Friday, December 6, 2013

What Do You Do When the Students Get Better?

The Crimson reporter who interviewed me for today's followup story on rising grades asked me if students had improved over the years I have been teaching at Harvard. By some interpretation of "better" the answer is surely yes. It's a question about a statistical distribution, but the system is so complicated that it is hard to untangle the various effects.

For example, as I mentioned earlier, Harvard has won the Putnam mathematical competition in three out of the past five years. So our student population way over represents the very pinnacle of mathematical gifts in the 18-22 year old world population. That doesn't mean that the left tail of the distribution has gotten shorter, or even that the mean has shifted. Nor does it say anything about who is turning up in any given class. In my own case, Computer Science courses seem to be drawing both more of the extremely gifted mathematicians and more of the very average (that is why I started teaching CS 20). But that is today. Ten years ago it was different and ten years from now it will be different again.

So giving everybody the grades they deserve has unpredictable statistical results. Most of responses to changes in the student population have nothing to do with grading, which just has to take care of itself. The better responses are educational. Over the years I have been teaching here (this is #40), I have turned several courses that used to be graduate level into undergraduate courses, because too many undergraduates were taking the graduate level courses and doing extremely well. My signature course, CS 121, is an undergraduate version of a graduate level course I myself took as an undergraduate. I first offered it (under the name Applied Mathematics 108) back in 1978 (Paul Spirakis and Oded Shmueli were my first TFs). CS 124 was an undergraduate version of a graduate course on algorithms. That started off as Applied Math 119 in the spring of 1982. Bill Gasarch was my first TF.

Michael Mitzenmacher, my faculty colleague who commented on my grading post below, unwittingly I'm sure posted himself on this topic yesterday on his own blog. There are too many students in CS 124. There are too many students in CS 121. And too many of them are too damned smart. Time to create a new course.

Go read Michael's blog. You will see there, live, the process of academic evolution in response to changes in the student environment in which we professors live.

Matt Yglesias has it right. This isn't a grading problem. It's an admissions problem!


  1. Replies
    1. I suppose you could look at it that way, but everything would have to grow in proportion: faculty, housing, dining halls, classrooms. (To be sure I would rather see the student body grow than some other parts of Harvard that have been growing!) But communities don't scale very well. In a residential college, distance matters. The possibility of knowing many of the people you live with matters. So while the pool certainly supports growth, and some growth in the student body would probably be feasible if done right, a bigger goose would not necessarily lay more golden eggs.

  2. High school grades and SAT/ACT scores jointly predict college grades, and most Harvard students had great test scores and high school grades. There can still be grade inflation at Harvard if, controlling for high school achievement, they earn higher grades at Harvard than they would have at other colleges. Harvard could ask the College Board to do such a study if it wanted to know the truth -- the College Board periodically studies the prediction of college grades from SAT scores to validate the SAT -- but it easier to just pat itself on the back for having such smart students.

    Why not list the average grade given in each course on the transcript so that transcript readers could assess where someone stood compared to his or her classmates?

    1. I see that Mark Bauerlein has made the same suggestion in an essay "A Solution to Galloping Grade Inflation" prompted by Harvard:

      "Let's add another grade to the transcript besides the individual grade. For each course listed, show the student's grade and also show the average grade in the course. it would give employers looking over a student's record a better picture of ability. A B+ in a course with an average grade of B looks a lot better than an A- in a course with an average grade of A."

    2. But it wouldn't do what Bauerlein claims, since there are some courses (Math 55 at Harvard would be an example) in which all the students are insanely talented. The gut that gives all As and the super-honors Math course that gave all As would still look the same on paper.

      There are actually statistically valid solutions to the problem of annotating transcripts or curving grades across the board. (See Valen's book from 2003, for example.) Unfortunately they are very difficult to explain, defeating the other objective, which is simple and effective communication.

  3. This comment has been removed by a blog administrator.