Analysing Firm-Level Labour Productivity Using Survey Data

Abstract

This paper investigates the determinants of firm-level labour productivity in the manufacturing sector using GAPS data. These data are from a stratified survey, where the strata are based on industry and firm size. The paper focuses on whether weights should be applied in the regression analysis. Augmented Cobb-Douglas production functions are estimated, where a set of dummies are used as proxies for firm-level knowledge stocks. The regression results show that there are significant differences between the parameters estimated by weighted least squares (WLS) and OLS, particularly for the variables union density and training expenditure. These differences can be caused by parameter heterogeneity (across strata); in theoretical terms this means that applying the same production function across all firms is not appropriate. Given this parameter heterogeneity, both the OLS and WLS methods do not estimate parameters of interest. Instead, there is a requirement to estimate sub-sample regressions. These are presented in the second part of the empirical results.