In order to test the factor structure more rigorously, the entire set of items are examined using confirmatory factor analysis (CFA). A confirmatory measurement denotes the relations of the observed variables to their posited underlying constructs that allow inter-correlating freely (Anderson and Gerbing, 1988). CFA provides an assessment of convergent and discriminant validity which aims at eliminating non-significant items (lower item-to-construct loadings). In the field of social sciences, the standardized factor loadings recommended minimum is usually 0.40 (Yli-Renko et al., 2001). A variety of CFA operations can be assessed by various literature; some concentrates on each of sole constructs (Daugherty et al., 2005; Ritter and Gemunden, 2004), another emphasize one exogenous (all-X model) and one endogenous (all-Y model) (Gold et al., 2001), and others highlight the holistic model (Calantone et al., 2002). In this study we adopt the method of isolated construct to estimate the factor loadings and fit indices of the measurement model.
Table 5.2 reveals the results of scale validation that factor loadings range from 0.63 to 0.95, and none of the indicators are non-significant (p<.05). As the significance of factor loadings indicates that the results of the measurement model performed very well; thus, the convergent validity is supported.
The measurement model is presented in an acceptable model fit and all of the fit-indices support the model which represents a reasonable fit to the data.
Table 5.2 Confirmatory factor analysis
Construct Dimension Standardized
Loading GFI NFI NNFI CFI
Social OKC
Structural knowledge capability 0.95 0.94 0.83 0.94
SKC-1 0.80
SKC-2 0.82(12.43)
SKC-3 0.63(9.71)
SKC-4 0.76(10.95)
Cultural knowledge capability 0.99 0.99 0.99 1.00
CKC-1 0.72
CKC-2 0.82(11.92)
CKC-3 0.85(12.20)
CKC-4 0.74(10.91)
Human knowledge capability 1.00 1.00 1.00 1.00
HKC-1 0.72
HKC-2 0.89(11.38)
HKC-3 0.78(11.32)
Technical OKC 0.98 0.98 0.95 0.98
TOKC-1 0.66
Knowledge sharing 0.86 0.95 0.93 0.95
KS-1 0.80
Organizational effectiveness 0.91 0.96 0.95 0.96
OE-1 0.87
t values are shown in parentheses.
Discriminant validity assesses a series of chi-square differences by comparing a model with the others, combining two explanatory constructs with the unconstrained model. We ensure that discriminant validity follows Calantone et al. (2002), for example, by combining human
knowledge capability and cultural knowledge capability into one construct which generates the result that the chi-square is 643.73 and degrees of freedom are 203. The chi-square difference with the original model is significant (∆χ2=109.94, df=4, p<0.1), which supports that these measures are distinct. This study examines each set of constructs in the measurement model and discriminant validity is supported in all test cases.
5.2.2 The measurement of second-order factor
In the framework a second-order factor measurement is developed. Social organizational knowledge capability is modelled as a second-order construct that is measured by the three first-order indicators including structural knowledge capability, cultural knowledge capability, and human knowledge capability. For validating the second-order model exists, five alternative models were examined in the confirmatory factor analysis (Xia and Lee, 2005) and the patterns are shown in Figure 5.1.
Item 1
(b) Model 2: one first-order factor
…
(c) Model 3: three uncorrelated first-order factors
… …
(d) Model 4: three correlated first-order factors
… …
(e) Model 5: a second-order factors
… …
Item 1
……
Social OKC
Figure 5.1 Alternative models tested in confirmatory analysis
Model 1 hypothesizes that each factor is measured by one item, which form eleven first-order factors to examine the model fitness. One first-first-order factor, a unidimensional construct, is a plausible model of underlying data structure which explains all the common variance among the 11 items assumed in Model 2. Model 3 supposes that 11 items form into three uncorrelated first-order factors that concerns reflect three dimensions: structural knowledge
capability, cultural knowledge capability, and human knowledge capability. Model 4 hypothesizes that three first-order factors account for the covariance among the 11 items and the correlations among these factors. Model 5 hypothesizes that the 11 items form into three first-order factors and that these three first-order factors are measured by a second-order factor, social OKC.
Testing with SEM software (LISREL), the results of alternative models are shown in Table 5.3. To assess the adequacy of the proposed model, a few fit indices can be observed through statistic analysis. For example, the ratio of chi-square (χ2) over the degree of freedom (df), normed chi-square, is used to measure the fitness of research model. Normally, the smaller value normed chi-square, the better model fit. An acceptable value of normed chi-square ratio is lower than 3 (Ritter and Gemunden, 2004), even lower than 5 (Xia and Lee, 2005). The results show that Model 1, 2, and 3 were not reasonable because most of their fit indices did not reach to the threshold criteria. The results of both Models 4 (normed chi-square = 2.84, GFI = 0.92, NFI = 0.97, and CFI = 0.98) and Model 5 (normed chi-square = 2.84, GFI = 0.92, NFI = 0.97, and CFI = 0.98) were acceptable because all fit indices meet the criteria.
Table 5.3 Results of second-order model
Model
No. Construct Dimension χ2 d.f χ2/d.f (<5.0*)
GFI (>0.9)
NFI (>0.9)
NNFI (>0.9)
CFI (>0.9) Model 1★ Null 1588.12 44 36.10 0.82 0.92 0.91 0.82 Model 2★ Social OKC (First order) 282.19 44 6.41 0.82 0.92 0.91 0.82 Model 3★ Three uncorrelated first order 427.64 44 9.72 0.77 0.88 0.86 0.89 Model 4 Three correlated first order 116.32 41 2.84 0.92 0.97 0.97 0.98 Model 5 Social OKC (Second order) 116.32 41 2.84 0.92 0.97 0.97 0.98
★ The model is not adapted.
* The threshold criteria based on the survey of Xia and Lee (2005).
Therefore, both models, 4 and 5, are adopted to examine the relationship between social OKC and knowledge sharing. Figure 5.2 presents the values of the parameter estimations of Model 4 (three correlated first-order factors). The model exhibits the data quite well because the factor loadings range from 0.64 to 0.86 and the correlations of constructs range from 0.75 to 0.77.
Then, the result of second-order factor model is shown in Figure 5.3. The results confirm that a second-order factor of social organizational knowledge capability exists. All first-order factors loaded on the second-order factor range from 0.86 to 0.88 and are strongly significant at p <.01.
SKC-1
Figure 5.2 Parameter estimations of three correlated first-order model (Model 4)
SKC
Figure 5.3 Parameter estimations of second-order model (Model 5)