Creating a 2000 IES-LFS Database in Stata

PROVIDE Project Technical Paper 2005:1

SAM one is more concerned about working with a database that reports the correct
expenditure pattern rather than the correct absolute expenditure levels, since external
control totals from the National Accounts (South African Reserve Bank) are used to scale
expenditure up or down in any event. However, working with the correct expenditure
levels is important when analysing poverty, since results (poverty lines and poverty
measures) have to be realistic.

• Van der Berg et al. (2003a) find that there is evidence of “sloppy work in the both
gathering and the management of data”. Some of this ‘sloppiness’ is dealt with when
expenditure totals are recalculated and expenditure items are mapped to new expenditure
categories for use in the SAM. Some obvious inconsistencies are also dealt with in various
ways where possible, although it is impossible to correct all errors resulting from
miscoding, misrepresentation by respondents and misinterpretation of questions. Section 3
contains more detail in this regard.

• Van der Berg et al. (2003a) estimate that about 25% of the records are unusable for many
purposes, mainly because total expenditure (including savings) and total income differ by
more than 30%.¹⁷ For the purpose of the SAM it is assumed that the larger of income or
expenditure is the correct welfare measure, and all income or expenditure items are scaled
up accordingly without changing the relative income or expenditure patterns of the
households. Van der Berg et al. (2003a) are also concerned about the food expenditure
reporting. In about 350 observations food expenditure was reported as zero (see section
3.2.7, which explains how food expenditure was imputed for households reporting zero
food expenditure).¹⁸

While these problems are disheartening and are cause for concern about the usefulness of the
dataset as a whole, the IES 2000 dataset remains, as pointed out by Van der Berg et al.
(2003), the most recent available data and one should attempt to work with it. In sections
2.4.2 and 2.4.3 some further research is conducted, first into aggregate income and
unspecified income (down “sharply”). He finds that the share of income attributable to the household as a
whole drops from 15.7% in 1995 to 9.9% in 2000, “suggesting worse income component measurement in
2000 than in 1995” (2003: 5).

¹⁶ This also does not say that the 1995 levels were correct. Some believe that 1995 had some inconsistencies of
its own. A recent study by Hoogeveen and Ozler (2004) does in fact compare the two surveys after
making various adjustments.

¹⁷ It is unclear why 30% or more is regarded by the authors as too large a difference.

¹⁸ The questionnaire asks respondents to give the actual value of expenditure or goods received in kind during
the last 30 days. It does not ask the value of consumption. This may be an explanation for the occurrence
of zero food expenditure.

17

© PROVIDE Project

<<   19   20   21   22   23   24   25   >>

More intriguing information