Area. This is reconfirmed in the descriptive statistics derived from
the questionnaire shown in Appendix 2, where it can be seen that
seeing the natural scenery dominates (mean ¼ 5.55) although
learning about different cultures does occupy the third place with
a mean score of 4.88. This implies that although not a primary
reason, the existence of cultural difference remains a strong
secondary or supporting factor when creating an image of Kanas as
a tourist place.
The degrees of specific interest in minority culture were then
tested by a section of the questionnaire that examined the extent to
which the respondents were actually satisfied with the separate
overall attributes of the region, and the degree to which they were
‘impressed’ by local cultures. With reference to degrees of appreciation
of local culture, it should be noted the level of non-response
to items begins to be an issue. While about 90% of respondents felt
able to make a comment about the architectural style of the local
Tuva homes, the degree of non-response differed across other
items, reflecting the degree and depth to which respondents
actually engaged with the Tuva people. Thus 526 respondents felt
able to comment about horse riding (it should also be noted that
horse riding was primarily controlled by local Kazakhs rather than
Tuva), and thus comments are confused as some felt they did mix
with Tuva, but others recognised they were Khazak or other nonlocal
groups.
The first test was to assess the determinants of satisfaction with
the visit to Kanas to better identify to what extent the culture of
minority peoples was being identified as an important contributor
to way visitors evaluated their visit experience. As noted, the data
were first examined for non-response patterns and these were
found to be random and relatively few in number. In assessing the
determinants of satisfaction, the software package, SPSS, permits
three ways of conducting linear regression, in one of which the use
of the mean item score to replace the missing response permits the
whole of the sample to be used. An alternative method is ‘listwise’
which means the computation is undertaken only for those
respondents who completed all the variables used in the calculation.
The final version, ‘pairwise’ will mean statistics are only calculated where a pair of completed variables exist. Enders (2010)
does not favour imputing mean scores, arguing that such a practice
adversely effects the correlation between variables. One of his
suggestions is to use the expectation maximisation algorithm
within missing likelihood estimation. Another is to undertake
stochastic regression to calculate an option for the missing data. A
series of comparisons were run and in practice many of the mean
scores differed little from imputed mean scores.