Introduction

The goal of this Choice-Based Conjoint analysis is to get a better grasp of the regional market and travellers’ buying decisions when it comes to booking flight tickets.

Attributes and levels

In this specific analysis, I have retained 4 principal attributes with respectively 4, 3, 3 and 2 levels, which are depicted in the table below.



Here is how it works:

• A set of 3 different profiles is shown to the respondents, who will have to pick one of the three options displayed (click on the Survey tab above to see how it looks).
• The same process is repeated 12 times with different profiles in order to detect the buyer’s preferences.

Keep in mind that if the analysis is too long, the respondents will get bored and hurry, which could potentially have a negative effect on the quality and accuracy of the data gathered. That is why I have chosen 12 iterations, which proved to be a good balance.

Creating profiles

According to the book “Multivariate Data Analysis, 7th edition” in the chapter Conjoint Analysis, no more than 6 attributes should be used for a Choice-Based Conjoint Analysis and a minimum number of profiles is also required. As a matter of fact, 72 profiles are possible in this example if I were to use a full factorial design. That is 4*3*3*2 levels. As we cannot require respondents to go through all of them, a solution exists which consists in selecting only a subset of profiles. The method is called “fractional factorial design” and has to respect 2 important aspects:

• Orthogonality: avoiding correlations between levels of an attribute
• balanced design : the attribute’s levels appear the same number of time

Here is the formula with regard to the minimum number of profiles that have to be defined taken from the book aforementioned:

• Minimum # of profiles = Total # of levels across all factors - # of factors + 1

In this case, it means that I need at least 9 profiles. In order to generate them and the 12 iterations of 3 profiles, I have used an add-in for Microsoft Excel called XLSTAT (some other alternatives such as Sawtooth Software can do the job as well).

Sample of ~100 people

After having collected the data from a sample of ~100 people, I ran various analysis to understand which attributes were the most relevant for the overall sample and subgroups based on demographic variables (age, activity, etc.). .

Results & Conclusions

As we can see in those charts, the most important attribute is the airport of departure whereas the number of stopover (0 or 1 in this case) is the less considered here.

That said, we can see that for the subgroup of people above 30 years old, the airline is the less important factor.

Surprisingly, the most important attribute for the subgroup “students” is the airport of departure and not the airfare as I thought it would be.

The extremely low popularity of Turkish airline against Swiss, Singapore and Emirates also came as a surprise. The company recently spent huge amount of money in various advertising campaigns to market the airline as one of best in the world. Nonetheless, it seems that Swiss people are still reluctant about flying with this fast-growing airline.