We need low-cost colleges

In an opinion piece for the NY Times, My Valuable, Cheap College Degree, Arthur C. Brooks relates an anecdote of his own education: failing out of college his first year, taking a gap decade, going back to school through low-cost correspondence courses from various state universities,  eventually getting a PhD, and becoming a professor and president of a “Washington research organization”.  (I’d be more impressed if I thought he meant a biotech or computer research job in Seattle, but I fear he means a political hack job in DC.)

His anecdote sounds like a classic “self-made man” story, but is actually about the substantial state subsidies for college education that were available 40 years ago, and that are fast disappearing under the current political belief that government should pay for nothing but guns and people to use them (the state budget in California is heavily weighted toward prison guards and police).

I’m in basic agreement that college has gotten ridiculously expensive—I would welcome a low-cost college education for my son like the one I got back in the 1970s.  Unfortunately, I don’t expect to see state governments coming to their senses in time, and I may have to send my son to a private college to get an adequate education.  (Currently his short list of about 20 schools is almost equally divided between public and private colleges and universities, but most of the public universities are out of state, raising their costs to the same level as the private ones.)

I expect that my son’s education is likely to cost me 2 or 3 years of my salary as a full professor (and that’s pre-tax salary). I’ve been saving 10% of my salary each year for the past 16 years in order to be able to pay for that education for my son, so we should be able to do it without his going into debt, which is good, since I’ll probably be retiring while he’s still in grad school.

The fact that the ratio of college tuition and fees to professorial salary is so high tells me that the cost is not due to high salaries for faculty (and I’m one of the lucky tenured faculty—most colleges are dominantly hiring “contingent” faculty now, who have no job security, low pay, and no voice in the running of the institution).  Other studies have show that the cost of public universities has not changed much—it’s just been shifted from the state to the student. In particular, the socially desirable subsidy of tuition for low-income students is almost entirely born by “return to aid” funding from other students, not from endowments or state subsidies.

Private colleges have had enormous increases in costs: largely related to the debt acquired in the amenities wars and the obscene salaries they pay their administrators and football coaches (though not their professors, who are only slightly better paid than public-university professors like me).

I don’t see online education as being a major cost saver in providing decent quality education.  Massive lecture classes are the least effective courses at the university (though relatively cheap to offer) and MOOCs just amplify the problem.  Those completing MOOCs are mostly the older, highly motivated students who have always been fairly easy to teach—there’s not much evidence that MOOCs are at all effective at educating the run-of-the-mill students who are the current main target of college education efforts.

Is coding for everyone?

One of the memes in computer education last year was “coding for everyone”.  For example, the year started with Mayor Bloomberg vowing to learn to code using Code Academy (how did that go for you, Mayor?), and there was Rip Empson’s post STEM Ed: CodeHS Wants To Teach Every American High Schooler How To Code, in which he reviewed the CodeHS online program to teach high schoolers how to program, and a host of other posts that I did not bother to bookmark all claiming the wonders of universal literacy in programming languages.

In May, there was a backlash: Jeff Atwood’s Coding Horror: Please don’t learn to code made the claim

The “everyone should learn to code” movement isn’t just wrong because it falsely equates coding with essential life skills like reading, writing, and math. I wish. It is wrong in so many other ways.

Atwood’s main points are that coding is not the goal, but just a tool in solving some types of problems, and that thinking about problems is more important than concentrating on this particular tool.  He’d rather have people like Bloomberg concentrate on doing the things they are good at rather than becoming half-assed programmers, and leave the programming to people who have the time and the inclination to do the job right.

I think that Atwood position, like Andrew Hacker’s claim that high algebra is pointless, is a deliberately extreme position taken to raise passionate responses and sell newspapers.  Hacker claims to want to lower the already low value of a high-school diploma by reducing the math required for it below the 8th-grade level currently required.  Atwood wants only professional programmers to learn to program.  Both appear to be protecting elite privileges by claiming that education is too much for the masses, and that they don’t really need it anyway.

Atwood makes an analogy to everyone learning to do plumbing, as if that were a ridiculous idea.  Personally, I think that everyone should learn a little plumbing: enough to clear a toilet or sink trap, or replace a faucet washer, and to know when to call in a professional plumber. So his analogy is an excellent one and completely undermines his conclusion.

Steve Myers, in Software developer revives debate about whether journalists should learn to code, makes the analogy to the “reporters vs. bloggers” war.  Unfortunately, that fight appears to have been settled by laying off most professional journalists and killing off a lot of newspapers, with the resultant degradation in the quality (though not the quantity) of news that people get.  As a blogger, I don’t think that discouraging blogging is the right way to support journalists, but if the analogy is a good one, then I can see why people like Atwood are attacking amateur programmers in fear for their jobs.

Myers quotes Andy in Don’t not learn to code

Even if someone who learned how to “code” never ends up as a “coder” professionally, they will be much better for having learned to think algorithmically and in the abstract. However, I bet that knowing how to write a quick script here and there will make them much more valuable in whatever profession they choose.

Andy makes the standard case for learning to program:

Do I always use software to solve problems? Definitely not. But do I often use the same problem solving techniques that learning to code taught me in order to solve other problems in my life and work situations? Definitely yes.

Can folks be taught to solve problems without using computer code? Definitely yes. But is learning how to “teach” a computer how to solve problems via code one of the best ways? I think yes as well.

In Definitions of “Code” and “Programmer”: Response to “Please Don’t Learn to Code”, Mark Guzdial points out

Most people who write code are not trying to create code solutions.  Most people who write code are trying to find solutions or create non-code solutions.  By “most people,” I do mean quantitatively and I do mean all people, not just professional programmers.  We know that there are many more people who write code to accomplish some task, as compared to professional programmers.

The point of the coding-for-all movement is that there are lots of times when a small amount of coding or scripting on a computer can solve an immediate problem without calling in a professional programmer.  Little snippets of throwaway code can be enormously useful for getting data into a form suitable for thinking about a problem, and the ability to create those snippets quickly is a very valuable skill.  I know I would feel very constrained if I had to wait for someone else to understand what I was trying to do and implement it for me every time I wanted to plot some data or model something.  I try to get the biologists and bioengineers that I teach to have some coding ability—enough to fit models to data, reformat data so that it is usable by other programs, and do simple analyses.  I don’t expect most of them to become professional coders (not even all the bioinformaticians need to be expert programmers, though most should be), but they all need enough skills to do the little tasks (the equivalent of clearing sink traps in plumbing) and to know when they need to call in the professionals.

Mark Guzdial takes particular exception to Atwood’s claim “You should be learning to write as little code as possible. Ideally none.”

And people who want to do interesting, novel things with computers should just wait until a software developer gets around to understanding what they want and coding it for them?  I could not disagree more.  That’s like saying that the problem with translating the Bible is that it made all that knowledge accessible to lay people, when they should have just waited for the Church to explain it to them.  “Please don’t learn to code” can be interpreted as “Please leave the power of computing to us, and we’ll let you know when we’ll make some available to you.”

As may be clear by now, I tend to side with those who would teach programming skills (and math) to everyone, even though not everyone will end up particularly good at programming or math.  I believe that I have enough skills and intelligence that I’m not threatened by an increase in the skills of others, so I don;t see any benefit in choking off education for others.

On the other hand, I don’t think that the coding-for-all movement will be any more successful than the math-for-all movement has been for the past century.  Math is taught to everyone and people are required to get through algebra and some geometry to get a high school diploma, but huge portions of the American adult population are still functionally innumerate. Even after ten years of learning math, many adults can’t add fractions or figure out that lottery tickets are a sucker bet.

The coding-for-all movement is starting from a much lower base—almost no elementary schools and very few high schools teach programming, and those that do generally have only one course in Java syntax, which will not get people very far in learning to solve problems.  So even if the coding-for-all movement is wildly successful and creates tens of thousands of teachers (coming from where, exactly?) teaching high-schoolers how to program, the fraction of people who are even marginally competent programmers will remain pretty small.  The online courses like CodeAcademy and CodeHS try to get around the shortage of teachers by having a computer do the work of teaching, but they are pretty miserable failures at teaching to even the low level of high school programming courses.  It’s nice that there are resources that will allow dedicated students to get around the curricular limitations of their high schools (current or past), but they are not going to lead to a world of universal coding competence.

