Discriminatory Beliefs and Gendered Task Selection

This study evaluates how peer discrimination informs preferences over working on tasks that often hold strong gender biases.
To address this an randomized control trial laboratory experiment is used with student participants.

Half of the participants are randomly selected to receive a discrimination signal. This discrimination signal captures the implicit gender biases in math of randomly assigned peer groups. The discrimination treatment is used to measure how preference to participate in math tasks change when individuals gain information about their peer's biases. Secondly, this paper uses the differences and treated and untreated individuals to observe differences in beliefs about their performance and their peer group's performance.

An example of the discrimination information presented to the treatment group is: ”In Experiment 1, you did an Implicit Association Test. An Implicit Association Test (IAT) is a psychological tool used to measure unconscious biases—attitudes or stereotypes people may hold, even if they are not aware of them. It works by asking participants to quickly sort words or images into categories, and the speed and accuracy of their responses reveal hidden associations in their minds. For example, if someone associates certain groups with positive or negative words more quickly, it suggests an implicit bias. On average your teammates showed a strong positive implicit bias toward women participating in math. Having an implicit bias for women and math means unconsciously associating women with strong mathematical abilities while being less likely to associate men with the same skill. An example of this might include assuming women are more likely to excel in math tasks. ”This signal differs based upon the gender of the respondent and the beliefs of their assigned groups. The discrimination treatment is used to measure how preference to participate in math tasks change when individuals gain information about their peer’s gender biases over the task type. The discrimination signal is constructed using a gender-science implicit association test prior to the start of the choice activities. The IAT produces Cohen’s D which is aggregated to the team level using the leave one out mean method. A threshold of 0.2 is set to indicate that a group has a negative implicit gender bias towards women, and a threshold of -0.2 is set to indicate that a group has a positive implicit gender bias towards women. Further thresholds are set for the level of biases which are slight/moderate/strong holding values (0.2-0.5] (0.5-0.8] (≥0.8) respectively.

The control group receives information about the first stage of the experiment and information about the average beliefs of past respondents. An example of the information that the control group receives is: "In Experiment 1, you did an Implicit Association Test. An Implicit Association Test (IAT) is a psychological tool used to measure unconscious biases—attitudes or stereotypes people may hold, even if they are not aware of them. It works by asking participants to quickly sort words or images into categories, and the speed and accuracy of their responses reveal hidden
associations in their minds. For example, if someone associates certain groups with positive or negative words more quickly, it suggests an implicit bias."

2025-02-14

2025-09-30

The primary outcome of interest is selection into the math task following receiving the discrimination treatment. This is a binary outcome where individuals have the option to choose math or the facial emotional recognition task. This will also be analyzed across difference in response for men and for women.

Difference in self-efficacy- Self-efficacy is measured using incentivized survey questions about the
number of questions that the respondent believes that they got correct in incentivized practice rounds, where
participants complete both the facial emotional recognition questions and math questions.

If groups have an average implicit score that is negative, that is they hold an implicit bias such that they more easily associate women with math these observations will be dropped for the main portion of analysis. These observations will be dropped due to them uncommon and thus will not have the power to separate out the positive bias for women from the negative bias.

Difference in self-efficacy, difference in beliefs about others performance, difference in secondary beliefs (social costs).

Self-efficacy is measured using incentivized survey questions about the number of questions that the respondent believes that they got correct in incentivized practice rounds, where participants complete both the facial emotional recognition questions and math questions. A comparison is made in these beliefs across treatment groups (i.e., individuals who receives the discrimination message and those that did not) and male and female respondents.

Difference in beliefs about other's performance: This outcome is measured by asking participants incentivized questions on average how many questions they thought the average (male/female) participant in the session got correct in the (math/FER) practice rounds. A comparison is made between male/female respondents in the treated and control groups.

Difference in secondary beliefs (social costs)-:This is measured through an incentivized survey instrument about a respondents beliefs on how their math ability is perceived by their teammates. Respondents are asked how they think each of their teammates ranked them in the previous question. A comparison is made between treated and untreated groups and male and female respondents.

This experiment uses a three part experiment (1) elicit beliefs (2) Complete an incentivized choice experiment (3) Complete a questionnaire exploring mechanisms and collecting demographic information.

This is a three part laboratory experiment:

1) The first stage is to elicit the gender biases of randomly assigned groups. This is done using the gender and science implicit association test (Greenwald et al., 1998). A score from this is aggregated to the team level and revealed to those that receive the treatment in the second stage of the experiment.

2) The second stage of the experiment participants complete three rounds of selecting and completing either a math task or a facial emotional recognition task. They are told that their objective is to get as many questions correct in thirty seconds, and they will receive pay for the total earnings of their team in one randomly selected round. Half of participants are randomly assigned to receive the discrimination signal after completing the second round of the task. This discrimination signal is produced in the first round of the activity.

3) Participants complete a questionnaire where they are asked about the experiment, demographic information, and explore secondary outcomes. The secondary outcomes use incentivized questions to explore how the discrimination information affects beliefs about own ability, peer's ability, and perception of peer's beliefs about own ability.

Treatment is assigned using a random number generator within Qualtrics, assigning values either 1 or 2. If assigned 1 the observation is in the control group, if assigned 2 the observation is in the treated group.

Individual level

No

NA

250-400

There will be between 125-200 pupils in both the treatment and the control group. Roughly half of the participants will be assigned to treatment group and half to the control group. Additionally, there will be roughly half male and half female participants.

Using the power twoproportions command in Stata, the minimum detectable effect size is 0.15 for a sample of 300 with an equal split between treatment and control groups. This will provide a power of .094 at the 95\% level. When comparing effects on men and women separately where the control and treated groups have 75 observations the minimum effect size is approximately .2 power of .87 at the 95% level.