Ah. Ok. You originally said “you know which kids are strong” and i thought you meant “you know Johnny does A work”.
It’s not really about it being possible for every kid to get an A. It’s just that if i saud “Hey, tell me what an A means in this class”, I’d expect a discussion of skills or content, not “the cluster at the top of whatever test”.
In my experience, while you do get different clusters for “studied” and “didn’t study”, you don’t get much in the way of clusters at a finer scale than that. You won’t get an “A cluster” and a “B cluster” and a “C cluster”, and so on: Within either the studied group or the didn’t-study group, you’ll get basically a continuum.
Beyond the study-or-not bimodality, the only other “cluster” you’re likely to get is from the end of the histogram where your last category has an “or higher” in it, from the students who are so high-performing that they can’t be meaningfully measured and compared by the assessments in the class. But depending on the class, you might not even get any such students in the class.
This is generally what I would prefer to always see. Everyone’s grade would be holistically determined by looking through their entire body of work as a student in the class and attempting to assess just how well the student had learned the concepts that were taught in the class. The problem is that this is really hard on teachers, especially those in large classes. It makes more sense in grad school when the problems are going to be harder and the class sizes are much smaller, and so it’s harder to determine how well someone should have done unless you managed to give the exact same lectures and gave the exact same test for a decade in a row, which is probably not a great idea. If you instead focus your questions on testing specific things that are important to the course material and are emphasized as being important, you can put that into a very difficult problem and see if the students are using their learning in the class while at the same time providing an extreme discriminator to find the absolute best students instead of having too many people score in the 90s.
The worst classes/exams I ever took were those in which the exams apparently tested for different things than what was taught in class. For example, instead of questions that were related to problems we did in class, we’d get a problem that was assigned as homework and not discussed at all in class, and required an idiosyncratic approach that had nothing to do with the bulk of the material in the course. If it were only one of those two things, I wouldn’t have a problem, but with both of those it’s a complete crap shoot as to whether a student who understands the class material is able to do the problem.
At times, colleges will skew the curve to force failing because they have too many students in a program. My sister faced this her freshman year in college when more students than expected accpeted the college offer. They always expect a percentage of accepted applicants to choose a different school, but that year they guessed wrong.
So students who did well on tests were still failed because the bar was set so high. Professors were told that they basically had to fail 15-20% of the incoming class.
I’ve never encountered that, but it is common in a lot of majors to put the hardest class in the first year, so those students who will eventually fail out do so quickly.
Yeah, weed-outs. Pretty normal in the hard sciences at least. Flunking 15-20% of your class as a whole sounds really weird and counterproductive to me. Colleges don’t want their students to flunk out of school, in general. That’s not good for their stats/rankings/etc, if they care about them. Plus no tuition from a flunked out kid.
This was a specific program at the universary. I forget the actual number but it was something like they normally have 100 students enter the program each year from 200 accepted applications. But this year, 140 accepted and they had no place to put them in the dorms or fit them in the classrooms. Some were offered places in other programs, but of the ones who chose to stay it was a bit of a bloodbath.
It was an unusual situation, for sure.
Colleges do have the option of just not accepting students into a major, if they’re set up that way*. I remember a classmate being told he couldn’t be a computer science major. He’d done ok in the lower level classes, but apparently ok didn’t cut it, given the demand.
*This might be easier at a liberal arts school where the vast majority of your classes are outside your major, so switching isn’t a huge time sink.
Re: bimodal distributions, I remember my organic chemistry professor drew a histogram on the board after he graded our first test. “Those of you over here are in danger of doing ok. Those of you over here, you might want to change your approach.”
Yeah, I assume this wasn’t a liberal arts school, but a specific school within a larger university offering a very targeted program. At a liberal arts school, they’d have no idea what sort of classes you’d actually be taking, and you don’t declare your major generally for at least one year, if not two. Once upon a time you would have to line up to register and the classes you wanted to take could be full by the time you signed up. Now you all get to try signing up at the same time online, but you might have the same issue of not having space in the classes you want if you’re too slow. At a specialized school, it’s not uncommon for everyone to be taking the exact same classes in the exact same order, because that’s what the curriculum is for the only program that this school offers, and so there’s no real “registration” for classes - you take exactly what the program says you take this semester. Then there’s real problems if too many people are admitted to the program.
