Questionnaire real estate is valuable. Sawtooth Software recommends that, for quality HB-MNL utility estimation, we ask enough choice sets with enough profiles each so that a respondent sees each level of each attribute at least six times. Using realistic artificial respondents with known utilities, we’ll measure the extent to which the HB estimation can recover the known respondent-level utilities when respondents see attribute levels 1-18 times. This will enable us to assess the extent to which parameter recovery degrades when respondents see each item fewer than 6 times and how much it improves when they see levels more than 6 times. Having quantified the quality of utilities at different levels of respondent effort we will be in a better position to make informed decisions about questionnaire length and cost. Moreover, we’ll have evidence about how many CBC questions we should ask each respondent.
