Estimating South African industry production functions with measurement error and parameter heterogeneity

Production functions are estimated for South African industries using a novel industry panel dataset that combines education and employment data from a series of household surveys with output and physical capital data from the South African Reserve Bank Quarterly Bulletins. We are particularly interested in estimating the effect of education on worker productivity. The analysis starts by exploring the pooled OLS estimates, and then continues to consider estimators that exploit less restrictive identifying assumptions that allow for industry fixed effects, endogeneity in the factors of production, measurement error, cross-sectional dependence and parameter heterogeneity. The pooled OLS results indicate that the returns to education are substantial and concave. Although the effect of education disappears when industry fixed effects are included in the regression, the POLS results are restored when also allowing for measurement error or parameter heterogeneity and cross-sectional dependence. The paper briefly discusses why the fixed effects and first difference estimators will be particularly sensitive to certain types of biases.