o therefore, removing either JS6 or ME1 will not impact the substantive
content of the construct as the ‘satisfaction’ component of the factor is
retained
o using the standardised residual covariance matrix, JS6 is removed on
the basis that it is associated with a far greater magnitude of
unmeasured variance than ME1 (7.109 versus 2.084)
Model CFA 3b:6
Model fit continues to improve but chi square remains <0.05 (χ^2 p = 0.010;
RMSEA = 0. 0 43; CFI = 0.983)
The SRC between JS5 and ME3 has risen considerably to 2.9
There are no specific substantive reasons to chose between retaining JS5
(satisfaction with promotional opportunities) and ME3 (I enjoy going to work)
JS5, however, has a lower factor loading than ME3 (0.653 versus 0.851) and
far greater magnitude of unmeasured variance associated with it (20.7 versus
3.6)
JS5 is removed
Model CFA 3b:7
Model fit is now satisfactory (χ^2 p = 0.100; RMSEA = 0. 0 32; CFI = 0.992)
All SRCs are satisfactory at between ±1.96
The modification index (MI) values suggest problems with unmeasured
variances between several indicator variables in the new Employee Attitudes
construct:
o the largest of these is between AOC1 and AOC2 (10.583)
o there is some item overlap between these items - AOC1 describes ‘being
part of the family (at work)’ and AOC2 describes ‘emotional attachment
(to work)’
o consulting the standardised residual covariance matrix reveals that
AOC1 has a greater magnitude of associated unmeasured variance
o AOC1 is removed
Model CFA 3b:8
Model fit is now good – the parameter estimates, model fit and construct
validity values are reported in Table 7 - 19.