219 /302
219
臺大管理論叢
第
28
卷第
2
期
Step 1: Calculate the direct-influence matrix according to the scores, and test their
reliability.
Step 2: Normalize the direct-influence matrix.
Step 3: Attain a total-influential matrix.
Step 4: Analyze the DEMATEL results. Identify thresholds to highlight the key
influence in the total-influence matrix, and calculate the prominences and
relationships to devise a relationship diagram for identifying the criteria in
the core, driving, independent, and influenced factor zones.
Step 5: Find the normalized total-influential matrix.
Step 6: Find the normalized matrix by using the dimensions and criteria.
Step 7: Build an unweighted supermatrix.
Step 8: Find the influential weights of the DANP.
Step 9: Obtain the global weights of the criteria by using the limit supermatrix of the
DANP.
4. Results and Analysis
4.1 DEMATEL and DEMATEL-Based Analytic Network Process Analysis
The questionnaire was conducted in accordance with the information listed in Table 4
to obtain the average input direct-influence matrices of the dimensions and criteria (Tables
5 and 6, respectively). The standard deviations of the input direct-influence matrices of the
dimensions and criteria were calculated as shown in Tables 7 and 8. Tables 9 and 10 list
the coefficients of variation of the input direct-influence matrices of the dimensions and
criteria, and only the coefficient of variation of C3–D3 was found to be greater than unity.
Using Forkman's approximation (Forkman, 2009) to conduct the chi-square test, of the
420 entries, only eight coefficients of variation of B3–D3, B4–D3, C3–D3, C6–D3, C8–
C1, C8–D3, D3–C3, and D3–C8 were neither sufficiently smaller than unity at the .05
significance level nor significantly larger than unity. In other words, the input data were
consistent among the respondents, and thus, we could perform the DEMATEL procedure
and ANP.