BUS603: Managing People

Before evaluating the appropriateness of the proposed structural model, as depicted in Figure 1, the current study first determined the degree of fit associated with several measurement models. The current study employed confirmatory factor analysis to evaluate goodness of fit associated with these measurement models. Three measurement models were compared. Only when a measurement model shows acceptable fit can it be used to evaluate the structural model.

Model 1 consisted of three latent variables (servant leadership, emotional intelligence and trust in the manager). Servant leadership was measured using the five dimensions associated with the construct, whilst emotional intelligence was measured by using its five subdimensions. Finally, trust in the manager was measured by 12 items related to the construct. From Table 1 it is clear that, although this model shows acceptable fit with regard to CFI (0.98), it shows lack of fit in terms of both the RMSEA and SRMR, which are higher than the recommended values. It was therefore decided to first employ item parcelling (for trust in the manager) to try and improve model fit. Item parcelling has the potential to improve model fit because it reduces the complexity of the measurement model and models with fewer variables have the potential for better fit. It is important to note that parcelling should only be done when the construct, in this case trust in the manager, is unidimensional, which has been established for this construct. Hence, Model 2 only differs from Model 1 in that trust in the manager was measured using four composite item parcels. Unfortunately, this did not result in much improvement of the fit of the model (See Table 1), especially with regard to the unacceptably high value associated with RMSEA.

In addition to using item parcels to improve model fit, researchers can consult modification indices. Modification indices may indicate the possible origin of the observed misfit associated with the measurement model. In the current study, the modification indices suggested that correlating the errors of certain dimensions related to the emotional intelligence construct (self-regulation with social skills, empathy and motivation, social skills with empathy and self-awareness with empathy) would lead to an improvement in model fit. This seems plausible given the fact that:

Hence, it is likely that, although these dimensions measure different aspects related to emotional intelligence, they still overlap in terms of measuring emotional intelligence in general. Hence, Model 3 only differs from Model 2 in that the errors of the previously mentioned emotional intelligence dimensions were correlated. From Table 1 it is evident that this has led to a much more appropriate measurement model, with all the approximate fit indices well within the suggested cut-offs. In addition, Model 3 had the smallest value associated with AIC, indicating that it was the better fitting model. It was therefore decided that the testing of the structural model would be based on Model 3.