Of Education Bubbles and Bad Graphs

The new school year is upon us, so there's been a lot of talk about academia and how it works recently. This has included a lot of talk about the cost of higher education, as has been the case more or less since I've been aware of the cost of higher education. A lot of people have been referring to a "Student Loan Bubble," such as Dean Dad, who points to this graph from Daniel Indiviglio as an illustration:

That post is a week old, which is a hundred years in blog time, and I wish I'd gotten to it sooner, because it's a terrible graph. Indiviglio says:

This chart looks like a mistake, but it's correct. Student loan debt has grown by 511% over this period.

He's half right. He's right that it looks like a mistake, but he's wrong about the reasons.

The problem with the graph isn't the scale of the increase, it's, well, everything else. The vertical axis has units of percent, but percent of what? The labels say "Cumulative Growth," but that doesn't really clear anything up. I mean, it strongly suggests that this is the increase in debt level over some initial value, but then it's impossible to make any sense of the three-quarter period in 2002-ish, when it goes negative. A negative growth rate might make sense, but a negative cumulative change makes no sense at all. To get that, you would need to be paying students a significant sum to go to college, wiping out their past loans completely. And you would need this change to have happened all at once, lasted three fiscal quarters, and then been reversed completely, bringing the debt back up to the previous trend line. There's another discontinuous jump, around the end of 2004, that doesn't make a lot of sense, either. So I don't know what the hell these numbers are that are plotted.

But the biggest problem here is that it's plotting some sort of percentage growth figure for two very different quantities (student loan debt and total household debt less student loan debt), without giving you the information you need to do a reasonable comparison of the two, namely the absolute starting value of these. He almost gives the right numbers in the text-- saying that student loans increased from \$90 billion to \$550 billion, but then he doesn't quite give the right number for the other figure, giving only the housing component, which he says increased from \$3.28 to \$9.98 trillion at its peak, an increase of about 200% where the graph shows an increase of only 150%.

These numbers are enough to make one important point, though, namely that the student loan debt is not actually that big a fraction of the total national debt load, contrary to the really alarming impression you get from that terrible graph: if we put everything in the same units, the student loan component of debt has increased from \$90 billion out of at least \$3,280 billion to \$550 billion out of something more like \$8,000 billion. Which is a big change, to be sure, but the way this is graphed gives a completely misleading impression. On top of the fact that the trajectory looks completely out of whack.

This is a shame, because the fact that the highlighted chart is garbage gets in the way of a very valid point: the cost of higher education has increased faster than inflation, and shows no real sign of slowing down. It is getting more expensive to attend college, and this is going to cause a problem eventually.

It's not like people in higher education aren't aware of this-- believe me, we talk about this a lot. And as bad as that graph is, it did at least prompt a good response from Dean Dad, highlighting some of the problems with discussions of this situation:

After slightly over a decade in higher ed administration, I'm increasingly convinced that change will have to come from outside. The forces of inertia from within are as powerful as they are shortsighted. They insist on continuing to frame a structural problem in personal terms.

To take one example, Benjamin Ginsberg seems to think that "deanlets" are the problem, which might make sense if their numbers weren't actually decreasing. Staff is increasing, but management is shrinking. And the staff increases are mostly concentrated in a few discrete areas: IT, financial aid, and students with disabilities. Prof. Ginsberg is invited to specify which of those he considers unimportant.

Among the blogs, you'd get the impression that the biggest problem facing higher ed was its overreliance on adjuncts. Put differently, you'd get the impression that colleges are too frugal. The preferred alternative usually offered is a dramatic and sustained increase in labor costs. From whence the money to pay these increased costs would come is usually left to the imagination.

Another really good discussion of this comes from the reliably excellent Timothy Burke at Swarthmore, whose post is too long to quote, but includes a really nice breakdown of the factors driving the cost. Both Burke and Dean Dad gloss a little too quickly over one of the biggest drivers, namely health care costs (in both posts, it gets brought up in the comments). Tuition has been increasing faster than inflation, true, but so has health care. In fact, the cost of health care for students, faculty, and staff is one of the major budget concerns here, and I suspect at most other institutions as well. If some way is found to control the growth of health care costs, then that will make it easier to control the growth of college tuition.

Ultimately, though, this is a really difficult problem, with no obvious solution. And it's been a really difficult problem with no obvious solution for a very long time. Back when dinosaurs roamed the earth and I was a sophomore in college, tuition at Williams broke the \$20,000 mark for the first time-- \$20,760. I remember the exact figure because an outraged student spray-painted that figure on the six pillars of Chapin Hall in the center of campus. There was a lot of worried discussion at the time about the increasing cost of higher education, and a lot of people saying that this couldn't go on indefinitely.

