ABSTRACT: Despite rapid growth in aggregate real income, Botswana continues to face endemic problems of unemployment, poverty and undernutrition in rural areas. Both the depth of poverty and its dispersion are considerable among the poor, especially in small villages and settlements. Poverty is a function of both the level of income and the degree of inequality of wealth. These are significantly related to differences in household size in adult equivalent, level of livestock ownership, educational attainment of principal income earner, and proximity to urban labour market. Poverty is gender differentiated, but it is more linked to gender-specific differences in these determinants of living standards. A simulation of how much these discriminates screen the poor shows that they are capable to capture a greater percentage of the poor when they are applied in combination than separately, but how they are combined is crucial in determining the extent of targeting the poor.
Botswana is the fastest growing economy in sub-Saharan Africa. This has been largely attributed to rapid growth in the mining sector (Botswana MFDP, 1991). But, despite its high rate of aggregate real income growth, the country faces sizeable unemployment and poverty, particularly in rural areas. Open rural unemployment averaged 23.5 per cent in the 1980s (Botswana CSO, 1986), which is considerably higher than the average for most sub-Saharan African countries. A considerable amount of household work time is also under-utilised in rural areas (Lipton, 1979; Mueller, 1985). Prevalence of income-based poverty measure averaged 55 per cent in the mid 1970s (Colclough and Fallon, 1983). It is plausible that this did not appreciably fall in the 1980s due to the effect of six consecutive years of drought. As it is evident from the report of the 1991 population census, the low standard of living in rural areas has precipitated rapid movement of labour, particularly working adult males, from impoverished small villages to larger villages and urban areas (Botswana CSO, 1991).
These twin problems of structural unemployment and poverty are closely linked to the poor performance of the agriculture sector. Agriculture, which is the mainstay of the rural economy, showed a marked decline in both relative and absolute terms in the 1980s (Botswana MFDP, 1991). Given that the income of the poor depends mainly on crop-related activities, which commonly share a strong rainfall-linked covarite risk, the decline in income due to drought was more than proportionate among the poor than the non-poor households (Watanabe and Mueller, 1984; Valentine, 1993). This income-widening effect of the drought was further reinforced by a more than a proportionate loss in livestock among small herds. Although the flow of private and public transfers largely offset the decline in farm income, the evidence is not conclusive as to how much such a poverty enhancing effect of the 1980s drought was neutralised through these income transfers.
The thrust of this paper is to study the extent and determinants of poverty in small rural villages in the early 1990s. These villages in general experience a slow growth in population due to net outmigration of working adults. They have more younger and aged population as compared to the adult population. And, within the working age category, in rural areas, there are more females than males, which indicates some age-gender selectivity in migration patterns. There is a great deal of income transfer to these villages to support the livelihood of the population even in normal non-drought years.
The paper begins with a definition and different measures of poverty. It follows the official definition of poverty. But it departs from previous studies (Colclough and Fallon, 1983; Watanabe and Mueller, 1984) in its measurements. First, it uses household consumption expenditure, instead of income, as a measure of command over basic needs. Second, it employs a more systematic approach in valuation of basic needs, which primarily involves estimation of some minimum food and non-food expenditure. Third, it provides estimates of a multivariate regression model to track the key determinants of poverty incidence in these villages. Following the tradition of Watanabe and Mueller (1984), the main characteristics of these households are revisited to trace if there are changes in the characteristics that distinguish the poor and non-poor households in these villages. Finally, it attempts to test how much the key poverty indicators screen the poor, and demonstrates their relevance for improving targeting practices.
This study draws on primary data collected from rural household surveys in eight "small" villages1. These were drawn mainly from the five districts running north-south in the eastern part of the country, which is the home of 80 per cent of the population. The final sample includes villages with some variations in ecology, population density and infrastructure development.
The survey was carried out during November 1991 and January 1992. Its main components included structured interviews of sampled households on their demography, time use, income and expenditure flows, asset holdings and credit transactions. In addition, the survey work involved interviews with village or community leaders to understand how the communities characterise the poor and what strategy the poor follow for their livelihood.
Definition of Poverty: The poor, according to the official definition, constitute households who are unable to meet their basic needs. The official poverty datum line (PDL) gives a monetary measure of the minimum cost of these basic needs for differently compost households (Botswana CSO, 1976). The procedure for determining PDL involves fixing a common basket of basic needs, valuation at market prices, and then applying age-and-sex specific weights to reflect variations in consumption requirements between households.
This study follows the same definition. That is, individuals are deemed poor if they cannot afford to meet their minimum basic needs, which include food, clothing, housing, etc. But it differs in its approach. First, it establishes its basket of basic needs from the actual 1991-92 household budgetary outlays. Second, the consumption basket of the low income households is used to anchor the poverty line. Third, it applies a more systematic approach to capture demography and price effects on determination of the minimum cost of basic needs.
