Updating Poverty Maps without Panel Data: Evidence from Vietnam

Asian Economic Journal 2009, Vol. 23 No. 4, 397–418 397 Updating Poverty Maps without Panel Data: Evidence from Vietnam* Nguyen Viet Cuong Received 2 February 2008; Accepted 11 June 2009 A household survey and a census can be combined to estimate a poverty map for small areas. Ideally, the survey and the census should be conducted in the same year. In several empirical applications, however, survey and census years can be different, which might make poverty estimates biased. Using data from the Vietnam Household Living Standard Survey 2002 and the 1999 Population and Housing Census, the present paper produces a 2002 poverty map for Vietnam and describes the biases when the survey and census years are not coincident. It is found that poverty estimates from the poverty mapping method taking into account the time difference between the survey and the census are quite close to survey-based estimates, at least at the regional level. Keywords: poverty measurement, poverty mapping, population census, household survey, Vietnam. JEL classification codes: I31, I32, O15. doi: 10.1111/j.1467-8381.2009.02020.x I. Introduction It is surely undeniable that poverty alleviation is what economic development is all about. Most developing countries have launched numerous poverty alleviation programs. To increase the effectiveness of these programs, the poverty targeting must be improved, and disaggregated measures of poverty are an important tool for poverty targeting. A recent study by Elbers et al. (2007) shows that the impact of budget transferring on poverty is larger when geographic targeting units are smaller (e.g. districts and villages). These findings are also mentioned in other studies, including Baker and Grosh (1994) and Bigman and Fofack (2000). Household surveys are important data sources for poverty measurement. However, household surveys have small samples, which cannot be used to estimate poverty measures for small areas such as districts or communes. To solve this small sample problem, Elbers et al. (2003) propose the ‘small area estima* Development Economics Group, Mansholt Graduate School, Wageningen University, Hollandseweg 1, 6706 KN Wageningen, the Netherlands. Email: [email protected]; hanhhuong@ fpt.vn. I would like to thank Roy Van Der Weide from the World Bank, and Duong Tuan Cuong and Tran Ngoc Truong from the Institute of Labor Science and Social Affairs, Ministry of Labor, Invalid and Social Affairs, Vietnam for their helpful comments on and suggestions regarding the present study. I am also grateful for helpful comments provided by anonymous reviewers of the Asian Economic Journal. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. ASIAN ECONOMIC JOURNAL 398 tion’ method to estimate poverty measures for small areas by combining a household survey and a census. The method estimates a functional relation between expenditure and household characteristics using a household survey, and then applies this estimated relation to a census to predict expenditure and welfare for small areas such as communes and districts. This method is widely applied to estimate disaggregated poverty and welfare measures, especially in developing countries (see Bigman and Fofack (2000) and Bedi et al. (2007) for review of the applications). Ideally, the survey and the census should be conducted in the same year. However, in empirical studies, the years of household surveys and population censuses can be different. For example, Kotzeva and Tzvetkov (2006) estimate poverty for Bulgaria using a 2001 census and a 2003 household survey. Another example is a study on Ecuador, in which the first poverty mapping combined a household survey in 1994 with a census in 1990, and the second poverty mapping combined a survey in 1999 and a census in 2001 (see World Bank, 1996, 2004). A poverty map for Morocco was produced using a 1998 household survey and a 1994 population census (see Litvack, 2007). If the years of a household survey and a population census are not close, it might not be possible to apply the expenditure equation that is estimated from the household survey into the population census. In other words, if the regression coefficients of the expenditure equation are not constant between the years of the household survey and population census, the poverty estimates cannot be interpreted for the census year or the survey year. The estimates might not be unbiased estimates of the poverty rate at any time.1 In Vietnam, the government has set poverty reduction as a major development goal. A huge amount of money has been spent on assistance programs targeted at the poor. In the period 2006–2010, the government plans to spend VND44 855 000bn (approximately US$2.8bn) on its poverty alleviation program.2 Therefore, a poverty map is a helpful tool for the government in the process of allocating funds for poverty reduction programs. Up to now, there have been at least two poverty maps that have been estimated using the small area estimation technique. The first one combined the Vietnam Living Standard Survey (VLSS) in 1993 and the Agricultural Census in 1994 to estimate poverty in rural areas of Vietnam (see Minot, 2000). The second one was constructed based on a VLSS in 1998 and a 33-percent sample of the 1999 Population and Housing Census (see Minot et al., 2003). In addition, in analyzing the effects of Vietnam’s access to the WTO on poverty, Fujii and Roland-Holst (2008) also apply the small area estimation method to provincial poverty rates using the 1998 VLSS and a 33-percent sample of the 1999 Population and Housing Census. The present paper has two main objectives. This first objective is to construct an updated 2002 poverty map for Vietnam using the 2002 Vietnam Household Living Standard Survey (VHLSS) and a 33-percent sample of the 1999 Population and 1. This issue will be discussed further in Section III (Methodology). 2. In January 2008, US1$ was approximately equivalent to VND16 000. