Re: what good is "percentage error"?

[Doug Craigen <dcc@cyberspc.mb.ca>]

On Tue, 13 May 1997, Rauber, Joel              Phys wrote:

> Yes, really. In the quantitative analysis labs I am familiar with, and where 
> the students are graded in this fashion; they DO NOT have a known result to 
> expect, because the answer is an unknown to them.  Therefore in their 
> attempts to get within x% error they do their darndest to reduce their 
> errors in measurement  and learn all sorts of subtleties to getting accurate 
> measurements; just as you hope they will.
>
> Accuracy is of paramount importance here and I think it is quite justifiable 
> to grade in part (or maybe even in whole, depending on circumstances) on 
> accuracy, which definitely means a percent error type criteria; at least in 
> part.

It appears that we are talking apples and oranges here.  My point has
been that we cannot replace proper accuracy estimates (a more positive
term than "error analysis") with a discrepancy between the obtained value
and an accepted value.  If you grade according to "percent error" as
compared to an accepted value, and the students know the value, you will
find everything they do is driven towards getting that value so that the
grading becomes a farce. I consider it perfectly legitimate to use
"unknowns" (which the marker knows but the student doesn't) as a test of
the true experimental skills of the student.  However, when doing so the
marker should also have some proper analysis skills.  Suppose that you
somehow did a measurement of 'g' without any of the students knowing the
"right" answer is 9.8.  Your own experience with the equipment says that
the best you can get out of it is a value of +/- 0.1.   Then students who
get 9.7, 9.8, and 9.9 should all receive the same mark for accuracy.
Markers very often are unaware of this distinction.