Setting Poverty Line: The absolute poverty line, which represents the monetary measure of minimum basic needs, was estimated in two steps. The first stage involved estimation of the food poverty line. This followed the work of Wasay (1977). The actual food basket of a reference low-income household (defined here as a household with average characteristics-household size, working adult family members, asset ownership-of the households at bottom 40 per cent of the income distribution) was first determined from household food consumption data. Such a food basket was then normalised so that the resulting per capita calories were equal to 2100 calories per day per person. Each of the scaled quantities was then valued at average village-specific market prices to determine the per capita food expenditure to meet the target calorie level. This set the food poverty line (Zf).
This procedure ensures that the reference consumption basket is anchored to the consumption preference of the low-income households. In addition, this has the advantage of excluding households with a propensity to consume expensive calorie sources (low income households are likely to consume low-cost calorie sources as compared to high income households). That is, it excludes those households which can afford but fail to meet their calorie requirements. This may be ascribed to inadequate information and/or food habits and customs.
The next stage involves estimation of the level of non-food expenditure by low-income households. Specifically, the measure of the non-food expenditure has to be consistent with expenditure patterns of those households at the poverty line rather than for the entire low-income households. This has also the advantage of excluding households which afford to meet their food requirements, but choose to allocate a high share of their income to non-food consumption. The estimation procedure followed the work of Ravallion and Bidani (1994). This involved estimation of non-food allowance for low-income households at the food poverty line1.
The total poverty line was then derived from the above two-stage procedure. The total poverty line for Botswana at RDA level of 2100 calories per capita was equivalent to a monthly per capita expenditure of Pula 38 at 1991 prices. In addition, two poverty lines were estimated to provide a lower variant (based on daily 1,700 calories per capita) and an upper variant (based on 2400 calories per capita). These lower and upper bounds allow for individuals to vary their calorie requirements. The values are within the 15 per cent from the RDA level, which Sukhatme (1977) found to be the mean coefficient of variability of energy requirements for individuals.
Poverty Aggregation: The next stage was to aggregate poverty among the sample population. Here, the three FGT (Foster, Greer and Thorbecke, 1984) poverty measures1 were applied to determine poverty incidence (P_=0), depth (P_=1) and severity of poverty (P_=2). The first measure gives the percentage of the population living below a poverty line. But such a measure fails to indicate the depth of poverty below the poverty line. The second measure (P_=1), which is the average proportional shortfall of consumption weighted by the head count index, gives the average shortfall of living standard below the poverty line. But the average depth measure of poverty is incapable of indicating the variation in depth of poverty among the poor. Thus, a poverty measure that is sensitive to inequality in distribution of income among the poor (_=2), which satisfies Sen's [1976] monotonicity and transfer axioms, is estimated to show the relative variation in poverty depth among the poor. These FGT measures have the additional property of being additively decomposable, which allows us to decompose poverty into its totally exhaustive subgroups.
The head-count measure shows a poverty incidence of 33 per cent at the low variant poverty line (Z=28.09 Pula per month per person). The incidence increases with poverty line; 46 per cent at Z=35.08 and 55 per cent at Z=40.54. But, it increases at a decreasing rate. The arc elasticities of the head-count measure with respect to poverty line exceeds more than unity between these poverty lines (1.5 between poverty line 28.09 and 35.08, and 1.2 between 35.08 and 40.54), but decreases with increase in poverty line.
The poverty gap index (P_=1) gives a mean consumption shortfall of 15.2 per cent among the poor. The average shortfall ranges between 9.5 and 20 per cent between the lowest and the highest poverty lines, respectively. The FGT severity index (P_=2) indicates poverty that is, the consumption shortfall below the poverty line, is unequal among the poor. The depth of poverty is thus much greater at the lower end of income distribution in these villages.
The earlier works of Watanabe and Mueller (1984) and Colclough and Fallon (1983) show that poverty is more prevalent in remote small villages than in large villages. Small villages with limited opportunity for gainful employment and income generation experience much lower living standards.
This study shows that there is considerable variation in the extent and severity of poverty ever within these villages. As it is shown in Table 1, poverty is more marked in two villages (Mahetwe and Moyiabana) where at least 60 per cent of the households were below the poverty line. Both the depth of poverty and its distribution are also greater in these villages. These two villages alone accounted for nearly 40 per cent of the population below the poverty line and contributed to 42 per cent of the inequality among the poor in the sampled survey villages.
These very poor villages are typically remote from major population settlements where there are growing and better paying employment opportunities. The population is highly dependent on low-paying income generating activities such as grass gathering and selling, beverage making and trading, sorghum stamping, and paid herding. There is high outmigration of male labour. Many villagers depend on income transfers, especially from private sources.
