The Effect of Self-awareness on Dishonesty (Wave 2)

We will recruit Amazon MTurk workers who will be randomized among multiple treatments. Each treatment requires the subject to write a description of a real-life event involving honesty or dishonesty which entailed different combinations of benefit and cost to themselves and/or others. There is also a control group who do not write descriptions. After this, they all face a series of tasks designed to reveal attitudes towards honesty. The structure of the basic design is almost identical to that in an earlier study (https://www.socialscienceregistry.org/trials/5142) however we will be varying the dishonesty task is a very precise way as described in the experimental design section below.

2020-07-22

2020-12-22

Honesty, measures of self-awareness (personality, integrity, fairness), other-regarding preferences, text.

The key honesty measure will be decisions made during the updated matrix puzzle game (that is described in the experimental design section and the included Analysis Plan in detail) in which subjects have the opportunity to lie. Personality is derived from the Big Five Inventory test. We use a separate questionnaire for integrity and fairness. The main other-regarding preference to be assessed is altruism using a dictator game. We will also analyse the text used in the descriptions to reveal other psychological and linguistic variables relevant to dishonesty and self-awareness.

Demographics, risk preferences.

Basic demographics (gender, age, education, etc.).

In the MTurk experiment, we first ask subjects to complete a number of questions that reveal information about their risk preferences, personality and attitudes as well as demographics. Next, we randomize subjects into different groups each of which requires the subject to write a description about a real-life event involving honesty or dishonesty. After this, they all face a task designed to reveal dishonesty. There is also a control group who do not write a description but face the same subsequent task. We also examine the scope for the demand effect. More detail about the tasks and our predictions about behavior are also provided in advance but will become available only on completion of the study.

Here we note details of the design and provide some behavioral predictions and contrast our earlier experiment, which was referred to as wave 1 (and is detailed in https://www.socialscienceregistry.org/trials/5142 and summarized below), with this new experiment, which we refer to as wave 2. There is much more detail provided in the included Analysis Plan. Our earlier experiment (wave 1) included a matrix puzzle task that required subjects to assess the number of pairs that total to 10: and then to report the number of pairs that they saw. They were paid based on the total number reported and were free to lie or tell the truth. In the wave 2 version of the matrix puzzle task, we maintain the same average payment in expectation but apply a mean-preserving spread which gives a bonus to those in the top half of the distribution. This shifts the payoffs in such a way that the psychological cost of lying is reduced in accordance with our hypothesis (which is based on Rabin, JEBO 1994). For the matrix puzzle task, we predict that the changes made in the wave 2 version should reduce the cost of lying and so we would expect an increase in dishonesty compared with wave 1.

By computer.

At the individual level.

No

0

400-800 subjects.

100-200 per treatment (there are 3 treatments and one control group)

Since we have already run one wave we elected to use the average effect size derived from Wave 1, but to adjust it as follows. First, we take the average of the effect sizes of different treatments that we calculated in the first wave of the study. For the matrix task, the average effect size was 0.72. Assuming that the variance of the population in the Wave 2 is the same as the Wave 1 of the study, we expect to observe an effect size of 0.72 in the modified version of the matrix puzzle. Technically this means we need only 32 subjects per treatment. However, last time data cleaning necessitated removing 20% of the data. Second, we wish to maintain comparability with wave 1 which featured two games, one of which featured a much lower average effects size (0.11). Therefore we plan to have 100 subjects per treatment which easily satisfies our initial average effect size calculation, then, after gaining a better idea of the variance of the underlying Wave 2 population (which might differ from Wave 1), we will potentially follow up with a larger sample size (rising to 200) if needed.

AnalysisPlan.pdf

We provide the first investigation of the relationship between self-awareness and dis-honesty in a multi-wave pre-registered experiment with 1,260 subjects. In the first wave we vary the level of awareness of subjects' past dishonesty and explore the impact on behaviour in tasks that include the scope to lie. In the second wave we vary the degree of competitiveness in one of our core tasks to further explore the interactions between self-awareness, (dis)honesty and competition. We also test for the experimental demand effect in order to rule it out. Our results suggest that in non-interactive tasks, self-awareness helps to lower dishonesty in the future. However, in tasks that are competitive in nature becoming more aware of past dishonesty raises the likelihood of dishonesty in the future. In other words, we show when making people aware of their own past dishonesty can help to reduce dishonesty and when it might back-fire.

Ceren Bengu Cibik and Daniel Sgroi, The Effect of Self-Awareness and Competition on Dishonesty, IZA DP No. 14256, 2020.

http://ftp.iza.org/dp14256.pdf