Wrong conclusion: E[Earnings|Language=OCaml] > E[Earnings|Language=Javascript]
More accurate conclusion: E[Earnings|Language=OCaml,School=Top10,Major=DualCS/Math] >> E[Earnings|Language=Javascript],
E[Earnings|Language=Javascript] >> E[Earnings|Language=OCaml,School=Who?,Major=Not DualCS/Math]
If you didn't explicitly control for it, your results probably only reflect selection bias.
For instance, let's a dev with a really unremarkable resume gets paid significantly less than his coworkers in a really hot subfield because all his coworkers have a top-10 stem degree, research with top professors, and work experience at really hot startups. It's likely if he is a decent negotiator the underpaid dev's compensation will eventually revert to the mean. That will be the compensation of an extremely marketable dev if he can sell himself as a dev with really valuable experience.