> The federal government has been adding indirect cost payments to research grants since 1947. Today, each university negotiates its own overhead rate—including one rate for facilities and one for administration—with the government every few years. Rates vary widely because of geography—costs are higher in urban areas—and because research expenses differ. Biomedical science, for example, often requires animal facilities, ethics review boards, and pricey equipment that aren't needed for social science. The base rate for NIH grants averages about 52%—meaning the agency pays a school $52,000 to cover overhead costs on a $100,000 research grant (making overhead costs about one-third of the grant total). Universities usually don't receive the entire 52%, however, in part because some awards for training and conferences carry a lower rate, and because certain expenses such as graduate student tuition don't qualify.
The author is also engaging in manipulative data presentation techniques (such as presenting the cost of a graduate student for five years, making it sound like an outrageously large $233,000, and not $46k/year, which when you factor in the number of hours a grad student typically works, is likely around federal minimum wage.)
If anyone would actually like to read up about the subject, here are some starting points:
https://www.nsf.gov/bfa/dias/caar/indirect.jsp
https://grants.nih.gov/grants/policy/nihgps/html5/section_7/...
If the NIH gives you a "$5M" grant, you get the $5M plus the F&A costs. I don't know how NSF does it.
There are a lot of rules around what grant money can and cannot be spent on, and the rules are specifically set up to try and prevent administrators from taking more than what is reasonable for keeping the lights on, staff paid, etc.
To everyone getting outraged in the comments about evil universities and schools 'stealing' research dollars: you are being manipulated by a computational researcher who likely is not even remotely aware of the costs of the infrastructure to support them, such as computational/storage clusters they utilize. Those clusters, if supported from F&A/indirect costs, often cannot have usage charges, because that would be "double dipping."