Hacker News Reader

Comment by kqr | Hacker News Reader

> it screws up the class curve

This is your problem right here. There is no rational argument for grading people on a class curve. That just makes the class a lottery and incentivises sabotaging for others to get an upper hand yourself.

View on HN

Curve grading is such a messed up thing. It leads to professors assuming you should get all A's in all of your classes basically, cuz of this nonsense from US unis.

I did an exchange year, and sent my grades over to a professor. Guy was like "What's up with all the non As" and I had to talk about how we do things differently here (giving points for right answers, and then adding them all up).

Of course there's always an overall curving happening on a high level because teachers choose how hard to make assignments or not, and ultimately grades are not really fundamentally important, but when people just have those to judge you on and choose to, it really fucks things up.

I don't get it, don't curves limit the number of people who get an A each semester? But potentially everyone in a class that isn't curved could get an A (or F)

it depends on the curve. "curve" is kind of vague here, sometimes a prof will curve by adding so many percent to everyone's grade to get enough people over a certain threshold

but also the prof will know if the class is full of overachievers and curve in a way commensurate to the class

Well for example, in my school, basically nobody would get 90% on an exam. Very good people would get 80% but the average is more around 60%. So you stuff that into the grade conversion for foreign universities and ... well lots of people get C!

I think that curves discourage cooperation and encourage zero-sum thinking. Curves are only really necessary when the professor is out of touch with the class and/or prerequisites.

I don't think that people would typically go to the lengths of trying to sabotage someone else, in practice it could just look like a bunch of people working separately and not cooperating at all

In an ideal world, there wouldn't be any curves and the coursework would be tailored sufficiently for every cohort of students for every semester

I think of helping each other as part of the reason we put students in the same room to begin with. So when a student asks another for help with understanding a concept and the other refuses that request, that's working against the purposes of group education, i.e. sabotage.

But I admit I might be a bit radical here compared to most people. (Who I am sure will argue that utilising teacher time to the maximum is the only reason to gather students in one room.)

I agree with you, but also most of the time when people ask each other for help it isn't to help understand a concept. It's usually for source code.

In the class I mentioned above, the prof and TAs weren't pissed off about people using each other to understand concepts or even high level design. They were pissed because people were copying source code verbatim, comments and custom debug messages and all

On a related note, if someone ever asked me for help with a concept or high level design I would be more than happy to oblige. But (most of the time) people aren't asking for that, they want source code that they didn't contribute to

this could be fixed by making assignments collaborative, but then the professor has no means of verifying where everyone is in terms of understanding

I've always loved curved classes because they allow very good professors to give real mindfuck exam questions that cause you to walk away from the test with a new perspective on the material.

Those exams were often the most edifying couple hours of the whole semester. You also got a clear sense of the difference between you and a real master of the material which is a helpful lesson in humility.

I went back to grad school recently and it seems like that mode of testing has gone out of style in the last ten years. The exams I took were geared more towards establishing a minimum bar of competence, more for my future employers' benefit than my own.

they allow very good professors to give real mindfuck exam questions that cause you to walk away from the test with a new perspective on the material

I think the problem with this is, although cool, very few professors are capable of doing this well and very few students benefit from it. IOW, it doesn't scale.

To my mind, oral finals or 'discussions" would repair the cheating issues quite well. But again, it's hard to scale and professors would need to be trained in how to do them.

You don't need after-the-fact curving for that, though.

When I set (free-form written, math/CS) exams, I always made a point of designing the exam in a way where I didn't really expect anybody to get more than 90% of the points. I also made sure that students knew this.

I always set grade brackets before grading (e.g., 80% of points gets you an A, below 35% is a failing grade, and so on). I always ended up with a pretty reasonable grade distribution.

I took a class that gave mindfuck questions and did not need curved grading.

It was simple. The weekly assignments contained 6 questions. One of those was the mindfuck question and wasn’t graded (if you solved it you got extra credit, but hadn’t been solved so far).

You don’t even need to hide the crazy questions. Offering them as extra credit is a simple solution that works well.

Curve grading does have a place. But that is standardized tests or placing entire cohorts to buckets. So we are talking of hundreds if not thousands of students. For individual schools or classes inside thereof it is wrong method. Either the students know enough of the course material to pass or they do not.

So true. Curve grading works well when the cohort is large and heterogenous. (Like, say, Finnish matriculation exams. >20k people in a single cohort, and each individual exam within the whole thing with at least 5k people in them.)

As far as I'm concerned, curve grading single courses is just as bad as stack ranking at work. And in here, with high concentration of engineers who loathe stack ranking, a disproportionate fraction are somehow in favour of per-course curve grading.

Blind spot? Stockholm syndrome? Go figure.

Fun thing with Matriculation examination. The small subjects are not on curve. Like for example Latin... Or higher level Russian where there is enough native speakers to have an effect on it.

