Tag Archives: education

The New Economics of Online Education

Leave a reply

Let’s stipulate up front that there is an old model of online for-profit education that has completely failed. The old model is this: take everything that used to be done in person, put it online, charge for it. There are some programs along these lines that are still kicking, and some traditional universities have online programs, but it seems clear now that the future of online education does not lie here.

A large problem with this model is incentive-compatibility. The incentives of the online provider are to enroll as many students as possible, especially since students are often subsidized, and then to let standards slip so that the students pass. If most students fail, they will be angry, and the commitment to keeping standards high is not credible. High standards are not a subgame perfect equilibrium.

The reason most universities are not for profit is that the profit motive strengthens the incentive to let standards slip, although it should be noted that standard-slippage has happened, albeit more slowly, even at non-profit universities. The great success stories in for-profit education are those that generate easily verifiable skills. For instance, if you want to learn German, you are probably better off going to Berlitz than to Harvard. When you finish at Berlitz, you know whether you got your money’s worth because you know whether you know German; this solves much of the incentive-compatibility problem. If Berlitz taught English literature (to English speakers), I don’t think it would be as successful as Harvard.

Enter Udacity. Unlike many other online education providers, Udacity has not tried to simply replicate the old in-person education experience online. Rather, they are adopting an entirely new business model. First, they have a radically different cost structure than traditional providers. Udacity does its grading by AI, so the marginal cost of another student is basically zero. It has a much higher fixed cost per course, because it needs to train the AI to grade homework and exams.

Since Udacity’s marginal cost of a student is zero, it can give away the courses for free if it can find another way to cover its fixed cost. I perked up when I read this:

All classes are currently free, and the goal is to keep it that way. When asked how it will make money, [Udacity co-founder] Sebastian [Thrun] pointed out that recruiting good technical talent is something that companies pay for. Udacity knows who the best students are and could pass them along to companies looking for new hires.

In Silicon Valley, finding talent is so difficult that recruitment bounties are often as high as 20 percent of the new hire’s first-year salary. This provides an opportunity for a new revenue structure for online education: vertical integration into job placement.

Udacity can sell its database of high-performing students to recruiters from tech firms. They can survey these students to see if they are interested in a job before they do so to ensure that the database is high quality in the interest dimension as well. This database is a very valuable asset, one that seems likely to cover Udacity’s fixed costs.

I think the Udacity model can succeed in subject areas where:

  1. Skills command a large premium in the labor market.
  2. Employers can verify skills in the medium run, so that they can decide whether to continue to pay for Udacity-educated recruits.
  3. It is possible to evaluate coursework using artificial intelligence (we’re not there for writing yet).

Udacity has started out with classes on computer programming, because that is what its founders know, but they seem eager to expand to new subject areas. I wonder if economics will be one of them.

Socially, there are some advantages to the Udacity model versus the traditional university system. Perhaps most obviously, we can avoid a lot of the student debt crisis. Since classes are free, there is no tuition expense, and since revenue is constrained by the labor market, the skills taught will need to correspond to those that are valuable. We can also save a lot of money on government subsidies for higher education; indeed, cutting off these funds now would hasten the adoption of new education business models.

Speaking generally, new contexts allow for new business models. People are often slow to recognize these new opportunities. We need to make sure as a society that we are not locking in old business models and preventing change. This means giving up control and being willing to be surprised by the outcome.

A Picture for Krugman

6 Replies

“The lights are going out all over America—literally.” So begins Paul Krugman’s latest NY Times column, in which he laments the crumbling of America’s foundations. The proximate cause of this decline, according to Krugman, is insufficient government spending on infrastructure and education. But the root cause is rhetoric:

How did we get to this point? It’s the logical consequence of three decades of antigovernment rhetoric, rhetoric that has convinced many voters that a dollar collected in taxes is always a dollar wasted, that the public sector can’t do anything right.

I could easily write a thousand-word post refuting this claim, but that would just be more antigovernment rhetoric. Instead, I want to show you a picture. The data comes from the BEA’s NIPA Tables (3.1, 1.1.4, and 7.1).

If antigovernment rhetoric over the last three decades were as effective as Krugman claims, what would this picture look like?

Should Schools be Run like Law Firms?

Leave a reply

There’s a lot of money in education. According to an article a friend posted on Twitter yesterday, Philadelphia spends $400,000 per classroom of 25 students per year. Other cities are similarly lavish.

