The first approach typically employs area estimates (e.g., data are at subnational administrative level), the second can be displayed with geo-referenced point data (representing enumeration area, communities, or villages), and the third combines area and point estimates in a model. Examples for each approach will be discussed.
Auxiliary data usually provide complete coverage of a country at various resolutions. They can be used to produce maps with relatively fine resolution, for example data from a population and housing census can be mapped at district (area) or village level (point). Possible indicators capture the social and economic dimension of human well-being and include education, housing, and 'basic needs' variables, and for selected countries income measures. Or, they can be at coarser resolution and include for example subnational statistics by administrative area (agricultural production, health and education statistics, etc.), maps (roads, water sources, location of clinics, etc.), and satellite imagery (land use, vegetative vigor, etc.). Possible indicators come from all three dimensions of human well-being.
Figure 3, page 34, presents two different approaches to map these data: (1) The auxiliary variables can be mapped individually, for example average household income at district level or variability of the growing season from a satellite image. (2) The auxiliary variables can be integrated, with the help of a GIS, to produce composite measurements, for example maps representing access, vulnerability, and economic diversity. Because of data availability, mapping auxiliary data is more often a presentation of variables that are linked to major causes and processes of poverty, and to a much lesser degree a presentation of direct poverty indicators such as the number of people below the poverty line or average household income.
The advantages of mapping auxiliary data are a good supply of international medium resolution datasets and expertise gained in food security and vulnerability mapping. Its disadvantages are the coarse resolution and our limited understanding of the causes and effects related to poverty, especially under local conditions. Another limitation arises when data are provided at and presented by administrative units. Since an area presentation disregards spatial heterogeneity within units, the produced maps are often too coarse to determine priorities within countries, develop and target specific poverty alleviation projects, or study causal relationships between poverty and marginal lands. Mapping auxiliary data is appropriate for international comparisons and broad priority setting exercises, and can become a useful input for high resolution poverty maps that model poverty estimates (see 5.3, page 56).
Three examples showing composite maps of auxiliary data - access, vulnerability, and economic diversity - will be discussed in more detail (5.1.1 - 5.1.3). Limited access to markets, high vulnerability to natural hazards, and a narrow economic base may be important dimensions of peripheral areas where the incidence and changes in poverty are greatly dependent on geographic factors.
Other examples of auxiliary data that could be mapped include population pressure and environmental degradation. Examples and data sources will be discussed in 5.1.4. 5.1.1 AccessAccessibility has been defined as the ability to interact with sites of economic or social opportunity, for examples markets, schools, and health facilities (Deichmann, 1997b). It is usually influenced by physical, economic, and social factors. The assumption is that areas with poor access or low market integration provide fewer economic opportunities for poor people to escape poverty, accentuate the severity of poverty, and may even contain a larger proportion of poor people. Deichmann has produced a detailed review of physical accessibility indicators which will be a useful background document for any poverty mapping exercise (Deichmann, 1997b).
Figure 4, page 37, and Figure 5, page 38, show accessibility maps for West Africa. Cities with a population greater than 200,000 were defined as major urban centers, representing major markets. Average travel time is then used as an indicator for market access and integration. Box 5, page 35, describes in more detail how to produce such maps.
The quality of the accessibility maps depends on the spatial accuracy of the road layer, correct road quality information, and realistic estimates of relative travel costs. The presented example used relatively coarse data and a simplistic model to define access. More sophisticated algorithms have been developed to calculate accessibility. For example, Deichmann has developed a program to generate a suite of accessibility indicators for each node in a network, and a lattice that estimates the "market integration" at each point (Deichmann, 1997). The West African Long-Term Perspective Study (WALTPS) demonstrated a strong correlation between population density, per capita agricultural production, and potential market integration, defined as the virtual price offered by the market for a standard basket of products (Ninnin, 1994). WALTPS assumed that the higher the virtual price, the greater the incentive to produce, on a sustained basis, an agricultural surplus and the greater the earning opportunities. The applicability of WALTPS throughout Africa needs further examination since there appear to be other factors than access that limit market integration such as farmers' capacity to sustain high outputs over the long-term.
