Before estimating the structural model, it is necessary to estimate and develop a
good-fitting measurement model and to ensure that the assumptions of
convergent and discriminant validity are met. Maximum Likelihood (ML)
estimation was used and missing values were replaced using AMOS’s model-based
imputation function.
The measurement model is initially specified with all of the indicator variables,
even though some of these (for ML, JP and DSB) were found not to perform well
in the previous model. By building each successive model iteration ‘from scratch’,
two issues are addressed. Firstly, that the model-based imputation of missing
values is based on the specific model that is being estimated and tested.
Secondly, this approach will confirm (or refute) the robustness of the
measurement of the individual constructs. That is, by estimating, testing and
modifying each model ‘from scratch’ it is possible to make an assessment of the
consistency of factor structures across models.
The estimates and model fit diagnostics for the initially-specified measurement
model, CFA 2:1 are described in Table 7 - 9.
Construct^ Item^ Standardised factor loading estimates
ML ME JP DSBMotivational
LeadershipML1 (^) .894 (^)
ML2 (^) .937 (^)
ML3 .894 (^)
ML4 .613 (^)
ML5 .615 (^)
Work Meaning
ME1 .879 (^)
ME3 .861 (^)
ME5. (^610)
ME6 .794 (^)
ME7.^303
Job Performance
JP1 .849 (^)
JP2 .859 (^)
JP3. (^556)
JP4.^493
Discretionary
Service Behaviour
DSB1 .799
DSB2 .774
DSB3 .674
DSB4 .696
Model fit statistics
χ^2 = 324.084; d.f. = 129; sig = 0.000
RMSEA = 0.084 (0.096; 0.073; pclose = 0.000)
CFI = 0.906
SRMR = 0.0736
CN (0.05) = 103