Chapter 3 Country Practices in Compiling Poverty Statistics I.P. David New York, 28-30 June 2005 Contents 3.1. Introduction [still to be drafted] 3.2. The Demand for Poverty Statistics [ AFRISTAT ] 3.2.1. The Demand for Poverty Statistics 3.2.2. The Widening of the Scope of Poverty 3.3. Income or Expenditure Based Measurement Methods 3.4. Direct Measures of Food Poverty 3.5. Non-Income Measurement Methods 3.6. Harmonizing Poverty Statistics Production in Developing Countries References [still to be finalized] 3.3. Income or Expenditure Based Methods Cost of Basic Needs (CBN) is Method Most Used. Split basic needs into food and non-food; estimate costs separately; 3 broad steps involved: 1. Specify dietary (energy) threshold (T); determine food basket satisfying threshold; fpl = cost of food basket. 2. Choose operational definition. of basic non-food needs; cost is non-food poverty line (nfpl). fpl + nfpl = tpl (total poverty line) 3. Compare PLs against metric: income or expenditure Remarks: Unit of analysis/observation is household; statistics are in per capita and national currency. 3.3.1. Specify a Food Poverty Threshold Dietary energy consumption used as proxy, based on simplifying assumption that if one gets enough energy, he/she gets enough of the other necessary nutrients. Nutrition/Research Institutes in Health/Science Ministries get into the act, guided by FAO-WHO recommendations or practice. Outputs include RDAs/RENIs (Table 1), energy threshold [T] (Table 2), and Food Composition or Conversion tables. Poverty statistics may be very sensitive to changes in T; Bangladesh, Philippines, Vietnam Table 3. Bangladesh Food Poverty Incidences from DCI Method and Two Energy Thresholds 2120kcal 1805kcal Difference 1983-84 62.6 36.8 25.8 1985-86 55.7 26.9 28.8 1988-89 47.8 28.4 19.4 . 1991-92 47.5 28.0 19.5 1995-96 47.5 25.1 22.4 Average - - 23.2 Year Table 1. Dietary energy RDAs, Philippines and Sri Lanka, in kilocalories Age groups Under 1 year 1-3 4-6 7-9 10-12 13-15 16-19 20-39 40-49 50-59 60-69 70 & over Philippines ----------------Male Female 700 700 1350 1350 1600 1600 1725 1725 2090 1930 2390 2010 2580 2020 2570 1900 2440 1800 2320 1710 2090 1540 1880 1390 Sri Lanka ----------------Male Female 818 818 1212 1212 1656 1656 1841 1841 2414 2238 2337 2300 2500 2200 2530 1900 2404 1805 2277 1710 2024 1520 1771 1330 Table 2. Dietary energy thresholds used by a sample of countries, 2000-2004 Threshold 2000 kcal 2030 2100 2124 2133 2138 2207 2238 2282 2250 2283 2288 2300 2309 2300 2436 2400 2470 2700 3000 Country Maldives, Philippines (but also specifies 80% of protein RDA which is equivalent of 50 milligrams per day). Sri Lanka Cambodia, China, Indonesia, Laos, Mongolia, Thailand, Vietnam, Fiji, Turkey, Armenia Nepal Madagascar Malawi Paraguay (all country) Oman Moldova Kenya Burkina Faso Albania Cameroon Jordan Iran Iraq Senegal, St, Kitt & Nevis, Morocco, Bahamas Belarus (all country) Sierra Leone Uganda 3.3.2 Construct Food Basket that Satisfies T Rank food items from consumption survey (based on value, quantity, or frequency of households reporting). Ranking is made from reference population, e.g. lowest quartile of hholds per capita income distn. Stopping rule: Food basket is the top items that provide T’≈ T kilocalories. Items range from 7 to 205 with a median 40. Multiply all items’ contributions by (T/T’) How many baskets? 3.3.3 Compute food poverty line (fpl). Let q1, q2, …, qf be the quantities of the f items in the food basket that supply e1 + e2+ … + ef = T’ kilocalories. Let p1, p2, … , pf be the unit prices of the f food items. fpl = (T/T’) ∑ qi pi in national currency. How many fpls? How to define reference population and what prices to use to ensure consistent welfare level in each domain? 3.3.4 Alternative Approaches Compute total expenditure and total kcalories consumed by the reference population. The ratio, price per kcalorie, can be multiplied by any choice of T to get as many fpls as there are choices. Eschews food basket, but requires complete array of expenditure and food composition (conversion) table for all food items consumed. (∑RDA) x price per kcal = household level fpl, where sum runs through the age by sex energy RDAs of household. This can be compared with total income or expenditure of household. This is traced to Prof. Kakwani, and tried in Laos, Thailand, Jordan. Avoids computing per capita values, but still in national currency. 