Let's assume, for a moment, that you believe the average and stdev of
performance for groups of students are the same.  Each group is
defined as "those students graded by a single grader."

then you can rescale each group as follows.  For grader A, all grades
in the group are scaled using:

grade' (grade prime) = (raw grade - mean_of_A_grp)/stdev_A

and so forth for group B, C, etc.

Then with all the grades together, decide what the overall average,
mean_overall, and stdev, stdev_overall, you would like them to have,
and rescale again:

reported grade = grade' * stdev_overall + mean_overall

this will work if you assert loudly that all subgroups are drawn from
t he same population (the average & stdev depends only on the grader,
not the students), and if your grading scale is truly linear.  A scale
with a maximum of 100% ain't linear at the top end, sorry.

I would be more concerned that the graders can interpret the answers
in such blatantly different ways.  Perhaps the students do the same,
which begs the question of the precision & usefulness of the
questions.  Reviewing the questions with your graders might tighten up
your (instructor's) part of the process.

Jay

Tristan Miller wrote:
> 
> Greetings.
> 
> It is sometimes the case that, when a large group of assignments are
> graded by different markers, one marker will grade much more or less
> harshly than the others.  So that students graded by said marker are not
> unfairly (dis)advantaged, it is the policy in some places to scale those
> marks up or down.
> 
> My questions, then, are as follows:
> 
> 1) Given that the n ideally-marked assignments (sample A) have mean grade
> x_A and standard deviation s_A, is there some function f() which can be
> applied indiscriminately to the m poorly-marked assignments (sample B)
> with mean grade x_B and standard deviation s_B, such that their mean and
> standard deviation are changed to x_A and s_A, respectively?
> 
> 2) Is f() unique?  Are there multiple such f()s?  An infinity of them?
> 
> 3) If f() is not linear, does it have a linear approximation?
> 
> 4) Is such a scaling even desirable?  It seems to me that, assuming a
> scaling is mandated, bringing the mean and standard deviation into line
> with the other assignments is the fairest possible way of adjusting the
> marks.  But then again, this is just a gut feeling of mine and I haven't
> gone through a rigorous proof.  I get the feeling that this question has
> already been asked and that the results and discussion are published
> somewhere. :)
> 
> --
> \\\  Tristan Miller
>  \\\  Department of Computer Science, University of Toronto
>   \\\  http://www.cs.toronto.edu/~psy/
> 
> .
> .
> =================================================================
> Instructions for joining and leaving this list, remarks about the
> problem of INAPPROPRIATE MESSAGES, and archives are available at:
> .                  http://jse.stat.ncsu.edu/                    .
> =================================================================

-- 
Jay Warner
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA

Ph:     (262) 634-9100
FAX:    (262) 681-1133
email:  [EMAIL PROTECTED]
web:    http://www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to