[RE][Rasch] Prameter to observation ratio?
stephen.humphry at uwa.edu.au
Thu Feb 14 17:49:37 EST 2008
Iasonas, I agree with you and think you make very good points. Targeting is
important for information, which is important for the precision of item
estimates. This becomes obvious in extreme cases -- for dichotomous items,
if almost all responses are 1s, while 0s are rare, then more responses will
Rasch (1977) described that "whole context for testing", to which you refer,
as the frame of reference. DIF and violations of local dependence, in any
form, are also important. Mike Linacre refers to targeting and DIF in the
RMT article to which you referred.
Purya, it seems you mention two different things. First, you mentioned the
parameter to observation ratio. To me, this implies the ratio of the number
of responses to the number of parameters; i.e. for both items and persons in
combination. In general, this is different from the "ratio of parameters to
persons" to which you later refer, though the two are of course related for
a given number of items. Anyhow, though interesting questions when 'all esle
is equal', I don't know how useful it is to ask them in general terms: I
think it's better to be concerned with good fit and acceptable standard
errors for the purposes at hand.
Dr Stephen Humphry
Graduate School of Education
University of Western Australia
35 Stirling Highway
CRAWLEY WA 6009
P: (08) 6488 7008
F: (08) 6488 1052
From: rasch-bounces at acer.edu.au [mailto:rasch-bounces at acer.edu.au] On Behalf
Of iasonas lambrianou
Sent: Thursday, 14 February 2008 3:07 PM
To: Purya Baghaei
Cc: rasch at acer.edu.au
Subject: [RE][Rasch] Prameter to observation ratio?
Altough I do not have any 'proof'/evidence, my suggestion is that your
sample size is related not only to the number of the items but also to the
type of the items, how well targeted the items are (a test should be
optimally of the right difficulty of the students), the absence of too many
misfitting response patterns (not many unmotivated or cheating or guessing
students) etc. So, in real life, its not just how many students we need per
item, but the whole context of the testing situation is important. To me,
there item calibration can only exist in some context. Of course, this is my
opinion based on my experiences, others wil hopefully be able to help you
with references and numbers.
---------[ Received Mail Content ]----------
Subject : [Rasch] Prameter to observation ratio?
Date : Wed, 13 Feb 2008 18:09:13 +0100
>From : "Purya Baghaei" <puryabaghaei at gmail.com>
To : rasch at acer.edu.au
Dear list members,
I vaguely remember something that I read or heard (don't know where, may be
dreamed) about parameter to observation ratio. Is there really such a thing?
Is it possible to stably calibrate say, 100 items with 50 persons?
Is there any rule of thumb for parameter to observation ratio?
I have read Mike's paper in RMT, where he convincingly specifies the right
number of persons for stable estimation but doesn't mention the ratio of
parameters to persons as a factor that can affect estimation.
I'd be thankful for any sources that have addressed this issue.
I was thinking of carrying out a simulation study and find out more about
Is it really worth doing?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch