With diversity in tech-hiring their goal, the Silicon Valley women who started Project Include will be tracking employment data. Their initial goal is to form a list of metrics, to gather the facts from 18 firms, to anonymize all and then share the results. As a baseline, those results can become a springboard for inspiring diversity.
I assume they are aware of Simpson’s Paradox.
Misleading Data
Our story begins with a study that cited the discrimination experienced by Dutch women who had applied for science research funding in 2010-2012. You can see that a higher proportion of the men who applied were successful:
The problem though was that a breakdown of the summary numbers contradicted the conclusion. If you looked category by category, there was no bias against women at all.
Simpson’s Paradox
To illustrate Simpson’s paradox, I found more detailed data for a similar situation at the University of California in a 1973 gender discrimination case:
Because 44% of all male applicants and 35% of all female applicants were accepted to their graduate schools in 1973, gender discrimination seemed rather obvious…until you took a closer look.
As you can see, more women applied to more competitive departments. Meanwhile more men competed for spots in departments with high acceptance rates. Combined, these stats conveyed the impression that women experienced discrimination. However, if we consider the relationship between the proportion of female applications and admissions rates the data point us in a different direction.
Similarly, in the more recent Dutch case, analysis of the data by a Dutch psychologist indicated no discrimination. But, as he adds, although the data indicate no bias, that does not mean it did not exist.
Our Bottom Line: Human Capital
Yes, finding Simpson’s paradox does not mean there is no gender discrimination. It does not mean that we are optimally accessing all human capital.
It just indicates that whether we are analyzing baseball batting averages or Silicon Valley hiring bias, we had better look closely at our numbers so that we can accurately target the issues.
