If anything, it should be a lower resolution system. If you measure a distance by crudely pacing it off, you don't say it is "134.581234 yards". You say it is about 130 yards.
Consider the following questions in thinking about how many significant digits should be expressed in a grade:
* Go back through every homework and test for your entire college career, and have them re-graded by the professors involved. Or consider doing that for a sample of 50 or 100 students. By how much does the GPA change ?
* Go back through every homework and test for an entire college career, and have it re-graded by a commitee of 10 experts in the field, who must all agree on every problem's grade after consulation with each other. By how much does the GPA change ?
* Does the average grade given out by a professor or other grader change if they are in their first semester of grading / teaching, versus following semesters ? If they grade late at night versus early in the morning ? If the name on the paper being graded is that of a good or poor student ? ( Presuming it is the same paper for both students -- this study was done by two students I know, who agreed to copy each other's papers for a couple of grades and then confronted the biased teacher and were punished for it.)
Now, if you are putting so many significant digits on those GPA's that non-student related stuff such as which professor they were randomly assigned to comes into play, then you are just fooling yourself. And that's only considering how good grades are at predicting THEMSELVES.
And, if you want to consider how good grades are at predicting anything else, except maybe "will I get into medical school" and other explicitly grade-related questions, I think you end up ignoring them all together. I would only care about whether or not they graduated high school, which college to a very loose degree, and whether or not they finished college. The presence or absense of a master's degree does not seem to predict anything (except maybe good grades as an undergrad).
If you assign a number to anything, some people will become fascinated by it. If you made a video arcade game, where you pressed a button once and it gave you a random score between one and a million, some people would stand there feeding in quarters until they were at the top of the high scorers list.
If you think that the grading system is low-resolution, then you aren't trying to use grades to predict anything useful about people, you are just fascinated by a very expensive and time consuming video game.
Consider the following questions in thinking about how many significant digits should be expressed in a grade:
* Go back through every homework and test for your entire college career, and have them re-graded by the professors involved. Or consider doing that for a sample of 50 or 100 students. By how much does the GPA change ?
* Go back through every homework and test for an entire college career, and have it re-graded by a commitee of 10 experts in the field, who must all agree on every problem's grade after consulation with each other. By how much does the GPA change ?
* Does the average grade given out by a professor or other grader change if they are in their first semester of grading / teaching, versus following semesters ? If they grade late at night versus early in the morning ? If the name on the paper being graded is that of a good or poor student ? ( Presuming it is the same paper for both students -- this study was done by two students I know, who agreed to copy each other's papers for a couple of grades and then confronted the biased teacher and were punished for it.)
Now, if you are putting so many significant digits on those GPA's that non-student related stuff such as which professor they were randomly assigned to comes into play, then you are just fooling yourself. And that's only considering how good grades are at predicting THEMSELVES.
And, if you want to consider how good grades are at predicting anything else, except maybe "will I get into medical school" and other explicitly grade-related questions, I think you end up ignoring them all together. I would only care about whether or not they graduated high school, which college to a very loose degree, and whether or not they finished college. The presence or absense of a master's degree does not seem to predict anything (except maybe good grades as an undergrad).
If you assign a number to anything, some people will become fascinated by it. If you made a video arcade game, where you pressed a button once and it gave you a random score between one and a million, some people would stand there feeding in quarters until they were at the top of the high scorers list.
If you think that the grading system is low-resolution, then you aren't trying to use grades to predict anything useful about people, you are just fascinated by a very expensive and time consuming video game.
Not that there's anything wrong with that.