Table 1: Measures of Total Poverty by Village, 1991-92
Village (District) |
Per Capita monthly expenditure (Pula/person) |
Poverty Index (%) P_=0 P_=1 P_=2 |
Contribution to Poverty (%) P_=0 P_=1 P_=2 | ||||
Gathwane-Digawana (Southern) |
67.14 |
27.2 |
8.3 |
3.3 |
14.4 |
12.5 |
12.0 |
Mahetwe (Kweneng) |
39.14 |
63.3 |
21.2 |
10.8 |
16.9 |
17.9 |
17.7 |
Malolwane (Kgatheng) |
66.75 |
43.5 |
12.1 |
4.7 |
16.1 |
14.4 |
13.5 |
Molyabana (Central) |
32.28 |
66.3 |
29.9 |
16.3 |
20.3 |
23.1 |
24.0 |
Makobeng (Central) |
44.83 |
55.0 |
20.4 |
9.5 |
13.7 |
13.9 |
15.1 |
Sechele-Makaleng (Northeast) |
48.38 |
42.1 |
10.3 |
4.0 |
18.6 |
18.2 |
17.7 |
in the survey villages the poor were in smaller size households than the non-poor. The sample from the 1991-92 survey shows that the poor whose per capita expenditure was less than 75 per cent of the poverty line, had, on average, 3.7 adult equivalent members. For the non-poor, whose income was at least 25 per cent above the poverty line, the average size was 5.7 adult equivalent members. Within the poor households, there were more dependants than working adult members, especially male adults. These characteristics were more pronounced among female-headed households. The small household size of the poor in these villages was, in fact, a reflection of the large number of female-headed households, which accounted for no less than 60 per cent of those sampled.
The evidence from the 1975 Rural Income Distribution Survey (RIDS) provides a similar pattern. Watanabe and Mueller (1984) found poor households to be numerous with high number of dependants from a large cross-section of the sample that included different households groups. Kossoudji and Mueller (1983), on the other hand, found the poor to be less numerous among female-headed households without working adults. The size of poor households is thus conditional on the extent to which female-headed households are represented in the sample. Poverty and household size are thus contingent on the type of household.
Poverty is more pronounced households with absent adult males, particularly female-headed households. The poverty measured from the 1991-92 survey shows a poverty gap (P_=1) of 18.0 and 9.4 per cent for households without adult-male support and with male-adult support, respectively. The respective mean values of poverty severity index (P_=2) were 9.0 and 3.6 per cent. Within female-headed households, the extent of the poverty gap and severity index were 24.4 and 13 per cent, respectively.
The works of Kossoudji and Mueller (1983) and Watanabe and Mueller (1984) also confirm that female-headed households are in general poorer than male-headed households, especially those without adult-male family labour. The absence of adult supporting males is particularly critical since there is an apparent male-biased selection in entry into better income generating activities, especially wage employment.
The poor tend to be older than the non-poor in the 1991-92 survey villages. These old poor families have many dependants, which is unique to rural Botswana. This is partly related to the common practice of having children before marriage and keeping them in rural areas with their grandparents. For example, there were no less than 24.3 per cent of the children in the 1991-92 survey villages who had absent mothers.
However, the findings from the 1975 based study show that this age-poverty link is not generally applicable since there are some old households who are better off because of their earning from their past investments and asset accumulation.
Households with educated working members have greater access to better paying employment. For example, wages in rural areas are positively related to the level of education (Mueller 1985). But the poor with working adults are generally less educated and skilled. Even those who migrate from these poor households are less endowed in term of educational attainment. Hence, they are less likely to move out of poverty.
There is some minimum threshold for education to make a difference in living standards. For example, Matanwabe and Mueller (1984) suggest a minimum of 10 years of education as the critical threshold level. But none of the other poverty studies attempted to determine such a threshold. Hence, it is not yet evident where the threshold for education should be to make a significant difference to living standards and to bring about substantial reduction of poverty.
The main form of asset accumulation in rural Botswana is possession of livestock, especially cattle. Households with cattle are better placed to engage in crop farming, to buffer food consumption in time of crop failure, and to obtain easy access to formal and informal credit markets. Hence accumulations of livestock is, an important form of saving to function as sources of income growth as well as income buffer.
The results from the 1991-92 survey show that the mean level of cattle equivalent ownership per person was 0.65 units for the poor and 1.3 units for the non-poor. Among those without livestock, the poverty gap averaged 17 per cent whilst for those with livestock it averaged 12.3 per cent. The poverty severity was also greater among those without cattle (8.4 per cent) as compared to the severity among owners (5.1 per cent).
Watanabe and Mueller (1984) found that ownership of smaller animals was more widespread than cattle, but it was not evident that poor households were able to raise more small animals than the non-poor to compensate for low holding of cattle.
Rural households are engaged in a variety of income generating activities. On average, a rural household has three to four income sources in a particular year. The main income sources originate from self-employment in agriculture (crop and livestock income), wage employment, and transfer income from private and public sources.
