Travel Reference
In-Depth Information
The details of the greedy algorithm and the mathematics behind the computer programming
are presented in the study of Yee
et al
. (2007). A simple example to explain how greedoid-based
dynamic programming works can be provided. Let's use the case of car purchasing again. Assume
the car buyers are using a lexicographic-by-aspect choice heuristic, which means the buyers will
select the cars with their most preferred attribute's aspect and then if there are ties, select the cars
with their second preferred attribute's aspect until they fi nd the fi nal choice. There are three
attributes with six aspects that are important for car buyers. These are the price (£18,000 and
£20,000), the colour of the car (red and black) and the brand of the car (Mercedes Benz
and Ford). There are eight combinations of the six aspects. And one respondent's preference
ranking on the eight possible combinations presented by stimuli cards is:
1
Price £18,000, Red colour, Benz
2
Price £18,000, Red colour, Ford
3
Price £20,000, Red colour, Benz
4
Price £20,000, Red colour, Ford
5
Price £18,000, Black colour, Benz
6
Price £20,000, Black colour, Benz
7
Price £18,000, Black colour, Ford
8
Price £20,000, Black colour, Ford.
By observing the preference ranking, it is possible to tell that this respondent uses a perfect
lexicographic-by-aspect choice heuristic, which means all the cars with red colour are put
forward before any other cars and then if there are ties, the ones which are Mercedes Benz are
ranked before other cars and then if there are still ties, the ones with lower price are ranked
before other cars. When the aspects are very small and the respondents are following a perfect
lexicographic heuristic, the lexicographic aspects order can be observed manually. But when
there are a relatively larger number of aspects and many respondents, it is too much work for
human analysis. The greedoid programme mimics the human observation analysis. Firstly, it starts
with one aspect and checks if all the cars with this aspect are ranked before other cars until it
fi nds the right aspect. And then the programme starts to check which aspect is the second
preferred aspect until the aspects order can sort all stimuli cards. In addition, most of the time,
the respondents are not following a perfect lexicographic heuristic, which means there is no one
lexicographic aspects order that can replicate the ranking exactly. In these cases, the greedoid
programme is able to fi nd the best-fi t aspects order that can replicate the closest ranking.
Although greedoid analysis is not able to provide the estimation of part-worth values of the
attributes, there are several advantages that make greedoid analysis a promising method to
estimate tourists' preference in destination decision-making. Firstly, it is a method that provides
a better insight of non-compensatory choice process by incorporating the principles of non-
compensatory factors rather than just adapting weighting schemes to imitate the output of
non-compensatory heuristics (Gabbott and Hogg 1994). When there are numerous alternative
destinations, tourists may tend to use a simplifi ed non-compensatory choice heuristic. Therefore,
the greedoid method can help us to explore the possibility of non-compensatory choice.
Secondly, compared with traditional conjoint analysis, the greedoid method requires less in terms
of respondent workload since it can deal with full-rank, consider-then-rank and rating tasks.
Moreover, the dynamic programming algorithm proposed by Yee (2007) substantially reduces
computation time and makes it feasible to identify the best lexicographic ordering for large
samples of respondents and moderately large numbers of aspects. Finally, greedoid analysis can
also identify the must-have aspect and the aspects that tourists used to eliminate the destinations.
