Chapter 5 DEA of Corporate Governance: The Model and its Application
5.4 Applying DEA
5.4.3 Test Trials of Data using DEA-Solver-LV
Test 1: Software Testing
Testing of the software and the data available verified that the program worked and that the variables chosen were correct. In the first test the data from business units A-F were used. This test had scores of all DMUs in each business unit averaged to represent a single score for each decision variable for each of the 6 business units tested. A summary of the results of this test from Appendix 3 is presented in Table 5.6 and shows that the test failed to discriminate because all DMUs were deemed to be equally efficient.
Table 5.6 Test 1 Summary for Business Units with Averaged Scores
DMUs 6 Business units A-F
Input Variables 8 Stakeid-soc reporting Output Variables 7 Corp imag-emplInvest
Result Test failed
Test 2: All Business Units, Average Scores and Reversed Variables
The second test was a trial to gauge whether the reversal of the input and output factors would provide any additional information. The only change to Test 1 was the reversal of input and output variables. A summary of the results from Appendix 4 shows that this reversal also failed to discriminate the factors under investigation.
Table 5.7 Test 2 Summary for Business Units with Averaged Scores but Reversed Input and Output Factors
DMUs 6 Business units A-F
Input Variables 7 Corp image-emplInvest Output variables 8 Stake Id-soc reportg
Result Test failed
Test 3: All DMUs in Cohort
The third attempt tried to elicit the factors that would be significant to the model, i.e. those factors that are determinants of efficiency. The segregation by business unit was abandoned and all DMUs (233) were presented for testing. Input factors were variables 17-39 in Table 5.3, and output factors were variables 9-16. The test failed to initiate.
gave a message to indicate the limit was breached and the test could not proceed. This is shown in Table 5.8.
Table 5.8 Test 3 for all DMUs
DMUs 233 All managerial units Input Variables 23 ComOpn-discitiz Output variables 8 StakeId-soc reportg Y-AF Result : DMUs≥ 50 (program failed)
Test 4: Reduced DMU Cohort
The fourth trial test overcame the 50 DMU limit by summarizing the data further and removing the DMUs which had the lowest level of employment service less than 5 years (variable 6 in Table 5.3), thus achieving a total DMU test sample of 48. The input factors of 23 variables and the 8 output variables were retained with the original intention of identifying the variables most relevant to the input and output factors. This test was able to run and computed a total of 1224 simplex iterations. However, the results were again inconclusive because all the DMUs displayed equal efficiencies of 1.00. These are summarized from Appendix 5 and shown in Table 5.9.
Table 5.9 Test 4 Sample of DMU Dataset to Overcome the 50 DMU limit
DMUs 48 Units A–F
Input Variables 23 ComOpn-discitiz Output Variables 8 StakeId-soc reportg Y-AF Result: All DMUs 100% efficient.
.
Tests 1-4 were all failures of the DEA application. Test failures require an analysis of the procedure, the data and the application. The Golany et al. (1989) procedure stipulates a careful selection of the variables in the first four steps: DMUs, input and output factors,, but an examination of the choices in the next step (5).
They suggest this evaluation of factors be done by subjective judgment, correlations and trial runs (see Section 3.7.1 validating the DEA Model). After four test failures this was the next step.
Test 5: One CSRMC Output Variable and Six Antecedent Variables
The analysis of the four trial-runs shown in tests 1-4 above and a reassessment of the data available suggests that the chosen DMUs are still appropriate but that input and output factors may have been mismatched. It was decided that the appropriate output factor should be the value expressed by Black’s aggregated score for the CSRMC variable (shown as variable 38 in Table 5.1). The input factors should be those variables seen as precursors to CSR and later discussed as the 6 antecedents. Those selected were the two communication, two ethics and the organisational support and justice variables (shown as numbers 18, 19, 21, 22, 28, 31 in Table 5.1) and identified by the notation (I). To overcome the limitation on the number of DMUs allowable under the DEA-Solver- LV program it was further decided that the DEA iterations would be conducted in two phases with two models as follows.
The whole cohort was to be tested but using the minimum, average and maximum scores for each decision variable of the DMUs in each business unit. This meant that every business unit (6) had three values, thus a total cohort of 18 DMUs were able to be run in the program (Appendix 6).
The whole cohort was to be tested iteratively by separate business units rather than the whole organisation. To ensure that the number of DMUs per business unit were acceptable at less than 50, units C and D which had 64 DMUs were trimmed to 49 and 48 by excluding DMUs which had a service history of less than five years (Appendix 7). The two tests above were conducted for the CCR-I and CCR-O models of DEA-Solver. Both models in both phases worked equally well in that input CCR-I or output CCR-O orientations made no difference. Phases one and two also revealed efficient and non- efficient DMUs. Test 5 was thus successful in showing the program worked and was able to discriminate between efficiencies of DMUs.
The summarized results of Phase 1 from Appendix 6 are shown in Table 5.10a and those for Phase 2 from Appendix 7 in Table 5.10b.
Table 5.10a Results of Test 5 for Minimum, Average and Maximum Scores in each Business Unit (from Appendix 6)
DMUs 18 Units A–F
Input Variables 6 Commacc-justice
Output Variables 1 CSRMC
Result: 4 DMUs 100% efficient.
Table 5.10b Results of Test 5 using DEA-Solver LV for Less than 50 DMUs per Business Unit (from Appendix 7)
DMU Cohort Number of DMUs Input Variables Output Variable Number of Efficient DMUs Unit A 6 6 CSRMC 3 Unit B 46 6 CSRMC 20 Unit C 49 (64) 6 CSRMC 9 Unit D 48 (64) 6 CSRMC 3 Unit E 41 6 CSRMC 16 Unit F 18 6 CSRMC 9 Total 208 Efficient 60
Aggregating the values of the business units allowed a rudimentary computation of the overall efficiency. A total of 60 DMUs of a cohort of 208 were efficient. This could only be verified by testing all DMUs across the organisation without segregating by business units, and therefore requiring the commercially available software for DEA-Solver-Pro. The DEA Professional Program available from SAITECH-INC.com was purchased.