## Tuesday, 6 September 2011

### BA@SIBMB: Conjoint Analysis

Differences between Conjoint Analysis and Perceptual Mapping

The table below summarizes the differences in inputs, outputs, and assumptions between individual models of choice reflected in the two systems - perceptual mapping and conjoint analysis.

Unique to Perceptual Mapping Shared Unique to Conjoint Analysis

1) Identification of ideal attribute levels Perceptions of products on attributes Ranking of all attribute levels, Trade off among profiles

2) General importance of attribute Importance of best vs worst attribute levels

Outputs

3) Perceptual map containing products Simulation of product choices Partworth values of each level of each attribute

& ideal points

Assumptions

Utility is symmetric around attribute level maps into NO constraints on form of partworth

ideal point unique utility to do it agin[que Attributes weights are modified by conjoint judgement

Attribute weights are modified Attribute weights are independent

by principal component: The level on one attribute doesnot change the

More correlated attributes get more weight utility of another.

Despite strong conceptual difference, both importance measures correlate very highly in practice. A big difference between the systems is the way they assess the utility of each attribute level. Perceptual mapping directly assesses the ideal level of each attribute and measures utility as a weighted deviation from that ideal. Conjoint analysis, by contrast, asks respondents to evaluate profiles or product descriptions and uses these judgments to infer the values of the attribute levels.

Conjoint analysis cannot directly produce maps but generates partworth functions that allow the analyst to visualize how much value an individual or a segment attaches to various attribute levels.

While conjoint analysis doesn’t product spaces, the information is there and, with commonly available discriminant analysis software, one could produce perceptual maps. Further, ideal points or vectors reflecting the partworth utility functions could be positioned in this space. Thus the input to most conjoint analyses can be used to produce perceptual spaces. For its part, the information in perceptual mapping can generate individual or aggregate partworth functions. Although, it is considered next, each of these will be in the inverted “U” shape rather than the unconstrained form of the conjoint partworth functions. Thus, although the two techniques do not differ critically with respect to their inputs or outputs, they do differ in important ways in their assumptions—and it is these assumptions that are likely to make a difference in the predictive power of the models.

Sampath Kumar Reddy,

Operations3