Alan McLean wrote:
>
> A number (>10!) of years ago I directed a subject with assignments and
> of course had this problem. (I say of course because there is always
> some variation among markers!) I went through an incredible amount of
> heart burn, trying to do the best thing for the students. I started by
> taking into account both class means and class standard deviations,
> using the linear transformation referred to before. Then I decided that
> the SD had very little variation, and simply leveled the class means.
Our usual solution in my department is (unconsciously?) based on the
dictum that a good experimental design ahead of time is better than
fancy fudging afterwards.
We rotate the graders among the sections by week, so that each student
gets graded roughly the same number of times by each grader. (We also do
the same thing on the final exams, with each instructor grading one or
more questions across the entire multisection course.
We also remove confounding variables by not grading the assignment, but
a weekly quiz based on the assignment. [However, in many courses it is
not a pure quiz, but a part of the learning process in which they are
permitted to ask for hints and have work checked before submitting it.
Then they get 85% of their course grade from strictly-invigilated,
closed-book or limited-note exams. It seems to work.]
-Robert Dawson
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
