217 /302
217
臺大管理論叢
第
28
卷第
2
期
Table 3 Profile of The Respondents (n = 30)
Respondent type
Respondent descriptions
Frequency Percentage (%)
Respondent’s working area
Financial holding company head
2
6.67
Bank
9
30.00
Insurance
7
23.33
Securities
9
30.00
Securities investment trust
3
10.00
Experience years
1~5
5
16.67
6~10
14
46.67
11~15
7
23.33
16~20
3
10.00
≥ 21
1
3.33
According to the interview results and the 21 criteria based on the RM-BSC
framework for risk management in the financial holding industry (Tseng et al., 2011), the
questionnaire was designed to assess the relative importance among the criteria. The
questionnaires were distributed to experts employed in the risk management departments
of FHCs in Taiwan. Afterward, the DANP was used to evaluate the strength of influence
and self-feedback of the four dimensions and 21 criteria in the RM-BSC (Figure 1). To
examine the consistency and agreement among the respondents, the standard deviations of
the input direct-influence matrices of the dimensions and criteria were calculated.
However, the standard deviation is meaningful only when the context of the mean data
(i.e., the average) is well understood beforehand. Alternatively, the coefficient of variation,
denoted by
C
v
, is dimensionless, and is comparable when the means differ among
populations.
C
v
is assumed to be positive, and a lower value is more desirable.
C
v
is
considered large when it is greater than unity.
McKay’s approximation (1932) and Forkman’s approximation (2009) are widely
used for statistically testing the hypothesis
H
0
:
C
v
≥
C
0
;
H
1
:
C
v
<
C
0
. In this study, we
analyzed the case of
C
0
= 1 to examine whether the data was sufficiently smaller than
unity. Given the independent observations
n
and sample coefficient of variation
ĉ
,
McKay's approximation was written as Equation (1), which is approximately a chi-square
distributed with
n
-1 degrees of freedom. Let
n
i
be the number of observations of sample
i
,
with a coefficient of variation
c
i
. Forkman's approximation was calculated as in Equation
(2), which is approximately a chi-square distributed with
degrees of freedom in
the presence of
k
samples. To provide a more accurate examination, Forkman's
approximation was applied in this study.