Box 5 Calculating Measures of AccessibilityThe presented example tries to characterize market accessibility in West Africa by producing a map of travel time to major urban centers. It is based on digital maps of settlements and roads from a database produced for the West Africa Long-Term Perspective Study (Brunner et al., 1995).In a first step, markets are defined by major urban centers that are cities with a population greater than 200,000. Information on the extent and quality of the road network becomes the second input. Different types of roads are assigned specific friction values corresponding to relative travel costs. For example, if a gravel road has a friction value of 4 and a paved road a value of 1, it will be four times as difficult or time consuming to cross the same distance on a gravel road than a paved road. GIS programs use for example "grid cell equivalents" (GCE) as the unit of measurement to calculate total travel costs.1 Moving through a grid cell with a friction value of 1 produces a GCE of 1. A total cost of 5 GCE could result from a movement through five cells with a friction of 1, or one cell with a friction of 5. Based on these friction values, cost-distance to the nearest market (metropolitan center) can be calculated. The GIS program generates a distance and proximity surface (also known as a cost surface) where distance is measured as the least effort required to move over a friction surface. Costs are determined radially from a set of source targets to the edges of the image. A specific average speed for different road types and non-road areas was assumed. To obtain a final cost surface with a more intuitive indicator than GCE, friction values were converted to travel time in minutes across each grid cell (average grid cell size was 6 km). The following table lists the different road friction and speed values for the dry and rainy (values in brackets [ ] ) season. It also provides the equivalent relative friction values.
Two cost surfaces, representing the cost-distance in minutes to the nearest market for the dry and rainy season, were calculated. These cost surfaces were then reclassified into six accessibility zones, identified as 0-1 hours, 1-3 hours, 3-6 hours, 6-10 hours, 10-20 hours, and more than 20 hours of travel time. The classified surfaces are shown in Figure 4, page 37, and Figure 5, page 38. ______________________ 1 The cost surfaces were computed with the Idrisi software using the COST module. COST incorporates two algorithms for the determination of cost distances -- a pushbroom algorithm and a growth algorithm. The COSTPUSH algorithm was used to generate the resulting cost image. It operates in the following way: 1. Distances are measured according to the minimum number of cells that must be traversed to move from that cell to the nearest source target. 2. Movements are in 8 directions from any cell, and diagonal movements produce a cost of 1.41 times the friction value. This concept of distance should be clearly distinguished from Euclidean distance. A more complex analysis could have employed the COSTGROW algorithm which allows to incorporate barriers, for example caused by customs delays at national borders, and maximum distance for travel. It also allows to calculate cost distances along the road network alone. |
The second example of mapping auxiliary data uses subnational administrative areas as the unit of analysis. It is based on a regional vulnerability assessment for the Sahel (Wright et al., 1995) which was discussed in more detail in 2.3.4 (see Table 2, page 14). The assumption is that highly vulnerable areas may contain a high proportion of poor people, expose the poor to frequent shocks, and trap them in chronic poverty.
The FEWS database includes data for 1,130 administrative units. Figure 6 to Figure 8, pages 40 to 42 show composite maps representing income structure, resource base, and current vulnerability, respectively. Income structure is based on district-level data. Resource base combines satellite imagery with an access measure to urban centers. Current vulnerability is a combination of satellite imagery, market data, and expert opinion. FEWS combined all three elements in an overall vulnerability index.
 |
||
Figure 4 Access to Major Urban Centers in West Africa - Dry Season |
||
Back to CONTENTS
The goal of developing a typology of administrative areas is to build an analytical framework by identifying areas with social, economic, or environmental characteristics that are of importance to poverty, human welfare, and agricultural research. These may include areas that are highly vulnerable to drought, have experienced rapid population growth, or suffer from a depleted natural resource base.
For example, the U.S. Department of Agriculture's Economic Research Service (ERS) has developed a typology of rural areas that has become the framework for statistical analyses, mapping, and graphical presentation of data. For more information on rural typology codes in the U.S., see Cooke and Mizer (1994) and Sommer and Hines (1991).
The availability of clearly defined typology codes which reduce the economic and social diversity of U.S. counties to few important policy relevant themes, has allowed analysts to examine the effects of national policy making on rural areas in more detail. For example, areas classified as "persistent poverty counties" have been compared to the rest of the country as for variables such as farm investment, potential ground water contamination, educational attainment, and local government capacity.
The detailed economic data that were used in the USA are not readily available at international level for most developing countries. However, a rough prototype typology of economic activities, based on existing data for West Africa, can be developed to show how a typology for a group of developing countries would look like (see Box 6, page 43). A plausible hypothesis, which will need further research, assumes that areas with a less diversified economy provide fewer economic opportunities for poor people to escape poverty and may even have a larger proportion of poor people than more diversified areas.
Figure 9 to Figure 12 (pages 46 to 49) are the resulting maps of this prototype typology. The map of metro and non-metro areas (Figure 9, page 46) used a previous accessibility map (Figure 4, page 37) as an input. This may be a useful approach to define rural areas for an international mapping activity, since no commonly agreed upon definition of rural areas exists yet. With additional data a more sophisticated typology could be developed. For example, non-metro areas could be further characterized by crop specialization, market access, and infrastructure variables.Other examples of typologies capturing
the economic resource base that could become a useful input for poverty mapping
include a map of agricultural production systems for the Greater Horn of Africa
produced by FAO (van Velthuizen
et al., 1995) and maps of food economy zones developed by Save the Children
Fund (Seaman et al., 1993; Save the Children Fund, 1996). Figure 13,
page 50, presents such food economy zones for Sudan. 
Other potential indicators that could be mapped include population pressure (land availability per capita), environmental degradation, and if known, causes of poverty (see Box 7, page 50). Various high quality large-area GIS databases have been published recently and could become a useful input for this approach:
For example, the CUI coverages for Mali, Burkina Faso, and Niger were intersected with village population databases for each country to estimate the amount of cultivated land per person per village (Brunner and Nielsen, 1997). Under conditions of extensive, low input agriculture, a shortage of cropland can cause forest clearing and expansion into environmentally marginal lands. But other empirical evidence links increased population density to increased agricultural productivity and environmental protection. A study of Machakos District, Kenya shows that despite a six-fold increase in population between 1930 and 1990, the rural population succeeded in raising productivity on both a per hectare and per capita basis, while controlling and even reversing the degradation of their natural resources (Tiffen, et. al., 1994)
Box 6 Developing a Typology of Economic ActivitiesThe typology was developed as follows. First, administrative units were classified as metro and non-metro. Next, economic activity codes were assigned to each unit. Finally, economic diversity per unit was determined by the number of activity codes assigned to each unit. Economic diversity was classified as single sector, dual sector, and three or more sectors dominant per administrative unit. These three categories, the codes, and a brief definition are summarized in the figure at the end of this box. Rural - Urban Continuum All administrative unit areas for West Africa were categorized as metro and non-metro areas. Of the 1,996 administrative unit areas, 447 were classified as metro and defined as areas that intersected with the 1 hour travel zone to cities with a population over 200,000. Of these metro units, 120 metro-core units were selected, based on all units whose center is located within the 1 hour travel zone. The remaining 327 metro units were classified as metro-periphery. All other 1,549 units are non-metro areas. Figure 9, page 46, shows a map of these areas. Economic Diversity and Activity Codes Nine economic activity codes were developed: Agriculture (very high, high, medium, and low), Protected Areas, Forest, Mining, Other, and Service. Economic activity was defined by the spatial extent or location of these activities within the administrative unit. This provides a starting point how an economic typology can be developed and allows to examine potential use for spatial and statistical analyses. It is recognized that with more detailed economic production data at subnational level - although it may not currently be collected consistently at