Graduate Student Guide
Graduate school preparation is the foundation for a successful career in statistics, and students preparing to enter the profession with a PhD are well encouraged to take some time to examine their goals before they finish.
Having a clear idea as to which path you wish to pursue after graduation allows you to work consciously towards the lifestyle you desire. While most people reading this guide will already be well on their way towards graduation, it’s worth mentioning the importance of the selection of a graduate program to the chance of success in a certain sub-area of the discipline.
The most famous graduate schools, for example, focus on fostering the next generation of statistical researchers. Their programs emphasize theoretical statistics and often encourage a high degree of specialization. Less famous programs aim at producing more applied statisticians who hope to find jobs in industry and government. Of course, many eminent researchers have been spawned by programs not mentioned near the top of the latest rankings, but if your aim is to do research at one of these highly-ranked departments, your best bet is to come from one as well. If your goals aren’t so focused on where you do your work, any of the top fifty should give comparably good training.
This section outlines the kinds of preparation one should undertake in graduate school. Much of the discussion is geared towards academic careers, but there is specific advice for other paths, and many suggestions are broadly applicable. Read Iglewicz (1998) for a summary of graduate programs.
Kafadar (1998) discusses career selection and Boen and Kjelsberg (1993) highlight what to do in graduate school given a particular career choice.
Courses to Take
For any career in statistics, one has to build a broad base in applied and theoretical statistics.
Virtually all departments steer students through the necessary classes, and we urge you to trust your advisor in selecting among the available options. Besides the traditional core curriculum, students can chart their course work with an eye to future employment. Don’t delay graduation excessively, but it often helps in the early years to have strength in a complementary discipline.
A few outside courses are often sufficient to accomplish this goal.
For academic research, it is wise to study as much mathematics as possible. Details depend on your interests, but popular choices are functional analysis, measure theory, combinatorics, and differential geometry. Real analysis and linear algebra are basic requirements.
If you have an interest in biostatistics, then medicine and biology are comparably valuable (Godinez 2002).
For industrial positions, it is a benefit to have a detailed understanding of the technology. Pharmaceutical companies are delighted to hire statisticians who know chemistry and biology; similarly, other organizations prefer candidates with a business or engineering background.
If one targets a career in a specific industry, doing a little relevant course work is sensible preparation. For government positions, appropriate background depends on the objective. For example, to work with the Environmental Protection Agency, one might study environmental science, biology or chemistry; for the Census Bureau, knowledge of demography/sociology would be useful, and for the Commerce Department, it would be appropriate to know a little finance.
For teaching positions, it often helps your marketability if you can teach basic courses in mathematics, such as linear algebra or differential equations. The value of such capability may have diminished recently, since statistics courses are gaining on mathematics for time in the undergraduate curriculum.
For consulting work (in or out of your current position) you would do well to have experience in a variety of outside disciplines. Knowing a little biology, epidemiology, engineering, agriculture, or economics will take you a long way.
You can sometimes glean relevant experience through consulting. Be alert to such opportunities in your department.
Skills to Acquire
Graduate school is the best place to attain computer literacy, professional writing skills, and experience in teaching and consulting. After graduation, there is little time or assistance for acquiring these skills. Most degree programs attempt to foster such breadth, and students should actively seek such opportunities.
Computer literacy is essential for a modern career in statistics. Some students focus exclusively on elegant math rather than confront the frustrations of computer hacking, but this is unwise. It is better to recognize the computational imperative early; in the long run, it saves time and improves the quality of one’s work.
Because of the rapid developments in computing, no specific hardware or software can be recommended. The most important computer skill is the ability to move easily between different computing environments. These environments consist of different subsystems, and competent statisticians can typically handle at least the following:
These create and manage computer files, and they link all the other facets of the computing environment. As one travels about in one’s profession, one is likely to encounter different systems, but the key concepts are fairly transportable. These days, a working knowledge of Linux/UNIX, Microsoft Windows, and/or Macintosh OS X will take you almost anywhere.
Examples include the Current Index to Statistics (www.statindex.org/CIS), Citeseer (citeseer.ist.psu.edu), Google Scholar (scholar.google.com) and the ISI Web of Science (isiwebofknowledge.com/). For researchers in biomedical statistics or statistical genetics, PubMed (www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=PubMed) can be useful.
