decision. None of the variables exceeded Curran et al.’s threshold of 7 for
univariate kurtosis (see Section 6.3 on data assumptions for details of
distributional normality).
Hair et al. (2006: 780) suggest that mean values for individual indicator variables
should not approach the upper or lower limits of the scale range. For these data,
the highest mean score was 4.45 (DSB2) and the lowest was 2.55 (SQ7). Hair et
al. did not offer any specific guidance regarding a threshold that indicates when a
mean value is too close to an upper or lower limit; in this case, because 4.45 is
not more than ‘halfway’ between the upper scale points 4 and 5 the score was
deemed to be satisfactorily distant from the upper polar extreme of 5. Using this
same logic, the lower bound mean score of 2.55 is sufficiently distant from the
lower polar extreme of 1 not to cause any concern.
Confirmatory factor analysis (CFA) of the individual constructs was used to check
the statistical significance and magnitude of the item loadings – that is, the
strength of association between each indicator variable and the construct that is
hypothesised to influence that variable. Once again, no specific cut-off value is
proposed by Hair et al. (2006: 780-781) regarding the level at which an indicator
is failing to perform adequately well at the pre-test stage. In a more general
context, Hair et al. (2006: 777, 795) note that factor loadings should preferably
be above 0.7 and at least 0.5 (see Section 7.2.5 below on assessment of
construct validity for more details on this). Accordingly, for the pre-test
assessment, indicator variables loading below 0.5 were considered as candidates
for removal from the survey.
Six indicator variables loaded below the 0.5 level (EM2 at 0.47; ME2 at -0.06; and
ME4 at 0.18; AOC3 at 0.36; DSB3 at 0.47; and DSB4 at 0.42). In total, two
variables (ME2 and ME4) had non-statistically significant factor loadings.
The indicator variable EM2 (I can make my own decisions at work) was removed
as it had a clear item content overlap with indicators EM1 (I can choose the best
way of doing my job) and EM4 (I have a great deal of control over my job) and
was removed from the survey. Indicator variables ME2 and ME4 loaded very
weakly and also failed to meet the criterion of statistical significance leading to
their removal from the survey. The indicator AOC3^7 loaded at only 0.36 and was
7
AOC3 was also the weakest-loading item (0.45) in Allen and Meyer’s (1991) study
from which the Affective Organisational Commitment scale is drawn.