Can an incentive-compatibility mechanism reduce hypothetical bias in smokers’ behavior? Evidence from a discrete choice experiment

Discrete choice experiments (DCEs) are increasingly used to provide evidence for tobacco policy. Typically, researchers collect stated preference (SP) data in these DCEs. A perennial issue with SP data is hypothetical bias; that is, what individuals say they will do in experimental settings may not fully accord with their corresponding, real-world behaviors. Researchers have developed a number of techniques to reduce hypothetical bias in the experimental setting. These can be broadly categorized into ex-ante and ex-post techniques. Papers in tobacco have applied ex-post techniques in an attempt to mitigate hypothetical bias, but as yet, no research has applied ex-ante techniques in DCEs in tobacco. In this study, an experiment is conducted to examine the impact of an ex-ante technique on smokers’ choice behavior. One of the underlying causes of hypothetical bias is when individuals’ incentives are misaligned with the research’s goals and respondents are not motivated to give truthful answers. A solution is to adapt the experiment to realign incentives, i.e. make the experiment incentive-compatible. Here, the impact of an incentive-compatibility mechanism is tested in a tobacco-based DCE. Respondents were assigned to either a treatment arm or a control arm of the experiment. In the treatment arm, respondents were instructed that for one individual, one of their choices would be selected. For this choice, $100 of the product selected would be given to the chosen respondent. In the control, no such instructions were given. To examine the impact of the treatment, a series of analyses were conducted. First, product and attribute preferences were interacted with the treatment condition. In this case, no response to the treatment was detected. Second, the impact of the treatment on the scale of utility was examined. Here, the estimated scale of utility was higher among the treated, and was statistically significant (p<0.01). In other words, choices of those exposed to the treatment were more sensitive to changes in attributes than choices of those that were not incentivized. This is perhaps surprising since we would expect to see a reduction in scale if behaviors were more reflective of those in the real-world (respondents in DCEs typically tend to overstate responses to attributes). The magnitude of the response of scale to the treatment was small and so predictions of choices are not impacted sizably. This result holds across a range of model specifications. These results raise several questions. If it is the case that the treatment induces revealed preferences, then the DCE without incentives is yielding decisions that are close to real behaviors, though responses are slightly dampened. On the other hand, it may be the case that the incentive induced an alternative behavioral response than revealed preference. Further, the impact suggests that the incentive itself was fairly weak, thus strengthening the mechanism may affect behaviors differently. In any case, this study shows how to use incentive-compatibility mechanisms in tobacco and that incentive-compatibility mechanisms can be used to alter SP behaviors in DCEs in tobacco.