The number of women reporting is remarkably small in some cases. Only ten women reported for the software engineer at facebook. Any woman that got hired for significantly more or less than the average would have a large effect on the data. Also looking at facebook you have a difference in experience of 0.7 average experience for women and 3.2 average experience for men. So you're not holding all other variables constant as you change the gender. If the average woman's salary would increase by 2.5% at the end of her first and second year then she would earn slightly more than the men with 3.2 years experience.
So what can we learn from the data? Basically nothing. Without large data sets it is impossible to come to reasonable conclusions about whats going on. To their credit Glass Door did not attempt to come to any real conclusions with the data. I guess they realized the data didn't allow for analysis and they decided to release it any way. Good way to get some attention I suppose... and I played right in their hands. Well played Glass door. Well played.