Model 1 tests hypotheses H 1 and H 2 which are articulated as:
H 1 (ML→JP): as employees experience greater levels of Motivational Leadership
they will also experience greater levels of Job Performance.
H 2 (ML→DSB): as employees experience greater levels of Motivational Leadership
they will also report greater levels of Discretionary Service Behaviour.
7.2.1 Measurement model specification
During Step 2 of the two-step procedure, the structural model specification
expresses the research hypotheses by linking the latent factors with uni-
directional connectors indicating causal relationships. During Step 1, however,
the measurement model is specified in such a way that all of the latent factors are
allowed to freely correlate using bi-directional connectors indicating non-causal
relationships. The focus of the measurement model is not on investigating the
causal relations between latent factors, but on ensuring that the individual latent
constructs (factors) are adequate in their role of measuring the concepts they are
intended to (convergent validity) and that each latent factor is in fact measuring a
unique construct (discriminant validity).
In specifying the measurement model, the researcher is focused on specifying –
from the theory that has been developed – which indicator variables load onto
which latent factors. In a congeneric measurement model (such as is being used
in this research), each indicator variable (the rectangles in Figure 7 - 2 ) loads onto
only one latent factor (the ellipses in Figure 7 - 2 ).
The 13 circular terms connecting to the indicator variables are error variances and
represent measurement error – this can be conceptualised as the inherent
unreliability in the capturing of data (introduced by sampling errors, mis-
interpretations of questions on the part of respondents, mistakes by respondents
etc). In SEM analysis, error variance is calculated as a product of (a) the
proportion of variance in the indicator not explained by the latent factor and (b)
that indicator’s variance (the average of the squared differences from the mean
for that indicator) (Garson 2011b). It is possible to specify links between error
covariance terms – the theoretical implications of this are that the researcher is
hypothesising that these linked error variances indicate that those indicators are
influenced not only by the latent factor, but also by some other common, but
unmeasured, factor.