therefore removed from the survey. Because DSB3 and DSB4 were not part of an
exploratory construct (DSB has been previously – and successfully - measured
using the same 4 indicator variables by Simons and Roberson 2003) it was
decided to retain these two indicators in the survey and to monitor their future
performance.
As discussed above in Section 5.3, there is no consensus on the use or non-use of
reverse-polarity items in psychometric scale design. Ray (1983) found evidence
supporting their use while both Herche and Engelland (1996) and Wong (2003)
found that reverse wording caused problems with scale unidimensionality and
construct validity.
Because the use of reverse worded items is not without problems (in particular,
Wong’s issues with reverse worded items in cross-cultural research), only three
indicators in this pilot survey were (somewhat cautiously) included with reverse
wordings. These indicators were ME2, ME4 and AOC3 and, indeed, each of these
performed poorly in the pilot survey. These particular indicators were chosen for
reverse wording because, firstly, AOC3 followed the original wording from Allen
and Meyer’s (1990) Affective Commitment scale. The ME2 and ME4 indicators
were chosen for reverse wording because the Work Meaning (ME) scale was the
only other scales that was both amenable to reverse wording and not based on an
existing set of indicator wordings.
The three indicators (AOC3, ME2 and ME4) were all removed from the survey and
the cautious conclusion drawn is that reverse worded statements do not work well
in hospitality service contexts where respondents are from a variety of cultural
backgrounds. (NB – ME2 and ME4 also negatively affected the unidimensionality
[see below] of the Work Meaning construct, reflecting Wong’s (2003) findings for
cross-cultural context psychometric research).
Regarding factor structures / unidimensionality, exploratory factor analysis (EFA)
was used to check that the indicator variables for each construct did indeed share
a single common cause (i.e. were unidimensional). The Employee Empowerment
(EM) indicators were an exception to this as these are hypothesised to measure
two discrete factors (EM-I, Empowerment – Influence; and EM-C, Empowerment –
Competencies). The EFA analyses were performed using maximum likelihood
estimation (the same estimation method used in the CFA analyses). Varimax
rotation of the factor matrix was applied to maximise the clarity of the factor