[Rasch] Items with various partial credit models
rprimi at uol.com.br
Sun Jun 18 10:23:44 EST 2006
Thank’s a lot. By using STKEEP my problem is solved.
I still have some theoretical doubts. Let me try to elaborate it.
Suppose that I have two groups of 4-items A and B. Then, A items are recoded
from 1234567 to 1111222 and the B items are recoded from 1234567 to 1112334.
Suppose two subjects with the following answers
S1 1111 1777
S2 7771 1111
Using the original answers they would have the same measures but different
IN and OUT fit’s because one of the response patterns will be more likely
based on the model expectations. But If I use the recoded answers, S1 will
have higher total score and correspondent higher measure than S2 because the
B items goes until a maximum of 4 while A items goes until a maximum of 2.
So although considering the original answers the two subjects have the same
total score they would come out with different measures under the new
I truly don’t know if this is a real problem or not. Do recoding with
different total maximum scores imply valuing items differently? Would be a
better choice to recode with 3333444 (setting STKEEP= NO) instead of
It would be nice hear what you have to say about this
From: rasch-bounces at acer.edu.au [mailto:rasch-bounces at acer.edu.au] On Behalf
Of Mike Linacre (RMT)
Sent: Friday, June 16, 2006 1:30 AM
To: rasch at acer.edu.au
Subject: Re: [Rasch] Items with various partial credit models
Thank you for contacting us.
Your two schemes differ because of the Winsteps default of "STKEEP=YES". If
you want them to be the same, set "STKEEP=NO"
So, assuming you are using different GROUPS= indicators for the different
item structures, then
For STKEEP= NO, IVALUEA="1114557" means "1112334" with the categories
treated as strictly ordinal.
For STKEEP=YES, IVALUEA="1114557" means "1114557" with the unobserved
intermediate categories 2, 3, 6 treated as sampling zeroes.
Please email me if you need more assistance.
At 6/16/2006, you wrote:
Dear list members
I am working with a personality test with 7 point rating scale items. But I
found that most items in fact work with 2 to 4 categories.
The question that I have is about recoding. I tried two coding schemes:
CODES = "1234567"
I expected to come up with the same results but I found out that the two
subjects estimated measures using the two coding schemes are not perfectly
correlated. In high end the same theta under Scheme 1 is associated with
little spread of measures when they are am no sure about using this coding
scheme. My question is: I s there a rule in this cases? Is there a problem
with the fact that i[Rasch] Items with various part.ems n some items the
maximum item score would be 4 while in other items would be 2, that is, with
the fact that they don s have the same maximum?
Editor, Rasch Measurement Transactions
rmt at rasch.org HYPERLINK "http://www.rasch.org/rmt/"www.rasch.org/rmt/ Latest
RMT: 19:4 Spring
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.1.375 / Virus Database: 268.8.3/360 - Release Date: 9/6/2006
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.1.375 / Virus Database: 268.9.0/368 - Release Date: 16/6/2006
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch