Skip to content

Combining census and survey data to construct a poverty map of South Africa

Literature › Works

Harold Alderman; Miriam Babita; Jean Lanjouw; Peter Lanjouw; Nthabiseng Makhatha; Amina Mohamed; Berk Özler; Olivia Qaba (2000). Combining census and survey data to construct a poverty map of South Africa. Chapter 2 of an edited volume (volume title not shown in extract).
Type book chapter
Scope Post-1994 South African economy
Method econometric
Period studied 1995–1996
Regions South Africa (national), nine provinces, 45 district councils, 354 magisterial districts, former homelands
Themes inequality poverty unemployment, methods data sources, bantustans removals influx, demography urbanisation, transition post apartheid
Citation key alderman2000

Argument

Income reported in the 1996 South African census is a poor and systematically biased guide to household welfare. It understates income relative to the 1995 Income and Expenditure Survey (IES), overstates poverty, and does so most in rural areas and former homelands. The authors instead regress log per capita expenditure in the IES/October Household Survey on household characteristics that also appear in the census (composition, education, occupation, housing, services) and apply the estimated coefficients to every census household. The imputed expenditure tracks IES means and poverty rates much more closely than census income does, and the gap no longer varies systematically with area of residence. This allows expected headcount poverty rates, with standard errors, to be produced for provinces, district councils and magisterial districts, which is useful for allocating the equitable shares grant to municipalities. The authors list the main assumptions to be tested next: normality, homoskedasticity and no small-area effects.

Evidence and method

1995 October Household Survey (29,700 households) merged with the 1995 Income and Expenditure Survey (28,585 households); Census '96 (over nine million households, income recorded in 14 brackets plus remittances); South African Reserve Bank private consumption aggregate; provincial, district council and magisterial district aggregates; methodology from Hentschel et al. and Elbers et al. (Ecuador).

Findings

  • Mean household income in Census '96 is well below IES income and expenditure. (p. 6)

    Both these estimates exceed the monthly income including remittances from the census income data. That average is R2 454.

  • Census income yields much higher poverty rates than the IES at the R800 household line. (p. 7)

    That is, the estimated poverty rate is over 80% higher in the census than the IES data.

  • Census income and IES expenditure correlate strongly at province level but rank provinces differently. (p. 7)

    the ordering in terms of income differ, hence the Spearman rank correlation coefficient is only 0,68

  • Census income understates welfare relative to the survey module, and the bias is not random. (p. 13)

    the income data collected in the census significantly understates the income or expenditure levels of the households measured by a detailed module in a household survey in South Africa.

  • The census-IES gap is larger in rural areas and former homelands than in urban areas, so census income overstates poverty in some areas. (p. 13)

    For households which live in areas classified as rural portions of former homelands or other rural areas, this gap is larger

  • Imputed census expenditure is close to IES mean expenditure. (p. 16)

    The average household expenditure from this imputation is R2 789 per month. This is only 6,4% below that in the IES.

  • Imputed expected poverty at R800 per household per month matches the IES national headcount. (p. 16)

    we find an overall expected poverty incidence of 28,5% for South Africa, a figure which is virtually identical to the corresponding headcount index value(28,4%)from the IES.

  • Provincial correlations between imputed census expenditure and IES expenditure are far higher than for census income. (p. 16)

    The correlation coefficient between the provincial averages of census imputed expenditures and that from the IES expenditure is 0,97, and the Spearman rank correlation coefficient is 0,93 (Table 2).

  • Unlike census income, the imputed measure shows no systematic bias by area type. (p. 16)

    unlike the average income and poverty estimates based on the census data there is no systematic pattern in the difference between the imputed expenditures and the IES data.

  • Policy motivation: the municipal equitable shares grant depends on counts of households below R800 per month, which are hard to observe. (p. 2)

    the grant is based, in part, on the number of households within the jurisdiction which have an income of less that R800 per month.

  • Model residuals are mostly heteroskedastic, so the initial homoskedasticity assumption does not hold and needs correcting. (p. 19)

    We also assumed initially that our residuals are homoskedastic. Further tests of this assumption showed us that in most of our nine regressions, the residuals are in fact heteroskedastic.