Science magazine touts the existence of strong and ubiquitous “implicit bias”, as well as the need to measure it and develop ways to eliminate it

March 5, 2023 • 10:45 am

This long new article in Science, one of the world’s premier science journals (or should I call it a “magazine”?), not only assumes that implicit (unconscious bias) is a real and pervasive thing, that it’s ubiquitous and has led to “structural racism” within institutions, that it can be assessed with a test (IAT: the “implicit association test”) that it diagnoses the severity of your bias correctly, and that making people aware of their hidden bigotry through “implicit bias training” will make them stop being bigots.

Every one of these assertions (particularly the last three) has been contested by psychologists, but you wouldn’t know that from reading the piece, which you can do by clicking below. Note that the point of this post is not to say that bias and bigotry don’t exist—you’d be a fool to claim that—but that there are grounds for questioning whether these biases are unconscious, can be diagnosed correctly with tests, and can be eliminated through training.

I’m not an expert on this topic, but I am aware that the notion of testing for and effacing unconscious bias has largely been dropped by experts—but not by “progressives”. For a documentation of the problems, with plenty of references, read this 2022 piece by Lee Jussim in Psychology Today (click on the screenshot). I’ll give his list of issues below, but note that even Scientific American has been allowed by an editor to publish a 2020 piece called “The problem with implicit bias training“, which contends that yes, implicit bias exists, but training doesn’t seem to get rid of it.

Here are Jussim’s reasons, each documented with references in the original article. The indented words and bolding are his (my comments flush left).

1. The peer-reviewed scientific literature has witnessed a great walking back of many of the most dramatic claims made on the basis of the IAT and about implicit social cognition more generally.

2. There is no consensually-accepted scientific definition of implicit bias.

3. The IAT measures reaction times, not things that most people think of as bias.

If you want to see what the IAT is all about, you can take a test here.

4. At best, the IAT measures the strength of association of concepts in memory.

5. The IAT may capture prejudice, stereotypes, or attitudes to some degree, but, if so, it is not a clean measure.

6. The IAT, as used and reported, has a potpourri of additional methodological and statistical oddities.

7. Many of the studies that use IAT scores to predict behavior find little or no anti-Black discrimination specifically.

8. Whether IAT scores predict behavioral manifestations of bias beyond self-report prejudice scales is unclear, with some studies finding they do and others finding they do not.

9. Procedures that change IAT scores have failed to produce changes in discriminatory behavior.

10. There is currently no clear evidence that implicit bias trainings accomplish anything other than teaching people about the research on implicit bias.

11. There is no evidence that IAT scores are “unconscious.”

This last one, which claims that people are very good at predicting their IAT scores, suggests that while people may be biased, it’s not unconscious (this is the main point of the “implicit” trope: that people may think they’re unbiased but they’re not, and therefore act out their racism constantly).

12. Critiques and discussions of its limitations or weaknesses are often not presented when the IAT is taught to introductory psychology students.

This is the main flaw with the Science article above. The only caveats it offers are that people’s scores aren’t often replicable, and that “simple interventions can dampen biases. . . but the changes are usually modest and don’t persist.” Well, that’s an admission of sorts, but the rest of the article is predicated on the existence of these biases and on the need for new ways to find them and eliminate them.

Jussim’s conclusion:

Here is my advice to you: Take an IAT or two (which you can here) if you have not already, just to see what the buzz is about. But now you are armed with enough information to reject any simple-minded proclamations about unconscious racism or the supposed power of implicit biases.

Again, Jussim gives copious references for his claims.

Now the one thing I’m not sure about is whether there really is a psychological trait of “implicit” bias that people aren’t aware of. It may be possible to have prejudices that linger in your unconscious (as opposed to prejudices you’re aware of but won’t admit).  When I saw the first claim below in the Science piece, for example—a study that’s gotten tons of press—I wondered if there really might be some unconscious bias leading to the effect, for the white doctors would surely assert that they don’t treat patients differently because of their race:

A 2020 study by Rachel Hardeman, a reproductive health equity researcher at the University of Minnesota’s Center for Antiracism Research for Health Equity, and colleagues showed Black newborns are twice as likely to die in the care of a white physician than a Black doctor, for instance. Another study from 2022 found women and people of color with chest pain wait longer to be treated in the emergency room compared with white men.

Of course there are competing explanations: the chest pain difference could be due to triaging of symptoms not based on race (I haven’t read the article). But I have read the other article, and the difference in death rate of black babies cared for by white versus black doctors really is a cause of deep concern: it’s a huge difference! Is it possible that it’s due to unconscious racism affecting treatment of infants? But before we go accusing white pediatricians of wholesale and strong racism that kills babies, however, we should immediately begin to pose questions about controls and the like. It turns out, though these results haven’t yet been published, that there apparently were no proper controls in this study, and there were several unassessed non-bias factors that could explain the results. We should withhold judgment about the infant-death study—and not tout it as an example of egregious racism—until other statisticians and physicians have weighted in in the literature.

