Combining light monitoring surveys with integrated surveys to improve targeting for poverty reduction : the case of Ghana

 77309 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. li 195-219 Combining Light Monitoring Surveys with Integrated Surveys to Improve Targeting for Poverty Reduction: The Case of Ghana Hippolyte Fofack Policymakers use poverty maps to design and assess poverty programs. The accuracy of these maps, which is critical for targeting, depends largely on the nature of the instru- ment used to construct them. Recently, in response to tight budget constraints, many countries have begun to construct poverty maps based on light monitoring surveys that rely on short questionnaires. This article shows that poverty maps constructed from such surveys are not accurate and could result in substantial leakage. Light monitoring surveys do include large samples that can help to target poverty programs at subregional levels. Combining these surveys with more detailed Integrated Surveys can help researchers reduce targeting errors significantly and build improved poverty maps with finer levels of disaggregation. Poverty analysis and the design of targeted programs traditionally have been based on comprehensive household surveys. Such surveys are conducted infre- quently, however, making it difficult to assess the effects of macroeconomic re- forms on poverty and income inequality in the short and medium term. Limited budgetary resources constrain governments from conducting these surveys more often. Policymakers fully recognize the need for reliable poverty maps, and several surveys have been designed and proposed as short-term alternatives to comprehen- sive surveys. These short-term instruments, which include Rapid Appraisal Meth- ods (Narayan and Srinivasan 1994) and Priority Surveys (Marchant and Grootaert 1991), are known as light monitoring surveys, light monitoring surveys are de- signed to quickly identify groups that interventions should target. In contrast, more comprehensive surveys are designed to conduct integrated poverty analysis and to draw inferences on welfare (see Ravallion 1996 and Deininger and Squire 1996). These surveys include household budget surveys, Living Standards Measurement Hippolyte Fofack is with the Africa Region at the World Bank. Hit e-mail address is hfofack® worldbank.org. Earlier versions of this article were presented at the International Workshop on Poverty Mapping, organized by the United Nations Environment Programme and Global Resource Information Database in Norway, and at the International Conference on Geographical Targeting, organized by Johns Hopkins University in Baltimore, Md., held in October 1998. The author would like to express his warm thanks to Lionel Demery, Jack van Hoist Pellekaan, and David Bigman for their useful comments and encouragement. He also is grateful to the article's anonymous referees for their valuable comments. © 2000 The International Bank for Reconstruction and Development/THE WORLD BANK 195 196 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 Study (^SMS) surveys (Grosh and Glewwe 1998), and Integrated Surveys (Delaine and others 1992). (For details on the design and implementation of these surveys, see Demery and others 1992 and Grosh and Munoz 1996.) Comprehensive Integrated Surveys are broad in scope, but they take a long time to administer and usually are based on relatively small samples in order to contain costs. The questionnaire collects data on most household and individual characteristics, providing extensive information on household income, credit, and savings; household enterprises; the value of durable, productive, and financial assets; agricultural livestock; food processing and consumption of own produce; food and nonfood consumption; and other expenditures. The survey is adminis- tered over the course of a year, with multiple visits to households to capture seasonality. Nonsampling errors are reduced by shortening the recall period. Using integrated surveys to target programs at the subregional level is difficult, how- ever, because the sample sizes are too small to allow sound inferences to be drawn. Light monitoring surveys are administered very quickly. They collect socio- economic data on a smaller set of variables than Integrated Surveys, but their sample size is typically much larger. These surveys are based on single visits to households. They thus do not capture seasonality in consumption patterns nor provide accurate estimates of consumption and income. The short questionnaire and single visit, however, reduce the time needed for data collection and allow for a larger sample size with better representation of different geographic areas. The limited coverage of expenditure items in light monitoring surveys has some major drawbacks for policy analysis, however! Short questionnaires focus on a few sets of goods; consumption aggregates based on these data are likely to pro- vide estimates of total expenditures that are lower than estimates from Integrated Surveys (Deaton and Grosh forthcoming). Moreover, the underestimation of to- tal expenditures is not likely to be uniform, shifting the Engel curve but preserv- ing welfare rankings. Rather, the bias varies significantly across households and regions, in part because regional differences in consumption patterns and changes over time—which are determinants of the distributions of income and expendi- tures—are not taken into account The distributions that emerge from the two types of surveys thus differ significantly. Despite evidence that they might generate inaccurate estimates of aggregate expenditures, light monitoring surveys have been used extensively for policy de- sign. In Sub-Saharan Africa, for example, World Bank poverty assessments draw major recommendations from these surveys (World Bank 1997). This article shows that welfare indicators estimated from light monitoring surveys are biased and that the bias affects not only the estimated magnitude of poverty but also its geographic distribution. The geographic distribution of poverty indicated by light monitoring surveys differs significantly from that indicated by Integrated Sur- veys, in part because per capita household expenditures, which are the basis for targeting, underestimate aggregate expenditures. Large differences in the level of systematic bias will yield inaccurate poverty maps; targeted income transfer schemes based on light monitoring surveys may result in leakage.  Fofack 197 This article investigates how comprehensive Integrated Surveys can be com- bined with light monitoring surveys to improve geographical targeting, make transfers for poverty alleviation.efficient, and sharpen the inferences on welfare measures that are drawn from light monitoring surveys. As core components of national statistical programs, light monitoring surveys and Integrated Surveys are both household-level surveys, with important similarities. Similarities in their sampling frames and sampling designs, as well as the proximity in their imple- mentation make their combination extremely appealing for poverty and policy analysis. L LIMITATIONS OF LIGHT MONITORING SURVEYS AS INSTRUMENTS FOR TARGETING To understand the implications of using light monitoring surveys as the basis for poverty analysis, I compare welfare indicators estimated from such surveys with those derived from Integrated Surveys. To allow full comparability, I ap- proximate aggregate expenditures as subsets of consumption items from the In- tegrated Survey. The study thus can be viewed as a counterfactual experiment, since estimates of total expenditures are known. Ideally, we would like to compare an Integrated Survey and a light monitoring survey that were conducted over the same period. But the surveys were never conducted concurrently in Ghana. Moreover, in countries in which light moni- toring surveys and Integrated Surveys have been carried out consecutively, varia- tion over