There are two recent studies of gender disparities in science and technology (referred to by the faintly awful acronym "STEM") getting a lot of play over the last few days. As is often the case with social-science results, the data they have aren't quite the data you would really like to have, and I think it's worth poking at them a little, not to deny the validity of the results, but to point out the inherent limitations of the process.
The first is a study of lifetime earnings in various fields that includes this graph showing that women with a Ph.D. earn about the same amount as men with a BA: That's pretty damning. But also a little deceptive, because this is a plot of "lifetime earnings," which means that they are necessarily doing a social-science analogue of astronomy to make this graph.
What I mean by that is that this is necessarily a retrospective study, looking back at how things were, in the same way that astronomers are looking at distant galaxies as they were back when the light was emitted, millions of years ago. The authors of the current study are following the procedure used by the Census Bureau for similar work, which they describe thusly:
What that means is that they're taking a cross-section back through time. In order to get the average salary for people at age 40, they are looking at the average salary of people who are 40 now (well, who were 40 in 2007-2009, but you get the idea). That means that you're looking at the salary of a 40-year-old Ph.D. who was hired somewhere in the 1995-2000 range, which is not necessarily the salary someone hired today should expect to be making when they are 40.
This is important for these kind of gap studies, because the data presented necessarily average over many years of changing conditions. To get the last few years of their hypothetical 40-year working life, they're looking at the salaries of Ph.D.'s who were hired in 1970(-ish). And I hope everybody will agree that the working environment was significantly different in 1970 than it is today, particularly when it comes to differential treatment of gender and race.
This makes interpreting this graph a little trickier than is being presented by most of the links to it I've seen. That is, even in the hypothetical fantasy-world case where gender disparities in salary and promotion have magically been fixed today, you would still expect to see a significant gap in this sort of lifetime earnings calculation due to the gender disparities that existed in the past.
If you want to really figure out what's going on with the earnings of different groups from this kind of study, you would need two such studies, preferably separated by a substantial interval in time. And, in fact, we have just that, in the 2002 Census Bureau report referenced in the current study (PDF), which includes this graph.