Tuesday, October 02, 2012
Baumol and Big Data
If you read only one book about higher education this year, read The Cost Disease, by William Baumol. It’s essential, brilliant, and even readable. And it answers an important question – why the costs of health care and education keep going up – more intelligently than anything else I’ve seen. (My book doesn’t come out until January, so he can have this year…)
William Baumol is an economist who gained fame in the 1960’s for discovering what came to be known as “Baumol’s cost disease.” It’s the observation that rates of productivity growth are uneven across industries, and that the products or services of the industries with lower productivity growth will gradually and inexorably become more expensive than those of the industries with higher productivity growth.
In the original paper, Baumol looked at a Mozart string quartet. It takes just as many musicians just as long to play today as it did two hundred years ago; in real terms, the productivity increase for string quartets has been zero. By contrast, the productivity growth in telecommunications, driven almost entirely by technology, has been astronomical. The rate of growth in telecom has been so high that companies can lower prices and increase profits at the same time. (Bless him, Baumol prefers to look at increased wages, rather than increased profits. Suffice it to say that actual telecoms don’t always work like that.)
Baumol isolates human labor as the key factor. (There’s a vague echo of Marx’ labor theory of value, though Baumol doesn’t acknowledge it.) Industries that have been able to replace expensive human labor with inexpensive machinery (when amortized over lots of production, anyway) have been able to lower costs for everyone. The share of GDP they’ve forfeited by lowering costs has moved over to what he calls the “stagnant” sector, which is the set of industries in which production is still very hands-on and therefore hard to speed up. Education and health care are the most conspicuous examples, though it also holds for live entertainment, restaurants, and legal services.
To take an easy example from the book, in the 1800’s, most Americans worked in agriculture. Now only about three percent do, yet that three percent produces far more food than the majority of the country was able to produce just a few generations ago. Technology has been the key. Even in manufacturing, though the job losses since the midcentury peak have been dramatic, there’s still plenty of stuff being produced; the key change has been that the producers have become so much more efficient that they don’t need as many employees.
Education and health care haven’t enjoyed those gains. They’re still producing in much the same way they did fifty years ago. As a result, salary increases for people who work in education and health care aren’t paid for through increased production; they’re paid for by charging more. Instead of the virtuous cycle of higher wages and lower prices that the more “productive” sector has enjoyed, we’ve endured the vicious cycle of lower wages and higher prices that comes from a long-term productivity squeeze.
Baumol is oddly content with that. He argues throughout the book that when the economy grows enough as a whole, the fact that a few sectors are taking up more of it really doesn’t matter much. I can’t share his confidence, since I can’t help but notice that middle class wages have stagnated for decades even in the face of annual productivity increases. We’ve had a self-reinforcing distribution problem that has concentrated the gains of productivity in the top tier, leading to a politics of resentment among the rest. (One could easily explain the election as the two parties offering different versions of who to resent.) Baumol is right that GDP as a whole can certainly handle increased tuition, but it doesn’t follow that therefore a middle class family can. To paraphrase Tug McGraw, ya gotta disaggregate.
Still, the real breakthrough of the book for me is the discussion of Big Data as a way around the cost disease. The cost disease happens – entirely without villains, corruption, or ill intent – when some sectors increase productivity faster than others for an extended period. Baumol notes that one way around the problem is to zero out productivity growth altogether. He’s right, mathematically, but I don’t think we want to live that way. So the other way around it – which the book explores in the context of health care, but largely ignores for education – is for the slow-growth sectors to pick up their pace.
Baumol contrasts car manufacturing to car repairing: the former allows for easy productivity improvements, but the latter doesn’t. As Tolstoy would have noted, every new car is the same, but every broken car is broken in its own way. (That’s from “Vanna Karenina,” about a minivan in a bad spot.) It’s harder to cut costs on the repair side because the diagnostic step still requires trained human intervention.
And that’s where we in higher ed have an opportunity, though Baumol doesn’t raise it himself. We’ve tried to use midcentury standardization – “take out your number two pencils” – with mostly terrible results. But we haven’t used Big Data more strategically, as a diagnostic.
Amazon.com does this all the time. When I log on, it offers suggestions based on what I’ve bought or looked at before. When I look at a particular book, it offers “others who bought this also liked…” It looks simple, but there’s an impressive amount of data crunching that makes that possible, and the data crunching has only become possible as computers have hit their stride.
