published population values for the specific population of interest that this
research focuses on (waiting staff in hotels with table service restaurants) or that
focus on the more specific constituencies (types of organisations) that form the
sample for this research (3 and 4 star hotels).
Assessing the likely impacts of non-response bias is clearly problematic in the
light of the difficulties arising from attempting to evaluate survey non-response
through comparisons of the sample characteristics with known values from
similar, but not identical, populations.
It is possible, however, to use SEM analysis to move beyond the subjective
estimation of non-response effects. Specifically, in order to explore whether or
not respondents’ demographic characteristics have any influence on the study
findings, multi-group SEM analyses will be performed to explore the differences
between groups. These multigroup analyses are reported in Section 7.15.
The benefit of SEM multi-group analysis in the context of survey non-response is
that, where a multi-group SEM model indicates that group membership does not
influence the measured relationships within a SE model, then we can be assured
that, even in the absence of a perfect understanding of the representativeness of
the sample, the estimated relationships are likely to be generalisable to the
population of interest.
6.6 Data screening
Prior to initiating the modelling work, the data were screened for missing values,
distributional normality, presence of outliers and an analysis of non-response
assessed how consistently the sample compares with the population of interest.
6.6.1 Treatment of missing data
Missing data are not uncommon in survey research (e.g. Byrne 2010: 353) and a
common source of missing data when using respondent-completed questionnaires
is non-response to one or more survey questions. Not surprisingly, an initial
examination of the completed data set for this research shows that there are
some missing values. Such non-response can occur where some respondents are