90487 v2 Poverty Maps of Bangladesh - 2010 Technical Report TABLE OF CONTENTS Acknowledgements 1. Introduction 2. Poverty Mapping A. Methodology B. Main data sources C. Technical Challenges D. Production of the Bangladesh Maps of 2010 3. Results at a Glance 4. Assessment of the Results in terms of Various Key Statistics A. Comparison of Poverty Mapping Estimates and HIES 2010 Estimates B. Explanatory Power of Consumption Models (Model fitness) C. Share of Variance of Residuals at Cluster Level D. Incidence of Trimming 5. Key correlates of Poverty Incidence in Bangladesh A. The poverty Map and Educational Attainment B. The poverty Map and Average Agricultural Wage Rate of Male Laborers C. The Poverty Map and Perceptions Map D. Poverty Headcount Rates VS Poor Population 6. Concluding Remarks Annex 1: 2010 Poverty Lines and Poverty Headcount Rate Estimates Annex 2: Key Results of the Poverty Mapping Exercise Annex 3: Standard Errors of the Poverty Estimates Annex 4: Results of Multi-Layer Analysis Annex 5: Detailed Methodology on SAE i Acknowledgements The World Bank team involved in preparation of the Poverty Maps included Faizuddin Ahmed, Naomi Ahmad, Mehar Akhter, Dean Jolliffe, MehrinMahbub, Iffath Sharif, Nobuo Yoshida, and Salman Zaidi (task leader), working under the guidance of VinayaSwaroop (Sector Manager, SASEP) and Johannes Zutt (Country Director, Bangladesh, Nepal, and Bhutan). The World Food Programme (WFP) team included MahabubulAlam, Nusha Choudhury, and KayenatKabir, working under the guidance of Christa Räder (Country Representative, WFP). The BBS team included Md. ZahidulHoqueSardar, Dipankar Roy, AKM Tahidul Islam, Md. Abdul Latif, Dinesh Roy, Enayet Hossain, Md. Rezaul Karim, and Md. Jibon Miah. Faizuddin Ahmed spearheaded the extensive analytical work to derive the regionally disaggregated poverty estimates using the 2010 HIES and 2011 Population Census data sets. MahabubulAlam and Faizuddin Ahmed prepared the various poverty maps presented in this report using GIS software. The team acknowledges the leadership and support of Mr. Md. Nojibur Rahman (Secretary Statistics and Informatics Division) and Mr. GolamMostafa Kamal (Director General, Bangladesh Bureau of Statistics). The team would like to thank members of the technical committee and the steering committee, who carefully reviewed both the methodology used as well as the results of the Poverty Mapping work. The team is also grateful to the UK Department for International Development (DFID) and International Fund for Agricultural Development (IFAD) for their financial support for the preparation and publication of the Poverty Maps. ii 1. Introduction Poverty mapping is a statistical exercise to estimate the incidence of poverty at sub-national levels to enable the government, civil society organizations, and development partners to accurately identify locations with a relatively higher concentration of poor people. Due to the considerable demand from policy makers, planners, and researchers for more disaggregated poverty estimates, the current poverty mapping exercise was initiated in September 2012 by the Bangladesh Bureau of Statistics (BBS), the World Bank, and the World Food Program (WFP) to produce reliable poverty estimates for key sub-national administrative units (zila and upazila) using data from both the 2010 Household Income and Expenditure Survey (HIES) and the 2011 Population Census.1 According to this latest population census conducted by BBS, the total population of Bangladesh was about 150 million. For administrative purposes the country is divided into 7 divisions, 64 districts, and 544 upazilas /thanas. Table 1 :Administrative Structure of Bangladesh Number of Number of Division Districts Upazilas Bangladesh 64 544 Dhaka 17 163 Chittagong 11 111 Rajshahi 8 70 Rangpur 8 58 Khulna 10 64 Sylhet 4 38 Barisal 6 40 The HIES is conducted by BBS every 4-5 years, and is the main source of data for official poverty related statistics in Bangladesh. The World Bank has provided extensive technical assistance to BBS over the past two decades to help improve the quality and timeliness of data collected through this survey. The official poverty estimates are computed from the HIES at the National and Division-level only due to the limited sample size of the survey. Table 2 : Upper and Lower Poverty Estimates for Bangladesh (2010 HIES)2 Headcount Poverty Rate (percent) Division Upper Poverty Line Lower Poverty Line Bangladesh 31.5 17.6 Dhaka 30.5 15.6 Chittagong 26.2 13.1 Rajshahi 29.7 16.0 Rangpur 42.3 27.7 Khulna 32.1 15.4 Sylhet 28.1 20.7 Barisal 39.4 26.7 Source: 2013 Bangladesh Poverty Assessment, World Bank. 1 2 The definition of the upper and lower poverty lines can be found in the 2013 Bangladesh - Poverty Assessment: Assessing a decade of progress in reducing poverty, 2000-2010 (http://documents.worldbank.org/curated/en/2013/06/17886000/bangladesh- poverty-assessment-assessing-decade-progress-reducing-poverty-2000-2010). 1 According to the latest 2010 HIES based estimates, poverty incidence in Bangladesh varies from a low of 26.2 percent in Chittagong division to a high of 42.3 percent in Rangpur division. Similarly, the incidence of extreme poverty (i.e. estimates based on the lower poverty line) varies from 13.1 percent in Chittagong division to 27.7 percent in Rangpur division. 2 2. Poverty Mapping A. Methodology The poverty mapping methodology used in this exercise is the so-called ELL method developed by Elbers, Lanjouw, and Lanjouw(2003) using Small Area Estimation (SAE) techniques. The ELL method, which has been widely tested and validated around the world, takes advantage of the strengths of both sources of data used in such exercises. The HIES includes the extremely rich data collected in an integrated household survey, including expenditure data. However this is for a relatively limited sample of households. On the other hand, the Population Census includes all households in the country, but collects data for a limited set of topics. The two data sets, HIES and Population Census, have common set of explanatory variables. The SAE technique uses the parameter estimates from a consumption model derived using the 2010 HIES data to predict/simulate consumption data for each census household. These predicted/simulated consumption data for all 2011 census households are then used to estimate poverty rates at the zila and upazila level using the same poverty lines used to derive the official poverty estimates using the 2010 HIES data. Poverty estimates for the Bangladesh Poverty Map and Extreme Poverty Maps were derived by using the upper and lower poverty lines in the HIES 2010 report published by BBS, which were also used in the World Bank’s latest Poverty Assessment.3The spatial distribution of poverty in Bangladesh at the upazila level is presented in the two maps below. 3 The methodology used to derive the poverty lines is briefly described in the Annex. Further details can be found in can be found in Bangladesh - Poverty Assessment: Assessing a decade of progress in reducing poverty, 2000-2010World Bank, 2013). 3 4 B. Main Data Sources The primary data sources generally used in the Small Area Estimation (SAE) method are a nationally representative household survey and a national population census. The Bangladesh Poverty Mapping of 2010 used the primary data of the 2010 Household Income and Expenditure Survey(HIES) and the 2011 Population Census.As a national statistical organization, Bangladesh Bureau of Statistics collects through HIES, a wide range of data covering many socio economic and demographic information such as detailed information on consumption and income , employment, ownership of asset, housing condition, access to education, health and sanitation etc. The sample size of 2010 HIES was 12,240 households, and most of the variables were representative at the division level. Through the 2011 Population Census BBS collected a wide range of data on household and individual characteristics, including employment, housing conditions, educational attainment, sources of drinking water, access to sanitation, electricity, etc. As a global practice, population census does not include consumption and income data. BBS collected census data covering around 32 million households in Bangladesh. C. Technical Challenges The previous poverty mapping exercise in Bangladesh was done using Population Census 2001 and HIES 2005 data. Back then, the long interval between population census 2001 and HIES 2005 posed a significant technical limitation and challenge. Applying SAE method, a consumption model was derived in HIES 2005 by regressing per capita consumption expenditure on a set of common variables present in both HIES 2005 and population census 2001. The model was then used to predict per capita household expenditure for each household in the census.This approach works well if one can assume that consumption pattern between 2001 and 2005 did not change much. While this assumption may be reasonable for a short interval of time, it is more problematic for a longer interval (e.g. the 4 year gap between 2001 and 2005) as many structural changes may happen during this period. Changes in consumption patterns can introduce biases in the poverty estimates and their standard errors derived from the ELL method. To mitigatethis potential problem, only time invariant variables – i.e. variables whose mean did not change much between 2001 and 2005 – were selected for the consumption model and this approach was found to be effective in reducing biases. As the current poverty mapping exercise of Bangladesh used 2010 HIES and 2011 population census data, the issue of long interval between census and survey could not come as a great technical challenge. But still then, every efforts were made to select those common variables whose means did not change much between 2010 HIES and 2011 population census. In fact, we computed the 95 percent confidence interval of all variables for HIES 2010. If the census mean of these variables fell within the 95 percent confidence ranges, we took those variables for the consumption modeling. Tarrozi and Deaton (2008) have highlighted a number of concerns with the ELL methodology, in particular as summarized below: 5 1) Differences in consumption patterns within a stratum can bias both the poverty estimates and the standard errors. The ELL method estimates a consumption model that is assumed to apply to all households within each stratum, the implicit assumption being that the relationship between the household consumption expenditures and its correlates is the same for all households within the domain, and that all remaining differences are due not to structural factors, but attributable to errors. This is not a minor assumption and is explicitly acknowledged as such in ELL (2003). 