirical underpinning could help potential users decide on its merits. The starting points are the countries' poverty lines that were converted to a common currency using PPP indexes based on consumption data in 1985 prices, indicated as z in $/person/month. The mean per capita private consumption, indicated as u (also in $/person/month in 1985 PPP prices), were likewise derived from the individual countries' national accounts data. Ordinary least squares regression of log(z) on (u and u2) provided a good fit to the data. Assuming that the u's are reasonably accurate, their strong empirical relationship with the z's implies that the z's are accurate as well. Somalia and India are at the bottom of this relationship, with z at $23/person/month, which may be interpreted as the line of extreme absolute poverty; further up the regression line is a group of countries (Bangladesh, Indonesia, Kenya, Morocco, Nepal, and Tanzania) that clusters around $31/person/month, a more generous and representative absolute poverty line. The next step is to fit a Lorenz curve on the monthly income or expenditure (survey) data of each country, likewise adjusted to 1985 prices, and then estimate the headcount ratio corresponding to $31/person/month (or any income point) or $1/person/day.

Headcount ratios from national poverty lines of DMCs and the corresponding estimates based on the $1/person/day definition are shown in Table 3 and Figure 2. Deviations from the 45-degree line of perfect match are naturally expected because, while the international poverty line is constant, the national poverty lines vary according to the local perceptions of what poverty means and the methods used to quantify it. Still, interesting observations emerge. The majority of the points fall below the 45-degree line, indicating that national estimates exceed the $1/person/day estimates, and often by considerable margins. A conspicuous exception is the PRC in 1995, in which the international poverty line gave an estimated poverty incidence (22 percent) at 3.5 times the national estimate (6 percent). In India, the international estimate (47 percent) was likewise 12 percentage points higher than the national estimate (35 percent).^3/ Conspicuously large deviations appear on the other side as well, including Kazakhstan (35 percent national estimate versus less than 2 percent international estimate), Sri Lanka (35 percent versus 4 percent), Pakistan (34 percent versus 12 percent), and Thailand (13 percent versus less than 2 percent).

Large deviations in both directions are the exceptions rather than the rule for Asia's developing countries, which raises several questions: Are the international poverty estimates for the individual countries useful, and for what purpose? Would a funding agency, such as the ADB, make use of 12 percent or 34 percent as a benchmark estimate in planning a poverty reduction strategy, for example, for Pakistan? Would Pakistan agree to replacing its 34 percent poverty incidence estimate with 12 percent? Are the World Bank estimates meant mainly as inputs to regional and global aggregations?

How do developing countries regard international poverty estimates? Consider an Asian developing country that last participated in an International Comparison Program (ICP) PPP price survey in 1985.^4/ Through models using other countries' data and its own GDP in local currency as inputs, its GDP in PPP prices and the PPP of its currency keep getting extrapolated to later years (at the World Bank or University of Pennsylvania). The World Bank obtains the country's most recent poverty line and private consumption GDP and combines these with other countries' to compute a global regression estimate of the poverty line as a function of private consumption in 1985 PPP prices. The World Bank uses the regression equation to predict the Asian developing countries' internationally comparable poverty line. The World Bank derives an empirical cumulative distribution function and Lorenz curve using either its own LSMS or the country's household income and expenditure survey (HIES) data set, and plugs into it the predicted poverty line to finally arrive at the proportion of the country's population below the international poverty line. More often than not, this figure turns out to be significantly different from the country's original estimate (Figure 2). Will the country find enough use for the international poverty line to continue supporting the exercise's data requirements? In other words, will the country renew its participation in ICP PPP surveys as well as continue its own HIES and poverty line estimation?

The incidence estimates based on the international poverty lines are rather dated because they require survey data from the countries as input (Table 3). Other reasons include the complexity of the estimation procedure and that the efforts were centralized outside the developing countries. Thus, in its present formulation, the international poverty line approach has limited applicability in quick monitoring of the impact of a crisis or a program on a country or group of countries

increase in poverty incidence in 1998 was rigorous. Because Thailand used the same methodology as in previous years, comparability was maintained with those estimates (National Economic and Social Development Board 1999). While the number of poor was reduced dramatically from 17.9 million in 1988 to 6.8 million in 1996, they increased to 7.9 million in 1998. In percentage terms, the incidence increased from 11 percent in 1996 to 13 percent in 1998.

According to official sources, Indonesia's poverty incidence also underwent a dramatic decline, from 40 percent in 1976 to 11 percent in 1996, representing an average drop of 1.4 percentage points per year. The number of poor people declined from 54 million in 1976 to 26 million in 1996, or 1.4 million people moving above the poverty line per year. The next official estimate is not due until 1999.^5/   In early 1998, the Central Board of Statistics (CBS) in Indonesia projected poverty incidence through mid-1998. To the extent that CBS produces the official poverty statistics, its projections are the closest to what can be regarded as an official or government assessment of the impact of the crisis on poverty incidence. CBS projections put the mid-1998 poverty incidence at 39 percent, or 79 million people (ADB 1999).^6/  A SUSENAS-type survey was conducted in December 1998, and the results were released in mid-1999. The poverty estimate indicates that the poverty incidence in December 1998 was 24 percent, meaning about 50 million poor. Although this estimate is not strictly comparable with the mid-1998 projection, some have mistakenly used the figures to indicate that the poverty incidence has declined from 39 percent to 24 percent. What can be inferred is that the poverty incidence increased from 11 percent in 1996 to 24 percent in 1998. If accurate, this estimate means that less than two years into the crisis the proportion of poor people slipped back to where it was more than 15 years ago, and the absolute number has increased by 1.5 times during the same period.

In the Philippines, the poverty estimates of the proportion of people below the poverty line declined slowly from 49 percent in 1985 to 38 percent in 1997. The less than 1 percentage point annual decline is overwhelmed by the 2.3 percent annual population growth rate: consequently, the number of poor people increased by one million, from 26.2 million to 27.2 million during the 12 year period (National Statistical Coordination Board 1999).

In the absence of an official estimate, the World Bank estimated that the number of poor people in the Philippines increased by 665,000 in 1998. This figure translates into 27.915 million poor people in 1998 compared with 27.274 million in 1997. Taking population growth into account, the World Bank estimate would cause a 1 percentage point rise in the poverty incidence from the 1997 level.

Malaysia's poverty incidence is estimated to have increased from 6.8 percent in 1997 to 8.0 percent in 1998 (Ali 1999).

In summary, little doubt exists that the crisis worsened poverty incidence in the four countries (Figure 3). The impact is the least in the Philippines and Malaysia, moderate in Thailand, and the greatest in Indonesia, which is the same order as the economic impact of the crisis on these countries.

The crisis and the resulting heightened priority for reducing poverty have drawn attention to the inadequacies of the available statistics on poverty and, therefore, on the need for a poverty statistical information strategy to address these inadequacies. The strategy should meet the information needs of the developing countries and funding agencies as well as those of other stakeholders, such as local governments and nongovernment organizations. The strategy should address the issues in the previous section in the context of poverty profiling, monitoring, and analysis.

The strategy rests on the assumption of close cooperation between country and funding agency brought about by a common goal. This appears reasonable because poverty reduction or elimination is already a goal of most governments. For example, upon assuming office in 1992, Philippine President Fidel V. Ramos declared his administration's goal was to reduce Philippine poverty incidence from 56 percent to 30 percent by the end of his six-year term. On the supply side, the funding agencies' roles will be funding and actively leading the development and promotion of the concepts and methodologies needed to implement the strategy. The countries will naturally be the producers and analysts of the primary data; any primary