Various methods for measuring the spatial location of the poor have been put forward in the literature and in practice, and most of them are under continuing development. Each of these methods has different data requirements and implementation costs.
This chapter provides a brief description of two widely known methods proposed by staff at the World Bank and researchers of other international agencies. One method is based on unit-level data and the other takes into account community-level data.
Poverty mapping based on household survey and census data is a method that has been used in a series of World Bank studies (Lanjouw, 1998; Hentschel et al., 2000; Elbers et al., 2001; and Deichmann, 1999).
This method, henceforward referred to as the Lanjouw et al. method, requires a minimum of two sets of data: household-level census data and a representative household survey corresponding approximately to the same time period as the census. It estimates the incidence of poverty in administrative areas (districts, subdistricts, counties, etc.) on the basis of the data of the household income and expenditure (HIE) survey and the population census. The measures of poverty estimated with this method are based on estimates of the individuals’ (or households’) incomes and these are therefore the common poverty measures that are used in welfare analysis. To determine these estimates, the method uses the HIE survey data and the data of the population census to estimate the per capita incomes of all the individuals in the country or the province (in other studies the dependent variable is the probability that the income per person falls below the poverty line).
Separate estimates are made for individuals in each administrative area and separate measures of poverty for each area are calculated with these estimates.
This approach combines an intensive survey containing information on consumption for a sample of areas and households with individual census data (or an extensive survey). The household survey data are used to estimate a consumption equation:
where yi is income/consumption expenditures per capita for household i, and xi is a vector of household characteristics. The choice of the indicators must be made so that the explanatory power of the vector (xi1, ..., xim) is sufficiently high. At the same time, only variables that are available both in the household survey and in the census data can be selected. The equation is used to predict the probability that households in a given area are poor for all the areas in the country, using the entire census data. Amore detailed description of the specific steps in this analysis is given below.
In the first step, the data of the HIE survey are used in an econometric analysis to estimate the functional relations between the level of the household’s income and key household and community characteristics. In the second step, these functional relations are used to estimate the level of income for all individuals/households in the country on the basis of the actual data from the population census. In the third step, these income estimates are used to determine the measure of poverty.
The practical application of this concept presents a number of econometric and computational challenges, including issues relating to the large size of census data sets, non-normality, spatial autocorrelation, and heteroscedasticity. Elbers et al. (2001) discuss these aspects in detail. In particular, the implementation of this method requires a huge database created by a well-designed and large-dimension survey on consumption/expenditure (HIE) and the census data. One quality of this methodology is the relative ease of checking the reliability of estimates that are built into the programs the World Bank provides to national poverty-mapping analysts. The size of standard errors in these estimates depends largely on the degree of disaggregation sought and the explanatory power of the exogenous variables in the first-stage model.
Another estimation method involves the use of average values from disaggregated geographical units, such as communities or small towns, instead of household-level data (Bigman et al., 2000). In this study, this method is referred to as the Bigman et al. method.
The Bigman et al. method has the advantage of less stringent data requirements. National statistical agencies may be more likely to release community averages upon request. Indeed, these data may be published. Apart from the difference in scale of the predictive model, the two estimation methods follow essentially the same steps.
The Bigman et al. method estimates an index of poverty or vulnerability for each community (district, subdistrict or county). This index is the probability that per capita consumption expenditures of households in a given community fall below the poverty line. In this case, the vulnerability is the incidence of poverty in the community. With this method, the index is estimated by using data from a wide range of sources that provide detailed information on the community and the area in which it is located. These sources can include: geographical information system (GIS) coverages that provide data on the geographic, topographic and climate conditions in the area; road maps that give data on the quality of the road infrastructure, the distance to the nearest market, etc.; the agricultural census that provides data on the crops grown in that area, the use of irrigation, fertilizers and pesticides in that area, etc.; and various surveys (such as the demographic and health survey) that provide data on the demographic and socio-economic characteristics of the population in these areas, etc. Data on community characteristics obtained from the population census can also be used.
This method combines the household survey with information on the areas in which the individuals reside (rather than on their personal data). For the analysis, information on the areas is collected from a wide variety of sources. The first step in calculating the vulnerability index for each community is to use the individual household survey data in order to estimate an income or consumption equation:
where yia is the income or consumption expenditures of individual i in area a, (xi1, ..., xim) is a vector of the household’s personal and family characteristics, and (zi1, ..., zak) is a vector of the area characteristics. From this logit analysis, one can derive estimates for the probability that an individual in area a is poor as:
where 
is
the average values of the individual characteristics (xia1,
..., xiam) for all the individuals in each of the areas
a. These relations can be used to predict the incidence of poverty
Pa in all small ‘areas’ of the country
(subdistricts, counties, etc.) on the basis of the information on the area
characteristics 
, which can
be obtained from a wide variety of sources.
While easier access to data makes this method attractive, the error associated with estimation for units of different sizes (in terms of population) has not been investigated thoroughly. Thus, it is not clear what the trade-offs are between ease of access and statistical reliability.
