MaxDiff, also known as best-worst scaling, is an approach for obtaining preference/importance scores for multiple items. Respondents are typically shown 2-6 items at a time (ex. ice cream flavors - chocolate, strawberry, blue moon, superman, cookie dough) and asked to indicate which is best and which is worst. The task is repeated many times, showing a different set of items in each task. Common applications include message testing, brand preference, customer satisfaction, or product features. The result is an ordered list with interval scaled utility scores for each item. These scores tell us how much unique value each item garners and can be transformed into ratio-scaled probabilities so that a utility score of 4 for an item is 2x as great as a utility score of 2.
MaxDiff is roughly comparable to a one-attribute, multi-level Choice-Based Conjoint (CBC) exercise. So once you learn MaxDiff, you are well on your way to learning Conjoint Analysis!
In conjoint analysis, we add more dimensions to the experiment, describing the product/ service with multiple attributes. So in our ice cream example, we would now have brand and price, in addition to flavor, to test. We vary the product features (independent variables) to build many product concepts. Then, we ask respondents to rate, rank or, most commonly, choose (depending on the type of conjoint analysis) which concept they prefer (dependent variable). Based on the respondents’ evaluations of the concepts, we can figure out how much unique value (utility) each feature adds to the product.
Common applications of conjoint analysis include designing new products, product line extensions, estimating brand equity, measuring price sensitivity (elasticity), and branding and packaging.
The biggest difference between conjoint analysis and MaxDiff is that in conjoint analysis, the rating or choice of a concept is based on the SUM TOTAL of its components (the items conjoined). It assumes an additive model, where the value of the overall product concept is equal to the sum of its parts. MaxDiff is not an additive model.
In terms of similarity, both techniques force trade-offs, which lead to greater discrimination among items. In addition, both techniques result in interval scaled utility scores for each item/ level tested, which can be transformed into ratio-scaled probabilities.