time and changes in the sampling frame and design make it difficult to assess the performance of these surveys from estimates of household welfare. In this study I make comparisons on households for which expenditure aggregates have been suitably constructed under the assumptions of both the full Integrated Survey and the approximated light monitoring survey. There is no time lag in data collection between the two surveys, and errors in measurement associated with variation in the sampling design are completely eliminated because the com- parisons are made on the same unit of analysis. Distribution of Expenditures from Integrated and Light Monitoring Surveys This study is based on the third in a series of surveys of living standards in Ghana (Ghana Living Standards Survey [GLSS] 3).1 The survey was administered to a sample of about 4,500 nationally representative households over the course of a year.2 It collected data on all dimensions of household welfare and economic behavior, including highly disaggregated and comprehensive data on household 1. The first two surveys, conducted in 1987 and 1989, are less comprehensive than the third survey. All three surveys are based on the master sample of enumeration areas defined by the 1984 population census. 2. A multistage stratified random sampling was used in selecting the sample. Initially, 407 dusters were enumerated, and households were selected with a probability proportional to the size, with 15 households drawn in each urban cluster and 10 households in each rural cluster. 198 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 income and expenditures. Data were collected on 107 food items, and the range of nonfood items was wider than in the first two surveys (Ghana Statistical Ser- vices 1996). Estimates from the third survey are considered more accurate than estimates from the first two, in part because estimates from the third survey are based on much shorter bounded recall periods (8 recalls at two-day intervals in rural areas and 11 recalls at three-day intervals in urban areas). Although Integrated Surveys collect extended information on both household income and expenditures, I use expenditure data as the measure of economic welfare, partly because nonsampling errors due to underreporting of income bias reported household income. There also are strong theoretical reasons for choos- ing expenditures over income (Deaton and Muellbauer 1980). Estimates of total household expenditures on food and nonfood constructed from the GLSS 3 data use 6 aggregates constructed from 17 subaggregates. These estimates account for all household expenses, including total household expenditures on rent (includ- ing imputed rent on owner-occupied, rent-free, or subsidized dwellings); con- sumption of home-produced food; and the value of wage income received by household members in the form of food. Other imputed expenditures include total wage income paid in-kind to household members, the value of produce of nonfarm enterprises consumed by households, and the use value of durable goods. The value of remittances made by households, as well as all other expenses, such as spending on education and household amenities, are also included in aggre- gate expenditures. Missing values and outliers are imputed on each variable based on a methodology developed by a team from the Development Economics Re- search Center at the University of Warwick (see Ghana Statistical Services 1996). All expenditure data are adjusted to account for inflation, which was relatively high during the survey period, using March 1992 Accra prices as the base. No adjustment is made for seasonal effects on household expenditures. I obtain an estimate of total household expenditures by summing across all household expenditure items, subaggregates, and aggregates on food and non- food. I first sum across items and subaggregates to obtain intermediate values for expenditure aggregates at the household level: (1) St- /-i where h represents households (ranging from 1 to the total sample size), and; = (1,2,...,N) represents the total number of items. P^ and Q^ represent the price and quantity of item;, a component of aggregate k consumed by a given house- hold h. The multiplicative factor Sfj is the frequency of purchase (recall period) of a given item within household h; X is the "frequency factor" or frequency of enumerator visits to households.3 3. The frequency factor X is determined from the surrey design. In the GLSS 3 survey households were visited 8 tunes at two-day intervals in rural areas and 11 times at three-day intervals in urban areas.  Fofack 199 A final estimate of total household expenditures for Integrated Surveys is ob- tained by summing across expenditure aggregates: (2) where k = (1,2,...yA) is the total number of subaggregates, and Y^ is the total household expenditure aggregate from the Integrated Survey for household h. We obtain estimates of total household expenditures based on light monitoring surveys by similar aggregation, albeit over a much smaller number of items and subaggregates. The components of aggregate expenditures are selected following the guidelines of the standard Priority Survey, which recommends limiting the collection of expenditure data to key food and nonfood items (Demery, Grootaert, and Hill 1991). Total expenditure aggregates constructed here are based on three aggregates: two nonfood aggregates, which include expenditures on education and health, and a food aggregate, which contains 10 key food items (corn, rice, cassava, plantains, beans, groundnuts, palm oil, sugar, salt, and meat). These expenditure items, especially spending on education and food, account for a large share of total expenditures. We can estimate total household expenditures from a Priority Survey by sum- ming across these three aggregates, after adjusting for inflation. The adjustment for inflation is necessary because light monitoring surveys are based on a single visit to households, and inferences on welfare are drawn assuming no seasonal variation in consumption patterns—an assumption that could limit the value of these surveys for poverty analysis. An estimate of total household expenditures from the light monitoring survey is provided by: a<A a<A n<N The frequency factor, X, does not appear in equation 3 because the data are collected in a single visit. In the light monitoring survey both the number of aggregates and the number of items are much smaller than in the Integrated Sur- vey (a < A and n < N). As a result total household expenditures are generally underestimated, and Yu^ < YIS. These aggregates are adjusted for household size to produce household per capita expenditures. Summary statistics on the distributions of these adjusted variables reveal im- portant differences (table 1). As expected, the level of welfare estimated from the light monitoring survey is much lower than that estimated from the Integrated Survey. At the national level mean per capita expenditures estimated from the light monitoring survey are about 8 percent of those estimated from the Inte- grated Survey. The magnitude of the difference in welfare is particularly great in rural areas. Mean per capita expenditures estimated from the light monitoring Table 1. Distribution of Per Capita Household Expenditures across Regions Integrated Survey Light monitoring survey Ratio of mean Share of Share of per capita Mean per national mean Mean per national mean expenditures of capita per capita capita per capita light monitoring Agroclimatic expenditures Coefficient expenditures expenditures Coefficient expenditures survey to Integrated region (cedis) of variation (percent) (cedis) of variation (percent) Survey Accra 260,309 155.69 1210 38,276.16 181.30 2.165 0.1470 Other urban 225,232 155.88 1.047 33,241.86 175.10 1.880 0.1476 Rural Forest 217,835 147.40 1.012 11,628.40 231.05 0.658 0.0534 Rural Coastal 213,641 178.55 0.993 8,769.98 269.86 0.496 0.0411 Savannah 188,691 150.81 0.877 8,674.90 35034 0.491 0.0460 All urban 233,889 157.68 1.087 34,484.26 179.13 1.951 0.1474 All rural 205,910 166.10 0.957 9,343.80 285.52 0.529 0.0454 National 215,186 163.87 1.000 17,678.70 278-50 1.000 0.0822 Source: Author's calculation* based on Ghana Statistical Services (1995,1996).  