• Tidak ada hasil yang ditemukan

CFA Results for Individual Constructs

Dalam dokumen PDF repository.nida.ac.th (Halaman 185-197)

DATA PREPARATION AND MEASUREMENT MODEL

5.6 CFA Results for Individual Constructs

The results of the CFAs of the individual constructs (business strategy, TM, satisfaction, commitment, POS, employee performance, and intention to quit) are discussed and presented in this section. The discussion here centres on the direction of the factor loadings (parameters), the fit indices, the critical ratios (C.R.), the significance of the path coefficients, the standardized regression weights, and the covariances and the correlations. All of these would indicate if a construct is fit to be used for further analysis and the fit constructs would then be used as the basis for constructing the overall measurement model.

5.6.1 CFA Results for Business Strategy

Business strategy was measured with eight items adopted from Zahra and George (2000). These eight items represent the dimensions of cost (three items), quality (two items) and innovation (three items). In this analysis it was expected that since the respondents answered this with respect to TM, the dimensions of quality and innovation would be positively related to TM while the dimension of cost would be negatively to TM. Business strategy was therefore subjected to CFA in order to find out if the data fit the model of three dimensions. The CFA results of the three dimensions of the model were as follows: the χ2 value was 16.7 with a degree of freedom of 17 at a p-value of 0.473, while the χ2/df was 0.984. The CFI = 1.00, IFI = 1.00, TLI = 1.00 and the RMSEA = 0.000. This shows that the data fit the model of the three dimensions well.

Further examination of the model showed that all of the CRs values were greater than 1.96, which indicated that all of the estimates were statistically different from zero and the null hypothesis that all estimates equals 0.0 can be rejected.

Additionally, the parameter estimates were positive and within the logical anticipated range of values (i.e. no estimate exceeded the value of 1.00). Again, the path coefficient from each latent construct to the observed indicators was significant (p <

0.000) and the standardised regression weight ranged from 0.67 to 0.90. This supported the validity and reliability of the items (Hair et al., 2010; Meyers, Gamst and Guarino, 2013; Tabachnick and Fidell, 2014). All covariances were significant and the correlations ranged from -0.20 to 0.54, which is considered reasonable and reliable.

I also conducted another CFA where all of the eight latent variables were loaded as one construct and the result were as follows: the χ2 value was 96.4 with a degree of freedom of 20 at a p-value of 0.00 while the χ2/df was 4.819. The CFI = 0.844, IFI = 0.847, TLI = 0.782 and RMSEA = 0.129. The factor loadings were also as follows: (Qua1= 0.34, Qua2 = 0.31, Inno1= 0.65, Inno2 = 0.77, Inno3 = 0.70, Cost1 = 0.44, Cost2 = 0.50, Cost3 = 0.52) and the factor loadings were statistically significant at p < 0.001. The model with one dimension (all eight items together) was compared with the model with three dimensions. Consistent with Zahra and George (2000), the CFA results showed that the three dimensions provided a better fit to the data than all of the eight items together. Accordingly, the three dimensions was used

for further analysis. Table 5.5 presents the results of the estimates, standard error, critical ratio, p-values and the AVE, and they all fall within the acceptable range.

Table 5.5 Parameter Estimates of Business Strategy

Latent Manifest Estimate S.E. C.R. p AVE

Quality Qua2 0.884

Qua1 0.803 0.109 8.136 *** 0.713

Innovation Inno3 0.674

Inno2 0.843 0.142 8.985 *** 0.556

Inno1 0.710 0.122 8.704 ***

Cost Cost3 0.900

Cost2 0.810 0.063 13.640 *** 0.687 Cost1 0.772 0.060 12.875 ***

Figure 5.2 Measurement Model for Business Strategies

5.6.2 CFA Results for Talent Management

Talent management was measured with five items as adapted from Luna- Arocas and Morley (2015). In their study, Luna-Arocas and Morley measured TM with five items representing the dimensions of talent alignment, talent application, talent competency, talent autonomy, and talent development. These five items were subjected to CFA. The fit indices showed that the model achieved a very good fit with the data (the χ2 value was 11.3 with a degree of freedom of 5 at a p-value of 0.046 while the χ2/df was 2.256. the CFI = 0.97, IFI = 0.988, TLI =0.975 and RMSEA = 0.074). The CR values were greater than 1.96, which indicated that all of the estimates were statistically different from zero while all of the parameter estimates were positive and within the logically anticipated range of values.

