Click here to return to the report index
Mapping Indicators of Poverty in West Africa

APPENDIX 3: Data documentation

3.1 POVERTY INDICATOR - HUMAN DEVELOPMENT INDEX

Dataset name:	West African Spatial Analysis Prototype (WASAP)
Available from:	Internet, at http://edcintl.cr.usgs.gov/adds/adds.html
Source:	Demographic and Health Survey (DHS), 1995 Text extracted from Brunner, 1997 (see attached paper)

Indication of poverty levels was evaluated in terms of quality of life, based on the Human Development Index (HDI).

For the purpose of assessing human development issues world-wide, U.N. Development Programme ranks nations based on the Human development Index (HDI). This index is a quality of life measure that includes both social and economic values. The HDI is based on three indicators (UNDP, 1996):

1. Life expectancy at birth
2. Educational attainment (measured by combination of adult literacy and school enrolment ratio)
3. Standard of living (measured by real GDP per capita)

The data used for this study was taken from the Demographic and Health Survey (DHS), funded by the U.S. Agency for International Development (USAID) and implemented by Macro International in collaboration with national statistics services. The DHS involves interviewing a randomly selected group of women between 15-49 years of age. The sample is selected in two stages. First a stratified random sample of enumeration areas is chose with equal probability of selection by urban and rural area. Second, a complete household listing is carried out in each enumeration area from which a number of households is chosen at random. The number of households chosen is proportional to the population of the enumerated area. Every sample (points in the dataset) is know as a cluster. On average there was 30 household surveyed by cluster or sample points

Surrogates Variables from DHS to represent Human Development Indicators:

Because none of the HDI indicators are explicitly captured by the DHS, the following surrogates were used:

1.

Life expectancy at birth

-->

Surrogate Indicator:

Child mortality

Child mortality is known to be correlated to life expectancy at birth. For this purpose the DHS variable used was the 'Proportion (%) of children who died before the age of 5 years' (WDEAD60P)

2.	Educational Attainment	-->	Surrogate Indicator:	Adult Literacy
		-->	and	Primary School Enrolment

The two surrogates are a) 'The proportion (%) of adult female household members aged 15-49 who can read' (WLITERP), and b) 'The proportion (%) of all children at the age of primary education attending school' (WSCPRMP).

3.

Standard of Living

-->

Surrogate Indicator:

Stunted Growth in Children

The DHS surrogate is 'Proportion (%) of children aged 1-35 months that are stunted' (WSTUNTP).

Studies show that the relationship between child nutritional status and poverty is strong at the lower end of the income range. Increasing GNP per capita from $300 to $600 is associated with a decline in the prevalence of underweight children from about 34% to 17% or a reduction of about 50% (U.N. ACC/SCN, 1992). Beyond $900, the effect of increasing incomes on nutrition diminishes. Because no country in the study has an average per capita income greater than $600, child nutritional status is used to estimate real GDP per capita and, by implication standard of living.

Stunting (low height-for-age) is an indicator of chronic undernutrition caused by inadequate long term nutrition and recurrent illness. It is an accepted global measure that can be used to compare the growth patterns of children across countries and across ethnic groups. Because stunting typically persists over years, this indicator is relatively insensitive to the fact that the survey were carried out over a ten year period (1986-1995).

Comments on the dataset (limitations, usefulness, further considerations)

• It is not clear that this kind of data would be available world wide. Therefore, the basic question of 'mapping poverty' remains. Ultimately it would be important in an extended project, to compare this technique of inferring levels of poverty using HDI, with other methods of evaluating poverty.
• It would also be useful to compare the surrogate variable for HDI to the TAC data used by the current ongoing ICRISAT study, which is country-level socio-economic data on developing countries.
• There were other variable available in the DHS, that weren't considered for the scope of this project, but that could be included in a more detailed study, to infer levels of poverty, such as: floor types, electricity in the house, water facilities, etc.
• Other limitations of the data: Firstly, the data collection spans over a period of 10 years, which may make some comparison between country questionable. Secondly, the DHS data are population based rather than geographically based, which means that they do not necessarily reflect a country's geographical diversity or terrain types. For this reason few clusters are chosen in the Sahara region in the Sahelian countries since these areas have small population. Thus, while the desert area may be critical from a geographical standpoint, it is undersampled in the DHS due to the sampling scheme which follows the population distribution.
• Finally, the sampling design of the DHS survey was to provides analysis on the national level by urban/rural residence, and therefore the data is designed to yield regional or provincial estimates. No DHS survey provide enough information to validly describe the situation at the cluster level (unique sample points). Therefore, in our particular pilot study, the results were more significant when using thematic classes that were more general. Although a more detail spatial analysis of agroclimatic zone, for example (derived from numerous themes soils, growing season, potential rainfall, etc.), would be a better representation of the true agroecological potential of individual area, it, on the other hand may be too detailed for the resolution of the DHS point dataset. This may also explain why some of the result (such as the accessibility level), did not show the expected trend or significant differences.

3.2 AGRO-CLIMATIC ZONES

Data Source: Data Exploration Tool (DET) CD-ROM, Corbett, J.D., 1996

It is often argued that the climate is a first order "determinant" of ecosystem character, with edaphic factors second, followed by human intervention and other natural disturbances (Scholes, 1995). For this reason we have focused mainly on the climatic component to determine agroclimatic potential. Agroclimatic and agroecological zonation schemes are standard tools used to target agricultural research and to set research priorities, yet, there is no unique standard and excepted definition of these classification.

