How Bias Works

It’s pretty clear that most bias happens at an unconscious level and that things like racial and gender bias are much more pervasive than we’d like to believe. Check out some of the implicit association tests available online if you don’t believe me. But some of the details of how this plays out are downright scary. Consider the medical example cited in Shankar Vendatam’s column on unconscious bias:

If you ask people whether men and women should be paid the same for doing the same work, everyone says yes. But if you ask volunteers how much a storekeeper who runs a hardware store ought to earn and how much a storekeeper who sells antique china ought to earn, you will see that the work of the storekeeper whom volunteers unconsciously believe to be a man is valued more highly than the work of the storekeeper whom volunteers unconsciously assume is a woman. If you ask physicians whether all patients should be treated equally regardless of race, everyone says yes. But if you ask doctors how they will treat patients with chest pains who are named Michael Smith and Tyrone Smith, the doctors tend to be less aggressive in treating the patient with the black-sounding name. Such disparities in treatment are not predicted by the conscious attitudes that doctors profess, but by their unconscious attitudes — their hidden brains.

There are a lot of quality issues in our health care system beyond the basic access issues, and in an increasingly diverse society this sort of thing is not helping to resolve them.



On the other hand, Austin Frakt points me to Amitabh Chandra and Douglas O. Staiger “Identifying Provider Prejudice in Healthcare”:

We use simple economic insights to develop a framework for distinguishing between prejudice and statistical discrimination using observational data. We focus our inquiry on the enormous literature in healthcare where treatment disparities by race and gender are not explained by access, preferences, or severity. But treatment disparities, by themselves, cannot distinguish between two competing views of provider behavior. Physicians may consciously or unconsciously withhold treatment from minority groups despite similar benefits (prejudice) or because race and gender are associated with lower benefit from treatment (statistical discrimination). We demonstrate that these two views can only be distinguished using data on patient outcomes: for patients with the same propensity to be treated, prejudice implies a higher return from treatment for treated minorities, while statistical discrimination implies that returns are equalized. Using data on heart attack treatments, we do not find empirical support for prejudice-based explanations. Despite receiving less treatment, women and blacks receive slightly lower benefits from treatment, perhaps due to higher stroke risk, delays in seeking care, and providers over-treating minorities due to equity and liability concerns.

So maybe doctors know what they’re doing after all. Glad to read something reassuring for once.