regional scale - a more accurate and robust economic typology could be developed. The category "other" is defined by areas that have no or minimal cropping (less than 2% of the land area of the administrative unit area is cropped), no major mining activity, no protected areas with significant areal extent, and no significant forest area. "Other" includes primarily desert and areas where herding plays an important economic role. All administrative divisions that had a mine, pipelines, and/or gas fields within their unit area were classified as having ongoing economic activities in the mining sector. Those administrative units that had more than 50% of the area covered by tropical moist forest were classified as having significant economic activities in the forest sector. This can include areas with degraded forest, for example resulting from intercropping with coffee and oil palms. All administrative units with more than 20% of the area under some sort of protection were identified as having some economic activities related to biodiversity and tourism. To estimate agricultural activities, a very simplistic approach was used. All areas that had more than 2% of the administrative land area under crops were classified as having ongoing agricultural activities. Agricultural activities were classified into four categories, expressing the percent of land area allocated to the agricultural sector. The exact thresholds for these four categories varied across six different vegetation zones. Such an approach makes adjustment for the latitudinal differences in resource endowment. The four categories (low, medium, high, and very high agriculture) were defined as outlined in the table below. A more sophisticated approach could have combined crop land use intensity (CUI) data with length of growing period maps. Classification of Agricultural Activities for Economic Typology
It was assumed that all metro areas include a significant service sector consisting for example of local and national government agencies, private and personal services, wholesale and retail trade, transportation, and public utilities. In a final step, the degree of economic diversification for the 1,996 administrative divisions was expressed. All units that were assigned only a single economic activity were classified as units with a single dominant sector (Figure 10, page 47). Those with two overlapping activity codes were categorized as having a mixed economy with two dominant sectors (Figure 11, page 48). All areas with three or more economic activity codes were classified as having a diversified economy (Figure 12, page 49). (continued) |
Box 7 Mapping Causes of PovertyA closer look at the causes and types of poverty may help to identify other indicators that could be mapped subnationally. For example, if a poverty assessment finds that the majority of the poor are landless laborers, pastoralists, and indigenous populations and information is available on the proportion of these population groups within each district, then mapping the location of these groups can become a useful proxy for the spatial distribution of poverty. The limitation of using such proxy indicators for poverty mapping is obvious: Causes of poverty must be known, and causes may vary from country to country. For example, size of land holdings may be a good poverty marker in Bangladesh, but not necessarily in the Sahel (Ravallion and Sen, 1994). Two reviews classifying causes and types of poverty, one by the Swedish International Development Cooperation Agency (Sida) the other by IFAD, are summarized below (Sida, 1996; Jazairy et al., 1992). Sida identifies four major interacting conditions that determine well-being or poverty status:
|
Back to CONTENTS
A brief example will demonstrate how geo-referenced DHS can be used to plot a map of enumeration areas and calculate indicators for new units of analysis (aridity zones). The data came from the USAID supported West Africa Spatial Analysis Prototype (see Box 8, page 54, for more detail).
5.2.1 Example - Child Nutritional Status and AridityThe presented example uses geo-referenced measures of child nutritional status from the DHS, aggregates them to new units of analysis (aridity zones), and examines the relationship between child nutritional status and aridity. Indicators of child nutritional status based on anthropometric measurements were selected because they represent a good indicator for the degree of development in a region and can be interpreted as a proxy for poverty. Experts from the Second Report on the World Nutrition Situation 1992 identified anthropometric measurements as the best general proxy for constraints to human welfare of the poorest, including dietary inadequacies, infectious diseases, and other environmental health risks. They proposed anthropometric measures as a strong predictor, at individual and population levels, of subsequent ill health, functional impairment and/or mortality (United Nations, 1992). Anthropometric indicators by themselves do not allow to identify the specific causes of growth retardation and wasting. Data were available from the following ten surveys: Burkina Faso (1993), Cameroon (1991), CAR (1995), Côte d'Ivoire (1995), Ghana (1988), Mali (1995), Niger (1992), Nigeria (1990), Senegal (1992), and Togo (1988).
