The Explanatory Power of Identity Models in Explaining Moral Balancing

Participants are assigned to one of 12 treatments. In the treatments, we vary two dimensions: 1.) the main outcome measure (4 options) and 2.) the actual or considered past (3 options).

The main outcome measure varies the main variable we want to measure. In the choice treatment (CT) individiuals face a moral choice, in the belief treatment (BT) we measure individuals' belief in being a high moral type, in the norm sensitivity treatment (NST) we measure the individiuals' norm sensitivity, and in the social norms treatment (SNT) we measure the social norms present in the CT.

The actual or considered past varies the task that participants face immediately before the main outcome measure. In the moral past treatment (MPT), individuals face a task that induces them to act morally, in the no past treatment (NPT), individuals receive an endowment without making a choice, and in the immoral past treatment (IPT), individuals face a task that induces them to act immorally.

A full description of the experimental design can be found in our pre-analysis plan.

2025-05-18

2025-08-31

In the CT: Whether or not a person donates in the donation game.
In the BT: The belief in being a high moral type according to the belief elicitation.
In the NST: The norm sensitivity according to the rule-following task.
In the SNT: The social norms of the CT according to the norm elicitation.

Construction of the belief in being a high-moral type: During the belief elicitation, participants state a number between 0 and 100. After the experiment, we normalize all answers to a scale ranging from zero to one.

Construction of norm sensitivity: During the rule-following task participants place 20 balls either in the blue or the yellow bucket. After the experiment, we calculate the norm sensitivity by dividing the number of balls in the blue bucket by the number of balls available.

Construction of the social norms: When judging the social appropriateness of an action, participants can rate each action with one of the following eight ratings: “Extremely Socially Inappropriate,” “Very Socially Inappropriate,” “Socially Inappropriate,” “Somewhat Socially Inappropriate,” “Somewhat Socially Appropriate,” “Socially Appropriate,” “Very Socially Appropriate,” and “Extremely Socially Appropriate. After the experiment, we convert the categorical ratings in the presented order into the following numerical scores: −1, −0.75, -0.5, −0.25, 0.25, 0.5, 0.75, 1.

The research project consists of a survey and two experiments: The pre-screening survey, the main experiment and the receiver experiment. The main experiment consists of two parts, separated by 24-48 hours.

Pre-screening survey:
In the first part, participants answer the questions on the altruism measure from the Global Preference Survey (Falk et al., 2018; Falk, Becker, Dohmen, Huffman, & Sunde, 2022). Based on their responses, we determine their moral type (high or low).

Part one of the main experiment:
In the first stage, participants encounter a belief elicitation (Karni, 2009) in which we measure their initial belief in being a high-moral type. In the second stage, participants encounter a rule-following task (Kimbrough & Vostroknutov, 2018) in which we measure their initial norm sensitivity.

Part two of the main experiment:
The stages encountered in part two of the main experiment differ across treatments. In stage one of the CT-MPT, participants assume the role of the sender in a sender-receiver game (Gneezy, 2005), which has been calibrated to induce truthfulness. In the first phase of the CT-NPT, participants are given a sum of money and cannot act. In the first stage of the CT-IPT, participants assume the role of the sender in a sender-receiver game, which has been calibrated to induce lying. In the second phase of the CTs, participants encounter a binary donation game in which they either donate a certain amount or keep their money. In stage three of the CTs, we elicit the (conditional) social norms of the actions possible in the previous two stages of their respective treatment using the Krupka & Weber method (2013). In the first stage of the BTs, subjects participate in the same tasks as in the CTs. In stage two of the BTs, subjects repeated the belief elicitation from part one. In the first phase of the NSTs, subjects again participated in the same tasks as in the CTs. In the second phase of the NSTs, subjects repeated the rule-following task from part one. In stage three of the NSTs, we elicit the (conditional) social norms of actions possible in the previous two stages of their respective treatment with the Krupka & Weber method. In stage one of the SNTs, we measure the same social norm as in stage three of the CTs. However, this time we use a between-design instead of a within-design.

Receiver Experiment:
In stage one, participants assume the role of the receiver in the sender-receiver game.

A full description of the experimental design can be found in our pre-analysis plan.

To assign participants to their treatments, we match participants to an entry in a random sequence of treatments based on their order of arrival. To create the random sequence of treatments, we create a sequence of blocks, each containing the treatments at a given frequency. Crucially, the order within each block is randomly shuffled (using the Python module random). As we approach the end of the data collection, we adjust the treatment frequency in each block to achieve our desired goal of usable observations.

Individual level.

No

We will collect at least 1940 usable observations from participants in the main experiment.

An observation from the CT, BT, and NST is usable if the participant behaves morally in the sender game of the MPT, if the participant is assigned to the NPT, and if the participant behaves immorally in the sender game of the IPT. An observation from the SNT is always usable.

We will slightly exceed this number since it is impossible to know the attrition between part one and part two of the main experiment in advance.

We will collect at least 1940 usable observations from participants in the main experiment. An observation from the CT, BT, and NST is usable if the participant behaves morally in the sender game of the MPT, if the participant is assigned to the NPT, and if the participant behaves immorally in the sender game of the IPT. An observation from the SNT is always usable. We will slightly exceed this number since it is impossible to know the attrition between part one and part two of the main experiment in advance.

We will collect data until we have at least 400 usable observations in the CT-NPT and 140 usable observations in each other treatment of the main experiment.

An observation from the CT, BT, and NST is usable if the participant behaves morally in the sender game of the MPT, if the participant is assigned to the NPT, and if the participant behaves immorally in the sender game of the IPT. An observation from the SNT is always usable.

We will slightly exceed these numbers since it is impossible to know the attrition between part one and part two of the main experiment in advance.

We used G-Power (Faul, Erdfelder, Lang, & Buchner, 2007) to determine the required sample size for each hypothesis. In each case, we used two-tailed tests, a significance threshold of 0.05, and a power level of 0.8. Furthermore, we assumed an even allocation between treatments. For Hypothesis 1, we calculated the required sample size for a test of proportions, assuming that we would observe the same donation rates as Gneezy et al. (2014). We determined that we would need at least 93 usable observations in the CT-MPT and CT-NPT to detect moral licensing and 70 usable observations in the CT-NPT and CT-IPT to detect moral cleansing. For Hypothesis 2, we calculated the required sample size to test the slope of a bivariate logit regression, assuming that we would observe at least a medium effect size (odds-ratio larger than 3.5), a probability of donating under H0 of 0.5 and that the belief in being a high-moral type is approximately normally distributed with a mean of 0.65 and a standard error of 0.25. We determined that we would need at least 390 observations in the CT-NPT. For Hypotheses 3 to 5, we calculated the required sample size to test the slope in a multivariate ordinary least squares regression with 2 to 5 explanatory variables, assuming that we would observe at least a small effect size (Cohen's f2 larger than 0.02). We determined that we would need at least 132 usable observations in the CT-MPT, CT-NPT, CT-IPT, BT-MPT, BT-NPT, BT-IPT, NST-MPT, NST-NPT, NST-IPT, SNT-MPT, SNT-NPT, SNT-IPT. Based on these power calculations, we concluded that we would need 400 usable observations in the CT-NPT and 140 usable observations in each other treatment.