Conjoint analysis

Conjoint analysis estimates how people value the separate attributes that make up an offer. Directly asking what someone will pay often produces weak evidence because real choices require trade-offs. A buyer may prefer premium quality, immediate delivery and many features, yet accept less when the combined price exceeds the budget. Conjoint research simulates that decision by decomposing a product or service into attributes—such as quality, delivery speed or sealing method—and varying the levels of each. Respondents choose among bundled alternatives, allowing the analyst to infer which features create value, what combinations are preferred and how price affects demand. The findings can guide product design, portfolio structure, communication and pricing. A simple envelope study, for example, could vary colour, sealing method and the presence of a window rather than asking respondents to value “the envelope” as one indivisible object.

When to use it

• Use it to estimate the relative value of product components and identify attractive price–feature combinations.
• Apply it when comparing existing and proposed designs, as in the office carpet-tile example below.

Origins

The method grew from mathematical work on conjoint measurement. French economist Gérard Debreu addressed the representation of preference in 1960, and mathematical psychologist R. Duncan Luce and statistician John Tukey formalised conjoint measurement in 1964. Early theory dealt with two attributes; later researchers developed designs and estimation methods for larger sets. Commercial research software, including Sawtooth products, made the technique more accessible, but experimental design and statistical interpretation still require specialist competence.

What it is

Each concept combines one level from every selected attribute and normally includes a price. The simplified envelope illustration contains two concepts, three attributes and two levels for each. Further alternatives could hold most features constant while changing colour or replacing a traditional seal with self-seal.

Real studies commonly use as many as seven attributes with four or five levels, producing hundreds or thousands of theoretical combinations. Experimental-design software selects an efficient subset—often about 30 bundles—rather than showing every permutation. A participant might complete five or six screens, choosing or rejecting among four or five alternatives on each. Statistical estimation converts the observed choices into relative utility values for each attribute level.

In the envelope illustration, concept B has a total utility of 85. That figure is not a score out of 100; it is meaningful only in relation to the other estimated utilities. The detail indicates a preference for white over brown, suggesting that a white, self-seal envelope with a window might outperform both displayed concepts if its price remained acceptable.

Developments of the model

Early conjoint studies used printed concept cards that respondents sorted into preferred and rejected groups. Computer-based administration expanded during the 1980s, and most studies are now conducted online. A typical design can include 30 evaluated offers. Modern variants can learn about a respondent before or during the exercise and adapt later choices accordingly. Adaptive choice-based conjoint might therefore show a specialist buyer in a large decision-making unit only the relevant attributes, while presenting a small-company owner with the full purchasing decision. The method provides a disciplined estimate of preference, but its limitations must shape the design:

Sample size:

reliable estimation commonly requires at least 100 qualified respondents and preferably 200. A target of 200 is stronger, while a minimum of 100 may be the practical limit in some business-to-business populations.

Number of attributes:

including too many dimensions, with price among them, weakens the exercise. Limit each attribute to roughly three to five levels so the design and resulting comparisons remain manageable.

Respondent fatigue:

repeated screens of similar, differently priced offers can blur together. Confused or tired respondents may stop evaluating the trade-offs and select arbitrarily simply to finish.