Considering rule of thumb (C) 6 cases with greater than 10 per cent missing
values are flagged for attention.
The next step is to examine the missing data for non-random patterns.
Six cases had a clear non-random pattern of missing data – specifically, these
respondents had not completed any of the 4 DSB (Discretionary Service
Behaviour) item statements. This may be an example of ‘end of questionnaire
attrition’ (see e.g. Hair et al. 2006: 55) as the four DSB items appear on the back
page of the survey form, separate from the other item scale statements.
Although, notably, 3 of these 6 respondents actually went on to complete the
demographic questions following the DSB item statements.
Hair et al. (2006: 55) note that, to a large extent, researcher judgment is
required in assessing the impact of missing data on any particular data set. For
this research, considering the requirement to avoid a concentration of missing
data on specific sets of questions/statements, an ad hoc guideline was introduced
to prevent concentrated missing data on individual latent constructs. Specifically,
it was decided to remove cases where missing values constitute more than one
third of the total values per latent construct. In practical terms, this means that
for latent constructs measured with 3 to 5 items, no more than 1 item per case
could be missing. For latent constructs with 6 to 8 items no more than 2 items
could be missing.
Five cases with instances of more than one third of values missing within a latent
construct were identified:
1 case was missing 2 DSB items (DSB has 4 items in total);
1 case was missing 4 ML (Motivational Leadership) items (ML has 5 items in
total);
1 case was missing 3 ML items and 3 JP (Job Performance) items (JP has 4
items in total);
1 case was missing 3 ME (Work Meaning) items (ME has 5 items in total); and
1 case was missing 2 ME items (ME has 5 items in total).
These 11 cases with non-random, or concentrated patterns of missing values
were removed from the data set leaving 213 cases. This results in the data set
containing only two cases where more than one value was missing per latent
construct. The first case was missing 2 values (33.3 per cent) from the Job
Satisfaction (JS) construct and the second case was missing 2 values (29 per