Dealing with earnings bracket responses in household surveys - How sharp are midpoint imputations?

Earnings functions form the basis of numerous labour market analyses. Non-response (particularly among higher earners) may, however, lead to the exclusion of a significant proportion of South Africa's earnings base. Earnings brackets built into surveys intend to maintain response rates. Econometric tools to incorporate brackets vary from "simplistic" imputation to interval regressions. Coefficient differences are investigated here to establish reliable remedies. Monte-Carlo simulations suggest that "simple" methods fail only under extreme skewness and when a substantial number of right-censored observations appear in the sample. Testing procedures applied to LFS data reveal that in practice coefficients are virtually invariant to the proposed methods.