Choice Architecture and Transparency

The main intervention is an announcement that makes it salient to all participants that nudges are potentially used in the formulation of the next question. This intervention is common to Choice Architects and Decision Makers.

We are also interested in the effects of the nudges themselves. This intervention affects only Decision Makers. Each nudge is thus also an intervention. We have three versions of the question. First, we have a simple question design that deliberately lists the low target first. Second, we have a system 1 nudge that preselects the high target both in the wording and in the software. Third, we have a system 2 nudge that encourages participants to determine which choice is better in terms of risks and benefits before answering the question itself.

Third, to study transition paths, we give all those Choice Architects in the control a delayed transparency treatment by asking them again to choose a question formulation but for another Decision Maker, who this time reads the transparency announcement before seeing the question.

2020-08-23

2020-10-31

For the Choice Architects, the Primary Outcome variable is "architecture".
For the Decision Makers, the Primary Outcome is called "choice".

Primary Outcome "architecture" is the Choice Architect's choice of the nudge: simple, system 1 nudge or system 2 nudge.
Decision Makers' Primary Outcome "choice" records whether they choose a high or a low performance target.

For Choice Architects, we also collect s0_effectiveness, s1_effectiveness, s2_effectiveness, s0_morality, s1_morality, s2_morality.
For Decision Makers, we also collect time spent on the choice page, tasks_correct, tasks_attempted, belief_B, liking_B, satisfaction, affected

Choice Architects' effectiveness variables measure Choice Architects' beliefs: how many people will choose the high target with each nudge. Morality variables measure how manipulative the Choice Architects view the nudges to be. For Decision Makers, we collect how long each participant spends time on the choice page, how many tasks they do correctly in the effort task that the target applies to, and how many they attempt. In the post-experiment survey we also collect which target they believed to have helped the Choice Architect more, what their attitude towards the Choice Architect is, how satisfied they were with the choice of target, and if they felt affected by the question formulation.

The experiment is a simple Choice Architecture Game with nudges.

In the second experiment, the goal is to test the Decision Maker hypotheses only on women, and to test the Choice Architect decisions on a more experienced group of people, namely, PhD students or PhDs in Social Sciences.

In the first stage, the Choice Architect participants choose how the Decision Makers will later be nudged (or not nudged). The main treatment is to inform the Decision Makers about this stage. The control is a short announcement that "Depending on your choice this part, another participant, Player B, may also get some money." In the second stage, the Decision Makers choose either a low or a high target in the next round of a real effort task. The low target is easier to reach but offers a lower reward than the high target. If the chosen target is not acheived, the Decision Maker earns nothing for the part. The Decision Maker is matched to a Choice Architect in treatment and shown the version of the question that the Choice Architect chose earlier. If the Decision Maker chooses the high target, the Choice Architect earns a bonus.

Randomization is done with a computer

Individual

No

Only 1 cluster

190 Choice Architects 420 Decision Makers Second experiment: 90 Choice Architects 150-180 Decision Makers

Aimed sample size by treatment cell:
95 for Choice Architects (2 cells) (total control over this, easily achieved)
70 for each Decision Maker cell (6 cells), target is on average (full balance not achievable or even desirable, target is to get at least 40 participants per each of the 6 cells)

Second Experiment:
90 Choice Architects, 45 in each of the two cells
150-180 Decision Makers, such that we get at least 30 in each of the 5 interesting treatment cells (simple-transparency is the uninteresting one in the 6 cells). Full balance is not likely, hence there is some uncertainty about how many Decision Makers will be required.

The minimum detectable effect size for the CA experiment is 20 percentage points. With a two-sided test, alpha = 0.05, beta = 0.2, starting from 15 % take-up rate, we need 73 observations per treatment cell, which is entirely achievable. The minimum detectable effect size for the Decision Maker outcomes is also 20 percentage points. However, starting from a rate of 25 %, a one-sided test, alpha = 0.05, beta = 0.2, we need 70 observations per each treatment cell. Second experiment: The minimum detectable effect size for the CA experiment is 23 percentage points, starting from 15 %, a onesided test, alpha = 0.05, beta = 0.20, needing 45 observations per each cell. The minimum detectable effect for the Decision Makers is 27 percentage points. Starting from a rate of 10 %, a one-sided test, alpha = 0.05, beta = 0.2, we need 30 observations per cell.