Chapter 6: Methodology and analysis: Exploratory factor analysis (EFA) and confirmatory
6.7 Model fit for the measurement model
In this CFA model, supplier costs and risk and supplier benefits were combined to produce one factor that was labelled as supplier net value. This was calculated as supplier benefits minus supplier costs and risks. This was done to reduce the number of factors and to produce one construct that measured the perceived net value of the relationship. The relational and cognitive social capital constructs were left as one factor in accordance with the EFA results. Trust and commitment, however, were separated into discrete constructs in the model as there was good theoretical evidence that these were distinct constructs
(Morgan & Hunt, 1994). These two factors would also be tested for discriminant validity in the CFA.
The measurement model for the relationship factors (Figure 6-1) showed an excellent model fit. The model was within the threshold values on all the goodness-of-fit, except for GFI (Table 6-11).
Table 6-11: Model fit criteria for relationship factors CFA Model Relationship Factors
Measure Measurement Model Threshold
Chi-square/df (cmin/df) 2.76 < 3 good
CFI 0.94 > 0.95 great; >0.9 traditional
GFI 0.87 > 0.95
AGFI 0.85 > 0.80
RMSEA 0.04 < 0.05 good; 0.05-0.10 moderate
PCLOSE 0.99 > 0.05
Table 6-12: Measurement statistics (after item deletions)
CR AVE 1 2 3 4 5 6 7 8 9 10 11 12 13 1. ProcDependence 0.81 0.68 0.83 2. Trust 0.93 0.69 0.15 0.83 3. SocialCap_RelCog 0.97 0.82 0.16 0.67 0.91 4. Commitment 0.81 0.58 0.11 0.78 0.55 0.76 5. Value 0.91 0.56 0.29 0.65 0.48 0.61 0.75 6. Cost 0.91 0.64 0.13 -0.41 -0.19 -0.45 -0.26 0.80 7. SatisfactionPerform 0.91 0.62 0.12 0.59 0.46 0.55 0.54 -0.32 0.79 8. SocialCap_Stuct 0.91 0.77 0.27 0.65 0.58 0.52 0.61 -0.18 0.47 0.88 9. SatisfactComm 0.87 0.69 0.03 0.55 0.45 0.53 0.44 -0.34 0.49 0.53 0.83 10. SpecificInvest 0.85 0.65 0.36 0.13 0.14 0.16 0.42 0.29 0.10 0.32 0.05 0.81 11. Loyalty 0.73 0.48 0.09 0.44 0.32 0.52 0.41 -0.40 0.32 0.34 0.34 0.12 0.69 12. SatisfactionPrice 0.79 0.56 -0.05 0.63 0.41 0.48 0.52 -0.42 0.50 0.39 0.50 -0.11 0.24 0.75 13. Power 0.59 0.42 0.29 -0.27 -0.14 -0.28 -0.02 0.56 -0.24 -0.04 -0.27 0.53 -0.22 -0.30 0.65
Notes (1) the diagonal entries express the variance extracted. The figures underneath the diagonal is the correlation between constructs.
There were a number of validity concerns where the criteria for discriminant and convergent validity were not met. The power construct initially had an AVE of 0.32 and a composite reliability score of 0.56. The loyalty construct initially had an AVE of 0.43. For both of these, the square root of the AVE was less than the absolute value of the correlation with another factor. The decision was made to delete some items to improve discriminant validity. The item deletions were based on measures with low factor loadings. Power1_Treat with loading of 0.54 was removed from the power construct and CMT_L1OptR with weighting of 0.54 deleted from loyalty (Table 6-12).
Table 6-13: Validity concerns in relationship constructs Construct Validity concern
Loyalty Convergent Validity: the AVE for loyalty is less than 0.50. Power Reliability: the CR for Power is less than 0.70.
Convergent Validity: the AVE for power is less than 0.50.
Following these deletions there were still some validity concerns (Table 6-13). Although these concerns remained, Hair et al. (2010) emphasise that the thresholds rules were only guidelines and suggests that more flexibility was acceptable especially when undertaking exploratory research. Following the deletions, loyalty had an AVE of 0.48, which was just below the suggested cut off of 0.5. In a similar way, power had an AVE of 0.42, which was significantly closer to the 0.50 cut off than before the item deletions (Table 6-2). Power also had a composite reliability score of 0.59, which was substantially lower than the 0.70
recommended. This indicated that there were some validity issues with the power construct and it was only suitable to be used in exploratory research and therefore should be treated with caution in making definitive predictions. The commitment construct also had some issues with discriminant validity; however, this was because of the high correlation between commitment and trust, which was 0.78. The square root of the AVE for commitment was 0.74, which was close to the value of the correlation between trust and commitment. These values indicate that the criteria for discriminant validity is close to being met.
