When it comes to hiring people, many organizations tend to aim for diversity (at least they say they do). People from different backgrounds could bring different perspectives to the same problem. On the other hand, people from different backgrounds might simply lack technical knowledge about the field the organization’s involved in. So do the benefits of diversity outweigh the costs?
This is a hard question to answer and it most certainly depends on the situation. This post discusses a specific instance of this question by presenting a new working paper studying the effects of ability dispersion on team performance.
Hoogendoorn, Parker and van Praag (2014) organized a controlled experiment where 573 students were split into 49 teams. The assignment to teams was random except for the ability of the students. Some teams had high ability dispersion, while others had medium and low dispersions. The students’ cognitive ability was measured by a Raven’s Progressive Matrices test, which is sort of like an IQ test and was significantly positively correlated with GPA in the experiment.
Ability dispersion was measured by the coefficient of variation, which is essentially the standard deviation expressed as a percentage of the mean. The CV for teams varied between .07 and .47 with a mean of .22. Thus indeed quite high variation was created in ability dispersion among teams.
The teams apparently had to create real businesses which they had to operate for some time (I think for roughly a year). They had to come up with a product to sell, and they had to get funding from shareholders (who were mostly relatives and friends). Bottom line is, according to the paper, they had to operate real small businesses.
So the authors deliberately created a variation in ability dispersion among teams, and could now see what levels of dispersion led to better performance.
First, by putting teams into low-, mid- and high-ability categories, the authors find that mid-ability teams performed the best, but this result is very sensitive to how we define the low, mid and high categories (i.e. where the cutoff points are). A more robust result arises when we split teams into low-, mid- and high-diversity teams (as measured by ability dispersion). In this case, mid-dispersion wins and now this result is insensitive to where the cutoff points are.
Second, the authors see how well certain performance indicators of the businesses (sales, profits, etc.) can be explained by ability dispersion and ability. They find that ability has no effect on performance, but dispersion does affect some variables significantly. Specifically, the effect of ability dispersion is strongest and highly significant on an indicator of whether the team had positive profits. It also predicts profits nicely, but fails to be a highly significant predictor for profits per share and sales (though the signs are as expected).
So what relationship is revealed in this part of the analysis? The authors find that the relationship between ability dispersion and profits is hump-shaped. This means that at low and high levels of dispersion performance is poor, while performance is best at mid levels of dispersion. Performance is maximized at a coefficient of variation of around .25 in this sample, which is slightly above the average CV of .22.
These results are based on OLS regressions and are insensitive to several robustness checks. The authors also fit a more flexible spline model with a cutoff at a CV of .25. This indicates that moving from a CV of .20 to .25 could increase profits by 200 EUR, and similarly moving from .25 to .30 could decrease profits by 200 EUR. The results are also robust to alternative measures of ability dispersion.
Finally, note that teams could dismiss members. It is found that the number of dismissals was lowest for medium ability dispersion teams (at a CV of .24), which is very close to the CV that maximizes team performance (.25). It may be that medium ability teams dismiss fewer people because they perform better, or it could also be the case that these teams are inherently more stable for some reason.
Thus it seems we have a nonlinear relationship between ability dispersion and performance. Diversity is good up until some point, then it becomes detrimental. To explain these findings, the authors build a simple model where teams are composed of a given number of individuals. The team has to solve a set of tasks. Members of the team can have various ability levels, but they display imperfect assortative matching. Assortative matching means that people with the same ability prefer to work with each other (i.e. on the same task) as opposed to people of different abilities. Imperfect here means that at each ability level there is at least one person who is only willing to work with people of the same ability, even if this means working alone. This means that if in a team there are people with d different ability levels, then that team will work on at least d different tasks. So if a team wants to work on many different tasks, it can be diverse. But if they want to concentrate on few tasks, it needs to have low ability dispersion because of assortative matching.
There are more productive and less productive tasks, i.e. tasks that contribute more or less to the total team output. The team can assign many people to one (very productive) task and few to no people to other (less productive) tasks, or it can spread out team members across tasks. The former is referred to as specialization and represents a situation where teams are not diverse, ability dispersion is low. The latter is diversification, and it represents higher diversity within the team. This is because the number of tasks the team concentrates on is directly related to ability dispersion as explained in the previous paragraph.
The team’s objective is to maximize the total output of the team, which is simply a sum of the performances on the different tasks. The authors therefore abstract from complementarities between different tasks.
It is illustrated with this simple model that whether diversification is beneficial depends on the productivity difference between the various tasks the team needs to perform. If the most productive task is much more productive than the second most productive task, then the team’s better off letting everyone work on the most productive task (and hence needs to be homogeneous as opposed to diverse). If the productivity difference is not so high, then letting some people work on task 2 pays off.
The reason is that putting more people on the same task has diminishing returns. So if the less productive task is not very unproductive, it might be more beneficial to let people work on that instead of having even more people work on task 1. So in sum, in environments where the importance/productivity/contribution of each task is relatively similar (different), team performance could benefit from more (less) diverse teams.
So there you have it. The optimal team diversity depends on how large the productivity differences are between the various sub-tasks the team has to perform.
While this is clearly a simplified analysis of a complex question, the results still hold if ability levels actually influence performance of team members (a fact that is abstracted from in the baseline model), and if the size of the team as a whole is endogenized. So all in all, this seems to be a theory that may have some usefulness in the business world despite its clear limitations. Obviously some more controlled experiments are needed. For instance, one could vary task productivity dispersion to see if the benefits of ability dispersion indeed drop when productivity differences among tasks increase. Some future work for experimental economists to undertake.