The recruitment includes 3,420 subjects via Amazon Mechanical Turk (‘mTurk”). Upon agreeing to participate to the study (described simply as a “computerized questionnaire”) and to receive $.75 upon its completion (the tasks altogether took about 5 minutes, making the implied hourly slightly above the mTurk average for North America), the subjects are randomly assigned to one of the following conditions:
• Control: subjects in this experimental condition are first asked to express an opinion about their support for legalizing payments for organ donors and their families. Then, they are administered a survey with questions about their demographics and socio-economic characteristics.
• Organ text: individuals in this treatment condition are first informed that a text will be administered to them, and a comprehension question on its content will be asked afterwards. The text reports information about the extent of organ supply shortage in the United States, and described a number of proposals that have been advanced to reduce such shortage, with references to the academic studies advancing these proposals. They include the implemented kidney exchange programs as well as studies on the effect of introducing monetary compensations of donors. After having read the text, a comprehension question is given to the subjects.
• “Placebo” text: this condition has the same structure as the organ text treatment; however, instead of a text on organ supply shortage, a “morally neutral” text, concerning the causes of the flu and remedies for it, as well as a related comprehension question, is provided to the subjects.
We employ the “Item Count Technique” to elicit support rates for organ payment. The ITC is based on not asking a question directly (e.g., “Would you support the implementation for regulated payments for organ donors or their families?”); instead, respondents are shown a set of statements, and are then requested to state how many apply to them. The control group is given a list of N “neutral” statements (i.e., non-sensitive in nature and not related to our topic of interest), whereas the treatment group is given N+1 sentences, of which N are the same as for the control group, and the additional item is the one of interest for the researcher. Thus the researcher cannot infer if a given respondent answered positively or negatively to a given item; only the total number of items that apply to an individual is identified. This preserves the privacy of the respondents and, together with the anonymity of the online survey, allays the concern that they might give what they perceive to be the “socially correct” answer. In our case, the hypothetical framework might lead to a downward bias if most respondents believe that paying for organs is generally considered morally wrong. With subjects assigned randomly to various experimental conditions, the choice of statements that are not perfectly correlated (so that most individuals do not agree with all of them or none of them, thus effectively revealing their opinion on each signal statement), and a large enough sample size, the difference in the average counts between treatment and control gives an estimate of the share of individuals in the population under study to which the phrase of interest applies. The key question in our survey asks the respondents to indicate how many of the listed statements apply to them. In the control condition, four statements were reported, and in the treatment conditions, the fifth statement is a sentence indicating that the respondent would support the establishment of a regulated system of payment for organs. Thus within each of the treatment conditions described above, subjects are further randomly divided in two subgroups, one receiving four statement, and one receiving five statements.
A similar design (but without a placebo group) is also implemented for two further activities: legalizing indoor prostitution, and legalizing slavery.