Do we really know what implicit bias is, and whether we have it?
This is the second episode on our two-part series on implicit bias; the first part was an interview with Dr. Mahzarin Banaji, former Dean of the Department of Psychology at Harvard University, and co-creator of the Implicit Association Test.
But the body of research on this topic is large and quite complicated, and I couldn't possibly do it justice in one episode. There are a number of criticisms of the test which are worth examining, so we can get a better sense for whether implicit bias is really something we should be spending our time thinking about - or if our problems with explicit bias are big enough that we would do better to focus there first.
Jump to highlights:
[accordion-item title="Click here to read the full transcript"]
Hi, I'm Jen and I host the Your Parenting Mojo podcast. We all want her children to lead fulfilling lives but it can be so hard to keep up with the latest scientific research on child development and figure out whether and how to incorporate it into our own approach to parenting. Here at Your Parenting Mojo, I do the work for you by critically examining strategies and tools related to parenting and child development that are grounded in scientific research and principles of Respectful Parenting. If you'd like to be notified when new episodes are released, and get a FREE Guide called 13 Reasons Why Your Child Won't Listen To You and What To Do About Each One, just head over to YourParentingMojo.com/SUBSCRIBE. You can also continue the conversation about the show with other listeners in the Your Parenting Mojo Facebook group. I do hope you'll join us.
Hello, and welcome to the Your Parenting Mojo podcast. Before we start this week's episode, I wanted to take a minute to thank you for being a part of this parenting journey with me and to share a quick update on where things stand with the podcast after four and a half years now. What is that saying? The days are long and the years are short? It certainly seems to be the case here. And well for some of you listening, this may be the very first episode that you're listening to, there are many others who have been with me for the entire 132 plus episodes that I've created to date. We're close to surpassing a million and a half downloads from all around the world, and my goodness, it's a bit strange to even say those words aloud given that I started the show with basically no idea whether anyone would be interested in listening. And it's such an honor to me when you recommend the show to your friends and to other parents at your daycare or preschool. When you share specific episodes that have helped you to find the answers that work with your family and your online communities. And I'm not exaggerating when I say that developing Your Parenting Mojo, which is now the podcast episodes, blog posts, courses, workshops, membership content is more than a full-time job. I have a very small team that helps me to keep my own sanity and my husband is now involved as well. Maybe one day he'll even listen to as many episodes as some of my most dedicated listeners have. Even my daughter now signs off on her videos at home with the brought to you by YourParentingMojo.com. And as the word continues to spread, more and more parents are making transformative lasting change using the methods that we talked about here on the show. We are big cohort in the current Taming Your Triggers workshop who are supporting each other and understanding the sources of their triggered feelings. And just a week or two in they were already reporting the ability to create space for responses to their children that fit with their values, where that just hadn't seemed possible before. And I'm very happy to share that the Parenting Membership open enrollment is just around the corner and will run from May 2 through May 12 in time for celebrating Mother's Day, while the learning membership is going to reopen later in the summer. The community of parents that have already enrolled in the Parenting Membership continues to grow and it's a great resource of knowledge and support and community to help anyone who's looking to apply the ideas and strategies that you hear on the podcast in your own family at home. So I'll share more as we get closer to open enrollment. But for those of you who want to learn more or to join the waitlist, feel free to go to YourParentingMojo.com/ParentingMembership. And thanks again to each and every one of you who have joined me on this parenting journey over the last four and a half years. I learn and I grow every day as a parent and it's amazing to share all this with you.
So let's get into our real topic of the day, which is part two of our miniseries on implicit bias. Today we're going to continue to open a real can of worms that's already half opened, and that I had no idea that I was about to open when I started researching this episode. So I've been interested in the connection between the brain and the body for some time now, and I'm planning a series of episodes to explore this topic in some depth. And one aspect of this is related to knowledge that we hold in our bodies rather than in our brains. And as I started to explore that idea, I was thinking about intuition, which we often experience as a felt sense in our bodies rather than a decision that we make in our brains. And that led me down a path toward understanding the role our gut plays in what we know. But another branch of this path led me to the topic of implicit bias because I was thinking, Okay, if we start relying on our brains a bit less and trusting our bodies a bit more, how can we know that this will lead us in a direction that's actually aligned with our values? What if implicit bias means that listening to our bodies actually takes us away from living our values because our bodies have implicit bias baked in so deeply?
Now I've read the book Blind Spot: The Hidden Biases of Good People by doctors Mahzarin Banaji and Anthony Greenwald some years ago, but I hadn't dug into the research behind it at the time. The book's premise is that we have visual blind spots, which are places where visual information is missing, but our brains fill in the gaps without us realizing it, which can be seen when a person whose visual cortex has been damaged and they can't see an object like a hammer in front of them, but if you ask them to reach out and grasp it, then they'll be able to do it. The author state that a parallel idea exists with non-visual stimuli as well, they say and I'm going to quote "Rather than an effective visual perception, this book focuses on another type of blind spot, one that contains a large set of biases and keeps them hidden just as patients who can't see a hammer can still act as if they do. Hidden biases are capable of guiding our behavior without our being aware of their role. What are the hidden biases of this book's title? They are for lack of a better term bits of knowledge about social groups. These bits of knowledge are stored in our brains because we encounter them so frequently in our cultural environments. Once lodged in our minds, hidden biases can influence our behavior towards members of particular social groups, but we remain oblivious to their influence. In this book, we aim to make clear why many scientists ourselves very much included, now recognize hidden bias blind spots as fully believable because of the sheer weight of scientific evidence that demands this conclusion. But convincing readers of this is no simple challenge. How can we show the existence of something in our own minds of which we remain completely unaware?" The book goes on to describe a test called the Implicit Association Test, or IAT, which was designed by Dr. Greenwald and modified by both authors over the years, starting when Dr. Banaji was Dr. Greenwald’s graduate student.
And so I'd known about what is the Race IAT for a number of years, there are actually now quite a few of them on topics ranging from age to religion to body weight. The landing page at implicit.harvard.edu, which is where these tests are housed, doesn't say much about the purpose of these tests. You have to click into the About the IAT page to see the description that "The IAT measures the strength of associations between concepts, for example, black people, gay people, and evaluations, for example, good and bad, or stereotypes, for example, athletic or clumsy. The main idea is that making your responses easier when closely related items share the same response key, the IAT scores based on how long it takes a person on average, to sort the words in the third part of the IAT versus the fifth part of the IAT. We would say that one has an implicit preference for thin people relative to fat people if they are faster to categorize words, when thin people and good share a response key, and fat people and bad share a response key relative to the reverse."
I think I’d even taken the Race IAT once before, a long time ago, which is the version of the test that most people assume you’re talking about when you mention the IAT. I don’t remember for sure what the result was but I assume it gave the result that it gives about 75% of people who take it, which is to say that they have implicit bias against Black people.