2) Misspecification in the error structure can lead to an overstatement of the precision of the poverty estimates. The poverty mapping software (PovMap2) developed by the World Bank research department, in its current configuration, can accommodate only two layers of errors: at the level of household and at the level of some unit of aggregation above the household. In the current poverty mapping exercise for Bangladesh, the mauza level was selected for this higher level of aggregation. However, as Tarrozi and Deaton have noted, there could be correlation in errors also at some higher level viz. union, upazila, or zila. If the ELL method is applied ignoring the existing correlationof errors at these higher levels, then the standard errors of poverty estimates can be understated, resulting in an overly optimistic assessment of their statistical precision. While one obvious solution to this issue is to allow for multi layers of errors during consumption modeling, thisis not a practical or feasible solution for practitioners using the PovMap2 software. Instead, special attention was given in the present exercise to undertake a set of validation exercises on the basis of indirect empirical evidence (Please see section 4 and 5). D. Production of the Bangladesh Poverty Maps of 2010 In poverty mapping procedure, two key aspects need to be paid special attention: o Selecting sound consumption models o Selecting the appropriate level of disaggregation As elaborated below, careful execution of the poverty mapping software is critical, despite the convenience and user-friendliness of the PovMap2 software. Although, the PovMap2 software makes this process easier by providing various statistics toguide users, each step must be checked carefully. Selection of Consumption Model: For the current poverty mapping exercise, a total of 18 consumption models,of which 16 correspond to the strata defined in HIES 2010, were createdto capture the regional variations,. The other two models relate to rural & urban areas of Rangpur, a new division that was created from Rajshahi division after 2010. According to the results of HIES 2010, it is evident that Rangpur division is a poverty prone area, with significant variation in poverty rates between it and the rest of Rajshahi division. Consequently the creation of two additional models for this new division was a logical next step. Levels of Disaggregation: The ELL method producesnot only poverty estimates but also standard errors associated with these estimates, which can be very helpful to practitioners when deciding on the appropriate level of disaggregation of the poverty estimates. Table 1 summarizes standard errors of the poverty estimates at four different levels and by percentile. The table shows that standard errors on average (median) at all levels are small. Even at the 95th percentile, the standard errors at all levels are reasonably low. However, the maximum number at union level reaches nearly 22 percentage points which is indeed quite high, 6 meaning that the 95 percent confidence interval of the corresponding poverty estimate has a range of +/- 44 percentage points from the poverty estimate—i.e. the true poverty rate of this union can be anywhere between 0 and 100 percent with 95 percent probability. This result indicate that Bangladesh poverty mapping estimates should preferably not be disaggregated beyond the upazila level (one level abovetheunion level) to avoid reaching an unacceptably low level of statistical reliability. Table 3 : Assessment of Simulation Results at Various Levels Standard Errors of Poverty Estimates (%) Percentile Stratum Zila Upazila Union Median 1.2 1.8 2.2 3.8 95 % 1.6 4.5 5.2 7.7 Max 1.9 6.2 10.7 21.5 7 3. Results at a Glance Division Level Poverty Estimates The poverty estimates derived through this Poverty Mapping exercise are quite close to those obtained from the 2010 HIES. The minor differences between these two sets of estimates (as summarized in the table below) are partly to be expected, since the methods used in the Poverty Mapping exercise match predicted consumption (i.e. not poverty rates) at the Mauza/Mahalla level in the two data sets – i.e. 2010 HIES and 2011 Population Census(see Annex 3). Table 4 :Comparison of Poverty Estimates: 2010 HIES and Poverty Mapping HIES Poverty Mapping Exercise 2010 Poverty 2010 Poverty Number of Poor (as % of Division Headcount (Percent) Headcount (Percent) overall population) Bangladesh 31.5 30.7 100.0 Dhaka 30.5 30.5 32.3 Chittagong 26.2 26.1 16.8 Rajshahi 29.7 27.4 11.6 Rangpur 42.3 42.0 15.0 Khulna 32.1 31.9 11.4 Sylhet 28.1 25.1 5.7 Barisal 39.4 38.3 7.3  Poverty estimates for Bangladesh based on both the HIES and the poverty mapping exercises show that Rangpur and Barisal divisions have the highest poverty incidence of poverty, while Chittagong and Sylhet have the lowest incidence.  Because of their large overall populations, Dhaka division (32.3 percent) and Chittagong division (16.8 percent) have the highest share of Bangladesh’s poor.  Compared to other divisions, Sylhet division has both the lowest headcount rate (25.1 percent) as well as the lowest number of poor people (5.7 percent of the country’s poor). Zila Level and Upazila Level Poverty Maps As the following maps illustrate, the resolution of spatial variation in poverty incidence improves considerably on moving from the division to zila level and upazila level poverty maps - there is considerable spatial variation in poverty incidence within individual divisions. 8 9 Key features of the variation in poverty incidence at the zila level are highlighted below: Table 5 :Variation in Headcount Rate by Division Division Average Poverty Rate Minimum Rate Maximum Rate 15.7 percent 52.6 percent Dhaka 30.5 percent Dhaka district Shariatpur district 9.6 percent 51.0 percent Chittagong 26.1 percent Noakhali district Chandpur district 16.6 percent 38.7 percent Rajshahi 27.4 percent Bogra district Sirajganj district 26.7 percent 63.7 percent Rangpur 42.0 percent Panchagarh district Kurigram district 3.6 percent 46.3 percent Khulna 31.9 percent Kushtia district Satkhira district 24.1 percent 26.0 percent Sylhet 25.1 percent Sylhet district Sunamganj district 19.0 percent 54.8 percent Barisal 38.3 percent Barguna district Barisal district 3.6 percent 63.7 percent Overall Bangladesh 30.7 percent Kushtia district Kurigram district Table 6 :Population, Number of Poor, and Headcount Rate in Bangladesh by District Division Zila #Upazilas Population Number Poor Headcount DHAKA 17 districts in total 163 45,762,841 13,972,226 30.5 SHARIATPUR 6 1,146,046 602,581 52.6 JAMALPUR 7 2,277,919 1,163,677 51.1 MYMENSINGH 12 5,046,655 2,547,466 50.5 SHERPUR 5 1,352,039 654,755 48.4 GOPALGANJ 5 1,158,216 494,505 42.7 RAJBARI 5 1,041,978 436,525 41.9 FARIDPUR 9 1,888,142 686,208 36.3 NETRAKONA 10 2,215,484 781,883 35.3 MADARIPUR 4 1,157,997 403,913 34.9 KISHORGONJ 13 2,880,856 872,936 30.3 TANGAIL 12 3,548,352 1,055,432 29.7 MUNSHIGANJ 6 1,409,831 404,116 28.7 NARAYANGANJ 5 2,847,240 744,072 26.1 NARSINGDI 6 2,180,550 517,906 23.8 GAZIPUR 5 3,235,402 626,340 19.4 MANIKGANJ 7 1,376,134 254,098 18.5 DHAKA 46 11,000,000 1,725,814 15.7 CHITTAGONG 11 districts in total 111 27,904,587 7,273,642 26.1 CHANDPUR 8 2,388,365 1,217,085 51.0 BANDARBAN 7 373,273 149,575 40.1 COMILLA 16 5,303,074 2,010,667 37.9 COX'S BAZAR 8 2,250,089 735,531 32.7 LAKSHMIPUR 5 1,699,556 530,006 31.2 BRAHMANBARIA 9 2,820,084 847,110 30.0 FENI 6 1,406,908 363,971 25.9 KHAGRACHHARI 8 599,899 152,808 25.5 RANGAMATI 10 576,536 117,111 20.3 CHITTAGONG 25 7,417,706 854,181 11.5 NOAKHALI 9 3,069,097 295,596 9.6 10 Division Zila #Upazilas Population #Poor Headcount RAJSHAHI 8 districts in total 70 18,252,001 5,008,480 27.4 SIRAJGANJ 9 3,070,468 1,188,618 38.7 NATORE 6 1,682,265 589,704 35.1 PABNA 9 2,503,504 789,824 31.5 RAJSHAHI 13 2,527,816 794,002 31.4 JOYPURHAT 5 900,984 240,529 26.7 NAWABGANJ 5 1,638,958 414,749 25.3 NAOGAON 11 2,581,033 434,989 16.9 BOGRA 12 3,346,973 556,065 16.6 RANGPUR 8 districts in total 58 15,482,473 6,508,616 42.0 KURIGRAM 9 2,051,530 1,306,462 63.7 GAIBANDHA 7 2,365,117 1,135,938 48.0 RANGPUR 8 2,727,182 1,260,516 46.2 DINAJPUR 13 2,930,171 1,111,722 37.9 NILPHAMARI 6 1,816,564 632,814 34.8 LALMONIRHAT 5 1,249,519 431,372 34.5 THAKURGAON 5 1,360,454 367,747 27.0 PANCHAGARH 5 981,936 262,046 26.7 KHULNA 10 districts in total 64 15,445,562 4,923,496 31.9 KUSHTIA 6 1,929,302 69,878 3.6 MEHERPUR 3 651,920 99,204 15.2 NARAIL 3 717,156 143,202 20.0 JHENAIDAH 6 1,758,472 434,194 24.7 CHUADANGA 4 1,121,841 310,218 27.7 KHULNA 14 2,218,316 860,533 38.8 JESSORE 8 2,722,805 1,062,501 39.0 BAGERHAT 9 1,447,373 619,480 42.8 MAGURA 4 912,168 413,786 45.4 SATKHIRA 7 1,966,209 910,500 46.3 SYLHET 4 districts in total 38 9,784,451 2,457,690 25.1 SYLHET 12 3,365,878 810,579 24.1 HABIGANJ 8 2,073,516 524,070 25.3 MAULVIBAZAR 7 1,901,486 488,895 25.7 SUNAMGANJ 11 2,443,571 634,145 26.0 BARISAL 6 districts in total 40 8,223,589 3,151,833 38.3 BARGUNA 5 884,747 168,098 19.0 PATUAKHALI 7 1,518,520 391,080 25.8 BHOLA 7 1,765,082 585,326 33.2 JHALOKATI 4 676,302 273,910 40.5 PIROJPUR 7 1,099,095 484,476 44.1 BARISAL 10 2,279,843 1,248,944 54.8  Poverty incidence in the 10 poorest upazilas in Dhaka division is 55 percent or higher; by contrast, poverty incidence in the 10 richest upazilas is less than 4 percent.  Similarly poverty incidence in the 6 poorest upazilas of Chittagong division is 50 percent or higher, while that in the 6 richest upazilas is less than 4 percent.  Even the poorest division of Bangladesh has considerable spatial variation in concentration of poverty: the incidence of poverty in the 11 richest upazilas in Rangpur division is lower 11 than the national average; by contrast, poverty incidence in the 7 poorest upazilas is more than twice the national average (i.e. it is 60+ percent).  