Making noisy data sing : a micro approach to measuring industrial efficiency

Technical, scale and allocative inefficiency are widely believed to plague the industrial sectors of developing countries. This paper presents a way to measure this inefficiency with imperfect data. There is great interest in documenting the patterns and magnitudes of inefficiency, so that appropriate corrective policies can be designed. This paper presents a new approach to analyzing plant efficiency that recognizes and deals with such data imperfections as measurement error, missing observations and selectivity bias. The author has developed full-information maximum-likelihood (FIML) estimators of production technologies that deal with missing data and measurement errors, making alternative assumptions about the missing data patterns and the timing of employment and decisions. These estimators yield indices of the returns to scale, means square deviation from the efficient frontier and - when labor is treated as endogenous - mean square deviation from efficient factor mixes. To gauge the performance of the alternative estimators, the author applies them to census data on Chilean industry, and compares the results with naive estimators that do not recognize data imperfections.