Leadership in Dishonesty - Theory and Experiment

Online experiment. Between-participant design. The experiment consists of two parts. Each of the two parts is paid out with a probability of 50%.
Part I:
Participants are randomly assigned to groups of two. Both group members observe the same outcome of a lottery, which can be high or low. The group members report sequentially and receive the high prize if and only if both of them report HIGH. The second mover (SM) is informed about the report of the first mover (FM) before their own report. The outcome of the lottery is high with a probability of 25%. The high prize is £1.50 and the low prize is £0.05. The outcome of the lottery is observable to the experimenter. There are two treatments:
In treatment EXOGENOUS, participants are randomly assigned to the roles of FM or SM. For the participants’ reports, we apply the strategy method by asking participants for their report in case they are assigned to the role of FM and SM.
In treatment ENDOGENOUS, one randomly chosen participant (the “principal”) decides on the order of the reports. The other group member (the “agent”) is informed that the principal decided on the order of the two reports. Principals are first asked to choose between FM and SM, and then for their report. We apply a softer form of the strategy method for principals, as their answers for the role they know that they do not choose anyway are meaningless. However, if they decide to be SM, we ask them for their decision in case the agent as FM reported HIGH. For agents, however, we apply the strategy method by asking participants for their report in case they are assigned to the role of FM and SM.

Part II.
Participants are not paired. They take part in the same lottery where the outcome is high with a probability of 25%. Again, the high prize is £1.50 and the low prize is £0.05. The outcome of the lottery is observable to the experimenter. Each participant receives the high prize if and only if they report HIGH.

The study will be conducted on Prolific. We restrict the sample to participants whose country of residence is either the USA or UK, who have at least 500 approved assignments, and whose submissions have been approved in at least 95% of the cases. We impose these restrictions to ensure a high data quality.

After the actual experiment we ask participants for their willingness to take risks on a scale from 0-10, and for their beliefs about the behavior of other participants.

2024-06-10

2024-06-30

All our outcomes relate to the case when the outcome of the lottery is low. We test the following predictions and hypotheses. “Predictions” refer to Propositions from our model, while “Hypotheses” are explorative.
Prediction 1a: In treatment EXOGENOUS, the likelihood that participants report HIGH is lower for participants who reported LOW in Part II of the experiment.
Prediction 1b: In treatment ENDOGENOUS, the likelihood that participants report HIGH is lower for participants who reported LOW in Part II of the experiment.
Prediction 1c: Considering both treatments together, the likelihood that participants report HIGH is lower for participants who reported LOW in Part II of the experiment.
Prediction 2: In treatment EXOGENOUS, the likelihood that participants report HIGH as FM is lower than the likelihood that they report HIGH as SM after HIGH by FM.
Prediction 3a: In treatment ENDOGENOUS, the likelihood that a principal self-selects to FM is lower for participants who reported LOW in Part II of the experiment when we neglect principals who self-select to FM and report LOW.
Hypothesis 3b: In treatment ENDOGENOUS, the likelihood that a principal self-selects to FM is lower for participants who reported LOW in Part II of the experiment.
Prediction 4a: The likelihood of two HIGH reports is higher in treatment ENDOGENOUS than in EXOGENOUS when we neglect principals who self-select to FM and report LOW.
Hypothesis 4b: The likelihood of two HIGH reports is higher in treatment ENDOGENOUS than in EXOGENOUS. (Note: If H4b is supported by the data, this implies that H4a is supported as well. We nevertheless need both hypotheses, as it may be that only H4a is supported by the data).

We test our predictions and hypotheses both non-parametrically and with regression analyses. In regressions, we control for age, gender, self-reported willingness to take risks, and the answers to the relevant belief questions.

Impact of the aforementioned control variables.

See description of the intervention.

See description of the intervention.

Computer (online experiment)

Experimental session for the treatment and individual for the specific role

No

900

900

100 paricipants in treatment EXOGENOUS, 400 participants in treatment ENDOGENOUS PRINCIPALS, 400 participants in treatment ENDOGENOUS AGENTS

For our minimum number of observations, we focus on Hypothesis 4b, that is, on the comparison of two HIGH reports in EXOGENOUS and ENDOGENOUS. In a pilot study with 409 participants, we found that, in 21.6% of cases both group members reported HIGH in EXOGENOUS compared to 38.9% in ENDOGENOUS. Assuming we can replicate this effect, we would need observations from at least 485 groups according to a two-tailed Fisher exact test with 5% error probability, 90% power and a 4/1 allocation ratio of ENDOGENOUS/EXOGENOUS (as some of our predictions exclude observations from ENDOGENOUS and we do not know the number of principals selecting into FM/SM, we aim to collect more observations in this treatment). Rounding up, this translates to 100 participants in EXOGENOUS (forming 100 groups) and 800 participants in ENDOGENOUS (400 principals and 400 agents forming 400 groups).