While Sylhet division is amongst Bangladesh’s most well-off regions, over 50 percent of the population of Gowainghatupazila lives below the national poverty line; similarly, 3 upazilas in Khulna division have a poverty rate of 50 percent or higher. Table 7 :Richest and Poorest Upazilas by Division Division Upazila Population #Poor Headcount DHAKA 163upazilas in total 45,762,841 13,972,226 30.5 3 Richest MOTIJHEEL 184,854 2,403 1.3 BIMAN BANDAR 8,350 109 1.3 DHANMONDI 129,815 1,776 1.4 5 Poorest DHOBAURA 195,175 113,586 58.2 GOSAIRHAT 156,416 91,244 58.3 DEWANGANJ 256,539 150,097 58.5 PHULPUR 599,947 352,742 58.8 NANDAIL 400,675 243,059 60.7 CHITTAGONG 111 upazilas in total 27,904,587 7,273,642 26.1 3 Richest DOUBLE MOORING 345,272 23 0.0 KOTWALI 292,683 778 0.3 PANCHLAISH 210,442 1,609 0.8 5 Poorest THANCHI 22,527 11,950 53.0 MATLAB DAKSHIN 207,963 111,577 53.7 HAJIGANJ 327,901 176,210 53.7 KACHUA 378,499 213,059 56.3 HAIM CHAR 108,032 66,222 61.3 RAJSHAHI 70 upazilas in total 18,252,001 5,008,480 27.4 3 Richest KAHALOO 221,474 25,878 11.7 SHAJAHANPUR 279,247 34,986 12.5 ADAMDIGHI 193,757 25,288 13.1 5 Poorest BERA 255,993 100,955 39.4 SHAHJADPUR 556,927 232,565 41.8 BELKUCHI 348,558 147,985 42.5 GODAGARI 329,218 145,338 44.1 CHAUHALI 159,283 72,494 45.5 RANGPUR 58upazilas in total 15,482,473 6,508,616 42.0 3 Richest TENTULIA 124,773 26,869 21.5 PIRGANJ 240,733 55,982 23.3 ATWARI 133,099 32,013 24.1 5 Poorest BHURUNGAMARI 230,639 150,246 65.1 ULIPUR 393,074 256,751 65.3 RAJARHAT 182,464 123,506 67.7 PHULBARI 159,682 109,358 68.5 CHAR RAJIBPUR 73,154 50,346 68.8 12 Division Upazila Population #Poor Headcount KHULNA 64 upazilas in total 27,904,587 7,273,642 26.1 3 Richest KUSHTIA SADAR 490,158 14,922 3.0 MIRPUR 328,710 11,003 3.3 BHERAMARA 198,999 6,790 3.4 5 Poorest KOYRA 193,301 94,980 49.1 TEROKHADA 115,758 57,420 49.6 CHITALMARI 138,261 69,143 50.0 SHYAMNAGAR 314,770 157,983 50.2 MOHAMMADPUR 207,394 105,304 50.8 SYLHET 38 upazilas in total 18,252,001 5,008,480 27.4 3 Richest DAKSHIN SURMA 249,563 25,712 10.3 BISHWANATH 229,614 28,794 12.5 SYLHET SADAR 789,139 112,655 14.3 5 Poorest JAINTIAPUR 161,103 55,827 34.7 JURI 148,268 53,806 36.3 ZAKIGANJ 234,557 91,510 39.0 KANAIGHAT 262,546 120,179 45.8 GOWAINGHAT 286,433 150,625 52.6 BARISAL 40 upazilas in total 18,252,001 5,008,480 27.4 3 Richest PATHARGHATA 162,556 20,977 12.9 BAMNA 79,209 13,505 17.1 MIRZAGANJ 121,192 21,536 17.8 5 Poorest BAKERGANJ 310,455 171,915 55.4 GAURNADI 185,815 103,038 55.5 MULADI 172,582 100,462 58.2 HIZLA 144,496 89,963 62.3 MHENDIGANJ 298,801 192,326 64.4 13 4. Assessment of the Results in terms of Various Key Statistics The Poverty Mapping exercise of 2010 also included a number of validation checks to assess the reliability and consistency of the poverty estimates: A. Comparison of Poverty Mapping estimates and HIES 2010 estimates HIES 2010 follows a stratified sample design whereby estimates derived from these data are representative at the division and urban-rural level. Figure 1 illustrates the consistency checks using 95 percent confidence intervals from both the HIES 2010 as well as the Poverty Mapping exercise. These two estimates will be consistent if their 95 percent confidence intervals are overlapping. As the figure below clearly shows, both estimates are overlapping in almost all the strata. Another interesting observation is that the range of confidence intervals of poverty mapping estimates is much narrower than that of the direct HIES 2010 estimates which indicates that the former(Poverty Mapping) estimates have much lower margin of error than the latter (i.e. the HIES 2010 direct estimates). Figure 1: Comparison between poverty estimates from HIES 2010 and SAE method Figure 1a bellow presents the scatter diagram of the poverty estimates from two sources at stratum level. If we look at the diagram, we see the dots falling almost in a straight line. This clearly indicates a very high correlation between the two poverty estimates. We have computed the simple correlation coefficient and found it to be 0.98, which is indeed a very high correlation. 14 Figure 1a :Comparison of poverty estimates showing degree of correlation between the two estimates. Scatter diagram showing correlation between poverty from two sources .5 .4 poverty_rate_hies2010 .3 .2 .1 Correlation= 0.98 .1 .2 .3 .4 .5 poverty_rate_sae2010 27 August 2014 15 B. Explanatory power of consumption models (Model fitness) The goal of the consumption model is to produce the most reliable and precise estimates of poverty. One measure of the specification of the model is model fit, measured by R-square and adjusted R-square. R-square measures the proportion of the variability in the target variable explained by the predictors. In other words, it provides information on how well a consumption model can predict the actual consumption expenditure of each census household. Adjusted R-square is a modification of R- square that adjusts for the degrees of freedom (df) in the model. Generally, the higher the R-square, the better predicted expenditure fits actual household expenditure. In the Bangladesh poverty mapping of 2010, both R-square and adjusted R-square are high (Figure 2). Regressions using cross sectional data normally tend to have lower R-square than panel or longitudinal data. Internationally, even with an R- square around 30 percent, some reasonably successful small area estimates have been produced. Figure 2: Stratum level R-square and Adjusted R-square 16 C. Share of Variance of Residuals at Cluster Level The consumption models in general capture only part of the variations in household expenditures,with the unexplained variations simply treated as residuals or errors. In the current poverty mapping exercise, these residuals are separated into two layers- the household layer and the cluster (mauza for rural area and mahalla for urban area) layer. The cluster effects are included in the consumption models since consumption expenditures can be affected by location specific factors that are common across all the households in the particular cluster or location of which some are observable and others are not. Since cluster effects can reduce the precision of the poverty or inequality estimates, great effort should be taken so that variation in consumption may be captured by observables as far as possible. If mauza or mahalla level effect is large, reliability of poverty estimates is low. A rough rule of thumb often used in this regard is whether the share of variance of this cluster effect to the total variance is 10 percent or less. One strategy for reducing the share of the variance of cluster effect is to include more area or location specific variables in the consumption models. In the current poverty mapping exercise, such location specific variables have been constructed by aggregating data from the Population Census. Besides, GIS data and data from Business Register 2009 at upazila level were also included in the regression models. This approach worked well in minimizing the share of cluster level effectfor the current Poverty Mapping exercise. Figure 3presents the share of the variance of cluster level effect by region (stratum): almost all strata except #3(Chittagong Rural), #8 (Dhaka Statistical Metropolitan Area) and #17 (Rangpur Rural) have the share of variance of cluster level effect less than 5 percent. 17 Figure 3 : Share of cluster level error 18 D. Incidence of Trimming The ELL method, at the time of simulation, randomly draw regression coefficients including residuals or errors from their corresponding distributions as estimated in the survey based consumption models. This random drawing of parameters (regression coefficients & errors) can at times pick up some extreme values, although with a low probability. Thus the simulated household expenditures may include a few outliers. Fortunately, PovMap2 software has the option to eliminate these outliers before estimating povertyorinequality indicators.This adjustment, often called "trimming", is required since a few outlier values can produce large biases, particularly in inequality statistics. However, trimming is a practical solution, and does not follow any rigorous statistical theory per se. Consequently, when constructing consumption models, care should be taken to do it in such a way as to minimize the need for trimming. Table 8 summarizes the incidence of trimming at different levels, and shows that at the level of stratum &zila, the incidence of trimming is low. Even at Upazila and Union levels, the incidence of trimming is also fairly low up to the 95th percentile. However, the maximum number at the union level is as high as 35 percent which means that more than one third of the simulated expenditures were dropped before estimating poverty headcount rates, further confirming that Bangladesh poverty estimations of 2010 should not be disaggregated below the Upazila level Table 8 : Incidence of trimming at various levels 5. Key Correlates of Poverty Incidence in Bangladesh As illustrated in the two sets of maps presented below, the poverty maps can also be compared with other geographic and regional characteristics that are likely correlated with poverty incidence. 19 A. The Poverty Map and Educational Attainment of Household Heads The maps below compare poverty rates with educational attainment of household heads (2011 Population Census data). Darker areas on the maps correspond to areas with high poverty rates and lower rates of completion of primary education. As the maps show, districts in north and southeastern Bangladesh whose poverty rates are high also suffer from low primary school completion 20 B. The Poverty Map and Average Agricultural Wage Rate of Male Laborers The maps below contrast poverty rates with average agricultural wages rates (BBS data), and illustrate the negative association between these two variables. Darker areas on the maps correspond to areas with high poverty rates and low wage rates. 