The Measurement Model

In the data analysis, the items related to their constructs in the final model were used in the confirmatory factor analysis using the statistical software package AMOS version 21. In a confirmatory factor analysis, the measurement model for each latent construct is created. A measurement model specifies the relations between the observed measures (question items in this thesis’ context) to their proposed underlying constructs (Anderson & Gerbing, 1988). The maximum likelihood (ML) method was chosen to estimate the difference between the observed and estimated covariance matrices as it is the most common procedure where a sample size is greater than 150 (Anderson & Gerbing, 1988; Hair, Anderson, et al., 2010).

The measurement model in this data analysis was evaluated by examining the factor loadings/regression weights of each item for statistical significance. As discussed in Section 5.7, the factor loadings should be at least 0.50 and above for adequate individual item reliability (Bagozzi & Yi, 1988). Items were dropped from consideration if their factor loadings were below the recommended level of 0.50. Table 16, shows the items that have been removed due to lower factor loadings.

Table 16: Items Dropped From the Model

Item

Factor loading

Items’ description

EC07

0.335

employees always share their experiences with colleagues from other departments.

TR05

0.375

firm will always try to treat me fairly.

Item EC07 is about sharing experience from other departments, whereas TR05 is related to trust in management. Both items have factor loadings lower than 0.50. In this thesis, the cut- off values of factor loading in CFA are equal to or greater than 0.50, which can improve the factor validity and the measurement model (Bagozzi & Yi, 1988). Several fit statistics were

141 | P a g e

employed to variably interpret the data, for details see Appendix H. The next section describes the fit statistics used in this thesis to represent the measurement model.

5.9.1 Parcelling of Items

Initially Kenny (1979) was accredited with an approach in which items are aggregated to provide a single indicator of a latent variable. This approach is known as item parcelling. Baggozzi and Edward (1998) suggest that item parcelling leads to fewer indicators and provides a better measurement model fit.

There are certain benefits and some disadvantages of using the item parcelling technique. The parcels may be more normally distributed as compared to the individual items (Hall, Snell, & Singer Foust, 1999). Further, the item parcelling technique is useful in small samples with comparatively lesser model parameters (Bagozzi & Edwards, 1998). Item parcelling can produce more reliable results and a better model fit (Kishton & Widaman, 1994). If the purpose of the parcelling is to improve model fit, then the research should not parcel it. However, if the concern is to parcel those items, measuring the same construct then the parcelling technique strengthens the results (Little, Cunningham, Shahar, & Widaman, 2002).

There are also some disadvantages of using the item parcelling technique. As item parcelling leads to fewer indicators, the SEM test, based on items parcelling, may not as rigorous a test as compared to the individual items (Bandalos, 2002). Item parcelling may lead to biased estimates of other parameters of the model (Hall, Snell, & Singer Foust, 1999). Two or more items can be parceled to improve the model fit in confirmatory factor analysis (Bandalos, 2002). Different parcelling strategies can be used to improve the model fit in the Structural Equation Modeling technique, including aggregating random or similar items (Hall, Snell, & Foust, 1999).

142 | P a g e

This thesis has used item parcelling based on aggregating two items in one parcel having similar meanings. EC12 was computed by adding EC8 and EC11; RR13 is computed by adding RR02 and RR03; and KS16 was computed by adding KS05 and KS06. Similarly, TR11 was computed by adding TR02 and TR06, whereas, IC08 was computed by adding IC01 and IC03. Details are shown in Table 17:

Table 17: Item Parcelling

Items and their Descriptions Parcel Item Parcel Measure

EC08: My organisation supports cross-functional team work for learning through collaboration. EC11: Our employees interact and exchange ideas with people from different areas of the company.

EC12 Learning in collaboration and Cross-functional

RR02: I am satisfied with the monetary rewards that I receive in exchange for the knowledge I give the organisation.

RR03: My feelings about the monetary rewards I receive for sharing knowledge are excellent.

RR13 Monetary rewards for sharing knowledge are good in my organisation

TR02: I can trust the people I work with to lend me a hand if needed.

TR06: I can trust the people in other departments to lend me a hand if needed.

TR11 I can trust the people in my organisation to lend me a hand if needed.

KS05: People in my organisation frequently collect knowledge of know-where or know-whom from other organisational members.

KS06: People in my organisation frequently share knowledge of know-where or know-whom with other organisational members.

KS16 People in my organisation collect and share knowledge of know-where and know-whom from other organisational members

IC01: I often develop new products and services that are well received by the market.

IC03: I often develop novel skills for transforming old products into new ones for the market.

IC08 I often develop new skills to develop new products for market

143 | P a g e

5.9.2 Goodness of Fit Indices

This thesis reports a number of goodness of fit indices for testing of measurement model and structural equation models (SEM). The common measures use the ratio of χ2 (statistics to the degree of freedom (DF), comparative fit index (CFI), goodness-of-fit index (GFI), adjusted goodness-of-fit index (AGFI), normed fit index (NFI) and root mean square error of approximation (RMSEA) (Hu & Bentler, 1999; Lin & Lee, 2005; Segars & Grover, 1998). In addition to these indicators, the standard root mean square residual (SRMR) is also recommended for assessing the goodness of fit model (Bryne, 2010; Raykov & Marcoulides, 2006).

As shown in Table normed χ2 (the ratio between χ2 and the degrees of freedom), which was used to assess model fit, was 4.00 at p<0.001. The recommended value is ≤3 for good model fit and 5.00 for acceptable fit (Ryu, Ho, & Han, 2003). Other fit indices also showed good model fit to the data set including goodness-of-fit index (GFI) and normed fit index (NFI) are both 0.80, and equal to the recommended cut-off level of 0.80. The root mean square error of approximation (RMSEA) was 0.80, which was below the cut-off level of 0.10 (Ryu, et al, 2003). Hence, the model showed an acceptable fit according to the data set, as shown in Table 18. Detailed results are shown in the Appendix H.

Table 18: Measurement Model Fit

Goodness-of-fit measures χ2 GFI NFI RMSEA RMR

Test statistics/df

Recommended values ≤ 5.00* ≥0.80* ≥0.80* ≤ 0.10* ≤0.08* CFA model 4.0 0.80 0.80 0.80 0.07

144 | P a g e