Travel Reference
In-Depth Information
on any combinations of attributes, even those not included in the original test (Hair
et al
. 1998).
In tourism contexts, conjoint analysis has also been widely applied (e.g. Bernoulli 1954; Dellaert
et al
. 1995; Dellaert
et al
. 1997; Basala and Klenosky 2001; Suh and Gartner 2004). Most of these
studies use conjoint analysis to estimate the importance of different attributes in order to infer
tourists' choice of activity packages or destinations. For example, Suh and Gartner (2004) used
conjoint analysis to investigate the preferences of international urban travellers from Seoul, Korea
with the aim to identify the relationship between preferences and expenditures for the attributes
or activities.
Conjoint analysis can be used to test different models based on relationships between
consumers' preferences and the nature of the attributes, which include the vector model, the
ideal-point model and the part-worth function model. The vector model describes consumers'
monotone preference on some continuous attributes. The most preferred value of an attribute
is at infi nity such as durability or price, more or less is always regarded better by consumers.
The ideal-point model is also known as the quadratic model, which is used to illustrate some
attributes like temperature of an environment. Too hot or too cold are both dis-preferred. The
ideal amount preferred is always at the moderate level. Some attributes for which the preference
pattern on them is uncertain (such as some categorical attributes like the mode of travel), the
part-worth function model is more suitable since it only estimates the importance of specifi c
levels within an attribute rather than assuming that any preference shape exists (Orme 2005).
Generally speaking, the part-worth function model provides the greatest fl exibility in allowing
different shapes for the preference function along each of the attributes (Hawkins
et al
. 1989).
After deciding a certain type of test module, there are always three essential steps involved in
conjoint analysis, which are data collection, questionnaire design and estimation. The common
ways used in previous conjoint studies for every stage are summarized below.
There are two main ways to collect the data required by conjoint analysis: the two-factor-at-
a-time procedure and the full-profi le approach. The two-factor-at-a-time procedure asks
respondents to rank the various combinations of each pair of factor levels from most preferred
to least preferred (Johnson 1974). This procedure is simple to apply and reduces information
overload on the part of the respondent (Hawkins
et al
. 1989). But this decomposition method
eliminates the infl uence of other attributes and it is not able to mimic the real selection situation
as much as the full profi le approach since respondents are only comparing different combinations
of two factors rather than two products.
The full-profi le approach (also referred to as the concept evaluation task) utilizes the complete
set of factors including product profi les consisting of all relative important product features gen-
eralized by previous literature or investigations, which are presented to respondents. Although it
will never be perfectly full-profi led and even the omitted attributes may generate bias, this
approach gives a more realistic description of stimuli. Additionally, whilst the two-factor-at-a-time
procedure provides only a set of rank orders, the full profi le approach can employ either a rank
order or ratings. However, since respondents need to process the information on every attribute,
it might lead to problems of information overload. Under the circumstances of information over-
load, respondents might try to simplify this task by ignoring variations in the less important factors
or even refuse to respond. Therefore, the full-profi le procedure is generally confi ned to, at most,
fi ve or six factors in any specifi c sort (Hawkins
et al
. 1989; Gabbott and Hogg 1994).
In recent years, a choice-based approach was developed based on the traditional full-profi le
approach. However, unlike the traditional method, respondents are not required to rate or rank
each profi le directly. By using an online survey, respondents need to select one preferred stimuli
among a subset of stimulus until enough information is obtained for sorting all profi les. This new
technical way is more similar to what buyers actually do in the marketplace. And it allows
