Unless you’ve walked a mile in their shoes---The effect of role switching on experimental subjects’ behavior

When an experimenter uses a game where different subjects assume different roles, she needs to pick and choose different elements of experimental design, one of which is the protocol for a random match. In particular, she has to decide whether to fix the roles throughout the experiment or reassign them every round. Under the fixed-roles protocol, a player continues to play the role she assigned at the beginning of the experiment, regardless of whether the groups are rematched or not. Under the changing-roles protocol, subjects experience different roles, and this experience can potentially have a significant effect on subjects' behavior. In this project, we investigate whether fixing or changing roles influences subjects' ability for hypothetical thinking.

2021-11-22

2021-12-31

1. The distribution of strategy profiles under four different treatments.
2. The frequencies of contributions (conditional on histories).

1. In each round of the game, there are 8 possible combinations of actions, such as everyone contributes or no one contributes, etc.
2. For the first-mover, this is simply the average contribution rate. For other players in the perfect information environment, this is the average contribution rate conditional on the history (the actions of all previous movers). For the followers in the imperfect information environment, this is the average contribution rate conditional on observing one contribution or not.

Players interact with each other in a sequential public goods game where each player decides to contribute or not in an exogenously designated order. This experiment consists of four treatments:

1. Fixed-role, perfect information (FP): At the beginning of the experiment, each player will be randomly assigned an order to decide (1st, 2nd or 3rd). This order will be fixed through the experiment. Each player will observe all previous movers' actions perfectly.

2. Changing-role, perfect information (CP): The only difference between this treatment and the FP treatment is that the order to decide will be randomly assigned at the beginning of each round.

3. Fixed-role, imperfect information (FI): In this treatment, players will know whether they are the first-movers or the followers. The followers cannot tell if they are the second or the third mover. Yet, the followers can observe the action made by the player immediately before him. At the beginning of the experiment, players will be randomly assigned as a first-mover or a follower. This role will remain fixed through the experiment.

4. Changing-role, imperfect information (CI): The only difference between this treatment and the FI treatment is that the role (first-mover vs. follower) will be randomly assigned at the beginning of each round.

The experiment consists of four parts. Prior to each part, the experimenter will direct the subjects to read the instructions and solve problems to test their understanding. Part 1 also contains an unincentivized trial round of the game.

Part 1 is the primary focus of our experiment. The subjects will be assigned to one of the four treatment groups and play the simple public good game introduced previously. Specifically, in any given round, each subject will receive one token and make a choice between investing it in a public project or keeping it in their private account. The total contribution to the public project is doubled and divided equally between the three players. We provide full feedback to all subjects on their payoffs and the actions taken by each player after each round.

The game each group plays will vary in information setting (perfect or imperfect) and the role-assignment protocol (fixed or changing-role). Under perfect information, subjects observe all previous contribution, while under imperfect information they learn the most recent contribution only. Note that, under imperfect information, there are essentially two asymmetric roles, the first mover and the follower. We will randomly reassign the order between Players 2 and 3 in the latter role in each round to prevent the players from learning their exact order.

Regardless of the treatment, the game is to be played sequentially for 60 rounds. We expect each round to take no longer than 30 seconds. The time will not be limited, but after 20 seconds, the screen will start flashing a "make your decision" sign to draw distracted subjects’ attention and avoid prolonging the experiment unnecessarily.

To limit reputation building, we reshuffle teams in each round. We emphasize this design feature in the instructions as it is crucial to maintaining the structure of a one-shot game and precluding reputation building. We first reshuffle teams within the same role in random role treatments and then reassign roles within each new team to keep the probability of playing with previous teammates again constant across all treatments.

The first 59 rounds use the direct response method to elicit decisions, with a caveat that we asked the followers under the imperfect information (Treatments 3 and 4) to construct a contingency plan for their role to preclude them from deducing their exact order from the timing of their play. In the last round, we use the strategy method for all subjects.

Parts 2 to 4 contain auxiliary measurement tasks. These tasks include the Cognitive Reflection Test (CRT), the Farsightedness Task and the Monty Hall problem. We introduce the latter two specifically because this game requires backward induction and, under imperfect information, Bayesian inference.

Randomization done by computer.

At the beginning of each treatment, players will be randomly assigned into one of the three clusters. Every cluster is independent of each other. There are 60 rounds in each treatment. In every round of every treatment, three players will be randomly assigned as a group. Under the fixed-role protocol, every player's position will remain fixed in all rounds. Under the changing-role protocol, each player's position will be randomly assigned at the beginning of each round.

Yes

12 clusters in total (3 clusters in each treatment). We plan to recruit 144 individuals. In each treatment, there will be 36 subjects who will be randomly assigned into one of the three clusters (12 subjects in each cluster). Every cluster is independent of each other in each treatment.

144 individuals

There are 4 treatments in total. Each treatment has 36 subjects. Each subject will participate in 60 rounds.