WP21/2009

The comparability of Census 1996, Census 2001 and Community Survey 2007

Stellenbosch Working Paper Series No. WP21/2009

Author(s): Derek Yu

JEL Classification: J00

Keywords: South Africa, Household survey, poverty, inequality, missing data, imputation

Download:  PDF (1.3 MB)


Abstract:

Census 1996 and Census 2001 are the only all-inclusive censuses conducted by Statistics South Africa (Stats SA) under the new democratic dispensation, providing information on demographics, educational attainment, migration status, labour market status, economic activities, income, housing, and access to household goods and services. However, when the cabinet took a decision that a census would not be conducted in 2006, a gap in information between Census 2001 and the next Census, which is scheduled to take place in 2011, was created. Later, a decision was taken to conduct the Community Survey (CS) in 2007, which was designed to provide information similar to the two censuses. The aim of this paper is to look at the trends in demographics, educational attainment, labour market status, income and non-income welfare (e.g., housing, access to household goods and services) across the three surveys, using the 10% samples from the 1996 and 2001 censuses as well as the data from the Community Survey 2007. With regard to changes in income welfare, the household income variable was derived differently in each survey. Besides, all three surveys had high proportion of households with zero or unspecified household income, and excluding these households from poverty and inequality analyses would lead to biased results. Hence, sequential regression multiple imputation (SRMI) was applied at both person and household levels to impute values for the households with zero and unspecified income, before the imputed household income values were used to derive per capita income for analyzing poverty and inequality trends across the three surveys. Finally, income welfare and non-income welfare were compared by dividing households into per capita income quintiles.