Friday, July 01, 2005

Conjoint Analysis

One of the most powerful and necessary tools available to a product manager or market researcher is conjoint analysis. Conjoint analysis helps you assess the relative appeal of various product features that would add to the cost of a product.

A recent project presented me with an ideal case for applying conjoint analysis. My company was doing market research for a client building condominiums at the northwest tip of downtown Austin. Floor plan mix (square footage and number of bedrooms of the various units), pool, concierge, finishes, storage space, etc. all contribute to the cost of building condominiums or to recurring maintenance fees.

To determine the relative appeal of each feature, I couldn't just ask prospective buyers to rate or rank them, since the appeal of each feature traded off against its cost. A buyer might have ranked a feature highly until he discovered how much it would cost. Furthermore, a buyer might have been willing to pay for each individual feature but might not have had the budget to buy all of them. Enter conjoint analysis.

I included in a survey two conjoint analysis questions. Each of them asked the respondent to choose among several pricing packages. Here is one of the questions:

"If you were to buy a condominium and had to select from the following configurations, which would you choose?"

a. 850 sq. ft., price $233,000, 1 bed/1.5 bath, stainless appliances, garage storage
b. 850 sq. ft., price $231,000, 1 bed/1.5 bath, standard appliances, garage storage
c. 1100 sq. ft., price $294,000, 2 bed/2 bath, stainless appliances, no garage storage
d. 1100 sq. ft., price $292,000, 2 bed/2 bath, standard appliances, no garage storage
e. 1400 sq. ft., price $384,000, 2 bed/2 bath, stainless appliances, garage storage
f. 1400 sq. ft., price $374,000, 2 bed/2 bath/study/2 story, standard appliances, no garage storage

After importing the responses to the conjoint analysis questions into Excel, I used the "Regression" statistical analysis tool to compute the relative appeal of each feature.

2 comments :

roger said...

In a conjoint analysis survey, you have to have at minimum as many questions in the survey as there are variables to choose from. Regression analysis from 2 inputs for the variables you are trying to solve for cannot predict consumer preference with any statistical significance. In general, you would want between 1.5 to 3 times as many questions as variables to achieve full saturation as well consistency in results.

rcauvin said...

As I understand it, the statistical significance of conjoint analysis lies not in the number of different questions, but in the number of different profiles. (Each multiple choice answer represents a profile.) You need to have more profiles than you have attributes.

One caveat is that a large number of attributes entails a large number of profiles. Presenting a large number of profiles and attributes to the user in a single question is confusing and burdensome to respondents.

Also, it does certainly help to cross-check the conclusions of conjoint analysis with other survey questions, and with the conclusions from qualitative research.