The concept of a poverty function is introduced, modelling the shortfall of
household consumption from the poverty line as a function of reduced form determinants such
as human capital and land holdings. The model is estimated using a tobit and data from Uganda.
Parameters from the model are found to be similar to those from consumption functions,
indicating that the poor receive comparable rates of return on their assets to the non-poor.
Education of both men and women appears to raise the welfare of the poor as well as the nonpoor,
in both urban and rural areas.
In fact, the relevant models of household consumption might be better referred to as `welfare functions', because 1
— unlike orthodox consumption functions — their emphasis is not relations with income but rather on the reduced
form determinants of income. Even if poverty is better analyzed by welfare functions than by poverty functions, the
earlier point about the lack of multivariate analyses of poverty remains true. Glewwe (1991) provides one of the few
examples of such welfare functions.
1
Studies of poverty using household survey data from developing countries have typically
concentrated on measurement issues and on describing the characteristics of the poor (see
Ravallion, 1992, for a review of the issues and Boateng et al., 1992, for one example of a `poverty
profile'). This work is sometimes technically very sophisticated and is a necessary first step for
evaluating interventions targeted at the poor (see Anand, 1983, for an illustration). However, there
are surprisingly few examples in the literature of multivariate analysis of the determinants of
poverty; of estimating what could be termed `poverty functions'. Such poverty functions might give
more insight into what type of interventions could reduce poverty. They would also permit more
valid inferences about the causes of poverty than are provided by the simple — often bivariate —
decompositions of poverty indices presented in conventional poverty profiles.
A common aggregate poverty index is the `poverty gap', that is to say the mean negative
deviation of consumption from some minimum acceptable level, the poverty line. Consequently,
if one aimed to model household poverty, a natural dependent variable is the household poverty
gap: how far, if at all, household consumption falls below the poverty line. Explanatory variables
could include those factors which might affect welfare including household variables such as
demographics, assets and education together with community-level variables such as prices and local
infrastructure. Perhaps the main objection to estimating such poverty functions is that it would be
better to model household consumption per se on the grounds that censoring the distribution of
consumption above the poverty line wastes information. The case for estimating poverty functions
rather than consumption functions requires that the information about the consumption of the
non-poor is not useful and could be misleading: that the poor may behave differently from the nonpoor
. If the hypothesised determinants of welfare, such as human capital and physical assets, have 1
different returns for the poor and non-poor, then consumption functions may give misleading
results for the analysis of poverty interventions. For example, consumption functions may show
schooling to have large returns on average, but if these arise for only the non-poor, then it would
be quite inappropriate to advocate expansion of education as a part of a poverty alleviation package.
Thus the choice between consumption functions and poverty functions in part depends on whether
people are poor just because they have less assets (including human capital) or whether they also
receive lower returns on their assets.
This paper compares the results of consumption and poverty functions using data from a
recent national survey of Uganda. Section 1 explains the poverty measure adopted and the methods
of analysis used. Attention is given to the choice of the appropriate econometric technique for
estimating a poverty function, given the censored nature of the dependent variable (the household
poverty gap). Section 2 presents the empirical results. Two broad inter-related issues are raised. The
first is methodological: is anything gained from modelling poverty per se rather than welfare (proxied
by consumption) in general? The second is substantive: what are the determinants of poverty and
welfare in Uganda? Particular attention is focused on the role of human capital, in view of the
current emphasis by many international agencies upon expanding education as one means of
reducing poverty (for example, see World Bank, 1990). Section 3 concludes the paper.