In spite of this spending, urban public schools are usually terrible. There is no simple solution to this problem (read Bill Easterly — 1, 2 — and insert “learning” and “education” for “growth” and “development”), but reflecting on the exorbitant amounts of money on the table helped me focus on the incentive problems encountered by the education industry. I realized that similar problems have been solved by law firms.

Imagine that there are two kinds of performance measures. One loosely correlates with what it is that we want from our agents and is publicly verifiable. Because this performance measure is publicly verifiable, we can use it as a basis for writing contracts or regulations or whatever. The second performance measure correlates more strongly with what it is that we want, but is verifiable only by people with specialized skill, such as fellow-agents.

This describes the situation in both education and law. In both fields, there are simple metrics that are publicly verifiable that loosely capture what it is we want from workers. In education, examples are test scores and graduation rates; in law, it’s billable hours. Nevertheless, the attempt to rely on these metrics alone to motivate production are doomed to failure: they do not correlate closely enough with what we actually want, quality education or legal services. The incentives are to game the system, to teach to the test, to bill too many hours, etc.

Fellow-agents are capable of evaluating each other on another basis, one which by assumption correlates more strongly with what we want from them. Left to their own devices, they will shy away from doing so, because it is unpleasant to evaluate others, potentially hurting feelings and rocking the boat. This is the problem that law firms have solved. They have done so by organizing as partnerships. Because a partner’s pay is dependent on the quality of the work done by the other partners and employees, there is an incentive to engage in mutual monitoring to ensure that one’s co-workers are in fact doing quality work.

Would mutual monitoring work in schools? I think so. Teachers know who the other good teachers are. Because schools are typically run in a top-down fashion and answer to external authorities, there is an incentive to keep this information private. But if schools were organized on the law firm model and teachers could keep the surpluses generated by quality improvements, I predict that there would be a lot of quality improvements. A prerequisite for this approach is that quality improvements generate surpluses; schools organized on this model would need to be private.

The partnership approach to school organization would solve some other problems as well. First, much of the money spent on education today is captured by administrators. This is true in both private and public schools. The partnership form makes agents sensitive to the amount wasted on administration. This increases the resources available to actually educate people. Second, mutual monitoring will improve teacher quality in three ways: by discarding bad teachers, by motivating existing teachers to excellence, and by drawing new talent into a higher-paying profession. Higher quality teachers, with higher marginal products, will earn more pay in market equilibrium. To the extent that complaints about “low” teacher pay are valid, adopting the partnership form should address them.

When discussing incentive problems, economists often talk about the mistake of rewarding what is measurable at the expense of what you actually want, which is often not measurable. Incentives, that is, can be too strong. Test scores and graduation rates are measurable, and the results of attempts to enact measurement-based accountability in education have been disappointing. The insight of the literature on partnerships is that you can reward on the basis of quality dimensions that are not precisely measurable. Organizing schools as partnerships might have a significant effect on the quality of education they produce.

An Answer for Grade Inflation

16 Replies

Grade inflation is a problem for several reasons. First, as GPAs increasingly get compressed into the 3s, the amount of information they contain within a given number of decimal points decreases. Second, the problem is exacerbated by the tendency for different academic departments to exhibit different rates of grade inflation. Third, negative externalities are clearly not being internalized. When the grades of lower-caliber students get inflated, it makes them appear similar to higher-caliber students. And since high grades are popular, professors who give high grades have fewer headaches relative to professors who are trying to stem the grade inflation tide. I have a simple solution for grade inflation. It’s a little radical, but I offer it in complete seriousness.

Grades and GPAs are a method of ranking performance. They are not a particularly good method of ranking, because giving one person a higher grade (or rank) does not immediately entail lowering the grade (or rank) of another person. This is the source of the externality problem. Fortunately, a better ranking system is already in use by another dimension of college life, NCAA basketball.

The system is called the Ratings Percentage Index, or RPI. You can find the latest RPI rankings here. The system works as follows. Every day, each team’s winning percentage (WP) is calculated. Then, each team’s opponents’ winning percentage (OWP) is averaged. Then each team’s opponents’ opponents’ winning percentage (OOWP) is averaged. The RPI is then a simple weighted sum: RPI = (WP × 0.25) + (OWP × 0.50) + (OOWP × 0.25).

