Among the small group of people who pay close attention to employment trends in science, technology, engineering, and mathematics (STEM) fields, the release each year of the Taulbee Survey report, from the Computing Research Association (CRA), is a notable event. The report is an exhaustive assessment of the state of training and employment in computer science (CS), emphasizing the Ph.D. level but including information on undergraduate and master's-level enrollment and degrees.

The 2014 Taulbee Report won't be released until May, but the CRA recently put out a "sneak preview" featuring data comparing CS degree production and enrollment, at the Ph.D. and bachelor's degree levels, from 2012 to 2013. There were two headline-grabbing results: In 2013, the number of CS Ph.D.s awarded grew to its highest level ever, up almost 8% over 2012 (itself a record year). And bachelor's-level enrollment rose by 21%, an increase that in any other field would seem astonishing, but in the volatile field of computer science isn't that rare: In the mid '90s, undergraduate CS enrollment grew 40% and 39% in consecutive years.

Those results mask a confluence of numbers that, to me, is even more interesting: New enrollment in CS Ph.D. programs is down 8.1%, despite a recent, quick rise in the number of bachelor's degrees granted in the field. Fewer of those more numerous bachelor's degree recipients, then, are choosing to go to graduate school in computer science, even as the number of still-enrolled undergraduates choosing the field explodes.

So, there are plenty of new graduates. The economy appears to be gaining steam. Computer science seems to be infiltrating many other scientific fields, and, with the new focus on "big data," is expanding into nonscientific fields. Why, under such conditions, would graduate enrollments suddenly start to fall?

There is an obvious answer: Whenever there are plenty of job opportunities relative to the number of prospective employees, they (the prospective employees) get to decide whether to take a job or go to graduate school, and there may not be enough of them to fill both pipelines. Graduate enrollment is falling because bachelor's degree graduates are finding, and taking, jobs.


CREDIT: Science and Engineering Indicators 2014
Select fields from Appendix Table 2-24, S&E Indicators 2014. The kink around 1007 is from a recalculation; there are two versions of the data for that year. (Click the image to enlarge)

That would also explain the undergraduate enrollment data. Clear-eyed undergraduates see graduates getting jobs, and they, too, pursue computer science training.

This answer may be obvious, but is it true? Couldn't there be some other reason why graduate enrollments are declining? Couldn't budget cuts, for example, be reducing the number of available Ph.D. training slots?

I asked around and found no evidence that graduate training opportunities are shrinking. Indeed, most graduate CS Ph.D. programs would prefer to grow, not shrink.

One of the people I contacted was Ed Lazowska, the Bill & Melinda Gates Chair in Computer Science & Engineering at the University of Washington, Seattle. (See our interview here.) Lazowska has a perspective on the field that is deep and broad—and also specific, since he sees up close what's happening at one of America's leading CS programs. I asked him what he thought of the new data. "Undergraduate enrolments and graduate enrolments are counter-cyclical," he wrote in an e-mail. "When there are lots of great jobs available, graduate enrolment falls, and undergraduate enrolment rises. When jobs are few, undergraduate enrolment drops, and graduate enrolment increases." Bingo. (It's important to note that this analysis indicates a healthy job market only for recent graduates. It has nothing to say about the job market for computer scientists who are further along in their careers.)

A look at the historical data—Taulbee Surveys from the last 20 years or so—reveals some complications. In the Ph.D. enrollment data, one must take into account the steady rise in the enrollment of foreign students in Ph.D. programs; in 2012 about 6 in 10 was a nonresident alien. One must also consider the ups and downs in the feeder population: The larger the number of bachelor's degrees granted, the larger the pool of candidates from which employers and graduate schools can draw. Finally, we really should be considering total graduate enrollment and not just enrollment in Ph.D. programs, but CRA's sneak peak doesn't provide data on master's degree enrollment. All this considered, the historical data does seem consistent, over the last 20 years, with Lazowska's countercyclical proposal.

Can an analysis like this be applied to other scientific fields? Why not? There's no reason to think that physics, chemistry, or the biosciences would be any different from computer science. In those fields, too, rising undergraduate enrollment coupled with declining graduate enrollment would indicate a healthy job market.


CREDIT: Science and Engineering Indicators 2014
Select fields from Appendix Table 2-23, S&E Indicators 2014. (Click the image to enlarge)

But, according to National Science Foundation (NSF) statistics, such conditions haven't existed at least since 2000. Since that year—the first year included in historical tables in the 2014 Science and Engineering Indicators—the rise in Ph.D. program enrollment has been slow but inexorable. Computer science is the only field where graduate enrollment has shown any real volatility, the only field where changing market forces seem to have made a mark.*

What about bachelor's degrees? The number of undergraduate degrees granted has risen almost as steadily as graduate enrollment, with some brief declines following 9/11 and fluctuations in small fields. Again, the overwhelming trend is a slow and steady rise.

In other words, in the traditional fields of science there's no indication of the kind of dynamism seen in the computer science data; there is no evidence that post-baccalaureate job opportunities have stolen away potential graduate students, in any of the traditional scientific fields, at any point between 2000 and 2011 (the last year covered by the latest NSF data). The picture that emerges instead is of a long-term steady state, where market forces are too weak to make much impact on the numbers. (For an explanation of why, see, for example, Michael Teitelbaum's analysis posted today at The Atlantic.)

The data also suggest, however, that employment in these traditional scientific fields isn't terrible. It would be encouraging for job seekers (and prospective scientists) to find in the data evidence of strong demand for their skills. It isn't there. But the steady rise in undergraduate enrollments—most of it preceding the intense pro-STEM propaganda we're seeing now—is slightly encouraging. If there were a true employment crisis in the major STEM fields—if new graduates weren't finding satisfactory jobs—then undergraduates would get the message and choose a different field. The Taulbee undergraduate enrollment data from 2002 to about 2007, and the NSF bachelor's degree data for computer science from 2005 through 2010, demonstrate that point.

It is possible (though not likely) that, for the first time in years, the turn toward computer science will significantly impact enrollment (and, in time, degree production) in the traditional scientific fields. Last year, according to the Taulbee "sneak peak," 12,500 more undergraduate students chose to study computer science than had done so the year before. That's about 2.5% of all the science bachelor's degrees granted in a given year. It will be years before NSF data will capture the effects of this apparent turn toward computer science, but when it finally does, it may show a downturn in degree production in other scientific fields. That prospect may terrify science policymakers, but it would be good news for job-seeking early-career scientists.

* Some caveats: NSF tracks undergraduate degrees, not enrollments, and the NSF numbers are for all graduate enrollments, not just in Ph.D. programs. Also, I'm looking only at enrollments in the natural sciences; engineering enrollments fluctuated in the middle of the previous decade. Those who are inclined to check my work—to look at the data for themselves—will notice—in Appendix Table 2-24 from the Science and Engineering Indicators—that the top-level data for Ph.D. enrollment started to decline in 2008. But a closer look shows that only two subfields turned down—psychology and "other health sciences"—and both rows are footnoted: In 2008, NSF got more aggressive in filtering out clinical-only graduate programs. The apparent decline in graduate enrollment in those fields isn't real.

 

Jim Austin is the editor of Science Careers. @SciCareerEditor on Twitter

10.1126/science.caredit.a1400071