I came back to the book again and poked around on both Dr. Banaji’s and Dr. Greenwald’s websites to see what they had been up to recently. I noticed that the most recent publication on Dr. Banaji’s website was co-authored with Dr. Yarrow Dunham, whom we met way back in episode 28 on how social groups form, which is linked to implicit bias. I emailed him and asked if he would be kind enough to introduce me to Dr. Banaji, which he was, and she quickly responded to say that while she was too swamped for an interview she’d be happy to answer three or four questions. So: so far, so good.
Now, Pretty often when I reach out to researchers I’ve been able to get an overview of the concept and of their work but I haven’t yet done a deep dive, because if they say ‘no’ then I will have wasted several days of prep work as I’d have to find someone else to ask and then have to do a deep dive into *their* work. So after Dr. Banaji agreed to answer some questions I reread Blindspot, and thirty peer-reviewed papers on implicit bias, and some really well-written articles in the popular press (and some not so well-written ones as well), and then things got a bit difficult. I want to give a special shout-out to journalist Jesse Singal, who wrote a really excellent piece on the topic of the research behind the IAT in an online magazine called The Cut. I had it open in my browser for a while as I was digging into the peer-reviewed literature and when I finally came back to it I found he had already coalesced a number of the thoughts that were swirling in my brain and had pointed to many of the studies I had by then found “by myself,” and directed me to a few more besides. He also reached out to a number of the researchers who have been involved in quite a drama of dueling analyses about the IAT which has played out in peer reviewed journals over the last decade or so (dramas in the peer reviewed journal world tend to unfold in slow motion!). It’s been a drama because there seems to be a great deal of uncertainty about whether the IAT measures what it says it measures, and whether that has any practical significance in the real world. I ended up actually writing the majority of this episode’s content to try to explain to myself what this web of findings was, and I sent the text to Dr. Banaji along with some succinct questions and she responded that she had never met an interviewer who had done more in-depth preparation and would I like to talk? So I said yes, I very much would, and you’ve already heard the outcome of that interview in a recent episode, where hopefully you got a good grounding in what the IAT is and how it works.
And we'll cover a bit more on that in a few minutes. But I have to say that I was highly hesitant to try to do anything that looked remotely like challenging the research on which the IAT sits because of something that appeared toward the end of Jessie Singal’s article in The Cut. He emailed Dr. Banaji with some questions about the kinds of methodological questions about the IAT that we’re going to look at in this episode, and he says she “repeatedly asserted that criticisms of the IAT come from a small group of reactionary researchers, and that questioning the IAT is not something normal, well-adjusted people do.”
He goes on to quote her directly: “Of course it annoys people when a simple test that spits out two numbers produces this sort of attention, changes scientific practice, and appeals to ordinary people. Ordinary people who are not scared to know what may be in their minds. It scares people (fortunately a negligible minority) that learning about our minds may lead people to change their behavior so that their behavior may be more in line with their ideals and aspirations. The IAT scares people who say things like “look, the water fountains are desegregated, what’s your problem.”
By and large I operate on the view that I need to pay attention to REAL criticisms of the IAT. Criticisms that come from people who are experts – that is[,] people who understand the science’s assumptions about response latency measures. People who do original work using such methods. I’m sorry to say this but we are all so far along in our work that I at least only read criticisms from people who are experts. I don’t read commentaries from non-experts. There’s too much interesting stuff to do and too many amazing people doing it for me to justify worrying about a small group of aggrieved individuals who think that Black people have it easy in American society and that the IAT work might make their lives easier. The IAT as you know is not about any one group, but this small group of critics ignore everything other than race, and it may be a good idea to find out why that may be the case.”
Now, if you've been listening for a while, I’m sure you can see the red flags going off in my conflict-avoiding mind already – I’m not an expert; I don’t do original work in this area, and thus she wouldn’t read any commentary or criticism that I proposed, which is why I didn’t get into any of this in the conversation with her. So I want to be clear what we’re trying to do here. This is NOT an expose of the IAT, or of Dr. Banaji. I’m really grateful that she took the time to talk with me, and I’m grateful to Dr. Yarrow Dunham for connecting us. But I believe strongly enough that the interview we conducted doesn’t tell the whole story of the IAT that I’m going to try to dig more deeply on it here so we can understand whether implicit bias is really something we should pay attention to.
So, in this episode my plan is to walk through what the research says about what implicit bias is, and how it’s measured, and why that’s important, and what implications this has for how we and our children interact with the world. I should mention that the research base on this topic is absolutely massive, and even what has now turned into two episodes isn’t really enough to do it justice.
So our story starts with the creation of the Implicit Association Test, or IAT, and that story is told in Dr. Banaji and Greenwald’s book Blindspot. A substantial chunk of the book is about convincing you that you have blindspots – that you and I, just like everyone else, don’t perceive things in the way they actually happened. So, for example, if you’re on a jury in a murder trial where the defendant is accused of drunk driving, and an eyewitness says that the defendant staggered against a table and knocked a bowl to the floor as they left the party they were both at then you might believe the eye witness, but if the eye witness says “On his way out the door the defendant staggered against a serving table, knocking a bowl of guacamole dip to the floor and splattering guacamole on the white shag carpet,” you’ll be more convinced of the defendant’s guilt. Vivid details make information more memorable (which, incidentally, is why stories in children’s textbooks often contain little snippets of vivid detail – they make the material easier to remember, which helps the child pass the test more easily).
We also make judgements about individuals based on how they look, and while we might think we’re pretty good at it we’re actually not researchers have asked research participants to judge the trustworthiness of people shown in photographs, they don’t do much better than chance at discriminating between Nobel Peace Prize recipients and criminals who have appeared on the TV show America’s Most Wanted. A lot of the time we make judgements about people based on what social group they are in, which is why Dr. Dunham’s work is relevant here – you might want to revisit episode 28 on how social groups form for a refresher on that.
And then a chapter of Blindspot talks about how often we say things that aren’t true, even if we think we’re a pretty truthful person, such as the answers to the question “how are you?” which was a big cultural shock to me when I got to the U.S. because the person asking the question had often already walked by by the time I could even think about an answer, because they didn’t really want to know the answer to the question, which means the answer we give is pretty much irrelevant. We might lie and say we don’t have any cash on us when a person on the street asks us for money or tell a telemarketer that we aren’t home when we were the one who answered the phone. We may lie to the doctor when they ask how often we exercise, or how many drinks we have each day, or whether we’ve ever used illicit drugs. We try to manage the impressions that other people have of us, so if a researcher were to ask us if we are *always* courteous, even to people who are rude to us, or if we would *always* declare everything at customs, even if it would cost us and we could never be found out, then we’re pretty likely to lie. So you can imagine what happens when researchers ask people whether they’re racially biased – of course they’re going to say ‘no.’ But is that really the truth?
So here’s where things start to get murky. The IAT was designed to test people’s attitudes toward people of different races by asking them to sort pairs of pictures and words and measuring how long it takes to do that. So there are positive words like “terrific” and “cheerful” and negative words like “poison” and “hate,” pictures of Black and White people, and the person taking the test is instructed to press the E or I keys to match the phrases in certain ways. (I should note, though, that I had problems with the test partly because I don’t believe all the words on the test that I was supposed to think are negative are actually negative. Sadness is just an emotion, after all, and I don’t think of it as necessarily ‘bad’ – so I had to remind myself that it was ‘bad’ every time I saw it. The children’s version of the test says that snakes and ants are ‘bad’ – and I know my six-year-old would disagree with that.)
