Monday, March 23, 2009
Grade Inflation at My Old Stomping Ground
Part I: No It Isn't Because of Better Students
There was a three part article on grade inflation in Duke's student newspaper last week. You won't find a better portrait of the problem of grade inflation on an individual college campus.
It's always a strange thing to read an article when you know a lot about the place and the subject being discussed and especially strange to read an article where you're quoted. Lots of times when that happens, I think the reporter just hasn't portrayed what's really happening very well. That's not the case this time. It's a dead on accurate portrait of the nature of grade inflation at my old stomping ground.
But it left many things out. There are only so many words and so much time. I've decided that this very well done article needs a supplement. This supplement might be two parts long. Here goes.
Duke has garden variety grade inflation for a highly selective institution. Grades are going up at a rate of about 0.1 per decade and the average grade is about 3.45 today. Why are grades rising? For a long time, Duke leadership was in complete denial that grades were rising because Duke faculty were grading easier. The idea was that it was all due to students getting smarter and better. Then they decided that maybe it wasn't due all to better students, but declined to say what else was driving rising grades.
I want to look at this assertion of Duke leadership. What component of rising grades is due to students getting smarter and better?
First off, why is grade inflation important at all? Why care? A grade is just a number. Who cares if the average grade is 3.0 or 3.2 or 3.4 or 3.6? Here's one reason. Grades are so high at Duke that classes often have no life. In many classes, students have little motivation to work hard because they know that just by going through the motions they can get a B+ or better.
I know this from personal experience. The energy in most of my classes couldn't have lit up a 2 watt bulb. The funny thing is that in my student evaluations, students often remarked how energetic my classes were relative to others. I was always amazed by those comments. There must have been corpses instead of students in those other classes. Students simply did not prepare. They did not work. Not all students. But most. The average number of hours Duke students study is 11 per week. That's it. Eleven measly hours a week.
In commenting on this lack of studying and especially on the fact that students at Duke study less in college than they did in high school, the authors of Duke's Campus Life and Learning Project of 2006 state: "In the transition to college, either students are becoming more efficient, or are working less hard at studies (having more demands on their time), or the difficulty of college work is less than high school, which is difficult to believe." It's not difficult to believe that college is easier than high school, actually. If you talk to students, that's exactly what many of them will tell you.
I once confronted Duke's past provost about this problem. We had a friendship for awhile; he said I reminded him of what he was like when he was young. His response to my concerns about the wan level of education we were providing was, "You're right but we're not doing anything that anyone else isn't." He came from Dartmouth. I took his response to indicate that Duke and Dartmouth were very similar. Publicly, in contrast, he was a full time cheerleader for the "rigor" that higher education provided. Privately the first time he taught a class at Duke, he was in a bit of shell shock over the amount of student whining over workload and grades.
The current provost, Peter Lange, is another cheerleader for the "rigor" that higher education provides. We've never had a friendship. As a matter of fact the last time I saw him, he glared at me, then turned his head and walked away. Be that as it may, in the 1990s before he was provost, he truly believed that grades were rising at Duke solely due to increases in student quality. He was delusional, but he was honestly delusional.
When Peter Lange became provost, he decided for some strange reason to try and prove that this relationship between grades and student quality existed. He did this work in 2002.
Some people are talkers. Some people are number crunchers. Peter Lange is a talker and he has no quantitative ability. I'm not saying he's dumb. He just has a blind spot for numbers and science; lots of smart people do. So to try and "prove" that rising grades were due to increases in student quality, Peter Lange needed help. He got some guy in his office to run some Excel spread sheets comparing GPA's and SATs over time. Above are essentially the data he used.
The SATs shown are relative to a baseline; as a condition of my receiving SAT data six years ago, I was told that I couldn't publish them. Here I've done the equivalent of what the Soviet Union used to do with geologic maps prior to publication. They would remove the coordinates. Like them, I've removed the actual values of the SAT scores. I'm not implying that there are similarities between the secrecy and paranoia of the old USSR and Duke leadership. Well, only in a joking sort of way.
Anyway, if you're naive and look at that graph you might think: 1) grades are up; 2) SATs are up; 3) they must be completely related. That's apparently what Peter Lange thought. He didn't do any research about the relationship between SAT scores and GPA. He apparently just decided to wing it. If he had done a little scholarship beforehand - he does have a Ph.D. after all - he would have looked at the data and stopped right then and there. There's a 0.37 rise in GPA. There's less than an 80 point rise in SAT. Those two numbers just don't match up well.
If you look at other studies, SATs and GPAs don't correlate very well and optimistically you can maybe get a 0.14 rise in GPA per 100 point SAT rise. The number is more likely 0.1. Also the admissions office at Duke - like admissions offices elsewhere - is increasingly pressured to cherry pick students on the basis of their high SATs to maintain their high US News ranking; it's likely that the 80 point rise in GPA partly says nothing about increases in student quality.
Plus look at the graph. The average SAT score of the student body rises when grades stay static for many years. The average SAT score then goes static when grades rise. It's a mess. And based on other studies, only about 0.08 of the observed rise can be accounted for by the increases in student quality inferred from SAT scores even if you ignore cherry picking of SAT scores.