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. UPDATING POVERTY A MAP WITHOUT PANEL DATA 399 Housing Census. Poverty maps are often updated using new population censuses and household surveys. The results using this method can be limited, because population censuses are conducted once every 10 years in most countries. Another approach is to use panel data from household surveys to update poverty maps for non-census years (e.g. see Emwanu et al., 2006).3 In the present paper, the poverty map is updated without panel data by using commune and household variables, which are the same or unchanged between the census and the survey. These variables are used to construct expenditure models. The second objective of the present paper is to show that poverty mapping without taking into account the time difference between the survey and census can result in high biases of poverty estimates. The research is structured into 5 sections as follows. Section II describes the data sources used in this study. Section III presents the method of small area estimation applied in Elbers et al. (2003) and illustrates the poverty estimation problem when the survey and census years are not close. Section IV presents empirical results, and some conclusions are drawn in Section V. II. Data Sources The present study relies on two data sources to estimate poverty rates at province, district and commune levels. The first is the 2002 VHLSS. This survey was conducted by the General Statistical Office of Vietnam (GSO) with technical support from the World Bank in Vietnam. The 2002 VHLSS covered 29 412 households, of which 6,876 households were located in urban areas, and 22 536 in rural areas. The basic sample frame of this survey was obtained from the 1999 Population and Housing Census. The selection of the sample followed a method of stratified random cluster sampling so that the sample was representative for national, rural, urban and regional levels. The 2002 VHLSS collected detailed information on household characteristics, including basic demography, employment and labor force participation, education, health, income, expenditure, housing, fixed assets and durable goods, and households’ participation in poverty alleviation programs. Expenditure per capita was collected using very detailed questionnaires. Expenditure includes food and non-food expenditure. Food expenditure includes purchased food and foodstuff and self-produced products of households. Non-food expenditure comprises expenditure on education, health-care expenditure, expenditure on houses and commodities, and expenditure on power, water supply and garbage. The second dataset is the 33-percent sample of the 1999 Population and Housing Census. This census was also carried out by the GSO with technical 3. This updating method requires panel data from household surveys. Suppose that panel data from two household surveys at time t1 and t2 are available, but population census data are available only at time t1. Expenditure at time t2 is estimated as a function of household variables at time t1 using panel data. Then, this estimated function is applied to the census to predict expenditure and welfare measures of census households at time t2. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 400 ASIAN ECONOMIC JOURNAL support from the United Nations Family Planning Agency and the United Nations Development Program. Data collected in the census include basic household demography, education, main occupation, housing characteristics and basic assets such as television and radio. The full census data are not made available by the GSO, and this study uses a 33-percent sample of the census. This 33-percent sample was selected by the GSO using systematic sampling of every third household on the list of households organized by administrative unit. The number of households in this sample is 5 553 811. III. Methodology Although the computation of poverty estimates according to the poverty mapping method of Elbers et al. (2003) is rather complicated, the idea of the method is straightforward. The method can be described in two main steps. In this first step, an expenditure equation is estimated using a household survey: ln ( yi ) = X iβ + ε i (1) where yi and Xi are expenditure and characteristics of household i in the survey, respectively; ei is a random disturbance term distributed as N(0,s2). Xi can be commune and household variables, and must also be available in the census. It should be noted that Xi can be commune or cluster variables, which are computed from the census and merged with the household survey. It should be noted that Equation (1) presents a simplified version of the model of Elbers et al. (2002) and Elbers et al. (2003). For simplicity, Equation (1) is used to illustrate the potential problem in poverty estimation when the survey and census years are not close. To estimate the poverty indexes, we will use the full model proposed by Elbers et al. (2002) and Elbers et al. (2003). In the full model, the error term is decomposed into cluster and idiosyncratic errors to capture the spatial autocorrelation of errors of households within a cluster, and the error terms are modeled to allow for heteroskedasticity. The more detailed model of Elbers et al. (2002) and Elbers et al. (2003) is presented in Appendix I. In the second step, the probability of being poor is predicted for all households in the census using the parameter estimates from Equation (1) and the X variables of these census households. Then, the predicted probabilities are averaged to estimate the poverty rate of a small area such as a commune or a district: 1 Hˆ = N N ⎡ ln z − X iβˆ ⎤ ⎥, σˆ ⎦ ⎣ ∑Φ⎢ i =1 (2) where N is the number of households in the small area, z is the poverty line, and F is the cumulative standard normal function. Ĥ is an unbiased and consistent estimator of the poverty rate, H.4 4. Instead of Equation (2), the poverty rate can be estimated using predicted expenditures. This method is presented in Appendix I. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 401 UPDATING POVERTY A MAP WITHOUT PANEL DATA Ideally, the survey and the census years should be the same or very close, so that there is no problem in interpreting poverty estimates. The condition that the X variables are the same for households in the census and the survey is also mentioned as an assumption on ‘measurement of predictors’ by Tarozzi and Deaton (2009).5 However, if the year of the survey is far from that of the census, it is not clear which year the poverty rate is estimated for. For example, in the present study we have the census at time t1, but the survey at time t2 (t2 > t1). Then, the expenditure equation is rewritten as: ln ( yi 2 ) = X i 2β 2 + ε i 2 , (3) where subscript 2 reflects time t2. If we use the parameter estimates in Equation (3) for poverty estimation, the poverty estimator in Equation (2) is expressed as follows: 1 Hˆ = N N ⎡ ln z − X i1βˆ 2 ⎤ ⎥. σˆ ⎣ ⎦ ∑Φ⎢ i =1 (4) It can be rewritten as: 1 N ⎡ ln z − X i 2βˆ 2 ( X i 2 − X i1 ) βˆ 2 ⎤ + Hˆ = ∑ Φ ⎢ ⎥ σˆ σˆ N i =1 ⎣ ⎦, = Hˆ 2 + R2 X i1 , X i 2 , βˆ 2 , σˆ , ( ) (5) where Ĥ2 is the unbiased estimator of the poverty rate at time t2 (denoted by Ĥ ), and R2 is the difference between Ĥ and Ĥ2, which is a function of Xi1, Xi2, β̂ 2 and σ̂.6 Taking expectation of Ĥ, we can get: E ( Hˆ ) = H 2 + E ⎡⎣ R2 X i1 , X i 2 , βˆ 2 , σˆ ⎤⎦ (6) ( ) Similarly, Equation (4) can be expressed as follows: ( ) ⎡ ln z − X βˆ X i1 βˆ 2 − βˆ 1 ⎤ i1 1 ⎥ ⎢ Φ + ∑ σˆ σˆ ⎥⎦ ⎢⎣ i =1 ˆ ˆ ˆ = H1 + R1 ( X i1 , X i 2 , β 2 , σ ), (7) ( (8) 1 Hˆ = N N and: ) E ( Hˆ ) = H1 + E ⎡⎣ R1 X i1 , X i 2 , βˆ 2 , σˆ ⎤⎦ , 5. The assumption on ‘measurement of predictors’ states that the X variables in the survey and the census should be close not only in time but also in terms of data collection methods (e.g. questionnaires, interview tools and data cleaning). In addition, Tarozzi and Deaton (2009) mention another assumption on area homogeneity in poverty mapping. This assumption implies that the expenditure function should be the same for all small areas so that it can be applied validly to the census to estimate welfare measures for these small areas.  ) can be specified using Taylor expansion. 6. A formula of R2 ( X i1 , X i 2 , βɵ 2 , σ © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. ASIAN ECONOMIC JOURNAL 402 where Ĥ1 is the unbiased estimator of the poverty rate at time t1 (denoted by Ĥ) and R1 is the difference between Ĥ and Ĥ1.7 Therefore, Equations (6) and (8) show that a poverty estimator that is based on data from a survey and a census at different time points is no longer an unbiased estimator of the poverty rate at time t1 or t2. The estimate will be close to the poverty rate at time t1 if the difference in the equation coefficients between t1 and t2 is small. In contrast, the estimate will be close to the poverty rate at time t2 if the change in the X variables during the period t1–t2 is negligible. For updating, the poverty map of interest should be for the t2 time. If the X variables in the regression are time-invariant, that is, unchanged during the period t1–t2, we will have Xi2 equal to Xi1. As a result, the bias E(R)2 will be eliminated. To update the poverty map for time t2, three conditions are required: (i) there are common variables in a census and a survey; (ii) these variables are time-invariant (i.e. unchanged during the year of the census and the year of survey or they are the same in the census and the survey (e.g. geographic information system (GIS) variables)); and (iii) these variables are correlated with expenditure so that relatively good expenditure models can be constructed. In addition to the poverty rate, two other popular measures of poverty are also estimated in this paper. These measures are the poverty gap index and the poverty severity index (see Foster et al., 1984). IV. Empirical Results This section presents the estimation of poverty indexes using the 2002 VHLSS and the 1999 Population and Housing Census. The estimation of the poverty estimates and standard errors follows the method of Elbers et al. (2002).8 IV.1 Variable selection In this study, the X variables come from three sources. This first source is commune variables that are constructed from the 1999 census. All the household variables are averaged at the commune level in the census. For example, we can calculate the percentage of households with television in all the communes. These commune means are merged with household data in the survey. The second set of explanatory variables is 1999 GIS district-level data. These data include elevation, slope, soil type, land cover, sunshine, temperature, humidity, rainfall, distance to town, length and density of road, the number and density of commune markets, and market payment to State. These variables, of course, can be merged with both the survey and the census. An advantage of the GIS and 7. s is assumed to be the same in the year t1 and t2. 8. In the present paper, we use a poverty mapping program called PovMap to estimate poverty indexes. This poverty mapping software has been developed by World Bank researchers to estimate poverty and inequality indexes using the method of Elbers et al. (2002). The software package can be downloaded at: http://iresearch.worldbank.org/PovMap/PovMap2/PovMap2Main.asp © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. UPDATING POVERTY A MAP WITHOUT PANEL DATA 403 cluster mean variables is the reduction of spatial errors in the consumption models (Elbers et al., 2002). The third set of explanatory variables is those available in both the census and survey, but they are required to be time-invariant or to have changed negligibly during 1999–2002. To select time-invariant variables, we use the 2002 and 2004 VHLSS panel data to assess changes in variables. The panel data cover 4008 households, which are representative at the national level.9 Based on the comparison analysis, the time-invariant variables selected include: household head’s age in 1999, head’s sex and ethnicity, basic education of head, head’s main occupation as manager, head’s main occupation (e.g. professional/technician), permanent house, no toilet, tap water, and at least one household member working in agriculture.10 Although the variables are selected based on the analysis of the period 2002–2004 instead of the period 1999–2002, they are expected to have remained unchanged or to have changed negligibly during 1999–2002. In addition, we generate interaction variables between all the above variables with urbanity to capture the difference in consumption models between the urban and rural areas. The total number of explanatory variables including interactions is 159.11 IV.2 Estimation of consumption models Once the explanatory variables are selected, consumption models can be estimated using data from the 2002 VHLSS. There are eight geographical regions in Vietnam. To allow for geographical heterogeneity, a consumption model is estimated for each region. In addition, to examine the sensitivity of the poverty estimates to different model specifications, for each region, we compare three different models, which vary in the number of explanatory variables included. They are regarded as large, medium or relatively small specifications. The large models use approximately 30 explanatory variables. The medium and small models have around 20 and 10 explanatory variables, respectively. In all models, only variables that are statistically significant at the 5-percent level are kept. Therefore, there are 24 expenditure regressions in total. It is found that the poverty estimates are very similar, especially for the medium and larger models. In the 9. The 2004 VHLSS was conducted by the GSO with technical support from World Bank in Vietnam. The questionnaires and survey design of the 2004 VHLSS are very similar to those of the 2002 VHLSS. The sample size of the 2004 VHLSS is 9189, which is representative at the regional level. It should be noted that we do not update the poverty map for 2004 because this period is rather long, and the time-invariant assumption might not hold. In addition, there are a number of new provinces and districts in 2004 compared to 1999. Some provinces, districts and communes are split or merged. This makes the merging of datasets very difficult. 