The entire premise of the IAT rests on the idea that when ideas are linked, like negative words and pictures of Black people, that a person who is biased against Black people will press the relevant key more quickly than a person who isn’t biased against Black people. Blindspot explains that “when categories can be linked to each other via shared goodness or badness, the shared property is what psychologists call valence, or emotional value. Positive valence can function as a mental glue that bonds these two categories into one.” The IAT aimed to bypass all the messiness of asking people about their biases toward certain groups of people, by “unveiling a type of mental content that we and other social psychologists at the time were just beginning to understand – hidden biases that could not possibly be tapped by asking questions because their possessors were unaware of having them.” About 75% who take the Race IAT are faster at linking the images of White people to pleasant words than they are at linking images of Black people to pleasant words. And now we get to the interesting stuff: Blindspot goes on to say that “The automatic White preference expressed on the Race IAT is now established as signaling discriminatory behavior even among research participants who earnestly (and we believe honestly) espouse egalitarian beliefs.”
Now, I'd like to pause for a moment here and say that of course, I don't doubt the existence of racism. I believe that structural racism exists, which is the system of structures that have procedures or processes that disadvantaged people of a certain race. An example of this would be systems that allow certain banks to fast track their clients applications for paycheck protection program loans that just happened to favor banks with a lot of white clients. I also believe explicit racism exists racism that is overt and intentional, and that's held on to an unconscious level. But does implicit bias exist and does it measure it? So let's dig in to find out more on that.
Before we get any deeper on this we should probably say that there are actually conflicting definitions of what implicit bias even is. I’ve seen two definitions from Dr. Greenwald himself – in one paper co-authored with Dr. Banaji he says that “the signature of implicit cognition is that traces of past experience affect some performance, even though the influential earlier experience is not remembered in the usual sense—that is, it is unavailable to self-report or introspection,” but in another written much more recently he says that “The currently dominant understanding of “implicit” among social cognition researchers is “indirectly measured.”’ So is it the thing we’re measuring, or the way we’re measuring it?
Now, I did ask Dr. Banaji about this when we talked and honestly I didn’t come out of it with much more clarity. I’m still not clear on whether implicit bias is something we’re aware of but is measured indirectly or is something that we’re not aware of at all, and that’s a pretty important distinction. If I’m fully aware of my racial bias but I’m just not willing to admit it to a researcher, or even check a radio button in the introduction to the IAT that says I believe I have racial bias which is why my bias needs to be measured indirectly, then the kinds of things I can do to reduce that level of bias are going to be a whole lot different than if I have no idea whether or not I even have implicit bias.
So there’s the issue of what implicit bias even is, but after we look at that we have to try to understand what effect having implicit bias has on behavior. Drs. Greenwald and Banaji and colleagues published a meta-analysis of the predictive validity of the IAT in 2009. Predictive validity tells you how well a certain measure can predict future behavior, and these authors found that the IAT had an overall correlation of 0.24 for prediction of behavioral, judgement and physiological measures in a sample of 14,900 subjects who had taken the IAT online, which was about double the predictive value of self-report measures like saying on a survey that you are prejudiced, which had a correlation of 0.12. To put that in context, psychologists usually call these correlations or ‘effect sizes’ of .1 “small,” 0.3 “medium,” and 0.5 “high,” so the self-report measures had a small effect size, meaning there wasn’t much of a relationship between what people said on a survey about their biases and how they actually acted. But the IAT had something approaching a medium effect size; in Blindspot this is played up quite a bit when the authors say that “The magnitude of this superiority of prediction by the IAT was not expected by anyone.” And then the fun began.
Another set of researchers led by Dr. Frederick Oswald at Rice University responded with a counter-meta-analysis which found an overall correlation of 0.15 between a person’s IAT score and their racial behavior, which is all of a sudden down near the “low” effect size of 0.1. They got this different result for three reasons – firstly the way that Dr. Greenwald did the first analysis suppressed variance in the studies they were analyzing, and secondly Dr. Oswald included more studies – some of which were missed by Dr. Greenwald, and some of which became available after Dr. Greenwald did his analysis. The third reason was that neuroimaging studies were included in Dr. Greenwald’s analysis, which is the practice of looking at a person’s brain activity while asking them to complete specific actions and correlating that to their score on the IAT. Neuroimaging studies often produce a high effect size, and this brought the average effect size up significantly in Dr. Greenwald’s analysis. Dr. Oswald concluded that “IATs, whether they were designed to tap into implicit prejudice or implicit stereotypes, were typically poor predictors of the types of behavior, judgments, or decisions that have been studied as instances of discrimination, regardless of how subtle, spontaneous, controlled, or deliberate they were.”
Dr. Oswald’s study was published in 2013, a full four years after Dr. Greenwald’s original meta-analysis, and two years later, in 2015, Dr. Greenwald and his colleagues responded to partially dispute the discrepancy as a result of different inclusion criteria and analysis techniques, but also to add three points to say that even if the effect sizes predicting behavior from an IAT are small, that they are still relevant for three reasons.
Firstly, small effect sizes still comprise significant discrimination. Dr. Greenwald and his colleagues did some analysis to convert between a statistical criterion of discrimination used by courts in the United States to test whether a protected class (identified by race, color, religion, national origin, gender, or disability status) has been treated in a discriminatory way. The courts use something called the “four fifths rule,” which says that if you are a member of a protected class like Black people and you have some favorable outcome like getting a job at a particular company less than 80% as often as another class of people, say people who are White, then you’ve experienced an adverse impact. You can translate that number to a correlation of .11, which you may recall is below the effect size that Dr. Oswald found for the IAT’s prediction of prejudiced behavior.
The second reason is that small effect sizes predict substantial discrimination in bases affecting many people, because so many people experience biased thinking, and so many people are on the receiving end. Dr. Greenwald calculated that the difference between police officers in New York City scoring a standard deviation below the mean and a standard deviation above the mean IAT score would reduce the discrepancy in the number of times police stopped Black people compared to White people by 9,976 stops per year.
And the third reason is that small effects can affect the same people repeatedly, in employment, educational, health care, and law enforcement settings so it’s still important to try to measure and reduce this impact.
All very well, said Dr. Oswald in his speedy reply just six months later, but there is very little evidence that any of these effects actually result in anything we can actually observe in the real world. They say, “No amount of statistical modeling or simulation can reveal the real-world meaning of correlations between IAT measures and lab-based criteria that are in the range of 0.15 to 0.25.” As an example of what they say is how misleading it can be to focus on overall effect sizes, they looked at 87 effect sizes and found that the mean correlation between the race IAT and microaggressions toward Black people was only 0.7, and if you put a band around that to say, “we’re 95% confident that the real number occurs within this band,” the band included zero – meaning potentially no effect at all.
