partworth

Partworth, in market research and decision modeling, refers to the estimated contribution of a specific attribute level to a consumer’s overall utility for a product or service. It is a coefficient in a statistical model used in conjoint analysis or discrete choice experiments to link attribute levels to the probability of option selection.

Part-worths are derived from experimental data through coding schemes such as dummy coding, effects coding, or effects-to-zero coding. The chosen coding determines how a base level is treated and whether part-worths are interpreted relative to that base or relative to an average. In many designs, the part-worths within an attribute are centered (sum to zero) so that the attribute’s overall impact reflects the relative desirability of its levels.

Interpretation centers on relative value. A positive part-worth indicates a level that increases utility relative to the reference, while a negative value suggests a less preferred level. The magnitude shows relative desirability, and the range of part-worths within an attribute (the difference between the highest and lowest level) is often used to compute attribute importance.

Estimation typically relies on respondents’ choices or preferences collected in surveys, using methods such as hierarchical Bayesian modeling, multinomial logit, or regression-based approaches. After estimation, part-worths can be summed across attributes to predict profile utility and, with a choice model, forecast market shares.

Limitations include dependence on the sample, study design, coding scheme, and model assumptions. Part-worths are not absolute utilities and are not directly comparable across studies with different scales or attribute coding. They may also be affected by interactions among attributes and the selection of attribute levels.