Whole stack ranking is just weird. Either the people are good enough to do the work or not. If not eventually fire them. On other hand at top the rewarding extra is an problem, but I don't know if you can do that without someone gaming it.

That's a good point, thank you. I wasn't aware of it, or at least didn't actively remember the fact. But it makes sense and circles on the same thing. Curve grading only makes sense when the cohort being evaluated is large and varied enough.

> Whole stack ranking is just weird. Either the people are good enough to do the work or not.

Even at the risk of veering quite far from the thread topic, I am not sure that's the only angle to look at things. In my experience, pretty much anyone who is genuinely curious and has the discipline to work at understanding how/why a ${THING} functions, is going to be good at their job. Some people just don't find what their thing is, and surprisingly many are in jobs where they are not even allowed to figure it out.

    As far as I'm concerned, curve grading single 
    courses is just as bad as stack ranking at work.

In my years of schooling, I never saw curve grading resemble "stack ranking" whatsoever.

In stack ranking you have to designate some team members as superior and some as bad. You could have a team of ineffective idiots and be forced to label some of them as high achievers, or a team of superstars and be forced to label some of them as bad. I agree that this is manifestly ridiculous.

I've never seen curve grading like that.

If you want to compare grades across semesters you have to either keep the difficulty of the tests constant or adjust the grades to compensate for how difficult the tests were relative to previous semesters.

BTW, I don't know if it's a named fallacy, but saying "there's no rational argument for X" is not a good argument against X. You can't prove a negative so you cannot know if there is not a rational argument for X.

Of course there’s a rational argument to using the distribution of scores to set the grades: If everyone in the class gets half the questions wrong on the final exam, you shouldn’t just fail the whole class! The test was probably just harder than expected

But in that same example, if many of the students get close to 100% on the final by cheating, then the remainder of the class will suffer

I disgree. How hard it is to pass a class should be an absolute. It shouldn't be easier for one cohort over another just due to the first being, on average, less prepared than the second.

If an exam turns out to be too hard the teacher can investigate, consult their peers, talk to their students and decide if it's worth to scale all grades by a factor to correct.

> How hard it is to pass a class should be an absolute.

It should be, but profs and TAs change from semester to semester. The coursework changes also. Also the profs, TAs, and coursework of prerequisites change too

> It shouldn't be easier for one cohort over another just due to the first being, on average, less prepared than the second.

A good professor will have the sense of discretion necessary to know when a class is half-assing on average

No, the proper response to that is still not a curve, it's to identify which block of questions wasn't appropriate and removing those from scoring or turning them into "bonus points" and similar measures.

That way you don't incur any of the (pretty severe) drawbacks of a curve but don't punish students for questions that were badly phrased or weren't properly taught.

Simply removing the offending questions after the fact doesn't solve everything. Students may have fruitlessly wasted a ton of time on those questions, causing their grade in other parts of the test to suffer.

Are how do you suppose an instructor can identify which questions were too hard when a big chunk of the class got them right by cheating?

A lot of my classes the average grade on a test could be as low as 30% with a P100 of 50%. The material was extraordinarily difficult. I’m glad it was because it stretched me in equally extraordinary ways to do my best. But I don’t see how they would have assigned grades without a curve. In most cases I agree with you, but in those classes I wouldn’t have changed to tests to dumb them down but I also wouldn’t have accepted score based grades. The curve felt fairly rational in the situation and I think the distribution of performance reflected a grade curve well enough. The GPA though for the department was depressed relative to others but it was also widely regarded and recognized internationally as extremely difficult.

    There is no rational argument for grading people on a class curve

It's an imperfect solution for a legitimate problem.

The counterargument is that it would be rare for a prof/teacher to absolutely nail the correct difficulty for an exam.

If they feel that their students' average exam grade fails to accurately reflect how well the class is learning the material, then some kind of adjustment makes sense... imagine a scenario where you have a classroom full of motivated and engaged students who are showing good understanding of the material, and yet the average exam score is 60%. This strongly suggests professor/teacher has erred and an adjustment is in order so that the scores better reflect actual mastery of the material.

There are of course also a lot of situations where grading on a curve is blatantly unfair and/or simply makes no sense.

This is why I'm generally against grades with more than three levels.

I would like to see a system where almost all (something like 99 %?) of students just plain "pass". Then we have the lower statistical anomalies (lowest 0.5 %) and upper statistical anomalies (highest 0.5 %) which fail and get an "excellence" type grade.

If the average score is too low or too high, that doesn't mean more students fail or perform excellently, that just means the teacher needs to recalibrate.

(Of course, these percentiles ought to be measured over as inclusive a reference class as practical. Ideally across schools and years simultaneously.)

That sounds promising!

Reply on news.ycombinator.com