And they went on to specifically take issue with the four fifths rule analysis, saying that Dr. Greenwald provided “no evidence to demonstrate that the IAT reliably predicts which employers will differentially apply a selection process to produce disparities in subgroup selection ratios that violate the four-fifths rule, and to our knowledge, no such evidence exists.” Further, they said that the example of police stops implies that a score on the IAT actually has a causal relationship with police stops, based on what is a weak correlation between IAT scores and behavior found in a lab situation in that experiment.
Dr. Oswald’s concerns are so pointed that I’ll allow him to explain them to you himself: “Sixteen years and hundreds of studies after the IAT was introduced, we know that the IAT reliably produces distributions of scores that are said by many IAT researchers to reveal large reservoirs of implicit bias against racial and ethnic minorities, but researchers still cannot reliably identify individuals or subgroups within these distributions who will or will not act positively, neutrally, or negatively toward members of any specific in-group or out-group. None of the points raised by Dr. Greenwald in the 2015 paper change this fact. With the low predictive validity of the IAT established, Dr. Greenwald’s 2015 paper offers examples and mathematical arguments aimed at showing how small IAT-criterion correlations could have substantial societal impact. However, the global effect size central to their presentations is disconnected from the contexts to which they seek to generalize and obscures important variation in the research findings. A more productive approach to modeling the societal implications of IAT scores would be to move past abstract debates on the real-world meaning of meta-analytic estimates derived from laboratory studies to conducting large-scale, well-controlled longitudinal investigations that model IAT prediction of socially meaningful criteria in organizations, schools, hospitals, and other contexts in which implicit bias is of direct concern. Assertions about the cumulative effects of small effect sizes should be counted not as evidence but as starting points for future efforts to identify substantively and theoretically important moderator variables and boundary conditions.” They conclude: Whether the small effects of unconscious bias that are suggested as at least possible from these meta-analysis will in reality grow, be contained or disappear in complex, real-world social systems is a question that should be resolved through vigorous empirical testing, not computer simulations and thought experiments that, by their nature, must rely on strong yet untested assumptions.”
So, that was the last part of the war of the meta-analyses that I could find in the journals, but the issue about whether we can use the IAT and if so what for, remains unsolved. Some of the researchers who were involved in the anti-IAT side of the debate say that we can’t definitively link the degree of behavioral bias to any specific IAT score, and neither can we use the distribution of scores to understand the prevalence or average magnitude of behavioral bias in any given group. Further, the zero point of IAT measures doesn’t map onto the zero point in behavior, which means that, on average, the IAT overestimates bias in a population and also overestimates the magnitude of that bias. This means it’s somewhat arbitrary as a metric, and there are a number of other arbitrary metrics in psychology so this isn’t inherently a problem, but it becomes a problem when we try to make statements about the prevalence or magnitude of an attribute, or if it’s used to draw inferences about an individual’s score on the test. If we’re going to do that we need to know how scores on the test map to behaviors and right now we have absolutely no idea if this is even possible. Dr.Neil Levy at the University of Oxford believes that a person’s score on the IAT will be a poor predictor of whether they will act fairly or unfairly toward a minority group member and that “one would do just as well, and often better, at the betting table by basing one’s bet on scores from explicit measures of prejudice than on IAT scores.”
The problem here is that Drs. Greenwald and Banaji do make exactly this inference in Blindspot. They describe a version of the IAT where the ‘bad’ categories of words were replaced with pictures of weapons and found that there was a strong association between images of Black people and images of weapons. They then go on to say, “In terms of ‘guilt by association,’ a Black = weapons stereotype is particularly consequential when it plays out in the interaction between citizens and law enforcement. Although it is difficult to say conclusively how great of a role race plays in the mistaken shooting of Black men, we do know that Black men experience the effects of such mistakes more often than White men,” and they go on to describe the murder of Amadou Diallo by police in New York City.
Now I’m not denying that race plays a role in the mistaken shooting of Black men. But here Greenwald and Banaji are lumping together explicit and implicit bias and saying, “we don’t know how big of a deal this is,” and then go on to say “An automatically operating Black = weapons stereotype may have played a role in the officer’s mistaking Diallo’s wallet for a gun.” It’s a big leap to say that bias *in general* plays a role in shootings of Black men by White police officers to saying without proof that it is implicit, rather than explicit bias that is important. They go on to say that “Several experiments support the idea that Black men carrying harmless objects such as cell phones are indeed more likely to be shot at mistakenly,” but the article that the book cites says that while police officers are more likely to react quickly than a non-police officer in the same experiment, police officers were more effective at differentiating between armed and unarmed targets, were faster to make those correct decisions, and whereas non-police officers made a biased pattern of errors, officers did not. The authors of that study stated that: Police occasionally shot unarmed targets, but they were no more likely to shoot an unarmed Black target than they were to shoot an unarmed White.” The data indicated that police are aware of and responsive to the target’s race, but they showed “no bias in the decisions they ultimately make: when deciding to shoot, they set statistically equivalent criteria for Blacks and Whites.”
So maybe implicit bias did play some small role, but if the ultimate decision made is one that does not discriminate, can we really say that implicit bias is as important a factor as Greenwald and Banaji say it is? Of course, this is one study of one group of police officers and we can’t necessarily extrapolate it to a broader population, but Greenwald and Banaji are attempting to do exactly this and to do it using a conclusion that seems to be in opposition to what the researchers actually found. When I was researching this specific point I found the same error cropping up in another publication – this time an annual State of the Research report by the Kirwan Institute at Ohio State University. The report cites Hillary Clinton’s mention of implicit bias during a 2016 Presidential Debate, where she argued that everyone is susceptible to it. The Kirwan Institute’s report says that “Senator Clinton also acknowledged the often weighty implications of implicit bias by asserting: ‘It can have literally fatal consequences.’ Supporting this latter statement is a considerable body of research that examines how law enforcement officers’ implicit bases can influence decisions regarding how quickly weapons are discharged and, quite significantly, at whom.”
It turns out that there actually is a considerable body of research that examines how law enforcement officers’ implicit biases can influence decisions regarding how quickly weapons are discharged and at whom, but not all of this research, and not even all of the papers that the Kirwan Institute specifically cites, actually do this. One of the papers they cited that “there was no evidence that target race biased a police officer’s ability to correctly shoot armed targets and not shoot unarmed targets” in an experiment although of course this happened in a lab, in safe conditions, sitting in a comfortable chair, and with enough time to make a decision that the participants could enact control over their decisions – so it’s possible the researchers weren’t even measuring implicit attitudes or actions at all. This confusion about implicit and explicit biases happens often in the literature – I found one paper where professors in a teaching college asked students in teacher training to read phrases like “We are considered to be the dominant group.” We come from two parent stable homes,” and then discuss how they felt about the activity, their comfort level with the discussion, and “why do you think it is important for teachers to recognize, understand, and deal with implicit biases?” They aren’t actually considering implicit biases, though – just explicit biases that the students hadn’t previously considered.
