2.1.1 Human Well-Being is Multidimensional

2.1.2 What Do We Measure?

1. How do we define human well-being?
2. What do we measure - the means or inputs of human well-being or well-being itself?
3. At what level of well-being is a person poor, or how do we determine a poverty line?
4. How do we combine measurements of well-being at the individual or household level to aggregated poverty measurements, both along one dimension and along different dimensions of well-being?

The literature refers to questions 1 - 3 as identification problem and question 4 as aggregation problem (Ravallion, 1992). Box 1 (page 5) and Box 2 (page 6) provide more detail to question 1. Question 3 and 4 are discussed in Box 3 (page 8) and Box 4 (page 9), respectively. Different ways to measure poverty are outlined below.

There are basically two methods to develop indicators of human well-being and poverty. We can measure the means or inputs of human well-being or well-being itself. Food consumption, income, and use of health services are indicators of means or inputs to human well-being. Nutritional status, life expectancy, and literacy rate are examples measuring the ends or outcome, that is well-being itself. Indicators of human well-being at the ends or outcome level are sometimes referred to as measures of human capabilities. Poverty is then defined as a lack of basic capabilities. For example, basic capabilities include a life free of avoidable morbidity, adequate nourishment, healthy reproduction, personal security, and participation in society (McKinley, 1997).

Some authors suggest that poverty indicators should focus as much as possible on people's capabilities, since means or inputs do not always lead to the desired outputs or capabilities (McKinley, 1997). On the other hand, selected outcome variables have the problem of not being able to determine the exact causes that are responsible for the observed results. For example, stunting of children is a good indicator for chronic undernutrition, but its exact causes may not be readily identifiable and include factors such as inadequate food supply, recurrent and chronic illnesses, or length of breastfeeding.

2.2 HUMAN WELFARE AND POVERTY INDICATORS - MAJOR DIMENSIONS

Most human welfare and poverty indicators can be grouped into three major dimensions of well-being: economic, social, and enabling environment. They are summarized in Table 1 (page 11). Indicators for these dimensions will be discussed in turn.

Table 1 lists three major approaches to produce poverty indicators under the economic dimension of well-being: current consumption expenditures, income, and wealth. Indicators of wealth are sporadically used and can be both an outcome or an input variable. For economists, the preferred measure to calculate poverty indicators is current household consumption expenditures, that is the aggregate expenditures on all goods and services consumed (including consumption from own production), valued at appropriate prices. If consumption of services can be expressed appropriately in monetary terms, then current consumption can capture elements of the economic and social dimension of well-being. Since household income of the poor can vary significantly over time (more than consumption) and such data are more difficult to collect in developing countries, it is usually a second choice for economists producing poverty indicators (Ravallion, 1992).

Household consumption expenditures as a measure of standard of living has its foundation in welfare theory. The total amount of money spent on the consumption of goods and services reflects the magnitude of utility derived from consumption. Each household's mix of goods and services is based on individual preferences. Prices provide the appropriate weighting for the chosen goods and services. It is assumed that a household maximizes its individual utilities and that a higher consumption of goods and services raises utility (Ravallion, 1992).

The limitations of using monetary measures to capture household well-being have been discussed extensively in the literature (McKinley, 1997). It includes difficulties of capturing non-marketed and non-priced goods, such as subsistence consumption and free social services, or other aspects important to human well-being, such as community resources, social relations, and the natural environment. Economists usually supplement their economic measures with other indicators such as literacy, infant mortality, and access to public services, to cover these non-monetary

aspects of household well-being. For example, see list of priority indicators in the World Bank' s (1992) Poverty Reduction Handbook and Operational Directive.

Monetary measures have the advantage of being easily comparable. They solve the problem of assigning weights to a mix of goods and services and allow to produce integrated poverty indexes that measure the depth and severity of poverty (see Box 4, page 9, for more detail).

Indicators under the social dimension of well-being in Table 1 include measures on nutrition, energy, sanitation and water, health and family planning, and education, covering both various means and outcome indicators. The strength of social indicators is that they provide a number of useful capability measures. For example, indicators of child nutritional status based on anthropometric measurements have been used as a proxy for the constraints to human welfare of the poorest, capturing dietary inadequacies, infectious diseases, and other environmental and economic constraints. Difficulties in aggregating different social indicators into a composite index is their greatest limitation (Lok Dessallien, 1995).

