A short, unsystematic treatise on grades

Matt Olsen posts an interesting question on grading, specifically with regards to school grading. What is a useful grading system? Notice I do not say fair, though I am of the opinion that these adjetives by definition describe the same system.

The primary goal of a grading system should be to give information on performance. As such, I am in favor of systems that are based exclusively on output rather than inputs (i.e., based on performance demonstrated rather than efforts exerted).

Let me say at this point that I have yet to see a perfectly implemented grading system (one that provides perfect information on performance) and indeed, I have seen some rather terrible systems. The lack of perfection stems from the inefficiency in evaluating performance. As humans we can only process so much information intelligibly and therefore generally sacrifice information on specific performance by aggregating grades (e.g., one grade for an entire test, one final grade for a class, one cumulative GPA, etc). To the extent grade users can glean enough information from these aggregated metrics, this is not such a grave problem.

One question often asked is whether to use an absolute grade scale or a "curve" scale. This depends on what information is most useful to the grade's user. Mastery of clearly defined material, if standardized appropriately, is best supported by an absolute grade. Using the curved scale, I may get an A+ in Thermodynamics without mastery granted everyone else does worse. The curved scale comes into play to eliminate variablilty caused by a particular class (e.g., bad professor, lots of snow days, etc).