3.3.5. Compute total poverty line (tpl) Define essential non-food basic needs, estimate cost (nfpl), and add to (fpl). Countries use one of three methods: List essential non-food needs, price each, and total cost is nfpl; tpl = fpl + nfpl. Example, Indonesia. Regression (World Bank). tpl = (2-a)fpl, where a is intercept of OLS reg of S = fe/te) on log (te/fpl) in reference population. Used in WB assisted countries. Engel’s coefficient. Compute (fe/te) from hholds within a narrow band around fpl; tpl = {2 – (fe/te)}fpl. Used by many other countries not dependent on WB-LSS. Comparisons of Three Methods List tends toward smaller tpl. Highly subjective, decisions on what to include/exclude subject to criticism or pressure. Different bundles for different groups, e.g. bus for urban, bicycle or motorized bike for rural, leads to different welfare levels? Regression and Engel’s coefficient more likely lead to comparable results. What to do when regression is not a good fit? What band around fpl, and how many regressions or coefficients? (See next slide). Figure 1. Ratio of Food Expenditures to Total Expenditures, 1994, Philippines fe/te 0.71 0.70 0.69 0.68 0.67 0.66 URBAN 0.65 RURAL TOTAL 0.64 0.63 0.62 +/- 2 +/- 5 +/- 10 Band +/- 15 +/- 20 Fourth Method of Incorporating nfpl Instead of adding {1-fe/te)}fpl to fpl and arrive at tpl = {2 - (fe/te)}fpl, a few developing countries (Philippines, some in ECLAC) use tpl = fpl/(fe/te) . This gives higher tpls: fe/te ----½ 2/3 ¾ 1 2 – fe/te --------1.50 1.33 1.20 1 te/fe ----2 1.5 1.25 1 3.3.6. Compute Incidence and Related Statistics Household and all M members with per capita income (expenditure) < fpl are food-poor. Replace fpl with tpl and you get absolutely poor. Design-based estimates of totals follow (e.g. y=1 if household is poor, 0 otherwise; and y = M if household is poor, 0 otherwise). Poverty incidence is not straightforward. Some countries use population projections as divisors (but these may not be available for certain domains of interest). Design-based estimates may be suggested, but these give different results in general. Very few countries, if any, have projections of the number of households. Problem of finding denominator not trivial; complicated by need to reconcile with implications on population projections. Philippines case. 3.3.6. Continued Household poverty incidence < population poverty incidence. Important to specify which. Serious questions about quality of basic data on food consumption (expenditure, quantity, unit prices), income and expenditure from traditional HIES. Limited empirical evidence point to different values obtained from different data capture methods and recall periods. Need additional studies. Results very sensitive to choice of divisor for per capita calculations. Countries expressed need for guidance in using adult equivalents (e.g. for food) and scale economy models (for income or expenditure). Majority still use unadjusted M. 3.3.7 Updating Poverty Measures. Food baskets, energy thresholds and reference populations seldom changed. fpl and tpl with list method can be updated anytime new prices become available, e.g. annually. The same regression intercept or Engel’s coefficient used to update tpl until the next HIES. Poverty incidences and counts can be updated only when a new HIES round is run because per capita income/expenditure is needed. (see next slide) This is sometimes confused with updating poverty lines, hence unduly heavy demand by users. HIES are very costly and complicated undertakings. 3.3.8. Estimating Trends or Changes For ratio (Y/X), V(Y/X) = V(Y) + V(X) – 2 Cov(Y,X) For change in ratio, . V(Yt2 – Yt1 ) = V(Yt2) + V(Yt1) – 2 Cov(Yt2,Yt1 ) where the y’s are ratios themselves. For inferences, Yt2 – Yt1 ± Z se(Yt2 – Yt1 ) may guard against hasty declaration that the war against poverty is being won, or else of search for kinks in the methodology when the observed change is small or negative. 