|
Field
Trial End Date
|
Before
December 20, 2024
|
After
December 20, 2026
|
|
Field
Last Published
|
Before
March 19, 2024 05:18 PM
|
After
September 16, 2026 02:08 AM
|
|
Field
Intervention (Public)
|
Before
A large literature shows that people tend to associate with people of the same gender, a phenomenon known as "Gender Homophily". This phenomenon might contribute to gender segregation across workplaces if people prefer interacting with same-gender coworkers.
We test this hypothesis in a lab experiment, in which we allow participants to endogenously sort into teams. We measure Gender Homophily by studying whether women choose to team up with women, and men with men.
Participants perform in pairs 3 rounds of a gender-neutral real-effort task with shared payoff. The task consists of counting the number of 0s in 10x10 matrices filled with 0s and 1s.
Before each round, participants are presented with 4 gender-balanced CVs, each representing a participant from their session. We allow participants to endogenously sort into teams, by asking them to rank these 4 CVs by order of preference. Rankings are used to match participants in pairs, following a serial dictatorship matching procedure.
This enables us to study whether women are more likely to team up with women, and men with men.
In each round, participants are paired with the same partner and must count 6 matrices within 30 seconds. Each partner independently counts the number of 0s in the matrix. After each matrix, one answer is chosen to determine the team's payoff for that specific matrix. We investigate how individual mistakes in teamwork influences Gender Homophily by varying the salience and visibility of mistakes across treatments:
In Treatment 0, participants receive feedback after each matrix on the team performance, without any individual-specific feedback. This treatment gives us a benchmark of the degree of gender homophily in a gender-neutral environment.
Treatment 1 renders individual mistakes more visible, by introducing an additional element of individual accountability. If an answer is incorrect, the name of the team member responsible is disclosed to the other participant.
Treatment 2 expands upon Treatment 1 by rendering individual mistakes even more visible. Not only is the name of the team member revealed for an incorrect answer, but this treatment also introduces an interactive component: the other team member sends a message in response to the error. This allows for direct communication between team members and introduces potential peer feedback or criticism to the experiment.
We hypothesize that homophily will increase with treatment number, as the salience and the visibility of individual mistakes grow.
We also investigate how homophily evolves in a multi-round context. Participants engage in three rounds of the same treatment, each time with a different partner selected from a distinct pool. This dynamic feature of our experiment will enable us to study whether participants update their behaviors based on experience, especially focusing on how homophily evolves after interacting with opposite-gender partners.
|
After
Update to the original experiment (AEARCTR-0012546, March 14, 2024)
We completed the experiment described in the original registration, with two additions to the initial design.
First, we added a fourth treatment, Treatment 3, which reverses the logic of Treatment 1 by making successful individual performance salient rather than mistakes. If the selected answer is correct, the name of the team member whose answer was selected is disclosed to the other participant. If the selected answer is incorrect, no individual-specific information is revealed.
The resulting experiment includes 512 participants, corresponding to four treatments, eight sessions per treatment, and 16 participants per session. Each session includes eight women and eight men, divided into four groups of two women and two men.
Second, we use performance in Round 0 as a predetermined measure of individual performance. In Round 0, participants complete the task individually, before any team interaction or feedback. They count the number of zeros in four matrices and receive no feedback on their performance. We use this measure to distinguish between high- and low-performing participants.
In the completed experiment, we find no evidence of gender homophily, on average, among men. However, we document gender homophily among women, which is driven by high-performing women. This pattern is constant across feedback regimes and is not explained by gender differences in performance or by gender differences in beliefs about performance.
These findings motivate a new extension designed to examine whether gender homophily among high-performing women persists when information on the performance of potential partners in Round 0 is disclosed.
|
|
Field
Intervention End Date
|
Before
December 20, 2024
|
After
December 20, 2026
|
|
Field
Primary Outcomes (End Points)
|
Before
Our primary focus is examining homophily through participants' ranking of potential partner CVs. We will test whether women assign higher ranks to female CVs, and men to male CVs.
We will track individual and team performance, along with beliefs regarding personal and partner's performance.
Post-experiment, we will also collect information on participants' sensitivity to negative evaluation and punishment.
|
After
- Our primary focus is examining homophily through participants' ranking of potential partner CVs. We will test whether women assign higher ranks to female CVs, and men to male CVs. We will retain the gender homophily index used in the original experiment (see explanations).
- Our interest is in how gender homophily changes with the gender gap in performance. We illustrate our theoretical predictions using the case of female homophily among high-performing women. Predictions with gender gap in performance (see explanations).
- We will track individual and team performance, along with beliefs regarding personal and partner's performance.
- Post-experiment, we will also collect information on participants' sensitivity to negative evaluation and punishment.
|
|
Field
Primary Outcomes (Explanation)
|
Before
|
After
- Gender Homophily index
We will retain the gender-homophily index used in the original experiment.
Each participant assigns ranks from 1 to 4 to two same-gender and two opposite-gender CVs. We assign Borda scores
s(r) = 3 if r = 1; 2 if r = 2; 1 if r = 3; 0 if r = 4
The gender-homophily index is
Gᵢ = s(rᵢ⁽¹⁾) + s(rᵢ⁽²⁾)
where rᵢ⁽¹⁾ and rᵢ⁽²⁾ are the ranks assigned to the two same-gender potential partners.
The index ranges from 1 to 5, with higher values indicating greater gender homophily.
We will use this aggregate measure to examine whether average homophily changes when information on potential partners' performance becomes available.
- Predictions with gender gap in performance: When male CVs display relatively higher performance than female CVs.
Case 1: Female homophily decreases with the gender gap in performance
This pattern would be consistent with lexicographic preferences: performance first; gender otherwise.
