Each year, in an attempt to alleviate poverty, government invests large parts of the budget
to provide infrastructure to poor households in South Africa. This not only necessitates an
understanding of the effectiveness of government’s infrastructure delivery rate to address
poverty in South Africa, but also raises important questions on how the poor can be
identified. In recent years, countries have moved away from traditional broad poverty
measures such as gross national income (GNI) per capita and Human Development Index
(HDI). Information on poverty and other household information are more often collected
through household surveys. From these surveys, monetary and non-monetary poverty
measures can be used to identify the poor. By making use of a monetary poverty measure
such as expenditure, per capita household expenditure can be calculated. Households are
divided into quintiles based on their per capita household expenditure, and the bottom 20
and 40 per cent are usually the benchmark for households to be identified as being poor.
This is analysed in terms of the poor’s access to services and other household
characteristics.
Qualitative regression models have gained more recognition in econometrics, especially
in the social sciences field. Information collected from household surveys is often
qualitative, or binary in nature. Due to the non-linear nature of binary-dependent variable
models, logit and probit models were appropriate for this study. The maximum likelihood
method, within the binary choice framework, was employed to determine the extent to
which infrastructure delivery and other household characteristics have an impact on
poverty. The results provided empirical evidence that infrastructure investment can
significantly reduce the likelihood that a household will be poor, given a set of
characteristics.