Travel Reference
In-Depth Information
are grouped based on their participation in each one of, say, ten vacation activities or benefi t
segmentation, where respondents have stated the importance of, say, seven key benefi ts people
may seek when going on a vacation and resulting segments include people who have similar sets
of benefi ts sought. The data-driven segmentation process requires a number of additional steps
as well as a number of additional methodological decisions to be made. As a consequence it is
often perceived as being 'more sophisticated'. The problem is that the greater number of steps
and methodological decisions can also lead to an increasing number of mistakes in the process.
Below is an outline of the key steps in this process including key methodological understanding
that is required and the type of decisions that are necessary.
Data-driven segmentation step #1: Choice of segmentation base
Just as management needs to decide which single criterion may be relevant to determine useful
segments, a decision about the set of questions to be used as a segmentation base in the data-
driven segmentation process is required before data is collected. This is critical to ensure that the
questions asked in the survey capture exactly what management believes are the key dimensions
by which tourists should be grouped.
Data-driven segmentation step #2: Data collection
This fi rst stage does not differ much from the commonsense approach. However, two key
decisions need to be made at this stage of the process which has implications later in the data
analysis step.
The fi rst is the number of questions (or items or variables) included which are intended to
be used as the segmentation base. So, for example, in case of a behavioural segmentation: how
many behaviours will be included? This is a critical decision because the number of variables that
can be used later in data analysis is not unlimited. A rule of thumb provided by Formann (1984)
in the context of latent class analysis, is that sample size needs to be 2
k
, where k indicates the
number of variables used in the segmentation base. So, according to Formann's rule, if the
number of behaviours included in the questionnaire is 15, the sample size required would be
higher than 32,768. At fi rst glance this appears to be a signifi cant restriction imposed on the data
analyst, but usually it only requires careful choice of survey questions, so rather than randomly
inserting a list of 30 behaviours in the questionnaire, managers should make considered decisions
as to which behaviours are actually relevant for the segmentation task at hand. If, indeed, it is
impossible to restrict the number of questions so as to be suitably low given the sample size, an
alternative is to select a subset of those questions (the most different ones) for the segmentation
analysis and then use the other questions when segments are described.
The second critical decision in the data collection stage is the choice of answer formats
(Dolnicar 2013). The vast majority of academic survey studies in tourism use so-called fi ve or
seven point Likert scales that ask respondents to indicate their level of agreement with a statement
in the questionnaire. This answer format is particularly tricky in the context of data-driven
market segmentation for two reasons:
1 It is prone to capturing response styles, such as respondents' tendencies to use the middle
('neither agree/nor disagree') or extreme options ('strongly agree', 'strongly disagree'). Such
tendencies contaminate the data set and can lead to artifi cial segments which are meaningless
in content; they actually just capture the response style. One such example is a segment
which has extremely high agreement with all statements.