3.3.9 Relative and Subjective Income Based Poverty Lines. Examples of Relative PLs in Developing Countries: 50% of the median per capita income (ECLAC) 40% 50% of the median per capita income (Oman) of both the mean and median per capita incomes (Iran). Relative PLs are more popular in the developed countries. Easier to measure, hence used more in poverty intervention than in monitoring. 3.3.9 Continued Examples of Subjective Poverty Lines ‘Self-assessed poverty’ approach , such as Philippines Social Weather Station asking heads of households their income, whether they consider themselves poor, and if so, how much more income they need so they will no longer think of themselves as poor. Egypt tried a similar approach but found that the method overestimates the extent of poverty because people’s expectations, especially the educated in the urban areas, exceed their current levels of living by a large margin. Philippines based on a small sample (1200-1500 households) and repeated quarterly; hence 12 time series points in the 3-year interval that official poverty statistics are produced. 3.4. Direct Measures of Food Poverty 3.4.1 Estimate empirical CDF of per capita energy consumption Let (ai) = 1 if ai ≥ 0 = 0 if ai < 0 F (t) = Σ πi-1 (t – xi) / Σ πi-1 3.4.1. Continued Example: Vietnam National Nutrition Survey, 2000 Energy cut-off % of population below cut-off < 1500 kcal 4.1% < 1800 kcal <2100 kcal 17.9% 45.1% Note: The official food poverty incidence from GSO was 12-13% in 2000 3.4.2. Household Size for Per Capita Calculations Example: Philippines. From Food Consumption Survey of the Food and Nutrition Research Institute. Table 5. Per Capita Energy Consumption Distributions (% of Population) Using M and M0.7 as Divisors, Metropolitan Manila - Philippines, 2003 Divisor/Cut-Off (kcal) <1500 <1800 <2000 <2100 Family Size, M 48.0 74.0 83.0 88.0 M* = M0.7 7.9 16.0 22.5 26.3 3.4.3. Eschewing per capita calculations ∑kcal < ∑RDA can be used directly to classify hholds and persons therein as either food poor or not. Energy gap = ∑w{∑RDA - ∑kcal} if {∑RDA - ∑kcal} > 0 = 0 otherwise The RDAs may be changed proportionately by ± 15% and ± 30% and end up with five points that give a picture of how food poverty behaves with RDA specifications. If countries have these, then comparable food poverty estimates can be easily interpolated for any choice of common energy threshold. No per capita calculations, no currencies, no prices, no reference populations. If desired, energy gap x price per calorie will provide energy gap in money terms. 3.5 Non-Income Measurement Methods Minimum Basic Needs (MBN) or Unmet Basic Needs (UBN) most popular among developing countries. The other approaches have not graduated beyond the small scale experimental or analytical phase. UBN indicators that are non-income and measure longer term outcomes or outputs serve as complement to CBN indicators that are income-based and measured from short-term inputs. Examples of UBN indicators are the MDG indicators minus the income indicators. 3.5.Non-Income Indicators, Continued Nearly all countries in ECLAC have UBN poverty monitoring systems in place . The number of dimensions and indicators depend on data availability, e.g. from censuses, surveys and administrative records. It is seldom that a new data collection system is initiated mainly for compiling UBN indicators. Broad categories are dwelling characteristics, access to safe water, and access to sanitation facilities and basic education. UBN systems also in place in many ESCAP countries. Bangladesh, for example, uses infant mortality as proxy indicator for the primary health care system, primary school enrollment rate for basic education, and housing characteristics (access to tap water, toilet facilities, electricity, and type of building material used) for living conditions. 