In 2004 two economists sent out thousands of fictional résumés in response to job ads in Boston and Chicago. The CVs were matched for quality. Only the names differed. Applicants with names like Emily and Greg received about 50 percent more callbacks than those with names like Lakisha and Jamal. Nobody involved had to hold a conscious grudge for that gap to appear. This is the kind of result that put implicit bias into everyday vocabulary.
What implicit bias means
The idea is that people carry automatic associations, absorbed from culture, that can shape snap judgments without their awareness. A hiring manager might sincerely believe in fairness and still read one CV as more promising than an identical one. The associations aren't necessarily endorsed. They're just there, like an accent.
The best-known tool for measuring them is the Implicit Association Test, introduced in 1998 by Anthony Greenwald and colleagues. It times how quickly a person pairs words and faces. Faster pairing of, say, one racial group with positive words than another is read as a stronger association.
What the test can and can't tell you
This is where the research gets less tidy than the headlines. Millions of people have taken the test online, and many are told they hold a bias they didn't know about. But scores for the same person can shift noticeably from one sitting to the next, so the test is a weak guide for an individual. It works better as a research tool when averaged across large groups.
Meta-analyses have found that IAT scores predict real behaviour only weakly. The relationship is small, and in some studies close to zero. That doesn't make the underlying phenomenon imaginary. It means the instrument is a blunt one, and researchers still argue about what it captures.
The evidence that doesn't depend on the test
Field experiments like the résumé study measure discrimination directly, by watching what employers, landlords or lenders do when only one detail changes. Many of these audit studies, across several countries and decades, find unequal treatment on the basis of race, ethnicity, gender or age. They don't prove the cause is unconscious. Some of it is deliberate. What they show is that the outcome is real whether or not the test works.
Does bias training work
Employers have spent heavily on training sessions. A large 2019 review of dozens of experiments found that interventions could shift scores on implicit measures a little, but those shifts didn't reliably carry over into changed behaviour, and effects tended to fade. A one-off workshop is unlikely to change a hiring pattern.
What seems to do better is changing the process. Structured interviews, where every candidate gets the same questions and is scored against set criteria, cut down the room for snap judgments. So does removing names and photos from early screening, and tracking outcomes by group so that gaps become visible. These fixes assume that bias will creep in, and design around it.
A fair reading
Implicit bias is a useful label for a real pattern, but the popular version overpromises. The test isn't a reliable personal diagnosis, and training alone is a thin remedy. Discrimination in hiring and other decisions is documented, though, and it responds better to better systems than to better intentions.
No comments:
Post a Comment