The Science paper documents other inequities (disproportional representation) in science and medicine, but, as always, we have to ask whether there are other explanations for them and not just bias, much less implicit bias. At times, as in the paragraph below documenting new ways to find implicit bias, there’s a telling assumption that it’s always there but sometimes hard to sniff out. This is, after all, science journalism (bolding is mine):

Others using computer software to research implicit bias in medicine are also struggling to give physicians meaningful feedback. Nao Hagiwara, a social and health psychologist at Virginia Commonwealth University, and her team are analyzing dozens of nonverbal and verbal communication behaviors, such as facial expressions and voice changes, in recordings of primary care physicians’ interactions with people who have type 2 diabetes. Their software hasn’t yet identified behaviors that could clearly be linked to bias or had an adverse effect in the patient’s outcome. One reason for this murkiness, Hagiwara suggests, is that multiple different cues likely interact to influence patient outcomes whereas studies so far tend to analyze the impact of only one behavior at a time.

Note that the failure of the test to detect “implicitly biased behaviors” doesn’t even mention that such behavior might not exist, but assumes that the researchers haven’t yet hit on the right metric or combination of behaviors (of course, that could lead to p-hacking). That paragraph itself is biased towards the existence of biased behaviors.

Finally, here’s Science‘s pronouncement (through  at the end that implicit bias exists, that it’s strong and pervasive, has resulted from “centuries of white supremacy”, and it’s structural.

Sustained implicit bias training for physicians should instead be the norm, some emphasize. Hospitals also need to monitor and collect data on health care outcomes for different groups in order to monitor equity, Sabin says. “You have to know where the disparities lie and then begin to work backwards from that.”

It won’t be easy, Hardeman says, noting that, at least in the United States, centuries of white supremacy and other forms of bigotry have resulted in deep-rooted stereotypes and other implicit biases. “Every single person should be thinking about doing this work,” she says. “But if they’re doing it within a system that hasn’t addressed its own biases and racism, then it’s not going to be fully effective.”

Among the many things to worry about in this article, including some of its disturbing assertions about racism that do bear investigating (e.g., the infant-death research), is the lack of balanced coverage of this topic and an almost complete neglect of the points Jussim notes above (his name isn’t mentioned in the piece). Such science journalism is, well, unscientific.

Here’s Jussim giving a remote Merton Seminar on the problems with implicit bias.

As always, read the article, Jussim’s piece and the literature he cites, dig further into the topic, and draw your own conclusions.

19 thoughts on “Science magazine touts the existence of strong and ubiquitous “implicit bias”, as well as the need to measure it and develop ways to eliminate it

  1. I’m not sure why, but I am reminded of :

    [ begin quote]
    [ emphasis my own ]

    And on the pedestal these words appear:
    “My name is Ozymandias, king of kings:
    Look on my works, ye Mighty, and despair!

    [ end quote ]

    From Ozymandias,
    By Percy Shelley, “Ozymandias”, 1819 edition[18]


  2. Re the infant-death study: such studies have to be careful about confounding factors. Suppose:

    1) The fraction of doctors who are black has been rising over time (on a quick google this is true, mostly from a marked rise in black women becoming doctors).

    2) The difficult cases (baby most likely to die) tend to get assigned to the most experienced, later-career doctors (of whom, given 1, a larger fraction are white).

    That alone could produce a differential death rate.

    (I’m not saying this is indeed part of the explanation, just that it’s the sort of thing that needs to be considered; on a quick scout of the paper they don’t seem to have controlled for seniority of doctor. The paper does say that the differential death rate is higher for more-complex cases, which you’d expect under the above hypothesis. Perhaps, though, others here know much better about this than I do.)

    1. The study was a retrospective trawl through an administrative database of hospital discharges in Florida. There were no chart reviews of the newborns who died to determine if failure to rescue played a role. (There were likely internal reviews of each death at the time. What I mean is that authors of this study didn’t look at the records of any individual infants.). Even if delivering mothers were assigned to paediatricians by the luck of the draw as to who was on call, it is still possible that babies in distress, who are the ones who die, were transferred immediately to (non-black) doctors in the ICU, who would then “own” the death. The death rate in ICUs is higher than any other place in the hospital. ICUs must be killing people, then, right?

      How did the authors determine the race of the doctors? Florida doesn’t register its licensed physicians according to race. The authors looked for pictures in various commercial databases like RateMyDoctor. If the doctor was “clearly” white or “clearly” black, they included that doctor-baby pair in the analysis. If the photo was some other race or “Damfino —s/he could pass as either”, they excluded that doctor-baby pair. And of course, photographically black does not necessarily mean legacy generational slavery black. It would include highly qualified immigrant doctors recruited for their expertise from African countries or the UK. Highly qualified doctors of other photographic races would be excluded.

      Interestingly, there was no evidence of a training effect. White doctors in hospitals that delivered a lot of black babies didn’t get any “better” at it than in hospitals that delivered only a few, counter to the benefits of experience we see everywhere else. Racist callousness must die hard, then.
      So all the usual problems of retrospective analysis of big data.

      But here’s a thought. Suppose it is true that, try as they might, white doctors just can’t reproduce the “special sauce” that black doctors bring to the care of black babies, who have terrible perinatal outcomes no matter who looks after them. Photographically white paediatricians outnumbered photographically black by 8:1. Who is going to look after all these black babies if whites are excluded (or get out of a business where they are not welcome)? Or, how black do you have to be to be allowed to train as a “black” pediatrician? Will a student from Ghana or Nigeria go ahead of or behind a student from Memphis?

    2. “The paper does say that the differential death rate is higher for more-complex cases, which you’d expect under the above hypothesis.”
      I’d say the seniority hypothesis would just mean that more of the complex cases get assigned to a white doctor. But the relative rates of complex-case neonates dying in the care of race W or race B should not be affected by this. (Unless their category “complex cases” is too broad and doesn’t capture decisive differences)

  3. Since so much crap is disseminated under the “DEI” label, I now tend to roll my eyes at any mention of this sort of thing—in other words, incessant hectoring on these matters has conditioned in me an implicit bias against phrases like “implicit bias”, and everything that goes with them. I wonder if acolytes of DEI-speak have ever thought about the effects of the incessant Marxist-Leninist hectoring that typified life in that galaxy far away, or have considered its effects on the free 1989 Sejm elections in Poland?

  4. For consultants, the truly excellent aspect about implicit bias is that it is a perfect nexus of financial opportunity with literally NO risk. A consultant will always find bias in her subjects (because they are human) and, if the bias is denied, even with facts, the consultant can claim that the client just doesn’t realize he is biased. But…, if the client will sign up for the consultant’s multi-week, high cost set of workshops, the bias can be ‘found’ and eliminated if the client does the hard work of educating himself. Oh, by the way, there is no metric to indicate when a successful outcome is achieved.

    Here’s your invoice, sir.

    1. This looks to me like the same sort of thing as the search for repressed memory. If you look hard enough you will “find” it, because of the large role simple mental imagery plays in the process, and the fact that the therapist is not looking for, nor has any idea what would constitute, evidence that would contradict the hypothesis. If at first you don’t find the repressed memory, accuse the patient of not cooperating and look more vigorously.

      The assumption that implicit racial bias exists- *that* is the primary implicit bias.

  5. This is incredibly disturbing. People who are committed to a dogmatic worldview (DEI) are enabled to study (validate) their dogmatism because of the current moral fashion. Going by trends on how poorly social science and medical research is conducted (refer to the economist for the latter), and given the bias of these people (or how easy it is to inject bias into this kind of research), it’s hard not to believe that the end result, even if it fails to show what it intended, would like the IAT, be touted as “evidence” of “structural racism” and “implicit bias”. With such a strong and (highly improbable) claim like this, this is bound to fuel further unnecessary division. Arguably, what’s being done here (I’m betting) would become a hallmark example of “scientific racism”.

  6. To add to the chorus, I’m at the point in my life where my assumption, based on 70+ yrs of “lived experience” (although no unlived experience), is that most of those who study implicit bias already assume it exists, and thus will continue indefinitely to find it. Meanwhile, they or their colleagues will offer unproven (but costly) remedies to fix it, which won’t work, as they’ll keep digging until they find more evidence that it still exists, so the trainers can keep on training, and claim perpetual “harm” and victimhood.

  7. Implicit bias (IAT) from a cognitive perspective:
    I’ll admit that I am a fan and supporter of Lee Jussim’s and admire the work he has done ferreting out bias and bull***t throughout social psychology. He’s a model all social scientists should emulate.

