Conjoint analysis basically utilises the fact that while human beings would love to have their cake and eat it too, they often settle for one or the other or for a little less of both because of money, time and other constraints. In other words, they trade-off one benefit against the other, depending on which is more important to them.
Let us take a light-hearted example. When looking for a spouse, a man may ideally want a wife who looks gorgeous, sings beautifully, and is wonderfully kind and considerate. As time goes by, he may decide that no such lady exists, or he may realise that such a lady, if she indeed exists, would not give him a second look. So, he would need to compromise on his ideal. He may decide that kindness matters a whole lot more than looks do, and that musical talent is also slightly more important than looks are. He may, therefore, get married to a lady who sings very well, and whose face hides a heart of gold. Or, he may get married to a lady who is nice looking but not gorgeous, and who cannot attract an audience with her singing but doesn't make the neighbours protest either, and who is kind and considerate to people, except when really riled. The trade-off is a highly individual decision and will vary from person to person.
Let us now take a more serious example. A car buyer could trade-off mileage for seating capacity, or power for mileage. This could be because no one has made a car that has a mileage of 40 kmpl, 2000 horsepower, and can seat a family of eight in comfort. Or, it could be because such a car does exist, but our car buyer cannot afford it.
Conjoint analysis basically tries to quantify this trade-off.
If the attributes and their levels were:
Mileage 10, 12, 14, 16, 18; Power 300, 400, 500, 600, 700; Seating capacity 4,5,6,7
In theory, there are a 100 possible product configurations possible from the above (mileage 14—power 400—seating 4; mileage 14—power 400—seating 5; and so forth).
Since we cannot research all the combinations, we use the conjoint analysis package to generate something called an ortho-plan. An ortho-plan selects a manageable number. It could be 18 or 20 or so; the software package decides that.
It is important that we use the ortho-plan package to generate this manageable number rather than doing it ourselves; if we just do it ourselves, analysis will be tough later.
The respondents are shown these configurations and asked to rate them in terms of preference and liking. That data is fed into the conjoint package.
These utilities can be used directly to predict a ‘liking' score for each of the theoretically possible 100 configurations.
The manufacturer can then decide where the cost of manufacturing and the ‘liking' score have an optimum fit, and that will most probably be the configuration to hit the market.
