DOES TECHNICAL EFFICIENCY CALCULATED BY SFA STATISTICALLY SIGNIFICANTLY VARIES IN DIFFERENT SAMPLES?

摘要

The aim of the paper is to analyse whether and how the results of Stochastic Frontier Analysis (SFA) changes if we use different samples of firms. We do not question the size of the sample, but the composition of it. SFA is frequently used method to calculate the technical efficiency of agricultural holdings and authors come often to different results. The results of SFA are sometimes used for policy-making and decision-making in agriculture and the results must be reliable. Agricultural holdings differ, and those differences can influence the results of SFA, i.e. the value of technical efficiency. We examined this issue empirically on a sample of 548 farms with 2268 observations for years 2013 to 2017. Accounting data were taken from Albertina database, data about acreage form LPIS. We chose randomly 5 samples per 500 observations (the number of farms differed as the panel was unbalanced). Technical efficiency for each sample was calculated by the same technique. Particularly True Fixed Effects model was used to model the Cobb-Douglas production function. Technical efficiency was calculated by Jondrow et al. (1982) method. Differences in medians of technical efficiency were tested by non-parametric Kruskal-Wallis test and Mood's median test. HO: there are no statistically significant differences in medians of technical efficiency was rejected in the first test but was not rejected in the other. We came to ambiguous results but nevertheless we can warn that the selection of the sample for calculation of technical efficiency might influence the results.