Experimental Design Details
Raven-style IQ-test
The IQ-test is used to provide a signal about a DM’s intelligence to the CAs. To generate the signal, and to ensure that CAs can relate to it, both types of participants take part in a Raven-style IQ-Test, a nonverbal tool for measuring abstract reasoning (Bilker et al., 2012). DMs have to solve ten Raven matrices in ten minutes and are informed that this type of test is frequently used to measure intelligence, and that it correlates with educational outcomes and future earnings. To reduce mental strain on CAs while still ensuring they understand the intelligence test well, they are presented with three questions (to be solved in three minutes) and are informed that these three questions convey the overall difficulty of the test. Sample questions were selected to cover the range of difficulties in the test and ensure significant variation in test results . Both DMs and CAs can earn 0,50€ per correct answer on the test.
Decision Makers
After solving the IQ-test, DMs face two choice-sets in the domains of risky-choice and time preferences for a total of four. In each domain, one choice-set is presented in a straightforward manner (easy frame) while the second choice-set requires some calculations to understand the payoffs (difficult frame). The calculations are chosen to somewhat mimic real-life situations, i.e., as an investment decision in the domain of time-preferences and a base-rate signal in the domain of risky choice.
In the domain of risky-choice, DMs choose between a fixed payment and two lotteries. One lottery may be more attractive than the fixed payment for risk-averse participants (with an expected value above the fixed payment), while the other lottery can only be the most attractive for risk-loving participants. In the easy frame, participants are informed about both, potential payouts for the lotteries (“the value of a winning ticket”) and probabilities of winning. In the difficult frame, participants are informed about the potential payouts, but the probabilities of winning are obfuscated so that they appear more attractive than they actually are: They can be calculated by a Bayesian individual with the given information, but are likely to be misjudged by individuals suffering from base-rate neglect. Both frames in the domain of risky-choice build upon work of Nielsen and Rehbeck (2022) who study decision-makers who violate common choice axioms they committed to and constructed a version of the difficult frame to induce such violations.
In the domain of time preferences, participants choose between receiving a payment today and receiving a payment in 6 months (or a combination thereof). The options are constructed so that more patient choices lead to higher payoffs overall. Consequently, only participants who are present-biased would prefer to choose the less patient options. In the easy frame, participants are informed about the payoff today and the payoff in 6 months. The easy frame is adapted from Ambuehl et al. (2021) who demonstrate that a significant number of CAs make impatient options unavailable for the DMs in their study. In the difficult frame, the payoff in 6 months is presented as an investment decision yielding a high, certain, interest rate. Knowing the value of the patient option requires participants to apply some (straightforward) calculations.
DMs rank each of the options in all four choice-sets from the best to the worst option. DMs are informed that one of the four choices will be randomly selected with equal probability to be payoff-relevant. In this case, a DM will receive her preferred option out of all the options made available to her by the corresponding CA. To avoid consistency effects on the part of the CAs there are two distinct but structurally very similar versions of each choice-set that are randomly assigned. In particular CAs randomly received Version A of the easy frames and Version B of the difficult frames or Version B of the easy frames and Version A of the difficult frames. DMs received the version of frames randomly assigned to their respective CAs.
Choice Architects
CAs main task is to judge the options in the four decision situations described above and thereby construct choice-sets for a DM that will take part in a future experiment. CAs are informed that the DM comes from the same participant pool (participating in experiments at the CLER) and that their decisions can influence the payoffs of the DM. Importantly, CAs own payoffs are not affected by the intervention choices they make for the DM: all CAs are informed that they receive a fixed bonus of 5€ for completion of the experiment.
CAs are first presented the two decision situations of the easy frame, one by one in random order. For each of the options in each situation, CAs decide whether to make the option available or unavailable for the DM. To avoid default effects, no alternative is pre-selected but CAs have to actively decide about each option. CAs are told that there are no right or wrong answers and asked to make their decisions based on their own judgement. However, CAs must keep at least one of the three options within each decision situation available to the DM.
Before judging the options, CAs are informed about the decision procedure of the DMs. In particular, CAs learn that DMs will have to rank all three options in each of the decision situations and that their payoff will be as the highest-ranked option made available by the CA. Thereby, CAs cannot influence the DMs decision process and the complexity of their task but only their outcomes.
Besides making options available or unavailable, CA are also given the possibility to provide feedback to the DM that they “not recommend” certain options. In this case, the DM is informed that a participant of a previous part of the experiment (the CA) has thought about the option and decided to recommend against choosing it. This feature provides CAs with a tool to give advice to the DM without restricting his choice set, and thereby rules out that CAs decide to make options unavailable simply because they have no other means of providing feedback.
Before being presented with the difficult frames, CAs learn that, for the remaining two choice sets, they will be provided with additional information that the DM himself will not be able to see. Table 2 provides an overview of the decision-making environment of the CA using the example of the difficult frame of the risky-choice decision, including the additional information only visible for the CA.
After receiving instructions and answering questions of understanding, but before constructing the choice sets, CAs take the three sample questions of the intelligence-test and are asked to evaluate their own performance . First, they have to estimate how many questions they answered correctly (absolute performance). Second, CAs are informed that a large group of similar participants of a former experiment at the Cologne Laboratory of Economics Research have participated in the exact same test before. CAs are then asked whether they belief to belong to the top or bottom 50% of participants (relative performance). CAs earn an additional 0,50€ for each correct self-evaluation.
Immediately before constructing the choice-sets in the task described above, CAs are informed about the performance of the DM for whom they were to construct choice sets. By random assignment, half of the CAs receive information that the DM belongs to the bottom half of participants (lower rationality), the other half of CAs receive information that the DM belongs to the top half of participants (higher rationality).