Fofack 201 survey in urban areas are about 14 percent of those estimated from the Inte- grated Survey; in rural areas the light monitoring survey estimate is about 4 per- cent of the Integrated Survey estimate. This bias is exacerbated by the fact that own-produced consumption, which accounts for a large share of consumption in rural areas, is not included in the light monitoring survey. The different scope of sampled items also biases the results. Increasing the number of consumption items in the light monitoring sur- vey slightly reduces the bias between expenditure aggregates in the two surveys, but the difference remains large, and the sizable urban-rural bias persists. When the number of sampled items is increased to 20, aggregate mean per capita ex- penditures in the light monitoring survey rise from 8 to 10 percent of aggregate mean per capita expenditures estimated from the Integrated Survey. Increasing the number of sampled items to 30 raises the light monitoring survey estimate to 11 percent of the Integrated Survey estimate. Increasing the number of items to 20 raises the light monitoring survey estimate from 14 to 20 percent of the Inte- grated Survey estimate in urban areas; in rural areas the increase is only from 4 to 5 percent.4 The persistence of a large urban-rural bias despite the increase in the number of sampled items may reflect the importance of own-produced food in rural areas. Estimates of total household expenditures from light monitoring surveys are based on just a few items and generally are biased downward. The downward bias is likely to shift the distribution of expenditures to the left. The difference in the scope of items and subaggregates may explain the large absolute difference between the distributions revealed by the two surveys. The variation in the struc- ture of these distributions and the large urban-rural bias may be due largely to the nature of the consumption items in the overall aggregate. Although consump- tion of own-production represents a large share of household consumption in rural areas, it is not accounted for in the light monitoring survey aggregate, partly because the rural economy has a low level of monetization and also because key consumption items are more tradable in urban areas, where induced transaction costs tend to increase their relative prices. Poverty maps constructed from the light monitoring survey reveal patterns different from those constructed from the Integrated Survey. Mean per capita expenditures are much lower across all agroclimatic regions, and the size of the bias is not uniform across regions. Measured by mean per capita expenditures, Savannah is no longer the poorest region of Ghana, and the differences in mean per capita expenditures between Savannah and Rural Coastal decline. These fig- ures represent about 50 percent of national mean per capita expenditures as esti- mated by the light monitoring survey. Moreover, the bias toward higher rural poverty is greater in the light monitoring survey than in the Integrated Survey. The rural mean estimate accounts for less than 30 percent of the urban mean in 4. The national mean per capita expenditure estimates based on 20 and 30 items are 22,883 crdis and 25,803 cedis. These estimates are much higher in urban areas (44,939 cedis and 50,858 cedis) and much lower in rural areas (11,944 cedis and 13,377 cedis). 2 02 THE WORIJ) BANX ECONOMIC REVIEW, VOL 14, NO. 1 the light monitoring survey, whereas it accounts for about 90 percent of the urban mean in the Integrated Survey. The variance in the distribution of household per capita expenditures is rela- tively high in the light monitoring survey estimates. Although the estimates of mean per capita expenditures are uniformly smaller in the light monitoring sur- vey than in the Integrated Survey, the coefficient of variation of per capita expen- ditures is higher across all agrocUmatic regions. The dispersion around the na- tional mean is greater in the light monitoring survey, where the ratio of regional to national estimates has a larger range. Differences in mean estimates are much smaller in the Integrated Survey, where the ratio oscillates around 1, suggesting that income inequality may be much higher in the light monitoring survey. The differences in variances also are measured in terms of the Gini coefficient and the Lorenz curve of inequality (figure 1). Figure 1 supports the hypothesis that light monitoring surveys record much greater income inequality than Integrated Sur- veys. The Gini coefficient approximated from the light monitoring survey is substantially higher (0.56) than the corresponding coefficient for the Integrated Survey (0.34). The relatively large variance observed in the light monitoring survey has im- portant implications for poverty analysis. The poverty gap is proportional to the income gap, and the severity index is proportional to the squared deviation from the poverty lines when a i l . And, unexpectedly, large variances are likely to increase the estimate of the poverty gap (and therefore the cost of poverty reduc- tion, which is proportional to the income gap). Figure 1. Lorenz Curve of Income Inequality Cumulative income share 1T 45-degrce line Light Monitoring Survey Integrated Survey 02 03 0.4 0.5 0.6 0.7 OS 0.9 Cumulative population share Source: Author's calculations.  Fofack 203 These large variations are also illustrated by the distribution of per capita household expenditures across expenditure quintiles, which differs significantly between the two surveys.5 The bottom 40 percent of the population accounts for less than 8 percent of total expenditures in the light monitoring survey and 20 percent of all expenditures in the Integrated Survey. In the upper quintiles, ex- penditure shares are overestimated in the light monitoring survey (figure 2). These differences in the distribution of expenditures across quintiles suggest that pov- erty maps constructed from a light monitoring survey may not be accurate be- cause the instrument tends to overestimate the welfare of the nonpoor and un- derestimate the welfare of the poor. Implications for Policy Analysis and Poverty Mapping The latest Ghana poverty profile uses two measures of poverty (Boateng and others 1990). The broader measure includes all people whose per capita expendi- tures are two-thirds or less of the national mean (equal to 132,230 cedis per person per year).6 The narrower measure of poverty includes all people whose per capita expenditures amount to no more than half of the national mean (equal Figure 2. Distribution of Per Capita Household Expenditures Per capita expenditure share D Light Monitoring Survey I Integrated Survey 0.4- 0.3- 0.2- 0.1 PTI l I I (Poorest) Expenditure quintiles Source: Author's calculations. 5. Expenditure quintiles for the Integrated Survey are derived from the full consumption aggregate adjusted for household size; expenditure quintiles for the light monitoring survey are constructed from consumption aggregates adjusted for household size, but with a limited number of items. 6. The same definitions were used in the 1987-88 poverty profile. The 1988 base values were adjusted for inflation and expressed in 1992 constant prices for the 1995 profile. The lower poverty line is equal to the same share of national mean per capita expenditures estimated from the full G LSS 3. 