Expected income mobility and antisocial behavior: an experiment with self-confidence and entitlement.

There are two treatments, respectively called “Random treatment” and “Effort treatment”. The two treatments differ in how the transition matrix is modified during the experiment. The transition matrix is a table that summarizes the probability with which every participant reaches a given level of income given a starting level of income. In this experiment, it is endogenously modified by individuals’ performance or luck. In the “Random treatment”, the transition matrix is modified according to the results of a lottery in which all subjects participate, whilst in the “Effort treatment”, it is modified according to the results of an effort test.

I have three possible income levels (5€, 7€, and 10.50€), then the transition matrix is a 3X3 matrix. In this experiment, it is endogenously modified by individuals’ performance or luck, in the sense that some probabilities are modified by a value of “p”, with “p” equal to 0.1, 0, or -0.1. The higher the value of “p”, the higher the expected income from each possible initial income. Individuals are divided into three terciles according to a score. Subjects in the first tercile receive p=0.1, participants in the second tercile p=0, and participants in the third tercile receive p=-0.1. In the “Random treatment”, the score is given through a lottery (a randomly drawn number from 0 to 100), whilst in the “Effort treatment”, the score is the performance subjects obtain in an effort test, which is the slider task (Gill and Prowse 2012).
This design enables measuring optimism in the “Random treatment” and self-confidence in the “Effort treatment”, collecting expectations about the outcomes of the lottery and the effort test.

2023-06-01

2023-09-28

The key outcome variables of interest in this experiment are the antisocial behavior decisions and the degree of the antisocial behavior decisions.

Antisocial behavior decisions are decisions in which a subject chooses whether to attack another subject or not. Then, it is a 0-1 decision.
The degree of antisocial decisions represent the sum the perpetrator of an attack wants to remove from the victim's earnings, in a range between 20 and 100 cents.

Self-confidence/optimism measures, a risk-aversion measure, political and religious orientation, and previous and current school information.

Self-confidence/optimism measures are derived from individuals’ expectations about their performance in a test or over a lottery relative to actual results. I collect absolute self-confidence/optimism, the difference between expected and actual results, and relative self-confidence/optimism, the difference between the expected and actual percentage of subjects that performed worse than oneself.
The risk-aversion measure is derived through a test, which is the Bomb Risk Elicitation Task (Crosetto and Filippin, 2013).
Political and religious orientation and previous and current school information are collected through a questionnaire.

The experiment uses a between-subjects design, and it is divided into eight experimental sessions. Participants, who are recruited online, are equally and randomly divided across the experimental sessions.
The experiment is divided into two parts. In both parts, subjects play in groups of three and decide whether to attack another group member’s outcome at a cost. In the first part of the experiment, subjects make attack decisions with no future mobility. Given that in the first round of instructions, mobility is not mentioned, subjects do not know that their income can change in the second part. In the second part of the experiment, attack choices are made knowing that the initial income can change according to the transition matrix. In the second part, attack choices are made twice for each scenario: not knowing the real transition matrix that will be applied to their initial income (i.e., with the value “p” unknown), and knowing it. Then, choices in the second part are different as the self-confidence effect fades away when the true value of “p” becomes known.
Before antisocial decisions are made in the second part, I collect individuals’ expectations regarding the results of the effort test or the lottery (depending on the treatment). These expectations will be used to determine self-confidence (in the “Effort treatment”) and optimism (in the “Random treatment”) measures. At the end of the experiment. a questionnaire is proposed to collect individual characteristics.

Before this main part of the experiment begins, subjects completed the Bomb Risk Elicitation Task (BRET), proposed by Crosetto and Filippin (2012) to measure risk attitudes.
In both treatments, the experiment is divided into two parts. Each part is preceded by the corresponding instructions, which are read aloud by the experimenters, and a comprehension test, highlighting the main concepts of each part.
In both parts of the experiment, subjects made antisocial choices. Subjects are randomly divided into groups of three subjects, that remain the same during the experiment. Subjects’ decisions can only affect the earnings of the members of their group. In each group, a subject has an initial income of 5€, a subject has an initial income of 7€, and a subject has an initial income of 10.50€. Initial incomes in a group are given randomly. During the experiment, subjects do not know their initial income. They see a table that shows their own and group members’ initial income. According to this table, they choose whether to attack another group member’s income or not. When they attack another group member, they choose the sum to be subtracted from the victim’s income in a range between 20 cents and 100 cents. Attacking costs 1/5 of the sum that has been subtracted from the victim. To maintain initial incomes unknown, subjects are shown three scenarios, that differ in the allocations of initial incomes. Every time subjects are asked to make an attack choice, they make a choice for each scenario. In each set of choices, only a choice is made correspondingly to the real initial income and can be used to determine earnings at the end of the experiment. As subjects do not know their real initial income, each choice has the same probability to be used.
As above-mentioned, the experiment is divided into two parts. In the first part, subjects make choices with no future mobility, whilst in the second part, subjects make choices with future mobility. In the second part, choices are made twice: not knowing the real transition matrix that will be applied to their initial income (i.e., with the value “p” unknown), and knowing it.
In both treatments, choices in the second part are preceded by the effort test, which is the slider task, proposed by Gill and Prowse (2012). In the “Effort treatment”, subjects give expectations about the test results before completing the slider task, knowing the test that is used and how it works, and after, to collect information about ex-ante and ex-post self-confidence. They are asked to give their expected number of correct answers and the percentage of participants they expect to have a lower score than their own. In the “Random treatment”, subjects give their expectations about the lottery results one time. They are asked to give the interval in which they expect their number to lie and the percentage of participants they expect to have received a lower drawn number than their own.
At the end of the experiment, subjects have decided on three sets of attack choices, each of which is composed of three attack choices (i.e., one for each possible initial income in each set). A set of antisocial decisions is randomly drawn. If the drawn set is the first one, the decision chosen in the first set considering the real initial income is applied to the initial income (given that in the first part mobility is not considered). If the drawn set is the second (third) one, the decision chosen in the second (third) set considering the real initial income is applied to the initial income (given that in the first part mobility is not considered).

Randomization by a computer, using the sofware z-Tree.

Individual

No

The treatment is not clustered.

192 students

96 students will be in the first treatment and 96 students will be in the second treatment.

In the power analysis, I use a power of 80%, 192 subjects, a standard deviation of 0.47, and given that subjects are equally divided into the two treatments. The standard deviation has been found through simulations of the experiment. I detected a minimum detectable effect size of 0.1899.