As an industry, we still use the artisanal mode of production. Until recently, there was a good argument that it was the best we could do. But I’m not sure that’s true anymore.
To me, the really exciting prospects offered by Big Data and MOOCs and, yes, outcomes assessment, is in helping us allocate human intervention most effectively. In health care, for example, evidence-based reforms have led to procedural checklists that have resulted in better patient outcomes at remarkably low cost. Instead of basing incredibly expensive institutions on anecdotes and self-reinforcing hunches, we have the option of starting to base them on what actually works. Instead of denoting learning in units of time – the credit hour – which defeats any possibility of productivity improvement, we could experiment with different ways of improving student capabilities. If we find a way to get a student to mastery faster, well, that’s productivity.
Oddly for such an insightful book, Baumol seems to take for granted that certain activities are simply trapped, and must be trapped forever. That strikes me as a failure of imagination. Since I don’t share his political complacency, I also don’t share his sense that, say, higher education will just keep on getting more expensive forever and that’s okay. It’s not okay – I have polls and an Occupy movement to prove it – and it’s not necessarily inevitable. We’ll need to be willing to make some pretty drastic changes, but that’s what smart people who intend to survive do. It’s how it’s done. We can do this.
Still, Baumol’s formulation is the only serious answer I’ve seen to why the costs of education, law enforcement, and health care have gone up drastically around the world for decades. It isn’t about this administrator or that one; if that were the issue, we’d see exceptions. The fundamental problem – the issue that colleges have tried to evade with adjuncts – is structural. It’s Baumol’s cost disease. Until we come to grips with that, we’ll be stuck in the politics of resentment.
However, I am less certain that the kinds of efficiency or productivity improvements that you envision are going to halt the cost growth. There have been plenty of productivity improvements in higher ed over the years; we have been one of the most aggressive industries in putting computers to work on our behalf. But the result has been a race to the top, not a race to the bottom: students and their parents have demanded that we use our increased capacity to deliver a higher "standard of care" (to use A&F's term), and have so far been willing to pay the cost given the increasing return that comes with a high quality college education. They grumble, but there are still lots who can afford to pay who choose to attend the best college they can afford, not the cheapest.
And on the average, I think Baumal is right that it is OK for a richer society to pay more for their artisanal services. In some global sense, the same productivity growth in manufacturing that drives cost disease also increases society's wealth at an even greater rate than needed to pay the cost of services. But the disaster, of course, is that society's wealth is increasingly distributed very inequitably, and higher education itself is at risk of splitting increasingly into a tier of wealthy schools for the wealthy, and poor schools for the poor. By focusing on cost, rather than equity, as the real crisis, we risk accelerating the movement of higher education away from its traditional role as an engine of opportunity and class mixing. The rich will always buy the best education they can afford, since the investment returns are so large, and there will always be schools competing to be "the best." But the populist rage that has been focused (mostly) on public institutions and their funding will ensure that the best the middle class will find is a "managed care" version of college that is increasingly distant in quality from that experienced by the wealthy.
In a world with increasing use of adjuncts and TAs, how can anyone pretend that there has not been an increase in student credit hours per instructional dollar assigned for instruction? You can argue that as a whole, higher education has a number of costs not contributing to whatever instructional outcome you want to measure, but if you narrow the field down to reasonable instructional costs, I think community colleges and adjuncts disprove Baumol's relevance to instruction.
Does the same relationship hold true for the price of running a university? Is it always equal to x times a certain salary, or does it increase faster than that?
As I have argued before, perhaps one COULD present data showing that the greatest productivity increases took place at large universities in that first decade, between 1962 and 1972 when larger classrooms and video delivery of instruction grew explosively, but I'd first have to be convinced that the cost part of the calculation over the past 50 years is reliable. Adjuncts were not so common in 1972. Is it correct to assign the full salary of the many professors whose time is spent doing research and graduate education is even relevant to a calculation of the cost of undergrad education just because that is how the university pays for it?
I know that students (and taxpayers) pay significantly more for the same English or Math class depending solely on where it is taught (i.e. the same person is teaching on both campuses), but I've never seen those comparisons in print or how they have changed over time.
I just wish I had as much spare time as administrators seem to have, so I could read and dissect the new book by Baumol et al. Does it explain why pay for doctors has increased faster than other artisans, such as engineers or adjunct instructors? Are there even any inflation-corrected tables of those salaries within a fixed reference frame such as a major R1 and its teaching hospital and its administration?
