There are some important implications of applying spatial statistical analysis to poverty-mapping studies.
First is the importance of taking the spatial dimension of the data into account. In fact, after having found a significant spatial correlation between the units, ignoring the spatial component in the regression analysis could lead to misleading estimates of the parameters. This may result in a large proportion of poor households being excluded from the allocation of transfers while a number of non-poor households might be declared potential beneficiaries.
Moreover, the use of spatial models in combination with the visual nature of the poverty maps, obtained from applying the spatial regression methods, may highlight unexpected relationships that would escape notice in a standard regression analysis.
One of the most important difficulties encountered in spatial analysis concerns the availability of skills and data. It would be useful to develop poverty maps by conducting the statistical analysis at a degree of disaggregation below broad regions; otherwise it is assumed implicitly that, within a region, the model of consumption is the same for all households irrespective of which province, county or community they reside in. However, all the data from the household survey (including measures of consumption) are usually based on samples. They are rarely sufficiently representative at low levels of aggregation to yield reliable estimates. At the same time, the census data, which are of sufficient size to allow disaggregation, have no information on consumption. Moreover, even where some subnational poverty data are routinely available, they are often not georeferenced. For this reason, the adoption of the small area estimation could offer a solution: a statistical technique that combines survey and census data to estimate poverty indicators for disaggregated geographic units (counties in this study). The only problem with this statistical tool is that many countries do not provide census and survey data simultaneously. Hence, when the temporal gap between these two kinds of data is large, census and survey information become incompatible.
In the context of data availability, other kinds of useful information about factors that influence poverty are not readily available for all countries, e.g. data on land use and the environment (e.g. for environmentally fragile ecological zones with no productive agricultural lands) and on accessibility. Accessibility concerns issues such as: access to health, water, education; facility location and infrastructure; transport, travel time and the road conditions of a zone. The availability of these kinds of data varies widely from country to country, and the design and selection of case studies must take this aspect into account.
The results of the fitted spatial model demonstrate the statistical significance of environmental variables. This suggests the presence of a poverty-environment relationship and hence the impact of environmental factors on the lives of the poor and on poverty reduction efforts. For this reason, environmental indicators could be an important tool for designing and evaluating poverty reduction strategies and they should be introduced into the statistical analysis.
The results obtained by the study have demonstrated the usefulness of the spatial analysis in poverty targeting compared with the classical statistical analysis. However, the significant data requirements of the spatial technique constitute a drawback compared with the simple regression analysis. The creation of an ad hoc database (georeferenced and geographically disaggregated) is time consuming. This problem could be overcome if all countries provided a unique database with socio-economic and demographic information, with data on food insecurity, geophysical parameters, etc. being as spatially disaggregated as possible and georeferenced.
Another problem of all polygon-based spatial analysis is the modifiable areal unit problem (MAUP): the areal units (administrative or political boundaries, agro-ecological zones, etc.) are arbitrary groupings and the data within each can be aggregated in an infinite number of ways (Nelson, 2001; Bigman and Deichmann, 2000). The implication is that different kinds of aggregation can lead to different results in the spatial analysis so that variables, parameters and processes that are important at one scale or unit are frequently not important or predictive at another scale or unit.
Adefinitive solution to minimize this effect remains to be found. Depending on the issue being examined, some methods have been used to solve the MAUP, for example: identifying the basic units and deriving optimal scales and zonal configurations (Openshaw and Taylor, 1981); conducting a sensitivity analysis (O’Neil et al., 1988); and adopting tools such as convolution filtering, different methods of zonification including extending the concept of areal units to measures such as time, accessibility, etc. (Nelson, 2001).
Concerning the concept and measurement of poverty, the analysis in this study adopted an index calculated using both consumption and the poverty line. Other kinds of index could be used as multiple indicators exist for poverty (Annex B). Poverty can be evaluated using: economic measures such as monetary indicators of households well-being (expenditures, income, consumption, etc.); demographic indicators (gender and age of head of the household, household size, infant mortality rates); and environmental and health measures (access to safe water and sanitation, time spent by household to collect water, cereal production for a family, prevalence of acute infections, disability adjusted life years) (Shyamsundar, 2002).
The choice of one indicator rather than another usually depends on the availability of the data, and on the practical implications in terms of time, costs and technical requirements for constructing the index.
The consequence of using a particular index is that different indicators can lead to different results of the analysis, and so to alternative poverty rankings.
One solution, albeit time consuming, could be to apply different kinds of indicators to the same analysis and then to compare and evaluate the implications of each index. In the application on the Ecuador data, the index adopted was the same used in other studies in order to enable the results of this study to be compared with results obtained using other common methods.