Development practitioners are beginning to broaden their field of potential poverty indictors and are looking at causes of poverty that are structural and systemic. Potential indicators seek to capture empowerment, governance, participation, and transparency of legal system and look at structural inequities and skewed processes that become an impediment to human well-being. Potential indicators are listed in Table 1 under the category enabling environment. Work in this area is still very preliminary because some of the concepts are not very well defined, are difficult to quantify, or cannot easily be isolated from poverty indicators in the other two dimensions (McKinley, 1997). Geographic factors may be one of those structural impediments and are summarized under the concept peripheral areas which are isolated areas where poverty is largely determined by geographic factors. Table 1 lists also some indicators that measure food security related vulnerability such as poor agricultural endowment and high environmental hazard under the enabling environment dimension.

Both vulnerability and peripheral areas will be discussed in more detail (2.3, page 12, and 2.4, page 17, respectively). Vulnerability and peripheral areas are not mutually exclusive and overlap to some degree with economic and social well-being. Vulnerability often combines information on hazards (drought, floods, etc.) with economic and non-economic measures of well-being, and may include indicators for peripheral areas as well. Peripheral areas combines information on economic and non-economic well-being with spatial factors.

A number of general limitations apply to many of the indicators listed in Table 1: They have problems capturing the dynamics of poverty. For many developing countries, data are not always available or accurate enough for detailed poverty analyses. International comparisons of national data are difficult since countries use different methods to calculate the same indicator. If measurements are at the household level, adjustments have to be made to address the imbalances within households, for example by disaggregating indicators by gender or using adult equivalent units to reflect different household sizes (Lok Dessallien, 1995).

2.3 VULNERABILITY

Vulnerability can be defined as the susceptibility of an individual, household, or community to external shocks and fluctuations. The concept of vulnerability has been applied extensively in famine early warning programs, food security assessments, and drought relief coordination, primarily in Africa. In this context, vulnerability is perceived as a concept that combines food security with risk factors that increase or decrease food security (FSTAU and FEWS, 1997).

Food security measures access to or control of resources required to meet basic needs for food, clothing, and shelter that are necessary to lead a healthy and active live. It includes two important components: food availability and food access.

Risk factors include an external and an internal element. The external element covers events or conditions that affect a household's food security, for example drought and economic change. They are referred to as shocks or hazards. The internal element deals with the coping resources of a household. These are factors that determine a household's ability to re-establish its food security once a shock has come to pass. For example, a subsistence farming family with few assets and limited skills which is experiencing frequent and severe droughts would be classified as a household with high risk factors, because it is exposed to frequent shocks and has low coping ability.

2.3.2 Types of Vulnerability

• Environmental risk (droughts, floods, and pests).
• Market risk (price fluctuations, wage variability, and unemployment).
• Political risk (changes in subsidies or prices, income transfers, and civil strife).
• Social risk (reduction in community support and entitlements).
• Health risk (exposure to diseases that prevent work).

2.3.3 Approaches for Measurement

A recent review meeting of food security and famine early warning experts distinguished between three major approaches to carry out vulnerability assessments: The top-down approach used by the USAID-financed Famine Early Warning System (FEWS) and two bottom-up approaches, used by Save the Children Fund (SCF) and by Agence Européene pour le Développement et la Santé (AEDES), respectively (Global Information and Early Warning System, 1997).

FEWS uses food security indicators linked to or reported by geographic area from which the food security situation at the household level is inferred. Data inputs include remotely sensed data, primarily measures of vegetative vigor and land use, and agricultural and socio-economic data at administrative level.

Conceptual frameworks for vulnerability assessments vary between countries, but generally include measures for chronic (baseline) vulnerability and acute vulnerability (reflecting the success or failure of the most recent agricultural season). Most indicators measuring baseline vulnerability could become an appropriate input for poverty assessments.

SCF conducts their vulnerability assessments in a two-step approach (Seaman et al., 1993). First, they delineate different food economy zones which are areas that are more or less homogeneous in their livelihood systems. Then, key informants from district to village level are interviewed to obtain information on the current food economy of poorer, middle, and richer households, thus producing an assessment for a larger geographic region. The methods used to delineate food economy zones could be adapted to develop a typology of economic activities. Figure 13, page 50, shows an example of such food economy zones in Sudan.

