The impact of peer pressure on effort and cheating: An experimental approach

2021-09-05

2021-10-30

NEW
We now consider the following treatments with cheating possibilities for the second mover (SM) (Treatment 1 is the treatment without cheating possibility).
T2: In the no-information treatment, the SM receives no information on the performance of the FM.
T3: In the information treatment, the SM receives information about the performance of the FM after the SM has finished the task but before the SM enters their outcome.
T4: In the message treatment, the FM can suggest what the SM should report. SM receives information about the actual performance of the FM.
T5: In the cheating treatment, the FM can misreport their outcome as well. Otherwise, T5 is identical to T2. The SM receives no information about the reported performance of the FM.
T6: In the cheating-information treatment, the FM can misreport their outcome as well. Otherwise, T6 is identical to T3. SM receives information about the reported performance of the FM after the SM has finished the task but before they enter their outcome.

In all treatments, our primary outcome is the report of SMs. We will compare SM reports among treatments, both with non-parametric statistics (Wilcoxon Rank Sum Test) and regression analysis. Specifically, we compare the SM behavior among treatments as follows:
Between treatment comparisons
Comparing the no-information treatment (T2) and the information treatment (T3) gives us the overall impact of information revelation on the degree of misreporting.
Comparing the message treatment to the information treatment reveals whether allowing the FM to send a message to the SM increases or decreases the overall degree of misreporting.
Comparing the cheating treatment (T5) and the cheating-information treatment (T6) gives us the overall impact of information revelation on the degree of SM misreporting in a setting where FMs can cheat as well.
Within treatment analyses
For the treatments with information revelation, our main independent variable of interest is the true performance (T3) or reported performance (T6) of the FM. In the message treatment T4, the other independent variable is the FM message.
In all regression, we also analyze gender effects and interact gender with the variables of interest. For gender, we analyze the impact of the SM’s own gender and the gender SMs are matched with on SM behavior.
In the message treatment T4, we run a regression of the FM report on the FM performance, gender, and there interaction.


OLD
Our primary outcome is always the regression coefficient for the FMs’ performance on the SMs’ performance. For all our analyses, we will run regressions with different combinations of covariates as robustness checks.
(i.a) Treatment 1: We run regressions of the SMs’ actual performance in Counting Letters as the dependent variable on the actual performance of FMs in the same task.
(i.b) Treatment 1: We run regressions of the SMs’ actual performance in Raven’s matrices as the dependent variable on the actual performance of the FMs in the same task.
(ii.a) Treatment 2: We run regressions of the SMs reported performance in Counting Letters as the dependent variable on the FMs’ actual performance in the same task
(ii.b) Treatment 2: We run regressions of the SMs reported performance in Raven’s matrices as the dependent variable on the FMs’ actual performance in the same task.

Between 5 September and 23 September, we collected data and as initially expected, our treatments had no effect. We tested only treatments in which we let participants solve Raven’s matrices. We so far have not collected data from a letter counting task. Performance differences are insignificant or significant as expected for those who were able to cheat in the task (participants in treatments where cheating was possible have higher average performance than participants where cheating was impossible).
These results confirm the results of the pilot that our current treatments will not lead to any significant results. We have decided to modify the approach in order to utilize our remaining funds more efficiently and to analyze more interesting behavioral insights. Our adjustments are as follows:
(i) We focus on the treatments with cheating, as the data shows that there will be no impact of the information revelation on effort.
(ii) For the cases with cheating, we add additional treatments that may potentially lead to more interesting results compared to our current treatments.
(iii) For budget reasons, we restrict attention to one real effort task. We have decided to take Raven’s matrices, as it seems reasonable to assume that, compared to a task in which participants count letters, subjects should intrinsically care more about their outcome, their self-image and about social image (reputation effects towards the FM).

(i) Our first secondary outcome is the comparison between FM performance and SM performance in treatment 1.
(ii) Our second secondary outcome is how demographics and personal attitudes influence the behavior. We will collect data on sex, age, education, personal income, and the traits risk tolerance and competitiveness.
(iii) Our third secondary outcome is the comparison between the impacts of the FMs performance in Counting Letters compared to Raven’s matrices.
(iv) Our fourth secondary outcome is the comparison of differences in actual/reported performance of SMs who had a same-sex or opposite-sex teammate.

The experiment will be conducted online and participants will be recruited using Prolific.
For the tedious task, we use a task for which participants count the occurrence of letters in senseless paragraphs. For the intellectually challenging task, we use Raven’s matrices. We adopt a between-subject design, i.e. each participant performs just one real effort task as either the First-Mover (FM) or Second-Mover (SM). We adopt a two-by-two design with the two variables “misreporting for SM possible/ misreporting for SM not possible” and “tedious task/intellectually challenging task”. In the treatment with misreporting possibility for SMs, SMs will learn about this possibility only after they have worked on the real effort task. This excludes that the misreporting possibility influences the actual performance.
We will run the experiments with the tedious and the intellectually challenging task separately. FMs will perform the task first, and their performance will be utilized for determining whether the average performance is larger with or without information on the partner’s performance. SMs will then be randomly assigned to FMs.
Details:
Subjects will have nine minutes to work on counting letters or to work on Raven’s matrices. The chance of receiving a fixed bonus payment of GBP 1.75 increases by 2 percentage points for each correctly counted letter or for each correctly solved matrix (depending on the assigned treatment). A pre-test showed that the expected payoff is then approximately the same for both tasks. Our payoff structure ensures that the marginal financial benefit from each correctly solved task is independent of the performance of the partner. This means that any impact of the FMs performance can be attributed to peer effects.

We perform the treatments for the challenging task (Raven's matrices) separately. Furthermore, FMs will perform the task before SMs. Still, our matching procedure ensures full random assignment as all participants receive an identical HIT when they decide to take part in the experiment.

Individual. No clustering.

No

0

1,040 individuals, thereof 520 first-movers and 520 second-movers We aim to collect 130 First-Movers (FMs) and 130 Second-Movers (SMs) for each treatment. Data collection takes approximately one week per treatment.

130 second-movers per treatment who are each matched sequentially with a first-mover. We should have a total of approximately 1,040 participants.