It is also very common to have access to electronic copies of journals (e.g. JSTOR www.jstor.org and Project Euclid projecteuclid.org). Many of these services require a subscription, some of which are available to members of professional societies. It is a good idea to create a website for yourself in order to make your papers available to the research community as well as to provide teaching resources for students in your classes.
Statistical Software Packages
There are a number of packages differing somewhat in their capabilities, but all basically offering the standard analysis techniques, as well as various routines for special analyzes such as clustering. Many are available on both UNIX and Windows systems.
Many packages have the ability to be able to write new statistical routines, and do simulation studies. Usually, they offer built-in graphics, matrix manipulation and other useful features.
At some point during your statistics career you will likely be asked to do at least some computer programming. In statistics, the most commonly used language is R but many different languages are used depending on the task and the individual. Other major choices include C, C++, Java, and Python. Programming languages are always being created and improved so keeping an eye on new developments can ensure that your skills stay relevant. Most graduate programs now require you to learn a programming language as a course requirement.
Document and Talk Preparation Packages
Skill with a document preparation package such as LaTeX has become a professional necessity. These enable one to write professional quality letters, spiffy curriculum vitae, and articles with complicated mathematical expressions. LaTeX packages like Beamer (latex-beamer.sourceforge.net) can help in the writing of computer-based talks.
Symbolic Manipulation Routines
This is software that can handle symbolic algebra, differentiation and integration. Some of the more useful packages also graph equations and perform matrix operations. Examples include Mathematica and Maple and sage math provides a free alternative.
Graduate school is the last chance to polish your writing style before public and professional embarrassment. Fortunately, statistics is a discipline in which an adequate writing style can be achieved without excessive effort. There is generally a stock phrase in the literature for many standard situations. For review articles or topics that demand a discursive tone, one must rely on one’s natural skill and such manuals as Strunk and White (2000), University of Chicago Press Staff (2003), or Higham (1998).
The main rules for professional prose are “Be brief” and “Be simple”.
Keep in mind that a well written paper is likely to make a bigger impact and is usually worth the effort. It will also make the referees happy while they read it. A well written article is the product of many revisions. No matter how eager you are to submit your article it pays to set it aside for a couple of weeks so that you can give it one last revision before sending it out to the journal. If at all possible, get someone to read your article before it goes out for review. The best way to get someone to do this for you is first to do it for them.
Foreign students face greater difficulties. They are strongly advised to work at improving their spoken and written English throughout their career, since that is usually their single greatest obstacle in all professional interactions. Both the IMS and the ASA are sincerely sympathetic with such difficulties, and the editors take special pains to help polish good articles written in poor English. Most universities offer English as a second language programs for foreign students, and all non-native speakers should enroll immediately, and participate throughout their graduate careers. Also, when writing an article, curriculum vitae or letter, foreign students should invariably ask a sympathetic native speaker to proofread the work before it is distributed.
As a graduate student, you can benefit from refereeing an article with the help of your advisor. Usually an advisor will be delighted to delegate such tasks. The customary formula for a referee report is given below (see Norton (1994) for advice with more applied articles):
- Summarize the content of the article and its context in the literature (one to three paragraphs);
- Point out strengths and flaws in methods, concepts, proofs and coverage;
- Point out smaller errors of fact, style, grammar, citation and spelling; if the article has potential, make constructive suggestions for the minimum necessary improvement; if it is not salvageable, indicate this in a socially graceful way;
- In a separate letter addressed to the associate editor, make a specific recommendation as to whether the article should be published immediately (very rare), published after certain specific revisions have been accomplished (most common), reconsidered after substantial change or extensions (quite common), or rejected entirely (quite common). It can help to specify those revisions which you consider mandatory for acceptance of the article, and those revisions that are less major. An article can be inappropriate for the Annals of Statistics, but entirely suited for The American Statistician. Similarly an article can be inappropriate for The American Statistician, but entirely suited for the Annals of Statistics.
Other skills that are useful to practice while still in graduate school include teaching, consulting, and giving seminars. These are discussed individually in later sections.
