A doctoral or PhD dissertation is supposed to be work that pushes the research frontier further. Obviously, not all dissertations are created equal and it is to be expected that some will push more that others. But they are all supposed to push. Well, do they? This is something that is quite difficult to measure as one needs to know where the research frontier lies and what the contribution of a dissertation is. For each dissertation, only few people can do this, and it is thus impossible to have an aggregate picture, unless you use a clever trick.
Sheng Guo and Jungmin Lee take publications in top Economics journals as the research frontier and look in which JEL categories they fall. They compare this to the JEL codes for US Economics dissertations and find a strong correlation controlling for the number of jobs available in the field. If you lag the publications by two years, the regression is just as good, which hints that dissertations react to the research frontier rather than the opposite (which is unfortunately undocumented), especially when you consider that with the long publication delays in Economics, the journals are in fact a few years behind the research frontier.
My interpretation here is not quite that of the authors, who really want to understand how students choose their field of study, given that they want to be on the job market with research on a hot topic. But when they start working on it, they do not know yet what will be hot. I am not sure this is quite such a conundrum, as seminars and conferences already give quite a good picture, and working papers as well. But it still looks like dissertations follow the trends instead of creating them.