Tuesday, December 8, 2009

On tight grading distributions

High schools in Virginia were faced with the need to "raise standards," so they did 2 things: (1) they instituted "SOL's" - standardized tests that, as far as I can tell, mainly give teachers a teast to teach to; and (2) tightened grade distributions so that each letter grade is only five units wide. Well, they did the same thing with the GRE around 2001.
I always knew this was a stupid idea (in both cases) but I could never quite put my finger on why until today: As Craig puts it on his blog, it will "reduce the variance in the scores and, unfortunately, reduce the signal-noise ratio in the scores." If the test is written so that the D students are able to get an 80-85, how am I supposed to distinguish between a D student who was lucky and a B student who was just unlucky on one or two questions? I'm all in favor of keeping a wider variance and letting the distribution fall where it may.

No comments:

Post a Comment