all SQ indicators loaded well (at >0.7) and all SQ factor loadings were
statistically significant;
the SQ construct, however, did not covary at a statistically significant level with
any other construct.
The model specification was modified in the usual way, with analyses of the
standardised residual covariances and modification indices being used to indicate
where modifications to improve this situation might be made. Ultimately, three
indicator variables were removed (SQ3, SQ5 and SQ7) from the SQ construct.
Following these modifications, the truncated SQ construct (consisting of indicators
SQ1, SQ2, SQ4 and SQ6) was estimated alongside the ML, EM, EPA, JP and DSB
constructs as per the final CFA specification used in Model 5b. However, no
statistically significant covariances between the SQ construct and any others were
found.
Because the modelling above was undertaken using the constructs present in
Model 5b, as a further check on the adequacy of the SQ construct, it was
modelled alongside the constructs used in the previous model iteration, Model 5,
which included the Work Values (WV) construct. Although it was not possible to
develop this model to a state of adequate fit, and SQ continued to covary at non-
statistically significant levels with the constructs as described above, SQ did
covary at a statistically significant level (p = 0.034) with the Work Values
construct. Accordingly, a CFA model containing only the SQ and WV constructs
was estimated. It was not possible, however, to develop a model that exhibited
satisfactory fit and in which the covariance between the SQ and WV constructs
remained statistically significant.
7.12 Post-development validation of the model
Two aspects of validation are dealt with in this section: (i) cross-checking the
model-based data imputation process; and (ii) cross-validating the model using a
split-sample multi-group analysis of invariance.