Writing a Dissertation
A dissertation is supposed to be a new contribution to the field. An important test of its success is publication. Publications tend to be most important in academics, but they can also accelerate one’s rise in industry and government. A good rule, therefore, is to write the dissertation with the goal of producing papers.
Typically, the milestones in one’s progress on the dissertation are:
Choosing the thesis advisor
This should be done in conformity with one’s career path. For those planning academic careers, the ideal advisor is someone you can work with, someone whose research interests match your own, and someone who is well known. For those aiming at industry or government careers, an advisor’s reputation is probably less critical.
Choosing the topic
One should explore a bit before settling on the topic, and one should be flexible on the results one seeks. Some students ask their advisors for a thesis problem, but this is stultifying and should be avoided. Seeking out one’s own problem diminishes the likelihood of being directed towards a topic which represents a straightforward untangling of minor details of the advisor’s ongoing research. The best way to proceed is to start doing small scale research in an interesting area, and write small scale papers for your advisor (always with an eye to publication). Out of this, a strong dissertation will usually emerge.
Presenting the thesis proposal
At most institutions the thesis proposal is a major milestone, but elaborate preparation is probably a waste of time. However, this is the first of many research talks you will give, and you should work at being able to give a good talk. If you are already doing research and writing mini papers, then the proposal will evolve naturally.
When your advisor says you are ready, just tie your ongoing work together and you probably have a very adequate proposal.
Selecting the committee
The usual committee is a group of four, including the thesis advisor and a member of another academic department. The committee should be chosen with an eye to future letters of recommendation and complementary research strengths. Some faculty are notorious for not reading dissertations in a timely manner, and such people are an aggravation.
Writing the dissertation
If possible, write articles, not a dissertation. If the rules of your university allow, give your articles separate chapter numbers, write a literature review chapter for the front and a future research chapter for the back. Investigate this possibility with your advisor and the graduate school at an early stage in your career. For style and grammar tips on writing a dissertation read Little (2004).
Defending the dissertation
If your articles have already been submitted, then the referee reports will have you well-armed for the defense. Almost surely, your advisor won’t let you schedule until success is certain. But do a practice talk for some knowledgeable friends and your advisor, and try to be relaxed, confident and mentally nimble at your defense. Probably someone will throw a ‘curve ball’ question, but even if you choke they’ll try to pass you.
Some people feel that the thesis advisor has the right to expect that your research will lead to a joint publication, and that a student should desire this too. Not only does it give you a start in your career and improve the odds of the paper’s being accepted, but it also puts the advisor’s reputation on the line besides your own. An added advantage of publishing with your thesis advisor is that you’ll have the moral support of an experienced professional as you steer your maiden effort through the pitfalls of publication.
Perhaps equally many people believe that it is undesirable to make your initial publication a joint effort with your advisor. A single author publication quells suspicion that the research was more the work of the advisor than of the student. Furthermore, working toward an independent publication puts you in the driver’s seat early on, and encourages your growth as a creative researcher.
Leaving with a Bang
When you start your job, research will compete with teaching, consulting, committees, meetings and projects for your time. (Post-doctoral positions are a pleasant exception.) For this reason, having a running start from graduate school is a big boost to your professional career. There are several things you can do to give yourself this boost.
One is to cultivate relationships with others in your field, both fellow students and professors. The more people you know, the more chances you have for collaboration and interaction. Also, most employers want three letters of recommendation, so you’ll want friendly relations with faculty other than your thesis advisor.
Finally, it helps to have a collaborative effort underway before leaving graduate school; this broadens your research and enables a more rapid publication rate. For academics, the most important part of a good start is to have some articles already submitted to journals. These usually come from your thesis; however, it is also possible to engage in other small projects concurrent with your thesis which can be submitted while you are in graduate school. Many book-length theses do not lend themselves to the extraction of publishable articles. In this case, it is doubly urgent that some of the article writing be finished in graduate school.
Having said all this, it is worth emphasizing that many successful statisticians did not leave school with published articles and a network of collaboration.
Funding and the seasonality of the job market are major factors that hurry one along. The points in this section represent strategy rather than requirements.