Travel Reference
In-Depth Information
more about these relevant attributes, it is necessary to quantify their importance, and the most
common approach used in tourism studies is a range of regression methods including simple
regression, multinomial logistic regression and conditional logistic regression. The following
section evaluates these approaches.
How important are these attributes during tourists' decision-making?
Regression analysis can provide more detail about the relative importance of each attribute
and how the value of total preference of destinations changes when any one of the relevant
attributes varies. The value of total preference can be indicated by the number of tourist arrivals
in the destination or by the assigned values of how much tourists prefer this destination.
Additionally, the relevant attributes can be derived from the hypothesis of researchers or previous
exploratory studies.
Different types of regression have different functions. If the interest is only on testing the
specifi c infl uence of a single attribute (e.g. price and climate) on the choice, a simple regression
can be used. The most common simple regression used in tourism destination choice studies is
linear regression which assumes that a change of the independent variable (the attribute) can
directly lead to the change of the dependent variable (the preference) and that the pattern
of change is linear. For example, if the independent variable is transport price and the
dependent variable is the number of annual arrivals of one destination, a simple linear regression
may be able to fi nd that transport price is inversely proportional to the annual arrivals and every
unit increase of the transport price will generate a 0.6 unit decrease of the number of annual
arrivals to this destination. Sometimes the infl uence of the attribute on the preference is not
linear but in a curve shape such as the temperature of the destination. The preference may start
to increase from a lower level of temperature until reaching the peak at a certain temperature
at which point it starts to decrease. In such situations, when linear regression is not suitable,
polynomial regression (e.g. quadric regression and cubic regression) can be used to explore an
infl uence relationship in any level of curvilinearity. And in this case, quadric regression is the
correct method for fi nding the ideal temperature that generates maximum preference.
However, due to the complexity of a destination as a product, it is rare that the fi nal destination
selected is only based on a single attribute. Therefore, simple regression is normally used to
confi rm the infl uence that a certain attribute plays during the decision-making. But in order
to gain a more comprehensive insight on the decision-making process, we may look into the
combined effect of a group of attributes together and hence a multi-regression approach is
required. Actually multi-regression is an extension of simple regression that incorporates two
or more independent variables in a prediction equation for a dependent variable. The study of
Sonmez and Graefe (1998) is an example that adopted both simple regression and multi-
regression techniques to test the effect of different demographic characteristics on risk perception
(multi-regression) and the infl uence of risk perception on the preference of foreign tourists
(simple regression). Other examples include the ordinary least square regression used to explore
the impact of personality on perceived destination values (Ekinci and Hosany 2006) and a multi-
regression of tourists visiting Australia (Crouch
et al
. 1992). In addition, signifi cance tests such as
ANOVA and T-test provide a way of measuring the quality of the fi ndings since they can indicate
to what extent the relationship found by the regression can be a product of mere coincidence.
Normally, regressions only deal with ratio data or at least ordinal data that can be regarded as
continuous variables. But in circumstances where the dependent variable is dichotomous or
categorical, general regressions are not enough. For instance, in situations where the research
seeks to investigate how perceived important attributes determine the fi nal choice of the tourists.