Online education for continuing education

Tom Katsouleas, Dean of Duke University’s Pratt School of Engineering, provided a very upbeat view of online courses (particularly MOOCS) in a blog post Who says online courses will cause the death of universities?.  He makes some assertions without evidence that I find a bit dubious:

The bottom line, though, is that while online education poses a challenge for universities, they will ultimately improve them.

…

And by moving lecturing online, MOOCs allow in-person time to be more interactive, dynamic, and valuable.

…

Despite the challenges to universities posed by MOOCs, there are great advantages to them as well. And the best universities will be able to capitalize on those advantages to provide the best value for their students – whether that value is online or in person. Despite what some see as a threat to higher education, MOOCs will only help it get better.

This is essential the administrative view of online courses: that they are new and cheap and must therefore be for the good of higher education.

I think that Dr. Katsouleas is right in asserting that the continuing education market for engineers is one of the places where universities will first see the effect of MOOCS, both because there are more MOOCs and more advanced MOOCs available in engineering than in other fields, and because many engineers seek continuing education even without certification.  I don’t think that the MOOCs will reduce MS degree seekers much, because there is still a very large premium for the MS certification in engineering, and the engineering MOOCs available still only go up to junior-level courses.  There are already many MS-level courses offered online, and I’m sure we’ll see an increase in them, but the market for them is too small for a MOOC, so there isn’t the advertising-for-the-provider benefit which drives most of the MOOCs currently, and we’re more likely to see the standard pay-per-course fee structure for them.

I think that we will see some colleges and universities starting to accept credits from MOOCs for undergrad courses, once the problems with cheating and low-quality assessment are fixed.

Cost of online education

In Changing Universities: Online Instruction, Budget Transparency, and the Cost of Education, Bob Samuels makes a crude estimate of the cost of online courses versus mega-lecture courses (the two approaches to education popular with university administrators, because of their relatively low cost).

I believe his numbers serious underestimate the costs of both, as he only counts the costs of lecturers, and not that of TAs.  It is true that lecturers are only paid about $7500 a course, but a course of 200 students is likely to have 4 TAs, at a cost of $11000 each (more fairly, $5800 each, since the rest goes back to the University as tuition).  So the direct instructional cost is more like $31k, or $155 per student, not counting benefits for the lecturer.  He also neglects a lot of the overhead costs, like amortizing the cost of classrooms and the costs of cleaning and maintaining buildings, roads, libraries, and other essential infrastructure.    He underestimates the cost of course development for traditional courses, though he includes it for online courses (which are, admittedly, far more expensive to develop).

I think he’s right that the University is not going to save significant amounts of money by offering big courses online, and that going online may actually raise costs, but his analysis is not careful enough to establish that.  The UC Online pilot project has spent a lot of money developing very few courses. The edX consortium has also dedicated huge amounts of money to deliver a very small number of classes—confirming my view that edX is primarily and advertising vehicle for MIT and Harvard, rather than an education-delivery vehicle.

The cost of developing traditional courses can be fairly high. I’ve spent about 400–500 hours developing a new lab course, even before preparing any of the lab handouts or ordering the parts for the students, which will take another 100 hours (more than full time over winter break). At my current salary and benefits, that is a development cost of about $30k. Delivering the course to 20 students will cost another $30k–40k (counting my time, ⅓ of a course for a lecturer, and an undergraduate group tutor), for a cost of around $3500/student. Of course, in future years, the class size will double, the development costs will have been amortized, and we may even alternate years between Senate faculty and a lecturer, reducing the cost to about $1000/student, as he estimated for somewhat smaller classes.

Incidentally, if you are interested in what goes into designing an engineering lab course, I’ve been making notes on the course design on this blog.   I’m up to 86 blog posts (about 200–300 pages) of notes for the course design—and I still haven’t started writing up the web pages or course handouts.

I’m still strongly of the belief that it is in courses like this lab course that the University is strongest, and that attempts to reduce the cost of education with mega-lectures and online courses reduce the quality much faster than they reduce the cost.

Faculty discussion of online courses at UCSC

This afternoon I attended a Faculty Senate panel discussion on the future of on-line courses at UCSC.  A couple of the panelists had already taught on-line courses, and their presentations were particularly interesting.