Similar to SCF, AEDES relies on interviews of key informants. The collected information is aggregated upwards and reported by administrative units.

Analysts conducting vulnerability assessments have adapted their approaches to reflect the differences in local risk factors that are caused for example by the physical geography, agroclimatic conditions, colonial history, agricultural production systems, infrastructure, economic and policy environment, and data availability. Tables 2 - 6 summarize the conceptual frameworks and indicators of five different vulnerability assessments in Africa (Ramachandran and Eastman, 1997). For some of these vulnerability assessments, all variables were combined to calculate a composite index, for example with the help of z-scores, in others the different variables were mapped for visual interpretation and overlay analyses. Selected components or variables of these assessments could be appropriated for poverty mapping.

The first example, a vulnerability assessment for four countries in the Sahel, differentiates between baseline and current vulnerability and is summarized in Table 2, page 14 (Wright et al., 1995). Two components, one reflecting the resource base and the other income structure, include variables that could be adapted for poverty analysis.

Table 3, page 14, a summary of a vulnerability assessment for Kenya in 1995, lists various variables under 'depth of household entitlement base,' which could be of interest for poverty mapping (USAID FEWS, 1995). Most of them are similar to those shown in Table 2 (access, livestock income, and cash crop income). The variable 'hectares of high potential land equivalent per capita' could be a good indicator of agricultural resource endowment in a poverty assessment for the CGIAR System.

The third example, a framework for famine early warning in Kenya, is less complex and shown in Table 4, page 15 (Hutchinson et al., 1992). The authors used an unusual variable, 'number of locally funded schools per capita per district,' as a measure for local wealth.

The vulnerability assessment for Zambia in Table 5, page 15, covers three dimensions: crop risk, market access, and coping strategies (USAID FEWS, 1994a). The variable 'share of drought resistant crops' suggests a link to agricultural research priorities. One could develop similar crop related variables, for example share of high yielding varieties or share of salt tolerant varieties in a poverty mapping activity for the CGIAR System.

A more recent vulnerability assessment for Zambia expands on the approach used in 1994 (WFP, 1996). It also differentiates between risk and coping variables and conducts analysis at the district level. Risk variables include comparable measures related to drought risk based on different NDVI indicators. Coping ability is grouped into a short-term and a long-term component. The short-term component includes variables similar to the ones used above: average per capita cereals production, average per capita cassava production, average per capita cash crop production, average per capita fisheries production, average per capita livestock offtake, other transfers, wild food gathering, and wages and salaries. Unlike the 1994 assessment, the income and assets data are more detailed (sometimes missing data were estimated) and were converted to per capita cereal equivalents. They were then aggregated to district level and compared to the minimum per capita cereal equivalent requirements to reach the poverty line in Zambia (based on the 1991 Priority Survey). The long-term indicators included 19 variables covering demographics, health and nutrition, access to infrastructure, and education.

In their vulnerability assessment for Zimbabwe (Table 6, page 16), the authors calculated a composite development indicator, comparable to UNDP's Human Development Index and Composite Poverty Index at national level (USAID FEWS, 1994b). They sought to capture the level of development at the district level and combined six variables: child mortality, female literacy, housing type, electrification, sanitation, and type of energy use.

Most of the vulnerability maps described above collect and analyze data at the administrative level and infer local (household level) conditions from these aggregated data. Such an approach produces rapid results at relative low cost, but often is too coarse for program design and targeting, especially when the subnational units are large in size or have a very heterogeneous population. FEWS is currently testing the integration of additional and more detailed data sets in their assessments. For example, a new research project for the Sahel tries to combine survey data with traditional food security variables (McGuire, 1997; Josserand, 1997). A recent vulnerability assessment for Malawi combined poverty data with the more traditional food security variables.

