When teachers curve grades, does it ever happen that someone ends up with a lower grade than their raw score? I.e. getting 80% but ending up with lower than a B-, getting 75% but lower than a C, etc?
Or does curving grades only ever help the students’ grade?
A genuine curve would certainly lower some grades sometimes, by pushing the actual distribution into an ‘expected’ shape.
Most people who talk about “grading on a curve” really do no such thing; they just slide the whole set upward uniformly.
Yes. In the classic implementation of grading on a curve, the bottom X% of the class gets an F, regardless of how well they did in an absolute sense. If the entire class scores at or above 90%, the people who got exactly 90% would presumably get Fs, since they are technically at “the bottom of the class” (and maybe 91% and 92% as well, depending on the other scores).
This is one reason why very few professors actually grade on a curve.
So teachers who say they “curve” grades really must mean “Since you all did worse than normal, what counts as an “A” is a lower grade than normal and the same for a B, C, etc.”?
I feel like what teachers call “curving” is just lowering the bar for what counts as an A, a B, etc.
I doubt any significant number of teachers curve grades based on a normal (bell) distribution, which would indeed possibly lower the scores of some.
Usually, teachers ‘curve’ grades when most of the class does worse than expected. The teacher assumes the test they made was too hard, and so, scores get bumped up and no one goes down.
Simply adding a specific number to the scores can work if everyone’s score was low. So if the top scorer was 60%, then everyone gets +40%. That’s not really a ‘curve’ as much as a very linear shift of all grades.
When you get a ‘curve breaker’ in the class, i.e., the one who scores 100% and everyone else failed, then you can’t bump grades like that and some teachers say, “oh well, can’t curve the grades because moriah got 100% and it wouldn’t look right to give him a 140% on the test.” (True confessions of a curve breaker here. 
)
But you can create a pseudo ‘curve’ (by tilting the linear bump) where the closer you get to 100%, the less you add and the further from 100% the more you add. E.g., this formula: New grade = 50 + [.5 x (old grade)]. In that formula, a 100% remains 100%, a 90% becomes 95%, a 70% becomes 85%, a 40% becomes 70%, and so on.
A true curve uses an exponential in the formula, or a trigonometric function (arctangents work well). And the math whizzes will use natural logs to get a true bell curve (which have the chance of lowering some grades).
.
I had a quite a few professors grade on the curve. They would give an X number of students an “A,” the next group and A-, then down from there.
When I was a TA, the only real question on where to draw the line was for C-, the minimum grade which the student needed in order to not have to retake the class in our engineering degree. The professor would also take into consideration the students’ efforts for those on the borderline. She didn’t fail or give a D to that many students, but enough that it did take work to get a C-.
We had a fairly competitive pre-engineering program, so the powers which be put a lot of pressure on the math, chemistry and physics departments to not allow grade inflation or grade creep. There was a limit of 110 students for the sophomore program and 100 students into the EE program itself. For the sophomore students, if you were in the bottom 10, you didn’t make the cut and would have to try to retake classes to improve the next year. If you couldn’t then you were out. I knew some people who had to give up EE and go into another program which was less competitive.
It doesn’t really lower a grade because it doesn’t matter the percent right you get, it matters where you are in the pack. The only time it would be particularly unfair is if everyone is very tightly clumped together, and that’s the fault of the instructor giving too many questions that are too easy or too hard and don’t differentiate enough.
I like this explanation of several methods.
I do it by an exam-by-exam basis. I might either add X points, or subtract from the total (e.g. if someone originally got a 75/100, the “curve” might be 75/95 or 79% now). The big disadvantage of those is when someone gets a 100%, they might end up with an inflated 110%. In those times, I will use on of the other methods outlined. In one course, grades tend to be very bimodal, so the flat methods don’t work.
I haven’t encountered a professor while teaching or as a student who really does a curve on their own. However, some departments seem to mandate one. Highly competitive ones, I suspect.
If someone’s final grade were an 89.7%, I would be very hard pressed to not give them an A-. I can be lenient/nice that way, although it also avoids anyone trying to dispute it or follow up.
As far as grade inflation goes: it’s a fairly arbitrary idea that “superior” work on a test/assignment/whatever = 90% correct. Deciding that no, on this test, 70% correct is superior work does not have to be inaccurate or inflated: it may just be a really difficult test. On the other hand, 90% on a very easy test may not be superior work at all. A percentage is just a number: it doesn’t tell you anything without context.
As a point of reference, the AP exams generally set 70-80% correct on the MC as the point that earns a 5, the highest possible score. (There is a free response as well: here, too, a student is doing very well to get 80% of the points). Depending on the exam, 5s are generally 5-10% of total test takers. So are they giving 5s too easily because you only have to get 80%, or giving them too rarely if only 5% earn them? I think neither, but it illustrates the problem with being too tightly married to the 90% = A mindset.
