I’ve spent 20 years studying bias at work. Here’s how to design an effective diversity training

The key is to arm bystanders to interrupt bias, so that the people experiencing bias don’t have to carry that load alone.

I’ve spent 20 years studying bias at work. Here’s how to design an effective diversity training
[Photo: Svetlana Aganina/iStock]

Bias trainings have a bad rap. And for good reason: Widely cited research by Frank Dobbin, Alexandra Kalev, and Erin Kelly found that diversity evaluations and diversity trainings for managers are often ineffective. It was a rigorous study, but it’s important to recognize what it found and what it didn’t. The researchers studied Equal Employment Opportunity filings and analyzed whether seven different sorts of diversity programs yielded progress on diversity.


Particularly during the period studied, many diversity trainings were sensitivity trainings with little basis in research. Ineffective trainings don’t work. Crap in; crap out. No argument there.

But bias trainings remain important for a simple reason: most people don’t know what bias looks like. That’s why I have spent 20 years studying how bias plays out in everyday workplace interactions. For example, the first basic type of bias I call “prove-it-again”: some groups have to prove themselves more than others.

One way prove-it-again bias commonly plays out is via stolen ideas. Our national study of lawyers found that white women report that other people get credit for ideas they originally offered at a rate of 20 percentage points higher than white men. Most prove-it-again patterns are triggered by race as well as gender: people of color report stolen ideas at a rate 18 percentage points higher than white men.

In far too many bias trainings, people sit around and have earnest conversations brainstorming what kinds of biases they hold. But here’s the problem: they don’t know.

That’s why I’ve developed a workshop that describes what bias looks like on the ground and involves a whole department, or company, sitting in groups of six brainstorming ways to interrupt bias when they see it. I ask people to consider how they themselves would do it, given their personality and their role in the company.

Typically, these don’t entail “calling out” as much as “calling in.” The key is to arm bystanders to interrupt bias, so that the people experiencing bias don’t have to carry that load alone.


Often when we are discussing the stolen idea, someone in the workshop will suggest that in this circumstance they’d say something like, “Hey, that’s gender bias. Don’t steal her idea!” I typically respond, “If you feel comfortable with that, that’s great. Does anyone have a different approach?” Inevitably, someone comes up with something like this: “I’ve been thinking about that idea ever since Latisha first said it, Rick. You’ve added something important; maybe here’s the step.” I was once leading a workshop for about 200 people at a large tech company, and the gentleman who offered this sort of answer was greeted with a round of spontaneous applause. Now that’s a good learning environment to persuade bystanders to take action.

What works in tech and many other fields is to present the research in a dispassionate way and allow people to brainstorm their own strategies for interrupting bias. This puts team members in a role they feel supremely comfortable in: Here’s a technical problem. Can you solve it? Their response: I can totally nail this.

When I gave the Individual Bias Interrupters workshop to STEM department chairs at one heavily science faculty, 100% of those who handed in evaluations reported that they were leaving with specific strategies for interrupting bias, and 83% reported that they were very likely to use what they learned going forward. The same department chairs had eaten alive a diversity consultant who gave a sensitivity-type training a year earlier.

One woman present during the workshop in the large tech company said: “This set new norms, so now we share language about what bias looks like and how we will interrupt it when we see it.”

Research shows that effective bias training is interactive and four hours long. Interactive, for sure. Four hours long? I believe the finding that this is the ideal length, but at many organizations, that will be a stretch. Individual Bias Interrupters runs for two hours and leaves people wanting more—always a good move.

A key advantage of interactivity is to defuse the dynamic where a trainer gets into fruitless arguments with workshop participants about whether or not bias exists in their company. What I find is that, when faced with a very concrete description of how bias commonly plays out, most people recognize that it’s happening.


If they don’t, colleagues are likely to have someone at their table of six who will clue them in. And in the rare case where someone becomes insistent that a specific pattern of bias does not occur, my reaction is: “That’s awesome. What if you were to meet this pattern outside the company, say from a supplier? Discuss what you would do then.” This keeps things from getting derailed by a single bias-denier.

But I find that bias denial is really, really, rare. Most people are relieved to find some simple scripts to help them navigate the workplace in a way that’s fair to everyone.

It helps a lot that the workshop is not just about gender. Many of the same problems that women face are also problems for people of color. Our national study of engineers found that about one third of white men reported that they have to prove themselves more than their colleagues of similar education and experience, but two thirds of both white women and people of color do.

Another example: People of Asian decent and women encounter a “modesty mandate,” or the cultural expectation they will be modest. In both groups it’s often internalized—since women and Asian people are often raised this way—and both groups also often encounter backlash if they are perceived to be immodest. That means that the same bias interrupters that help women tackle the modesty mandate can help people of Asian decent, too.

First-generation professionals will benefit, too, because people from blue-collar families are typically taught that “boasting and self-promotion and credit hogging are wrong and unseemly,” to quote one “class migrant”—a professional whose father was a bricklayer. One study found that white men from elite backgrounds were 12 times more likely than white men from non-elite backgrounds to get a callback for a professional job.

The modesty mandate means that women, people of Asian decent, and first-generation professionals will often be unduly modest in the self-evaluations that are a standard of many performance appraisal processes. That’s why I’ve written a simple two pager to level the playing field, which is open-sourced online.


The fact that “class migrants” often encounter prove-it-again bias is particularly important in tech, because computer science and engineering tend to attract first-generation professionals. I have been in workshops where the first time I talk about class-based bias, some white men who have been staring sullenly out the window reorient their chairs, reassess me, and listen up. That’s just one reason it’s important not to demonize white men in this process.

I am also careful to talk about the ways that prescriptive bias disadvantages men as well as women. Prescriptive stereotypes depict the “good woman” as modest, self-effacing, and nice, which leads to bias against women who are competitive and ambitious—aka the Hillary problem. But prescriptive bias also leads to bias against modest men because they are violating the prescriptive stereotypes that men should be competitive, ambitious, assertive, and direct. Discussing specific readily recognizable ways bias impacts many different groups can help engage more people in disrupting it.

It’s a pity bias trainings have gone out of fashion, because people can’t begin to interrupt bias until they know what it looks like on the ground. But it’s important not to oversell any bias workshop. You can’t change a company’s culture by doing anything once.

To keep bias at bay, you need to build organizational bias interrupters into your basic business systems. We have built open-sourced toolkits for interrupting bias in hiring, assignments, performance evaluations, and meetings, with metrics for establishing a baseline and measuring progress, and simple evidence-based tweaks to interrupt bias before it happens, or correct it soon afterward.

It’s time to replace both old-fashioned sensitivity trainings and the last generation of bias trainings based on the Implicit Association Test. The IAT-based trainings are certainly better than the earlier sensitivity trainings, but they had two important drawbacks: they focused on cognitive processing instead of giving people a clear picture of how bias plays out on the ground, and they typically provided no guidance whatsoever about how to interrupt bias. People need both.

Joan C. Williams is Hastings Foundation chair and director of the Center for Work Life Law at the University of California, Hastings College of the Law.