40 / 414
服務主導邏輯之共同生產：前置因素與結果因素
40
4. Results
4.1 Structural Equation Modeling
The hypothesized relationships in the model were tested simultaneously through SEM.
The standardized path coefficients of the structural model as estimated by LISREL 8.52 are
given in Table 2. The fit of the model was acceptable (chi-square (219) = 1054.79,
p
= 0.00;
GFI = 0.86; CFI = 0.93; PNFI = 0.80; NNFI = 0.91; RMSEA = 0.08; RMR = 0.08). The chi
square divided by the degree of freedom value was higher than the criterion of 3, and GFI
statistic was below the 0.9 threshold of acceptability. However, other fit indices that adjust
for model complexity and sample size need to be considered. In the above model, NNFI and
CFI were greater than the suggested 0.9. These two fit indices indicate a good fit of SEM
model, especially for a model with such a large number of constructs (Hoyle and Panter,
1995; Morgan and Hunt, 1994; Taylor and Todd, 1995). Actually, the aforementioned model
was actually purified by analyzing the modification indices provided by LISREL software.
By correlating some of the indicator error variances (X1 and X6, X12, and X23) and adding
paths from confidence benefits to social benefits, these changes resulted in a GFI of 0.92 and
a lower chi-square value. However, these additional improvements suggested by the
modification indices were not implemented because of a lack of theoretical support. The
squared multiple correlations (SMCs) for endogenous variables reveal as follows:
co-production = 0.31; special treatment benefits = 0.17; social benefits = 0.28; confidence
benefits = 0.23; share of wallet = 0.37.
4.2 Hypotheses Testing
Hypothesis 1 stated that asset specificity would have a positive effect on co-production.
The effect of asset specificity on co-production was significantly positive (
γ
= 0.279,
p
<
0.05), and therefore, H1 was supported. Hypothesis 2 stated that quality of customer
interaction would have a positive effect on co-production. The effect of quality of customer
interaction on co-production was significantly positive (
γ
= 0.172,
p
< 0.05), and therefore,
H2 was supported. Hypothesis 3 stated that decision-making uncertainty would have a
positive effect on co-production. The effect of decision-making uncertainty on co-production
was significantly positive (
γ
= 0.231,
p
< 0.05), and therefore, H3 was supported.