Travel Reference
In-Depth Information
(e.g.
Mplus
; Muthén and Muthén 2007) that allows for sound statistical modelling of ordered
categorical indicators.
Equivalent models
Very few applications consider equivalent models. Even when a hypothesized model fi ts the data
well, it is likely that the same variance-covariance matrix may be reproduced by a variety of
alternative models 'that are indistinguishable from the original model in terms of goodness-of-fi t
to sample data' (MacCallum
et al
. 1993: 185). Consequently, a satisfactory goodness of fi t does
not automatically prove that a model is correct. Therefore, a theory-guided approach comparing
competing and theoretically justifi ed models and cross-validating the fi ndings is essential.
Multiple items
Although assuming that multi-item constructs outstrip single items is true with regard to
measurement error (Steenkamp and Baumgartner 2000), it often urges the analyst to invent
largely redundant items for a variable that might easily be measured directly. An example of a
reasonably directly measured variable can be found in Oh (2003). The author operationalizes
price unfairness by the difference between the reference and the actual price.
Refl ective or formative indicators
In most SEM studies constructs are implicitly declared as refl ective, even though this may not be
theoretically sound in every instance (Diamantopoulos and Winklhofer 2001). Only recently
(Murphy, Olaru and Hofacker 2009), a research note has called for rigour regarding the
specifi cation of refl ective and formative constructs and authors have started to incorporate
formative constructs into their models, e.g. Song, van derVeen, Li and Chen (2011) who develop
a formative satisfaction index.
Alternatives to covariance-based modelling
One such alternative that has recently recaptured attention is PLS Path Modelling (PLSPM).
First applications in tourism marketing date back to the early 2000s and have experienced
growth in the last four to fi ve years. A recent simulation study demonstrates that, whereas
covariance-based SEM (CBSEM) outperforms PLS in terms of consistency, PLSPM is clearly
preferable for small sample sizes. Additionally, PLSPM is superior when it comes to statistical
power, which makes PLSPM attractive where the research goal is identifying rather than
confi rming relationships (Reinartz
et al
. 2009). PLSPM may be complemented by covariance-
based latent variable modelling when analyzing secondary data on tourism destination
competitiveness (Mazanec and Ring 2011).
Unobserved heterogeneity
Another often neglected issue in SEM is unobserved heterogeneity. The adjective 'unobserved'
points to lacking prior knowledge about the causes of heterogeneity forcing the analyst into a
data-driven method. Usually, authors fi rst segment respondents and then apply multi-group
analysis to test for differences between these segments (see e.g. Barroso Castro, Martín Armario
and Martín Ruiz (2007) for a combination of Latent Class and subsequent Path Analysis). In