Depends on the major. All liberal arts schools say that you don’t have to declare your major right away, but in the sciences, at least, if you don’t start off on track, you don’t stand a chance. When I was in undergrad, the astronomy majors were required to take more credits each semester than most students were allowed to.
Going back to the OP, fundamentally there are two approaches to grading:
Measure each student according to a (mostly) objective standard of expected performance.
Measure each student relative to the other students. Whether that’s across the microcosm of a single section of a single semester class at a single school, or across everybody of that age nationwide as in the SAT, etc.
Before anyone can usefully start talking about what’s the right answer, it helps to know which question the grade is supposed to be answering.
Some examples of “curving” amount to trying to split the baby or answer a blend of both questions at once. Which is not necessarily a silly approach.
One more anecdotal data point. In my academic life, that term was used to mean adjusting what “100%” meant after the scores came in. Usually, any questions the best score taker missed were tossed.
Failing was still an option, but no teacher I had ever got approval for a grading system that guaranteed some number of failures based on a bell curve. Similarly, any professor that insisted on arbitrarily limiting the quantity of A’s, B’s, etc. awarded got pushback.
The SAT is actually a pretty good example of this. It’s hard to say if it’s norm or criteria referenced because it’s basically criteria-referenced but designed to get a bell curve, given what they know about what students can generally do. So the scale is set before a particular test is administered (so not a curve), but set because they can predict how kids will do and what it needs to be to make a curve (so curved).
On a purely practical level, “throwing out a question” used to be pretty much logistically impossible, even if it turned out that a question was literally wrong. You had to go back through tests by hand, and recalculate scores and if you have 100 tests, it’s just really not worth it–better to just give everyone 3 points, and if that means a kid who got the wrong question “right” gets a 3 pt bump they didn’t deserve, it’s not going to move the needle for the course average.
However, you can now buy $6 apps that scan your MC and will allow you to change the key post-test, updating all the scores. Any such software will also tell you which items failed to discriminate, and you can throw those out as well. However, very few education programs teach about test development, so that’s not a way that teachers generally think about tests. Even math teachers, in my experience.
Speaking as a math teacher, apps that scan papers are mostly off of my radar because doing multiple choice questions on paper at all is now mostly off of my radar. If you’re going to be giving an assignment that can be graded by a computer, why not just give it in electronic format to begin with?
But yes, most all-electronic assessments also enable things like this. There have been a few times I’ve had to go back to a homework assignment and cancel out a question because it turned out that it needed something we didn’t cover in class, or because the person who created the question made a typo and it wasn’t accepting the right answer, or whatever.
Its funny you day that, because our mayh department is the most resistant to going digital. They want everything on paper.
But yes, my point was that even if you are doing it on paper, its trivial now to get hold of platforms that will allow you to adjust things across the board after the fact. That was unthinkable 10 years ago. A lot of the point of a “curve” was a teacher having no other practical way to adjust a test that had a couple bad questions.
I used to work down the street from ETS, so I’m an expert. 
I know they test the hell out of new questions, sometimes including some that didn’t get scored. When I was in Grad school one of our professors was a consultant for the then new Computer Science Achievement tests, and some of us got to take a sample version of it.
I’ve seen lots of ways of doing this. At MIT just about all tests were curved, because many of them were very hard and they figured if you got in you probably deserved one. You could fail a course, but you had to work pretty hard to do so. Some of my friends worked hard in this way. Typically they’d show the curve when the test was returned.
In my first grad school I TAed for a large class, and we worked out the curve based on the data from the test or assignment. I think most classes did this.
In my second grad school I was an instructor for a term and gave a very hard test. The students did not do well, and freaked out because the tradition in the department was that there were preassigned cutoffs for A, B, etc and they all got Cs. I reassured them that they were not going to all fail, since the fault was more in my test than in them.
At Illinois the class I mostly TAed was assembly language, which was where we sorted those who did fine with Fortran, barely, in CCS101 from those who knew what they were doing. We graded the hell out of the first programming assignment, to scare them into using structured programming (which was new back then.) We threw out the grade later, but we got much better assembly code from doing this.
Experimental standardized-test questions can, and are, slipped into examinations in years before they are officially scored, so they do know in advance how kids will do on them.
One of my grad school classes, it was the first time the professor had ever taught. The first test, the highest score was 27%.
It was the Physical Therapy Department of Sargent College at Boston University.