[doctorpangloss]

Back at Workpop, an hourly jobs website, we actually observed a negative correlation between educational attainment and getting interviews. This was from data that did not use a recommender system at all, so it lacked the serious confounding “feedback loop” that may occur in TripleByte’s case.

After comparing to a much smaller set of broad-market resumes, it was clear that you’re right: people using online job finding tools are actually shut out, for some reason, from conventional job channels where the intuitive rules apply.

But yes, obviously there’s something wrong with the data if the salary gap observed was so small. I generally like TripleByte and its writing, but it seems like they really got something 200% wrong here.