Attitudes About Postdoctoral Training for Statisticians Evolve

Allan KarAlan F. Karr is director of the National Institute of Statistical Sciences and a professor of statistics and biostatistics at The University of North Carolina at Chapel Hill. He is a Fellow of the American Statistical Association and Institute of Mathematical Sciences and an elected member of the International Statistical Institute.

Statistics, like many other professions, has not been rapid in responding to changes in its scientific and human resources environment. However, one rather dramatic change has occurred over the past 25 years: the emergence of postdoctoral training. I believe this change has benefited everyone—the postdocs; those who mentor them; and the organizations in academia, government, and industry that become their employers.

In the physical sciences, postdoctoral training has been the norm for years. Statistics is not one of the physical sciences, however. Nor is our situation exactly like those of either of our sister disciplines, mathematics and computer science. In math, postdoctoral training is de rigeur, while it remains rare in computer science.

Cynics say postdocs constitute a labor force that is much more highly trained than graduate students and not much more expensive, but this misses the point. In settings in which postdocs are common, faculty understand that complex, high-impact research will not get done without the skills and focus of postdocs. Departments know graduate study alone is too blunt a filter to identify talented potential faculty. Nearly everyone realizes the teaching and service responsibilities borne by new faculty members can interfere with the essential transition from being an advisor-dependent graduate student to being an independent researcher with a distinctive research identity and strong funding. The early career scientists who become postdocs welcome the opportunity to emphasize research productivity, from papers to initial proposals. It really is an “everyone wins” situation.

One influence has been the array of career paths available to statisticians entering the profession. Some individuals and organizations think being a postdoc is irrelevant or superfluous. In fact, postdoctoral training is as valuable for statisticians working in government or industry as for those working in a university, because the independence it engenders is necessary everywhere.

The centrality of postdocs to statistics emerged contemporaneously with two of the other most notable trends in our profession over the past quarter-century: the increasing attention to cross-disciplinary research and the now-pervasive influence of computing. Many postdocs have more cross-disciplinary training and inclination than their mentors, and most are more adept at computing, often to the point it may not always be clear who is the mentor and who is the mentee.

How did this happen? Several organizations led the way by example. These include the National Institute of Statistical Sciences (NISS), the Geophysical Statistics Program at the National Center for Atmospheric Research and, more recently, the Statistical and Applied Mathematical Sciences Institute. The Division of Mathematical Sciences at the National Science Foundation has provided both dedicated awards for postdocs and support for postdocs across its entire research portfolio.

Of these organizations, I know NISS best. Our experience is substantial: More than 75 early-career statisticians and other scientists have been postdocs here. Despite strong self-selection, we receive dozens of applications for each position. Every NISS postdoc has exerted major influence on the research in which he or she has been engaged. Some have reshaped floundering efforts into more focused, productive ones. Our postdocs have always shared ideas among themselves, often with dramatic results. Working with collaborators and clients has improved our postdocs’ communication skills. The placement record is perfect, reflecting the growing appetite among employers for statisticians who have been postdocs. Indeed, this year, all four of our “graduating” postdocs secured tenure-track faculty positions.

So, are postdocs now part of the statistical infrastructure? Not quite. Employers across all sectors still show a tendency to talk the talk when it comes to hiring only those who have been postdocs. Most new degree recipients will turn down a postdoc in favor of a tenure-track faculty position, or a “permanent” position in industry or government, even though this should not be necessary. Any university should be willing, as NISS has, to allow a newly appointed faculty member to take leave for a year (or longer) in order to hold a postdoctoral position.

The current economic situation is anguishing, but we can react in positive ways. I urge all who have, or can generate, resources to support postdocs to make maximal use of the current opportunity to broaden understanding of the value of postdocs. Having robust and productive postdoctoral programs will enlarge the future of statistics. More to the point, we simply face a unique chance to make a profound scientific and personal difference to our profession.