Secrets, statistics and implicit bias

Employers are increasingly trying to get their staff to ask themselves a difficult question: “Am I secretly racist?” They want them to think harder about this and other common unconscious biases and take action to overcome them.

To do this, they may use tests that examine the risk of unconscious biases as part of diversity training programmes. So I sit down one afternoon, hoping I too will not be found out as being unconsciously racist.

Clicking on to the Project Implicit site, where millions of people have done one of the many tests designed by Harvard University, I choose to start with weight. I cannot bring myself to head straight to the race test, the most commonly used test on the site.

Do I hate fat people? The test aims to tell this by making me click the same button for good words and fat people and the same button for bad words and thin people, and then reversing the association. My response times would show my true feelings.

The test finds I have a “slight automatic preference for thin people over fat people”. Uh-oh.

I am comforted, however, by the fact that a large number of people seem to be more anti-fat than I am: I can see that 31 per cent of test takers strongly prefer and 28 per cent moderately prefer thin people.

I don’t think I dislike fat people and I don’t think I prefer white people to black people. Yet these tests report that the majority of people surveyed hold these implicit biases. I prepare myself for my fingers giving away an underlying racism. This second test is similar, with pictures of black and white faces mixed up with good and bad words.

Looking at the result, my data suggest I have a slight automatic preference for European Americans over African Americans. Again, more than half the respondents are more racist than me — with 24 per cent showing a strong automatic preference for white faces and 27 per cent showing a moderate preference.

But I do not want to be racist at all.

I call Joelle Emerson, founder and chief executive of Paradigm, a company based in San Francisco that works with Fortune 500 companies and tech start-ups to help them create more diverse and inclusive organisations.

Tests such as these and accompanying unconscious bias workshops have been adopted by companies including Facebook, Google and Microsoft, which even publish advice on the training for other companies to follow.

I tell Ms Emerson that I have just taken my first implicit bias tests.

“That’s never a good feeling,” she says.

Ms Emerson thinks the implicit bias tests can be useful to raise awareness of the “mental shortcuts” that everyone has developed because of the biases they have been exposed to over a lifetime.

The measure does not directly reflect an exact number that is unwavering

More important, she says companies often call her after they have done awareness training and want to take it further. She encourages people taking Paradigm’s workshops to be tactical and focus, for example, on how to mitigate their bias ahead of interviewing candidates for jobs.

Ms Emerson pushes companies to think about giving training to staff just before their performance reviews, to encourage people to consider how unconscious bias could lead them to provide different levels of feedback, such as giving more on personality to women than on technical or leadership ability.

I take another test to see how I score on a subject where I could be deemed to be in a disadvantaged category. Taking the “gender career” test means seeing if I implicitly associate words about career with men, and words about family with women. Now I am revealed to have a “moderate automatic association for Male with Career and Female with Family”.

This seems complex to me, however. As a woman with a career but no kids, I had felt I might associate career with both men and women, but somehow I still mainly associate family with women.

I contact Liz Redford, a researcher and consultant with Project Implicit, who urges me not to think of the test as diagnostic. The tests have shown predictive outcomes in groups of about 200 people, but for individuals they are useful only as an educational tool. Even then, it could be counterproductive: at least one paper has shown that telling people their scores can make them defensive or denigrate the test, if they disagree with them.

Moreover, test results can change over time. The more you take the race test, for example, the less racist you appear, which may be because awareness of your thoughts influences the outcome. “Could my mood have affected me taking the test?” I ask, noting I had a headache.

“That’s totally possible,” Ms Redford says. “The measure does not directly reflect an exact number that is unwavering. It is sort of like blood pressure . . . There’s a baseline level score at any point which it fluctuates around.”

Ms Redford warns that even “well meaning” companies need to make sure they understand the science behind the test and be more cautious in how they use it. “They could be miscommunicating things that aren’t true to thousands of people,” she says.