204 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 to 107,188 cedis per person per year). I use the lower poverty line (z B = 107,188) as the cutoff point in the Integrated Survey, because maximum targeting is easily achieved at the lower and upper ends of the distribution, where the within-group variance and the probability of household misclassification are lower than in the middle of the distribution. Similarly, I define the relative poverty line to be half of mean per capita expenditures as approximated by the light monitoring survey (ZLMS = 8,839). I use that measure for cross-sectional analysis, looking at varia- tions in poverty rates across regions in the two surveys during the same period. I do so because total household expenditures aggregated from the light monitor- ing survey show a large urban-rural bias—the rural expenditure aggregate is sub- stantially lower—and because using an upper poverty line would exacerbate the scope of rural poverty. In order to assess the performance of the light monitoring survey as a targeting instrument, I estimate the headcount, poverty gap, and severity indexes from distributions in both the light monitoring survey and the Integrated Survey (table 2)7 Performance is assessed by the probability of type I and type II errors, as well as by the rate of mistargeting. The probability of a type I error is defined formally as e ; = [P(yy e P I y; e P)], where P represents the set of poor households or individuals (yy) and P represents the set of nonpoor households or individuals. A type I error can be referred to as an error of inclusion because it indicates the probability of classifying nonpoor households or individuals as poor. The probability of a type II error is defined as ew = [P(y; e VI yt € P)]. A type LI error can be referred to as an error of exclusion because it gives the probability of classifying poor households or individuals as nonpoor. The rate of mistargeting depends on the size of these two errors. Perfect targeting is achieved when the rate of mistargeting is equal to 1.0, implying that both surveys classify the same number of individuals as poor. Perfect targeting occurs when the errors of inclu- sion and exclusion are both close to 0. Let £*M(£/, EH) be the rate of mistargeting expressed as a function of the error of inclusion (e;) and the error of exclusion (en). This rate is a number between 0 and n, where n < °°, that is, 0 < £KM(EJ, %) < n. When mistargeting results largely from the error of inclusion, CRM^J. eH) > 1. When the error of exclusion is much larger than the error of inclusion, the rate of mistargeting is confined between 0 and 1, that is, 0 < £,RM(£I> 6//) < !• When estimates of aggregate per capita household expenditures from the light monitoring survey are used as the basis for constructing poverty maps, Rural Coastal and Savannah remain the poorest regions in Ghana. The magnitudes of the differences across regions indicated by the two surveys vary significantly, however. In GLSS 3 urban expenditures exceed rural expenditures by just 4 per- 7. The welfare indexes are selected from the Pa class of poverty indexes (Foster, Greer, and Thorbedce 1984), which measures different dimensions of poverty depending on the value of a. When a = 0, the P a represents the headcount index; when a « 1, it measures the poverty gap index. The indexes provide estimates of the severity of poverty when a > 1.  Table 2. Indexes of Extreme Poverty and Rate of Mistargeting across Regions Light monitoring survey Integrated Suwey Light monitoring survey Agroclimatic Headcount Poverty Severity Headcount Poverty Severity Type I error Type II error Rate of region index gap index index index gap index index probability probability mistargeting Accra 0.0678 0.0274 0.0158 0.1429 0.0259 0.0072 0.0369 0.1121 0.47 Other urban 0.0865 0.0325 0.0180 0.1657 0.0350 0.0118 0.0423 0.1214 0.52 8 Rural Forest 0.5543 0.3029 0.2060 0.1673 0.0324 0.0097 0.3978 0.0108 3.31 Rural Coastal 0.6731 0.3722 0.2493 0.1887 0.0409 0.0135 0.4889 0.0045 3.57 Savannah 0.6810 0.4236 0.3082 0.2471 0.0546 0.0173 0.4752 0.0414 2.76 All urban 0.0819 0.0313 0.0175 0.1600 0.0327 0.0106 0.0410 0.1191 0.51 All rural 0.6506 03753 0.2604 0.2043 0.0438 0.0141 0.4648 0.0186 3.18 National 0.4621 0.2612 0.1799 0.1897 0.0401 0.0129 0.3243 0.0519 2.44 Source: Author'* calculations based on Ghana Statistical Service* (1995,1996). 206 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 centage points. In contrast, the light monitoring survey indicates a difference of 55 percent. Similarly, the light monitoring survey classifies more than 68 percent of the population of the Savannah region as extremely poor (falling below the lower poverty line), while the Integrated Survey classifies just 25 percent of the population as poor. The same variations are observed in other regions, with pov- erty overestimated in rural areas and underestimated in urban areas. A poverty map constructed from the GLSS 3 data has a higher urban headcount than a poverty map constructed from light monitoring survey data, partly be- cause the GLSS 3 accounts for consumption of own production in rural areas, which reduces the urban-rural bias and increases national mean per capita ex- penditures (and therefore the extreme poverty line). Thus the light monitoring survey underestimates the scope of urban poverty. However, in the comprehen- sive Integrated Survey the headcount index is much higher in urban areas, and targeting may be justified as well. Poverty in Sub-Saharan Africa is generally much higher in rural areas, where opportunities for income generation are much more limited than in urban areas. Extreme differences such as those revealed by the light monitoring survey—which classifies more than 65 percent of the rural population and less than 8 percent of the urban population as Living in extreme poverty—are nevertheless unexpected. These large urban-rural differences reflect differences in the error of inclusion, which is much higher in rural areas than in urban areas (0.47 compared with 0.04). The size of the error is directly proportional to the rate of misclassification; the larger is the error of inclusion, the higher is the rate of misclassification. In urban areas the error of inclusion is low, and the error of exclusion is much higher (0.11). Mistargeting results from large errors of exclusion, which occur be- cause poor households are underrepresented in the sample of intended beneficia- ries. Large errors of inclusion in rural areas result from underestimating total house- hold consumption and overrepresenting the poor population in the approximated light monitoring surveys, which causes nonpoor households to be surveyed as in- tended beneficiaries for targeted interventions. Mistargeting is relatively high un- der the light monitoring survey design, in which the population identified for the targeted intervention is nearly two and a half times larger than the true population estimate (table 2). This high rate of overall mistargeting is inflated by the rural rate of misclassification, which is more than three times the true number of intended beneficiaries. The variations in the rates of mistargeting across other rural regions (Rural Forest, Rural Coastal, and Savannah) are not significant. The amount of leakage is directly proportional to the rate of misclassification. It will be lower in urban areas, where differential rates are smaller. The number of extremely poor people mistargeted by the light monitoring survey is about half the true targeted population in urban areas; in rural areas the number of misclassified people is three times the number of primary beneficiaries. The dol- lar amount of leakage that occurs as a result of poor targeting by the light moni- toring survey is more than three times the amount required to alleviate extreme poverty in rural Ghana (table 2).  