Diversity only marginally boosts accuracy of group’s predictions

February 14, 2018
Contact: Jared Wadley jwadley@umich.edu

Icons of diverse people

ANN ARBOR—Diversity for boards, juries and other influential decision-making teams can help ensure that the interests of a diverse population are fairly represented and addressed.

But for situations that call for predictions or estimates, there is typically little performance benefit for using a diverse group compared with one with similar individuals, a new University of Michigan study found.

U-M psychology researchers used surveys and simulations to replicate previous findings indicating that small crowds can be wiser than individuals, but it matters far less whether the crowd is demographically homogeneous or diverse.

The study’s authors, Stephanie de Oliveira Chen and Richard Nisbett, looked at social diversity—individual characteristics, such as gender, race, interests, religion and background—and cognitive diversity, which takes into account people’s judgment or how they think.

People often believe that social diversity boosts cognitive diversity in a group. However, there can be substantial cognitive diversity within demographically homogeneous groups.

Researchers measured the social factors and judgment among study participants who completed nine judgment tasks. For example, they predicted the points that would be scored in a rival football game, votes for presidential candidates, opinions on political statements and ratings on two dozen books.

From people’s estimates in the various tasks, crowds were created by averaging eight randomly selected participants—which then comprised diverse and homogeneous groups.

Among the findings:

  • The homogeneous groups produced nearly the same results as the diverse group when it involved numerical judgments. Diverse groups were sometimes more accurate, but typically by a small margin.
  • People who expect social groups to think differently in these types of judgments may be erroneously stereotyping. The differences between how social groups think can be much smaller than expected.

“In other words, not all women think alike, not all liberals think alike, and so forth,” said Chen, a postdoctoral researcher and study’s lead author.

The researchers note that the findings do not mean that no social factors correlate with any types of judgment. The connection between demographics and numerical judgment is often weak, thus making it challenging to say in every situation that diverse crowds are wiser than homogenous ones, said Nisbett, the Theodore M. Newcomb Distinguished University Professor Emeritus.

The findings appear in the Proceedings of the National Academy of Sciences.

For the article, contact the Proceedings of the National Academy of Sciences News Office at pnasnews@nas.edu or 202-334-1310.