The Arithmetic of Bias

The NYT recently ran a great opinion piece, by Jessica Nordell and Yarnaa Serkez, about the long-term impact of bias on women in the workplace. The magic was in the math.

It wasn’t that the piece called out egregious examples of discrimination, or identified companies or people that were really bad actors (although there was some of that); it was that it called attention, via a simple simulation, about how even small levels of bias – whether conscious or unconscious – can accumulate over time and lead to very large effects. In other words, it wasn’t trying to pin issues on particular bad actors or motives, but flagging systemic issues we might be otherwise blind to.

And that’s something we should think about too, as we write about complex systems – to resist the temptation to just look for for bad guys but instead to help readers really understand how the world works, even if terrible outcomes are the result of small flaws or human frailty.

The piece features a simulation of a company, NormCorp, where employees are promoted based on their performance reviews. You know, more or less like a regular company.

NormCorp is a simple company. Employees do projects, either alone or in pairs. These succeed or fail, which affects a score we call “promotability.” Twice a year, employees go through performance reviews, and the top scorers at each level are promoted to the next level.

So if all things are fair, men and women progress at the same rate through the company. But what if there’s some in-built bias in the system that regularly rates women slightly lower then men? It doesn’t have to be intentional, or the result of bad motives, or even conscious. It just has to exist – and it doesn’t even have to be against women. It simply has to be systemic. Maybe managers have a (unconscious) preference for people that look or sound a certain way, or who have gone to a certain university, or come to work early, or socialize after hours. Whatever.

When we dig into the trajectory of individual people in our simulation, stories begin to emerge. With just 3 percent bias, one employee — let’s call her Jenelle — starts in an entry-level position, and makes it to the executive level, but it takes her 17 performance review cycles (eight and a half years) to get there, and she needs 208 successful projects to make it. “William” starts at the same level but he gets to executive level much faster — after only eight performance reviews and half Jenelle’s successes at the time she becomes an executive.

Our model shows how large organizational disparities can emerge from many small, even unintentional biases happening frequently over a long period of time. Laws are often designed to address large events that happen infrequently and can be easily attributed to a single actor—for example, overt sexual harassment by a manager — or “pattern and practice” problems, such as discriminatory policies. But women’s progress is hindered even without one egregious incident, or an official policy that is discriminatory.

That’s a really important point worth repeating: “…women’s progress is hindered even without one egregious incident, or an official policy that is discriminatory.” And of course we do need to call out egregious incidents or discriminatory policies when we find them; but it’s just as important to help readers understand unconscious bias as well.

It can be tough, I know; we’re used to writing character-driven narratives that engage readers rather than wonky analyses that feature math and simulations. The journalistic form closest to doing this well are interactive graphics, and certainly we can do a lot more in that promising space.

And/or write more stories like this opinion piece.