3.5 Non-Income Indicators, Continued UBN approach is far from widespread in Africa. Only three of the 10 members of the Economic Community of Western African States (ECOWAS) acknowledged having a UBN system in place. The main poverty dimensions considered are basic education, primary health, and housing characteristics such as access to safe water, toilet facilities and type of building materials used. UBN methods at times brought down to sub-national levels. China monitors community level indicators, such as percent of villages accessible by road, percent with land line phone connection, and percent with electricity, illiteracy rate, child enrollment rate, and labor migration rate. 3.5. Non-Income Indicators, Continued Producing composite indexes from indicators expressed in different units of measure is a perpetually subjective and difficult task. This has not stopped some international agencies from compiling them; e.g. HDI and other indexes in UNDP-HDR. These indexes, however perhaps have more value as advocacy tools and less as monitoring tools especially at the national and sub-national levels. Few developing countries, if any, compile composite UBN indexes, preferring to use the indicators individually and collectively in much the same way that they are used to monitor progress in the MDGs. 3.6. Harmonizing Poverty Statistics Production Harmony = Synchronized timing; comparability; balance between supply and demand. Internationally, it also means improving capacities in the statistics-deficient countries. National statistical information systems have evolved to a point that countries follow similar updating cycle and sequencing for certain parts of their socioeconomic databases; e.g. censuses every ten years, demographic surveys 3-5 years, agri surveys every year or season, etc. This evolution has enabled IMF to formalize the periodicities of statistical series in its General Data Dissemination System (GDDS) and Special Data Dissemination System (SDDS). Poverty statistics, however, are not covered in these systems. 3.6. Harmonizing , Continued Poverty reduction implementing agencies want statistics at smaller domains and updated more frequently (usually yearly), than what NSOs can provide (excepting censuses). One strategy is NSO continues doing poverty monitoring surveys say every 3 to 5 years which are the sources of official statistics; NSO helps agencies plan and implement their poverty information gathering program, so that longer- term, comparability is improved; however, the agencies’ data should not be used to produce aggregates for domains where NSO official statistics exist. International agencies generally want annual national data, and will project, intrapolate or extrapolate otherwise. This is ok, as long as these are for global comparison/analysis only.. 3.6.3. Main Sources of Non-Comparability; Possibilities for Improvement. Different dietary energy thresholds (Table 2). True within country also, e.g. India. Possible improvement: estimate per capita energy consumption CDF. For food poverty, consider Kakwani’s approach or ∑kcal < ∑RDA criterion. Food baskets vary. Very difficult and not practical to recommend one food basket. Possible solution: per capita energy consumption CDF, combined with use of adult equivalents based on age by sex RDAs; or Kakwani’s approach or ∑kcal < ∑RDA criterion for food poverty. Definition and measurement of non-food basic needs vary. Suggestion: Use either regression of Engel’s coefficient, combined with use of adult equivalents possibly based on a scale economies of need model. 3.6. Main Sources of … Continued Countries split between income and expenditure. Recommendation: Each country sticks to one, but do some empirical research to find out likely difference in poverty levels between income and expenditure metrics. Use scale economies of need for per capita calculations. For food poverty, consider ∑kcal < ∑RDA for determining the food poor. Method of data capture varies. Very difficult to get agreement. Sustainability a very important factor (e.g. Vietnam going back to old method). Try combination of objective and recall methods; e.g. combine food weighing (subsample) with face-to-face interview (main sample). More evidence from empirical research needed to guide on data capture decision. Thank you!