The Impact of Gender Diversity on Team Communication, Team Performance, and Preferences for Teamwork

We implement an online experiment that induces exogenous variation in the gender composition of teams working on a complex real-effort team task. Each experimental session consists of two stages. In stage 1, we form randomly composed teams of four who meet in an online browser-based chat room. The teams' gender composition varies between all-male, all-female, and mixed (two males and two females). While working on the team task, subjects can communicate via a group audio chat that is recorded by us. In stage 2, we form randomly composed pairs of two subjects who have not met in stage 1. Pairs meet in the chat room for a short period of time. During that time, the subjects perform a simple task that requires them to talk to each other. Once the audio chat is closed, we inform subjects about the possibility that they will work on another task similar to the one in stage 1 for 15 minutes, and ask subjects to state their preference for working on the task individually or in a team with the subject they met in the chat room. We also elicit subjects' beliefs about productivities (individual and team productivity), and beliefs about team communication and interaction should the subject work with the potential teammate.

2021-07-01

2022-01-31

Quantitative communication measures at team level (stage 1):
- Number of words spoken
- Number of contributions
- Team-level Herfindahl indices over number of words and number of contributions
- polarity of team communication

Subjective communication measures at team level (stage 1, all measured individually on a 5-point Likert scale):
- Averages of survey responses to question on how positive the team communication was
- Averages of survey responses to question on how cooperative the team communication was
- Averages of survey responses to question on enjoyability of team work

Quantitative communication measures at subject level (stage 1):
- Number of words spoken
- Number of contributions

Subjective communication measures at subject level (stage 1, all measured individually on a 5-point Likert scale):
- Survey response to question on how positive the team communication was
- Survey response to question on how cooperative the team communication was
- Survey response to question on enjoyability of team work

Measure for team performance: Number of correctly solved problems in stage 1

Preference for teamwork: Binary indicator for preference to work with potential teammate in stage 2, as opposed to individual work

Productivity beliefs (stage 2):
- Belief about own output (number of correctly solved problems out of 20) if subjects works individually
- Belief about potential teammate's output (number of correctly solved problems out of 20) if teammate works individually
- Belief about team output (number of correctly solved problems out of 20) in case of teamwork with potential teammate

All outcomes not labelled primary outcomes will be considered secondary outcomes (see pre-analysis plan for details).

The experiment will be implemented as an online experiment with students of a large public university in Germany. Many students enrolled at the university are registered with a university-wide mailing list used by researchers to invite registered students to surveys and other research-related tasks that can be conducted online. To collect our data, we invite random subsamples of students on the mailing list via email to participate in an online session at a specific time.

Each session consists of two stages. In stage 1, we form randomly composed teams of four who meet in an online browser-based chat room to work on a complex real effort task for 30 minutes. The teams' gender composition varies between all-male, all-female, and mixed (two males and two females). The real effort task consists of ten single-choice problems. Each team member receives a bonus for each problem provided that all team members mark the correct answer. While working on the task, subjects can communicate via a group audio chat. All participants have given us permission to record the audio chat for research purposes. Stage 1 ends with a short survey that subjects fill out individually. In stage 2, we form randomly composed pairs of two subjects who have not met in stage 1. Pairs meet in the chat room for a short period of time. During that time, the subjects perform a simple task that requires them to talk to each other. Once the audio chat is closed, we inform subjects about the possibility that they will work on another task similar to the one in stage 1 for 15 minutes, and ask subjects to state their preference for working on the task individually or in a team with the subject they met in the chat room. Before eliciting the preference, we inform subjects about a random draw with three possible outcomes: (a) both subjects who met will work on the task individually, irrespective of their stated preferences; (b) their stated preferences will be implemented; and (c) they will not work on the task at all. We also elicit subjects' beliefs about productivities (individual and team productivity), and beliefs about team communication and interaction should the subject work with the potential teammate. We then implement the random draw regarding the task, and (if determined by the draw) subjects work on the task. The experiment ends with a brief survey.

Subjects are invited for sessions taking part at a specific time. Using a link provided in the invitation email, they enter a webpage. A page with basic instructions informs subjects that they take part in a scientific study, that they will interact with other subjects via an audio chat, and that the audio will be recorded for research purposes. Only subjects who give explicit consent to the recording and to the linking of the data generated during the experiment with administrative data from the university registry can participate. Subjects are then redirected to a microphone test. Next, subjects see a screen with instructions. We inform subjects that they will receive a show-up fee of EUR 10 and that they will work on a series of 10 single-choice problems in a team together with three other subjects they will meet in an audio chat room. We also inform them about a piece rate of EUR 1 for each correctly solved question, and that the piece rate will only be paid if all team members mark the correct statement. We add a statement that further money can be earned in a later stage. We then connect the team members via the chat room for stage 1. Teams have some time to say hello to each other. In the chat room, the subjects of a team are labelled from 1 to 4 (the order is randomly determined). Each team member's number is shown as an avatar in the audio chat and we visualize who is currently speaking. This enables subjects to directly address each other. Next, we instruct the subjects to familiarize themselves with further written technical instructions, including how to reconnect if the internet connection is lost. After that, the information material for the first block of problems is displayed, and teams start working on the problems. With 10 problems and 3 minutes to work on each problem in stage 1, teams work on the task for a total of 30 minutes (net of the time given for reading the information material).