Four indicators of child nutritional status were used:Stunting measures chronic undernutrition and is an indicator for the long-term effects of undernutrition in a population. Low height-for-age is primarily caused by inadequate nutrition over a long time period, but is also affected by recurrent and chronic illness. Wasting is a measure for acute undernutrition and represents nutritional deficits immediately preceding the survey. It may also be caused by recent episodes of illness or an acute food shortage. Low weight-for-age (underweight) takes into account both acute and chronic undernutrition, but cannot distinguish between children who are underweight because of stunting or wasting. The
proportion of children underweight and stunted indicates the share of children who are experiencing at the same time chronic and acute nutritional deficits Aridity was defined using the aridity index (AI) ranges given in the World Atlas of Desertification (United Nations Environment Programme, 1992). The index is defined as the ratio of precipitation to potential evapotranspiration and then classified into six aridity zones: hyper-arid, arid, semi-arid, dry-subhumid, moist subhumid, and humid. Data came from the International Center for Research in Agroforestry (ICRAF) in Nairobi.
Box 8 Geo-Referencing Survey DataThe USAID supported West Africa Spatial Analysis Prototype (WASAP) was motivated by a desire to add value to the DHS so that clusters could be grouped to new units of analysis and used in broad international analyses. Data have been geo-referenced for 12 countries in West Africa. The DHS is funded by USAID and implemented by Macro International Inc. in collaboration with country statistical services. The DHS is a national sample survey designed to provide information on fertility, family planning, and health. The survey involves interviewing a randomly selected group of women between 15-49 years of age. Typically, the DHS is selected in two stages. First, a stratified random sample of Enumeration Areas (EAs) is chosen with equal probability of selection per region or urban/rural area. Second, a complete household listing is carried out in each EA from which a number of households is chosen at random. The number of households chosen is proportional to the population of the EA. The DHS reduces sampling costs by sampling a relatively large number of households from few EAs. The sampled EAs are known as clusters. Most of the DHS clusters were mapped by the U.S. Bureau of the Census (BUCEN) at the request of USAID's Regional Office for West and Central Africa (REDSO/WCA) in Abidjan. With the exception of the Côte d'Ivoire and Mali DHS which used GPS to geo-reference the survey clusters, no survey explicitly mapped the clusters. BUCEN located each cluster by linking the name of the settlement with the settlement names, and associated latitude/longitude coordinates, given in the Defense Mapping Agency (DMA) Gazetteers. For settlements not found in the Gazetteers, the coordinates in degrees and minutes were read from a map. About 85% of the clusters were mapped using the Gazetteers (BUCEN, 1996). If the settlement name did not appear in either the Gazetteer or map, the cluster was located in the capital of the administrative unit within which the cluster lies. Administrative centers typically have the same names as their administrative units. In Nigeria, where some have different names, the cluster was located in the capital of the higher level administrative unit. As a result of this hybrid procedure, there may be differences of up to 10 km (50 km in Nigeria) between the estimated and actual location of each cluster. |
5.2.2 Aggregating Cluster Data to New Units of Analysis
Anthropometric data for the ten countries include 2,250 clusters, of which 1,038 are urban and 1,212 rural. Each cluster includes about 30 households. Although the household data have been aggregated to the cluster level, there are too few observations to derive statistically valid estimates at the cluster level (MacroInternational, 1996). To be of any real value, the cluster level data need to be aggregated to higher levels, such as administrative units or agro-ecological zones. However, to aggregate the data to higher levels, it is not sufficient to sum the values for each cluster and then divide by the number of clusters per region. This would merely produce the average of cluster level values that are themselves averages, and not the average for the population per region of interest. A second problem arises from the fact that the DHS is collected using a nationally, not locally, representative sample, i.e., the probability of selection of a household for interview is not constant across the whole country. To aggregate cluster level data to new units of analysis, corrections need to be made for differences in the probability of selection using appropriate sampling weights. These are the inverse of the probability of selection and need to be applied when new averages are calculated from the DHS data. These weights are included in the DHS cluster files. The indicators in this example were calculated using an ArcView script written by Trevor Croft (Macro International Inc.). The script produces means and standard error for the selected variables and warns users when too few clusters are aggregated that would yield an unrobust measurement. Proportions for children aged 3-35 months were calculated using weights that take into account the size of the sample relative to the background population. 5.2.3 ResultsThe results of aggregating the data to new units of analysis are presented in Table 9, page 55, that show the mean and standard error of the incidence of each indicator per aridity zone expressed as a percentage of children sampled, disaggregated by urban and rural clusters. Clusters of the hyper-arid zone were combined with those of the arid zone because it contained too few clusters (6) to produce statistically significant results.
