(Note: This is the third in a series about graduate life in statistics, co-written by Mike and Greg. For links to all articles in the series, click here).
1. You’re on your own
Sure, you are going to take classes that are taught by professors, but you are the one responsible for learning the material. If you have a great professor, that’s wonderful, as it will probably be a lot easier to grasp the material and to do well on exams. If the professor is terrible, however, you still need to learn the material. And in college, you could learn that material, take a C, forget it and never think about that stuff again. In grad school, however, you are still responsible for that material, and in many cases its going to show up on your qualifying exams and/or general exam. (Shhhh: once you pass these, you can forget most of the material.)
Also, “learning” something as an undergraduate is a lot different that actually learning something in graduate school. As an undergraduate, you can just memorize something and regurgitate it for a test. In grad school, you might need that stuff to finish your dissertation. So it’s a little bit more important, and it’s a different kind of learning. I mean I took three chemistry classes and I know about ionic and covalent bonds, but I KNOW the Neyman-Pearson lemma and I KNOW the Central Limit Theorem and how to prove it, for instance (Well I used to anyway, I passed my general exam in 2010).
Relatedly, the reality with graduate school is that teaching evaluations and teaching quality are not the primary determinant of tenure for professors at most research institutions. On average, the incentives to be a really good teacher are lower for the professors teaching your graduate school courses than they were for the ones teaching your undergraduate courses. As a result, the onus is often on the graduate student to learn the material on his or her own.
2. Free time is important, and how you handle your free time will go a long way towards determining your success in grad school
You’ll learn a ton in graduate school. You’ll learn a ton in classes. But you’ll also learn an immense amount of stuff outside of the classroom.
Before you pass your exams, you need to spend a lot of this free time studying for exams. I can’t emphasize this enough. I was the kind of student in high school and college who didn’t really need to study that much to pass my math exams. One day for most exams was enough. This is not the case in graduate school (and if it is, you’re really, really smart). In the winter that I took the qualifying exam, I would wake up, read a little bit, and then study until about 3 or 4am. I did this for about 8 weeks with very few days off. I have never, ever learned so much and I passed the exam. Putting this time in is absolutely crucial. In fact, my advisor tells all of his students that your chances of passing the exam are simply a function of the time you spend studying. I mostly agree with that.
Once you pass the exams, you’ll have a different kind of free time; time to write, time to read, time to pick a dissertation topic (see part II). But also, its this free time in which you should…
3. Be prepared to code.
Be prepared to code. And then code some more. And then code some more. And then finally, after a few days – or, in some instances, weeks or months- you finally might get the convergence, property, or the result you hoped for. And when that happens, you still have more coding to do; to redo or revise your manuscript, and then again when you want to make your code ready for public consumption (which you should!)
If you’re in statistics or biostatistics, this often means learning to write R code (though Python, C++, etc. are also very useful). Basically, entire dissertations will be done writing R code. And while some people will have written code before graduate school, few are proficient when they enter. As statistics graduate students, your dissertation will be composed of methods that no one has ever used or properties that no one has ever shown. And because your work is new, the code will be new, too! This point sort of nails home points made in 1 and 2– while you’re on your own, use that free time to learn how to code. It’s really, really important.
Lastly, it is also not out of the realm of possibility to be expected to know several languages simultaneously. Many collaborators will want to use SAS, Stata, SPSS, or MATLAB. For example, NC State, which has one of the nation’s biggest and best statistics programs, also has a strong relationship with SAS. And so when NC State’s Marie Davidian wrote her longitudinal data analysis book, it’s no surprise that much of it was coded in SAS. Thus, many students around the country will learn longitudinal data in SAS, even if its not their language of choice. Expect this type of situation to come up at some point – even if you despise such a computing language, you might have to be good enough of at it to survive a class or two.
4. Get funded.
Funding means that you don’t have to work a job, which allows you to have the free time (see point 2) that you need to really learn this stuff and put your name out there in the literature. And getting paid to go to school is pretty awesome. It’s way better than crippling debt, imho. I can only speak for myself, but if you have to choose between to graduate programs and one is going to fund you and the other isn’t, I can’t think of many situations where I’d advise someone to go to the school that wasn’t going to fund you. If you belong in graduate school, someone, somewhere will fund you. You might just have to move across the country and give up on the idea of going to Harvard.
While most graduate programs will fund doctoral students throughout their careers as PhD students using research or teaching assistantships, there is also funding to apply for once you get to graduate school, even if you arrive already funded. For example, the National Institute of Health funds pre-doctoral students through F31 grants. These grants will allow you to forego any teaching or research requirement at your university, and to focus entirely on your research area (important caveat- there are restrictions as far as what types of students are eligible for F31’s).
While funding rates are relatively low compared to past years, the grant-writing experience is great preparation for life in academia after graduate school (or so I’ve been told. Repeatedly). Even better, funding rates for F31’s are higher for pre-doctoral grants than they are for post-doctoral ones, and there is no harm or cost in applying. Here’s a graph of the pre (F31) and post doctoral funding rates, via the NIH and ASA (and as a sidenote, it might also be the only time you’ll see an Excel time series plot on this blog!)
Final note on grants, Part I: the overwhelming majority of grants will not get funded on the first submission. Expect a second submission, which isn’t a big deal because most everything will have already been written.
Final note on grants, Part II: grants require three letters of recommendation from faculty members. Faculty members are busy. As a result, it’s not unheard of for the students to actually write their own letters, having to describe themselves in the process. Yes, this is awkward.
in the next part of this series, we will explore “What I wish I had learned in a graduate program in statistics”
Greg and I talked about some of this at JSM but it was more focused more on material than life lessons. I haven’t read through your whole series yet, but the thing is that it is essentially true. You CAN find time to do things in grad school so long as you get your stuff done and aren’t working a true full-time job. The stuff I didn’t do in grad school was on me but was not a function a time but a function of foresight, I could have done the same stuff and still got to the hockey games I wanted.
I suppose some of the balance is a function of talent. If you have to work your fingernails to stumps to graduate then the life balance will be out.
As for code. A statistician, unless in the rare position of being at a world class institution or a popular professor, will need to be his own code master and he/she is RARELY given proper tutelage. I think in hindsight the prefect PhD level stat has obtained at minimum a minor in Computer Science. I’m not asking anybody to go that way, but the foundations of algorithmic thinking and knowing how to find and operate with computational documentation are skills that will allow a statistician to fly and flourish. In the end, whether it be papers or otherwise, a research statistician can only move as far as his two feet will take him. After all, in the end, you don’t usually generate money, you’re on your own to carry your body.
Of course, separately, I believe a well functioning gov’t/industry research group should have support for strong computational roles. Its only natural in the evolution when one is not bound to the traditional restrictions of the university cultural and promotional setting.