Thursday, July 25, 2013

Ph.D. Attrition

Leonard Cassuto writes in the Chronicle an article Ph.D. Attrition: How Much Is Too Much? He presupposes the answer with the subtitle "A disturbing 50 percent of doctoral students leave graduate school without finishing".

The 50% goes over all fields but the numbers in computer science are somewhat in that range. Computer Science has different issues than humanities and theoretical CS has not quite the same issues as the rest of CS. Certainly we lose several students to start-ups and high-paying jobs. But what about the ones that just have trouble in grad school.

Cassuto writes
Perhaps they lack the temperament to work on their own (which undergraduate work does not test as severely as graduate school does), or perhaps they lack, say, the mathematical chops necessary to succeed at advanced physics. But there will be a number—and if admissions committees do a good job, it will be very small—who won't be able to finish because they're not up to the demands of the task.
Having read through many graduate applications through the year there are very few, perhaps on average one or two a year, that will clearly succeed through graduate school. Almost without exception those students go to MIT or Berkeley.

For the rest of us, you have a choice. You can either take someone who will probably work their way to a Ph.D. but with uninspired research, or those you can take a risk with a student who might have strong potential. Some of those students become great scientists, some of them flame out. You get a higher attrition rate by taking risks but that's not a bad thing.

If you do take a risk in admissions you need to encourage students to "pursue other opportunities" once you realize they won't make it. That's a process that too many of us try to avoid, so we don't take those risks as much as we should.


  1. It appears to me that another issue that we do not like to address is how often talented and bright students are discouraged by sub-optimal advising. Perhaps this is not often an issue in TCS, where most of us tend to advise a small number of students, but what about larger systems groups? What about chemistry, biology, etc? How much talent goes wasted?

    Talent is a difficult resource to deal with, and fresh graduate students are human beings with different levels of maturity. Finding an appropriate advisor-student match at the personal level, more than at the subject level, still remains one of the major challenges of grad school. Sometimes, going to the highest-ranked school one is admitted to is not necessarily the best choice.

    Of course, some smart people find a way out no matter what, but some fantastic colleagues are who they have become only thanks to their amazing advisors ...

  2. Is the 50% drop out rate really a problem.
    If they drop out QUICKLY, like in the first 2 years, and get (say)
    a masters degree, thus didn't waste that much of their time or
    the schools time, thats OKAY. They might not have wasted ANY of
    their time since they learned some other stuff and got a Masters.
    They may not have wasted ANY of the schools time if they did a short
    project and/or was a TA.

    If they drop out after more than that then its a bigger problem.

    I want to say that the solution is to have a tough qualifying exam
    system or some mechanism to get students who shouldn't be there
    out after \le 2 years. But there are problems with that also.

  3. My advisor wins the heart of students by doing a very good job at teaching. It takes about 2 years (after the qual, so 4 years into the program) to dawn on the student that this dude might be inept as a researcher -- however, the dude puts such a good front that the student always doubts his own potential. After 2 years, the prelim deadline arrives and the student is forced out of the program.

    Those who survive the dude realize his game within 1 year, by which point they start ignoring him, ask their own questions and obtain their own results (to which the dude bullies his name of course).

  4. Some departments/groups have a "weed them out" mentality that can lead to really high drop out rates. Lets face it some departments have a hostile atmosphere. I know of a large math department that is internationally known for treating first year students like products on assembly line to be weeded out. The physics and CS departments at the same school have completely different cultures even though they are all located right next to each other.

    I would say that qual exam scores are not a reliable indicator of who will do good research. Maybe algorithms/complexity is different, but so much of research requires patience and persistence and the ability to work on your own, which just isn't measured in the exam system. I know a prof in the PL area who makes his students take masters level classes in algebra, topology, and logic and will swear that students that excel at his school's breath requirements in CS but flounder in the math don't do good work in any area of CS, but it doesn't really matter about the results in the CS quals. I'm sure some of the systems folks will disagree, but it's an interesting argument.

    One thing that shocked me over the years was that how many students enter a PhD program with no idea of why they really want to be there. Assuming it's a top program they all tend to have good grades and did some summer internships that bored them and decide to try for the PhD. I've met kids coming strait out of big midwestern schools who start a PhD because they like the whole "college town" thing -- being in the marching band, being in a town of 30K other 20 year olds, etc. They have no idea what grad school is about and never seem to succeed. A good percentage of the international students seem to know that this helps game the US visa program.

    We all know really good candidates who get turned off by bad advising. What happens when Jane comes to Alpha University to work on X with prof Y and she just doesn't fit with that group. If she still wants to work in X, but there aren't other professors in that area, she probably drops out. When an advisor goes off to work in a start up or leaves the university or for whatever reason just seems to be inattentive the attrition rate can be real high. If a professor has ten or more students I will guarantee they have a high attrition rate. The prof with four students probably does better with the student who needs early direction.