    However, there is more to the Implicit Bias debate than is contained in his article of your review. My dissertation at Oxford University in 1986 showed that individuals’ access to the information they used to perform complex tasks (computer games) is often unavailable for verbal reports (i.e., it is unconscious). Here’s a subsequent publication: Porter, D.B. (1991). Computer games and cognitive processes: two tasks, two modes, too much? British Journal of Psychology, 82. 343-357. This finding is common in experimental cognitive psychology. Bob Zajonc, Donald Broadbent, and Nobel Laurette, Daniel Kahneman among many others have demonstrated a dissociation between what people say and what they do (i.e., Kahneman’s Thinking Fast and Thinking Slow).

    Experimentally, differences in conditions are usually greater than differences between subjects. This is a falty assumption of most “trainers” and “training programs.” IAT scores change as the situation changes not as individuals are changed through “learning”. In the previous decade, 30 of my undergraduate research students at Berea College earned competitive recognition for the quality of their senior research projects. Many of these projects used the Harvard Implicit Associations Test as a dependent measure of the effect of changing situations.

    Here is one example: Two groups of students (black and white males) performed a series of games on a computer with a partner who was a confederate of the experimenter and had mastered these tasks. Half of the subjects performed the tasks competitively (a victory for one player was counted as a loss for his playing partner) and half performed the tasks cooperatively (the scores of the two partners together would be used to determine prizes awarded at the end of the experiment). Half of the confederates were white and half the confederates were black. This was a 2X2X2 design with race of the subject, race of the confederate, and performance condition as the three dichotomous independent variables. Subjects completed IATs before and after the experiment (it takes about 15 minutes). The results showed that all subjects’ biases against other races increased when they had been repeatedly beaten in competitions with a member of the other race but decreased when they had been cooperating with a member of different race. The situation (whether competitive or cooperative) had no effect on IAT scores when the subjects were paired with members of the same race.

    Implicit biases are real (and often unconscious). However, they vary across situations to a much greater extent than they do across individuals. Perhaps the best training programs would be ones that helped individuals recognize the situations which might increase or decrease their biases.

      1. If “implicit bias” means “any bias of which people are unaware,” and includes things like Kahneman&Tversky heuristics, self-serving biases, hindsight biases, confirmation biases and the like, sure. It has been apparent that people do not always have conscious access to their cognitive processes (bikeriding anyone?) since Nisbet & Wilson’s classic 1977 article.

        None of this is what is usually meant when they use the term “implicit bias” in 2023. Implicit bias, as in the article Jerry reviews, involves various isms (racism, sexism), demographic phobias (Islamo, trans) and the like. Jerry’s critique is a bullseye.

  8. It’s unfortunate that Science is taking a position on implicit bias, even though the evidence is weak for the effectiveness of measurement and for the value of efforts at remediation. It’s not a good look for a journal that is otherwise so rigorous. The editors must be feeling the DEI pressure. In so doing they are failing to uphold the standards of the discipline that is their namesake. Sad.

  9. I have a conspiracy theory that the belief named “implicit bias” hijacks more credible/and testable forms of bias, such as “availability bias”, gaining near instant respectability, and then reinforcing itself via an “availability cascade”

    “An availability cascade is a self-reinforcing cycle that explains the development of certain kinds of collective beliefs. A novel idea or insight, usually one that seems to explain a complex process in a simple or straightforward manner, gains rapid currency in the popular discourse by its very simplicity and by its apparent insightfulness. Its rising popularity triggers a chain reaction within the social network: individuals adopt the new insight because other people within the network have adopted it, and on its face it seems plausible. “

  10. If academics (authors) created a list of journals that they would refuse to submit articles to, en masse, many if the issues you post about could be reversed. One time high and mighty publications would be left with drek while lesser known ones would grow, and competition would force change. This would require a galvanizing force like you to create focus. But it is doable.

  11. I accept that there is often implicit bias that is not amenable to simple introspection… but what if some implicit bias is unfounded and some is well founded?

    I might have a bias about walking through a red light area at night for fear of violence –
    and because it is ‘implicit’ I might not realise that I have modified my behaviour to avoid such areas ‘automatically’. But is the bias well founded or unfounded? Is my biased behaviour primarily ‘racist’ or not? And would training cause me to change my behaviour?

  12. One of the many topics addressed by skeptics is whether believing in the supernatural/God is instinctive. Do we feel as if there are higher powers? Is it normal to interpret significant events in our lives as rewards and punishments, or perhaps signs of what we’re supposed to do?

    The answer seems to be ‘yes.’ And what brain scans apparently revealed was that most of the time we unconsciously override these brief superstitious defaults with reason. Atheists obviously do it more often. We don’t “know there’s a God.” We ignore unconscious biases and instantaneously substitute assessment.

    Does this have any connection to implicit racial biases? I don’t know. But the assumption that an instinctive reaction is what we “really believe” and we’re going to act on it needs to be carefully examined.

Leave a Reply