10. Age is not time-invariant. The variable ‘age of household head’ is defined as age in 1999. We can calculate age in 1999 for the 2002 VHLSS by subtracting 3 from the age in 2002. 11. The full list of the explanatory variables is not reported in the paper, because it is very long. The list can be provided on request. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 404 ASIAN ECONOMIC JOURNAL Table 1 Estimates of poverty rate (P0): in percent Regions Red River Delta North East North West North Central Coast South Central Coast Central Highlands North East South Mekong River Delta Total VHLSS 2002 Mapping method 1 Mapping method 2 VLSS 1998 21.9 (1.0) 36.0 (1.4) 67.9 (3.4) 43.4 (1.8) 25.2 (1.7) 56.1 (3.3) 13.4 (1.0) 23.4 (1.1) 29.0 (0.5) 20.7 (0.8) 33.8 (1.0) 68.6 (1.6) 41.0 (1.1) 24.2 (1.2) 53.7 (1.3) 11.1 (0.6) 20.2 (0.8) 27.3 (0.5) 29.4 (1.1) 45.1 (0.9) 73.6 (1.4) 52.3 (1.2) 39.1 (1.4) 54.0 (1.7) 14.0 (0.8) 33.2 (1.1) 36.0 (0.4) 29.3 (3.7) 62.0 (6.2) 73.4 (8.6) 48.1 (5.2) 34.5 (6.5) 52.4 (9.7) 12.2 (3.1) 36.9 (3.0) 37.4 (1.7) Notes: Figures in parentheses are standard errors. For the estimates based on the 2002 Vietnam Household Living Standard Survey (VHLSS) and 1998 Vietnam Living Standard Survey (VLSS), the point estimates and standard errors are corrected for sampling weights and cluster correlation. The standard errors of the poverty mapping are calculated using the simulation method of Elbers et al. (2002) with 200 replications. present paper, regression results of the large models are presented in Tables A1–A4 of Appendix II.12 The adjusted R2 of the models are rather high. The highest R2 value is 0.7 for North West Vietnam. Three regions have models with R2 of approximately 0.63, and two regions have models with R2 of approximately 0.55. The lowest R2 values are 0.45 and 0.35, which are for Mekong Delta and North Central Coast, respectively. Given that the explanatory variables include a large number of cluster and GIS variables, the R2 results are very promising. IV.3 Poverty estimates The estimated consumption models are applied to the 1999 census to predict the poverty rates at different levels. Tables 1, 2 and 3 present the estimates of the poverty rate, the poverty gap index and the poverty severity index at the regional level, respectively. In these tables, ‘mapping method 1’ refers to the updating 12. We use the PovMap program to estimate poverty. Communes are specified as clusters in modeling the location effect. The poverty estimates are similar when districts are selected as clusters. The full regression results are not reported in this paper, but they can be provided on request. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 405 UPDATING POVERTY A MAP WITHOUT PANEL DATA Table 2 Estimates of poverty gap index (P1) Regions Red River Delta North East North West North Central Coast South Central Coast Central Highlands North East South Mekong River Delta Total VHLSS 2002 Mapping method 1 Mapping method 2 VLSS 1998 0.042 (0.003) 0.087 (0.087) 0.240 (0.016) 0.104 (0.006) 0.060 (0.007) 0.185 (0.016) 0.030 (0.003) 0.047 (0.003) 0.070 (0.002) 0.039 (0.002) 0.071 (0.003) 0.228 (0.008) 0.097 (0.004) 0.050 (0.004) 0.160 (0.006) 0.026 (0.002) 0.036 (0.002) 0.063 (0.001) 0.060 (0.003) 0.119 (0.004) 0.285 (0.011) 0.143 (0.006) 0.103 (0.006) 0.178 (0.009) 0.032 (0.003) 0.075 (0.004) 0.093 (0.001) 0.062 (0.012) 0.176 (0.027) 0.222 (0.019) 0.118 (0.007) 0.102 (0.034) 0.191 (0.059) 0.030 (0.013) 0.082 (0.009) 0.095 (0.007) Notes: Figures in parentheses are standard errors. For the estimates based on the 2002 Vietnam Household Living Standard Survey (VHLSS) and 1998 Vietnam Living Standard Survey (VLSS), the point estimates and standard errors are corrected for sampling weights and cluster correlation. The standard errors of the poverty mapping are calculated using the simulation method of Elbers et al. (2002) with 200 replications. method that is used in the present paper. It should be noted that we also estimate the poverty indexes using consumption models with all household variables, including time-invariant ones, to examine the biases when the difference in the years of the survey and the census are not taken into account. This way of poverty mapping is referred to as ‘mapping method 2’ in what follows. Because the 2002 VHLSS is representative at the regional level, we can examine the accuracy of the poverty estimates from the poverty mapping methods by comparing them with those based on the 2002 VHLSS.13 Table 1 shows that poverty mapping method 1 produces estimates of the poverty rate very close to those based on the 2002 VHLSS. For the poorest region, North West, the poverty estimates from the 2002 VHLSS and mapping method 1 are 68 and 69 percent, respectively. These corresponding figures are 13 and 11 percent for the richest region, North East South, respectively. Therefore, the poverty map updating using 13. This study uses the expenditure poverty line constructed by the World Bank and the GSO. This poverty line was first estimated for 1993 using the 1993 VLSS. Poverty lines in the following years are estimated by deflating the 1993 poverty line using the consumer price index. Regional price differences and monthly price changes over the survey period were taken into account when the poverty lines were calculated. The poverty lines for the years 1993, 1998 and 2002 are equal to VND1 160 000, VND1 790 000 and VND1 917 000, respectively. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 406 ASIAN ECONOMIC JOURNAL Table 3 Estimates of poverty severity index (P2) Regions Red River Delta North East North West North Central Coast South Central Coast Central Highlands North East South Mekong River Delta Total VHLSS 2002 Mapping method 1 Mapping method 2 VLSS 1998 0.012 (0.001) 0.029 (0.002) 0.104 (0.009) 0.035 (0.003) 0.022 (0.004) 0.079 (0.008) 0.011 (0.002) 0.014 (0.001) 0.024 (0.001) 0.011 (0.001) 0.023 (0.002) 0.097 (0.005) 0.033 (0.002) 0.016 (0.002) 0.068 (0.004) 0.009 (0.001) 0.011 (0.001) 0.024 (0.000) 0.018 (0.001) 0.044 (0.002) 0.133 (0.008) 0.054 (0.003) 0.039 (0.003) 0.077 (0.005) 0.011 (0.001) 0.024 (0.002) 0.034 (0.000) 0.021 (0.005) 0.067 (0.013) 0.087 (0.010) 0.041 (0.003) 0.045 (0.019) 0.096 (0.038) 0.012 (0.006) 0.027 (0.004) 0.036 (0.003) Notes: Figures in parentheses are standard errors. For the estimates based on the 2002 Vietnam Household Living Standard Survey (VHLSS) and 1998 Vietnam Living Standard Survey (VLSS), the point estimates and standard errors are corrected for sampling weights and cluster correlation. The standard errors of the poverty mapping are calculated using the simulation method of Elbers et al. (2002) with 200 replications. time-invariant variables is very encouraging in the study, at least at the regional level. However, poverty mapping using both time-invariant and time-variant variables (i.e. mapping method 2) yields very different estimates. For all the regions, the point estimates are higher than those based on the 2002 VHLSS. They seem close to the poverty rate for 1999. These findings suggest that the X values changed significantly during the period 1999–2002, whereas the coefficients of X in the consumption model changed slightly during this period. The estimates tend to be close to the 1999 poverty rates rather than the 2002 poverty rates. To investigate this issue more carefully, the poverty rate for 1998 is estimated using the 1998 VLSS. For the whole country, the poverty estimates from mapping method 2 and the 1998 VLSS are 36 and 37 percent, respectively. For some regions, the poverty estimates from mapping method 2 are very similar to the 1998 poverty estimates. However, for four regions, including North Central Coast, South Central Coast, Central Highlands and North East South, the point estimates from mapping method 2 are even higher than those based on the 1998 VLSS. Therefore, combination of a household survey and a census in different years can result in point estimates of poverty that do not lie between the poverty estimates of the survey and the census years. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. UPDATING POVERTY A MAP WITHOUT PANEL DATA 407 The estimates of the poverty gap and severity indexes using mapping method 1 are rather similar to those based on the 2002 VHLSS (Tables 2 and 3). Mapping method 2, again, produces higher point estimates of poverty indexes than the 2002 VHLSS. The poverty estimates at the provincial level are presented in Table 4. There were 61 provinces in Vietnam in 2002. Lai Chau Province has the highest point estimate poverty rate, at 80 percent. This province belongs to North West region. Ho Chi Minh City, which is the richest province, has a point estimate poverty rate of 1 percent. The map of the poverty rate estimates is presented in Figure A1 in Appendix II. Although the 2002 VHLSS is not sampled to be representative at the provincial level, its sample size is rather large and can be used to estimate the poverty rate at the province level to examine the accuracy of the estimates from the poverty mapping. Figure 1 is a scatter graph of the poverty rate estimates based on the 2002 VHLSS and on mapping methods 1 and 2 for all provinces. It shows that mapping method 1 produces estimates relatively close to those based on the 2002 VHLSS, whereas mapping method 2 tends to yield higher estimates than the 2002 VHLSS. In Figure 2, we examine whether the point estimates from the poverty mapping methods lie within the 95% confidence interval of the poverty estimates based on the 2002 VHLSS. The provinces are ordered from the poorest according to the poverty estimates from the 2002 VHLSS. Figure 2 shows that for 52 out of 61 provinces, the estimates from mapping method 1 are within the 95-percent confidence interval of the survey-based estimates. However, the hypothesis on equality of poverty estimates from the mapping and those based on the 2002 VHLSS is not rejected at the 5-percent significance level for 60 provinces. For mapping method 2, the number of provinces with poverty estimates inside the 95-percent confidence interval of the survey-based estimates is only 26 (out of 61 provinces). The hypothesis on equality of poverty estimates between mapping method 2 and the 2002 VHLSS is not rejected at the 5-percent significance level for only 35 provinces. Finally, the estimates of district poverty rates are mapped in Figure A2 in Appendix II. The variation in district poverty rates within provinces seems relatively high, especially for provinces in central regions (i.e. North Central Coast, South Central Coast and Central Highlands). V. Conclusion The paper combines the 1999 Population and Housing Census and the 2002 VHLSS to estimate poverty measures for 2002 using the poverty mapping method of Elbers et al. (2003). Because the survey and census years are different, the selection of the explanatory variables in consumption should be paid special attention. To produce unbiased estimates, the explanatory variables should be unchanged or have changed negligibly during the period 1999–2002, or be from © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. Province code Ha Noi Hai Phong Ha Tay Hai Duong Hung Yen Ha Nam Nam Dinh Thai Binh Ninh Binh Ha Giang Cao Bang Lao Cai Bac Kan Lang Son Tuyen Quang Yen Bai Thai Nguyen Phu Tho Vinh Phuc Bac Giang Bac Ninh Quang Ninh Lai Chau Son La Hoa Binh Thanh Hoa Nghe An Ha Tinh Quang Binh Quang tri Poverty rate (P0) (%) Poverty gap (P1) Poverty severity (P2) Estimate Standard error Estimate Standard error Estimate Standard error 4.4 14.8 25.8 16.1 19.1 29.2 29.5 32.0 29.3 62.1 45.4 52.6 57.5 42.4 36.6 45.1 28.9 32.5 26.5 25.5 19.9 13.3 80.3 67.4 63.8 43.9 43.6 42.8 35.5 36.2 0.6 1.0 1.4 1.3 1.6 2.2 1.6 1.9 1.7 3.6 3.0 2.3 3.2 2.5 1.9 1.7 1.8 1.9 2.1 2.0 2.4 1.2 1.5 2.5 1.8 1.8 1.6 1.8 2.1 2.4 0.0073 0.0263 0.0483 0.0265 0.0322 0.0550 0.0581 0.0628 0.0579 0.1611 0.1111 0.1456 0.1577 0.1006 0.0824 0.1183 0.0605 0.0603 0.0483 0.0471 0.0324 0.0248 0.3268 0.2030 0.2066 0.1037 0.1067 0.0979 0.0804 0.0899 0.0012 0.0025 0.0039 0.0028 0.0036 0.0058 0.0046 0.0053 0.0048 0.0144 0.0121 0.0112 0.0138 0.0095 0.0065 0.0073 0.0057 0.0055 0.0054 0.0049 0.0053 0.0028 0.0143 0.0130 0.0101 0.0066 0.0059 0.0062 0.0068 0.0083 0.0019 0.0072 0.0137 0.0069 0.0085 0.0157 0.0171 0.0184 0.0172 0.0487 0.0376 0.0539 0.0583 0.0336 0.0267 0.0428 0.0187 0.0183 0.0134 0.0133 0.0082 0.0071 0.1571 0.0788 0.0854 0.0352 0.0375 0.0321 0.0265 0.0325 0.0004 0.0008 0.0014 0.0009 0.0011 0.0021 0.0017 0.0019 0.0018 0.0067 0.0054 0.0057 0.0066 0.0042 0.0028 0.0036 0.0023 0.0021 0.0019 0.0016 0.0016 0.0010 0.0109 0.0069 0.0061 0.0030 0.0028 0.0027 0.0028 0.0037 408 101 103 105 107 109 111 113 115 117 201 203 205 207 209 211 213 215 217 219 