Fofack 207 The estimated cost of eradicating extreme poverty (nzpi) is proportional to the poverty gap, where px is the estimate of the poverty gap. The poverty gap esti- mated from the light monitoring survey is about six times higher than the GLSS estimate at the national level. As a result, more than six times as much money would be needed to eradicate extreme poverty if the light monitoring survey were used as the basis for poverty analysis (27,998 cedis per capita per year compared with 4,298.5 cedis).8 The potential costs to the central government and local authorities are considerable, because poor targeting and improper identification of intended beneficiaries increase the amount of leakage and the amount of re- sources allocated for poverty alleviation. n. PREDICTING CONSUMPTION EXPENDITURES TO IMPROVE TARGETING The accuracy of light monitoring survey data can be improved by using pov- erty predictors, correlates of expenditures that are used to impute household consumption. Fofack (1997) proposes a methodology for deriving national pov- erty predictors that could be used to improve targeting. Poverty predictors and their corresponding weights are estimated from Integrated Surveys and used to predict total household expenditures. These predicted values then serve as the basis for conducting poverty analysis and for constructing poverty maps from light monitoring survey data. The prediction error is low, and high rates of suc- cessful classification are achieved. Moreover, a test of stability shows that the poverty predictors and their corresponding weights are stable over time.9 Recently, attempts have been made to exploit the wide coverage of population censuses by combining them with household surveys (Hentschel and others 1998). Although combining these tools is appealing—especially given the scope for geo- graphical targeting—the data requirements for capturing the large proportional variance in welfare could be enormous. The method proposed by Hentschel and others uses a large number of regressors from the census to predict welfare. Here, I take a different approach, based on data reduction. The poverty correlates used to predict welfare are reduced to a set of minimum core variables that can be easily collected with minimal measurement error. To model household consumption for poverty analysis, the best correlates of welfare are derived from the GLSS 3 survey using correlation analysis and regres- sion models. The model assumes that the conditional expectation of the response, given the covariates, £(ylxlv..pt:A), is related to the linear predictors by the re- 8. This is a hypothetical scenario included for illustrative purposes only. Direct income transfers are not cornerstones of policy in Sub-Saharan Africa. The cost of such measures would be enormous and would worsen fiscal deficits. 9. The stability of the poverty predictors and the corresponding weights estimated from GLSS 3 (1992) are assessed by applying the regressors from GLSS 3 to GLSS 1 (1987) and GLSS 2 (1989) to predict the level of welfare in those years. Success rates of ranking households in the same expenditure quintiles are as high as 95 percent, despite the time lag between these surveys. For details on the implementation of the test of stability of these regressors, see Fofack (1997). 208 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 spouse link function h(x,Q). Because the variance of total, household expenditures across and within regions is large, a logarithmic transformation is applied to the response to make the relationship between y and the x's linear. This transforma- tion stabilizes the error variance, reduces asymmetry in the distribution of error terms, and improves the prediction. The structural form of the correcting model is specified by equation 4: (4) Y = X'p + e where Y is total household expenditures transformed to the log scale, X is the model matrix containing the vector of regressors, P is the vector of estimated parameters relative to continuous and discrete-level variables, and e are the er- ror terms, which are distributed as N ^ o 2 ) . The poverty predictors are predomi- nantly discrete-level variables. Most continuous variables with strong predictive capabilities are dichotomized to discriminate between poor and nonpoor house- holds. These dummy regressors are constructed- and included in the model to capture the effects of qualitative independent variables. In order to account for selection bias in choosing predictor variables, I use a conditional maximum likelihood estimation method to select predictors. Unlike other selection criteria, the conditional maximum likelihood method is based on the expected overall discrepancy. Because the omission bias in the fixed model becomes an additional residual variation, the method produces an unbiased esti- mator of the discrepancy.10 The best poverty predictors are those that signifi- cantly increase the explanatory power of the model. That is, if (xi,X2>—>*/) is the initial set of poverty correlates and * ;+1 , Xy+t for k * 1 are potential poverty pre- dictors, the variable Xy+1 will be selected over x ; + k if (5) Ify - E(y I xi,xz,...,xj,xj+1)f < I[y,- - £(y I x1,x2,...,xj,xj+k)f Initially, I assume that all predictor variables are available for inclusion in the model. I then proceed by elimination, using the stepwise selection method with a minimum level of significance. I remove a given independent variable from the model only when a marginal increase in the percentage variance of the response explained by the model as a result of that variable's inclusion is smaller than the marginal increase attributed to the inclusion of any other independent variable: (6) Rz ijXj I ' ^^^ ' ' I I ' /-i Applying this selection procedure to the model iteratively produces an optimal model with 10 core poverty predictors. The optimal model has very few continu- 10. Other criteria used to select the subset of predictors include the Sp and the Mallows Cp criteria. The Sf is a model selection criterion that mmimct that the response variable and the predictors are jointly normally distributed. The Cp method, due to Mallow* (1973), assumes that the predictor variables are fixed and not random. For more details on model selection see Linhart and Zucchini (1986).  Pofack 209 ous variables; those that remain are either dichotomized or discrete-level vari- ables. This characteristic is likely to reduce errors that arise because of long recall periods. Moreover, the accuracy of targeting increases because the poverty pre- dictors and the weighted coefficients are estimated from Integrated Survey data and are imputed using information collected at the household level during the administration of the light monitoring survey. The poverty indexes thus are no longer just a function of aggregate household expenditures, but also depend on the estimated regression coefficients: t =f(y,z), foik = 0,1,2, is the poverty line. The poverty predictors are derived at the national and regional levels and are used in conjunction with the corresponding weights to predict total expendi- tures. Predicted expenditures, expressed as the weighted sum of the poverty pre- dictors, are then used as the basis for constructing poverty maps, classifying re- gions for poverty analysis, and targeting. The poverty predictors are able to explain more than 65 percent of the pro- portional variance observed in the welfare measure reported. The proportional variance explained by the model is high at the national level and at the regional level when the model is calibrated to derive poverty predictors for each agroclimatic region. (The appendix lists the deriyed poverty predictors at the national and regional levels. The use of different poverty predictors is intended to reflect dif- ferences in consumption patterns.) l HI. EMPIRICAL RESULTS To assess the accuracy of poverty maps constructed from the improved light monitoring survey—that is, the light monitoring survey in which total expendi- tures have been modeled using poverty predictors—I estimate the incidence of poverty and poverty-related indicators for different agroclimatic regions using predicted expenditures constructed from the model. I compare these estimates with the poverty indicators derived from GLSS 3 using the same poverty line for the predicted and measured consumption aggregates (since in the absence of pre- diction errors, the means of these two distributions are the same). The differences in the poverty estimates decrease substantially in both urban and rural areas when the poverty predictors are used to model household expen- ditures (table 3). The error of inclusion, for example, which was as high as 0.48 in the approximated light monitoring survey, falls to just 0.13 or less when the poverty predictors are the basis for poverty analysis. Moreover, the significant decrease in this error is not accompanied by an increase in the error of exclusion, which remains low across all agroclimatic regions.  