And here we are, 20 years later, and tuition at Williams is \$42,938, with miscellaneous fees bringing the total cost to more than \$54,000. And we're still having the exact same conversation about the increasing cost and how it can't go on.

While it's undeniably true that this can't go on forever, it's also true that nobody has the foggiest idea how much longer it can go on. The real reason that tuition continues to increase rapidly is, ultimately, that parents and students continue to pay the higher tuition. (I said this at a faculty event once, and a colleague from economics said "Congratulations. You are now an honorary economist.") The rate of increase will slow the moment that it becomes clear that people are being priced out of the market, and we haven't hit that point yet. Where that point is, nobody can really say until it starts to happen.

Which is, in some sense, the very definition of a bubble. And yet, calling it a bubble doesn't give any clear idea of a path forward. In some sense, it argues for continuing to increase tuition at a high rate for as long as you can get away with it, so as to have some cushion when things start to come apart.

To end on a bit less of a downer note, though, let me leave you with the half-joking explanation an administrator offered for why the \$50,000+ price tag for a year at an elite private school is actually a bargain. This person noted that the college-age child of a mutual acquaintance had recently had surgery, and the bill for one night's stay in the hospital-- not the surgery itself, just the hospital stay-- was \$11,000. Compared to that, college is a bargain-- yeah, tuition, room, and board may run you \$55,000, but that covers you for nine months. The per-day cost of college is around 2% of the per-day cost of an overnight hospital stay, making it much, much cheaper to be a student than to be sick.

Tags

More like this

While I agre it's a horrible graph, wouldn't "cumulative growth" here just mean "accumulated change relative to the baseline level"? So a "cumulative growth" of -10% in some year would mean that the level in that year was 90% of the baseline level? Better, I'd think, to follow the usual practice of plotting everything with baseline year = 100, and better still to use a log scale, but not total nonsense.

Cosma @1: You are entirely correct that "cumulative growth" is a valid concept, even if it is misleadingly presented here. But as Chad points out, some of the numbers for the student loan trace don't pass the smell test. You don't go from a cumulative growth of (eyeballing it) +40% to -20%, or -10% to +50%, in a single quarter without some major external event. Perhaps there were some redefinitions of what constitutes student debt (a temporary one for the 2002-03 academic year, and a permanent one in 2004), but in that case it makes no sense to plot it as a cumulative series--the point of such a series is to look at changes in some well-defined quantity.

By Eric Lund (not verified) on 24 Aug 2011 #permalink

Why does the graph label it "Cumulative Growth of Household Debt Less Student Loans" and "Cumulative Student Loan Growth" not "Cumulative Growth of Student Loan Debt".

Is it actually looking at total debt from student loans or just the amount of student loans given per year?

Cosma: Pretty much "What Eric Lund said." I agree that the sensible definition of the quantity on the vertical axis is:

100*(Current total debt- Initial total debt)/(Initial total debt)

and that using that definition, a negative value just means that the total debt at that time is less than the initial debt.

But that's the thing that doesn't make sense. For that to be the case, the student loan debt nationally would've needed to increase slowly to about \$140 billion, drop suddenly to about \$80 billion, stay there for three quarters, than jump back to about \$150 million, increase slowly to maybe \$160 billion, then jump suddenly to \$225 billion.

The only way that makes sense is if the definition of "student loan debt" changed three times during the span of the graph. But in that case, it's completely inappropriate to plot the data in the manner, because you're not using the same definition for it at the end of the window that you were at the beginning of the window.

Now, you could fairly say that student loan debt has increased from \$225 billion in 2004 to \$550 billion today (assuming the relatively smooth data since the big jump in 2004 means that the definition hasn't changed in that interval), which is still a big increase. While it has the virtue of being an honest description of the situation, though, it's not nearly as effective for stoking a sense of alarm and driving hits to The Atlantic's economics blog.

If those are supposed to be quarters (I honestly can't read the numbers on the horizontal axis, which is of course another strike against the plot), then yeah, those swings are completely implausible.

The illegibility of the axes isn't the fault of the plot-maker. They were legible on the original site, but I had to scale it down to fit the layout here, which makes them much harder to read.

The individual points are, in fact, quarterly values.

Economic arguments about tuition cost only make sense if education is merely a good, i.e. a business service. If people at least in part are paying for something more than a service (and I would suspect in part that most people put at least some value in education beyond merely economic advancement) than relying on market forces to control the price will not work.

The cumulative student loan debt curve would likely be a running time average of new debt being issued.