Even more troubling, Dr. Banaji and one of her graduate students, Tessa Charlesworth, have written more recently about a decline in implicit racism among people taking the IAT. In their paper they write that “over the past decade, explicit race attitudes have moved toward neutrality by approximately 27%. Implicit race attitudes have moved in the same direction but a slower rate than explicit attitudes (changing toward neutrality by 17%).” They discuss other similar changes happening in the other IAT tests although at a slower rate. On the surface, this sounds great! But Dr. Hart Blanton at Texas A&M University poked around under the hood at the scoring system and found that test takers are scored on a scale of -2 to +2, with anything above 0.65 or below negative 0.65 indicating a ‘strong’ link. But it turns out that between 2002 and 2007, “the architects of the IAT changed the criteria for classifying people into the different preference categories (at the same time they adopted a new scoring algorithm).” If they had continued using the criteria they had been using, 60% of the people taking the IAT in 2007 would have been diagnosed as having a “strong automatic preference for Whites over Blacks.” This paper was published in 2008 and I haven’t yet seen a paper that addresses the 2019 claims about the recent shift toward neutrality but we can assume that this shift in calculation method is responsible for at least some of the reported change in levels of implicit prejudice. And we should also acknowledge here that 40% of the people who take the IAT report doing it to fulfil an educational requirement. If we can extrapolate from here to assume that these are mostly college students then as with so much of psychological research we’re essentially sampling college students and assuming that the results are applicable to all of mankind. Some researchers even analyzed the IAT data by location, which they get from respondents providing their ZIP code. They assessed levels of bias at the state and county levels, but if 40% of your respondents are living on college campuses, the situations they encounter may not be representative of the country as a whole.
One of the most persistent problems with the IAT is that it doesn’t actually tell us who is likely to engage in discrimination, or what they are likely to do. Dr. Blanton says “no study has linked specific IAT scores to observable, behavioral outcomes reflective of implicit prejudice.” Even Dr. Greenwald, the original designer of the test, has said (buried deep in an academic paper) that: “IAT measures have two properties that render it problematic to use them to classify persons as likely to engage in discrimination. These two properties are modest test-retest reliability, and small-to-moderate predictive validity effect sizes. Attempts to use such measures diagnostically for individuals therefore risk undesirably high rates of erroneous classifications.8 These problems of limited test–retest reliability and small effect sizes are maximal when the sample consists of a single person (i.e., for individual diagnostic use), but diminish substantially as sample size increases. Limited reliability and small-to-moderate effect sizes are therefore not problematic in diagnosing system-level discrimination, for which analyses often involve large samples.”
Now let’s pick that apart a little bit. The ‘modest test-retest reliability’ refers to the fact that you can take an IAT test today and get one result, and take it again tomorrow and get another result. I’m not kidding. And the rest-retest stability was particularly low for the implicit measures of race bias, more than the other types of IATs. Another meta-analysis found that less than 20% of the variability in implicit bias can be explained by an individual’s level of implicit bias a few weeks earlier. This actually introduces a puzzle – we know that young children show levels of implicit bias that are similar to adults. This evidence is typically interpreted as evidence that people learn implicit biases early and retain them for life (Dr. Dunham’s work supports this), but if biases aren’t stable across a month, how can they be stable across a lifetime?
And following on from this, as even Dr. Greenwald himself says that these small to moderate effect sizes aren’t so problematic in diagnosing system-level discrimination (except when you change the scoring criteria without telling anyone, I suppose), but the “problems of test-retest reliability and small effect sizes are maximal when the sample consists of a single person.” And, of course, individuals take the IAT and see how their score compares to everyone else’s. Individuals read Blindspot and want to know what they can do about their score. Many of the outcomes that researchers study aren’t even discrimination, and when they are measuring things that have the potential to capture the concept of discrimination, there is little evidence of individual differences in discriminatory behavior, so there is little evidence that the IAT can meaningfully predict discrimination. So part of the problem is that we need better tests to understand whether the IAT can actually measure discrimination and a team that did one meta-analysis concluded that if we try to do this, we are essentially chasing noise. The average effect of discrimination measured by these studies is zero, and the results are widely inconsistent between different studies. And even when we’re talking about population-level effects, this is still correlational data. Even if we could say that the IAT measures implicit bias, we don’t have actual data saying that implicit bias leads to discrimination. It’s possible that discriminatory actions could lead to some internal changes that result to a different score on the IAT.
Not only does the test not predict individuals’ behavior, but you can also change a person’s score on the test just by manipulating the context. So implicit racial bias scores can be shifted by interacting with an African American experimenter, listening to rap music, or looking at a photo of Denzel Washington.
This doesn’t necessarily mean that we should never use the IAT; even critics agree that the IAT can still be a useful tool “for researchers, educators, managers, and students who are interested in attitudes, prejudices, stereotypes, and discrimination.” Using the “IAT as a tool to learn about automatic associations can be a good way to start a rich discussion about attitudes, prejudices, and discrimination” can still be helpful. Dr. Banaji has been quoted as saying that “''We like to think of it as an unconsciousness-raising tool for increasing awareness or self-analysis. It should not be used to select individuals for jobs or to select a jury.'' And maybe it isn’t being used for exactly those things, but there’s a lot of grey area in between. Dr. Greenwald makes himself available to provide expert testimony on “implicit bias in suits based on race, gender, ethnicity, or age discrimination.” Law professors and sitting federal judges cite IAT research conclusions as grounds for changing laws. The National Center for State Courts and the American Bar association have launched programs to educate judges, lawyers, and court administrators on the dangers of implicit bias in the legal system.
So now I want to turn to what this means for our own thinking toward others and what we discuss with and teach our children.
Overall, what we seem to be talking about is whether our actions are under our control. Dr. Neil Levy at the University of Oxford had argued that control isn’t an either/or situation; we can have more or less control over our behavior on a continuum, but if we believe that we have to be in control to be responsible for our behavior, then we have to have sufficient control over our behavior to be responsible for it. So what happens if we act in accordance with our implicit beliefs, which are not under our control? Should we still be held accountable for our actions if we say we believe in inequality but we act in ways contrary to this? This is especially problematic when the rewards for stating one belief and the punishments for stating another belief are so high, so we can’t necessarily rely on what people tell us they believe on a survey.