Table 3. Indexes of Extreme Poverty and Rate of Mistargeting by Region with the Improved Light Monitoring Survey Improved light monitoring survey Integrated Survey Improved light monitoring survey Agroclimatic Headcount Poverty Severity Headcount Poverty Severity Type I error Type II error Rate of region index gap index index index gap index index probability probability mistargeting Accra 0.0878 0.0163 0.0040 0.1429 0.0259 0.0072 0.0212 0.0763 0.615 Other urban 0.1344 0.0274 0.0076 0.1657 0.0350 0.0118 0.0623 0.0936 0.811 O Rural Forest 0.0988 0.0159 0.0039 0.1673 0.0324 0.0097 0.0537 0.1220 0.591 Rural Coastal 0.1868 0.0357 0.0103 0.1887 0.0409 0.0135 0.1036 0.1056 0.990 Savannah 0.2787 0.0461 0.0116 0.2471 0.0546 0.0173 0.1261 0.0946 1.128 All urban 0.1229 0.0247 0.0067 0.1600 0.0327 0.0106 0.0522 0.0894 0.768 All rural 0.1999 0.0351 0.0094 0.2043 0.0438 0.0141 0.1008 0.1053 0.978 National 0.1743 0.0317 0.0085 0.1897 0.0401 0.0129 0.0847 0.1001 0.919 Source: Author's calculations based on Ghana Statistical Services (1995,1996).  Fofack 211 The poverty indicators estimated on the basis of the GLSS 3 data and the im- proved light monitoring survey data are similar (figure 3). At the national level the absolute relative error is less than 0.081, probably because the difference in the headcount indexes estimated from the Integrated Survey and the predicted welfare function is small. This relatively small difference is due largely to mea- surement errors in reported household expenditures and sample size effects. The sample size at the subregional level is the smallest in Accra and the Rural Forest region, where the magnitude of the difference is largest. The prediction error is inversely proportional to the sample size, however, suggesting that the error should decrease substantially in the actual light moni- toring survey, where the large sample size and exhaustive coverage allow much greater representation at the regional and subregional levels. Part of the predic- tion error also could be attributed to measurement errors in reported household expenditures. This part of the prediction error is associated with the survey de- sign and implementation and could be large depending on sample size and nonsampling errors. In contrast, measurement errors are washed away by the instrumented variable; targeting might even become more accurate when based on predicted welfare. A potential increase in accuracy is another reason for using instrumented consumption in poverty analysis. Figure 3. Headcount Index Headcount index 0.8 0.7 0.6 1 0.5 1 1 1 0.4 0.3 0.2 0.1 ll 1 ll"l II Accra Other Rural Rural Savannah All urban All rural National urban Forest Coastal I Integrated Survey Q Improved light Monitoring Survey light Monitoring Survey Source: Author's calculations. 212 THE WORLD RANK ECONOMIC REVIEW, VOL. 14, NO. 1 The differences between headcount indexes remain when predicted expendi- tures are used for poverty analysis. The bias toward higher rural poverty is pre- served in the regional ranking, and the ranking of the five agroclimatic regions is consistent across the two ranking criteria (Integrated Survey data and improved light monitoring survey data with predicted expenditures). The magnitude of the differences across agroclimatic regions is also preserved in the improved light monitoring survey. The incidence of poverty is highest in the Savannah and Rural Coastal re- gions, where 28 and 19 percent of the population, respectively, are extremely poor. The largest poverty gap is in Savannah, suggesting that if the population were distributed uniformly across regions, the volume of transfers needed to eradicate poverty would be largest in this region. The Savannah and Rural Coastal regions also have the lowest literacy and enrollment rates, and both depend on food crops as their main source of income (Ghana Statistical Ser- vices 1995). In contrast, in the Rural Forest region, where income sources are more diversified, the incidence of poverty is much lower. The poorest of the poor are thus uneducated small farmers residing mainly in the Savannah and Rural Coastal regions, where employment opportunities are Limited. To the extent that education variables (public school enrollment, proportion of school- age children enrolled in school) appear to be good proxies for income in these regions, the placement of public infrastructure may be a good means of mak- ing transfers to the poor. I also assess the performance of the proposed method by the size of errors of inclusion and exclusion, as well as the rate of mistargeting. While the error of inclusion across agroclimatic regions is generally much higher in the approximated light monitoring survey dian in the improved light monitoring survey (figure 4), the error of exclusion is generally much lower (figure 5). The magnitude of the difference between the errors of exclusion in the two surveys is small, however, in part because mistargeting is due largely to high errors of inclusion. The rate of mistargeting falls substantially when the improved light monitor- ing survey is used (table 3). In fact, except in the Rural Forest region, targeting is almost perfect. Perfect targeting is achieved in all Rural and Rural Coastal areas when predicted expenditures are used as the basis for constructing poverty maps. Compared with the light monitoring survey, the gains in accuracy are significant. Even in the Rural Forest region, where the error of inclusion is relatively high (0.12) and the rate of misclassification is slightly different from unity, the gap is less than 40 percent. Moreover, the rate of mistargeting is less than 1, implying that poor targeting is due largely to a high error of exclusion. The rates of mistargeting £RM(E/J £//) are particularly high in rural areas when the approximated light monitoring survey is used to construct the poverty map. These high rates are attributable mosdy to large errors of inclusion. Significant improvement is achieved when predicted expenditures are used as the basis for targeting (in the improved light monitoring survey).  Fofack 213 Figure 4. Probability of Error of Inclusion Probabflily 0.8 0.6 Accra Rural Rural Savannah All urban All rural National Forest Coastal • Light Monitoring Survey Improved Light Monitoring Survey Source: Author's calculations IV. IMPLICATIONS FOR GEOGRAPHICAL TARGETING The method described sharpens the accuracy of poverty maps and allows policymakers to target beneficiaries at subregional levels. The approximated light monitoring survey is constructed from the more comprehensive GLSS 3 survey. The sample size corresponding to the approximated light monitoring survey is dictated by the GLSS 3 design, just as the level of disaggregation is determined by the actual GLSS 3 sample size. The geographic profile of poverty provides living standards indicators for the various agroclimatic regions. Efficiency could be improved and leakage reduced significantly if smaller geographical units could be targeted (Baker and Grosh 1994). The Priority Survey design recommends a large sample size for targeting smaller administrative units (Grootaert and Marchant 1992).11 Improving the accuracy of the welfare function by predicting household expenditures should enable researchers to exploit the large sample size to achieve geographical target- 11. The Kenya Welfare Monitoring Survey (1994) was based on a sample of 12,000 households. The Ghana Core Welfare Indicators Survey (1997) was based on a sample of 15,000 households. 