Traditional Conjoint Design The goal of most conjoint studies is to determine ideal product configurations by analyzing the relationship of underlying product characteristics to respondents' preferences for differently-configured products. Research staff identify the presumably salient underlying attributes of a product or service, and the unique levels of each attribute, and the principles of experimental design are used to create an array of hypothetical products, each composed of a unique profile of the different attributes. Ideally, a fully-crossed factorial design would be used to test all possible combinations of the different product attributes. If, for example, there are five product attributes, each of which has two underlying levels, then a total of 2 to the fifth or 32 unique combinations exist. Practical considerations usually preclude using a fully-crossed factorial design, and a main-effects model is tested with a more efficient design such as a fractional factorial or a Graeco-Latin square. In the example of 5 product attributes, each with two levels, a very common design would involve the testing of a half-replicate fractional factorial with 16 unique product combinations tested. Once the unique product combinations are established, conjoint studies have typically collected data via the use of either:
This illustrates why some means of allowing the respondent to directly view the unique product configurations is required. With the ranking methodology, it is extremely difficult to accurately communicate over the telephone the nuances of 16 unique products, each composed of five (or possibly more) different characteristics, and keep each product differentiated from the other products. After listening to each of the sixteen product descriptions, it is then necessary for the respondent to go back and rank each product hopefully keeping straight their in mind what the different products are and their preferences for each. For the paired-comparisons model, a telephone survey is often difficult because of the amount of time required to go through each of the possible comparisons. Since there are a significant number of repeats in the presentation of the description for each option, the amount of time required to go through the entire process on the telephone is prohibitive. The conjoint analysis transposes the usual subject (row) by object (column) database into an object (row) by subject (column) database where the number of cases in the dataset equals the number of unique combinations evaluated by subjects as part of the conjoint exercise. The variables in this transposed dataset, assuming an individual-differences model that also estimates an overall average set of utilities, include: (1) one variable for each subject describing his/her ranking on that combination, (2) the average rank across subjects for that combination, and (3) one variable for each of the underlying attributes that describe the hypothetical product. |