Travel Reference
In-Depth Information
Discrete Choice Models (2 per cent) were not considered in the 1988-2008 survey unless
they explicitly dealt with tourists' or managers' individual choice alternatives and decisions.
Choice Models on individual level were addressed as early as 1993 in a purely theoretical paper
(Wie and Choy 1993) and in an empirical study on modal choice (Winzar, Pidcock and Johnson
1993). After a fi rst empirical application to modelling destination choice (Morley 1994) it took
tourism research quite a while to adopt random coeffi cient specifi cations of choice models
(Nicolau and Más 2005).
Despite their long tradition and their unmistakable merits in handling the qualitative (less
than interval-scaled) data that abound in tourist behaviour research some methods are extremely
sparse. Examples are Latent Class Analysis (LCA), Canonical Correlation, Log-Linear Models or
the Analytical Hierarchy Process (AHP, introduced by Thomas Saaty in 1977). Saaty's eigenvector
method got mentioned in an
Annals
paper (Calantone and Mazanec 1991) but had been
practically applied by the Austrian National Tourism Organization as early as 1984 (with a little
help from their friends in academia).
Applications of Expert Systems are rare. After the early example of an application to
computer-assisted travel counselling (Hruschka and Mazanec 1990) there was a fairly long
period of silence until the advent of trip recommender systems became fashionable. Other
examples of neglected fi elds with a promising problem-solving potential are Fuzzy Set Theory
or Classifi cation and Decision Trees.
Tourism marketing research was fairly quick in embracing Neural Networks. Among the
early adopters Mazanec (1992) and Pattie and Snyder (1996) used a feedforward network with
backpropagation learning to classify tourists into market segments and to forecast visiting
behaviour. Other neural network architectures such as the Self-Organizing Map (SOM) and its
descendants in the fi eld of Vector Quantization (VQ) must be taken into account to portray the
full picture of neurocomputing methodology. Genetic Algorithms (GA) were not quite as prolifi c
though their 'tourism' history began no later than 1997 when Taplin and Qiu decided to use GA
methodology for estimating the parameters of a route choice model. A little later Hurley,
Moutinho and Witt (1998) published a 'stand-alone' genetic application.
A method such as (non-naïve) Meta-Analysis is still under-utilized in tourism marketing
research. An early example such as Geoff Crouch's (1995) analysis of tourism demand models
would have deserved a larger number of followers. Several other methods were subject to very
late detection by tourism marketing. Social and Semantic Network Analysis is one of them and
does not yet enjoy a long tradition of tourism marketing applications. Interestingly, the fi rst
example found in the 1988-2008 period did not analyze sociometric or semantic data but
tourists' drive patterns (Shih 2006). Shortly after, the 'ordinary' applications to social or semantic
networks follow suit (Pan and Fesenmaier 2006).
Judging the frequency of usage over time the number of SEMs, and to a lesser extent, of CFAs
has risen exponentially. These two categories will be examined more closely to verify whether
the enormous gain in popularity also implies qualitative improvement and has contributed to
advancement in theory building. CFA, besides its natural role within SEM, has also been
instrumental as a routine method in scale construction and validation. Both aspects will be
considered in a separate evaluation section.
Clustering methods show a persistent record of applications over the years. This prompts the
question whether there is a discernible improvement of how the various decisions an analyst has
to make during a cluster analysis are justifi ed and substantiated. Hierarchical and partitioning
methods will be explored later.
The lower than average usage frequency of Choice Modelling does not correspond to
the success story apparent in core marketing. However, choice models tend to appear in
