leadership and motivation in hospitality

67. and their associated standardised residual covariances (Byrne 2010: 86; and,
    see also Shumaker and Lomax 2004: 177). This respecification is often carried in
    an iterative fashion while the researcher monitors the overall effect of individual
    changes on the broader model. Further, any model modifications should be
    carried out in the light of appropriate consideration of the substantive changes
    (i.e. the theoretical implications for the model) that they will bring about.

So, having developed a measurement model which exhibits good fit and provides
evidence for acceptable convergent and discriminant validity, Step 2 of the
procedure develops the structural model to provide an assessment of predictive
validity. Part of the rationale for the two step procedure is described by Jöreskog
and Sörbom (1993) who write that:

The testing of the structural model, i.e., the testing of the initially

specified theory, may be meaningless unless it is first established that

the measurement model holds. If the chosen indicators for a construct

do not measure that construct, the specified theory must be modified

before it can be tested. Therefore, the measurement model should be

tested before the structural relationships are tested.

(Jöreskog and Sörbom 1993: 113)

During Step 2, the structural model is developed following similar principles of
iterative improvement as with the measurement model. In the case of the
structural model however, the modifications typically take the form of
respecifications to the pattern of between factor (rather than within factor)
parameters^5. Once again, this process should be strongly related to and guided
by consideration for the substantive implications of any changes made
(Schumaker and Lomax 2004: 71; Kline 2005: 65; Hair et al. 2006: 797-798).

The five stage procedure described by Schumaker and Lomax (2004) throughout
their guide to SEM analysis is based on the Anderson and Gerbing Two-Step
procedure and breaks each of the two step into the following five stages:

1. model specification;


2. model identification;

5
It is possible to make changes to the factor structures (i.e. removal of indicator
variables) during the development of the structural model, however, following any
such change it is then necessary to re-estimate the measurement model to ensure
that the requirements of construct validity continue to be met.