because it DOES NOT control for hours worked nor experience, and lumps up narrow specialties with wide specialties together in a single "finance".
There is a huge difference in finance as a "bank teller" and finance as a "investment banker at Wall St".
This is a problem of large scale population level wage research, it misses very important confounding variables and lumps up everything they failed to explain as some magical gender pay gap.
This is the epitome of how low replicability social sciences research is done: download dataset from JSTOR, load it in Stata/Matlab, run some regressions and call it a day.