Perhaps the labor intensive part of another industry would clarify what has changed? For example, it seems to me that the ratio between the salaries of senior executives and top engineers has changed radically in the past 50 years. What part of health care cost can be attributed directly to time spent with the doctor, and how much of that cost is overhead such as malpractice insurance and patient management and corporate management?
This is where I see problems for health care and education. Clearly, a big data approach is both necessary and useful, but the kinds of data we have access to are very limited and largely of the wrong kind, especially in education. We have lots of outcome data (i.e., "purchases" like course registrations and grades) and basic demographic data but very little data on motivation, satisfaction, etc., plus we have almost no longitudinal data (how do students do when they leave school?). I think the big data approach will be helpful, but it's also going to generate a lot of garbage. It will take very thoughtful and careful administrators to navigate what "recommendations" get spit out.
1. The need to support high-technology. A college classroom used to need only desks, chalk, and a blackboard, but now you need a multimedia projector, WiFi (ethernet plugs at every seat), a computer for every student, power outlets everywhere, and even the blackboard has usually gone high tech.
2. All of those federally-mandated and state-mandated rules and regulations that have been imposed on colleges and universities add to the cost. Colleges have to maintain staff to show that they are in compliance with this regulation and that regulation. A lot of the so-called "administrative bloat" that many people in academe complain about can be blamed on the need to show compliance with this thicket of rules and regulations.
3. The requirements of obtaining and maintaining accreditation. A lot of time, cost, and effort must be expended in order to keep the accrediting agencies happy.
4. Top-tier research universities such as Harvard, Yale, Princeton, or the University of Chicago must maintain and support an extensive research program for their faculty. The support of research is quite expensive, especially in the experimental sciences and in medicine. Not only top-tier institutions, it now seems that just about every college or university seems to want to turn itself into an “R1” institution, in order to jack up their ratings on the US News and World Report annual rankings.
5. The cost of healthcare and pensions. The rapidly rising cost of medical insurance for the faculty, the administrators, and the staff adds to the cost. Pension costs are also rapidly rising.
6. The cost of liability protection. There are gaggles of lawyers waiting to pounce on just about every accident or infraction, and colleges and universities must protect themselves against being sued by an irate student, an angry parent, an injured employee, a person who was sexually harassed, or even by an assistant professor who was denied tenure. Universities now have to maintain legal teams to handle these issues (as well as issues like disputes over intellectual property, etc.), and they have cumbersome procedures which consume time from faculty and administrators to give "due process" to everyone.
7. Declining state and federal support! This is a HUGE factor in state colleges.
8. The star system salary bloat. It's now not uncommon for the football coach, the basketball coach, the university president, plus several high-level administrators to get a million dollars a year, 10-20 times what a typical faculty member gets. The same thing happens with faculty in very competitive fields like biotech or business. This just wasn't the case in the past. It's not that the typical English or math professor is getting a whole bunch more (certainly not true for the adjuncts), but the "spread" has been greatly exaggerated.
9. Poor investment performance for endowment funds.
DD, I didn't follow how these two ideas are related. Could you expand on this point? To me, these are independent facts, one having to do with economic reality and the other having to do with the suppression of the labor movement through changing laws.
The $1 million challenge now is how to find an algorithm that is good for both subsets, and also keeps track of how different people use the star system.
In a very real sense, I worry that people don't realize just how laborious 'big data' is. We're very close to the point where the limitation in genome sequencing isn't the chemical reactions, but the analytical power (ok, arguably, we're way past that point; there are a lot more genomes than annotations for genomes out there!).
If the best solution Baumol's cost disease is itself highly labor intensive, sooner or latter all the money will be spent on figuring out how to spend less money. How's that for administrative bloat?
"I am at a top tier public research university and the only money 'invested' in research is the part of overhead that the university is required by federal mandate to reinvest."
That is simply not true. You are overlooking the large part of your salary that is unrelated to undergraduate teaching duties and productivity. You are not competitively hired, compensated, or granted tenure based on the number of undergrads that pass through your classroom to a degree each year. Quite the opposite.
What other explanation is there for requiring significantly higher fees for the exact same class taught by the exact same adjunct instructor just because of the brand name on the buidling?