More specifically, the path coefficient from each latent construct to the observed indicators was significant (p < 0.000) and the standardised regression weights ranged from 0.71 to 0.82. Again, all of the covariances were significant and the correlations ranged from 0.38 to 0.83, which is considered reasonable and reliable as well (Hair et al., 2010; Meyers, Gamst and Guarino, 2013; Tabachnick and Fidell, 2014). Table 5.6 presents a summary of the results in terms of the estimates, standard error, critical ratio, p-values, and the AVE.

Table 5.6 Parameter Estimates of Talent Management

Latent Manifest Estimates S.E. C.R. p AVE

TM → TM5 0.700

→ TM4 0.800 0.132 10.558 ***

→ TM3 0.710 0.120 9.837 *** 0.568

→ TM2 0.730 0.118 9.911 ***

→ TM1 0.820 0.126 10.842 ***

Figure 5.3 Measurement Model for Talent Management

5.6.3 CFA Results for Perceived Organisational Support

Perceived organisational support was measured with six (6) items adapted from Eisenberger et al. (1997). These items were subjected to CFA and the result showed that all of the indices achieved a very good fit. Specifically, the χ2 value was 52.9 with a degree of freedom of 9 at a p-value of 0.001, χ2/df=5.876, IFI = 0.946, TLI = 0.910, CFI = 0.946 and RMSEA = 0.070. The factor loadings were also acceptable, ranging from a low value of 0.63 to a high value of 0.89, and all of the items had a statistically-significant relationship with their factors (p < 0.001).

The direction of all of the estimated parameters was positive, and the critical ratios (t-test) for all of the estimated parameters exceeded the benchmark of +1.96, which were also found to be statistically significant and the standard error (SE) was not excessively large or small. The covariances were significant and the correlations ranged from 0.61 to 0.90, which is considered reasonable and reliable as well. This is shown in table 5.7.

Table 5.7 Parameter Estimates of Perceived Organizational Support

Latent Manifest Estimate S.E. C.R. p AVE

POS6 0.673

POS5 0.627 0.100 8.712 ***

POS → POS4 0.757 0.092 10.313 ***

→ POS3 0.885 0.096 11.724 *** 0.598

→ POS2 0.828 0.094 11.124 ***

→ POS1 0.835 0.102 11.208 ***

Figure 5.4 Measurement Model for Perceived Organisational Support

5.6.4 CFA Results for Employee Satisfaction

Employee satisfaction in this study was measured with three items developed by Seashore et al. (1982). These three items were subjected to CFA and it was found that the model was just identified because of the number of indicators in the model.

However, both the internal consistency of the scale and the convergent validity were high, while the Cronbach alpha was above 0.80 and the AVE was 0.782. Therefore, the factor loadings were good as they ranged between 0.76 and 0.99 with all having a statistically-significant relationship with their factor loadings at p < 0.001. Again, all of the goodness-of-fit indices indicated excellent fit (the χ2 value was 0.00 with a

degree of freedom of 0 at a p-value of 0.001, IFI = 1.000, TLI = 1.000, CFI = 1.000 and RMSEA = 0.000).

Thus, a just identified model was perfect. The CR values were greater than 1.96 which indicated that all of the estimates were statistically different from zero while all of the parameter estimates were positive and within the logically-anticipated range of values. All covariances were significant and the correlations ranged from 0.63 to 0.90, which is considered reasonable and reliable as well. This is shown in table 5.8 and figure 5.5.

