This is an interesting study. I think there are a few points/gaps that need to pointed and clarified.

(I did a similar study, and found no gender wage gap: https://medium.com/@parasu/on-the-gender-wage-gap-at-the-university-of-michigan-a-narrative-through-numbers-d36128311668#.jhd5lj9tp)

1. What’s the actual accuracy of the gender prediction? Even randomly selecting male/female would give me a 50% confidence…
2. Why not use name as predictor of gender? The U.S census database has number of males and females for a given name. You could easily set your own confidence level and predict the gender that way. Much easier and more accurate
3. I’d be interested in seeing some significance testing on the difference in median salaries. The gap tends to disappear if you consider statistical significance
4. Very peculiar that the median salaries have a wave-like form, which is somewhat unrealistic. I think the experience needs to be binned properly to overcome this