One had taught a hybrid course where half the students attended live lectures and the other half watched videos of the lectures.  Both halves had required weekly hands-on discussion sections, so the course wouldn’t scale to MOOC sizes.  The bottom line was that there was no significant difference in performance between the on-line and live-class halves of the class, and that students spent a lot less time looking at the videos than predicted.  (The class has been offered 4 times to about 300 students each time, so this was not a small sample.)

The other professor is currently teaching a tiny boutique class (14 students, I think he said), using software that lets him lecture from his office, with a whiteboard window, a little web-cam video feed, and a chat window.  I’ve used similar software in conversations with the Global Physics Department (whose meeting tonight I missed, because of the panel discussion, but they were just discussing the College Board’s plan to split AP Physics B into two courses, which I’m not all that interested in).  When I gave a presentation to the GPD, I found it very difficult to present material on the whiteboard, talk, and watch the chat box all at the same time.  I asked the professor about this problem at the reception afterwards, and he said that with a small, quiet class, he can usually keep up, but if everyone chats at once, stuff scrolls off screen before he can read it.  He thinks that the technology might scale up to 60 students with a very non-interactive lecture style and sleeping students (I exaggerate his description), but not beyond that.

Another professor presented a course that is going to be offered soon that takes the form of a self-paced e-book (on calculus).  He showed a couple of features of the e-book, and I think that it has many of the bells and whistles that math bloggers have expressed an interest in seeing in math e-books.  Personally, I did not find the examples he showed very appealing, but I’m not part of the target audience. (He also loves math history, which I have always found to be a tedious addition to math books, so I’m really not part of the target audience.)

Some of the panelists just raised questions for us to think about, though they went by so fast that I don’t think anyone in the audience will remember more than one or two of them—the questions they were thinking about before coming to the meeting.  I hope that the Committee on Teaching or the Committee on Educational Policy will send out the list of questions as e-mail.

One thing that disturbed me about this meeting was the average age of the attendees. I think I was well below the median age there, and I’m turning 58 this week.  If we are talking about the future of online education at the university, then we absolutely need to be talking with the people who will be the faculty in that future.  It can’t be only us old farts who will retire in the next decade (and the professors emeriti, who have already retired)—where were the assistant and associate professors?  I’d be very surprised if there were more than 4 assistant or 6 associate professors there.

My personal feeling is that UCSC should not invest large amounts of money in online education.  It does not seem to be much cheaper than conventional teaching methods, and UC does not have a good track record for providing infrastructure cheaply, nor for running businesses.  I think that UCSC should be concentrating its shrinking resources on the things where there is enormous value added by being a UC: on lab courses and small seminar courses where students get direct hands-on experience and interact with faculty.  If this means outsourcing the teaching of the 1000 students a year taking precalculus,  well, that’s too bad, but high schools and community colleges can teach those courses ok.  I don’t believe that UC should be teaching precalc—certainly not to a quarter of each incoming cohort!

Unfortunately, the budgetary pressure in recent years has been towards eliminating small grad courses and expensive-to-teach lab courses, and creating more and more mega-lecture courses.  These mega-lecture courses are relatively easy to replace with MOOCs, since the teaching in mega-lectures has already been degraded almost to the level of video lectures, with no interaction for most students. Once you start moving to a factory model of education, it starts becoming “obvious” to outsource the production to cheaper labor elsewhere, or to look for “economies of scale” that allow you to mass-produce a course.  I’m not convinced that there are economies of scale in education—I don’t think that it is really more cost-effective to teach 1000 students at once than 20 students at once.  You can make the course cheaper per student, but the cost in quality is pretty high.

The calculus e-book looks like a promising alternative to big lecture courses, though I suspect that not that many students will slog through it without someone holding their hands and cheering them on.  Even my son, who is very interested in math and quite good at it, finds it much easier to learn in the context of a class with regular meetings and feedback from the teachers than in a self-paced course with the same content—lack of time-management skills ruins self-paced courses for most students.  Of course, there is no reason that e-book has to be used in a self-paced course, but adding math coaches or teaching assistants to the course raises the cost of offering it to nearly the levels of a conventional  course. Furthermore, the time, money,  and effort involved in creating such an e-book means it is unlikely that UCSC will create many such resources.

The chair of the Committee on Educational Policy suggested that there would be a market for on-line courses in bioinformatics from UCSC, since UCSC is an acknowledged world leader in bioinformatics.  And it is true that there might be a market, but as the teacher for our core graduate bioinformatics course, I don’t think that our quality of education would survive a transfer to on-line format.