221 223 225 301 303 305 401 403 405 407 409 Province name ASIAN ECONOMIC JOURNAL © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. Table 4 Poverty estimates at the provincial level 801 803 805 807 809 811 813 815 817 819 821 823 27.3 2.0 0.0648 0.0068 0.0224 0.0029 2.8 34.2 29.1 21.4 22.7 15.4 45.0 57.3 45.8 0.6 28.0 40.9 28.7 14.6 4.0 8.1 25.9 8.6 1.1 2.3 1.9 1.6 1.9 1.3 2.9 1.3 1.7 0.2 2.4 3.4 2.3 1.9 0.7 0.8 2.0 1.0 0.0037 0.0777 0.0630 0.0362 0.0448 0.0325 0.1380 0.2192 0.1318 0.0010 0.0719 0.1152 0.0706 0.0267 0.0063 0.0145 0.0712 0.0153 0.0019 0.0073 0.0056 0.0038 0.0045 0.0032 0.0143 0.0076 0.0076 0.0004 0.0085 0.0146 0.0079 0.0043 0.0013 0.0019 0.0073 0.0024 0.0008 0.0275 0.0211 0.0097 0.0138 0.0110 0.0493 0.1038 0.0517 0.0003 0.0268 0.0453 0.0255 0.0076 0.0016 0.0041 0.0251 0.0043 0.0005 0.0033 0.0026 0.0012 0.0017 0.0013 0.0073 0.0055 0.0041 0.0001 0.0039 0.0073 0.0037 0.0015 0.0004 0.0007 0.0033 0.0008 16.2 18.4 13.0 11.9 15.4 16.3 23.1 14.4 36.0 27.8 20.2 24.9 1.6 1.8 1.6 1.2 1.4 1.4 1.8 1.5 2.3 2.0 2.3 2.5 0.0302 0.0346 0.0241 0.0207 0.0280 0.0293 0.0487 0.0264 0.0709 0.0616 0.0414 0.0344 0.0040 0.0043 0.0036 0.0027 0.0034 0.0035 0.0053 0.0034 0.0075 0.0064 0.0060 0.0062 0.0087 0.0100 0.0069 0.0056 0.0079 0.0082 0.0153 0.0075 0.0237 0.0202 0.0128 0.0101 0.0014 0.0015 0.0012 0.0009 0.0012 0.0012 0.0020 0.0012 0.0032 0.0026 0.0022 0.0022 Note: The standard errors are calculated using the simulation method of Elbers et al. (2002) with 200 replications. 409 © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 501 503 505 507 509 511 601 603 605 701 703 705 707 709 711 713 715 717 Thua Thien Hue Da Nang Quang Nam Quang Ngai Bin Dinh Phu Yen Khanh Hoa Kon Tum Gia Lai Dak Lak Ho Chi Minh Lam Dong Ninh Thuan Binh Phuoc Tay Ninh Binh Duong Dong Nai Binh Thuan Ba Ria Vung Tau Long An Dong Thap An Giang Tien Giang Vinh Long Ben Tre Kien Giang Can Tho Tra Vinh Soc Trang Bac Lieu Ca Mau UPDATING POVERTY A MAP WITHOUT PANEL DATA 411 410 ASIAN ECONOMIC JOURNAL 0 Estimates from mapping method (%) 20 40 80 60 100 Figure 1 Estimates of provincial poverty rate using the 2002 Vietnam Household Living Standard Survey (VHLSS) and mapping method 2: 䉭, P0 method 1; 䊊, P0 method 2; —, 45 degree line 0 20 40 60 Estimates from the 2002 VHLSS (%) 80 100 0 Poverty estimates (%) 20 40 60 80 Figure 2 Estimates of provincial poverty rate using poverty mapping methods and 95-percent confidence interval of estimates based on the 2002 Vietnam Household Living Standard Survey (VHLSS): 䉭, P0 method 1; 䊊, P0 method 2; —, lower and upper bounds 0 20 40 60 Provinces the same source of data, such as cluster mean variables (based on the census) and GIS variables. This mapping method also allows for updating poverty maps when a census is not available at the time of a household survey. The results show that the poverty estimates from this mapping method are rather close to the estimates based on the 2002 VHLSS, especially at the regional © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. UPDATING POVERTY A MAP WITHOUT PANEL DATA 411 level. In contrast, poverty mapping that does not take into account the difference in time points of the survey and the census produces very different estimates from the survey-based estimates. They tend to be closer to the 1999 poverty rates. This might be because the household variables changed much more than the relationship between the household variables and consumption changed during the period 1999–2002. Therefore, if we focus on the explanatory variables whose coefficients in the consumption equation are less changed or unchanged during the time, we can produce a poverty map for 1999 without a survey in that year.14 Appendix I: Small Area Estimation Method The small area estimation method of Elbers et al. (2002, 2003) can be described in two steps. The first step estimates an expenditure equation using data from a household survey. The expenditure equation is expressed as follows: ln ( yic ) = X icβ + uic + ηc , (A1) where yic and Xic are expenditure and characteristics of household i in cluster c in the survey. Unlike Equation (1), the error term is now decomposed into household idiosyncratic component, uic, and cluster component, hc, to capture a withincluster correlation of the error term. This step estimates not only the means of coefficients in (A1), but also the variances of the coefficients and idiosyncratic and cluster errors. In other words, the distribution of coefficients and idiosyncratic and cluster errors are estimated from the first step using the household survey data. In the second step, Monte Carlo simulation is applied into the census to estimate the poverty and inequality indexes. In each simulation, specific values of regression coefficients and idiosyncratic and cluster errors are randomly drawn from their distribution, which is estimated in the first step. Let β̂2 , ûics , and η̂cS denote the drawn values of the coefficients and idiosyncratic and cluster errors in the s-th simulation, respectively. Then, the predicted expenditure for a household in the census is given by: ˆ S. (A2) ln ( yˆ s ) = X βˆ s + uˆ s + η ic ic ic c Poverty and inequality indexes of a small area can be expressed as a function of household expenditures and the number of households in that area; that is, W = f ( yic , n) , (A3) where n is the number of households in a small area. Hence, after each simulation, we can obtain an estimate of the index: (A4) Wˆ s = f ( yˆ s , n) . ic 14. We do not attempt this mapping method because the 1998 VLSS is available and very close to the year of the census. The map using the 1998 VLSS and the 1999 census was conducted by Minot et al. (2003). © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 412 ASIAN ECONOMIC JOURNAL The simulated expected value for an index is the mean of S simulations: S Wˆ = ∑Wˆ s . (A5) s =1 Finally, the variance of Ŵ is calculated from the sample of S simulated values. Appendix II: Regression Results and Poverty Maps Table A1 Regression on per capita expenditure Red River Delta and North East Explanatory variables Iintercept Head age in 1999 Head age squared (in 1999) A least one household member working in agriculture Head ethnic minorities Head primary school Head post secondary school Head leaders/managers Head professionals/technicians Head clerks/service workers Permanent house % arable land Log of head age Commune ratio of head post secondary Number of markets per commune Minor road density Market payment to State % natural land % planted land Annual rainfall Commune ratio of people with primary school Commune ratio of people with lower