As a follow-up to my previous post on the institutional distribution of NIH training funds, I have analyzed 1 year's worth (FY2013) of data from the NSF Graduate Research Fellowship program available through NSF FastLane. This data set lists those who were offered awards (but did not necessarily accept them if they got a "better offer" although this is likely rare) as well of their baccalaureate institution, their field of study, and their current institution.
The distributions of current institutions and baccalaureate institutions are shown below:
As might has been anticipated, the distribution of current institutions is relatively narrow with 25 institutions accounting for 50% of the 1842 fellows but with a total of 271 institutions represented by at least one fellow. The distribution of baccalaureate institutions is somewhat broader with 58 institutions accounting for 50% of the fellows and 462 institutions contributing at least one fellow.
Of the 1842 fellows, 511 listed their field's of study as Life Sciences in some form. The distribution of the current institutions for the Life Science fellows correlates reasonably well with the overall pool with an correlation coefficient of 0.82. However, some life science-focused institutions such as UCSF represented at a higher level than in the fellow pool overall.
Further inspection revealed that the distribution of Life Science NSF fellows correlates reasonably well with the distribution of NIH F32 post-doctoral fellowship as shown below:
The correlation coefficient here is 0.80.
These distributions have implications with regard to what would happen if more graduate students were supported by individual fellowships rather than research grants or institutional training grants. In addition, these distributions suggest those institutions that are most attractive to trainees who are likely to have the most options, depending on how one interprets the relationships between award probability and other factors.
(Updated with revised first figure)