7.2.9 Measurement model: first respecification (CFA 1:2)
The model is now respecified without the inclusion of ML4, ML5 and JP3 and DSB4
and the estimates for the respecified model are described in Table 7 - 5.
The model fit statistics have all improved in comparison with the initially specified
model (CFA 1:1). The model chi-square, however, remains below 0.05 indicating
that the model does not fit the data as well as it might. The AVE values for each
of the factors have improved; the CR value for ML has increased and for JP and
DSB, the CR values have decreased slightly. The respecified model also performs
satisfactorily with regard to discriminant validity – the minimum AVE is 0.604 and
the maximum squared inter-factor correlation estimate is 0.195. The
requirements for nomological validity continue to hold with all factor covariances
at statistically significant levels (p = <0.01).
Construct Item Standardised factor loading estimates
(^) ML JP DSB
Motivational
Leadership
ML1 .903
ML2 .9 57
ML3^ .873
Job Performance
JP1 .868^
JP2 .857^
JP4^ .459
Discretionary
Service Behaviour
DSB1 .839
DSB2 .776^
DSB3^ .618^
Average variance
extracted (AVE) 0.766^ 0.618^ 0.604^
Construct
reliability (CR) 0.908^ 0.820^ 0.819^
Model fit statistics
χ^2 = 41.830; d.f. = 21; sig = 0.013
RMSEA = 0.059 (0.088; 0.027; pclose = 0.282)
CFI = 0.983
SRMR = 0.0446
CN (0.05) = 185
Table 7-5 Estimates for CFA 1:2
One factor loading is flagged for attention – DSB3 has dropped from 0.677 to
0.618 prompting an examination of the standardised residual covariance (SRC)
matrix to see if there are any other problems associated with DSB3. The
standardised residual covariance matrix reveals that DSB3 shares a residual
covariance value of 2.689 with JP4. Because SRC values greater than 1.96, and
particularly those greater than 2.58, are indicative of a component of the model
that does not correspond well with the data, a useful modification here will be to