My “lectures” are very interactive—I try to get students to derive things like the Smith-Waterman algorithm and the forward-backward algorithm for HMMs from reasoning about how to break problems into sub-problems for dynamic programing.  I could present the algorithms in a textbook-like way in a quarter the time, eliminating the long waits for students to digest and idea and suggest a next step, eliminating the cold calls, eliminating the checks for understanding at every key point, … .  I can teach a group of 20 students in the same room with me, but I’d lose most of the useful feedback in an online setting.  I’d also lose the chats with students between classes—e-mail and forums do not bring up the same issues that come up when I stop by the grad office to get more hot water for my tea. Even recording my extemporaneous presentations would flatten them—I’m likely to be just enough nervous about making mistakes on camera that I’d play it safer, doing pre-canned examples, rather than riskier live-action math and algorithms that show how I think about problems, rather than just showing “the solution”.

Just Monday, when I was presenting a numeric example of computing HMM probabilities, I made a serious mistake that amounted to multiplying by two transition probabilities instead of just one in the first step.  It was caught by one of the students, and I could correct it and go on.  Today, after we together derived the more general recurrence relation for the forward algorithm, one student suggested an optimization that wouldn’t quite work, and I could point out that it was exactly the same as the mistake I had made near the end of Monday’s lecture.  With an online course, either the mistake wouldn’t have happened in the first place (if I polished my examples before presenting them, following a script rather than extemporizing), or the students would not have had the involvement to correct me or to propose optimizations that didn’t quite work.  Having a small class that has been encouraged to present ideas, to challenge me when I may be making a mistake, and to ask questions when they don’t understand is crucial to my teaching style, and having a record of the class is likely to ruin that.

I sometimes deliberately make mistakes and hope for the students to catch them—if they don’t, I have to spend more time stepping them through the pitfall, so that they can see it and avoid making the same mistake themselves.  At the beginning of the quarter, the students were pretty shy about saying anything, but I now have over half the class participating on a regular basis,  and even the weaker students are willing to ask about potential errors, though they ask more timidly than the stronger students, since there is a bigger chance that they are misunderstanding something, rather than pointing out my error.  Encouraging the students to correct my mistakes does get me more feedback about misunderstandings, when their attempts to correct something that is actually already correct highlights where they did not quite grasp a concept.

Even if we could somehow magically provide online all the visual cues and social interaction of the face-to-face classroom, I don’t think that we could scale up other aspects of the course: I’m already spending almost all my weekends providing detailed feedback on programs and papers for a class of around 16 students.  If we scaled the class up by even a factor of 2, we’d lose that detailed feedback, which I see as an essential part of the homework.  For many of the seniors and grad students, my reading of their programs and papers is the first time any professor has read any of their work closely—and they desperately need to hear how to fix their in-program documentation or how to reorganize their sentences to avoid flow problems.

Incidentally, in my other class (which includes many of the same first-year grad students), the students just finished doing 10-minute presentations on techniques from Teach Like a Champion.  Tomorrow, before we start reviewing the video recordings of their presentations, I think I’ll have them try to think about which of the techniques they presented that they have seen me use in the bioinformatics core course.  This year they presented Circulate, Ratio, Cold Call, the Hook, Pepper, Warm/Strict, Wait Time, No Warnings, Check for Understanding, Stretch It, Positive Framing, No Opt Out, Board=Paper, Call and Response, and Begin at the End.  I think that they’ll find that I use about half of those on a regular basis. (I leave it to my readers to guess which of these I don’t use much—those who had me as an instructor a decade or more ago might make different guesses than those who’ve had me recently, as I’ve gotten better about some things.) Note that most of the teaching techniques in Teach Like a Champion are difficult to apply in an online course.

I’m not planning to teach any on-line courses in the near future, and I’ll be putting my efforts into creating more of the interactive, lab-style courses that are difficult to replicate on-line (like the Applied Circuits course I’ve been designing for the past 5 months).  I think that the future of the university is in these high-interaction-level courses—artisanal education, not mass-produced factory education.  There will undoubtedly be a huge market for the Wal-marts of education, but that’s not where I want to work, nor where I want my son to be a student.