Tuesday, December 13, 2005

da curve, boss

So, what is it about US schools that imprints students with the idea that grade curves are numerical absolutes?
And what idiot set 90/80/70 as some sort of "standard" percentile for the A/B/C grade?

It took me a surprising long time to realise that a lot of students though I added an arbitary percentage to their numerical grades to "set the curve"... I suppose what is done can be considered functionally equivalent to doing so, except in so far as it is not.

The person setting gradable tasks and who does the grading controls the mean and range of grades, and to some extent the shape of the curve, although it is very hard to force a curve for a large class far from either a single or a double gaussian.

Personally I feel a mean percentage grade around 60% is good, with an A represented by maybe 70% or higher, though for psychological reasons it is near impossible to actually set curves that low in actual classes.
Bunching grades in a narrow percentile range makes individual grades vulnerable to random errors and single stupid mistakes and is a degraded indicator of true performance.


Anonymous Anonymous said...

Do you realize that the normal curve is not actully normal as on .001% of the population is not in the class for for another normal curve(entrence exam)

11:30 AM  
Blogger Steinn said...

Not sure what the point is.
From observation, for a "large enough" a class of undergraduate students, the distribution of numerical grades is almost always a good approximation to a "normal" or gaussian distribution, or two superimposed gaussian distributions.
That is just what you see when you measure it.
There are usually non-gaussian tails, in that there are somewhat more outliers for very low and very high grades than you would expect.

Certainly undergraduates are not an unbiased sample of the general populations; for PSU they are a biased sample of about 15% of that age group which did best in certain kinds of tests and other activities, for other universities there is a similar but quantitatively different bias.
But the classes tend to sample the PSU population well, for large enough (probably more than 20-30 students) class.

11:52 AM  

Post a Comment

<< Home