214 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 Figure 5. Probability of Error of Exclusion Probability 0.16 0.12 0.08 0.04 Accra Rural Rural All urban All rural National Forest Coastal i • Light Monitoring Survey Improved Light Monitoring Survey Source: Author's calculations. ing with minimum leakage at a level of disaggregation well below the agroclimatic region. The causes and determinants of poverty, as well as the sources of large dispari- ties across agroclimatic regions, are variable. At the aggregate level differences in the potential for income-generating activities and wage inequality may constitute important factors; at the regional and district levels human capital, access indica- tors, and location of infrastructure may be more critical. Light monitoring sur- veys collect good data on access indicators (location of schools, health centers, and water supply). The relatively large sample size of such surveys may provide opportunities for geo-referencing information at subregional levels, thereby im- proving the potential for analysis beyond fixed geographical boundaries. Fur- ther, overlaying improved poverty maps atop maps of local infrastructure (schools, health clinics, hospitals, water supply facilities, and roads) may improve the un- derstanding of poverty dynamics, shed more light on the possible constraints to growth and poverty reduction, and improve priority setting, impact assessment, and policymaking.  Fofack 21S Poverty predictors, which include food and nonfood consumption variables, are strong correlates of welfare. They can serve as the basis for targeting by commodity and by welfare indicator, especially if indirect transfer schemes are used to reduce poverty.12 Targeting by indicator is based on the ability to easily identify a few key variables that are highly correlated with household income and expenditures. The use of regional poverty predictors, which are correlates of welfare, may be particularly appealing in Sub-Saharan Africa, where there are important differences in the determinants of welfare across agroclimatic regions. Taking these differences into account in selecting correlates might improve the efficiency of transfers and the allocation of public expenditures for poverty reduction.13 The poverty predictors also can serve as a basis for targeting by commodity because they include food items. This targeted scheme draws on the. differences observed in the consumption baskets of the poor and nonpoor. Its objective is to reduce the cost of commodities that are consumed heavily by the poor through targeted subsidies. Although poverty predictors are derived according to agroclimatic region rather than poverty, the methodology is flexible and could be used in multiple steps—that is, poverty predictors could first be used to predict household expenditures, and expenditures in turn could be used to differentiate between the poor and the nonpoor. A cross-sectional analysis that focuses on the variation in the consumption of the poor by agroclimatic region could be a start- ing point for investigating the causal link between variation in the depth of pov- erty and the nature of poverty correlates. Future research will have to explore the links between these correlates and poverty at the regional and district levels, and determine how a better understanding of those associations could be used to channel scarce resources to the most needy. V. CONCLUDING REMARKS Many developing countries are confronted with widespread poverty and have limited resources for poverty alleviation. To minimise leakage, policymakers must have accurate arid detailed poverty maps that allow identification of the poor at finer levels of disaggregation than the agroclimatic region. Some developing countries have used light monitoring surveys, which have large sample sizes and are less expensive to administer than other types of sur- veys, as the basis for constructing disaggregated poverty maps. This study shows that the cost of mistargeting associated with the use of such surveys is significant and can outweigh the savings generated from their lower administrative costs. Light monitoring surveys underestimate aggregate expenditures, the basis for dif- 12. Other methods of targeting include targeting by income and self-targeting. For a survey of targeting methods with applications to developing countries, see Glewwe (1992); Kanbur, Keen, and Tuomala (1994); Bigman and Fofack (forthcoming). 13. Glewwe (1992) uses bousing indicators as the basis for targeting in Cdte d'lvoire. Ravallion (1989) uses landownership as the basis for designing land-contingent transfers for poverty alleviation in Bangladesh. 216 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 ferentiating between the poor and the nonpoor. Moreover, underestimation is not uniform across regions. As a result, welfare indicators and poverty maps derived from light monitoring surveys may not always be consistent with the actual distribution of poverty. This article shows that combining more detailed surveys, which have compre- hensive income and expenditure data, with light monitoring surveys yields im- proved poverty maps that are disaggregated at a level below the agroclimatic region. The rate of mistargeting is reduced substantially when poverty predictors derived from more comprehensive surveys are used to model total expenditures based on data from light monitoring surveys. These poverty predictors are household-level variables that are available in both Integrated Surveys and light monitoring surveys. Over the past few years demand for poverty maps that are disaggregated at a level as low as the district has been growing in developing countries, particularly in Sub-Saharan Africa. This demand has been prompted both by the need for a more accurate picture of the geographic distribution of poverty and by the move toward decentralization, which increasingly channels resources to communities. As the demand for more disaggregated information continues to grow and bud- getary and resource constraints tighten, methods that optimize the use of light monitoring surveys while improving the accuracy of targeting will be in increas- ing demand. The method proposed here recommends that light monitoring sur- veys and more detailed comprehensive surveys be combined to improve poverty mapping and geographical targeting. The accuracy of household welfare predicted from the model depends on the base point of the prediction, the stability of the poverty predictors, and their corresponding weights. Although modeling consumption significantly reduces errors of inclusion and exclusion, the level of targeting attained in the various agroclimatic regions is imperfect because of unavoidable prediction errors. These errors are attributable largely to measurement errors in reported expenditures and might be washed away when the instrumented variable is used. The stability of the predictors over time is another important question. Fofack (1997) assesses stability using surveys conducted under the same sampling frame. It would be worth investigating how this stability is affected by variation in the sampling frame. The possible implications for poverty mapping are also worth examining. Finally, efficiency in the transfer and allocation of resources could be improved by combining geographical targeting with another form of targeting, such as targeting by commodity or by indicator. Poverty prediaors are correlates of expenditures and could serve as a vector of transfers if the dynamic between these correlates and poverty were better understood. Appendix. National and Regional Poverty Predictors National level Urban areas Rural areas Expenditures on soap Expenditures on soap Expenditures on soap Number of spouses Number of spouses