Testosterone and diversity

linear regression
When students work on group projects, they work better when their groups are diverse. This study asked if that depends on the testosterone level of the group, and if high-testosterone groups benefit less from diversity.
Author

Alex Reinhart

Published

June 18, 2018

Data files
Data year

2018

Motivation

Some research has suggested that groups of people working on a task can do better if the group is more diverse, since diverse group members can suggest more creative ideas and make better decisions. At the same time, diverse groups can have more conflict than less diverse groups, possibly eliminating those benefits.

This dataset is from a study that attempted to understand these factors and how they relate to the testosterone levels of members of the group. People with high levels of testosterone tend to be competitive and may try to dominate their groups, leading to conflict, while people with low testosterone may be more cooperative. So do groups with low testosterone work better together, and benefit more from diversity, than those with high testosterone?

The study involved 370 students working on their Master of Business Administration degrees at Columbia Business School. The students were randomly assigned to groups of three to six people, and their testosterone levels were measured. The groups then did a week-long project where they pretended to run blood-testing laboratories, with the goal of making their laboratories perform better than those run by other groups in the class. Group members had to collaborate with each other to decide on their strategy. At the end of the project, the researchers recorded how well each group’s laboratories did, how much money they made, and so on.

The result, according to the study, was that

The findings suggest that diversity is beneficial for performance, but only if group-level testosterone is low; diversity has a negative effect on performance if group-level testosterone is high.

Data

The first data file contains individual data: the age, gender, and ethnicity of each student, along with their testosterone and cortisol levels. The second file reports the performance of the teams on their projects. There were 370 students in 74 groups.

Data preview

hormone-diversity-individual.csv

hormone-diversity-teams.csv

Variable descriptions

hormone-diversity-individual.csv
Variable Description
ID Participant ID number
team.id ID number of the team this participant belonged to
Age Age, in years
Gender Gender (Male or Female)
Ethnicity Ethnicity of the participant
Cortisol Participant’s cortisol levels, nMol/L
Testosterone Participant’s testosterone levels, pg/mL
log.cortisol Natural logarithm of the participant’s cortisol level
log.testosterone Natural logarithm of the participant’s testosterone level
Country Country of citizenship of the participant
hormone-diversity-teams.csv
Variable Description
team.id Team ID number
team.size Number of people on the team
final.performance The team’s final performance score
time.of.day The time of day the team’s hormone sample was collected (hh.mm)
females Number of females in the group
final.cash Total cash earned by the team
final.contracts Total number of contracts won by the team
final.reorders Total number of reorders won by the team
final.rank Team’s final rank at the end of the project, relative to other teams in their class section
interim.performance Same as above, but measured at Day 5 of the study (missing for some teams)
interim.cash
interim.contracts
interim.reorders
interim.rank

Questions

  1. The original paper measured the diversity of each group using something called “group faultline analysis”, which looks at the group members’ genders, countries of origin, and ethnicities to calculate a diversity score. The calculation is somewhat involved, so we’ll make a simpler score. For each group, calculate the number of unique gender-ethnicity-country combinations (such as female-white-Russia or male-Indian-USA) among the group members, and store this with the other group information such as team size and performance. Also calculate the average testosterone level for each group.
  2. Do exploratory data analysis to explore the composition of groups, the typical amount of diversity, and the typical amounts of testosterone. Note particularly that the data includes the logs of the cortisol and testosterone levels as well as the raw levels; does your EDA suggest you should use the logs or the raw values?
  3. The data also includes participant ages. It’s possible age is related to hormone levels, as is gender, and that both are related to final performance (perhaps older team members have more relevant experience, for example). Consider carefully whether you want to include these variables in the model, and how you should include them—average team member age? Gender proportion?
  4. Build a model predicting group performance (final.performance) using the group’s diversity score (be sure to control for the size of the group) and its average testosterone level. Is there an interaction between the two? Do your results resemble those presented by the original study?
  5. The dataset also includes cortisol measurements. Cortisol is a hormone with many effects and roles in the body, but is commonly known as a “stress hormone” because its levels often increase because of sleep deprivation, intense exercise, and stress. Curiously, though cortisol is included in the data, it is not mentioned in the paper apart from an appendix. Include it in your analysis. Do stressed groups have better or worse performance? Does stress change the effect of diversity?

References

M. Akinola, E. Page-Gould, P. H. Mehta, Z. Liu (2018). Hormone-Diversity Fit: Collective Testosterone Moderates the Effect of Diversity on Group Performance. Psychological Science 29 (6): 859-867. https://doi.org/10.1177/0956797617744282

Data available from https://osf.io/8eqtc/