What is the nature of the enrollment expansion today?
It’s like déjà vu all over again.
—attributed to Yogi Berra

Since 2007, the number of students taking computer science courses and declaring computer science majors has been rising rapidly. To a large extent, the expansion mirrors the earlier expansionary periods that occurred from 1979 to 1984 and from 1994 to 2001, in the years just before the preceding crashes. In particular, the factors fueling the increase in student interest are similar. The last seven years have been an extremely exciting time in computer science, with the proliferation of computing into ever more facets of everyday life. The ubiquity of smart phones and the applications that run on them, the enormous successes in machine learning, the rise of big data, and so many other advances all make computer science extremely attractive. Moreover, the seemingly boundless opportunities for employment that computing careers offer—particularly when coupled with the enormous uncertainty facing other aspects of the economy—are certain to draw more students into computer science, just as those same factors have in the past.

The sections that follow trace the dimensions of the current expansion and its likely consequences, exploring both the similarities and the differences from the earlier cycles.

The dimensions of the current expansion

Over the last four years, several members of the field who remember the history of the two previous cycles, including myself, have sought to raise awareness of the danger of rapid expansion in enrollment with no commensurate increase in teaching capacity. In 2011, I published an editorial entitled “Meeting the challenges of rising enrollments” in ACM’s Inroads magazine.20 In that piece, I review the history of the earlier crises and end with the following warning:

In the 1980s, the inability to hire new faculty made it impossible for most departments to satisfy the increased student demand. As a result, institutions were forced to discourage student interest by adopting such strategies as limiting the size of the computer science major or staffing courses with inadequately trained outsiders. . . .
    As a nation, we cannot afford to repeat the failures of the early 1980s. As we emerge from a decade in which far too few students chose to major in computer science, it makes no sense to frustrate the renewed student enthusiasm by turning yet another generation away because of a lack of resources. The economy needs more people with computer science training, and we have a collective responsibility to prepare students for those positions.

More recently, Ed Lazowska from the University of Washington, Jim Kurose at the National Science Foundation (on leave from the University of Massachusetts), and I have given joint talks in various venues about the growing capacity problem, including the NSF Future Directions in Computer Science Education summit in January 2014, the National Center for Women in Information Technology summit in May 2014, and the Computing Research Association’s biennial conference at Snowbird in July 2014. The slides from our presentation, entitled “Tsunami or sea change?: Responding to the explosion of student interest in computer science,” offer an overview of the capacity crisis and are available from http://lazowska.cs.washington.edu/NCWIT.pdf. The slides document the rapid growth in the number of computer science majors at several leading research institutions, both public and private, as shown in Figure 5.

Major growth figure

The expansion seen at these institutions is even more rapid than aggregate statistics show because many universities insulate themselves from changes in demand by controlling admissions to the major. For example, institutions like Carnegie Mellon and the University of Washington experience lower variability because those institutions admit students directly into the computer science program. The number of applicants changes along with the national pattern of student interest, but the number of students actually admitted remains much more stable.

The slides also document the continued strength of the job market for graduates with strong computational backgrounds. Figure 6, for example, shows the projections from the Bureau of Labor Statistics for job growth and job openings (job growth plus replacement) for the STEM (science, technology, engineering, and mathematics) sectors of the economy from 2012-22. By both measures, most of the employment growth over the decade is in the computing disciplines, which account for 71 percent of job creation and 57 percent of job openings in STEM.

BLS job growth

The “Tsunami or sea change” presentation ends with the warning that “we have seen this movie before, and it wasn’t pretty.” In previous cycles, “universities did not respond adequately,” increasing the importance of doing a better job this time around.

Today—as in past years in which student enrollments in computer science increased rapidly—faculty openings exceed the number of applicants for those positions. From July 2014 through June 2015, more than 700 distinct advertisements of computer science faculty positions at North American universities and colleges were posted on the ACM employment web site. Many of these ads listed multiple searches (sometimes as many as five) at the same institution, so the total number of open positions is larger than the number of advertisements. Although the precise number is impossible to determine because many of the listings use imprecise phrases like “several positions” or “multiple positions,” it appears that the number of open computer science faculty positions in 2014-15 was around 1000.

According to the Computing Research Association’s most recent Taulbee survey, North American institutions produced 1,651 computer science Ph.D.s in 2014.21 Of this number, 244 (15 percent) accepted faculty positions at North American institutions. By this calculation, the current rate of Ph.D. production is sufficient to fill about one of every four open positions.