Number of spouses Asset score Asset score Asset score Percentage of school-age children enrolled Percentage of school-age children enrolled Percentage of school-age children enrolled in school in school in school Expenditures on meat Expenditures on meat Expenditures on meat Ownership of land Ownership of land Ownership of land Consumption of bread Percentage of household members employed Ownership of goats and sheep Ownership of poultry Use of toothpaste Number of household members per room Export crops Percentage of children enrolled in public school Ownership of farm Number of household members per room Percentage of literate household members Ownership of cattle Accra region Other urban Rural Forest Asset score Expenditures on soap Expenditures on soap Expenditures on meat Number of household members employed Consumption of bread Percentage of household members employed Expenditures on meat Asset score Expenditures on rice Asset score Use of toothpaste Number of household members who Percentage of school-age children Number of spouses completed secondary school Expenditures on bread Expenditures on meat Expenditures on soap Percentage of household members who Percentage of household members who completed Percentage of school-age children completed secondary school secondary school Number of children under five Use of toothpaste Expenditures on rice Use of toilet paper Ownership of land Number of household members per room Percentage or children enrolled in public school Use of toilet paper Use of toilet paper Rural Coastal Savannah Expenditures on soap Expenditures on soap Asset score Number of spouses Percentage of school-age children Consumption of bread Expenditures on meat Percentage of school-age children Number of spouses Ownership of sheep and goats Percentage of children enrolled in public school Use of toothpaste Consumption of bread Expenditures on meat Use of toilet paper Asset score Ownership of poultry Use of toilet paper Number of children under five Gender of head 218 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 1 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Baker, Judy L., and Margaret Grosh. 1994. "Measuring the Effects of Geographical Tar- geting on Poverty Reduction." Living Standards Measurement Study Working Paper 99, World Bank, Washington, D.C. Bigman, David, and Hippolyte Fofack. Forthcoming. "An Overview of Targeted Schemes for Poverty Reduction in Developing Countries." In David Bigman and Hippolyte Fofack, eds., Geographical Targeting for Poverty Alleviation: Methodology and Ap- plications. Washington, D.C: World Bank. Boateng, E. Oti, Kodwo Ewusi, Ravi Kanbur, and Andrew D. McKay. 1990. A Poverty Profile for Ghana, 1987-1988. Social Dimensions of Adjustment in Sub-Saharan Af- rica Working Paper 5. Washington, D.C: World Bank. Deaton, Angus, and Margaret Grosh. Forthcoming. "Consumption." In Margaret Grosh and Paul Glewwe, eds., Designing Household Survey Questionnaires: Lessons from Ten Years of LSMS Experience for Developing Countries. Washington, D.C: World Bank. Deaton, Angus, and John Muellbauer. 1980. Economics of Consumer Behavior. Cam- bridge, U.K.: Cambridge University Press. Deininger, Klaus, and Lyn Squire. 1996. "A New Data Set Measuring Income and In- equality." The World Bank Economic Review 10(September):565-91. Delaine, Ghislaine, Lionel Demery, Jean-Luc Dubois, Branko Grdjic, Christiaan Grootaert, Christopher Hill, Timothy Marchant, Andrew -McKay, Jeffrey Round, and Christo- pher Scott, eds. 1992. The Social Dimensions of Adjustment Integrated Survey: A Survey to Measure Poverty and Understand the Effects of Policy Change on House- holds. Social Dimensions of Adjustment in Sub-Saharan Africa Working Paper 14. Washington, D.C: World Bank. Demery, Lionel, Christiaan Grootaert, and Christopher Hill. 1991. "Annotated Ques- tionnaire and Listing Forms." In Timothy J. Marchant and Christiaan Grootaert, eds., The Social Dimensions of Adjustment Priority Survey: An Instrument for the Rapid Identification and Monitoring of Policy Target Groups. Social Dimensions of Adjust- ment in Sub-Saharan Africa Working Paper 12. Washington, D.C: World Bank. Demery, Lionel, Jean-Luc Dubois, Christiaan Grootaert, and Tim Marchant. 1992. "An- notated Integrated Survey Questionnaire." In Ghislaine Delaine, Lionel Demery, Jean- Luc Dubois, Branko Grdjic, Christiaan Grootaen, Christopher Hill, Timothy Marchant, Andrew McKay, Jeffrey Round, and Christopher Scott, eds., The Social Dimensions of Adjustment Integrated Survey: A Survey to Measure Poverty and Understand the Ef- fects of Policy Change on Households. Social Dimensions of Adjustment in Sub- Saharan Africa Working Paper 14. Washington, D.C: World Bank. Fofack, Hippolyte. 1997. "Using Poverty Predictors as Expenditure Proxies for Ranking Households for Poverty Analysis." Proceedings of the International Statistical Insti- tute 1997. Ankara: State Institute of Statistics. Foster, James E., Joel Greer, and Erik Thorbecke. 1984. "A Class of Decomposable Pov- erty Measures." Econometrica 52(3):173-77. Ghana Statistical Services. 1995. "Ghana Living Standards Survey Report on the Third Round (GLSS 3)." Accra.  Fofack 219 . 1996. "Measuring Household Income and Expenditure in the Third Round of the Ghana Living Standards Survey (1991/92): A Methodological Guide." Accra. Glewwe, Paul. 1992. "Targeting Assistance to the Poor: Efficient Allocation of Transfers When Household Income Is Not Observed." Journal of Development Economics 38(2):297-321. Grootaert, Christiaan, and Timothy J. Marchant. 1992. "SDA Socioeconomic Informa- tion System." In Ghislaine Delaine, Lionel Demery, Jean-Luc Dubois, Branko Grdjic, Christiaan Grootaen, Christopher Hill, Timothy Marchant, Andrew McKay, Jeffrey Round, and Christopher Scott, eds., The Social Dimensions of Adjustment Integrated Surveys: A Survey to Measure Poverty and Understand the Effects of Policy Change on Households. Social Dimensions of Adjustment in Sub-Saharan Africa Working Pa- per 14. Washington, D.C.: World Bank. Grosh, Margaret, and Paul Glewwe. 1998. "Data Watch: The World Bank's Living Standards Measurement Household Surveys." journal of Economic Perspectives 12(l):187-96. Grosh, Margaret, and Juan Mufioz. 1996. A Manual for Planning and Implementing the Living Standards Measurement Study Survey. Washington, D.C.: World Bank. Hentschel, Jesko, Jean Olson Lanjouw, Peter Lanjouw, and Javier Poggi. 1998. "Com- bining Census and Survey Data to Study Spatial Dimensions of Poverty." Policy Re- search Working Paper 1928. Development Research Group and the Poverty Reduc- tion and Economic Management Network, World Bank, Washington, D.C. Processed. Kanbur, Ravi, Michael Keen, and Matti Tuomala. 1994. "Labor Supply and Targeting in Poverty Alleviation Programs." The World Bank Economic Review 8(2):191-211. linhart, Heinz, and Walter Zucchini. 1986. Model Selection. Wiley Series in Probability and Mathematical Statistics. New York: John Wiley. Mallows, C. L. 1973. "Some Comments on Cp." Technometrics 15(4):661-75. Marchant, Timothy, and Christiaan Grootaert, eds. 1991. The Social Dimensions of Adjustment Priority Survey: An Instrument for the Rapid Identification and Monitor- ing of Policy Target Groups. Social Dimensions of Adjustment in Sub-Saharan Africa Working Paper Working Paper 12. Washington, D.C: World Bank. Narayan, Deepa, and Lyra Srinivasan. 1994. Participatory Development Tool Kit: Mate- rials to Facilitate Community Empowerment. Washington, D.C: World Bank. Ravallion, Martin. 1989. "Land-Contingent Poverty Alleviation Scheme." World Devel- opment 17(8):1223-33. . 1996. "How Well Can Method Substitute for Data? Five Experiments in Poverty Analysis." The World Bank Research Observer 11(2):199-221. World Bank. 1997. "Status Report on Poverty in Sub-Saharan Africa 1997: Tracking the Incidence and Characteristics of Poverty." Institutional and Social Policy Department, Africa Region, World Bank, Washington, D.C. Processed.