secondary Commune ratio of head not working Spouse primary school Commune ratio of spouse lower secondary school Commune ratio of spouse upper secondary school Ratio of spouse leaders/managers Spouse professionals/technicians Spouse clerks/service workers Commune ratio of spouse not working Commune ratio of agricultural workers Commune ratio of flush toilet Have no toilet Commune ratio of having TV Tap water Length of roads Urban * Commune ratio of permanent house Urban * Main road density Urban * Commune ratio of post secondary Urban * Commune ratio of spouse post secondary Urban * Spouse clerks/service workers Urban * Commune ratio of clean water Urban * Permanent house Number of observations Adjusted R2 Red River Delta North East Coefficient Standard error Coefficient Standard error 11.817 0.1104 -0.0007 -0.1852 — -0.1972 0.1809 0.2394 0.1729 0.1357 0.1509 -0.3202 -1.8462 — 0.0538 — — — — -0.0007 1.4252 — — -0.1032 — 0.6007 0.3956 0.2079 — 0.3918 — 0.2687 -0.1189 0.6463 0.1063 — -0.5017 0.0111 3.9083 -4.4247 0.2105 0.1505 — — — 0.968 0.0183 0.0001 0.0136 — 0.0179 0.0279 0.0394 0.0291 0.0284 0.0124 0.0511 0.4166 — 0.0148 — — — — 0.0001 0.1665 — — 0.0189 — 0.1653 0.1132 0.0282 — 0.0816 — 0.0597 0.0253 0.0677 0.0247 — 0.0879 0.0024 0.3052 0.5457 0.0425 0.0372 — 5240 0.622 11.908 0.1131 -0.0007 -0.2088 -0.0644 -0.1272 0.1434 0.1668 0.2598 0.1340 0.1701 — -1.9968 -0.6134 — 0.0154 0.0000 -0.0045 0.0080 0.0001 — 1.3177 -0.7115 -0.1043 -0.6305 — — 0.1769 0.2904 — -0.2838 — -0.1914 0.2544 — 0.0000 — — — — -0.3111 — 0.1275 — — 0.905 0.0192 0.0001 0.0169 0.0163 0.0160 0.0327 0.0411 0.0327 0.0296 0.0195 — 0.4059 0.1456 — 0.0026 0.0000 0.0008 0.0010 0.0000 — 0.2152 0.1147 0.0165 0.1290 — — 0.0286 0.0427 — 0.0528 — 0.0195 0.0574 — 0.0000 — — — — 0.0616 — 0.0311 4722 0.54 Note: —, not applicable. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 413 UPDATING POVERTY A MAP WITHOUT PANEL DATA Table A2 Regression on per capita expenditure North West and North Central Coast Explanatory variables Intercept Head age Head age squared Member working in agriculture Head ethnic minorities Head primary school Head post secondary school Head leaders/managers Head professionals/technicians Head clerks/service workers Permanent house Log of head age Commune ratio of working people Annual rainfall Commune ratio of head primary school Commune ratio of head post secondary Commune ratio of head clerks/services Commune ratio of head agriculture Commune ratio of head unskilled workers Spouse primary school Commune ratio of spouse upper–secondary school Spouse professionals/technicians Spouse clerks/service workers Commune ratio of spouse leaders Commune ratio of spouse professionals/technicians Commune ratio of other toilet Have no toilet Commune ratio of having TV Commune ratio of tap water Commune ratio of unclean water Commune ratio of permanent house Commune ratio of temporary house Percentage of areas with 0–4% slope Length of roads Track density Average temperature Urban * Commune ratio of flush toilet Urban * Main road density Urban * Tap water Urban * Permanent house Urban * Agricultural worker Urban * Head ethnic minorities Urban * % bare land Urban * % planted forest Urban * % natural forest Urban * Spouse primary school Urban * Spouse clerks/service workers Number of observations Adjusted R2 North West North Central Coast Coefficient Standard error Coefficient Standard error 7.696 — — -0.2620 -0.2800 -0.1298 0.2143 0.3229 — — 0.3073 — 1.6349 -0.0004 0.6661 -2.9430 1.6003 — — — 1.1221 0.4564 — — -1.5014 — -0.1676 0.5738 -0.3276 — — -0.2733 0.0027 0.0000 -0.0415 — — 0.1364 — -0.1999 0.2835 0.4111 0.0229 -0.0519 — -0.3834 -0.3389 — — 0.236 — — 0.0599 0.0399 0.0289 0.0813 0.0608 — — 0.0532 — 0.4339 0.0001 0.1080 1.1234 0.6166 — — — 0.4782 0.0913 — — 0.6982 — 0.0389 0.1156 0.1324 — — 0.0744 0.0012 0.0000 0.0111 — — 0.0328 — 0.0905 0.0905 0.0950 0.0051 0.0191 — 0.1504 0.1171 892 0.704 8.019 0.0882 -0.0005 -0.2446 — -0.1245 0.1477 0.2837 0.2860 0.1640 0.1650 -1.5274 0.8698 — — — — -0.3332 -1.2351 -0.1381 — 0.2302 0.1527 6.5745 — -0.1902 -0.1485 0.4570 — -0.1227 0.2814 — — — -0.0195 0.1159 -0.8253 — 0.2937 — 0.1645 — — 0.0072 -0.0054 — — — — 1.207 0.0216 0.0001 0.0209 — 0.0203 0.0419 0.0486 0.0449 0.0558 0.0244 0.4794 0.2109 — — — — 0.0809 0.3762 0.0201 — 0.0383 0.0457 2.3600 — 0.0420 0.0247 0.0641 — 0.0434 0.1103 — — — 0.0041 0.0201 0.1163 — 0.0506 — 0.0456 — — 0.0017 0.0015 — — 3238 0.448 Note: —, not applicable. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 414 ASIAN ECONOMIC JOURNAL Table A3 Regression on per capita expenditure for South Central Coast and Central Highlands Explanatory variables South Central Coast Coefficient Intercept Head age Head age squared Member working in agriculture Head ethnic minorities Head primary school Head post secondary school Head leaders/managers Head professionals/technicians Head clerks/service workers Permanent house Main road density Minor road density Commune ratio of head post secondary school Spouse primary school Commune ratio of spouse upper–secondary school Ratio of spouse leaders/managers Spouse professionals/technicians Commune ratio of other toilet Have no toilet Commune ratio of having TV Commune ratio of unclean water Tap water Commune ratio of permanent house Commune ratio of temporary house Average temperature Average humidity Annual sunshine duration Urban * Commune ratio of permanent house Urban * Commune ratio of temporary house Urban * Head primary school Urban * Head upper–secondary Urban * Commune ratio of post secondary Urban * Spouse clerks/service workers Urban * Spouse not working Urban * No toilet Urban * Commune ratio of no toilet Urban * Permanent house Urban * % natural forest Urban * Market payment to State Urban * Number of market per commune Urban * Commune ratio of working people Urban * Track density Number of observations Adjusted R2 3.225 0.0207 -0.0002 -0.1246 -0.2850 -0.1250 0.1123 0.2521 0.1052 — 0.3091 — 0.0247 -2.4498 -0.0412 — — 0.2406 -0.1617 -0.1726 0.5474 — 0.1237 2.3632 — 0.1621 — — -2.2481 — — 1.8170 — — -1.1951 -0.1842 0.5674 — 0.0030 0.0000 — -1.0368 0.0358 — — Standard error 0.992 0.0031 0.0000 0.0174 0.0480 0.0187 0.0481 0.0643 0.0457 — 0.0315 — 0.0035 0.4716 0.0179 — — 0.0468 0.0346 0.0188 0.0569 — 0.0291 0.4324 — 0.0397 — — 0.5210 — — 0.3456 — — 0.2045 0.0491 0.1182 0.0008 0.0000 — 0.2385 0.0113 2715 0.564 Central Highlands Coefficient Standard error 2.161 — — — -0.3943 — 0.1881 0.5003 0.3632 0.3730 0.1902 -0.1097 — — -0.0903 3.2856 0.4471 0.3519 — -0.2312 — -0.3281 — — 0.2035 — 0.1029 -0.0012 — -0.8021 -0.1934 — -5.9178 0.3345 — — 1.1918 0.1931 — 0.0000 0.3012 — — — — 4.450 — — — 0.0299 — 0.0909 0.0979 0.0724 0.1048 0.0651 0.0304 — — 0.0282 0.4095 0.1511 0.0843 — 0.0302 — 0.0648 — — 0.0653 — 0.0497 0.0002 — 0.1749 0.0600 — 1.0121 0.1346 — — 0.2474 0.0915 — 0.0000 0.0699 — 0.0407 1040 0.625 Note: —, not applicable. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 415 UPDATING POVERTY A MAP WITHOUT PANEL DATA Table A4 Regression on per capita expenditure for North East South and Mekong River Delta Explanatory variables Intercept Member working in agriculture Head ethnic minorities Head primary school Head post secondary school Head leaders/managers Head professionals/technicians Head clerks/service workers Permanent house Log of head age Number of markets per commune Minor road density % natural land Commune ratio of head upper–sec school Commune ratio of head unskilled workers Commune ratio of head not working Spouse primary school Spouse professionals/technicians Spouse clerks/service workers Commune ratio of spouse leaders Commune ratio of spouse professionals/technicians Commune ratio of spouse clerk/services Commune ratio of spouse agricultural worker Commune ratio of flush toilet Commune ratio of no toilet Have no toilet Commune ratio of having TV Commune ratio of having radio Commune ratio of tap water Commune ratio of clean water Tap water Commune ratio of permanent house Commune ratio of temporary house Percentage of areas with 0–4% slope Track density Average temperature Average humidity Annual sunshine duration Urban * Head age squared Urban * Commune ratio of semi–permanent house Urban * Minor road density Urban * Commune ratio of spouse post secondary Urban * Spouse leaders/managers Urban * Market payment to State Urban * Number of market per commune Urban * Track density Number of observations Adjusted R2 North East South Mekong River Delta Coefficient Standard error Coefficient Standard error 5.218 -0.1438 -0.3038 -0.0848 0.1105 0.2893 0.1877 0.1608 0.2589 0.1392 — -0.0379 0.0019 — 1.2785 — -0.1166 0.1576 — — — — — — 0.2092 -0.1738 0.6010 — — 0.1521 0.2676 — — — — 0.1644 — -0.0009 -0.0001 0.2230 0.0448 2.6107 — 0.0000 -0.1982 0.0051 — — 0.284 0.0185 0.0392 0.0171 0.0422 0.0659 0.0399 0.0339 0.0226 0.0292 — 0.0068 0.0004 — 0.3361 — 0.0186 0.0415 — — — — — — 0.0472 0.0246 0.0793 — — 0.0357 0.0233 — — — — 0.0114 — 0.0001 0.0000 0.0585 0.0068 0.3784 — 0.0000 0.0252 — 3806 0.634 12.119 -0.1105 -0.2360 -0.1485 0.2637 — — — 0.3152 0.2007 0.0460 0.0202 — 0.9903 — -0.9142 -0.1171 0.1885 0.1615 12.9119 -4.1154 -1.1150 -0.5226 0.3612 — -0.0982 — 0.4305 -0.2089 — 0.1022 — -0.3294 -0.0119 0.0089 — -0.0399 — 0.0000 — — — 0.4868 — 0.0909 — — — 0.722 0.0144 0.0251 0.0128 0.0455 — — — 0.0252 0.0201 0.0115 0.0053 — 0.2539 — 0.1307 0.0123 0.0419 0.0401 4.2230 0.9595 0.2232 0.0569 0.0810 — 0.0144 — 0.0951 0.0505 — 0.0188 — 0.0603 0.0020 0.0027 — 0.0087 — 0.0000 — — — 0.1591 — 0.0174 — 5830 0.349 Note: —, not applicable. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 416 ASIAN ECONOMIC JOURNAL Figure A1 Estimates of provincial poverty rates in 2002 using the poverty mapping method North East North West Red River Delta North Central Coast Poverty rate (%) 0–10 10–20 South Central Coast 20–30 30–40 40–50 50–60 60–70 Central Highlands 70–80 80–90 90–100 North East South N Mekong River Delta 200 0 200 400 km © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. 417 UPDATING POVERTY A MAP WITHOUT PANEL DATA Figure A2 Estimates of district poverty rates in 2002 using the poverty mapping method North East North West Red River Delta North Central Coast Poverty rate (%) 0–10 10–20 South Central Coast 20–30 30–40 40–50 50–60 60–70 Central Highlands 70–80 80–90 90–100 North East South N Mekong River Delta 200 0 200 400 km © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd. ASIAN ECONOMIC JOURNAL 418 References Baker, J. and M. Grosh, 1994, Poverty reduction through geographic targeting: How well does it work? World Development, 22, pp. 983–95 Bedi, T., A. Coudouel and S. Simer, 2007, More than a pretty picture: Using poverty maps to design better policies and interventions. World Bank, Washington, DC. Bigman, D. and H. Fofack, 2000, Geographic targeting for poverty alleviation: Methodology and applications. World Bank Regional and Sectoral Studies, Washington, DC. Elbers, C., T. Fujii, P. Lanjouw, B. Ozler and W. Yin, 2007, Poverty alleviation through geographic targeting: Hhow much does disaggregation help? Journal of Development Economics, 83, pp.198– 213. Elbers, C., J. Lanjouw and P. Lanjouw, 2002, Micro-level estimation of welfare. Policy Research Working Paper No. WPS 2911. World Bank, Washington, DC. Elbers, C., J. Lanjouw and P. Lanjouw, 2003, Micro-level estimation of poverty and inequality. Econometrica, 71, pp. 355–64. Emwanu, T., J. Hoogeveen and P. Okwi, 2006, Updating poverty maps with panel data. World Development, 34, pp. 2076–88. Foster, J., J. Greer and E. Thorbecke, 1984, A class of decomposable poverty measures. Econometrica, 52, pp. 761–6. Fujii, T. and D. Roland-Holst, 2008, How does Vietnam’s accession to the World Trade Organization change the spatial incidence of poverty? Policy Research Working Paper No. WPS 4521. World Bank, Washington, DC. Kotzeva, M. and A. Tzvetkov, 2006, Regional dimensions of poverty in Bulgaria: The challenges of poverty, pp. 86–104. National Statistical Institute, Sofia. Litvack, J., 2007, The poverty mapping application in Morocco. In: More than a Pretty Picture: Using Poverty Maps to Design Better Policies and Interventions (eds Bedi T., Aline C. and Kenneth S.), pp. 208–24. World Bank, Washington, DC. Minot, N., 2000, Generating disaggregated poverty maps: An application to Vietnam. World Development, 28, pp. 319–31. Minot, N., B. Baulch and M. Epprecht, 2003, Poverty and inequality in Vietnam: Spatial patterns and geographic determinants. Final report of project ‘Poverty Mapping and Market Access in Vietnam’ conducted by the International Food Policy Research Institute Washington, DC, and the Institute of Development Studies, the University of Sussex, Brighton, UK. Tarozzi, A. and A. Deaton, 2009, Using census and survey data to estimate poverty and inequality for small areas. Review of Economics and Statistics, 91 (forthcoming). World Bank, 1996, Ecuador poverty report. World Bank Country Studies Series. World Bank, Washington, DC. World Bank, 2004, Ecuador poverty assessment. Report 27061-EC, Poverty Reduction and Economic Management Sector Unit, Latin America and the Caribbean Region, World Bank, Washington, DC. © 2009 The Author Journal compilation © 2009 East Asian Economic Association and Blackwell Publishing Ltd.