There are some systematic differences in income sources between the poor and non-poor households. The earlier studies based on the 1975 Rural Income Distribution Survey (Colclough and Fallon, 1983; Kossoudji and Mueller 1983; and Matanwabe and Mueller, 1984) show that incomes from livestock production and wage employment are more concentrated among the non-poor households. The poor depend mainly on low return income sources, which are also greatly dependent on highly variable rainfall (crop production, gathering, beer-making, and so forth). However, the poor are largely compensated for such low and variable earnings through greater access to transfer income, which tends to be positively related to the degree of rainfall failure (Lucas and Stark 1985).
The evidence from the 1991-92 survey shows that income from crop farming barely exists in these villages (Table 2). The poor are engaged in low paying manufacturing and trading activities. Wage employment contributes a significant share of income for both the poor and non-poor households. But the contribution of wage income to the poor was largely from the low paying special public provisioned rural work scheme in the villages. Private transfers are widely spread among the poor and non-poor households. The dominance of transfer income in these sampled households indicates a prevalence of earlier high migration from these villages.
Table 2: Rural Household Income by Source at Sample Mean (%)
|
1974-75a |
1985-86b |
1991-92c | |
|
|
|
Poor |
Non-poor |
Self-employment |
|
|
|
|
Farm - crop |
9.7 |
3.2 |
2.7 |
2.6 |
Farm - livestock |
29.0 |
20.2 |
15.2 |
18.9 |
Non-farm |
25.8 |
13.4 |
22.0 |
18.6 |
Wage employment |
21.5 |
34.5 |
33.8e |
28.6 |
Road employment Other wage |
|
|
|
|
Transfers |
14.0 |
28.7 |
26.3 |
31.3 |
Source: Colcough and Fallon [1983]; HIES [1988]; and 1991-92 Rural Survey, [1991-92]
Note: a Nationally Representative rural income survey and weights were readjusted after excluding the imputed value of an owner-occupied dwelling (see Valentine [1993]).
b. Nationally Representative Household Income and Expenditure Survey (HIES).
c. Household Survey in 8 small villages of rural Botswana. The poor are defined as individuals with income below 75 per cent of the poverty line. The non-poor as individuals with a mean income of 25 per cent above the poverty line.
d. Income from the road work scheme accounts for 27.1 and 19.8 per cent of the
poor and non-poor, respectively, in 1991-92.
The income sources observed in these small villages is markedly akin to the pattern that emerged during the drought years of the 1980s (Table 2). As the analysis of the 1985-86 Household Income and Expenditure Survey (HIES) shows, the decline in crop and livestock income was partially offset by an increase in wage income and increased public and private income transfers (Valentine, 1993). The wage income was mainly generated through labour intensive relief work schemes, which were part of the Government's drought relief programme. Both private and public income transfers also increased in the drought years.
These income sources are not distribution neutral. Following the procedures in Lerman and Yitzhaki (1985) and Shalit (1986), the total Gini coefficient is decomposed as,
Gy=_ wiGiRi
where Gy is total Gini Coefficient, wi is the share of income i in total income, Gi is source specific Gini Coefficient, and Ri is correlation ratio between income i and total income. The estimates of these components are given in Table 3. In addition, there are two summary measures in the Table-relative concentration coefficient (gi), which indicates whether an income source has equalising (if less than one) or widening (if greater than one) or neutral (if it is equal to one) effect on income distribution, and factor inequality weight (wigi), which measures how much a specific income source contributes to overall income distribution.
Table 3-Decomposition of Total Income Inequality by Source
(1991-92 survey)
Source |
Income share(wi)a |
Source Gini (Gi)b |
Correlation ratio(Ri)c |
Concentration coefficient(gi)d |
Factor inequality weight (wi gi)e |
Farm income |
0.19 |
0.71 |
0.63 |
1.06 |
0.21 |
Trading-A |
0.05 |
0.85 |
0.21 |
0.42 |
0.02 |
Trading-B |
0.06 |
0.98 |
0.85 |
1.99 |
0.11 |
Non-farm-other |
0.09 |
0.56 |
0.52 |
0.69 |
0.06 |
Wage-project |
0.16 |
0.71 |
0.33 |
0.55 |
0.09 |
Wage-other |
0.13 |
0.88 |
0.59 |
1.24 |
0.16 |
Transfers |
0.32 |
0.68 |
0.67 |
1.09 |
0.35 |
Source: Rural Survey, 1991-92.
Notes: a wi = _i = Ratio of mean income of i th source _i to the mean of total income (_).
_
b Gi = 2 Cov(Yi,ri) where Cov(yi, ri) is the covariance between income
n_i level of i th income and its income source rank ( ri).
c Ri =Cov(Yi,r)/ Cov(Yi,ri), where Cov(yi, r) is the covariance between
source incomei and total income rank (y ).
d gi Ri(Gi/Gy) = Relative concentration coefficient of i th source in overall income inequality.
e wigi = Factor inequality weight of the i th source in overall inequality.