21 C. Comparison between the Poverty Maps and Perceptions Map The reliability of the poverty maps can be evaluated by comparing the estimates obtained from the poverty mapping exercise with the results of the Perception Survey on Relative Prevalence of Poverty commissioned by the WFP. This was carried out in 2014 in 16 districts (BagerhatBandarbanBograChandpurGazipurHabiganjLakshmipurManikganjNaogaonNilphamari PanchagarhPatuakhaliRangamatiSirajganjSunamganj, Thakurgaon)across Bangladesh. The participatory key informant based pair wise comparison tool was used as the main instrument for the perception survey. The approach involves structured interviews with key informants to capture both qualitative perception and quantitative estimates on poverty within an upazila. They key informants sequentially compared one geographic region with another, hence the tool has been named pair wise comparison. Inherent in the tool is the consistency check with purpose of validating the informant’s knowledge.Knowledgeable respondents from 3 groups (local government body and local people, local government officials, and NGO workers, UN field officials and community workers) in each of the 16 districts covered by the survey were asked to estimate the poverty rate in each of the upazilas in their particular district, yielding a total of 125 observations on estimated poverty rates for the 125 upazilas covered. The upazila level poverty estimates from the Poverty Mapping exercise derived using both the upper and lower poverty lines can be regressed against the poverty estimates obtained from the Perceptions survey to ascertain the degree of correlation between the two sets of independent estimates.4 Since there is some debate about whether such a regression should be run with a constant term or alternately whether the constant term should be suppressed, we report both sets of estimates below (i.e. with and without the constant term included in the regressions). Lower Poverty Line Estimate Upper Poverty Line Estimate No Constant With Constant No Constant With Constant Term Term Term Term Number of observations 125 125 125 125 F (16, 109) 94.04 - 90.90 - F (16, 108) - 7.53 - 5.94 Prob.> F 0.0000 0.0000 0.0000 0.0000 Adj. R-squared 0.9225 0.4574 0.9200 0.3892 Coefficient of HCR 1.3923 0.9010 0.8202 0.4575 t-statistic 8.76 5.19 8.42 3.45 Prob.> | t | 0.000 0.000 0.000 0.001 The summary results reported above reveal a very high correlation between the two sets of poverty estimates (i.e. upazila level poverty rates obtained from the Perceptions survey and Poverty Mapping exercise respectively). In all instances, the positive correlation between the two sets of estimates is statistically significant at the 1 percent significance level, indicating there is very close conformance between these two independent exercises in the ranking of upazilas based on estimated poverty levels. This is true both for the poverty estimates based on the upper as well as the lower poverty lines. Overall, these results provide strong corroborating evidence of the robustness of disaggregated poverty estimates obtained from the Poverty Mapping exercise. 4 When running the regressions, we included dummy variables at the district level to purge the effect of idiosyncratic differences between the various respondents in levels of their poverty estimates. 22 D. Poverty Headcount Rates Vs Number of Poor Population The Poverty headcount rate refers to the proportion or percentage of poor population living bellow the poverty line. It reflects the incidence of poverty in a specific area. It is the most popular indicator of poverty. However, one can also look at the absolute number of poor population. Maps bellow compares the number of poor with the percentage or proportion of poor population. The map on the right shows the dot density of poor population by geographic location and the map on the left shows the percentage of the poor or headcount rates by Zila. Geographic patterns of poverty in Bangladesh vary by location. There are some locations/areas, for instance, near Dhaka (circled in red)where poverty headcount is low but absolute number of poor population is large as can be seen from the map on the right by the greater number of dots within the same location/area. On the contrast, Bandarbanzila in Chittagong Hill Tract region (circled in green) record a high poverty headcount rate but the size of the poor population is very small. Also, there are areas, for instance, Rangpur region (circled in black) where both poverty rate and number of poor population are relatively high. Zila level poverty maps showing headcount rates and number of poor population 23 6. Concluding Remarks Poverty Mapping is a powerful tool for identifying and monitoring pockets of affluence and poverty across the country. The resultant maps provide a rich information base and can be used to help policy makers and development partners better plan their resource allocations, which in turn can contribute to faster and more effective poverty reduction. The usefulness of poverty maps can be further reinforced by combining them with other geo-referenced databases such as maps of human development indicators, maps of natural disasters, and maps of the impending impacts of climate change. 24 Annex 1: 2010 Poverty Lines and Poverty Headcount Rate Estimates The Bangladesh Bureau of Statistics (BBS) organized a committee of expertsto produce the official 2010 poverty estimates. The 2010 poverty estimates were based on a Cost of Basic Needs (CBN) methodology, and were derived by adjusting the 2005 poverty lines to reflect changes in the cost of meeting basic needs. The adjustments to poverty lines (PL) for 2010 were obtained by: (i) updating 2005 food poverty lines with food inflation rates calculated from unit values of HIES 2005 and HIES 2010 data; and (ii) re- estimating the non-food poverty line using HIES 2010 data to adjust for the non-food allowance. Using the CBN method, the PL represents the average level of per capita expenditure at which individuals can meet basic food and non-food needs. The CBN method is implemented in three steps. In the first step, the cost of a fixed food bundle is computed. In the case of Bangladesh, this bundle consists of eleven food items that include: rice, wheat, pulses, milk, oil, meat, fresh water fish, potato, other vegetables, sugar, and fruits. The bundle provides the minimal nutritional requirements that correspond to 2,122 kcal per day per person. In the second step, two different non-food allowances for non-food consumption are computed: the lower non-food allowance (the median amount spent on non-food items by households whose total consumption is approximately equal to their food-poverty line) and the upper non-food allowance (the amount spent on non-food items by households whose food consumption is approximately equal to their food-PL). In the third step, the food and non-food allowances are added together. The sum of the food and upper non-food allowances constitute the upper poverty line (UPL), while the sum of the food and lower non-food allowance constitutes the lower poverty line (LPL). Table 9 : Bangladesh Harmonized Poverty Lines 2000, 2005, and 2010 2000 2005 2010 Region LPL UPL LPL UPL LPL UPL Barisal (Rural) 580 714 753 926 1284 1485 Barisal (Muni.) 643 764 800 951 1419 1963 Chittagong (Rural) 619 733 753 891 1404 1687 Chittagong (Muni.) 643 827 749 963 1495 1825 Chittagong (SMA) 639 978 766 1171 1479 1876 Dhaka (Rural) 563 651 728 842 1276 1497 Dhaka (Muni.) 625 742 749 890 1314 1793 Dhaka (SMA) 678 855 806 1018 1406 2038 Khulna (Rural) 511 582 652 743 1192 1435 Khulna (Muni.) 561 690 670 825 1262 1680 Khulna (SMA) 582 773 706 938 1348 1639 Rajshahi (Rural) 511 598 656 766 1236 1487 Rajshahi (Muni.) 575 707 696 857 1312 1585 Rajshahi (SMA) 576 682 722 856 1223 1556 Sylhet (Rural) 560 661 697 822 1240 1311 Sylhet (Muni.) 666 843 806 1020 1286 1558 Source: Bangladesh - Poverty Assessment: Assessing a decade of progress in reducing poverty, 2000-2010 World Bank, 2013. The table below presents the poverty and extreme poverty headcount rate estimates for Bangladesh (i.e. the proportion of the population that is deemed to be poor and extremely poor respectively) using the 25 above poverty lines and the 2000, 2005, and 2010 HIES data sets. The poverty rates are estimated from the survey data by computing the proportion of the country’s population whose per capita expenditures are below the UPL, while the extreme poverty rates are estimated by computing the proportion of the country’s population whose per capita expenditures are below the LPL. Table 10 : Poverty Headcount Rates based on the 2000, 2005, and 2010 HIES Poverty Extreme Poverty 2000 2005 2010 2000 2005 2010 National 48.9 40.0 31.5 34.3 25.1 17.6 Urban 35.2 28.4 21.3 19.9 14.6 7.7 Rural 52.3 43.8 35.2 37.9 28.6 21.1 Source: All estimates are CBN based on HIES 2005, updated for 2010, and back-casted for 2000. 2010 update: survey-based food prices and non-food allowance re-estimated using “upper� poverty lines. Official Poverty Lines estimated for HIES (2000, 2005, and 2010). 