Purpose: Knowledge about decision-making processes is likely to provide important insights for the development of multi-attribute health state descriptions. Moreover, taking account of applied heuristics while modeling valuation data is assumed to improve fit. However, little is known about how to incorporate data regarding respondents' heuristics into valuation studies. We therefore conducted a valuation study in which we mapped applied heuristics and explored simple weighted regression techniques.

Methods: Pain profiles, containing 7 attributes, were constructed based on real pain data. Valuations for these profiles were obtained by means of a Time Trade Off technique. Each respondent valuated 12 unique profiles. At the end of the valuation task, respondents were asked to indicate which attributes they had been taking into consideration and to rank them, at which it was allowed to give several attributes equal rankings. Applied heuristics were mapped and the aggregated data were modeled using a Tobit model without as well as with taking account of these heuristics (i.e. converting the values for the attributes that had not been taken into account into 0). Full models were described and log likelihoods of the models were compared.

Results: In total, valuations for 814 unique pain profiles were obtained from 68 respondents. The majority of respondents applied a unique ranking of attributes (78%). The number of attributes ignored varied from 6 out of 7 (6% of respondents) to 0 (18%). Most respondents had taken 3 out of 7 attributes into account (32%). Contrary to our expectations, overall fit of the weighted model did not improve (log likelihood of –4318 versus –4272 for the unweighted model).

Conclusions: Heuristics applied while valuating pain profiles as indicated by respondents are very diverse. Assuming that respondents are well able to indicate which attributes they had ignored during the valuation task, the results imply that simple weighted regression techniques can not be used if one wants to take account of applied heuristics. Non-linear and/or Bayesian modeling approaches will have to be explored to see whether these will improve results.