Case 2: Female homophily does not respond to the gender gap in performance
This pattern would be consistent with preference-based homophily that persists despite differences in potential partners’ performance.
Case 3: Female homophily increases with the gender gap in performance
This pattern would be consistent with motives such as gender inequality aversion or altruistic motives toward female partners.
|
|
Field
Experimental Design (Public)
|
Before
We use a between- subject design in an individual decision-making study. Each participant is randomly assigned to one treatment and will earn at least 5€. A session comprises 16 participants, divided into 4 gender-balanced groups.
The experiment comprises 3 parts.
In the first part of the experiment, participants fill in a short demographic survey. They are asked their gender, whether they were born in Ile De France and whether their average daily commuting time is above or below 30 minutes. This information is used to create a CV for each participant. After filling the form, participants are asked to choose a fictitious name that will be used as an individual identifier throughout the experiment. Participants who report being women select from a list of female names, with each chosen name removed from the list to prevent duplicates. Similarly, those who report being men pick from a list of male names, ensuring no two participants have the same name.
In the second part of the experiment, participants perform the real-effort task individually, to get familiarized with the task.
They are presented with four 10x10 matrices filled with 0s and 1s. They are given 30 seconds per matrix to count the number of 0s. They earn 1 point if they correctly count the number of 0s in the matrix.
In the third part of the experiment, participants perform three rounds of the real-effort task in pairs.
We allow them to choose their partner before each round. More precisely, they are presented with 4 gender-balanced CVs, each representing a participant from their session. Each CV contains the fictitious name chosen by the participant, as well as their gender, whether they were born in Ile de France, and whether their daily average commuting time is below or above 30 minutes. We randomize the order in which CVs appear (whether a female or a male CV appears first) across rounds. Participants are presented with different CVs across the rounds.
Participants rank the 4 CVs from 1 (most preferred partner) to 4 (least preferred partner). Rankings are used to match participants in pairs, following a simple strategy-proof procedure. One participant is chosen at random and is assigned to their top-ranked partner. Another participant is then randomly selected. This participant is paired with their top-ranked partner. If that partner is already paired, the participant is paired with their next preferred choice. This process continues until everyone is paired.
Participants are then informed of their partner's identity, receiving the fictitious name assigned to their partner. Before beginning the counting task, they are asked to estimate both their own and their partner's performance from the previous round. These guesses are incentivized: participants earn 1 point for each correct prediction.
After making their guesses, they start the real-effort task. They count 6 tables filled with 0s and 1s.
Both partners see the same table filled with 0s and 1s. They both have 30 seconds to count all the 0s. One partner's answer is randomly chosen as the team's response. If this answer is correct, both earn 1 point; if not, both earn 0 points.
At the end of the round, participants are asked how much they enjoyed performing the task with their partner.
We vary the type of feedback received across treatments:
In treatment 0, participants are informed after each table whether the selected group's answer is correct. The identity of the partner whose answer was selected is not revealed.
As in Treatment 0, participants in Treatment 1 are informed after each table whether the selected group's answer is correct. If the group's answer is correct, the identity of the partner whose answer was selected remains undisclosed. However, the identity of the partner is revealed if the group's answer is incorrect, making individual errors more visible.
Treatment 2 expands upon Treatment 1 by rendering individual mistakes even more visible. Not only is the name of the team member revealed for an incorrect answer, but this treatment also introduces an interactive component: the other team member selects a message from a list and send it to their partner in response to the error. The list contains 4 messages, which vary by the emphasis put on the mistake. Participants are presented with 2 comprehensive messages "No stress, this matrix can be tricky!", "Don’t worry, mistakes happen!"; and two messages that hold the partner more accountable for the mistake: "I know it’s tough, but try to focus more.", "It’s crucial to get the exact count, you should try another counting method."
At the end of the experiment, participants perform a post-experimental survey, which consists of 3 parts.
The first part consists of incentivized guesses about performance, to test whether men and women are perceived as performing equally well in the counting task. Participants are told that in a previous session, two female players were assigned the names "Jade" and "Louise"; and two male players the names "Gabriel" and "Léo". They are asked to rank them from highest performer (rank 1st) to lowest performer (rank 4th). They can give the same rank to different players. They earn 1 point for each correct rank.
The second part asks participants to indicate how useful each trait (name, gender, region of origin, commuting time) was for ranking the CVs.
The third part consists of questions aimed at measuring participants' sensitivity to negative evaluation and punishment.
At the end of the session they are paid in cash, depending on the number of points they have accumulated throughout the experiment, with 1 point corresponding to 1 euro.
|
After
In the completed experiment, we find no evidence of gender homophily, on average, among men. However, we document gender homophily among women, which is driven by high-performing women. This pattern is constant across feedback regimes and is not explained by gender differences in performance or by gender differences in beliefs about performance.
These findings motivate an extension designed to examine whether gender homophily among high-performing women persists when information on the performance of potential partners in Round 0 is displayed.
The extension keeps the experimental setting as close as possible to the original design but adds predetermined performance to the information displayed in potential partners' CVs.
As in the original experiment, participants first complete Round 0 individually and without feedback (Treatment 0). They then complete three rounds of partner rankings. In each round, participants rank four CVs, two females and two males, from most to least preferred. Motivated by the fact that CV rankings are stable across rounds, we randomly select one of the three rounds for implementation, and participants are matched according to their ranking in that round to perform the collective task in pairs.
The main difference with the original experiment is that each CV additionally reports the potential partner's performance in Round 0. The main question is whether high-performing women continue to prefer female partners when male potential partners perform better.
|
|
Field
Planned Number of Observations
|
Before
450 participants.
|
After
200 participants.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
150 participants per treatment.
|
After
200 additional participants in the control treatment, with predetermined performance displayed in the CVs.
|