Fairness preferences in the face of limited information

2017-09-20

2017-09-29

Our main outcome variable of interest is the degree of inequality implemented by the spectators, deciding how to distribute points earned by pairs of workers via Amazon Mturk. (See the uploaded preplan for more details)

The experiment will consist of two parts. In the first part workers recruited via Amazon Mturk will work on a real effort task. In the second part, spectators recruited from the first years student population at the NHH Norwegian School of Economics will decide on how certain amounts of bonus payments will be distributed over two randomly paired Mturk workers from the first part of the experiment. The first part: Real effort task conducted by Mturk workers. Workers will be recruited via the Amazon Mturk account of the Choice Lab research group. The task the workers will perform is a qualtrix implementation of the ‘encription task’ introduced for experiments by Benndorf, V., Rau, H.A., and Sölch, C. (2014). In this task workers are shown three letters and at the bottom of the screen all letters of the alphabet, in random order, with below each letter a random number between 100 and 1000. For the three letters shown, the workers must fill out the corresponding number. After this, they can continue to the next screen where they will receive a new three letters. Workers will have exactly 15 minutes to work on the task. Workers will earn points depending on how many times they correctly filled out the three codes on a screen. Next to this, for each worker individually, an integer ‘random factor’ will be drawn from a normal distribution with mean 0 and standard deviation of 20. The sum of the worker’s task performance and the random factor will add to be the ‘production’ that each Mturk worker will bring to the second part of the experiment. Each Mturker who finishes at least 20 will earn 2 dollar as a fixed fee and workers earn a bonus depending on a number of factors being; their performance and random factor, a random other worker’s performance and random factor and the distribution decision made by a spectator. The Mturk workers are informed about this procedure.

The second part: Distributional choices by spectators. The second part of the design will be conducted at the Norwegian School of economics on the 27th 28th and 29th of September. Students from the first year course “Business Ethics” will be recruited via the course.All first year students at the NHH are expected to follow the Business Ethics course and participation in the experiment is made a compulsory part of the course. They will be instructed that the experiment results will be used for actual research purposes, that they will receive a payment for participating in the experiment, which will depend in part on their decisions, and the experiment will take place in a separate time and location compared to the classes for the course. The participants will be asked to perform a number of tasks.

Experiment sequence

• Participants will read instructions and answer a comprehension questionnaire

• Participants will answer incentivized questions regarding two Bayesian updating scenario’s, a ‘disease test’ scenario and the ‘taxi cab color’ scenario. We expect participants updating with a high degree of 'base-rate neglect' to show a quantitatively (and possibly) qualitatively different treatment effect. This task will pre-classify participants along the base-rate weighting dimension.

• Participants are then informed about the encryption task Mturk workers performed in the week before the experiment. The participants will be shown a histogram distribution of the amount of correct encriptions the different workers managed to complete in the given time.

• Participants will then receive information on the productions (in the each treatment) and performances and random factors (in the Full Information treatment) regarding a set of 10 randomly chosen pair of Mturk workers from the first part of the experiment. For each pair, the participants will be asked to make a distributional decision, splitting the total production points over the two Mturk workers.

• Participants in the Limited Information treatment are then elicited their beliefs regarding the individual performances of the 10 pairs they made a decision for.

• After this participants will make a number of further, non-incentivized choices that will be used for purposes of the course.

• The Participants will fill out a post-experimental questionnaire which will include questions on background variables (age, gender, political preferences). After this participants will be payed out and leave the experiment.

We plan to recruit around 800 workers on Mturk to work on the task to match with the approximately 450 expected participants for the spectator decisions.

Randomisation of the 'random factor' in the Mturk point production is done via the computer. Matching of Mturk worker pairs is done via computer randomisation.
Treatment randomisation is done via the fact that spectator participants when entering the lab will randomly draw a table number.

Random factor (number drawn from a discrete normal distribution)
Mturk to Spectator matching. The unit here is a list of Mturk pairs per Spectator
Treatment unit (2 treatments)

No

No clusters

We expect between 400 and 500 spectator decisionmakers

2 treatments, between 200 and 250 spectator decisionmakers per treatment

see uploaded pre-analysis plan for the specific details. Using a simulation apporach based on results form previous expeiments and the decision model from Cappelen et al (2007,AER) we calculate a power of 95% assuming 450 decisionmaking Spectators and sufficient variance in Mturk real effort task performance and drawn random factor.

Pre-analysis plan Fairness and Limited Information September 2017