Table 9 Nutrition Indicators by Aridity Zones in West Africa|
| |||||||
|
|
|
|
|
|
| ||
| URBAN AND RURAL | |||||||
| NUMBER OF CLUSTERS | 178 | 806 | 189 | 476 | 601 | 2,250 | |
| NUMBER OF CHILDREN | 1,971 | 8,967 | 2,193 | 5,072 | 6,155 | 24,358 | |
| Stunting (%) | 34.9 [1.6] | 41.6 [1.4] | 36.3 [1.8] | 32.7 [2.1] | 28.0 [0.9] | ||
| Wasting (%) | 21.2 [1.1] | 17.7 [0.8] | 11.1 [1.1] | 8.4 [0.8] | 7.2 [0.6] | ||
| Underweight (%) | 45.6 [1.6] | 45.1 [1.4] | 36.1 [1.7] | 30.1 [2.7] | 25.4 [1.1] | ||
| Stunting and Wasting (%) | 8.1 [0.8] | 8.2 [0.6] | 4.5 [0.7] | 2.6 [0.4] | 2.8 [0.4] | ||
| URBAN | |||||||
| NUMBER OF CLUSTERS | 36 | 371 | 92 | 203 | 336 | 1,038 | |
| NUMBER OF CHILDREN | 358 | 3,098 | 677 | 1,643 | 2,730 | 8,506 | |
| Stunting (%) | 21.7 [2.7] | 25.8 [2.0] | 31.4 [4.2] | 25.8 [1.4] | 22.6 [1.4] | ||
| Wasting (%) | 17.3 [2.8] | 15.7 [1.3] | 6.8 [1.6] | 7.7 [0.8] | 6.3 [0.9] | ||
| Underweight (%) | 29.6 [3.1] | 30.0 [1.7] | 25.2 [2.9] | 22.7 [1.7] | 19.4 [1.5] | ||
| Stunting and Wasting (%) | 4.2 [1.0] | 4.4 [0.7] | 2.0 [0.7] | 2.8 [0.4] | 1.9 [0.5] | ||
| RURAL | |||||||
| NUMBER OF CLUSTERS | 142 | 435 | 97 | 273 | 265 | 1,212 | |
| NUMBER OF CHILDREN | 1,613 | 5,869 | 1,516 | 3,429 | 3,425 | 15,852 | |
| Stunting (%) | 36.8 [1.8] | 45.4 [1.6] | 38.7 [1.7] | 35.5 [2.7] | 30.3 [1.1] | ||
| Wasting (%) | 21.8 [1.2] | 18.1 [0.9] | 13.2 [1.6] | 8.7 [1.1] | 7.6 [0.8] | ||
| Underweight (%) | 47.7 [1.8] | 48.7 [1.6] | 41.5 [2.0] | 33.6 [3.5] | 27.9 [1.4] | ||
| Stunting and Wasting (%) | 8.6 [0.9] | 9.1 [0.8] | 5.8 [1.1] | 2.4 [0.6] | 3.2 [0.4] | ||
Note: The values in brackets [ ] show the Standard Error for each average.