26 Annex2: Key Results of the Poverty Mapping Exercise Barisal Rural [STRATUM 1] Estimated Variables Description of the variables Coefficient _intercept_ constant used in the model 7.7164246 CHLD0YRP Proportion of 0yr child in the household -0.8390937 DELECTRIC_1 Household with access to electricity 0.1853818 DHSEC_EDU_1 Head of household with higher secondary education 0.3646295 DNSLATRINE_1 Household with access to non-sanitary latrine -0.2067275 DOTH_HOUSE_1 Not a pucca or semi-pucca house -0.3057381 DOTH_WATER_1 Other than tap or tube-well water 0.2872642 DSEC_EDU_1 Head of household with secondary education 0.1675674 HD_AGE Age of the head of household in years 4.40E-03 MEMBER Household size -6.60E-02 N15_59YRP Proportion of 15-59 yr. persons in the household 0.4898133 N60PLUSP Proportion of elderly persons (60+) in the household 0.4891679 P11_15FSCHNEW Proportion of female children attending school 0.4865016 TPE_CONS Total persons engaged in construction sector for 2009 at upazila level -3.83E-03 _ZL$DPGRA_EDU_060 Zila=06 and head of the household not a post graduate -0.4370122 _ZL$DPGRA_EDU_420 Zila=42 and head of the household not a post graduate -0.3089488 _ZL$DPGRA_EDU_790 Zila=79 and head of the household not a post graduate -0.3639935 Barisal Urban [STRATUM 2] Estimated Variables Description of the variables coefficient _intercept_ Constant used in the model 6.848 CHLD1_4P Proportion of Children aged 1-4 yrs. In the household -0.8179 DELECTRIC_1 Access to electricity in the household 0.3266 DHD_MARIED_1 Head of the household married 0.3355 DHD_SEX_1 Head of the household is a male person 0.4747 DJSEC_EDU_1 Head of the household with junior secondary education -0.1964 DPUCCA_1 Main house of the household is pucca (cement/concrete) 0.3795 HD_AGE Age of the head of household in years 0.0077 HD_EDU Education grade completed by head of household 0.0477 MEMBERSQ Squared household size -0.0068 ZL_06 Dummy for zila=06 -0.3761 ZL_42 Dummy for zila=42 -0.2451 27 Chittagong Rural [STRATUM 3] Estimated Variables Description of the variables Coefficient _intercept_ Constant used in the model 7.68 DELECTRIC_1 Access to electricity in the household 0.19 DHD_LIT_1 Head of the household is literate 0.17 DJSEC_EDU_1 Head of the household with junior secondary education -0.08 DNOLATRINE_1 Household with no latrine facility -0.23 DNSLATRINE_1 Household access to non-sanitary latrine -0.10 DOWNED_HH_1 Household owned the house 0.12 DPRI_EDU_1 Head of the household with primary education -0.12 DPUCCA_1 Main house of the household is pucca 0.33 DSEMI_PUCA_1 Main house of the household is semi-pucca 0.20 MEMBER Household size -0.13 MEMBERSQ Household size squared 0.01 N15_59YRP Proportion of 15-59yr. Persons in the household 0.55 N60PLUSP Proportion of elderly people(60 +) in the household 0.44 P11_15FSCHNEW Proportion of female children aged 11-15yrs. attending school 0.40 P11_15MSCHNEW Proportion of male children aged 11-15yrs. attending school 0.41 P6_10MSCHNEW Proportion of male children aged 6-10yrs. attending school 0.19 ZL_13 Dummy for zila=13 -0.38 ZL_19 Dummy for zila=19 -0.31 ZL_75 Dummy for zila=75 0.22 _ZL#012$MEMBER Dummy for zila=12 & household size -0.02 _ZL#030$MEMBER Dummy for zila=30 & household size -0.06 _ZL#030$TPE_MFG Dummy for zila=30 & total persons engaged in manufacturing 0.00 _ZL#051$TPE_MFG Dummy for zila=51 & total persons engaged in manufacturing 0.00 _ZL$DHD_LIT_510 Dummy for zila=51 & head not literate 0.22 _ZL$DHD_LIT_511 Dummy for zila=51 & head literate 0.36 Chittagong Urban [STRATUM 4] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 8.269 CHLD1_4P Proportion of 1-4 yrs. Children in the household -0.4084 DELECTRIC_1 Access to electricity in the household 0.1501 DGRA_EDU_1 Head of the household is a graduate 0.3995 DJSEC_EDU_1 Head of the household with junior secondary education 0.1086 DOWNED_HH_1 Household owns the house 0.1854 DPGRA_EDU_1 Head of the household with post graduate education 0.5051 DSEC_EDU_1 Head of the household with secondary education 0.2458 MEMBER Household size -0.2037 28 MEMBERSQ Household size squared 0.0102 PRENTED_HH_M Proportion of rented households at mauza level 0.4349 PRENTED_HH_UZ Proportion of rented households at upazila level -0.5855 TPE_SERV Total persons engaged in the service sector at upazila level 0 ZL_13 Dummy for zila=13 -0.2853 ZL_30 Dummy for zila=30 0.4325 ZL_51 Dummy for zila=51 -0.4439 ZL_75 Dummy for zila=75 0.3614 Chittagong SMA [STRATUM 5] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 8.6232 DHDNMUSLIM_1 Head of the household is a non-muslim -0.1752 DPUCCA_1 Main house of the household is pucca 0.3666 DSLATRINE_1 Access to Sanitary latrine 0.3569 MEMBER Household size -0.3085 MEMBER2 Household size squared 0.0194 TPE_MFG Total persons engaged in manufacturing at upazila level 0 TPE_MINING Total persons engaged in mining at upazila level -0.0034 TPE_SERV Total persons engaged in service sector at upazila level 0.0001 TPE_TRADE Total persons engaged in trade sector at upazila level 0.0001 TPE_TRANS Total persons engaged in transport sector at upazila level -0.0001 Dhaka Rural [STRATUM 6] Estimated variables Description of the variables Coefficient Intercept Constant used in the model 7.1576 DELECTRIC_1 Access to electricity in the household 0.1253 DGRA_EDU_1 Head of the household is a graduate 0.1643 DHD_MARIED_1 Head of the household is married 0.0937 DHSEC_EDU_1 Head of the household with higher secondary education 0.1897 DISABLEP Proportion of disable persons in the household -0.2741 DSLATRINE_1 Access to sanitary (hygeneic) latrine in the household 0.1213 DTUBEWATER_1 Access to tube-well water in the household 0.1469 MEMBER Household size -0.155 MEMBER2 Household size squared 0.0086 N15_59YRP Proportion of 15-59 yrs. persons in the household 0.5016 N60PLUSP Proportion of elderly (60 +) persons in the household 0.5067 P11_15FSCHNEW Proportion of female children (11-15 yrs.) attending school 0.2686 P11_15MSCHNEW Proportion of male children (11-15 yrs.) attending school 0.4514 29 P6_10FSCHNEW Proportion of female children (6-10 yrs.) attending school 0.1855 PEMPLOYED Proportion of employed people in the household -0.1921 PLITERATE Proportion of literate persons in the household 0.4321 PNOLATRINE_M Proportion of households with no latrine at mauza level -0.3788 POTH_WATER_M Proportion of households access to other water at mauza level 0.8625 TPE_SERV Total persons engaged in service sector at upazila level 0 ZL_48 Dummy for zila==48 0.2313 ZL_56 Dummy for zila=56 0.1661 ZL_68 Dummy for zila=68 0.2553 ZL_72 Dummy for zila=72 0.1665 ZL_86 dummy for zila=86 -0.1547 _TPE_MFG$DELECTRIC#0 Interaction of TPE_MFG & household with no access to electricity 0 _ZL$DELECTRIC_331 Dummy for zila=33 and household with access to electricity 0.27 _ZL$DELECTRIC_820 Dummy for zila=82 and household with no access to electricity -0.1506 _ZL$DELECTRIC_931 Dummy for zila=93 and household with access to electricity 0.2243 _ZL$DHD_LIT_350 Dummy for zila=35 and head of the household not literate -0.1655 _ZL$DHD_LIT_611 Dummy for zila=61 and head of the household is literate -0.0647 _ZL$DHD_LIT_671 Dummy for zila=67 and head of the household is literate 0.2356 Dhaka Urban [STRATUM 7] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 7.6498 CHLD0YRP Proportion of 0 yr. children in the household -0.7232 CHLD1_4P Proportion of 1-4 yr. children in the household -0.6534 DELECTRIC_1 Household with access to electricity 0.2486 DGRA_EDU_1 Head of the household is a graduate 0.3505 DHD_LIT_1 Head of the household is literate 0.1328 DHD_MARIED_1 Head of the household is married 0.2186 DHD_SEX_1 Head of the household is a male person 0.1527 DHSEC_EDU_1 Head of the household with higher secondary education 0.4334 DPGRA_EDU_1 Head of the household with post graduate education 0.542 DPUCCA_1 Main house of the household is pucca 0.2883 DSEC_EDU_1 Head of the household with secondary education 0.2388 MEMBER Household size -0.1501 MEMBER2 Household size squared 0.0081 PNSLATRINE_UN Proportion of households with non-sanitary latrine at union level 0.291 TPE_SERV Total persons engaged in the service sector at upazila level 0 TPE_TRANS Total persons engaged in the transport sector at upazila level -0.0005 ZL_35 Dummy for zila=35 -0.2673 ZL_39 Dummy for zila=39 -0.4058 ZL_89 Dummy for zila=89 -0.4535 30 Dhaka SMA [STRATUM 8] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 12.2286 CHLD1_4P Proportion of 1-4 yr. children in the household -0.6243 DHD_LIT_1 Head of the household is literate 0.2128 DPUCCA_1 Main house of the household is pucca 0.3018 DRENTED_HH_1 House is rented -0.2187 DRENTFREE_1 House is rent free -0.3184 HD_AGE Age of head of the household in years 0.004 MEMBER Household size -0.2165 MEMBER2 Household size squared 0.0133 P11_15MSCHNEW Proportion of male children aged 11 - 15 yrs attending school 0.4581 POWNED_HH_UZ Proportion of households owning a house at upazila level -4.326 PRENTED_HH_UZ Proportion of households with a rented house at upazila level -3.7776 TPE_SERV Total persons engaged in service sectoe at upazila level 0 _ZL#067$TPE_MFG Zila=06 and total persons engaged in manufacturing sector 0 _ZL$DSP_LIT_261 Zila=26 and spouse literate 0.1239 Khulna Rural [STRATUM 9] Variables Description of the variables Coefficient Intercept Constant used in the model 7.0173 CHLD0YRP Proportion of 0 yr. children in the household -0.8907 CHLD1_4P Proportion of 1 - 4 yr. children in the household -0.5747 DELECTRIC_1 Household access to electricity 0.1529 DGRA_EDU_1 Head of the household is a graduate 0.3468 DHSEC_EDU_1 Head of the household with higher secondary education 0.3417 DRENTFREE_1 House is rent free -0.1592 DSEC_EDU_1 Head of the household with secondary education 0.1193 DSLATRINE_1 Household has access to a sanitary latrine 0.1078 DSP_LIT_1 Spouse of the head of the household is literate 0.07 MEMBER Household size -0.0276 N15_59YRP Proportion of 15 - 59 yr. people in the household 0.5395 N60PLUSP Proportion of elderly (60+) people in the household 0.2505 ZL_18 Dummy for zila=18 0.1348 ZL_44 Dummy for zila=44 0.1801 ZL_50 Dummy for zila=50 0.5322 ZL_57 Dummy for zila=57 0.2821 ZL_65 Dummy for zila=65 0.3075 31 Khulna Urban [STRATUM 10] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 6.7661 DGRA_EDU_1 Head of the household is a graduate 0.4683 DHDWID_DIV_1 Head of the household is widowed or divorced -0.2731 DHSEC_EDU_1 Head of the household with a higher secondary education 0.181 DJSEC_EDU_1 Head of the household with a junior secondary education -0.1523 DOWNED_HH_1 Household owned a house 0.0899 DSLATRINE_1 Household has access to sanitary latrine 0.1395 MEMBER Household size -0.0686 N15_59YRP Proportion of 15 - 59 yrs. People in the household 0.2544 PELECTRIC_UN Proportion of households with access to electricity at union level 0.5335 PLITERATE Proportion of literate in the household 0.7097 ZL_44 Dummy for zila=44 0.2076 ZL_50 Dummy for zila=50 0.4239 ZL_55 Dummy for zila=55 0.2281 ZL_57 Dummy for zila=57 0.2212 Khulna SMA [STRATUM 11] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 9.8197 CHLD1_4P Proportion of 1 - 4 yr children in the household -0.9315 DSP_LIT_1 Spouse of the head of household is literate -0.5637 DTUBEWATER_1 Household access to tube-well water -1.511 MEMBER Household size -0.3283 MEMBER2 Household size squared 0.0258 N60PLUSP Proportion of elderly (60+) people in the household -0.5934 POTH_WATER_UZ Proportion of households with access to other water 15.7517 _DSP_LIT#1$HD_AGE Interaction of spouse literate with age of head 0.021 Rajshahi Rural [STRATUM 12] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 7.2713 CHLD0YRP Proportion of 0 yr, children in the household -0.992 CHLD1_4P Proportion of 1 - 4 yr, children in the household -0.7355 CHLD5_14P Proportion of 5 -1 4 yr, children in the household -0.1974 DELECTRIC_1 Household access to electricity 0.1689 DHDNMUSLIM_1 Head of the household is a non-muslim -0.1644 DSLATRINE_1 Household access to a sanitary latrine 0.158 32 HD_AGE Age of the head of household in years 0.0031 HD_EDU Education grade of the head of household 0.0292 MEMBER Household size -0.0403 N15_59YRP Proportion of 15 - 59 yr. people in the household 0.2591 P11_15MSCHNEW Proportion of 11 - 15 yr. male children attending school 0.4315 ZL_10 Dummy for zila=10 0.1353 ZL_64 Dummy for zila=64 0.2467 ZL_70 Dummy for zila=70 0.2006 Rajshahi Urban [STRATUM 13] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 6.1221 CHLD5_14P Proportion of 5 - 14 yr. children in the household 0.8567 DELECTRIC_1 Household access to electricity 0.2616 DHD_LIT_1 Head of the household is literate 0.651 DJSEC_EDU_1 Head of the household with junior secondary education -0.5902 DPRI_EDU_1 Head of the household with primary education -0.5387 DSEC_EDU_1 Head of the household with secondary education -0.3653 DSP_LIT_1 Spouse of the head of household is literate 0.1154 N15_59YRP Proportion of 15 - 59 yr. people in the household 1.3953 N60PLUSP Proportion of elderly (60 +) people in the household 1.2067 ZL_38 Dummy for zila=38 0.2864 ZL_69 Dummy for zila=69 -0.2215 ZL_70 Dummy for zila=70 0.2237 _ZL#038$MEMBER2 Zila=38 and household size squared -0.0164 _ZL#088$MEMBER2 Zila=88 and household size squared -0.0046 Rajshahi SMA [STRATUM 14] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 7.765 CHLD1_4P Proportion of 1 - 4 yr. children in the household -0.7448 DNSLATRINE_1 Household access to a non-sanitary latrine -0.1336 DPUCCA_1 Main house of the household is pucca 0.2769 MEMBER Household size -0.1263 _DHSEC_EDU#1$TPE_MFG Head with higher secondary edu. & total persons in mfg. 0.0003 _DTAP_WATER$DELECTRIC_01 Household not using tap water & access to electricity 0.3542 _HD_AGE$MEMBER2 Interaction of age of head & household size squared 0.0001 33 Sylhet Rural [STRATUM 15] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 6.9912 CHLD1_4P Proportion of 1 - 4 yr. children in the household -0.3627 DELECTRIC_1 Household access to electricity 0.2158 DHDNMUSLIM_1 Head of the household is non-muslim -0.1818 DHD_LIT_1 Head of the household is literate 0.1221 DNOLATRINE_1 Household with no latrine -0.1281 DOWNED_HH_1 Household owned the house 0.1456 DPUCCA_1 Main house of the household is pucca 0.264 DRENTED_HH_1 Houshold with a rented house 0.3454 DSP_LIT_1 Spouse of the head is literate 0.1148 HD_AGE Age of the head of household in years 0.0048 MEMBER Household size -0.1564 MEMBER2 Household size squared 0.0077 N15_59YRP Proportion of 15 - 59 yr. people in the household 0.4352 P11_15FSCHNEW Proportion of 11 - 15 yr female children attending school 0.8364 P11_15MSCHNEW Proportion of 11 - 15 yr male children attending school 0.8373 PELECTRIC_M Mauza level census mean of households with access to ectricity -0.281 PPUCCA_M Mauza level census mean of households with pucca house 0.5721 PTUBEWATER_UZ Upazila level census mean of households using tube-well water 0.5283 Sylhet Urban [STRATUM 16] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 8.1775 CHLD1_4P Proportion of 1 - 4 yr. children in the household -1.0735 DGRA_EDU_1 Head of the household is a graduate 0.1811 DHDNMUSLIM_1 Head of the household is a non-muslim -0.2484 DHD_LIT_1 Head of the household is literate 0.5252 DHSEC_EDU_1 Head of the household with higher secondary education 0.3007 DISABLEP Proportion of disable person in the household 1.06 DNSLATRINE_1 Household access to non-sanitary latrine -0.2589 DPGRA_EDU_1 Head of the household is a post graduate 0.3533 MEMBER Household size -0.0703 Rangpur Rural [STRATUM 17] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 10.1256 DRENTFREE_1 The house of the household is rent free -0.1067 HD_EDU Education grade of the head of household 0.036 34 MEMBER Household size -0.0235 N15_59YRP Proportion of 15 - 59 yr people in the household 0.6397 N60PLUSP Proportion of elderly (60 +) people in the household 0.693 P11_15FSCHNEW Proportion of 11 - 15 yr. female children attending school 0.4768 P11_15MSCHNEW Proportion of 11 - 15 yr. male children attending school 0.7226 PSEMIPUCCA_UN Union level census mean of the households with a semi-pucca house 0.883 PTAP_WATER_M Mauza level census mean of the households with access to tap water -11.6155 Mauza level census mean of the households with access to tube-well PTUBEWATER_M water -3.2322 _ZL#049$DHD_LIT_MEAN_UZ Zila=49 &upazila level census mean of households with literate head -0.9172 _ZL$DGRA_EDU_270 Zila=27 & head of the household not a graduate -0.1687 _ZL$DGRA_EDU_320 Zila=32 & head of the household not a graduate -0.2286 _ZL$DGRA_EDU_850 Zila=85 & head of the household not a graduate -0.2053 _ZL$DGRA_EDU_941 Zila=94 & head of the household is a graduate 0.9244 Rangpur Urban [STRATUM 18] Estimated Variables Description of the variables Coefficient Intercept Constant used in the model 6.8095 CHLD1_4P Proportion of 1 - 4 yr. children in the household -0.3287 DELECTRIC_1 Household has access to electricity 0.2445 DGRA_EDU_1 Head of the household is a graduate 0.495 DHD_LIT_1 Head of the household is literate 0.1494 DHSEC_EDU_1 Head of the household with higher secondary education 0.1775 DPGRA_EDU_1 Head of the household is a post graduate 0.5013 DPUCCA_1 Main house of the household is pucca 0.1909 DSEC_EDU_1 Head of the household with secondary education 0.1951 DSP_LIT_1 Spouse of the head of household is literate 0.0989 MEMBER Household size -0.0486 N15_59YRP Proportion of 15 -59 yr. children in the household 0.3885 N60PLUSP Proportion of elderly people (60 +) in the household 0.5888 PELECTRIC_UZ Upazila level census mean of households with access to electricity 0.5859 ZL_94 Dummy for zila=94 0.2348 35 Annex3: Standard Errors of the Poverty Estimates The HIES utilizes a stratified two stage survey design where primary sampling units (PSUs) are selected within each stratum at the 1st stage, following which individual households are randomly selected within each selected PSU at the 2nd stage.APSU is usually a natural cluster of households—i.e. a Mauza in rural areas and Mahalla in urban areas. In the Poverty Mapping exercise Mauza/Mahalla, which is the lowest administrative unit in Bangladesh, was treated as a cluster.Thisfollows Elbers,Lanjouw, and Lanjouw (ELL), who recommend using the lowest possible administrative unit as cluster for estimation purposes.For comparison purposes, standard errors of poverty estimates derived using the Upazilaas cluster were also computed, and are presented in Table 12. Point estimates of poverty using Upazila as the cluster are in general similar to those estimated with Mauza/Mahalla as the cluster, but with higher standard errors.Table 11 presents a comparison of standard errors of the poverty estimates computed using these two different levels of aggregation. Table 11 :Comparison of standard errors at different level of disaggregation Standard Errors of Poverty Estimates (%) Mauza/Mahalla as cluster Upazila as cluster Percentile Stratum Zila Upazila Stratum Zila Upazila Median 1.2 1.8 2.2 1.6 3.1 7.1 95% 1.6 4.5 5.2 3.4 7.3 13.7 Max 1.9 6.2 10.7 4.0 11.2 20.4 36 Table 12 :Comparison of Poverty Estimates and Standard Errors Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila BAGERHAT SADAR 0.359 0.359 0.014 0.058 CHITALMARI 0.500 0.490 0.019 0.069 FAKIRHAT 0.364 0.358 0.019 0.059 KACHUA 0.425 0.426 0.019 0.075 MOLLAHAT 0.461 0.454 0.023 0.071 MONGLA 0.419 0.422 0.017 0.050 MORRELGANJ 0.465 0.465 0.016 0.063 RAMPAL 0.411 0.405 0.018 0.070 SARANKHOLA 0.480 0.466 0.028 0.087 ALIKADAM 0.429 0.413 0.045 0.124 BANDARBAN SADAR 0.308 0.313 0.024 0.064 LAMA 0.410 0.380 0.031 0.098 NAIKHONGCHHARI 0.460 0.440 0.037 0.114 ROWANGCHHARI 0.329 0.323 0.036 0.100 RUMA 0.423 0.398 0.037 0.100 THANCHI 0.530 0.514 0.042 0.107 AMTALI 0.228 0.205 0.017 0.057 BAMNA 0.171 0.153 0.017 0.045 BARGUNA SADAR 0.192 0.183 0.014 0.049 BETAGI 0.196 0.165 0.016 0.047 PATHARGHATA 0.129 0.127 0.012 0.031 AGAILJHARA 0.511 0.498 0.023 0.075 BABUGANJ 0.487 0.467 0.022 0.081 BAKERGANJ 0.554 0.540 0.019 0.070 BANARI PARA 0.522 0.497 0.023 0.085 GAURNADI 0.555 0.550 0.018 0.062 HIZLA 0.623 0.591 0.021 0.076 BARISAL SADAR (KOTWALI) 0.499 0.511 0.019 0.047 MHENDIGANJ 0.644 0.624 0.021 0.070 MULADI 0.582 0.569 0.022 0.072 WAZIRPUR 0.521 0.513 0.023 0.075 BHOLA SADAR 0.492 0.519 0.030 0.070 BURHANUDDIN 0.283 0.263 0.017 0.064 CHAR FASSON 0.282 0.295 0.021 0.073 DAULAT KHAN 0.303 0.312 0.020 0.059 LALMOHAN 0.278 0.253 0.018 0.063 MANPURA 0.328 0.318 0.027 0.081 TAZUMUDDIN 0.223 0.213 0.024 0.063 37 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila ADAMDIGHI 0.131 0.121 0.013 0.031 BOGRA SADAR 0.176 0.150 0.008 0.024 DHUNAT 0.198 0.214 0.022 0.061 DHUPCHANCHIA 0.132 0.127 0.014 0.035 GABTALI 0.156 0.156 0.019 0.039 KAHALOO 0.117 0.116 0.015 0.038 NANDIGRAM 0.161 0.161 0.016 0.044 SARIAKANDI 0.216 0.210 0.022 0.053 SHAJAHANPUR 0.125 0.115 0.012 0.032 SHERPUR 0.157 0.152 0.019 0.042 SHIBGANJ 0.169 0.160 0.017 0.043 SONATOLA 0.237 0.228 0.020 0.045 AKHAURA 0.269 0.246 0.023 0.071 BANCHHARAMPUR 0.273 0.260 0.031 0.096 BIJOYNAGAR 0.358 0.337 0.031 0.110 BRAHMANBARIA SADAR 0.260 0.245 0.022 0.066 ASHUGANJ 0.218 0.205 0.033 0.091 KASBA 0.255 0.228 0.025 0.076 NABINAGAR 0.305 0.287 0.028 0.077 NASIRNAGAR 0.437 0.417 0.034 0.101 SARAIL 0.311 0.305 0.035 0.089 CHANDPUR SADAR 0.455 0.414 0.042 0.082 FARIDGANJ 0.466 0.463 0.050 0.100 HAIM CHAR 0.613 0.606 0.056 0.129 HAJIGANJ 0.537 0.509 0.047 0.099 KACHUA 0.563 0.522 0.051 0.137 MATLAB DAKSHIN 0.537 0.515 0.042 0.089 MATLAB UTTAR 0.499 0.509 0.051 0.091 SHAHRASTI 0.505 0.499 0.050 0.119 ANOWARA 0.155 0.166 0.019 0.049 BAYEJID BOSTAMI 0.092 0.083 0.012 0.014 BANSHKHALI 0.279 0.284 0.025 0.084 BAKALIA 0.049 0.042 0.011 0.013 BOALKHALI 0.105 0.100 0.015 0.028 CHANDANAISH 0.135 0.136 0.016 0.046 CHANDGAON 0.169 0.172 0.022 0.022 CHITTAGONG PORT 0.124 0.104 0.020 0.018 DOUBLE MOORING 0.000 0.000 0.000 0.000 FATIKCHHARI 0.176 0.179 