Page 199 - Dark Shades of Istria
P. 199
9.2 Additional Methodological Explanations
commemorative events, which means that all recognisable/representative
events were analysed; the inclusion of the entire population and the man-
ner of data collection indicate high external validity/generalisability. Ex-
ternal validity can also be evaluated through the academic relevance of
the topic, which is systematically described in the previous chapters from
a multi- and inter-disciplinary perspective. These descriptions demon-
strate that both the history and the present of Istria as a problematic re-
gion are interesting for academic investigation and thus this research up-
grades the previous academic findings. Hence, the construct validity de-
rives predominantly from the mixing of the memory and dark tourism
theory (although some other aspects were included as well), reinforced
by historical facts and contemporary Istrian memorial practices that re-
flect the regional reality (categorical variables were developed on these
bases). Internal or criterion validity (also statistical validity) was tested
with the nonparametric χ2 Test of Independence to determine whether
there is an association between categorical variables – see the following
sub-chapter.
The TwoStep cluster analysis with the log-likelihood measure in the
s p s s system was employed in order to handle categorical variables
(codes).⁶ The TwoStep cluster analysis can operate with mixed type at-
tributes, although such analyses can be problematic – different combina-
tions of the variables can impact the results, which is especially evident
when continuous variables are included; the sample size has nearly no
effect (Bacher et al., 2004, p. 13).
In addition, testing both nominal and binary variables produced prac-
tically the same results (Rezankova et al., 2006, pp. 611–612). Conse-
quently, in order to minimise the procedural impact on the number of
clusters, we decided to use only nominal and (symmetric and asymmet-
ric) binary variables. Both Akaike’s Information Criterion (aic) as well as
Bayes Information Criterion (bi c) were implemented for the automatic
clustering algorithm, which determines the number of clusters – we fol-
lowed the practice of Kayri (2007) and recommendations of Sarstedt and
Mooi (2014, pp. 299, 304). The degree of stability was tested this way as
well.
⁶ It can be used to handle categorical and continuous variables: nominal, ordinal and di-
chotomous (Rezankova et al., 2006; Rezankova, 2009, p. 220; Kayri, 2007; Sarstedt &
Mooi, 2014; Bacher et al., 2004). More can be found on the i bm Knowledge Center (i bm
Knowledge Center, n.d.).
199
commemorative events, which means that all recognisable/representative
events were analysed; the inclusion of the entire population and the man-
ner of data collection indicate high external validity/generalisability. Ex-
ternal validity can also be evaluated through the academic relevance of
the topic, which is systematically described in the previous chapters from
a multi- and inter-disciplinary perspective. These descriptions demon-
strate that both the history and the present of Istria as a problematic re-
gion are interesting for academic investigation and thus this research up-
grades the previous academic findings. Hence, the construct validity de-
rives predominantly from the mixing of the memory and dark tourism
theory (although some other aspects were included as well), reinforced
by historical facts and contemporary Istrian memorial practices that re-
flect the regional reality (categorical variables were developed on these
bases). Internal or criterion validity (also statistical validity) was tested
with the nonparametric χ2 Test of Independence to determine whether
there is an association between categorical variables – see the following
sub-chapter.
The TwoStep cluster analysis with the log-likelihood measure in the
s p s s system was employed in order to handle categorical variables
(codes).⁶ The TwoStep cluster analysis can operate with mixed type at-
tributes, although such analyses can be problematic – different combina-
tions of the variables can impact the results, which is especially evident
when continuous variables are included; the sample size has nearly no
effect (Bacher et al., 2004, p. 13).
In addition, testing both nominal and binary variables produced prac-
tically the same results (Rezankova et al., 2006, pp. 611–612). Conse-
quently, in order to minimise the procedural impact on the number of
clusters, we decided to use only nominal and (symmetric and asymmet-
ric) binary variables. Both Akaike’s Information Criterion (aic) as well as
Bayes Information Criterion (bi c) were implemented for the automatic
clustering algorithm, which determines the number of clusters – we fol-
lowed the practice of Kayri (2007) and recommendations of Sarstedt and
Mooi (2014, pp. 299, 304). The degree of stability was tested this way as
well.
⁶ It can be used to handle categorical and continuous variables: nominal, ordinal and di-
chotomous (Rezankova et al., 2006; Rezankova, 2009, p. 220; Kayri, 2007; Sarstedt &
Mooi, 2014; Bacher et al., 2004). More can be found on the i bm Knowledge Center (i bm
Knowledge Center, n.d.).
199
