Experimental Design and Conjoint Analysis

Executive Summary

One of the most important and useful tools in all of marketing research is the ability to conduct studies using conjoint analysis. Full profile conjoint analysis is a way of using experimental design and full/partial factorials to come up with a set of cards that represents a given product’s features and levels (profiles). A person will go through and rank these cards in order of preference, from which the researcher can later derive representational utility for each combination of features and levels. It is not hard to see that as you include more features or levels per feature the more possible combinations there are and the less feasible it becomes to have a person go through and rank the cards. To compensate for this, experimental design narrows the total number of possibilities (factorials) that would be on each card to a manageable set of cards between 8 and 16 in most cases. The model is limited in that the more you use fractional factorials, the less your predictive capability becomes. For example, if I have a product that has four features with four levels per feature the total number of possible combinations is 256 (4x4x4x4). The model will use partial factorials to narrow the required number of cards to 16 and gives the researcher predictive capability to estimate utility levels for any combination—even one that has not necessarily been represented on the cards.

The demand for marketers that are familiar with conjoint analysis is consistently high. Software designed to conduct the necessary computations and algorithms to accurately conduct conjoint studies sells for many thousands of dollars—thus, making it a very expensive tool. This past winter semester, Dr. Smith introduced us to a basic DOS based tool that assisted in setting up full-profile conjoint analysis. He mentioned that the tool was over 20 years old, yet due to the lack of “affordable” alternatives, it was all we could use. Due to my marketing ambitions and desire to learn how to conduct conjoint studies, I thought to make a VBA based solution for the problem. In the process, I recruited a statistical expert in Brent Taylor to assist me with the project. In continued research on the scope of this task, we were disappointed to learn that solutions currently available to help one through the experimental design approach were in excess of 10,000 lines of code. Due to the sheer size of this undertaking, we have simplified the model so that instead of outputting utility levels, we can simply use the cards to determine functional direction a firm should undertake when considering a new product offering, change to an existing product, or other line/brand extension. Dr. Smith, owner of Qualtrics, has already communicated interest in our research.

Write-Up

VBA-File