The salary database Comparably has released a new study exploring the pay gap between men and women in the tech industry. Among its most interesting findings is that the gap is largest for women early in their careers, with women under 25 earning on average 29% less than men their age, while the gap drops to only 5% for workers over 50.
The study adds to similar recent results published by Glassdoor, who found last November that the average female programmer made nearly 30% less than her male counterpart.
Get Data Sheet, Fortune’s technology newsletter.
Comparably doesn’t attempt to explain why the pay gap drops as workers age, but it can be interpreted in at least two ways. First, it may reflect women progressively overcoming bias or some other disadvantage over the course of their careers. Alternately, it may be a sign of broader changes in the industry or society over the course of recent decades.
Writing at TechCrunch, job search executive Tim Cannon argues for the first view, saying that women “start behind and spend their careers just trying to catch up” on the salary scale, mainly, in his view, because of implicit bias in the way jobs are listed and salaries are negotiated.
But a look back at history shows that women currently starting their careers in tech face more obstacles than did women who are now in their fifties. As the filmmaker Robin Hauser Reynolds highlighted in the 2013 documentary CODE: Debugging the Gender Gap, women made up a much greater proportion of computer science graduates thirty years ago—37%, compared to just 14% in 2013.
NPR’s Planet Money argued in a 2014 report that that decline, starting in the mid-1980s, coincided with the aggressive marketing of home computers to young boys. That both helped computers become more culturally associated with men, and gave young women relatively less access to the tools to learn computer science from a young age.
The decline meant that women starting their careers today have fewer female role models and colleagues than did women now in their fifties. The decline of women in tech also makes it much easier for negative stereotypes about women to permeate the male-dominated industry. Those factors may contribute both to women’s lower wages, and to the fact that women are more likely than men to leave the field even after they get a job.
Of course, this doesn’t mean that changing the language in job descriptions or outlawing questions about applicants’ salary history, as Cannon suggests, wouldn’t help the situation. But it will take a lot more than that to reverse three decades worth of steady erosion of women’s status in the tech workplace.