Table 6-15: Shared variance, variance extracted and correlations between constructs for the supplier factors (before item deletions)
CR AVE 1 2 3 4 5 6 7 8 9 1. ProcDependence 0.78 0.64 0.80 2. SuppMotivation 0.81 0.38 0.13 0.62 3. SupplierAbility 0.91 0.71 0.14 0.47 0.84 4. CustomerFocus 0.81 0.48 0.19 0.38 0.24 0.70 5. SelfDirect 0.69 0.43 0.13 0.11 -0.15 -0.01 0.66 6. SupplierProfit 0.75 0.53 0.01 -0.10 -0.01 0.04 -0.11 0.72 7. SupplierComm 0.84 0.73 0.22 0.38 0.24 0.34 0.00 0.02 0.85 8. MarketUncertanty 0.68 0.42 0.19 0.20 0.07 0.23 0.12 0.01 0.13 0.65 9. SuppDependence 0.80 0.67 0.21 0.13 -0.01 0.12 0.24 -0.07 0.29 0.09 0.82
Notes (1) the diagonal entries express the variance extracted, the figures underneath the diagonal is the correlation between constructs.
Table 6-16: Validity concerns supplier constructs (before item deletions)
The data in Table 6-15 identified some validity issues (Table 6-15). Reliability was an issue for both self-direction and market uncertainty, although the composite reliable value for self- direction was 0.69 and the value for market uncertainty was 0.68, both of which were very close to the composite reliability threshold of 0.70 indicating that they both have reasonable reliability. There were convergent validity issues for customer focus, supplier motivation, self-direction and market uncertainty. The value for customer focus was 0.48, which was very close to the 0.50 threshold and, therefore, not a significant issue. However, the values for supplier motivation, self-direction and market uncertainty were all below 0.44, with the lowest being supplier motivation. As a result, items were considered for deletion on these constructs based on low factor loadings. One item was deleted from the self-direction construct. This item was “Selfdirect3_Constraint” (factor loading: 0.58). Deleting this increased AVE from 0.43 to 0.49. One item was deleted from the market uncertainty construct. This was “UncertMkt3_Price” (factor loading: 0.55) and this increased AVE from 0.42 to 0.55. Two items were deleted from supplier motivation construct; these were
Construct Validity concern
Supplier Motivation Convergent Validity: the AVE for supplier motivation is less than 0.50. Customer Focus Convergent Validity: the AVE for customer focus is less than 0.50. Self- direction Reliability: the CR for self-direction is less than 0.70.
Convergent Validity: the AVE for self-direction is less than 0.50. Market Uncertainty Reliability: the CR for market uncertainty is less than 0.70.
“SuppPerf4_AWelfare” (factor loading: 0.51) and SuppPerf5_NoPremium (factor loading: 0.46), which increased the AVE for from 0.38 to 0.45 (Table 6-17).
Table 6-17: Shared variance, variance extracted and correlations between constructs for supplier factors (after item deletions)
CR AVE 1 2 3 4 5 6 7 8 9 1. SupplierDependence 0.78 0.64 0.80 2. SuppMotivation 0.80 0.45 0.14 0.67 3. SupplierAbility 0.91 0.71 0.14 0.49 0.84 4. CustomerFocus 0.83 0.56 0.19 0.38 0.24 0.75 5. SelfDirect 0.66 0.49 0.13 0.16 -0.12 0.01 0.70 6. SupplierProfit 0.75 0.53 0.01 -0.10 -0.01 0.04 -0.11 0.73 7. SupplierComm 0.84 0.73 0.22 0.36 0.24 0.33 0.01 0.02 0.85 8. MarketUncertainty 0.69 0.55 0.20 0.15 0.08 0.21 0.09 0.04 0.11 0.74 9. SupplierDependence 0.80 0.67 0.21 0.13 -0.01 0.12 0.22 -0.06 0.29 0.05 0.82
Table 6-18: Validity concerns following item deletions
Following the item deletions, there were still some validity concerns even though most values had significantly increased (Table 6-18). Supplier motivation had an AVE of 0.45, still below the 0.50 cut off. In contrast, self-direction had AVE of 0.49, which was very close to .50. There were reliability issues for both self-direction and market uncertainty. This was only an issue for self-direction with a composite reliability score of 0.66. The score for market uncertainty was 0.69, which was only marginally below the 0.70 cut off.
Following the deletion of these items the validity of the model was considered acceptable. This was on the basis of the assertion of Malhotra and Dash (2011) who noted that, "AVE is a more conservative measure than CR. On the basis of CR alone, the researcher may conclude that the convergent validity of the construct is adequate, even though more than 50% of the variance is due to error” (Malhotra and Dash, 2011, p.702). Furthermore, Ping (2007)
explains that a low AVE may be acceptable for first time exploratory models that are incorporating new measures.
Construct Validity concern
Supplier motivation Convergent Validity: the AVE for supplier motivation is less than 0.50 Self-direction Reliability: the CR for self-direction is less than 0.70.
Convergent Validity: the AVE for self-direction is less than 0.50. Market uncertainty
6.8.2
Invariance testing
Appendix C presents the data from the analyse (invariance test) to see if the model produced the same results across the sheep beef and venison groups. If the constructs do not meet the test of invariance then they may be measuring different latent constructs for each group. This showed that the factor structure and loadings are sufficiently equivalent across the three groups.
6.8.3
Conclusion: Confirmatory factor analysis
This chapter presents the results of the CFA. The majority of the constructs showed
sufficient discriminant and convergent validity as well as reliability. There were some latent factors that did not meet these criteria. These were addressed by deleting some items with low factor loadings. Following these item deletions there was a significant increase in these variables, however there were still some marginally below the recommended values. These were considered to be acceptable to use in the SEM.