Bias≠Discrimination

A recent Nature Magazine article (Scale dichotomization reduces customer racial discrimination and income inequality) by a team from Yale University, Rice University and University of Toronto - Rotman School of Management reports the results of a very interesting study on how the rating scale on online platforms affects racial discrimination, and how moving from a typical 5-star scale to a 2-point rating scale can reduce discrimination.

An accompanying commentary provides the figure below: the authors "analyzed performance-rating data from an online platform in which customers can hire workers for home services and then rate their satisfaction with the worker. The platform made a sudden switch from using a five-star rating system to a binary ‘thumbs-up or thumbs-down’ system. Before the switch, workers who are people of colour were less likely than white workers to receive the top rating; but after the switch, ratings equalized. Further experiments showed that the binary rating system did not enable the expression of racist beliefs in the same way that the five-star rating did, and that the rating equalization translated to an equalization in income".

But the commentary swapped the word "discrimination" for the word "bias", including in the title (Racial bias eliminated when ratings switch from five stars to thumbs up or down). But they are very different things: bias is (implicit or explicit) belief; discrimination is the tangible consequence of bias, that is, people being treated differently. The binary rating system shows that it reduces or even eliminates discrimination (a good thing) but I doubt that it reduces or eliminates bias. Switching to the binary rating scale is the equivalent of rounding to the nearest integer: on a 5-star scale, you'll assign 4 stars, but when translated into a 0-1 scale, it is 0.8, which is rounded up to 1. The feeling is still the same, but the outcome is different.

Still, let's take the victory: this "decision architecture" is a wonderful behavioral economics nudge that can produce real, tangible results. It is especially powerful in situations of unconscious biases.

PS: Bias, unconscious bias and discrimination are three of the expressions, among a collection of them, used to flag "woke" activities in the federal government (according to a The New York Times article today).