The issue here is that there are at least three potential mechanisms for the connection between the IAT and behavior: it could be measuring automatic effects of associations on behavior, overlapping influences that are independent but related and both are involved in producing behavior, and cooperative causation, and/or that automatically activated associations produce conscious judgements that play some role in guiding judgements and behavior. Using stereotypes can actually increase the probability that a judgement about a person will be accurate, but they raise issues of fairness. For example, we aren’t supposed to use a person’s race to decide their guilt in an American courtroom, and insurance companies in the European Union aren’t supposed to membership in a certain group to determine premiums. So there’s something of a tension here – using stereotypes can make us more accurate but can be perceived as being less fair. Upholding fairness may seem to decrease accuracy. Dr. Banaji and her former student Jack Cao illustrated an example of how accuracy and fairness can be simultaneously achieved. Consider the statistic that doctors tend to be male, and the statistic that nurses tend to be female. Now imagine a charity that invites medical professionals to an event based solely on whether they are a doctor or nurse. If someone is a doctor, that person is likely to be invited. If someone is a nurse, that person is unlikely to be invited. Given these premises, the charity is more likely to invite a male than a female since the former is more likely to be a doctor. However, this is only the case when the charity does not know if the person in question is a doctor or nurse. Once the person’s profession becomes known, gender ceases to be of relevance and therefore should not be used. With respect to who will be invited, a female doctor should be treated the same as a male doctor—even though doctors tend to be male. Likewise, a male nurse should be treated the same as a female nurse—even though nurses tend to be female. So from this perspective it’s less about whether or not to use stereotypes, and whether or not to be fair, but to use stereotypes where it increases accuracy and fairness as soon as we have enough information to stop using stereotypes. Of course, this makes most sense when the categories are not highly charged, like doctors and nurses – notably, there’s no IAT related to doctors and nurses. Yes, there’s a certain cachet attached to the stereotype of being a doctor that might not be there for nurses, but this isn’t like a stereotype of “this person has characteristics that mean they are likely to attack me.”
From these macro issues we start to get into some of the more ‘micro’ issues. One way that researchers use to test the IAT is to teach adults or children about two made-up groups of people that differ on some characteristic and to say that the person about to take the IAT is a member of one of these groups and to ask them to pair the good and bad words with their ‘own’ group and with the other group. Results have been found on both ends of the spectrum here, with some researchers finding that children quickly learn the new groups and answer on the IAT as if the other group were inferior. But other studies find that even after telling the children over and over again that their group is ‘good’ and the other group is ‘bad,’ children failed to acquire implicit attitudes. So I’m wondering: among the group who learned quickly, is it possible that this was actually implicit learning that happened? Is it possible that the learning was actually explicit? And for the group that didn’t, how can we go from this to explaining how children pick up implicit attitudes from the world around them even when an adult isn’t telling them exactly what to think?
Another study by Dr. Banaji and her grad student Tessa Charlesworth found that children judged that a picture of a face that was manipulated to appear more trustworthy would be more likely to take a positive action like sharing than a face that seemed untrustworthy, and a face that appeared to be dominant would be more likely than a submissive-looking face to perform dominant behaviors like pick up heavy things, and a face that seemed competent would be more likely to draw the best pictures compared to an incompetent-looking face. The researchers observed that films and cartoons often use extreme examples of facial characteristics, and a face that looks trustworthy is always heroic and performs good behaviors. Previous research has found a connection between faces and behaviors in classic films but no research has yet confirmed that this where children learn about which facial features indicate which characteristics. And of course there’s lots of research on the proportion of characters in movies who have light skin compared with characters who have darker skin, and how the characters with light skin are usually in the leading roles and the characters with darker skin are much more likely to be the villain.
I did also wonder whether our own manipulations of our faces when our children are very young and we exaggerate our interest in every tiny movement they make could be related to it as well, but I’m not aware of any research on this. This overall line of thinking did make me wonder about whether there’s any point trying to shift our children’s thinking on it, since they get so much reinforcement of the ideas about who is trustworthy and who isn’t from our culture. We would essentially have to watch *every* cartoon and video with them and pick up on these things ourselves and talk with our children about them every time. A line of research that’s related to this topic is based on a study that aimed to induce an implicit bias, and then told participants the information was inaccurate (it had been caused by a computer glitch). The participants reported shifting their thinking away from the inaccurate information when they were asked about it, but when the researchers tested their implicit thinking, the old information was still there. The researchers hypothesized that implicit evaluations may form and change slowly and require the repeated experience of pairing information over time, and that simply telling a person “you got bad information” isn’t enough to change the implicit thought. It’s possible that the more vivid the newly provided information, the more easily the implicit idea will be changed, for example by doing something like pairing a control face with screams and a target face with silence, but obviously it’s very hard to say how this might transfer to knowledge updating in the real world where we are bombarded by stimulus about many topics at once, and somehow we decide which parts to retain and which to reject.
In a study that Dr. Banaji did with another of her students, Benedek Kurdi, did a review of interventions that aimed to change a person’s implicit bias. Of the 17 interventions they looked at, eight shifted implicit evaluations of Black people toward neutrality, while nine produced no changed. Of the eight that were effective, five were what the researchers described as model-free – they were simple, and only required the participants to hold two pieces of information – to do something like pairing pictures of Black people with positive stimuli or outcomes. The remaining three that were effective required only the simplest mental models, including one with a vivid story in which the protagonist was assaulted by a white American and saved by a black American. Among the nine models that were ineffective, eight required having a complex causal model of the environment, including a model of another person’s mind, a model of a positive encounter with an outgroup member, and a model of racial injustice.
There are two outstanding questions for me here: firstly, are the researchers pulling the right levers to try to change implicit bias? And even if they were able to do it successfully in the study, using the simplest approaches, would this actually translate out to the real world where situations ARE complex and must often be assessed quickly? And if we could successfully do either of those things, would the changes hold for any significant period of time? In another paper, Dr. Kurdi observes that “Attempts at creating long-term change in implicit evaluations are rare and usually unsuccessful. Failures to achieve durable change may, at least in part, be attributable to the fact that most studies to date have used long delays of days or at least hours between learning and test during which (a) forgetting may have occurred and (b) participants may have spontaneously encountered sources of interference.” Of course out here in the real world we call that interference “life,” and it does unfortunately get in the way of our performance in psychology research studies. But perhaps the interventions should take this into account when they devise the study, rather than seeing it as something that needs to be controlled and eliminated to the greatest extent possible. In another meta-analyses of hundreds of interventions to change implicit bias that was co-led by Dr. Lai, Dr. Banaji’s former grad student who is now the director of Project Implicit, as well as Dr. Brian Nosek who has worked a lot with Dr. Banaji on IAT studies over the years, concluded that the methods we’ve tried up to this point haven’t done much. Dr. Forscher, who was the other co-lead on the study, told Jesse Singal that “Based on the evidence that is currently available, I’d say that we cannot claim that implicit bias is a useful target of intervention.”
And following on from this, there are a group of researchers who see implicit bias not as something that necessarily *is* or *isn’t* in a person, but something that is a “social phenomenon that passes through the minds of individuals but exists with greater stability in the situations they inhabit.” This perspective also helps us to explain Drs. Greenwald and Banaji’s findings that the IAT is actually a pretty reliability measure for large populations, even though its reliability is low for individuals. There’s quite a bit of research on what is known as the ‘wisdom of crowds.’ If you think about showing a jar of jellybeans to a crowd and ask them to guess how many are inside, the average of the individual guesses is likely to be more accurate than even the most accurate individual. One researcher describes what’s happening like this: “Crowds are “wise” because each individual is likely to have partial true knowledge as well as erroneous biases that are largely random. When independent judgments are averaged, the random variations are aggregated away, leaving the true knowledge to emerge as the central tendency of the distribution. For any single person, that true knowledge might be fleeting and changeable. Before looking at an urn of marbles or considering the likelihood of a geopolitical event, a respondent may have never thought about the question before. And the next day, they might have forgotten, or changed their mind. But whatever partial information is available to the sample at the time of judgment is sufficient to create surprisingly accurate and stable estimates in the aggregate.” From this perspective, “thoughts and feelings that spontaneously pass through our minds are easily interpreted in a variety of ways because they have no fixed meaning. They are simply concepts that came to mind, often in response to environmental cues.”