Gini Coefficient for aggregate income per capita, Gy = _wiGiRi is 0.42
As Table 3 shows, the incomes that accrue mainly to the non-poor have a relative concentration coefficient of more than one. That is, incomes from livestock, skilled wage employment, and self-employment in large commercial transport and trading business (Trading-B) contribute to an increase in overall income inequality. Among these income sources, livestock accounts for the largest share of the overall income distribution, as indicated by the factor inequality weight. It has thus a considerable income widening effect. The effect of income transfers on neutralising the income inequality of these income sources is weak since it is neutral in its distribution effect. The income sources of the non-poor thus not only contribute to difference in level of income, but also to inequality in income distribution. Prevalence of poverty in these villages is thus a making of both differences in level and inequality of income distribution.
The main explanation for such systematic variation in income sources between the poor and the non-poor lies in a difference in productivity of labour, which is largely ascribed to a difference in access to assets, especially livestock. As the work of Matanwabe and Mueller (1984) shows, there was little difference in work time allocated between the poor and non-poor to explain variation in income sources. Composition of labour, particularly male adults with access to wage employment, which is a return to education, explained some of the systematic variation in income sources. But, the effect of access to assets, particularly livestock, was by far the most important factor in explaining variation in income level and composition among poor and non-poor households.
The food expenditure patterns for households differentiated along the expenditure-poverty continuum are summarized in Table 4. Food constitutes the main component of household consumption expenditures. It accounts, on average, for 68 per cent of the total household budgets for the poor. Its share is lower among the nonpoor, but it still averages more than 50 per cent of total expenditure. Both groups allocate no less than 70 per cent of their food budgets on three food groups: cereals, meat and other animal products, and sugar and sweets. There is, hence, no notable difference in the composition of their food baskets.
Table 4: Comparison of Food Share as per cent of Total Expenditure by
Commodity Groups and Distance from Poverty Line, 1991-92
Distance from Poverty Line |
| ||
Food group |
Below 75 per cent |
Between 75 and 125 per cent |
Above 125 per cent |
Food share |
68.1 |
64.5 |
54.1 |
Cereals |
33.1 |
25.7 |
18,.7 |
(Maize) |
(10.2) |
(8.8) |
(5.8) |
(Sorghum) |
(14.9) |
(8.1) |
(4.7) |
Livestock products |
10.2 |
14.1 |
15.3 |
Sugar and sweets |
7.9 |
6.8 |
5.0 |
Daily calorie per capita |
1,624 |
1,882 |
2,470 |
However, these households differ in the extent of their food allocation response to change in income level. The food shares for cereals and sugar decline with income, which indicates that these food groups have low income elasticity. In fact, for sorghum, both the level of expenditure the food share decreases with increases in income level, which indicates that it is an inferior food crop. On the other hand, the food share for animal products (for example, meat and milk products) and some cereal crops and products (for example, wheat flour, bread, and rice) increase with income level. These food items are income elastic and, hence, constitute relatively a larger share of food budgets among the non-poor households.
Despite the large allocation of household budgets to food, the level of food consumption, as measured in daily calorie availability per person, is below the recommended daily allowance (RDA=2,100 calories per person) for individuals at and below the poverty line. Based on the 1991-92 weekly recall consumption data, the poor consumed on average 1,624 calories per day per capita (those with an income level below 75 per cent of the poverty line) as compared to 1,882 calories per day per person around the poverty line (those with an income within plus or minus 25 per cent of the poverty line), and 2, 470 calories per day above the poverty line (those with an income at least 25 per cent above the poverty line). The poor are thus markedly undernourished in terms of their calorie consumption.
Why these households consume such a low level of calories cannot be explained fully by the income level alone. There is reason to suspect that the choice of diet in these rural villages is not driven purely by the desire to meet the calorie goal. For example, the difference between expenditure and calorie (quantity) elasticities averages 0.28 which indicates that households purchase food for reasons other than pure calorie considerations. There is a tendency even among the poor to consume expensive food, which, if necessary, can be reallocated to obtain a higher level of calories for the same level of income.
The results so far indicate that the poor in rural Botswana have distinguishing identifiers. Poor household are small in terms of adult equivalent household size. The poor are less educated and skilled. They have low asset holdings, especially livestock, which is an important determinant of wealth in rural Botswana. These characteristics are more pronounced among the female-headed households, especially among those with little supportive transfer income.
How much then, do these factors explain variation in poverty among the population in small rural villages? None of the past studies have attempted to isolate and measure the effect of these variables on the incidence of poverty. Hence, this study attempts to fill in the gap by estimating a multivariate regression model.