  5. I think had school for many is just a culture shock plain and simple. I think there's a misunderstanding that just because you can program well or multiply large numbers that you should get a PhD in cs or math respectively. I definitely don't think those skills are unnecessary but I think that they're somewhat perpendicular to the theoretical skills that are needed in grad school.

    Then to take it a step further, some students (ans dare I say most grad school students) are used to being the best of the best and the concept of asking for help or being tutored if a foreign concept to them. I compare it to being a first round draft pick in the nfl and expecting that college success to carry over. I mean, I'll say God bless you if you're the one who makes it through that rookie season (or rookie contact) without growing pains, but most players have them. And I'd think that most students have them as well, especially in that grad school atmosphere where there is sometimes a much larger gap between what's covered in the text (if there is a text) and the lecture notes. So like we hear of athletes, many early graduate students spend countless hours trying to master their craft. Question often becomes how good are they at teaching themselves and/or how good are they at asking questions and asking for help?

  6. I think that in TCS also a weak candidate can eventually get his/her PhD, since the advisor always has the choice to select an easier though less important topic. It may be some purely academic variation of something already known, some combinatorial tinkering, etc. I think every researcher can come up with a wealth of such doable, though not really important topics, some by-products or fall-out of research. On the other hand this is not really interesting to supervise and you may waste your time on behalf of the student. Much better would be a student who solves some of your research problems or advances the borders. But this is probably the rare case of the top players and definitely not the average.

  7. To the previous anonymous. I don't see why TCS is singled out in this regard. Every PhD program faces the same issues. Some times the student is not strong enough, some times they have bad luck in finding a thesis topic and some times the advisor does not do their job well and so on. It is no different than in any other aspect of life that we have a continuum of achievement in terms of what ever metric one chooses.

  8. After graduation, the pro-vice-chancellor told me (to motivate me to bring students to the university), he said:"One good thing to know is that the british government has done a study on how the Ph.D. process is organized, and our university has got number one."

    I replied:"I'm not surprised, that's why I sacked my supervisor."

    He laughed:"Ha Ha Ha..."


    Rafee Kamouna.

  9. As background to this post, please let me say that I am one of the many people who use (and like) the BibDesk/megafile method of organizing citations. It's worth filling in the "keyword" field carefully … the sweet spot is a keyword that is associated to 10-50 citations.

    In regard to PhD attrition, three recently added keywords (because they are sharply up-trending in the literature) are "naturality", "reconsolidation", and "immiseration"

    The literature associated to these three keywords, in aggregate, supports the integrative thesis that the first keyword (namely "naturality" of mathematical expression) can concretely guide the second keyword (namely "reconsolidation" of integrative STEM capabilities) in service of remediating the third keyword (namely the generation-long immiseration of the STEM enterprise that particularly afflicts young people).

    Summary At present "naturality" is used mainly by mathematicians; the word "reconsolidation" is used mainly by cognitive scientists; and "immiseration" is used mainly by economists and historians (Marxists especially!). Yet in the 21st century, for solidly practical reasons (as it seems to me), the STEM professions should embrace all three of these notions, and make them their own.

    1. Is this an automatically generated text to increase links?

    2. The post recognizes burgeoning STEM trends:

      • 9,907 MathOverflow queries that contain the (mathematical) word natural, and

      • 584 PubMed articles (almost all of them recent) that contain the (neuro-cognitive) word reconsolidation, and

      • The once-rare economic word immiseration, which is broadly the subject of several of Lance's and Bill's recent posts, and is becoming (regrettably) the generic lot of STEM graduate students.

      To what degree can any one of these three topics be analyzed in isolation from the other two?

      That is a question regarding which it is best that everyone not think alike. An outstanding resource for reflection is Bill Thurston's On Proof and Progress in Mathematics, which vividly illuminates all three topics.

  10. Do you know what they call a student below the ability level and/or the temperament to do solid work on their own for a Ph.D. who gets through his program on the sympathy of faculty after many years of kicking around the department?


    We should not see a Ph.D. as an entitlement. It is bad enough that a B.S. is becoming easier to push through because of mandatory pass rates.

  11. It is probably really helpful to be blunt and express concern for the student who might not make it, because just encouraging the student to pursue other opportunities might not be enough to push him to realize that he might not make it.

  12. "Academic attrition" memes are becoming culturally prevalent — sometimes as satirical expressions that are merely discomfiting to high-level academic establishments — sometimes as expressions of mathematical and cognitive reconsolidation that enjoy massive popularity and are so eerily beautiful as to be scarcely comprehensible