However, the majority of the missing values appeared systematically
across an entire range of questions. Therefore
we interpreted them as an active and systematic choice
of no opinion (i.e., ‘not relevant/lacking knowledge’) and
not as missing values in the ordinary sense. Consequently,
we did not replace missing values in our analysis. A technical
and compelling argument for not replacing missing
values in cluster analysis, with, for example, the series
mean, is that clusters, from which the proper mean calculation
should be executed, are, a priori, unknown (Everitt,
1993). Replacing missing values in the data set at the outset
would have had a distorting effect (either normalizing
or randomizing, depending on the method) on the formation
of ‘true’ clusters. On these grounds, we found it
sounder to use the smaller but complete data set (N = 71)
that resulted from the list-wise deletion of both missing
values and the response category ‘not relevant/lacking
knowledge’.18 Relating the 71 observations to the other
163 observations shows that there are generally only limited
and insignificant differences between them regarding
factors such as municipality size, financial strength, and
the amount of outsourcing (see Table A1). However, size
is an exception when the two large cities of Stockholm
and Gothenburg are included amongst the 71 municipalities;
when comparing medians in size amongst the groups,
no difference appears. Furthermore, relating only the 71
observations to the total population yields a response rate
of 24 per cent that is comparable with other cluster studies
(e.g., Henri, 2008) and research on IORs (e.g., Poppo
et al., 2008). Although missing-value analysis indicates that
the sample is missing completely at random (MCAR) the
potential presence of unobserved variables causing NMAR
(not missing at random) and other selection bias problems
discussed above causes us to treat the study (N = 71
sample) as exploratory and to see the results as indicative
relating to the general Swedish municipal population
(inference).