Source: World Resources Institute, September 1997
As expected, stunting, wasting, and underweight are highly correlated. Both short and long-term indicators show a gradual decline toward the richer and more fertile coastal zone. The incidence of malnutrition is significantly higher in rural than in urban areas.
The proportion of stunted children is the highest in the semi-arid zone for rural clusters and in the dry subhumid zone for urban clusters. The results would suggest that the greatest constraints to human welfare caused by the interaction of factors such as dietary inadequacies, infectious diseases, and other environmental and economic constraints are within these two ecological zones.
To test this proposition, the relationship for one specific indicator, the proportion of children stunted in rural areas, was examined. Of the four indicators, stunting is probably the most robust measure. Stunting typically persists over years and is not reversible. Thus, it is relatively insensitive to the fact that the surveys were carried out over a seven year period (1988-95). The analysis was limited to rural clusters because of their higher dependence on agriculture and subsistence food production with the associated high risks of crop failure for rainfed agriculture. Figure 15, page 57 (map of sample clusters) shows the spatial variation of stunting. Stunting varies greatly across the semi-arid zone, ranging from lower values in Senegal, comparable to the coastal areas in Ghana and Côte d'Ivoire, to the highest values in eastern Niger and northern Nigeria. The highest concentration of stunting is found around Kano, Nigeria, an area characterized by high levels of agricultural intensification and environmental protection (Adams and Mortimore, 1997), suggesting that a rising agricultural productivity, on a per capita or per area basis, does not preclude low levels of human development.
Aggregating malnutrition indicators by aridity zone is only one example how new units of analysis can be created with geo-referenced survey data. Although there appear to be clear spatial trends at aggregated and disaggregated level, further analysis is required to understand some of the underlying factors contributing to this spatial pattern. This could be done for example with the help of multivariate analyses at the cluster level that include in addition to aridity zones other explanatory variables from the DHS (family size, mother's education, etc.) and other sources (market access, cropping system, policy environment, etc.).
Other variables from the geo-referenced DHS data have been mapped by UNEP/GRID-Arendal and include selected variables related to human development (UNEP/GRID-Arendal, 1997). Geo-referenced survey data could also be aggregated to composite indexes, for example to a cluster-level Human Development Index (HDI), which combines child malnutrition, adult female literacy, and female school enrollment, and a Household Assets Index (HAI), which combines water source (piped, well, etc.), quality of housing, and means of transport.
Back to CONTENTS
Different statistical techniques have been developed to extrapolate from survey data and produce poverty measures at larger scale, for example at the community or village level. These techniques typically demand data with a wide geographic coverage, such as census data and national digital maps, which are incorporated into a multivariate prediction model that estimates poverty measures for much smaller areas than the original sampling of the survey was designed for (small area estimation).
Figure 16, page 59, summarizes how auxiliary and survey data can be combined to produce poverty maps with the help of such statistically techniques. Two examples of modeling poverty estimates that could be used in sub-national poverty maps will be discussed in 5.3.1, page 58, and 5.3.2, page 60. Example 1 combines survey and census data for Ecuador. Example 2 uses a GIS to integrate data from household surveys, community surveys, and national digital maps for Burkina Faso.
A third example of modeling local poverty estimates is methodologically different from the two examples above. Section 5.3.3, page 62, describes how CIAT is testing a method to extrapolate from local participatory assessments to a larger geographic area. The case study for Ecuador demonstrates how household survey data can be combined with census data to predict per capita expenditure figures for all households in the census (Hentschel et al., 1997). It required two datasets:
The modeled poverty estimates matched closely the poverty rates calculated from the household survey data alone. A comparison of this consumption-based poverty measure and a composite basic needs index revealed significant discrepancies in identifying the same poor households.
