Your mentorship program is solving the wrong five years

Here's an uncomfortable thought experiment. Every leadership program, sponsorship initiative, and "lean in" seminar aimed at women in STEM assumes the gap opens on the job — in performance reviews, in who gets stretch assignments, in who gets passed over for VP. What if that's mostly not true? What if the gap is already locked in before anyone's badge photo gets taken?

One economist went looking for the answer, and the data says: yes, pretty much.

The headline number. The National Science Board counted 36 million people in the U.S. STEM workforce in 2023. 65% men, 35% women — a gap most of us have seen a dozen versions of but a snapshot only tells you the gap exists. It can't tell you when it opened, and that's the question that actually points to a fix.

So someone tracked it. Jamin Speer, an economist at the University of Memphis, followed the same people from high school into mid-career — pulling from the National Longitudinal Survey of Youth and roughly 1.2 million STEM graduates in census data — and broke the total gender gap into six stages to see which ones actually did the damage.

Two stages ran away with it:

Everything else is a rounding error by comparison. Staying in a STEM major once you've started contributes 16%. Gaps in high school prep contribute just 8%. And women's higher college attendance rates actually work against the gap, narrowing it by 14%.

Read that again: both decisive stages are over before most careers even start. And it's not the students you'd expect walking away. As Speer puts it, "little of the gender gap in STEM can be explained by differences in ability or level of preparation. The gap in major choice is driven by the most STEM-ready students." The women choosing something else are, disproportionately, the ones most obviously capable of doing the work.

Now the part that should sting a little. If the gap is set by 22, what happens to it once women are actually in the workforce? Nothing much. Speer compared early-career and mid-career STEM graduates and found the gap doesn't widen — if anything, it narrows slightly. His own words: the data show "no big changes in the gender gap in STEM between early- and mid-career."

Sit with that for a second. Sponsorship programs. Negotiation coaching. Executive presence training. An entire advice industry, pointed almost exclusively at the one stretch of the pipeline where the measured gap barely moves.

What's actually inside that 35%. The topline number hides three very different labor markets. Women are 65% of workers in science-and-engineering-related occupations — mostly health care. They're 27% of science and engineering proper — the STEM most people actually picture. And they're just 11% of middle-skill technical jobs that don't require a bachelor's degree. One number, three worlds.

The one late-career figure that did move — the wrong way. In May, 50/50 Women on Boards released its Q1 read on the Russell 3000: women now hold 29.9% of board seats, down from a peak of 30.4% a year ago. This isn't a survey — it's a census of 2,843 companies and 26,077 seats.

The how matters more than the headline: women gained 138 seats and lost 83, a net of +55. Men gained 404 and lost 218, a net of +186. And of the seats women did pick up, 86.2% were newly created — not seats that used to belong to a man. Boards kept growing (new appointments jumped from 411 to 647), and men took most of the new real estate.

The number that simply doesn't exist. Zoom into pharma and medicine manufacturing — an industry that's roughly half women — and the leadership data disappears entirely. The only industry-specific count ever published came from BIO, fielded once between November 2021 and January 2022: 34% women on executive teams across 39 companies, 20% women CEOs across 84. BIO didn't even publish a board number, citing too small a sample. No federal agency breaks this out by industry. Four and a half years later, nobody has run it again.

Before you take any of this to a slide deck, four caveats:

Speer himself flags the mid-career finding as the shakiest part of his own paper — "of all the stages, this is the one least suited to my data" — since the ACS isn't a panel survey, so it's comparing different people at different ages, which tangles up experience with cohort effects. Treat the "flat mid-career" result as the least secure claim in this whole piece.

It's also accounting, not causation. If an 18-year-old correctly reads a hostile ladder ahead and picks a different major, that shows up in the data as a "major choice" problem — when it's really a ladder problem wearing a major-choice costume. No study we found can cleanly separate the two.

And the Science Board's own numbers complicate Speer's persistence stage: among students who started in STEM and finished some degree, 57% of women ended up with a STEM degree, versus 53% of men — the opposite direction from what Speer's decomposition would predict. The likely explanation is a definitional mismatch between the two datasets, but it's an open thread, not a resolved one.