They then argue that whether a person has biased information in their minds isn’t sufficient for it to actually emerge. The situation the person finds themselves in determines whether it actually comes out. This could explain why young children display the same biases as adults even though individuals show little stability, because they reflect the same biases that were present in the context in which they were sampled last time, but to me this raises more questions than answers because I assume that *in most cases* individuals are taking the test in the same environment as they took it last time (whether that’s at work or home or school or in a research lab) so we would expect to see stability among a majority of IAT respondents’ answers, not change. If we do accept this line of reasoning, though, then we should consider that people may actually be able to understand and report on their implicit biases but because these only appear in certain situations, they don’t come to mind when the researcher says, “what implicit biases do you have?”. They may just be thoughts and feelings that pass through our minds and are interpreted in a variety of ways depending on the situation because they themselves have no fixed meaning. If the study participant interprets their experiences differently than the way the researchers ask the question, both participant and researcher may believe that implicit biases are not available to be examined when actually they could be under different circumstances.
The suggestion that flows from these observations is that instead of screening people, or trying to change biased ideas in people, we should change situations. Ambiguous situations are ripe for implicit bias to creep into decision making, so doing things like hiding the social categories of the people being evaluated is one option. In the consulting company I used to work for we had to use “they” when describing candidates rather than “he” or “she,” and in Blindspot, the authors note that symphony orchestras have dramatically increased the proportion of women they hire by having candidates sit behind a screen when auditioning. Using a rubric when evaluating candidates prevents people from making a decision based on biased information and then backing into a rationalization of their decision. The basic thing we can do is to realize that the environments we’re in aren’t neutral, and that the potential for biased decision making is the default condition in most situations, especially in situations that are time-pressured and implicit biases are likely to influence spontaneous behavior, rather than explicit biases predicting controlled behavior. As we’ve seen in the research on talking with children about race, taking a ‘colorblind’ perspective is unlikely to actually result in meaningful change. These authors suggest that “Affirmative strategies for increasing inclusivity, diversity, and the visible presence of women and minority group members in positions of authority may be necessary to offset the constant “background radiation” of systemic bias that gives rise to widespread stereotype accessibility.”
other authors have taken these ideas a step further and said that we actually need to incorporate the mental processes and the situation. And this view has important implications for whether and how interventions to change people's level of bias are going to be effective. To the extent that follow up assessments occur in contexts that are different from the intervention, the event intervention may seem to be ineffective. And I would argue the interventions that take place in the artificial environment of a lab are thus unlikely to generalize to the outside world.
Other authors have taken these ideas a step further and said that we actually need to incorporate the mental processes AND the situation, and that this view has important implications for whether and how interventions to change people’s level of bias are going to be effective. To the extent that follow up assessments occur in contexts that are different from the intervention, the intervention may seem ineffective. And I would argue that interventions that take place in the artificial environment of a lab are thus unlikely to generalize to the outside world. So implicit bias isn’t a simple characteristic of the person, or a pure reflection of the situation. It is a reflection of the person within a given situation and it seems to me as though we’re going to need to work both ends of that to actually make a difference here. Finally, as we observed early in this episode and return to now, even if we can point to changes in implicit bias, there are only very weak links between measures of implicit bias and people’s actual behavior.
So where do we leave this? Well it’s in a bit of an uncomfortable place for me. We clearly have big problems with bias in the U.S. and in many other places, related to race and a whole host of other topics like gender and weight and age. Sticking with the issue of race, as we have done throughout this episode, a good deal of racial inequality is driven by structural factors that we have put in place – policies like redlining that restricted Black people from buying houses in certain areas, and then policies that preferentially offered predatory loans to Black people that later resulted in foreclosure. There’s also the explicit bias that helps to create structural inequality which says that it’s OK for funding for early childhood care to be cut while funding for prisons increases. Those policies are in place because a lot of people think it’s OK to treat people unequally and then blame unequal outcomes on the people who got the short end of the stick, and in my mind that’s a much bigger issue that we can actually understand and get our arms around and address rather than trying to pin down this nebulous concept of implicit bias.
I also want to acknowledge that I’m coming back to the topic of gut instinct or intuition or intuitive decision making. I was expecting to find more of a connection between the literature on implicit bias and intuition, and I was surprised not to find one. Based on what I’ve read so far, my hypothesis is that intuition is a form of pattern recognition, and that when that pattern recognition is based on judgements about other people then it becomes what is known as bias. I believe there’s a connection between patterns that we learn to recognize that have physical sensations associated with them which we then generalize over time, so when our caregivers are the same race as us and we see them on a regular basis and then make us feel good when we see them, we associate seeing people of our race and having good feelings, and we don’t have those associations with people of other races. But you’re learning alongside me on this one, and I reserve the right to update this idea in a future episode, so we’ll have more to come on this. If you know anyone who is deep in the weeds on this research then please do reach out and put me in touch, because it’s been a hard topic to find much information on beyond some conjectures in books about the wisdom that’s held in our guts.
So thanks for sticking with me on this. I know it's been a lot to get through, it was for me as well. And in some ways, I'm disappointed not to have fully been able to sort it out. But I guess in summary, what I'm trying to do is to not worry so much about my daughter's implicit bias, and really to focus mostly on explicit bias and the situations and types of experiences that I'm exposing her to, and to have that be the grounding of our anti-racist work rather than worrying too much about what might be going on that I can't see or predict or control and neither can she see or predict or control those things either. So if you'd like to read the more than 30 papers that I read to prepare for this episode, you can find them at YourParentingMojo.com/ImplicitBiasRevisited.
Thanks for joining us for this episode of Your Parenting Mojo. Don't forget to subscribe to the show where YourParentingMojo.com to receive new episode notifications, and the FREE Guide to 13 Reasons Your Child Isn't Listening To You and What To Do About Each One. And also join the Your Parenting Mojo Facebook group. For more respectful research-based ideas to help kids thrive and make parenting easier for you. I'll see you next time on Your Parenting Mojo.
Banaji, M.R., & Greenwald, A.G. (2002). Blindspot: Hidden biases of good people. New York: Delacorte.
Blanton, H., & Jaccard, J. (2008). Unconscious racism: A concept in pursuit of a measure? Annual Review of Sociology 34, 277-297.
Blanton, H., Jaccard, J., Strauts, E., Mitchell, G., & Tetlock, P.E. (2015). Toward a meaningful metric of implicit prejudice. Journal of Applied Psychology 100(5), 1468-1481.
