Things to consider about masters programs in statistics

Here is a somewhat blunt set of general thoughts and recommendations on masters programs in statistics. I don't intend for them to be polemic, though some of them may sound like that.

I am going to assume that you are interested in a terminal masters degree to later go into industry and are not interested in potentially pursuing a doctorate. Please do not take this reply as authoritative, though.

Below are several points of advice from my own experiences. I've ordered them very roughly from what I think is most important to least. As you choose a program, you might weigh each of them against one another taking some of the points below into account.

1. Try to make the best choice for you personally. There are very many factors involved in such a decision: geography, personal relationships, job and networking opportunities, coursework, costs of education and living, etc. The most important thing is to weigh each of these yourself and try to use your own best judgment. You are the one that ultimately lives with the consequences of your choice, both positive and negative, and you are the only one in a position to appraise your whole situation. Act accordingly.

2. Learn to collaborate and manage your time. You may not believe me, but an employer will very likely care more about your personality, ability to collaborate with others and ability to work efficiently than they will care about your raw technical skills. Effective communication is crucial in statistics, especially when communicating with nonstatisticians. Knowing how to manage a complex project and make steady progress is very important. Take advantage of structured statistical-consulting opportunities, if they exist, at your chosen institution.

3. Learn a cognate area. The greatest weakness I see in many masters and PhD graduates in statistics, both in industry and in academia, is that they often have very little subject-matter knowledge. The upshot is that sometimes "standard" statistical analyses get used due to a lack of understanding of the underlying mechanisms of the problem they are trying to analyze. Developing some expertise in a cognate area can, therefore, be very enriching both statistically and professionally. But, the most important aspect of this is the learning itself: Realizing that incorporating subject matter knowledge can be vital to correctly analyzing a problem. Being competent in the vocabulary and basic knowledge can also aid greatly in communication and will improve the perception that your nonstatistician colleagues have of you.

4. Learn to work with (big) data. Data sets in virtually every field that uses statistics have been growing tremendously in size over the last 20 years. In an industrial setting, you will likely spend more time manipulating data than you will analyzing them. Learning good data-management procedures, sanity checking, etc. is crucial to valid analysis. The more efficient you become at it, the more time you'll spend doing the "fun" stuff. This is something that is very heavily underemphasized and underappreciated in academic programs. Luckily, there are now some bigger data sets available to the academic community that one can play with. If you can't do this within the program itself, spend some time doing so outside of it.

5. Learn linear regression and the associated applied linear algebra very, very well. It is surprising how many masters and PhD graduates obtain their degrees (from "top" programs!), but can't answer basic questions on linear regression or how it works. Having this material down cold will serve you incredibly well. It is important in its own right and is the gateway to many, many more advanced statistical and machine-learning techniques.

6. If possible, do a masters report or thesis. The masters programs associated with some of the top U.S. statistics departments (usually gauged more on their doctorate programs) seem to have moved away from incorporating a report or a thesis. The fact of the matter is that a purely course-based program usually deprives the student of developing any real depth of knowledge in a particular area. The area itself is not so important, in my view, but the experience is. The persistence, time-management, collaboration with faculty, etc. required to produce a masters report or thesis can pay off greatly when transitioning to industry. Even if a program doesn't advertise one, if you're otherwise interested in it, send an email to the admissions chair and ask about the possibility of a customized program that allows for it.

7. Take the most challenging coursework you can manage. While the most important thing is to understand the core material very, very well, you should also use your time and money wisely by challenging yourself as much as possible. The particular topic matter you choose to learn may appear to be fairly "useless", but getting some contact with the literature and challenging yourself to learn something new and difficult will make it easier when you have to do so later in industry. For example, learning some of the theory behind classical statistics turns out to be fairly useless in and of itself for the daily work of many industrial statisticians, but the concepts conveyed areextremely useful and provide continual guidance. It also will make all the other statistical methods you come into contact with seem less mysterious.

8. A program's reputation only matters for your first job. Way too much emphasis is put on a school's or program's reputation. Unfortunately, this is a time- and energy-saving heuristic for human-resource managers. Be aware that programs are judged much more by their research and doctoral programs than their masters ones. In many such top departments, the M.S. students often end up feeling a bit like second-class citizens since most of the resources are expended on the doctoral programs.

   One of the brightest young statistical collaborators I've worked with has a doctorate from a small foreign university you've probably never heard of. People can get a wonderful education (sometimes a much better one, especially at the undergraduate and masters level!) at "no-name" institutions than at "top" programs. They're almost guaranteed to get more interaction with core faculty at the former.

   The name of the school at the top of your resume is likely to have a role in getting you in the door for your first job and people will care more about where your most advanced degree came from than where any others did. After that first job, people will care substantially more about what experience you bring to the table. Finding a school where lots of interesting job opportunities come to you through career fairs, circulated emails, etc., can have a big payoff and this happens more at top programs.

A personal remark: I personally have a preference for somewhat more theoretical programs that still allow some contact with data and a smattering of applied courses. The fact of the matter is that you're simply not going to become a good applied statistician by obtaining a masters degree. That comes only with (much more) time and experience in struggling with challenging problems and analyses on a daily basis.