The dependent variable in the regression equation is defined by a continuous variable, a ratio of income to total poverty line. For a household at the poverty line, the ratio is unity. For those above (below) the poverty line, the ratio is greater (less) than unity. The lower bound of the ratio is zero. Where the sign of an explanatory variable is positive, it has the effect of either moving the household closer to the poverty line from below (if the household is below the poverty line to start with) or crossing the poverty line (if it is already near the poverty line) or moving away from the poverty line (if it is already above the poverty line). A variable with a positive sign thus indicates improvement in living standard. A variable with a negative sign has the opposite effect on living standard.
The multivariate regression estimates and associated standard errors are given in Table 5. The F-test is significant at less than one per cent probability. That is, the independent variables as a group explain a significant part of the systematic variations in the normalised per capita expenditures (R2 = 23%). The signs of the individual coefficients, which have t-values of at least 2, confirm the patterns observed in the descriptive analysis.
Table 5: Ordinary Lease Squares Estimates of Consumption Expenditure-to-Poverty Ratio1
Variable and definition Parameter estimate
(Standard error)
Age of household head in years -0.01
(X = 51, SD = 15) (.01)
Age of household head squares 0.85E-04
(0.13E-03)
Sex of household head 0.01
(1 if male: _ =0.376 SD =0.48) (.08)
Education of household head 0.19***
(1 if at least some primary: X = 0.47 SD =0.48) (0.75)
Size of household in adult equivalent 0.24***
(X = 4.56 SD = 2.04) (0.07)
Size of household squared -0.01*
(0.01)
Dummy for livestock ownership 0.15**
(1 if middle tercile) (0.08)
Dummy for livestock ownership 0.19**
(1 if upper tercile) (0.09)
Share of income transfer to total household income 0.27**
(X = 0.25 SD = 0.27)3 (0.14)
Dummy for village location (1 if close to major [ ]) 0.34***
X = 0.75 SD = 0.43 (90.08)
Constant -0.82**
(0.41)
F-Value 9.63***
R2 0.23
N 275
Notes: (1) The dependent variable is the natural log of the ratio per capita household consumption expenditure to total poverty line.
(2) *** = significant at 1 per cent probability
** = significant at 5 per cent probability
* = significant at 10 per cent probability
(3) The share of transfer income with respect to total household income may introduce the problem of endogeneity and hence a bias in the parameter estimate of the model. The equation was restricted with this variable and there was no significant differences in the estimated parameters.
Household size defined in adult equivalent units has a significant positive but diminishing effect on income-poverty ratio. That is, large households with more able- bodied adults are able to move upward on the income-poverty continuum. There are at least two plausible factors that contribute to such positive association between household size in adult equivalents and indicator of living standards. First, these large households are better endowed with working labour able to engage in local income generating activities. Where the local labour market is demand-constrained, these households have a greater propensity to migrate in search of non-local employment.
Evidence on characteristics of migrants from these villages shows that they come mainly from large households. These migrants are often in age group 21-45 years, better educated, and engaged mainly in wage employment. Second, there is a potential economy of scale in consumption in large households. For example, the results obtained from the estimated calorie demand equation, which includes an interaction effect between income and household size1, shows that the calorie elasticity with respect to household size falls (in absolute terms) from 0.26 to 0.23 as household size increases by 26 per cent, holding income constant at mean level. Such a gain in consumption means that the level of minimum expenditure necessary to meet the basic needs of large households is lower than small households, with a similar level of income.
Ownership of livestock contributes to a better living standard. The marginal effect relative to those in the lower tercile of the ownership distribution (the reference group) is slightly stronger for those households at the upper tercile of the distribution. In general, ownership of livestock, especially large animals like cattle, contributes to a better living standard. Those with livestock are more likely to benefit from cultivating a large area, having access to better education and wage employment, and better opportunity for obtaining credit for long term investment in productive income.
Education of the main income earner has a similar effect to ownership of livestock. But there was no evidence of a quadratic effect of education on poverty reduction. The effect of education is much stronger than the impact of livestock ownership, except for households at the upper tercile of the distribution of livestock ownership.
Like livestock ownership and education, access to transfer income (both cash and gifts combined) makes a positive contribution to the income position of a household. The marginal effect is not substantially different from these other income-promoting asset (livestock and human capital) variables.
The effect of gender on the income-poverty ratio is diminished when these other factors are controlled, particularly differences in number of working adults and ownership of livestock. Household size and composition, and access to assets, particularly livestock, are more important in explaining variations in living standard than gender of a household per se.
The effect of the village dummy is positive and statistically significant. In fact, the marginal effect due to location of households is quite strong, holding other factors constant. Households in the poorest villages had markedly lower income-to-poverty ratios compared to the other villages in the group. These are not only linked to differences in household and individual characteristics, but also to the type and size of the labour market. The labour market in these poorest villages is dominated by employment in low wage activities. In addition, the link to places with larger labour markets is constrained by poor physical infrastructure.
