Happy Equal Pay Day! Using Analytics to Address the Gender Pay Gap

Did you know that today (April 4, 2017) is Equal Pay Day?  I didn’t either until I woke up to this stream of tweets:

There’s been a lot of talk in the news and politics about equal pay laws: the idea being that women should get paid an equal amount as their male colleagues for the same work. There have been a host of studies (here’s one by Pew Research Center) that show a stubborn persistence in the gender pay gap over the last fifty years. For example, last year, a black women would only make $0.65 for every dollar that a white man would make performing that same task! The response to this in a number of states has been legislation mandating equal pay for equal work.

What does all this have to do with the INFORMS Analytics Conference, you ask? Well, it just so happens that yesterday Margrét Bjarnadóttir from the University of Maryland, College Park and PayAnalytics gave a talk entitled Closing the Gender Pay Gap With Analytics. In this talk, she described an analytical model designed to help companies close the gender pay gap between their male and female employees, while attempting to minimize costs to the company and maintain some sense of fairness.

It seems like the solution to this problem is pretty simple: give raises to women until your pay gap is gone! However, what Dr. Bjarnadóttir highlighted in her talk (and accompanying paper, found here is that there is a complicated feedback loop that gets formed when trying to associate pay and work output. In particular, paying a particular individual X more will have a simultaneous effect on the pay gap as well as on the traits that are used to determine what “equal work” means. This means, to quote her paper, that “men who strongly typify male employees in terms of [traits related to high pay] also tend to have large differential influence because paying them more increases the explanatory power of these traits.” Thus, in such models, paying men more is the cheapest way to close the gender pay gap.

This was a pretty eye-opening revelation to me. It turns out that there are all kinds of unintended consequences in so-called optimal methods of closing the pay gap. Here’s another: “A firm can reduce the explanatory power of gender and thus the gender pay gap by cunningly finding other ways to explain women’s low relative pay. One particularly relevant strategy is to create a job classification scheme to ‘ghettoize’ women apart from men in the firm’s organizational structure; the resulting job classification scheme – as encoded in a set of dummy variables in the log-wage regression – would absorb much of the explanatory power of gender, reducing the gender pay gap without raising the compensation of women at the firm.”

Dr. Bjarnadóttir continues in her paper to do a robust statistical analysis of these trends, perform simulations, and then offer advice for companies seeking to close their pay gap that is more in line with the spirit of “equal pay for equal work”. So, on this #EqualPayDay, it’s worth taking some time to think about not just the end goal, but the methods that we take to get there–and how analytics can be used both for good and for ill in the process.