There is absolutely no reason to equate an 80% with a B- or a 75% with a C. Those are artificial numbers. Who decreed that one equals the other in all cases? Why should it be so?
Grading on a curve was designed to free instructors from this artificial equation and give them leeway to assign grades based on the class’ actual performance. It may work and it may not, but the same thing can be said about equating percentages and grades.
This is typical in my experience.
I was in some STEM college classes where the mean grade was 50%-ish and several people got 20’s and the effect of the “curve” was to raise everyone’s grade to the range used by school’s official percentage-> letter grade mapping chart (e.g. 90-100 = A). I don’t recall ever being in a truly statistically curved class where even a raw score of 93% could be an F if everyone else got a 94 and above.
I’ve heard tell that there are some teachers who use a curve for part of the grade range, but guarantee a minimum grade to people who actually try. For example, the curve might only apply for A-D and the F grade was reserved for people who didn’t turn stuff in, quit attending, or failed to achieve even a minimal level of achievement. This protected against people failing outright if they would have been, say, B or C material except for the fact that the class happened to be packed with geniuses, people with uber-leet advanced study skills, and people with super-above average effort motivation.
In my experience, most classes tend to have a bimodal distribution of grades: You get a hump near the top from those who studied and otherwise put in the effort, and then you get a hump lower down from those who didn’t. In a good class, that lower hump will be small, and in an easy class it’ll be close to the upper hump, but it’ll still generally be detectably there. And it’s generally pretty absurd to try to treat a bimodal distribution the same way you’d treat a single-modal distribution, in any way (much less actually reshaping it into a single-mode distribution).
Don’t some of the cheaper law schools do exactly that? They grade on a strict curve so that the bottom 10 to 20 percent are GUARANTEED to flunk out after the first year, cutting the absurd class size down to normal and pocketing 1 year of tuition money? Or is that just a myth?
The problem with grading on a true curve is it assumes criterion-based data is actually normed-based data. Look at is this way, if I teach my students how to tie a tie, why should there be an assumtion that 16% will have no clue how to do it, 34% will kind of get it, 34% will almost be able to do it correctly and 16% will actually learn how to tie the tie.
I see what you are saying, but most courses aren’t a one off task. Tying a tie isn’t the same as learning macro economics.
However, I agree that a hard curve shouldn’t be established so that a certain percentage of students automatically fail.
As far as I know it’s a common practice. At my school the bottom 14% were guaranteed to fail, and then be kicked out. This leads to a very intense competition and lots of gamesmanship. On one writing assignment, one kid got a 4/5, all of us got a 5/5, so he got an F.
From what I can remember about my college, grading on a “curve” actually meant, in practice, that if the average raw score was less than 75 (the typical numerical score for a C) then the difference was added to everyone’s average for purposes of assigning a letter grade. It never meant assigning grades in a way that conformed to an actual curve.
There is no such beast at my school, and why should there be? Maybe 5 to 20 percent should flunk out. Maybe everyone will be a good lawyer. Why set an arbitrary number to flunk students?
And that’s the problem with curves that try to force a bell curve or ranking strata.
If somebody got a straight-up perfect score, why would you boost any of them?
The question in the title suggests that there is a predetermined way to go from numerical score to letter grade, but there isn’t. Grading on a curve doesn’t lower anyone’s grade, nor does it raise anyone’s grade. It’s the very process of determining what the grades are in the first place. In high school there might have been a chart that said “90% or better is an A, 80% to 90% is a B” or whatever. As material gets more challenging, though, this doesn’t work anymore.
At the university level, the exams and homework assignments each year will typically differ and the quality of the instruction can also differ. There is no way to know ahead of time that 75% is a magical score above which someone deserves one letter grade and below which they deserve another. In large classes, the best way to see what scores people should be expected to get for this incarnation of the course is to see what scores people actually got for this incarnation of the course. In other words: you have to look at the distributions of scores to even start assigning letter grades.
It comes down to what you want a letter grade to represent. Grading on a curve means that you want a letter grade to represent how this student handled the material in relation to other students. That’s very useful to know. If the final exam ended up being a doozy and everyone’s numerical scores are lower this year, this won’t affect what “A” and “B” mean, which makes the letter grades more informative across years.
If I see a grade of “C” on a transcript from, say, Columbia, I know what that means: this student had some trouble relative to other typical Columbia students on the material in this course. I do not need to worry if it was just a hard exam and everyone else also got C’s. Likewise, if I see a grade of “A”, I know that this student handled the material well in comparison to typical Columbia students, as opposed to wondering if the instructor just made the class too easy this time around.
The pass/fail threshold is a separate matter, as failing means something very specific, namely that the student didn’t master the material well enough to even claim credit for it. One can easily handle setting the pass/fail line manually, but the full distribution of A+/A/A-/B+/etc. needs some form of calibration (“curving”) to have any usefulness at all.