Lange had his assistant run some regression analysis on the data. Based on that analysis, Lange made the claim that half of the observed increase in GPA was due to increases in student in quality. How you can say that 1/2 of all the observed GPA rise can be explained by better students on the basis of the data shown above is beyond me. To explain half of the observed SAT rise you'd have to have a 0.25 coefficient between GPA and a 100 point rise in SAT. No one has observed that anywhere. It's fiction.
Let's put it another way. Suppose there were two students at Duke, one with a 1200 SAT and one with a 1400 SAT. What would be the expected GPA difference between the two? I actually have some experience advising Duke students with numbers like these at the freshman and sophomore level. Lange's analysis would say that the expected GPA difference would be 0.5. My experience says the number is about 0.2, which is in line with large data sets that have looked at such relationships elsewhere.
This is what a good relationship between a cause and effect should look like. I've plotted Duke's GPA versus the average private school tuition in the US over time in 2007 dollars. Duke's retail tuition is higher than this, but many students don't pay retail so this probably isn't a bad approximation for the actual average tuition students shell out. The one thing I've done to massage the tuition data is that, rather than plot the outright tuition in a given year, I've used the average tuition that freshmen to seniors first had to face as freshmen. The idea is that everyone responds to the initial sticker shock of tuition more than they respond to subsequent tuition increases.
Essentially the hypothesis being "tested" is that grades at Duke are rising because students are buying better grades. That's a distasteful hypothesis I know. I don't believe that this is actually what is directly driving grade inflation. But it is a far better model for explaining rising grades than better students. It's an amazing linear relationship actually, "explaining" over 97 percent of the variation in grades.
One problem with testing this hypothesis is that we don't really have a control. And we don't really have much variation in changes in tuition. Tuition and grades were both flat in the 1970s, but we can't shut off tuition rises now and see what happens. It would be an interesting experiment to keep tuition at Duke flat for a decade, though. I don't know if grades would stop rising as a result, but parents sure would be happier.
While the above curves don't prove anything, they are consistent with the hypothesis that grades started to rise at Duke (and nationwide) in the 1980s in part because college costs became burdensome relative to the earning power of Americans. Psychologically, the attitude of the American public toward college changed as a result. They used to defer to the powers that be to do what was best. But once they started to pay all that cash, they started to feel like they were part owners of the institutions their children were attending. If they were paying all that money, they wanted something tangible in return.
Prior to the 1980s, students in academe were acolytes. Since the 1980s, they have been consumers of a product. Academe used to be primarily in the knowledge business. Now it is primarily in the customer satisfaction business. One outcome of that transition is higher grades.
Now getting back to the "students" are better hypothesis; grades are definitely rising at Duke. Here's what happens if you try to plot GPA versus SAT over the time period 1976-2006. It's a bit messy. But if you want you can fit a linear line to it. You would be asserting that a 100 point change in SAT produces a 0.5 change in GPA. You're off in lala land if you choose to make such an assertion.
Want to explain half of the observed GPA by increases in SAT? Then you have to use Lange's coefficient of 0.25. It's a preposterous coefficient. Half of the rise in GPA at Duke is not due to better students. To make such a statement - and Peter Lange has repeatedly made that statement publicly - isn't consistent with either the data or with other studies elsewhere. It's fiction.
Lange certainly isn't the only college administrator in the US to have made claims based on fictional relationships between SAT scores and grades. In other places, the argument sometimes goes something like this. SAT scores have increased x percent. Grades have increased about the same percent. Voila! We have correspondence.
One problem with this kind of analysis is that GPAs are a highly damped indicator of what is happening grade-wise at an institution. For example, over the time period 1970-2006, GPAs increased 18 percent at Duke. SATs increased about 10 percent. So the naive analysis says that about about half of the GPA rise is due to better students. But wait a minute, if you want to compare percentages in an ad hoc way like this, here's another one for you: over that same time period, the frequency of A grades increased by over 100 percent at Duke. Looking it at this way, increases in student quality account for less than ten percent of the observed grade rise. Both analyses are silly, though. Ad hoc comparing of percentages is dumb no matter what measures you use.
In the recent Duke Chronicle report on grade inflation, Peter Lange told the reporter that he would need to analyze the data again before he decided to do anything about grade inflation. Lange is quoted as saying, "We have not looked at this matter in six years. Given the other things going on, we are not doing a major study of it." What would further analysis accomplish, especially considering Lange's obvious ineptitude with analyzing data? What would a "major study" tell him that isn't already obvious?
It's clear Lange and the rest of Duke's leadership don't want to touch this problem. They don't want to do the work. They are far more concerned with "protecting the brand" than with solving substantive problems. As a result, the quality of education at Duke University continues to suffer.
In the last year I was at Duke, a bigwig administrator made a pronouncement at some meeting (I heard this second hand) that I had "no respect for the institution of Duke University." That is absolutely not true. I have a great deal of respect for the institution of Duke and for higher education in general. What I don't respect are leaders who through cravenness and fecklessness run an institution into the ground. With regard to undergraduate education, that is exactly what has happened due to poor leadership at Duke and elsewhere.
But enough finger pointing for now. Can the problem of grade inflation at Duke (and elsewhere) be solved? Definitely. That's what I'll talk about next time.