Affirmative action for statistics Ph.D.’s?

Someone pointed me to this note by John Langford:

Graduating students in Statistics appear to be at a substantial handicap compared to graduating students in Machine Learning, despite being in substantially overlapping subjects.

The problem seems to be cultural. Statistics comes from a mathematics background which emphasizes large publications slowly published under review at journals. Machine Learning comes from a Computer Science background which emphasizes quick publishing at reviewed conferences. This has a number of implications:

1. Graduating statistics PhDs often have 0-2 publications while graduating machine learning PhDs might have 5-15.

2. Graduating ML students have had a chance for others to build on their work. Stats students have had no such chance.

3. Graduating ML students have attended a number of conferences and presented their work, giving them a chance to meet people. Stats students have had fewer chances of this sort.

In short, Stats students have had relatively few chances to distinguish themselves and are heavily reliant on their advisors for jobs afterwards. This is a poor situation, because advisors have a strong incentive to place students well, implying that recommendation letters must always be considered with a grain of salt.

This problem is more or less prevalent depending on which Stats department students go to. In some places the difference is substantial, and in other places not.

One practical implication of this, is that when considering graduating stats PhDs for hire, some amount of affirmative action is in order. At a minimum, this implies spending extra time getting to know the candidate and what the candidate can do is in order.

My first thought is that if CS graduate students really have 5-15 publications, then, yeah, maybe I would like to hire these people–at least they seem to have learned some useful skills! I don’t know how different the fields really are–I remember, about 10 or 15 years ago, one of my senior colleagues in the statistics department telling me that one of my papers must not be very good, because I had published only one paper on the topic. (His reasoning: if it were really such a good idea, it should be possible to get several papers out of it.)

Beyond this, I’ve always assumed that Ph.D. students in certain fields (computer science, also I’m thinking of economics and physics) are simply smarter, on average, than students in more run-of-the-mill fields such as statistics, math, political science, and sociology. I don’t mean that this is always true, but I’ve always thought (with no particular hard evidence) that it was true on the average. Of course, it’s important what you learn at these programs too. Top MBA programs are incredibly hard to get into, but that doesn’t mean I want to hire an MBA to do research for me.

But, as I’ve written before, the cool thing about a postdoc (compared to a faculty position, or for that matter compared to admissions to college or a graduate program) is that you’re hired based on what you can do, not based on how “good” you are in some vaguely defined sense. I like to hire people know how to fit models and to communicate with other researchers, and my postdocs have included a psychologist, an economist, and a computer scientist, along with several statisticians.

9 thoughts on “Affirmative action for statistics Ph.D.’s?

  1. Probably, it could be good for statistics students if we were encouraged by stats departments to try to write a paper for a machine learning journal.

  2. As a computer science graduate student, I can confirm that we are expected to submit a MINIMUM of one paper per year at the top conference in our area. The more prolific students write more, so the 5-15 number is quite realistic.

    OTOH, the tradeoff between journals and competitive conferences is something that's being debated a lot, at least in my area of CS (sorry, can't find a link). I think many people would like to move towards a journals-with-quick-turnaround model. One reason is that submitting research on a deadline is a great way to publish mistakes; another is the high burden on reviewers, which means papers with errors are more likely to get through. So perhaps in a few years stats and ML will find more of a middle ground in this respect.

  3. Hi. Why do you think fields such as economics, physics or computer science tend to attract, on average, smarter students? Also why math is not included? Of course, I'm just asking for a guess, not hard evidence. I'm just curious to hear about this from you. Thank you,

  4. I'm not entirely clear on which jobs you're discussing, but if you mean industry, it seems highly application dependent — but one place where ML researchers continually outshine statistics students (at least, based on my interactions here at MIT and with stats students at Harvard) is in the area of being able to _write code_. Understanding algorithms, algorithmic complexity, computing theory, large-scale software-systems, building complex infrastructure with abstraction, all are crucial components of the average CS curriculum that the CS-ML student has under their belt, and would seem to make them more appropriate for work in data-intensive industries.

    I'm continually shocked by stats students lack of understanding of computers and computation — few know of anything outside R, and write abysmal R code to boot. This isn't meant to suggest that there don't exist stats students who can't code or don't understand CS, but I get the impression that they're in the minority.

    A lot of statisticians seem to have a mentality of "programming the computer is a necessary evil", which most of my friends in machine learning lack. And with many contemporary interesting statistics problems (I'm looking at you, Google) requiring mastery of insane quantities of data, I'm not surprised the ML students are at an advantage.

  5. At my school (CMU), I really don't see stats and ML people applying for the same jobs very often (academic or otherwise) so I'm not too worried. For example, I'd bet my house that there are no ML students interested in what currently qualifies as my dream job, and I have no interest in working for Google, so things feel pretty harmonious. If there's any sibling rivalry, it's over the fact that ML people have no quals and higher stipends.

  6. Eric's comment interesting. I've always considered myself someone for whom a computer implementation with correctness verification is a tedious part of the job. I'm all for organized thought when it comes to laying out a software system, but software dev seems so inchoate to me that it's hard to strongly identify with any given way of doing things. I'd love any pointers to effective methods for thinking about software design. Most of what I've seen is fairly religious doctrine.

  7. This is an interesting post. I always imagined that it would be EASIER to publish in statistics than in other fields , simply because statisticians work across so many disciplines that the variety of papers and the subsequent volume of journals that the work can be published in is greater.

    I don't really know much about the internal workings of Machine Learning but I would imagine that if writing some kind of small but useful algorithm gets you published, it would be pretty easy to crank out papers.

    I think what is missing from this analysis is the relative newness of Machine Learning as compared with statistics. With much fewer people in the field and journals needing to fill their pages with SOME kind of research (perhaps even research of lower quality) it makes sense that Machine Learning grad students would face much lower levels of competition than statisticians trying to get papers in JASA etc.

    As for econ & physics & CS students being smarter, even assuming that that's true, intelligence doesn't necessarily translate into productivity, although that might be an interesting study…maybe correlating IQ with # of publications? Hmmm..

Comments are closed.