0.016 0.067 HALISHAHAR 0.056 0.051 0.010 0.015 38 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila HATHAZARI 0.011 0.008 0.004 0.004 KOTWALI 0.003 0.001 0.002 0.002 KHULSHI 0.011 0.009 0.004 0.004 LOHAGARA 0.183 0.195 0.021 0.060 MIRSHARAI 0.134 0.152 0.015 0.049 PAHARTALI 0.300 0.312 0.049 0.048 PANCHLAISH 0.008 0.005 0.004 0.005 PATIYA 0.081 0.097 0.011 0.040 PATENGA 0.039 0.031 0.009 0.008 RANGUNIA 0.140 0.149 0.014 0.044 RAOZAN 0.085 0.091 0.009 0.027 SANDWIP 0.191 0.194 0.021 0.057 SATKANIA 0.152 0.155 0.015 0.053 SITAKUNDA 0.115 0.135 0.013 0.040 ALAMDANGA 0.260 0.247 0.036 0.064 CHUADANGA SADAR 0.292 0.267 0.030 0.057 DAMURHUDA 0.271 0.258 0.034 0.074 JIBAN NAGAR 0.291 0.279 0.036 0.080 BARURA 0.379 0.355 0.028 0.094 BRAHMAN PARA 0.399 0.374 0.035 0.119 BURICHANG 0.333 0.321 0.030 0.119 CHANDINA 0.412 0.398 0.030 0.086 CHAUDDAGRAM 0.344 0.340 0.027 0.101 COMILLA SADAR DAKSHIN 0.333 0.335 0.022 0.072 DAUDKANDI 0.385 0.381 0.028 0.099 DEBIDWAR 0.414 0.394 0.027 0.089 HOMNA 0.383 0.374 0.037 0.100 COMILLA ADARSHA SADAR 0.244 0.249 0.017 0.053 LAKSAM 0.374 0.365 0.028 0.094 MANOHARGANJ 0.471 0.464 0.034 0.111 MEGHNA 0.373 0.364 0.042 0.122 MURADNAGAR 0.450 0.439 0.031 0.112 NANGALKOT 0.451 0.448 0.033 0.117 TITAS 0.377 0.375 0.038 0.111 CHAKARIA 0.285 0.268 0.022 0.086 COX'S BAZAR SADAR 0.262 0.259 0.020 0.056 KUTUBDIA 0.311 0.292 0.044 0.103 MAHESHKHALI 0.402 0.382 0.034 0.107 PEKUA 0.309 0.280 0.034 0.096 RAMU 0.343 0.331 0.027 0.093 39 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila TEKNAF 0.382 0.367 0.034 0.095 UKHIA 0.378 0.351 0.033 0.104 ADABOR 0.125 0.192 0.025 0.080 BADDA 0.134 0.143 0.033 0.067 BANGSHAL 0.094 0.075 0.028 0.053 BIMAN BANDAR 0.013 0.105 0.010 0.052 CANTONMENT 0.015 0.018 0.008 0.018 CHAK BAZAR 0.107 0.101 0.028 0.054 DAKSHINKHAN 0.246 0.165 0.061 0.075 DARUS SALAM 0.142 0.193 0.027 0.081 DEMRA 0.199 0.165 0.043 0.079 DHAMRAI 0.228 0.255 0.018 0.080 DHANMONDI 0.014 0.010 0.006 0.012 DOHAR 0.239 0.284 0.021 0.091 GENDARIA 0.093 0.101 0.027 0.056 GULSHAN 0.033 0.089 0.013 0.049 HAZARIBAGH 0.122 0.131 0.020 0.061 JATRABARI 0.116 0.134 0.029 0.069 KAFRUL 0.070 0.099 0.015 0.053 KADAMTALI 0.150 0.141 0.040 0.063 KALABAGAN 0.101 0.082 0.026 0.051 KAMRANGIR CHAR 0.220 0.201 0.057 0.085 KHILGAON 0.137 0.138 0.028 0.063 KHILKHET 0.147 0.155 0.033 0.076 KERANIGANJ 0.259 0.259 0.035 0.099 KOTWALI 0.059 0.040 0.014 0.037 LALBAGH 0.160 0.142 0.028 0.067 MIRPUR 0.067 0.069 0.014 0.040 MOHAMMADPUR 0.040 0.021 0.015 0.021 MOTIJHEEL 0.013 0.004 0.008 0.007 NAWABGANJ 0.211 0.263 0.021 0.106 NEW MARKET 0.037 0.034 0.017 0.034 PALLABI 0.120 0.122 0.020 0.062 PALTAN 0.027 0.004 0.015 0.006 RAMNA 0.038 0.026 0.012 0.026 RAMPURA 0.102 0.131 0.025 0.063 SABUJBAGH 0.116 0.118 0.027 0.060 SAVAR 0.340 0.289 0.101 0.099 SHAH ALI 0.157 0.159 0.023 0.075 SHAHBAGH 0.015 0.019 0.007 0.020 40 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila SHYAMPUR 0.129 0.151 0.031 0.064 SHER-E-BANGLA NAGAR 0.077 0.145 0.022 0.065 SUTRAPUR 0.046 0.034 0.011 0.030 TEJGAON 0.053 0.050 0.019 0.033 TEJGAON IND. AREA 0.067 0.119 0.021 0.062 TURAG 0.251 0.196 0.039 0.090 UTTARA 0.037 0.064 0.015 0.036 UTTAR KHAN 0.249 0.160 0.083 0.077 BIRAMPUR 0.359 0.402 0.032 0.111 BIRGANJ 0.431 0.471 0.044 0.156 BIRAL 0.388 0.479 0.049 0.153 BOCHAGANJ 0.384 0.428 0.044 0.124 CHIRIRBANDAR 0.385 0.477 0.043 0.146 FULBARI 0.338 0.426 0.041 0.109 GHORAGHAT 0.418 0.483 0.036 0.126 HAKIMPUR 0.389 0.382 0.031 0.112 KAHAROLE 0.443 0.476 0.049 0.136 KHANSAMA 0.465 0.474 0.058 0.153 DINAJPUR SADAR 0.282 0.335 0.021 0.097 NAWABGANJ 0.373 0.466 0.045 0.157 PARBATIPUR 0.397 0.430 0.040 0.138 ALFADANGA 0.299 0.342 0.023 0.106 BHANGA 0.335 0.378 0.018 0.086 BOALMARI 0.393 0.427 0.017 0.080 CHAR BHADRASAN 0.358 0.382 0.047 0.078 FARIDPUR SADAR 0.383 0.398 0.042 0.070 MADHUKHALI 0.305 0.357 0.022 0.089 NAGARKANDA 0.359 0.379 0.022 0.089 SADARPUR 0.369 0.392 0.025 0.084 SALTHA 0.421 0.438 0.029 0.086 CHHAGALNAIYA 0.259 0.213 0.026 0.069 DAGANBHUIYAN 0.163 0.259 0.028 0.081 FENI SADAR 0.186 0.230 0.022 0.065 FULGAZI 0.318 0.253 0.031 0.089 PARSHURAM 0.306 0.244 0.025 0.067 SONAGAZI 0.445 0.375 0.032 0.096 FULCHHARI 0.581 0.637 0.053 0.143 GAIBANDHA SADAR 0.448 0.490 0.041 0.146 GOBINDAGANJ 0.454 0.506 0.036 0.144 PALASHBARI 0.448 0.543 0.042 0.144 41 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila SADULLAPUR 0.510 0.522 0.046 0.162 SAGHATA 0.528 0.537 0.043 0.143 SUNDARGANJ 0.476 0.546 0.050 0.145 GAZIPUR SADAR 0.221 0.226 0.107 0.073 KALIAKAIR 0.110 0.137 0.022 0.041 KALIGANJ 0.157 0.144 0.019 0.054 KAPASIA 0.270 0.188 0.021 0.077 SREEPUR 0.144 0.160 0.023 0.046 GOPALGANJ SADAR 0.411 0.429 0.021 0.071 KASHIANI 0.391 0.424 0.021 0.092 KOTALIPARA 0.436 0.432 0.023 0.080 MUKSUDPUR 0.465 0.483 0.020 0.091 TUNGIPARA 0.426 0.439 0.026 0.080 AJMIRIGANJ 0.326 0.319 0.019 0.082 BAHUBAL 0.241 0.291 0.016 0.089 BANIACHONG 0.276 0.308 0.015 0.096 CHUNARUGHAT 0.275 0.278 0.014 0.069 HABIGANJ SADAR 0.169 0.186 0.011 0.050 LAKHAI 0.252 0.293 0.020 0.086 MADHABPUR 0.259 0.255 0.014 0.072 NABIGANJ 0.268 0.261 0.014 0.066 AKKELPUR 0.269 0.273 0.011 0.046 JOYPURHAT SADAR 0.260 0.259 0.010 0.045 KALAI 0.256 0.259 0.013 0.043 KHETLAL 0.261 0.263 0.015 0.052 PANCHBIBI 0.283 0.277 0.013 0.051 BAKSHIGANJ 0.504 0.493 0.029 0.102 DEWANGANJ 0.585 0.556 0.033 0.089 ISLAMPUR 0.550 0.503 0.023 0.076 JAMALPUR SADAR 0.498 0.494 0.024 0.072 MADARGANJ 0.555 0.553 0.020 0.075 MELANDAHA 0.472 0.450 0.016 0.089 SARISHABARI UPAZILA 0.447 0.438 0.018 0.080 ABHAYNAGAR 0.360 0.361 0.017 0.056 BAGHER PARA 0.425 0.430 0.020 0.084 CHAUGACHHA 0.428 0.443 0.018 0.087 JHIKARGACHHA 0.389 0.390 0.020 0.075 KESHABPUR 0.420 0.433 0.017 0.071 JESSORE SADAR 0.353 0.333 0.013 0.057 MANIRAMPUR 0.402 0.391 0.019 0.077 42 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila SHARSHA 0.408 0.404 0.018 0.072 JHALOKATI SADAR 0.377 0.363 0.023 0.064 KANTHALIA 0.342 0.371 0.029 0.079 NALCHITY 0.465 0.460 0.020 0.066 RAJAPUR 0.420 0.419 0.024 0.069 HARINAKUNDA 0.260 0.254 0.022 0.056 JHENAIDAH SADAR 0.239 0.228 0.018 0.054 KALIGANJ 0.240 0.221 0.019 0.056 KOTCHANDPUR 0.202 0.214 0.016 0.045 MAHESHPUR 0.236 0.230 0.022 0.064 SHAILKUPA 0.282 0.279 0.020 0.059 DIGHINALA 0.225 0.226 0.034 0.091 KHAGRACHHARI SADAR 0.195 0.164 0.020 0.058 LAKSHMICHHARI 0.310 0.317 0.035 0.107 MAHALCHHARI 0.214 0.192 0.033 0.096 MANIKCHHARI 0.301 0.271 0.046 0.109 MATIRANGA 0.283 0.268 0.028 0.092 PANCHHARI 0.234 0.214 0.040 0.097 RAMGARH 0.326 0.273 0.033 0.080 BATIAGHATA 0.405 0.402 0.016 0.063 DACOPE 0.445 0.443 0.019 0.071 DAULATPUR 0.345 0.258 0.020 0.066 DUMURIA 0.372 0.381 0.016 0.069 DIGHALIA 0.393 0.379 0.024 0.066 KHALISHPUR 0.411 0.255 0.026 0.058 KHAN JAHAN ALI 0.319 0.277 0.028 0.063 KHULNA SADAR 0.355 0.216 0.021 0.061 KOYRA 0.491 0.494 0.021 0.080 PAIKGACHHA 0.424 0.421 0.017 0.068 PHULTALA 0.337 0.328 0.025 0.060 RUPSA 0.369 0.372 0.015 0.068 SONADANGA 0.193 0.205 0.023 0.057 TEROKHADA 0.496 0.488 0.026 0.070 AUSTAGRAM 0.337 0.281 0.034 0.070 BAJITPUR 0.282 0.236 0.026 0.064 BHAIRAB 0.339 0.287 0.022 0.055 HOSSAINPUR 0.330 0.279 0.034 0.061 ITNA 0.349 0.300 0.034 0.082 KARIMGANJ 0.271 0.251 0.027 0.079 KATIADI 0.316 0.287 0.029 0.079 43 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila KISHOREGANJ SADAR 0.276 0.248 0.022 0.061 KULIAR CHAR 0.327 0.306 0.033 0.057 MITHAMAIN 0.352 0.328 0.038 0.098 NIKLI 0.300 0.261 0.038 0.082 PAKUNDIA 0.261 0.242 0.024 0.058 TARAIL 0.261 0.242 0.033 0.073 BHURUNGAMARI 0.651 0.635 0.052 0.191 CHAR RAJIBPUR 0.688 0.681 0.057 0.168 CHILMARI 0.611 0.685 0.050 0.154 PHULBARI 0.685 0.656 0.056 0.204 KURIGRAM SADAR 0.580 0.593 0.035 0.130 NAGESHWARI 0.650 0.669 0.040 0.146 RAJARHAT 0.677 0.683 0.066 0.196 RAUMARI 0.570 0.646 0.049 0.181 ULIPUR 0.653 0.667 0.048 0.169 BHERAMARA 0.034 0.028 0.008 0.017 DAULATPUR 0.040 0.035 0.009 0.021 KHOKSA 0.047 0.054 0.008 0.024 KUMARKHALI 0.040 0.035 0.008 0.020 KUSHTIA SADAR 0.030 0.027 0.005 0.015 MIRPUR 0.033 0.029 0.007 0.019 KAMALNAGAR 0.187 0.341 0.040 0.102 LAKSHMIPUR SADAR 0.456 0.240 0.047 0.068 ROYPUR 0.167 0.250 0.022 0.066 RAMGANJ 0.214 0.237 0.022 0.053 RAMGATI 0.304 0.355 0.038 0.079 ADITMARI 0.360 0.370 0.029 0.128 HATIBANDHA 0.381 0.347 0.033 0.129 KALIGANJ 0.353 0.324 0.030 0.123 LALMONIRHAT SADAR 0.313 0.331 0.021 0.116 PATGRAM 0.333 0.394 0.025 0.116 KALKINI 0.332 0.349 0.019 0.096 MADARIPUR SADAR 0.350 0.351 0.019 0.090 RAJOIR 0.314 0.334 0.025 0.102 SHIB CHAR 0.388 0.390 0.021 0.110 MAGURA SADAR 0.430 0.421 0.012 0.054 MOHAMMADPUR 0.508 0.508 0.016 0.070 SHALIKHA 0.442 0.448 0.015 0.067 DAULATPUR 0.294 0.297 0.039 0.094 GHIOR 0.137 0.159 0.024 0.071 44 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila SREEPUR 0.450 0.437 0.019 0.078 HARIRAMPUR 0.181 0.177 0.030 0.073 MANIKGANJ SADAR 0.187 0.185 0.029 0.067 SATURIA 0.150 0.152 0.027 0.072 SHIBALAYA 0.158 0.166 0.027 0.065 SINGAIR 0.181 0.184 0.025 0.052 GANGNI 0.158 0.175 0.020 0.054 MUJIB NAGAR 0.136 0.135 0.023 0.051 MEHERPUR SADAR 0.151 0.176 0.018 0.048 BARLEKHA 0.257 0.221 0.015 0.066 JURI 0.363 0.290 0.021 0.073 KAMALGANJ 0.267 0.264 0.014 0.067 KULAURA 0.281 0.228 0.016 0.064 MAULVIBAZAR SADAR 0.167 0.185 0.010 0.064 RAJNAGAR 0.223 0.245 0.013 0.084 SREEMANGAL 0.293 0.274 0.014 0.066 GAZARIA 0.268 0.290 0.023 0.066 LOHAJANG 0.336 0.323 0.031 0.096 MUNSHIGANJ SADAR 0.308 0.289 0.020 0.068 SERAJDIKHAN 0.288 0.262 0.023 0.087 SREENAGAR 0.263 0.264 0.027 0.097 TONGIBARI 0.251 0.263 0.024 0.078 BHALUKA 0.311 0.343 0.024 0.110 DHOBAURA 0.582 0.575 0.034 0.105 FULBARIA 0.526 0.514 0.036 0.107 GAFFARGAON 0.439 0.487 0.024 0.116 GAURIPUR 0.506 0.516 0.030 0.093 HALUAGHAT 0.503 0.538 0.025 0.126 ISHWARGANJ 0.560 0.549 0.029 0.092 MYMENSINGH SADAR 0.523 0.503 0.057 0.081 MUKTAGACHHA 0.433 0.466 0.021 0.095 NANDAIL 0.607 0.595 0.048 0.092 PHULPUR 0.588 0.588 0.023 0.102 TRISHAL 0.478 0.482 0.026 0.110 ATRAI 0.135 0.147 0.013 0.039 BADALGACHHI 0.150 0.161 0.012 0.047 DHAMOIRHAT 0.179 0.190 0.012 0.041 NAOGAON SADAR 0.174 0.205 0.009 0.028 NIAMATPUR 0.194 0.201 0.013 0.046 PATNITALA 0.186 0.197 0.012 0.041 45 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila PORSHA 0.217 0.226 0.017 0.051 MANDA 0.147 0.149 0.013 0.044 MAHADEBPUR 0.156 0.163 0.013 0.048 RANINAGAR 0.133 0.139 0.013 0.043 SAPAHAR 0.214 0.217 0.016 0.048 KALIA 0.233 0.217 0.022 0.055 LOHAGARA 0.199 0.216 0.020 0.061 