Subjects with rheumatoid arthritis (RA) enrolled in an online panel were asked to evaluate pairs of treatment alternatives with different attributes. Half of the sample saw a cheap-talk text. Preference parameters were estimated using random-parameters logit models to account for unobserved taste heterogeneity. The models also were estimated in willingness-to-pay (WTP) space instead of conventional utility space. Cheap talk not only affected the coefficient on the cost attribute, but also preferences for other attributes. WTP estimates were generally lower in cheap talk sample, except for the most important attribute and a 2-level attribute. Subjects who were presented with cheap talk discriminated between the adjoning attribute levels better than the subjects in the control sample.