In general, these regression results indicate that the key variables that are likely to distinguish between the poor and non-poor are size of household in adult equivalent, education of principal income earner, ownership of livestock, access to transfer income, and location of household (local labour market conditions). Households with better labour and asset endowments, and access to transfer income are more likely to have a higher living standard, as measured by ratio of income-to-poverty line. Neither gender nor age of household head are critical on their own in explaining the incidence of poverty.
These key poverty indicators can be easily utilized to improve the current targeting practices in the country1. For example, if households without livestock are considered as poor, this indicator alone captures 38.1 per cent of the poor in the sample. If it is combined with other indicators, the method of combination matters in terms of the type and size of the poor identified. Suppose the additional indicator is female-headed households with no male support. Then, if only those who satisfy the two conditions simultaneously are selected, the percentage drops to 33 per cent and selects only those who are female-headed with no male support and no livestock. On the other hand, if the two are joined horizontally (that is, each selects independently, the percentage becomes 42.4 per cent (33 per cent are those who satisfy both conditions, 4.3 per cent are female-headed with livestock, and 5.1 per cent are male-headed with no livestock). The intermediate case is where one indicator is used sequentially within the other. For example, one type of such subsetting is to select female-headed households with nomale support within those with no livestock (this captures those 5.1 percent male-headed households with no livestock and 33 per cent female-headed households with no livestock, but misses the 9.4 per cent female-headed households with livestock). Another sequencing is to select those with no livestock among female-headed households. This form of targeting captures 33 per cent female-headed with no livestock, but misses those male-headed households with no livestock (5.1 Per cent). Note that order is important in the latter approach since the percentages missed are not symmetric. All these approaches use overlapping techniques, but the method of application gives different results. Which one of these methods is appropriate depends on the understanding of differences in depth of poverty and the policy objective of anti-poverty programmes.
The expenditure-based poverty measure shows a substantial presence of poor households in the small villages of rural Botswana. Despite rapid economic growth since the 1970s, poverty is still prevalent in rural areas, particularly among the population living in areas where employment opportunities besides high-risk farming are limited. The poor often engage in diverse income-generating activities, but the returns to their labour are low. Furthermore, because of the strong link of these activities to variability of rainfall, they face covariate income risks. As evident from the 1991-92 survey villages, the bulk of the poor depend on private and public income transfer schemes.
Rural poverty is closely related to low adult equivalent household size, level of education of working members, size of ownership of physical assets, especially livestock, and access to supportive income transfers. These characteristics are particularly pronounced among female-headed households with no healthy working male support. However, it is misleading to link poverty strongly to gender, since the critical factors that explain why poverty is prevalent among female-headed households is not due to gender per se, but more due to lack of supportive male adult labour, lack of assets, constrained access to credit and income transfers, and/or a dependency on low earning migrants.
Poverty is unequal among the poor. A simple configuration of the above set of poverty indicators shows several categories of the poor, which include, for example: (i) no male support, no assets, and no transfers; (ii) no male support but with some access to assets and/or transfers; (iii) working adults but no assets and/or transfers; (iv) working adults but with little assets and/or access to transfer income. These different categories of the poor differ not only in terms of their head counts, but also in the severity of their poverty. The very poor are typically without income earning adults, and have little access to assets and transfer income. These are disproportionately concentrated among the female-headed households.
The indication so far from the poverty targeting practices in Botswana shows that policymakers follow the method of horizontal concoction of different indicators. The national welfare programmes provide work for those eligible to work at low wages, direct cash transfer to those adults who cannot work but with no other means of livelihood (no livestock, no transfers), and food as well as health services to the medically vulnerable (children under 10 years and mothers who are pregnant and lactating). These programmes provide a maximum coverage of the poor since it joins horizontally those with work but poor, those without work but poor and those outside the labour force. Such broad coverage entails large budgetary costs, which the country is now trying to save through improving its targeting mechanisms (MFDP, 1991). However, the selection of the poor is not invariant to the method or technique applied, which, in turn, is linked to the objective of poverty reduction. It is crucial for the policymakers to clarify these issues as they search for cost-effective targeting mechanisms.
An important caveat that should be emphasised, but missing in this paper, is the substantial progress the country has made on improving the non-income indicators of welfare of the poor, particularly in reducing infant mortality, child malnutrition, and illiteracy. The country has the most dense health and school infrastructure in sub-Saharan Africa, where 85 per cent of the rural population was within 15 km of a health facility in 1990 (NDP 7). About 80 per cent of the rural villagers had also access to potable water by 1990 (NDP 7). These services are universally available to the rural population. Such progress on social indicators, however, is yet to be matched by a similar reduction in food poverty. The current level of food adequacy is insufficient to maximise the synergy between food intake and health inputs to produce good nutrition and health for the rural population.
1. The settlement patterns in rural Botswana show distinct features which between small and big villages. Villages with population of more than ten thousand are defined as big villages. Typically, the small villages experience low population growth rates, and tend to be poorer, on average, than the big villages (Colclough and Fallon, 1983).
