Via Ted Frank we have this:
The implicit association test ... is an excellent example [of the replication crisis in social "science"]. Banaji and Greenwald claim that the IAT, a brief exercise in which one sits down at a computer and responds to various stimuli, measures unconscious bias and therefore real-world behavior. If you score highly on a so-called black-white IAT, for example, that suggests you will act in a more biased manner toward a black person than a white person. Many social psychologists view the IAT, which you can take on Harvard University’s website, as a revolutionary achievement, and in the 20 years since its introduction it has become both the focal point of an entire subfield of research and a mainstay of diversity trainings all over the country. That’s partly because Banaji, Greenwald, and the test’s other proponents have made a series of outsize claims about its importance for fighting racism and inequality.
The problem, as I showed in a lengthy rundown of the many, many problems with the test published this past January, is that there’s very little evidence to support that claim that the IAT meaningfully predicts anything. In fact, the test is riddled with statistical problems — problems severe enough that it’s fair to ask whether it is effectively “misdiagnosing” the millions of people who have taken it, the vast majority of whom are likely unaware of its very serious shortcomings. There’s now solid research published in a top journal strongly suggesting the test cannot even meaningfully predict individual behavior. And if the test can’t predict individual behavior, it’s unclear exactly what it does do or why it should be the center of so many conversations and programs geared at fighting racism.