SOFTWARE FOR EVALUATING CONFIDENCE INTERVALS IN DEA PROBLEMS

CIDEA is a program designed to run on unix machines and which allows the user to solve Data Envelopment Analysis (DEA) problems and assign confidence intervals to the estimated efficiencies.

Both constant and variable returns to scale DEA models may be solved following the CCR method (Charnes et al., 1978) and the BCC method (Banker et al., 1984) respectively. Output orientation is assumed. The confidence intervals are estimated using the first algorithm of Simar and Wilson (2004), imposing a fixed bandwidth (which may be chosen by the user).

The package is designed to run on unix machines supported by NAG routines. It is delivered in a zip file, cidea.zip, which comprises two files CCR and BCC. These two files are then used to solve CCR and BCC problems respectively. Once you have downloaded the zip file onto your PC, unzip it and ftp its contents over to a unix machine. You will then need to set permissions by typing the following lines at the unix prompt:

chmod u+rwx CCR
chmod g+rwx CCR
chmod o+rwx CCR
chmod u+rwx BCC
chmod g+rwx BCC
chmod o+rwx BCC

Data to be used in the exercise must be stored in a file named tlt.txt, and it must be set out according to the following format. Each row of the data file represents a separate decision making unit. Inputs are listed first, followed by outputs. Data must be arranged in F10.4 format. The total number of inputs and outputs must not exceed 10, and a maximum of 5000 decision-making units is permitted.

Upon execution, the program prompts for the number of decision making units, the number of inputs, the number of outputs, desired bandwidth, and then prompts to see if the user wishes confidence intervals to be estimated. Note that the program runs much faster if CIs are not required. Output is then sent to a file named dea.out.

This method of computing confidence intervals does not guarantee that the point estimate of the efficiency lies within the estimated confidence interval. Neither does it guarantee that the bounds of the estimated confidence interval lie within the unit interval. The program does not correct for what Simar and Wilson refer to as bias (since Simar and Wilson counsel against doing so) and in consequence the lower bound of the confidence interval for efficient decision-making units should always be unity.

You are free to download and use this software without charge. However please note that it comes without warranty or support of any kind beyond this document; you use it at your own risk. In addition, it is requested that if you use the software to generate published results, you acknowledge this using the following citation:

Johnes, Geraint (2004) CIDEA: software for evaluating confidence intervals in DEA problems, mimeo, Lancaster University Management School, available from http://www.lancs.ac.uk/people/ecagj/cidea.html.

If you wish to access the files of FORTRAN source code, they are available here for the CCR and BCC programs respectively. You may wish to do this if, for example, you wish to use the program for problems involving more than 5000 decision-making units; this is possible if you amend the dimensioning statements. You may use this code in part or whole in any software that you yourself publish, but if you do so you must: (i) acknowledge me (Geraint Johnes) as co-author of the new software; (ii) inform me; and (iii) supply me with a copy of the new software gratis.

References:

Banker, Rajiv D., Charnes, Abraham and Cooper, William W. (1984) Models for estimationof technical and scale inefficiencies in data envelopment analysis, Management Science, 30, 1078-1092.

Charnes, Abraham, Cooper, William W. and Rhodes, Edwardo (1978) Measuring the efficiency of decision making units, European Journal of Operational Research, 2, 429-444.

Simar, Léopold and Wilson, Paul (2004) Performance of the bootstrap for DEA estimation and iterating the principle, in Cooper, William W., Seiford, Lawrence M. and Zhu. Joe (eds) Handbook on Data Envelopment Analysis, Boston: Kluwer.