2. The estimation procedure involves the following steps: (1) estimate food share equation; (2) evaluate the share equation at a point where household total per capita expenditure is equal to the food poverty line; and (3) derive the non-food share at food poverty line. The resulting total poverty line is then derived as Z = Zf + Zn, where Zf is food poverty line and Zn is non-food allowance. Since Zn = (1 - _) Zf, the total poverty line can be presented as Z = (2 - _) Zf, where _ is estimated from the food share equation evaluated at a point where per capita consumption expenditure is equal to food poverty line.
3. The FGT measure is specified as:
P_= 1 _ ( Z - Yi) )_,
n i=1 Z
where Z is the poverty line, Yi is the consumption expenditure per capita for the ith household, q is the number of individuals (households) having an income below the poverty line, n is the total population, and _ is a poverty aversion measure.
4. CALPC= 2215.9 + 2260 In(PCXP) + 3191 In(HHS)
(5303.7) (621) (1140)
- 976 In(PCXP) In(HHS) - 182 HHSX + 4314 In(VPMZ)
(292) (109) (1427)
- 1735 In(VPSR) - 1905 In(VSVG) - 1198 In(VPBF) - 659 DVL
(1047) (1136) (742) (696)
F=10.95 R2 = 0.25 N=276
where CALPC=per capita daily calorie availability, PCXP=predicted monthly (w) per capita expenditure, HHS=household size, HHSX=household sex (1 if male), VPMZ=village price of maize per kilogram, VPSR=village price of sorghum per kilogram, VSVG=village price of sugar per kilogram, VPBF=village price of beef per kilogram, and DVL=dummy for village (1 if 3 or 5).The figures in parentheses are standard errors.
5. There are at least three ways of configurating these indicators. Suppose there are two indicators A and B. The first approach is to take individuals that are at the intersection of A and B. That is, A _ B. The second approach is to join them horizontally so that the total selected is the union of those in A but not in B, those in B but not in A, and those at the intersection of A and B. That is, A _ B. The third approach is to take an overlap within a subset. For example, select B within A or vice versa. That is, B _ A. The three approaches differ in the extent and the type of the poor selected through these indicators.
Botswana CSO (Central Statistics Office) (1976). Rural Income Distribution Survey 1974/75. Gaborone, Botswana: Government Printer.
_____________ (1986). CSO Labour Force Survey 1984-85.
_________ (1988). Household Income and Expenditure Survey (HIES) 1985/86. Gaborone, Botswana: Government Printer.
_____________ (1991). Population Census, preliminary results.
Botswana MFDP (Ministry of Finance and Development Planning) (1991). National Development Plan (NDP7), 1991-97.
Colclough, C., and P. Fallon (1983). Rural Poverty in Botswana: Dimensions, Causes and Constraints: In D. Ghai and S. Radwan (eds.), Agrarian Policies and Rural Poverty in Africa. Geneva: ILO.
Foster, J., J. Greer, and E. Thorbecke (1984). A Class of Decomposable Poverty Measures. Econometrica 52: 761-765.
Kossoudji, S., and E. Mueller (1983). The Economic and Demographic Status of Female-Headed Households in Rural Botswana. Economic Development and Cultural Change 31 ( 4).
Lerman, R. and S. Yitzhaki. (1985). Income Inequality Effects of Income Source: A New Approach and Application to the United States. Review of Economics and Statistics 67(1).
Lipton, M. (1979). Employment and Labour Use: in Botswana, Final Report. Gaborone: Central Statistical Office, March.
Lucas, R. E. B., and O. Stark (1985). Motivations to Remit: Evidence from Botswana. Journal of Political Economy 93 (5).
Mueller, E., (1985). The Value and Allocation of Time in Rural Botswana: In C. Dov, R. E. B. Lucas. and E. Mueller (eds.), The Household Economy of Rural Botswana - An African Case, World Bank Working Paper 715.
Ravallion, M. and B. Bidani (1994). How Robust is a Poverty Profile? The World Bank Economic Review 8.
Sen, A. (1976). Poverty: An Ordinal Approach to Measurement. Econometrica 44.
Shalit, H. (1985). Calculating the Gini Index of Inequality for Individual Data. Oxford Bulletin of Economics and Statistics 47( 2).
Sukhtame, P.V. (1977). Malnutrition and Poverty 9th Lal Bahadur Shastri Memorial Lecture. New Delhi Indian Agricultural Research Institute.
Valentine, T. R. (1993). Drought, Transfer Entitlements, and Income Distribution: The Botswana Experience. World Development 21(1): 109-126.
Wasay, A., (1977). An Urban Poverty Line Estimate. Pakistan Development Review, 15 (1).
Watanabe, B. and E. Mueller (1984). A Poverty Profile for Rural Botswana. World Development, 12(2).