For the purpose of our pilot study, the Agroclimatic Zones were developed based on the ratio of mean annual precipitation to potential evapotranspiration (P/PE). This gives an indication of potential water availability for plant growth, and therefore to agriculture. It is generally excepted that ratio of P/PE below 0.5 have no to low potential for agriculture.

Although the zones do not incorporate agricultural data, they account for most of the agroecological variability in West Africa. All the analysis is done for rainfed areas, since there is no significant not-flood recessional irrigation in the region, with the exception of the Office du Niger irrigated rice project in Mali.

The Agroclimatic zones were derived following the aridity index (AI) rages given in the World Atlas of Desertification (Arnold, 1992).

Precipitation / Potential Evapo-transpiration (P/PE)	Agroclimatic Zones
< 0.05	Hyper-arid
0.05 to 0.20	Arid
0.20 to 0.50	Semi-arid
0.50 to 0.65	Dry sub-humid
0.65 to 1	Moist sub-humid
> 1	Humid

The map for Africa was developed using a set of high resolution precipitation and potential evapotranspiration surfaces (Corbett et al., 1996). The DET climatic surfaces have a cell size of 0.05 decimal degrees (6km at the Equator), are based on a 60 year time-series, and were derived using a high resolution digital elevation model to yield monthly mean temperatures to within 0.5 oC and monthly mean precipitation to within 10-30% (Hutchinson et al., 1996).

From these basic agroclimatic zones, length of growing period, land cover types and Agricultural application can be derived (van Velhuizen et Verelst, 1995, Ehrlich and Lambin, 1996)

Comments on the dataset (limitation, usefulness, further considerations)

• This is a good quality and reliable dataset, and has much information available for a global analysis, or for a more detail analysis base on growing season length, temperature etc.
• There are other dataset that were considered but not used (because of the short period of this initial pilot study), such as the FAO Agroecological Classification Scheme. Water availability (P/PE) and temperature are the key factors in this classification, used to determine crop suitability for rainfed agriculture. It would be interesting in a more detail study to compare the FAO map with the one developed for this pilot project.
• In another study conducted by IFPRI, titled "Should the Indian Government Invest more in Less-Favoured areas?, 1996", the criteria used to determine agroclimatic region were a) soils, b) rainfall, and c) length of growing season. Again, replicating their technique and comparing with the output for this project, would be a interesting possibility for a more detailed study.

3.3 LAND DEGRADATION

Dataset name: Global Assessment of Soil Degradation (GLASOW)
Source: Deichmann et Eklundh, 1991, GRID Nairobi

The GLASOD was conducted by the International Soil Reference and Information Centre (ISRIC) at Wageningen, Netherlands.

The database contains information about the degradation within a map unit as reported by numerous soil experts around the world through a questionnaire. It includes information on the degree, extent, type, cause and rate of degradation. Yet for this study, a severity variable was used which combines the degree and the extent of the land degradation. This basically qualifies the strength at which degradation processes act. The severity variable is divided into 4 classes: low, moderate, high and very high.

Details on the methodology can be read in Deichmann and Eklundh (1991), and in Oldeman (1991).

3.4 POPULATION

Dataset name: Africa Population Database
Source: Deichmann et Eklundh, 1991, GRID Nairobi
http://grid2.cr.usgs.gov/africa/

Population density was used with the assumption that the higher the population is, the higher the stress and demand is on the land, and the more marginal the land is in terms of potential for crop production.

The population density map was produced by Deichmann, 1996, and is available from USGS at the following web site: http://grid2.cr.usgs.gov/data/download.html.

More detail information on methodology, process, and data quality and accuracy, is also available at this site.

3.5 ACCESSIBILITY

Source: GRID-Arendal, digitized from IGN maps at 1:1000000

A derived accessibility surface was used was used to evaluate the influence of a socio-economic factor, on poverty. The assumption is that marginal or unfavourable land will have lower accessibility to markets, to infrastructure, financial institution etc.

The accessibility surface (grid) is a output from the accessibility program (Uwe Deichmann, 1997) which was run on the Mali and Burkina roads and villages coverage.

The index (ranging from 0 to 10) is a measure of average accessibility or potential for interaction (note the "potential" since it's not based on observed interaction/travel behaviour). It gives an idea of the concentration of population within the neighbourhood of each point while considering the quality of the transport network. So, the closer (as measured in actual network travel time) a point is to a large town or a number of large or medium sized towns, the higher the index. So, a large town far away has the same influence as a small town close by.

The indicator is calculated for each node in the network. Since the network was relatively sparse, a network grid was overlaid with a spacing of 3 minutes on top (created using GENERATE with the GRID option). Then the accessibility program was run using the negative exponential distance weighting (see for more details: Uwe Deichmann, 1996, Asia and Africa pop database documentation). The threshold was set to 3 1/2 hours travel time (so for each node, all villages are considered that can be reached in 3 1/2 hours). The node's id, x, y, and value are written to an ASCII file. After running the program, the values for the nodes are interpolated in to a regular surface of 1 minute (0.01666667 decimal degrees). The natural log of that surface was then taken to compensate for the skewed distribution of the cell values.