Table 5.8 Parameter Estimates of Satisfaction

Latent Manifest Estimate S.E. C.R. p AVE Satisfaction → Satis3 0.763

→ Satis2 0.991 0.01 21.308 *** 0.782

→ Satis1 0.885 0.01 15.091 ***

Figure 5.5 Measurement Model for Satisfaction

5.6.5 CFA Results for Employee Commitment

As indicated earlier, employee commitment in this study was regarded as affective commitment. Therefore, employee commitment was measured using three items from the scale developed by Meyer, Allen and Smith (1993). A CFA was performed on these items and the model was just identified. As such the fit indices

were as follows: (the χ2 value was 0.0 with a degree of freedom of 0 at a p-value of 0.000, IFI = 1.000, TLI = 1.000, CFI = 1.000 and RMSEA = 0.000). Again, all of the CR values were greater than 1.96, indicating that all of the estimates were statistically different from zero.

The parameter estimates were all within the logically-anticipated range of values while the path coefficient from each latent construct to the observed indicators was significant (p < 0.000) and the standardised regression weight ranged from 0.70 to 0.99. This supported the validity and reliability of the items (Hair et al., 2010;

Meyers, Gamst and Guarino, 2013; Tabachnick and Fidell, 2014). All of the covariances were significant and the correlations ranged from 0.52 to 0.84, which is considered reasonable and reliable as well. This is shown in table 5.9 and figure 5.6.

Table 5.9 Parameter Estimates of Commitment

Latent Manifest Estimate S.E. C.R p AVE Commitment → Comit3 0.991

→ Comit2 0.704 0.049 14.547 *** 0.813

→ Comit1 0.980 0.121 40.311 ***

Figure 5.6 Measurement Model for Commitment

5.6.6 CFA Results for Employee Performance

Employee performance is generally considered as a multi-dimensional concept (Borman and Motowildo, 1993; Sonnentag and Frese, 2002). A review by Koopmans et al. (2011) showed that employee performance is made up of four dimensions: task, contextual, and adaptive performance, as well as well as counterproductive behaviours. One of the central arguments in this study is that most of the studies on TM and employee performance (Abbasi, Sohail, Cheema, Syed, 2010; Chami- Malaeb, 2012; Kumari and Bahuguna, 2012; Wurim, 2012; Kaur, 2013; Luna-Arocas and Morley, 2015) did not research employee performance according to these dimensions. It was therefore not clear which of the dimensions had been studied previously. Mensah (2015) has argued that TM may relate to the various dimensions of employee performance differently. Therefore, this study attempted to find out the relationship between TM and these four dimensions of employee performance.

Employee performance was therefore subjected to CFA to find out if the data fit the model of the four dimensions. The CFA result of the four dimensions of the model was as follows: the χ2 value was 847.0 with a degree of freedom of 225 at a p- value of 0.000 while the χ2/df was 2.50. The CFI = 0.934, IFI = 0.935, TLI = 0.920 and RMSEA = 0.080. This showed that the data fit the model of the four dimensions.

Further examination of the model showed that all of the CR values were greater than 1.96, which indicated that all of the estimates were statistically different from zero.

Also, the parameter estimates were positive and within the locally-anticipated range of values. Again, the path coefficients from each latent construct to the observed indicators were significant (p < 0.000) and the standardised regression weight ranged from 0.53 to 0.95. This supported the validity and reliability of the items (Hair et al., 2010; Meyers, Gamst and Guarino, 2013; Tabachnick and Fidell, 2014). All of the covariances were significant and the correlations ranged from -0.26 to 0.61, which is considered reasonable and reliable as well.

