I wonder how random the graduate admission process is in the United States. I am mostly interested in the fields of computer science, and PhD programs (i.e., not Master programs).

One way to quantify it would be following a similar methodology as in this study (mirror), which showed that half the papers appearing at NIPS would be rejected if the review process were rerun. Of course it may not be flawless, but might give some decent approximations.

I am looking for referenced numbers, not guesses.

  • 3
    As candidates tend to to apply at multiple schools of the same level (5-10 I'd say at least), and the same people tend to get in or not in, it's probably not very random: a very specific student profile is preferred somewhat consistently across professors at different schools. This seems very similar to the quoted experiment; but I doubt you'll find actual numbers from some such experiment. Mar 21, 2016 at 16:57
  • @gnometorule Good point, yes the experiment could be on student side (probably easier to get the data there). Mar 21, 2016 at 16:59
  • 2
    I guess this is very similar to many conferences where the number of papers that can be accepted is much smaller than the number of submitted papers: the outstanding papers usually make it, the crappy papers are weeded out pretty reliably, and for all the good-but-not-outstanding papers in between the choice is influenced by a multitude of factors (and becomes somewhat random). Mar 21, 2016 at 21:18
  • 2
    In general, the sheer number of applications reviewed by admissions committee members in some departments will inevitably lead to decision fatigue.
    – ff524
    Mar 22, 2016 at 2:48
  • @ff524 I recall having read some similar studies, but I don't recall such a high variation. Impressive. Mar 22, 2016 at 2:58

1 Answer 1


There is without a doubt some randomness in the process, but I find your question impossible to answer. In the question, you implicitly assume that the "quality" of an application is a well-defined concept that could be used to replicate an experiment.

But this isn't the case. The only way you could do this is by submitting N "equally good" applications to the same graduate programs in the same year, and expect the same outcome for all. But this is clearly not a workable concept. Rather, different graduate programs may look for different candidates (for example because they have different areas of strength), and possibly may want to balance areas in successive years. In other words, the "quality" of an application is a concept that can only be measured against the current needs of individual graduate programs. You can't repeat the experiment at a different place or different time and expect the same outcome.

  • 3
    Actually, people do that type of experiment when researching discrimination in hiring and other markets. In addition, this is not an answer. Mar 21, 2016 at 22:26
  • 1
    It's true that one can make these sorts of experiments in a controlled setting using test subjects. What I wanted to point out is that to do it in the real world, using actual graduate school admissions committees is going to be exceptionally complicated and unreliable. One can say that this doesn't answer the original question, but it may shed light on why there is not likely going to be a lot of factual data. Mar 22, 2016 at 2:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .