Handbook
Gradient Handbook / Products / Conjoint Experiments / Conjoint Designs / Conjoint Vocabulary Edit this page

Conjoint Vocabulary

Vocabulary

Since people love to complicate simple things, there are a few conjoint related terms that have a bunch of synonyms that might confuse you. Below is a list of possible synonyms (in brackets) for conjoint terms (the ones we should use are in bold) with a short description.

  • Version [block]: a set of conjoint tasks that will be presented to a single respondent. The tasks within the version are optimized in a way that the combinations of attribute levels are unique and balanced (the levels of each attribute appear a more or less equal number of times). Usually, we have a smaller number of versions than we have respondents. As a consequence, we will show the same set of questions to more than one respondent.

  • Task [set, question, conjoint card]: a choice of alternatives we show to a respondent and ask to pick up one (or ask to attribute some values to each of the alternatives). A set of tasks (usually 8-12) makes a Version.

  • Alternative [concept, product, bundle, item, profile]: a composition of attribute levels. Usually, we show at least two competing alternatives within a task that differ by at least one attribute level. The competing alternatives make a task.

  • None alternative[constant base alternative, base alternative, opt-out]: an alternative for which we usually do not show attributes. WARNING: None alternative does not necessarily mean “I don’t like it”. Also alternatives Other or Other from brand X are technically None alternatives. Also if you are using Dual Response None (so the question if the respondent would actually buy the alternative is a followup) it should not be counted in the design.

  • Sample size: Sample size relates to both: the Overall number of respondents in the survey or the number of smallest subsample. This number affects mostly the standard errors of the model (the larger the sample size, the smaller the errors), so as good practice we can run the designs for both numbers to assess how well our model performs for the smaller and larger number.