To be sure whether the data fit the four dimensions of employee performance to be used in this study, all of the items were loaded once to be compared with the four dimensions. The results of the overall CFA were given as follows: the χ2 value was 475.1 with a degree of freedom of 104 at a p-value of 0.000, χ2/df =4.568, IFI = 0.783, TLI = 0.748, CFI = 0.781 and RMSEA = 0.124. The factor loadings of the

first-order CFA ranged from -0.42 to 0.75 and they were statistically significant at p <

0.001.

In comparing the overall CFA model to the four dimension CFA model, both models had a good χ2/df but in terms of other indices, the overall model poorly fit the data. Therefore, based on other indices, it can be seen that the data fit the four dimension model better than the overall CFA model. Hence, the four dimension CFA model was used in subsequent analysis. This is illustrated in table 5.10 and figure 5.7.

Table 5.10 Parameter Estimates of Employee Performance

Latent Manifest Estimate S.E. C.R p AVE

Task Task5 0.665 0.120 9.278 ***

→ Task4 0.531

→ Task3 0.867 0.142 10.281 *** 0.539

→ Task2 0.575 0.113 8.334 ***

→ Task1 0.943 0.147 10.586 ***

Contextual → Cont8 0.659

→ Cont7 0.584 0.119 8.031 ***

→ Cont6 0.834 0.147 9.440 ***

→ Cont5 0.842 0.140 9.387 ***

→ Cont4 0.549 0.140 7.488 *** 0.517

→ Cont3 0.547 0.141 7.512 ***

→ Cont2 0.621 0.154 8.051 ***

→ Cont1 0.721 0.153 8.649 ***

Adaptive → Adapt6 0.677 0.092 10.862 ***

→ Adapt5 0.752

→ Adapt4 0.779 0.088 12.075 *** 0.568

→ Adapt3 0.771 0.093 12.091 ***

→ Adapt2 0.828 0.090 12.595 ***

→ Adapt1 0.706 0.084 10.909 ***

CPWB → Count4 0.915

→ Count3 0.954 0.038 28.108 *** 0.739

→ Count2 0.951 0.038 27.735 ***

→ Count1 0.899 0.044 23.319 ***

Figure 5.7 Measurement Model for Employee Performance

5.6.7 CFA Results for Intention to Quit

Employee intention to quit was measured with four (4) items developed by O’Reilly, Chatman and Caldwell (1991). After subjecting these four items to CFA, the results of the fit indices were as follows: (the χ2 value was 8.7 with a degree of freedom of 2 at a p-value of 0.013, χ2/df=4.343, IFI = 0.986, TLI = 0.957, CFI = 0.986 and RMSEA = 0.120). All of the factor loadings were good, ranging from 0.55

to 0.92 and all of the items had a statistically significant relationship with their factors (p < 0.001) and also in the expected direction. However, the χ2 value and the RMSEA indices did not fit the data well and required further modification. An examination of the modification indices showed that item two (Quit2) did not fit the data and had to be deleted.

The re-run showed that the model was just identified because of the number of indicators in the model. However, the data fit the model well. All of the factor loadings were good, ranging from 0.62 to 0.95, and all of the items had a statistically- significant relationship with their factors (p < 0.001). The fit indices were as follows:

IFI = 1.00, TLI, 1.00, CFI = 1.00 and RMSEA = 0.070. Again, the critical ratio (t-test) for all the estimated parameters exceeded the benchmark of +1.96 and was statistically significant, and the SE was not excessively large or small. The covariances were significant and the correlations ranged from 0.48 to 0.95. This is shown in table 5.11 and figure 5.8.

Table 5.11 Parameter Estimates of Intention to Quit

Latent Manifest Estimate S.E. C.R P AVE

Quit → Quit4 0.77

→ Quit3 0.99 0.09 13.78 *** 0.722

→ Quit1 0.77 0.07 12.75 ***

Figure 5.8 Measurement Model for Intention to Quit

Dalam dokumen PDF repository.nida.ac.th (Halaman 185-197)