In the Malawi vulnerability mapping exercise, FEWS collaborated with the Malawi Ministry of Economic Planning and Development and their Poverty Monitoring Systems (WFP et al., 1996). They used statistical methods to combine biophysical and socioeconomic indicators. The study proceeded in the following four steps:

1. The first step defined a conceptual framework that captures the causes (risks of producing and acquiring food), responses (resilience to withstand these risks), and outcomes (results of food insecurity and vulnerability processes) of food security.
2. With the help of an expert panel, 154 subnational units (Extension Planning Areas) were grouped into five major, discontiguous geographic clusters. These clusters were differentiated by key variables related to food production and acquisition and included a cluster dominated by maize production, a cluster with mixed-agricultural production systems, a cluster dominated by large-scale farms, a cluster where households were largely dependent on non-agricultural income, and a cluster dominated by major urban centers.
3. With the help of principal components analysis, indicators measuring vulnerability outcomes (variables related to poverty, access to food, malnutrition, and coping strategies) were combined into three composite measures of vulnerability outcomes for all 154 subnational units.
4. The final step involved a detailed analysis to identify the most important factors determining these vulnerability outcomes with the help of multivariate regression.

2.4 PERIPHERAL AREAS

• Why do poor areas exist?
• How much does geography determine the spatial concentration of poverty?
• If geography determines the concentration of poverty, can we identify areas where poverty is largely determined by geographic factors? These locations could then be defined as peripheral areas.

Analysts examining the causes and spatial clustering of poverty, generally point to individual or structural explanations. Individual explanations concentrate on human capital (education, skills, etc.) and endowments of productive resources. Structural explanations focus on structural factors that constrain opportunities. They include constraints imposed by the economy, social system and geography, for example limited job supply, discrimination, and poor natural resource endowments (Crump, 1997).

Ravallion (forthcoming) summarizes explanations for poor areas under two theoretical models, one named "individualistic" and the other "geographic." Both will be discussed in turn.

The individualistic model assumes that people are highly mobile and migrate to or remain in poor areas because of specific wage and price incentives. Poor areas are thus a consequence of personal decisions and, if they persist over time, reflect local resource endowments, rents for housing and land, etc. In some cases, poor areas could also result from a time lag in the adjustment process of labor markets because individuals are unable to migrate or are delaying their relocation.

Poverty researchers using an individualistic model try to identify causes of poverty at the individual level. They do not attribute any causal significance to spatial inequalities in resource endowments (geographic capital), although they see differences in geographic endowment as the sorting mechanism that leads to spatial poverty concentrations. Consequently, they would target their anti-poverty measures towards improving the endowment of individuals, for example by providing training opportunities (Ravallion, 1996b).

In Ravallion's geographic model, the mobility of individuals is restricted and poverty has a causal link to geography. Local factors such as climate, soil type, infrastructure, and access to social services change the marginal returns of investments, for example to a given level of education. Barriers to migration ensure that these differences persist.

The degree to which individual or geographic factors are causing poverty has implications for developing a strategy of agricultural research aimed at improving the situation of the poor in marginal areas. If geographic factors play an important role, then geographic targeting of agricultural research to the poor in marginal lands can become a useful tool to address poverty issues. If individual characteristics explain most of the local poverty, and individuals are free to migrate, then the mobility of people and capital will limit the success of targeting marginal lands.

Each of the two theoretical models has shortcomings in explaining the spatial clustering of the poor. The two models have not been compared sufficiently yet. Typically, either one or a combination of individual and structural factors are identified as causes for poverty and its spatial concentration (Miller, 1996).

Ravallion (forthcoming) cites a number of studies that support a geographic model of poverty. Empirical research on poor areas in China and Bangladesh shows significant geographic effects on living standards after controlling for non-geographic characteristics (Jalan and Ravallion, 1997; Ravallion and Wodon, 1997).

A study on migration and poverty in the USA confirmed that the spatial concentration of poverty is a reflection of differences in economic opportunities. In this study, poor people migrated to poor areas, because they faced an overall lack of opportunities throughout most areas. High poverty areas provided them with small but real economic opportunities, for example a greater availability of low skill jobs and inexpensive housing. The authors concluded that the causes of poverty differ from the factors leading to its spatial concentration. A spatial association between poverty rates and the social and economic characteristics of high poverty areas does not always point to the root causes of poverty. Thus anti-poverty interventions need to be targeted within and outside these areas. They cannot be successful by concentrating efforts solely on high poverty areas (Nord et al., 1995).