Although the ratio of applicants to open positions is less than the one-in-seven shortfall of the early 1980s, the number of unfilled positions is significantly larger in absolute terms. If the number of Ph.D.s is sufficient to fill only a quarter of the open positions, then the number of positions that cannot be filled from this pool is around 750. Unlike other fields, computer science has no reserve labor force in the form of Ph.D.s who received their degrees in prior years but who have been unable to find positions. Some positions, of course, will attract faculty members from lower-ranked institutions who “trade up” to more prestigious employment. That flow of existing faculty up the ladder of institutional prestige, which is usually referred to as churn in discussions of the academic labor market, means that some of the 750 open positions will indeed be filled, but only in a way that leaves vacancies in other institutions that will have to be filled in future years. The only way to ensure stability is for the number of new faculty entering the workforce to keep pace with the rate of departures and the growth of the field.

Where are we heading?

Figuring out how to respond to the current expansion is complicated by the fact that history does not provide us with a clear sense of how the situation will evolve from here. The two expansionary periods we have seen before—which were quite similar in form—were followed by collapses that were qualitatively different. Are we heading for another capacity collapse similar to the one the field experienced in 1984, or will we be saved by a downturn in the high-tech industry that sends students scurrying away to other fields? And, perhaps more importantly, is there any way to predict the actual outcome?

Given the uncertainties of economics, it is clear that the answer to the second question is no: there is no way to predict with confidence exactly how economic forces will play out in the high-tech industry and how those forces will affect enrollment patterns. Many analysts believe that the situation in the technology sector is substantially different from the “irrational exuberance” of the dot-com bubble. As The Economist reported on July 25,22

Greed, profligacy, tiny companies with outlandish valuations: it is not hard to detect echoes of the turn of the century, when the dotcom bubble burst spectacularly and America’s economy stumbled as a result. But to see history as about to repeat itself is to miss how deeply things have changed. Today’s technology businesses are selling services and products from which they already generate income, rather than just saying that one day they might. And the group of people doing the investing is much smaller now than it was then. The risks are on fewer shoulders.

This assessment, of course, is by no means authoritative. The important point is that there is certainly no evidence to suggest that there will be a downturn in the high-tech industry that doesn’t affect the economy as a whole. More importantly, it is foolhardy to assume that there will be such a downturn and that academic computer science will be saved thereby. We don’t know what is going to happen, and it is therefore important to prepare for what might well be a capacity collapse similar to that of the 1980s.

There is, however, a reasonable interpretation of history that makes the differences in the mid 1980s and the early 2000s less important. Rather than looking at the character of the downturns, I believe it is more productive to focus on the pace of the enrollment increases that preceded those periods of collapse. In the early 1980s, the late 1990s, and again today, computer science departments face a rate of expansion that is much faster than university departments ordinarily sustain. Those periods of expansion, moreover, coincide with extremely tight labor markets for faculty, which makes it difficult to respond to the increase in student load, even if the institutional will to do so is there. In the absence of extraordinary measures that most universities have been unwilling to undertake in the past, the rates of increase during the boom years are simply unsustainable. And, in a marvelously succinct principle generally attributed to Herbert Stein, economics tells us that

If something is unsustainable, it will stop.

Stein’s principle does not tell us how an unsustainable phenomenon will stop, only that it will. The unsustainable buildup of the early 1980s ended with a capacity collapse. The unsustainable expansion of the dot-com era ended with the collapse of the dot-com bubble, and with it, the enrollment crisis that had threatened to overwhelm academic departments.

The current rise looks very much like the previous ones and exists against a backdrop of faculty shortages that bears all the hallmarks of past expansions. Unless new strategies are implemented at a scale that has not been attempted in the past, this expansion too will stop. Our foresight may not permit us to understand the precise mechanism of the collapse, but those details will matter very little to those who suffer from its effects.

What strategies are currently in progress?

As the effects of increasing enrollments become more evident, professional societies have undertaken several initiatives to address the capacity problems. These initiatives include the following:

Why has a concerted response from the community been so slow in coming?

Despite being several years into the latest period of skyrocketing enrollments, efforts to address the problems are just now getting off the ground. For those of us who have lived through past crises, this delay reflects an unfortunate change in the community’s understanding of the problems. In the early 1980s, academic computer scientists had a solid appreciation of the dangers they faced from massive increases in enrollment. The same was true in the late 1990s—an understanding all the more vivid because memories were still fresh from the capacity collapse of the mid 1980s. This time around, however, it has been much harder to get universities, departments, and individual faculty to recognize the risks, despite the accumulation of additional historical experience.

In my view, the failure in this cycle to benefit from the lessons of history arises from a confluence of several factors:

The fact that people who are responsible for making decisions that affect the future of computer science education are less aware of the problems of the past makes it harder for the field to act with a common purpose. That historical myopia, however, in no way reduces the importance of finding a way to forestall a repeat of the capacity collapse of the 1980s that cut the number of qualified computer scientists nearly in half. Our society cannot afford to repeat that mistake.