How can we adapt a system based on wins and losses to a classroom setting, where “games” are exams, papers, and so on? The answer is surprisingly simple. At the end of every semester, instead of submitting grades to the registrar, the professor will submit an ordinal ranking of students. For each pair of students, being ranked more highly counts as a win, being ranked lower counts as a loss. Suppose there is a class with 25 students. In effect, every student is playing 24 pairwise games. The top student in the class is 24-0, the second student is 23-1, and so on. The worst student is 0-24. If a student takes three such classes and ranks first, third, and fifth for the semester, her record will be 66-6. Her winning percentage will be 91.7.

This already goes a long way toward curbing grade inflation. Professors won’t feel pressured to give high grades because they can’t give universally high grades. They can rank students. Ranking one student more highly will cause another student to rank lower. But the real benefit of the system comes from the other components of the RPI.

Under the current system, a student can earn a high GPA by taking only very easy classes. With an RPI system, classes won’t be easy in the sense of offering high average grades, but they may still attract inferior students. For instance, the reality of college education is that engineering classes attract more talented students than education classes. If you wanted to boost your winning percentage, you could do so by taking a lot of education classes. However, the other components of the RPI account for this. If you take only classes with less talented classmates, your OWP and OOWP will be very low, as long as there is not complete separation of students between programs (we should expect this—if a school is running two distinct programs, then it naturally ought to have two separate ranking systems).

One way to ensure that there is not complete separation is to require an extensive general education curriculum. But even if this is not present, the system incentivizes integration between programs. If a student in an easy major wishes to improve her rank, she will have to take difficult electives and work very hard to beat out the students who are more talented than those with whom she is used to competing. Contrast this to the GPA system. If a student in an easy major wishes to improve her GPA, her incentive is to take easy electives, probably in her major area.

Another advantage to the RPI system is that it solves the first mover problem. If a school wishes to combat grade inflation within the current GPA system, it will need either to do it in coordination with other schools or to subject its students to a disadvantage as the seriousness of its fight against grade inflation only gradually becomes apparent (in this latter respect, it is similar to the problem faced by the central bankers of countries that suffer from monetary inflation). But switching to the RPI system is immediately credible. The old scale is not comparable, and most of the externalities are internalized. A school that switched to the RPI system would only disadvantage its students insofar as they would have to explain the system to people who had never heard of it.

Obviously, any system can be gamed. This one is no exception. Nevertheless, I think it is a clear improvement over the GPA system. I hope that in spite of its radicalness, it will garner some consideration from the people who control such things.

Bribes for Privatization

4 Replies

The privatization of education is near the top of my policy wishlist. While I’d prefer radical privatization, I think that’s essentially impossible any time soon; but vouchers seem to be gaining support. Here is my proposal for putting them over the top.

We bribe people.

My plan is to bribe the educational establishment (teachers and administrators) in failing school districts with the assets of the failing schools. The local government could implement a voucher system, and under cover of some made-up reason like retaining the spirit of the public schools, give the building, land, and other assets of the local schools to the teachers and administrators. Equity could be determined by whatever method the establishment thought was fair, such as seniority. The new equity holders could decide whether to organize as a for-profit school, a non-profit school, or to divest the assets by selling them to entrepreneurs who wanted to start a new school or use the assets for other purposes.

I am convinced that this is good public policy. Consider the following two premises:

P1. Many schools, especially in the inner cities, produce little or nothing of value.

P2. The educational establishment is able to block any real efforts at reform.

If you buy these two premises, then the bribe is actually costless to the local government. Think about the property rights in the school assets as divisible into rights of control and residual claimancy. If you accept P2, then you concede that the establishment already has rights of control. Nothing can be done with the assets without the consent of the establishment. If you accept P1, then you must acknowledge that the right of residual claimancy is worthless to the local government. Giving residual claimancy to the establishment does not adversely affect the local government or its citizens because it wasn’t doing them any good in the first place. If in exchange for nothing the government gets real educational reform, then it is good public policy, no?

One may object that this is rewarding failure, which it is. But the status quo rewards failure on an ongoing basis. At least under my proposal, there is an end to the rewarding of failure.

Or is there? This is the real danger. If, after divesting the bribe, the establishment is able to resurrect demand for a public school system, then the local government will be in worse shape than where it started. It will continue to have schools that fail and it will have to pony up more money to reestablish the public schools. That would be a disaster, but nevertheless, I think the upside on the plan is good enough that someone should give it a try.