A detailed study of high poverty areas, however, could identify the opportunity structure that attracts and keeps poor people (Nord et al., 1995). This structure is often seen as an impediment for people trying to escape from poverty or for the effectiveness of anti-poverty interventions (spatial poverty trap). Spatial poverty concentrations may be intensified by further discrimination or exclusion. For example, a bank decides not to extend its credit programs to high poverty areas. The subsequent lack of access to financial services will impact local economic development, increasing the differences between poor and non-poor areas (Leyshon, 1995).

There is some empirical evidence to defend the concept of a spatial poverty trap. A study for rural China supported the idea of a spatial poverty trap (Jalan and Ravallion, 1997). Studies examining urban poverty concentrations in the USA have used the neighborhood effects hypothesis which assumes that the prospects for leaving poverty are partly influenced by the neighborhood, e.g. access to education and other services, and its social environment, e.g. values of local communities affect individual aspirations and expectations (O'Regan and Wiseman, 1990).

While empirical research has shown geographic effects on the level of poverty and the rates of poverty reduction, more research is needed to examine cause and effects in more detail. The studies cited above could isolate a geographic effect and identify peripheral areas, but could not identify the specific characteristics of these areas that were responsible for the level of poverty. There are many possible interacting geographic factors such as access to markets, availability of goods and services, limited agro-climatic conditions, low quality of labor force, and poor entrepreneurial environment (Rasmussen, 1986). Whether it is one of these factors or a combination of them that contributes to the spatial clustering of poverty depends very much on the local situation.

Without additional research, it is difficult to identify universal characteristics that could explain the geographic effects on poverty or define peripheral areas. It can be assumed, however, that isolation is one important element of the factors explaining geographic poverty effects. Isolation may not always cause poverty or be among the major geographic determinants of poverty. Isolation will certainly accentuate poverty and increase the risk of natural disasters. Isolation may even reduce the motivation to produce, because markets are not close by to sell additional outputs or consumer goods are not available to purchase with the proceeds of marketable surplus (El Sherbini, 1986).

Isolation can be measured with different access indicators. This could include access to resources, land, infrastructure, irrigation, technology, transport, communication, social services, or labor and capital markets.

Choosing a specific conceptual approach to define poverty determines what type of data to compile and map. For example, using an economic definition of poverty requires comparable household income or expenditure data that are usually collected through household surveys such as the World Bank's Living Standards Measurement Study (LSMS) surveys. However, such survey data are not available for all developing countries. Ultimately, the TAC and the different CGIAR centers need to agree upon a poverty definition that meets the following criteria:

• It is generally useful for CGIAR's envisioned exercise of setting priorities and analysis,
• provides a sufficient pool of data,
• is relevant to agricultural research, and
• allows to compile poverty statistics at the subnational level.

A case study using household survey data for Côte d'Ivoire compared seven indicators of human well-being: household or per capita consumption, per capita income, per capita food consumption, food ratio (proportion of household budget spent on food), food consumption expressed in calories, anthropometric data (weight-for-height and height-for-age measurements) of children, and basic needs (households are defined as poor if their basic needs for food, shelter, clothing, health care, and education are not met). The study simulated targeting households by each of the seven indicators. The results showed that each indicator identified different population groups as poor with very little overlap between the indicators, except for the economic measures which had higher correlation coefficients (Glewwe and Gaag, 1988).

Similarly, a study examining the benefits of geographic targeting compared income-based poverty measures from household surveys with a poverty map for Venezuela. The poverty map was based on a composite index developed with the help of a principal components analysis of 32 indicators such as unemployment, literacy, and access to safe water. The composite poverty map and the economic indicators differed significantly in their ranking of states (Baker and Grosh, 1994).

The findings of the studies above were confirmed by differences in targeting results between a consumption-based poverty measure and a composite basic needs index (access to water, access to sanitation, access to waste disposal, education, and household crowding) in Ecuador. At the regional level, both methods came up with similar rankings of broad geographic regions, but rural areas appeared poorer under the basic needs indicator. A comparison of household rankings, however, showed less correspondence between the two alternatives. For example, households in the poorest quintile under the basic needs methods did not completely match those grouped into the poorest quintile under the consumption measure. Only 40.8 percent in the bottom quintile of the households identified with the basic needs method would be also included in the bottom quintile using the consumption based indicator (Hentschel et al., 1997).