NARAIL SADAR 0.173 0.155 0.018 0.039 ARAIHAZAR 0.344 0.355 0.024 0.074 SONARGAON 0.213 0.224 0.012 0.070 BANDAR 0.209 0.343 0.036 0.104 NARAYANGANJ SADAR 0.279 0.312 0.052 0.102 RUPGANJ 0.225 0.246 0.013 0.050 BELABO 0.219 0.188 0.035 0.068 MANOHARDI 0.227 0.204 0.029 0.069 NARSINGDI SADAR 0.228 0.225 0.028 0.061 PALASH 0.222 0.185 0.024 0.038 ROYPURA 0.294 0.267 0.032 0.081 SHIBPUR 0.189 0.166 0.031 0.056 BAGATIPARA 0.316 0.313 0.023 0.056 BARAIGRAM 0.361 0.365 0.018 0.051 GURUDASPUR 0.370 0.378 0.018 0.056 LALPUR 0.357 0.367 0.017 0.053 NATORE SADAR 0.318 0.320 0.019 0.045 SINGRA 0.378 0.374 0.015 0.052 BHOLAHAT 0.208 0.221 0.021 0.061 GOMASTAPUR 0.261 0.272 0.020 0.054 NACHOLE 0.242 0.250 0.018 0.048 CHAPAI NABABGANJ SADAR 0.254 0.264 0.016 0.048 SHIBGANJ 0.260 0.264 0.021 0.053 ATPARA 0.316 0.355 0.038 0.105 BARHATTA 0.352 0.369 0.041 0.108 DURGAPUR 0.302 0.382 0.049 0.083 KHALIAJURI 0.372 0.409 0.043 0.101 KALMAKANDA 0.376 0.392 0.040 0.105 KENDUA 0.409 0.379 0.042 0.079 PURBADHALA 0.354 0.351 0.038 0.084 DIMLA 0.352 0.361 0.031 0.147 DOMAR 0.313 0.360 0.025 0.113 JALDHAKA 0.435 0.427 0.027 0.120 46 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila KISHOREGANJ 0.309 0.360 0.027 0.122 MADAN 0.416 0.415 0.042 0.084 MOHANGANJ 0.343 0.368 0.035 0.078 NETROKONA SADAR 0.308 0.312 0.030 0.084 NILPHAMARI SADAR 0.364 0.341 0.023 0.125 SAIDPUR 0.277 0.279 0.017 0.062 BEGUMGANJ 0.059 0.088 0.009 0.037 CHATKHIL 0.048 0.070 0.008 0.027 COMPANIGANJ 0.076 0.113 0.014 0.042 HATIYA 0.160 0.193 0.025 0.079 KABIRHAT 0.124 0.156 0.018 0.058 SENBAGH 0.054 0.085 0.010 0.039 SONAIMURI 0.050 0.081 0.008 0.032 SUBARNACHAR 0.187 0.228 0.025 0.088 NOAKHALI SADAR 0.102 0.128 0.014 0.047 ATGHARIA 0.312 0.318 0.013 0.052 BERA 0.394 0.389 0.013 0.043 BHANGURA 0.335 0.325 0.011 0.051 CHATMOHAR 0.314 0.311 0.013 0.061 FARIDPUR 0.315 0.318 0.017 0.043 ISHWARDI 0.262 0.260 0.012 0.041 PABNA SADAR 0.278 0.286 0.011 0.045 SANTHIA 0.331 0.334 0.011 0.055 SUJANAGAR 0.354 0.353 0.012 0.058 ATWARI 0.241 0.290 0.024 0.106 BODA 0.266 0.295 0.021 0.110 DEBIGANJ 0.342 0.364 0.025 0.129 PANCHAGARH SADAR 0.242 0.305 0.023 0.107 TENTULIA 0.215 0.298 0.026 0.129 BAUPHAL 0.240 0.219 0.014 0.056 DASHMINA 0.218 0.204 0.020 0.068 DUMKI 0.220 0.225 0.019 0.052 GALACHIPA 0.260 0.237 0.016 0.058 KALA PARA 0.203 0.216 0.018 0.052 MIRZAGANJ 0.178 0.140 0.016 0.047 PATUAKHALI SADAR 0.369 0.395 0.024 0.071 NAZIRPUR 0.515 0.531 0.026 0.077 PIROJPUR SADAR 0.427 0.427 0.017 0.047 NESARABAD (SWARUPKATI) 0.433 0.445 0.020 0.074 ZIANAGAR 0.491 0.504 0.025 0.064 47 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila BAGHA 0.336 0.364 0.011 0.039 BHANDARIA 0.420 0.433 0.023 0.067 KAWKHALI 0.522 0.536 0.022 0.075 MATHBARIA 0.380 0.389 0.026 0.072 BAGHMARA 0.294 0.309 0.011 0.049 BOALIA 0.241 0.173 0.019 0.035 CHARGHAT 0.314 0.335 0.012 0.042 DURGAPUR 0.257 0.275 0.013 0.044 GODAGARI 0.441 0.460 0.010 0.056 MATIHAR 0.333 0.288 0.027 0.062 MOHANPUR 0.249 0.268 0.013 0.058 PABA 0.334 0.316 0.012 0.046 PUTHIA 0.268 0.284 0.013 0.047 RAJPARA 0.244 0.185 0.020 0.037 SHAH MAKHDUM 0.309 0.263 0.024 0.045 TANORE 0.357 0.393 0.010 0.044 BALIAKANDI 0.397 0.322 0.029 0.117 GOALANDA 0.505 0.421 0.032 0.094 KALUKHALI 0.396 0.315 0.032 0.110 PANGSHA 0.457 0.384 0.025 0.092 RAJBARI SADAR 0.387 0.321 0.024 0.090 BAGHAICHHARI 0.248 0.223 0.037 0.096 BARKAL 0.261 0.221 0.037 0.108 KAWKHALI (BETBUNIA) 0.234 0.199 0.037 0.094 BELAI CHHARI 0.347 0.333 0.064 0.128 KAPTAI 0.122 0.113 0.027 0.059 JURAI CHHARI 0.193 0.177 0.042 0.080 LANGADU 0.293 0.303 0.038 0.122 NANIARCHAR 0.212 0.194 0.033 0.094 RAJASTHALI 0.205 0.196 0.039 0.086 RANGAMATI SADAR 0.073 0.075 0.014 0.037 BADARGANJ 0.483 0.495 0.032 0.122 GANGACHARA 0.583 0.520 0.040 0.113 KAUNIA 0.450 0.503 0.026 0.096 RANGPUR SADAR 0.371 0.382 0.017 0.078 MITHA PUKUR 0.454 0.489 0.030 0.138 PIRGACHHA 0.497 0.483 0.038 0.152 PIRGANJ 0.469 0.517 0.038 0.135 TARAGANJ 0.524 0.507 0.040 0.135 BHEDARGANJ 0.563 0.493 0.062 0.104 48 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila DAMUDYA 0.479 0.406 0.058 0.111 GOSAIRHAT 0.583 0.534 0.061 0.105 NARIA 0.481 0.419 0.053 0.098 SHARIATPUR SADAR 0.498 0.445 0.049 0.095 ZANJIRA 0.540 0.452 0.057 0.110 ASSASUNI 0.484 0.477 0.016 0.066 DEBHATA 0.431 0.425 0.023 0.062 KALAROA 0.460 0.450 0.013 0.056 KALIGANJ 0.480 0.483 0.014 0.062 SATKHIRA SADAR 0.431 0.431 0.013 0.051 SHYAMNAGAR 0.502 0.498 0.016 0.054 TALA 0.452 0.447 0.015 0.063 BELKUCHI 0.425 0.423 0.013 0.051 CHAUHALI 0.455 0.458 0.015 0.066 KAMARKHANDA 0.325 0.326 0.015 0.059 KAZIPUR 0.362 0.368 0.014 0.057 ROYGANJ 0.394 0.399 0.012 0.064 SHAHJADPUR 0.418 0.414 0.012 0.062 SIRAJGANJ SADAR 0.367 0.364 0.010 0.043 TARASH 0.358 0.370 0.014 0.061 ULLAH PARA 0.366 0.361 0.012 0.050 JHENAIGATI 0.369 0.411 0.030 0.101 NAKLA 0.468 0.484 0.023 0.076 NALITABARI 0.418 0.455 0.022 0.091 SHERPUR SADAR 0.558 0.516 0.024 0.074 SREEBARDI 0.491 0.476 0.024 0.104 BISHWAMBARPUR 0.304 0.351 0.025 0.099 CHHATAK 0.236 0.257 0.012 0.064 DAKSHIN SUNAMGANJ 0.244 0.299 0.017 0.087 DERAI 0.262 0.323 0.016 0.085 DHARAMPASHA 0.255 0.352 0.022 0.102 DOWARABAZAR 0.299 0.308 0.019 0.093 JAGANNATHPUR 0.210 0.268 0.014 0.070 JAMALGANJ 0.246 0.329 0.021 0.097 SULLA 0.283 0.362 0.023 0.098 SUNAMGANJ SADAR 0.251 0.282 0.014 0.072 TAHIRPUR 0.312 0.388 0.022 0.110 BALAGANJ 0.197 0.180 0.013 0.066 BEANI BAZAR 0.159 0.155 0.010 0.051 BISHWANATH 0.125 0.164 0.012 0.059 49 Poverty Headcount Rate Standard Error of Estimate Cluster based on Cluster based Cluster based on Cluster based Name of Upazila Mauza/Mahalla on Upazila Mauza/Mahalla on Upazila COMPANIGANJ 0.345 0.354 0.021 0.102 DAKSHIN SURMA 0.103 0.133 0.010 0.050 FENCHUGANJ 0.169 0.182 0.017 0.062 GOLAPGANJ 0.149 0.161 0.012 0.057 GOWAINGHAT 0.526 0.354 0.035 0.088 JAINTIAPUR 0.347 0.279 0.020 0.081 KANAIGHAT 0.458 0.245 0.041 0.070 SYLHET SADAR 0.143 0.149 0.007 0.028 ZAKIGANJ 0.390 0.250 0.032 0.074 BASAIL 0.197 0.168 0.018 0.072 BHUAPUR 0.344 0.291 0.014 0.071 DELDUAR 0.243 0.205 0.016 0.069 DHANBARI 0.370 0.305 0.018 0.075 GHATAIL 0.287 0.246 0.015 0.077 GOPALPUR 0.293 0.261 0.014 0.069 KALIHATI 0.235 0.241 0.017 0.078 MADHUPUR 0.364 0.304 0.016 0.082 MIRZAPUR 0.267 0.252 0.020 0.073 NAGARPUR 0.399 0.313 0.015 0.094 SAKHIPUR 0.260 0.219 0.023 0.068 TANGAIL SADAR 0.317 0.270 0.026 0.060 BALIADANGI 0.265 0.298 0.034 0.132 HARIPUR 0.297 0.345 0.035 0.135 PIRGANJ 0.233 0.296 0.029 0.110 RANISANKAIL 0.258 0.324 0.035 0.118 THAKURGAON SADAR 0.286 0.267 0.020 0.112 50 Annex 4. Results of Multi-Layer Analysis Table13 bellow presents the results of a multi-layer analysis done with multilevel mixed-effects linear regression using the STATA command XTMIXED. The objective of this analysis is to see impact of the errors at the union and upazila levels. The table shows the shares of variance of errors by stratum at different levels of clustering. The table clearly indicates that most of the errors (more than 95%) are concentrated at the layers or levels of household and mauza and PovMap2 (software used for poverty mapping) can take explicit account of these two levels of errors. Table 13 :Shares of Variance of errors in each layer (%) Three layers Two layers One layer model model model Stratum UZ UN MZA HH All UZ MZA HH All MZA HH All Barisal (Rural) 3 0 0 97 100 3 0 97 100 3 97 100 Barisal (Urban) 0 2 2 96 100 0 4 96 100 4 96 100 Chittagong (Rural) 7 0 0 93 100 7 0 93 100 7 93 100 Chittagong (Urban) 0 1 1 98 100 0 3 97 100 3 97 100 Chittagong (SMA) 0 1 1 98 100 0 2 98 100 2 98 100 Dhaka (Rural) 1 3 4 92 100 1 7 92 100 8 92 100 Dhaka (Urban) 0 2 2 96 100 0 4 96 100 4 96 100 Dhaka (SMA) Not Converging Khulna (Rural) 2 1 1 96 100 2 2 96 100 4 96 100 Khulna (Urban) 0 0 1 99 100 0 1 99 100 1 99 100 Khulna (SMA) 0 1 1 98 100 0 2 98 100 2 98 100 Rajshahi (Rural) 0 2 2 96 100 0 3 97 100 3 97 100 Rajshahi (Urban) 0 2 2 96 100 0 3 97 100 3 97 100 Rajshahi (SMA) 0 2 2 96 100 0 4 96 100 4 96 100 Sylhet (Rural) 0 2 2 96 100 0 4 96 100 4 96 100 Sylhet (Urban) 0 1 1 98 100 0 2 98 100 2 98 100 Rangpur (Rural) 0 4 3 93 100 0 7 93 100 7 93 100 Rangpur (Urban) 2 2 2 94 100 2 4 94 100 6 94 100 51 Annex 5: Detailed Methodology on SAE Box 1: The Small Area Estimation Method Developed by ELL (2003) The method proposed by ELL has two stages. In the first part, a model of log per capita consumption expenditure ( ln y ch ) is estimated in the survey data:  ln y ch  X ch �  Z �  u ch  where X ch is the vector of explanatory variables for household h in cluster c, is the vector of regression coefficients, Z  is the vector of location specific variables, � is the vector of coefficients, and u ch is the regression disturbances due to the discrepancy between the predicted household consumption and the actual value. This disturbance term is decomposed into two independent components: u ch  � c  � ch where a cluster-specific effect, � c and a household- specific effect, � ch . This error structure allows for both a location effect – common to all households in the same area—and heteroskedasticity in the household-specific errors. The location variables can be any level – Zila, Upazila, Union, Mauza, and Village – and can be drawn from any data sources that include all locations in the country. All parameters regarding the regression coefficients ( � , � )and distributions of the disturbance terms are estimated by Feasible Generalized Least Square (FGLS). In the second part of the analysis, poverty estimates and their standard errors are computed. There are two sources of errors involved in the estimation process: errors in the estimated regression coefficients ( �ˆ , �ˆ ) and the disturbance terms, both of which affect poverty estimates and the level of their accuracy. ELL propose a way to properly calculate poverty estimates as well as their standard errors while taking into account these sources of bias. A simulated value of expenditure for each census household is calculated with predicted log ˆ expenditure X � ch  Z �ˆ and random draws from the estimated distributions of the disturbance terms, � c and � ch . These simulations are repeated 100 times. For any given location (such as a zila or an upazila), the mean across the 100 simulations of a poverty statistic provides a point estimate of the statistic, and the standard deviation provides an estimate of the standard error. 52 53