Since the study is still work in progress, no high resolution poverty maps have been produced yet. The authors presented poverty rates for 8 different regions in Ecuador. They discussed producing more detailed poverty maps, for example for the 400 cantons or over 1,000 'parroquias' (parishes), a relative easy aggregation since estimates are available at the household level.
5.3.2 Burkina FasoThe Burkina Faso case study combines spatial data at the village level and modeling expertise and is being carried out by the Human Resources and Poverty Division of the Africa Technical Department at the World Bank in collaboration with a team from Oxford University and I-MAGE, a Belgian GIS consultancy. The project tries to develop criteria for better targeting of public investments and projects to poor rural communities in Burkina Faso (Bigman, 1996).
The purpose of this project is to identify criteria which allow narrow geographic targeting, that is identifying and reaching poor villages and communities. The study combines data from household surveys, community surveys, and national digital maps in a Geographic Information System (GIS) and then uses econometric modeling to develop criteria for narrow geographic targeting.
The research project is based on various observations of rural communities in Africa. Poor people tend to live with poor people. In selected countries, the majority of the rural poor often concentrate in a relatively small number of villages where the majority of residents are poor. For example, in Nigeria nearly 80 percent of the rural poor live in 25 percent of the villages. A comparison of income differences in rural areas of Ghana, Kenya, Nigeria, and Uganda showed that the overall income inequality between individuals can be mostly explained by income inequality between communities and to a much lower degree by income differences between individuals within communities (Bigman, 1997).
The analytical design of the study is as follows:Source | Data | Coverage |
| Household Income and Expenditure Survey 1994 |
| national sample of 8,628 households |
| Household Income and Expenditure Survey 1994 |
| national sample of 351 villages |
| Agricultural Survey 1993 |
| national sample of households |
| Demographic and Health Survey 1993 |
| national sample of 5,706 households and 230 clusters (enumeration areas) |
| Demographic and Health Survey 1993 |
| national sample of 5,706 households and 230 clusters (enumeration areas) |
A comparison of regional poverty rates showed a close match between estimates from the prediction model and those from the Household Income and Expenditure Survey alone (Bigman et al., 1997). A preliminary poverty map at the village level demonstrated significant spatial heterogeneity that is not captured in the coarse map based on the Household Income and Expenditure Survey which presents poverty rates by five agro-climatic zones, as well as for urban and rural areas.
| Level of Aggregation | Data | Source and Coverage |
| Village |
| 1985 Census; 7,131 villages; national coverage |
| Village |
| 1995 database of Ministry of Water Management; 6,299 villages; national coverage |
| Village |
| 1996 database of Ministry of Water Management; 1,981 villages; 5 south-western provinces |
| Health centers |
| 1995 database of Ministry of Water Management; around 1,000 health centers |
| Primary schools |
| 1995 database of Ministry of Education; 1994-95 school year, national; coverage of 3,233 primary schools |
| Meteorological Stations |
| 1995 database of Directorate of Meteorology; national sample |
| Hydrological Infrastructure |
| 1995 database of Ministry of Water Management; national coverage |
| Hydrological Infrastructure |
| 1996 database of Ministry of Water Management; 5 south-western provinces |
| Roads |
| 1992 IGN map; national coverage |
| Province |
| 1995 Markets Data Base; national coverage |
| Department |
| 1995 database of Ministry of Agriculture |
Because of the qualitative nature and the local scale of participatory poverty assessments, they have generally been perceived to complement more traditional top-down approaches based on questionnaires and household surveys. Researchers at CIAT have developed a method to extrapolate from local participatory assessments to a larger geographic area, thus making local perceptions of poverty the basis for poverty assessments. Their method is based on work carried out in Tanzania and Colombia and tries to overcome the two traditional shortcomings of participatory poverty assessments: lack of quantification and representativeness (Ravnborg and Guerrero, 1997). Ravnborg et al. have prepared a manual that outlines a nine-step approach to scale up local poverty assessments. The method is currently being tested in Honduras, and includes the following major steps (Ravnborg et al., 1997):
