The Impact of Multiple Imputation of Coarsened Data on Estimates on the Working Poor in South Africa

South African household surveys typically contain coarsened earnings data, which consist of a mixture of missing earnings values, point responses and interval-censored responses. This paper uses sequential regression multivariate imputation to impute missing and interval-censored values in the 2000 and 2006 Labour Force Surveys, and compares poverty estimates obtained under several different methods of reconciling coarsened earnings data. Estimates of poverty amongst the employed are found not to be sensitive to the use of the multiple imputation approach, but are sensitive to the treatment of workers reporting zero earnings. Multiple imputing earnings for all workers with missing, interval-censored or reported zero earnings, the proportion of workers earning less than R500 per month falls by almost a third between 2000 and 2006.