Travel Reference
In-Depth Information
Conclusions
If tourism marketing research were deprived of Regression-Based models and Exploratory
Factor Analysis, about half of the applications of advanced analytical methods would vanish.
Heavy usage of regression and related methods is easily explained by convenient access to
secondary data describing international tourist fl ows, and demand modellers loving their favourite
pet of forecasting. With EFA the situation is different. The large majority of EFA, or actually
PCA, are fed with cross-sectional primary data. EFA/PCA typically serves as a precursor for later
confi rmatory analysis. As a deterministic method of tentative data reduction it is clearly focused
on defi nition and concept formation and thus signals an infant stage of theory development.
A similar diagnosis pertains to the highly popular method class of clustering procedures
(accumulating hierarchical and partitioning routines). Classifi cation results per se explain nothing
unless they become building blocks in subsequent cause-effect hypothesizing. Nevertheless,
classifi cation for its own sake remains attractive for tourism marketing, which shares problems
such as the 'curse of dimensionality' with most other applications of marketing segmentation.
Hence, new approaches such as Biclustering are highly welcome (Dolnicar et al . 2012).
Tourism marketing research has caught up with other social science disciplines as far as SEMs
are concerned. In relative terms, this method class has grown most rapidly and this exerts a ben-
efi cial infl uence. While the best-practice requirements of methodologically sound SEM are not
always strictly met, the quest for specifying cause-effect relationships on structural level strength-
ens explanatory power in the long run. Tourism marketing research has already begun to climb
next levels of SEM, e.g. accounting for unobserved heterogeneity and nonlinear relationships
(Mazanec 2007b). Discrete heterogeneity has been tapped with SEM-LCA models; managing
continuous heterogeneity with Bayes models and Markov-Chain-Monte-Carlo estimation are
being discovered.
Regarding Discrete Choice Modelling tourism marketing is lagging behind core marketing
where it has been an avenue for mainstream research for more than three decades, strongly
driven by an abundance of scanner panel data. Unfortunately, tourist panels are rare. On the
other hand, the repeated measurements collected with longitudinal data are necessary to
guarantee that an up-to-date specifi cation of an advanced choice model, say, a fi nite mixture
multinomial logit model, is identifi ed and can be safely estimated. At least, there are encouraging
developments such as the choice experiment of Chaminuka et al . (2012) or Huang and Pengs's
(2012) ingenious combination of Fuzzy Sets and Item Response Theory.
Several of the methods yet rarely encountered in tourism marketing research bear a promising
potential. A prominent example is nontrivial (i.e. based on Graph Theory) Social and Semantic
Network Analysis which offers a variety of application opportunities for marketing intelligence.
Information exchange networks emergent in Internet communities are a natural fi eld of
application becoming more widely acknowledged.
Some new analytical methods have received quick acceptance in tourism marketing, though
the absolute number of applications is still small. For example, the various methods of Vector
Quantization and Topology Representing Networks are likely to survive and expand as researchers
realize that they produce a meaningful partition where traditional clustering may detect nothing
(see the recent example of a quality of live segmentation by Dolnicar, Yanamandram and Cliff
(2012). Considering optimization procedures Data Envelopment Analysis comes to mind. When
economics re-discovered this method (originally developed in the 1970s) a couple of years ago it
became quickly adopted for effi ciency measurement of tourism businesses and destinations.
Method combinations are ambiguous. As indicated above, they may originate from obsolete
usage of methods. A striking example is data reduction with Principal Components and
Search WWH ::




Custom Search