A general treatment of non-response data from choice experiments using logit models

A new approach is developed for the treatment of 'Don't Know' (DK) responses, within Choice Experiments. A DK option is motivated by the need to allow respondents the opportunity to express uncertainty. Our model explains a DK using an entropy measure of the similarity between options given to respondents within the Choice Experiment. We illustrate our model by applying it to a Choice Experiment examining consumer preferences for nutrient contents in food. We find that similarity between options in a given choice set does explain the tendency for respondents to report DK.