Brown, E.L., Vesely, C.K., & Dallman, L. (2016). Unpacking biases: Developing cultural humility in early childhood and elementary teacher candidates. Teacher Educators’ Journal 9, 75-96.
Cao, J., Kleiman-Weiner, M., & Banaji, M.R. (2017). Statistically inaccurate and morally unfair judgements via base rate intrusion. Nature Human Behavior 1(1), 738-742.
Carlsson, R. & Agerstrom, J. (2016). A closer look at the discrimination outcomes on the IAT Literature. Scandanavian Journal of Psychology 57, 278-287.
Charlesworth, T.E.S., Kurdi, B., & Banaji, M.R. (2019). Children’s implicit attitude acquisition: Evaluative statements succeed, repeated pairings fail. Developmental Science 23(3), e12911.
Charlesworth, T.E.S., Hudson, S.T.J., Cogsdill, E.J., Spelke, E.S., & Banaji, M.R. (2019). Children use targets’ facial appearance to guide and predict social behavior. Developmental Psychology 55(7), 1400.
Charlesworth, T.E.S., & Banaji, M. (2019). Patterns of implicit and explicit attitudes: I. Long-term change and stability from 2007-2016. Psychological Science 30(2), 174-192.
Chugh, D. (2004). Societal and managerial implications of implicit social cognition: Why milliseconds matter. Social Justice Research 17(2), 203-222.
Cvencek, D., Meltzoff, A. N., Maddox, C. D., Nosek, B. A., Rudman, L. A., Devos, T. Dunham, Y., Baron, A. S., Steffens, M. C., Lane, K., Horcajo, J., Ashburn-Nardo, L., Quinby, A., Srivastava, S. B., Schmidt, K., Aidman, E., Tang, E., Farnham, S., Mellott, D. S., Banaji, M. R., & Greenwald, A. G. (in press). Meta-analytic use of Balanced Identity Theory to validate the Implicit Association Test. Personality and Social Psychology Bulletin.
Forscher, P.S., Lai, C.K., Axt, J.R., Ebersole, C.R., Herman, M., Devine, P.G., & Nosek, B.A. (2019). A meta-analysis of procedures to change implicit measures.
Gawronski, B., & Bodenhausen, G.V. (2017). Beyond persons and situations: An interactionist approach to understanding implicit bias. Psychological Inquiry 28(4), 268-272.
Goode, E. (1998). A computer diagnosis of prejudice. The New York Times. Retrieved from https://www.nytimes.com/1998/10/13/health/a-computer-diagnosis-of-prejudice.html
Greenwald, A.G., & Lai, C.K. (2020). Implicit social cognition. Annual Review of Psychology 71, 419-445.
Greenwald, A.G., & Lai, C.K. (2020). Implicit social cognition. Annual Review of Psychology 71, 419-445.
Greenwald, A.G., Banaji, M.R., & Nosek, B.A. (2015). Statistically small effects of the Implicit Association Test can have societally large effects. Journal of Personality and Social Psychology 108, 553-561.
Greenwald, A.G., Poehlman, T.A., Ulhmann, E.L., & Banaji, M.R. (2009). Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity. Journal of Personality and Social Psychology 97(1), 17-41.
Kang, S.K., DeCelles, K.A., Tilcsik, A., & Jun, S. (2016). Whitened resumes: Race and self-precentation in the labor market. Administrative Science Quarterly 1-34.
Kurdi, B., Gershman, S.J., & Banaji, M.R. (2019). Model-free and model-based learning processes in the updating of explicit and implicit evaluations. Proceedings of the National Academy of Sciences 116(13), 6035-6044.
Kurdi, B., Mann, T.C., Charlesworth, T.E.S., & Banaji, M.R. (2019). The relationship between implicit intergroup attitudes and beliefs. Proceedings of the National Academy of Sciences 116(13), 5862-5871.
Kurdi, B., & Banaji, M.R. (2019). Attitude change via repeated evaluative pairings versus evaluative statements: Shared and unique features. Journal of Personality and Social Psychology 116(5), 681.
Kurdi, B., Seitchik, A.E., Axt, J.R., Carroll, T.J., Karapetyan, A., Kaushik, N., Tomezsko, D., Greenwald, A.G., & Banaji, M.R. (2019). Relationship between the Implicit Association Test and intergroup behavior: A meta-analysis. American Psychologist 74(5), 569.
Leech, T.G.J., Irby-Shasanmi, A., & Mitchell, A.L. (2019). “Are you accepting new patients?” A pilot field experiment on telephone-based gatekeeping and Black patients’ access to pediatric care. Health Services Research 54, 234-242.
Mann, T.C., Kurdi, B., & Banaji, M.R. (2019). How effectively can implicit evaluations be updated? Using evaluative statements after aversive repeated evaluative pairings. Journal of Experimental Psychology: General 149(6), 1169.
Mitchell, G., & Tetlock, P.E. (2017). Popularity as a poor proxy for utility: The case of implicit prejudice. In S.O. Lillenfeld & I.D. Waldman (Eds), Psychological Science Under Scrutiny: Recent Challenges and Proposed Solutions (p.164-195). Hoboken: Wiley-Blackwell.
Oswald, F.L., Mitchell, G., Blanton, H., Jaccard, J., & Tetlock, P.E. (2013). Predicting ethnic and racial discrimination: A meta-analysis of IAT criterion studies. Journal of Personality and Social Psychology 105(2), 171-192.
Oswald, F.L., Mitchell, G., Blanton, H., Jaccard, J., & Tetlock, P.E. (2015). Using the IAT to predict ethnic and racial discrimination: Small effect sizes of unknown societal significance. Journal of Personality and Social Psychology 108(4), 562-571.
Papageorge, N.W., Gershenson, S., & Kang, K. (2016). Teacher expectations matter. IZA Discussion Papers No. 10165, Institute for the Study of Labor (IZA), Bonn.
Payne, B.K., Vuletich, H.A., & Lundberg, K.B. (2017). The bias of crowds: How implicit bias bridges personal and systemic prejudice. Psychological Inquiry 28(4), 233-248.
Porter, S., England, L., Juodis, M., ten Brinke, L., & Wilson, K. (2008). Is the face a window to the soul? Investigation of the accuracy of intuitive judgements of the trustworthiness of human faces. Canadian Journal of Behavioral Science 40(3), 171-177.
Rae, J.R., Newheiser, A-K., & Olson, K.R. (2015). Exposure to racial out-groups and implicit race bias in the United States. Social Psychological and Personality Science 6(5), 535-543.
Reyes, R.M., Thompson, W., & Bower, G. (1980). Judgemental biases resulting from differing availabilities of arguments. Journal of Personality and Social Psychology 39(1), 2-12.
Singal, J. (2017, January 11). Psychology’s favorite tool for measuring racism isn’t up to the job. The Cut. Retrieved from https://www.thecut.com/2017/01/psychologys-racism-measuring-tool-isnt-up-to-the-job.html
Staats, C., Captosto, K., Tenney, L., & Mamo, S. (2017). State of the science: Implicit bias review. Kirwan Institute for the Study of Race and Ethnicity, The Ohio State University.