WORKING PAPER NO. 98/3
COMPARING TECHNIQUES FOR MEASURING EFFICIENCY AND PRODUCTIVITY OF AUSTRALIAN PRIVATE HOSPITALS
(Richard Webster, Steven Kennedy and Leanne Johnson)
The Australian Bureau of Statistics (ABS) is trying to improve its measurement of inputs, outputs and productivity for the non-market sector and, more generally, for the services sector of the Australian economy. In 1996-97, this project focussed on the health services industry. The analysis reported in this paper applies a range of firm-level efficiency-measurement techniques to a unit record dataset for the Australian private hospital industry. Firm-level analyses of this kind are being applied by influential members of the ABS user community. This private hospitals study has three aims:
- to explore the differences in assumptions made by the various techniques and the differences in results they yield;
- to test the assumptions (relating to homogeneity of the industry, economies of scale, etc.) that underlie ABS standard methods for analysing aggregate productivity; and
- to understand the ways in which the characteristics of a dataset can affect the application of these analytical techniques.
Two types of techniques are used in the analyses: a non-parametric technique known as Data Envelopment Analysis (DEA), and two parametric technique - Stochastic Frontier Analysis (SFA) and Ordinary Least Squares (OLS) regression. The benefits and shortcomings of each technique are discussed in general terms, then each is applied to a number of model specifications using different combinations of input and output variables drawn from the private hospitals dataset.
In this analysis the DEA technique is not robust to changes in the number or construction of variables. Conclusions about the relative efficiency of sub-samples and the efficiency ranking of individual hospitals change appreciably when the choice of variables is altered. Thus, if DEA is to be used for monitoring the performance of individual firms or for assessing patterns of efficiency across the whole population of firms, extrinsic judgements must be brought to bear when selecting the input and output variables.
Results from the parametric estimation techniques (OLS and SFA) also suggest a lack of robustness to changes in model specification. Conclusions about the structure of production, the pattern of productivity and the performance of individual hospitals can all change when the model is altered. Analyses of sub-populations (characterised by hospitals' size or profit-making status) indicate that individual hospitals may be engaging in substantially different activities from one another. This brings into question the validity of an aggregate productivity analysis of the kind traditionally applied by the ABS.
The analysis also highlights the inability of the dataset and our models in combination to completely characterise the private hospitals industry. In part, this is due to shortcomings of the frontier estimation techniques. However, it also suggests minor changes to the private hospitals census which could enhance the value to analysts who are interested in developing measures of unit level hospital efficiency.
An earlier version of this paper was presented to the ABS' Methodology Advisory Committee where Annette Dobson was the discussant. The authors would also like to thank Tim Coelli, Kathy Kang, Marelle Rawson, John Goss, Ken Tallis, Ben Phillips and Keith Carter for helpful comments and assistance with this research project. Responsibility for any mistakes or omissions is entirely our own.
This page last updated 10 August 2004