Once the work on the task is completed, the chat room is closed and subjects answer survey 1 individually. The survey contains several items that elicit the subjects' perceptions of team communication and interaction (whether team communication was positive, whether it was cooperative, whether working jointly on task was enjoyable, plus items on quantity of communication, distribution of shares of speech, and interruptions), the belief about how many problems the team has answered correctly, and the belief about how much one has contributed to team output. We also ask subjects to indicate how many of the other subjects in their team they believe were female. In order not to reveal the purpose of the study, we also ask subjects how many of the other subjects in the team they believe are enrolled in certain fields of study, and how many they believe have completed at least two semesters at university.

At the beginning of stage 2, we inform subjects about a further stage of the study, and that they will earn an additional flat payment of EUR 2. We inform subjects that they will meet another subject in the chat room for one minute. Subjects are randomly assigned to form pairs of two. To form pairs, we randomly determine pairs of first-stage teams and randomly match subjects across teams. This makes sure that the newly formed pairs of two consist of subjects who have not met in stage 1. To make sure that subjects in a pair talk to each other, we display a random five-digit number on each subject's screen. Subjects are instructed to fill in the other subject's number in a field on their screen. Once the chat room is closed again, we inform subjects about the possibility that they will work on another task similar to the one in stage 1 for 15 minutes, and ask subjects to state their preference for working on the task individually or in a team with the subject they met in the chat room. The instructions also explain the random draw that determines which of the following scenarios will be implemented: (a) both subjects who met will work on the task individually, irrespective of their stated preferences; (b) their stated preferences will be implemented as follows: they work as a team if they both indicated this as their preferred option, and they both work individually otherwise; or (c) the subjects will not work on the task at all. The subjects then state their preferences for teamwork (binary choice between individual work or teamwork). The subjects' beliefs about their individual productivity is elicited as follows: we ask subjects to imagine they would work individually on a task similar to the one they have worked on in stage 1, but that the task would comprise 4 blocks of 5 problems each, giving a total of 20 problems. Given a piece rate of EUR 1 per correctly solved problem, we ask subjects to indicate how many problems they believe they would solve. Similarly, we ask subjects to indicate how many problems (out of 20) they believe the other subject would solve correctly when working on the task individually. Finally, we ask the subjects to state how many problems they believe they would solve when working with the other subject in a team under conditions as in stage 1. In addition, we elicit beliefs about team communication and interaction should the subject work with the potential teammate. For that purpose, we use similar items as in stage 1. Specifically, we use items measuring subjects' beliefs about whether the communications would be positive, whether it would be cooperative, and whether working jointly with the potential teammate on task would be enjoyable. Finally, survey 2 elicits the subjects' perception of the potential teammate's gender. In order not to reveal the purpose of the study, we also ask subjects if they believe the potential teammate is enrolled in certain fields of study and has completed at least two semesters at university.

The random draw regarding the subjects' actual work on the task will be parameterized such that (a) the probability for individual work irrespective of stated preferences is 5\%, (b) the probability for the stated preferences to be implemented is 5\%, and (c) the probability for no work on the task at all is 90\%.

The experiment ends with survey 3, eliciting the Big-5 personality traits. We use a 15-item survey to measure openness, conscientiousness, extraversion, agreeableness, and neuroticism.

Randomization by a pre-programmed algorithm

First stage: Individuals are randomly assigned into teams of four
Second stage: Individuals are randomly assigned into pairs of two

No

We are unsure how many subjects from the pool of students registered on the mailing list we will be able to recruit for the experiment. We believe a final sample size of between 200 and 400 first-stage teams to be realistic.

First-stage team level: between 200 and 400 teams. Subject level: between 800 and 1600 subjects As described in the pre-analysis plan (see subsection 3.1.4, "Check if First-Stage Task Is Gender Neutral"), before starting the data collection on the team task, we collected 55 observation on subjects who worked individually on the first-stage task. We use this data to test if individual performance on the task differs by gender. To match the expected sample size regarding the first-stage team task (about 300 teams with complete communication data), we will collect about 250 additional individual observations. This data will be collected starting from December 20, 2021.

First-stage team level: between 67 and 134 per team of given gender composition (all-female, all-male, mixed)

Subjective assessment of how positive team communication was in stage 1: With a sample of 200 (300, 400) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition is 0.20 (0.16, 0.14). Subjective assessment of how cooperative team communication was in stage 1: With a sample of 200 (300, 400) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition is 0.18 (0.15, 0.13). How much the subjects liked to work with their teammates in stage 1: With a sample of 200 (300, 400) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition is 0.27 (0.22, 0.19). Number of correctly solved problems in stage 1 at team level: With a sample of 200 (300, 400) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition (for instance, mixed vs.~all-male teams) is 0.83 (0.68, 0.59). Regarding estimations at the level of the individual subject, we report minimum detectable effect sizes for gender-specific estimations of the effect of being in a mixed team, relative to being in a gender-homogenous team. Hence, in each estimation, we use about half of the overall sample of subjects. Subjective assessment of how positive team communication was in stage 1: With a sample of 800 (1200, 1600) subjects overall (about half of them females), the minimum detectable effect for differences between teams of a given gender composition is 0.26 (0.21, 0.18). Subjective assessment of how cooperative team communication was in stage 1: With a sample of 800 (1200, 1600) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition is 0.24 (0.20, 0.17). How much the subjects liked to work with their teammates in stage 1: With a sample of 800 (1200, 1600) first-stage teams, the minimum detectable effect for differences between teams of a given gender composition is 0.37 (0.30, 0.26). Standard deviations in all-male teams are slightly larger, leading to slightly larger minimum detectable effects for male subjects.

Pre-Analysis Plan