[Rasch] Why isn't the person separation better?

Stephen Humphry stephen.humphry at uwa.edu.au
Wed Sep 1 22:12:50 EST 2010

Trudy, I may be missing something in your post, but good for a given separation in no way implies the separation should be high. The separtion depends on targeting, actual separation (as Tom said), and the discrimination at thresholds.

If you want to improve the separation index, systematically remove items with low threshold discrimination, which will tend to have higher Outfit. Items that previously had a higher discrimination may then have an acceptable discrimination, and the separation will tend to increase. This, in my view, is what should be done routinely rather than treating the average (roughly speaking) discrimination of the initial item set as the reference and omitting items with high (as well as low) discrimination. To fit the Rasch model, the criterion is uniform discrimination, not uniform discrimination that is necessarily equal to the initial 'average' discrimination.

Having said that, you don't necessarily want the separation too high--it can indicate unwanted dependencies from response set, halo effect or other such response tendencies.



From: rasch-bounces at acer.edu.au [rasch-bounces at acer.edu.au] On Behalf Of Tom Bramley [Bramley.T at cambridgeassessment.org.uk]
Sent: Wednesday, 1 September 2010 3:58 PM
To: Rasch Listserve
Subject: Re: [Rasch] Why isn't the person separation better?

Dear Trudy,
Perhaps the problem is simply that there is not enough 'true' variance in your sample of participants.  What is the range and standard deviation of the raw scores (which I assume are on a scale of 0-64)?

Tom Bramley
Assistant Director, Research Division
Assessment Research & Development

Cambridge Assessment
1 Regent Street, Cambridge, CB2 1GG
Direct Dial: 01223 553985

Cambridge Assessment is the brand name of the University of Cambridge Local Examinations Syndicate, a department of the University of Cambridge. Cambridge Assessment is a not-for-profit organisation.

From: rasch-bounces at acer.edu.au [mailto:rasch-bounces at acer.edu.au] On Behalf Of Stephanou, Andrew
Sent: 31 August 2010 23:56
To: Rasch at acer.edu.au
Subject: [Rasch] FW: Why isn't the person separation better?

Forwarded to the Rasch listserv on behalf of Trudy Mallinson, trudy.mallinson at usc.edu

-----Original Message-----
From: mallinso at usc.edu [mailto:mallinso at usc.edu] On Behalf Of Trudy Mallinson
Sent: Wednesday, 1 September 2010 3:00 AM
To: rasch
Subject: Why isn't the person separation better?

Dear Listserv,

I am analyzing a small data set (about 200 participants, both admission and discharge currently in the dataset) and 32 items.  Everything about the instrument looks good - the items mostly fit, the rating scale (3 points) mostly works. The range of persons seems to form a fairly normal distribution, items are targeted on the people (mean -0.28 logits) with good range (min -4.07 logits, max 3.55 logits).  But the person separation is less than stellar at .84.  There are a few items where the middle category of the rating scale is "submerged" but not disordered.  When I combine it with one of the other categories, the SD increases a little but so does the error so it's a wash in terms of the separation.  I tried removing the misfitting items (not very misfitting at 1.3 - 1.4 infit MNSQ) same thing, increased SD, increased error, no change in separation.  I wondered if it could be a dimensionality problem - the PCA suggests it's unidimensional though.  I separated out the items into two groups.  Item psychometrics continue to look OK.  Separation for the vision items .82, for tactile items .82.  I created an xy plot of the person measures from the two set so items.  Scores clearly track together although the shape of the coordinates is basically a parallelogram.

Can any one suggest why, when everything about the items looks good, the separation is not great, and all the things I've tried don't seem to improve it in any meaningful way?

Thanks for your thoughts,


Trudy Mallinson, PhD, OTR/L, NZROT

Assistant Professor

University of Southern California

Division of Occupational Science and Occupational Therapy

1540 Alcazar Street, CHP 133, Room 101F

Los Angeles, CA 90089-9003

PH: (323) 442-2950

trudy.mallinson at usc.edu

Click here<https://www.mailcontrol.com/sr/wQw0zmjPoHdJTZGyOCrrhg==> to report this email as spam.

If you are not the intended recipient, employee or agent responsible for delivering the message to the intended recipient, you are hereby notified that any dissemination or copying of this communication and its attachments is strictly prohibited. If you have received this communication and its attachments in error, please return the original message and attachments to the sender using the reply facility on e-mail. Internet communications are not secure and therefore Cambridge Assessment (the brand name for the University of Cambridge Local Examinations Syndicate, the constituent elements of which are CIE, ESOL and OCR [Oxford Cambridge and RSA Examinations is a Company Limited by Guarantee Registered in England. Registered office: 1 Hills Road, Cambridge CB1 2EU. Company number: 3484466]) does not accept legal responsibility for the contents of this message. Any views or opinions presented are solely those of the author and do not necessarily represent those of Cambridge Assessment unless otherwise specifically stated. The information contained in this email may be subject to public disclosure under the Freedom of Information Act 2000. Unless the information is legally exempt from disclosure, the confidentiality of this email and your reply cannot be guaranteed.

This message has been scanned for viruses by BlackSpider MailControl<http://www.blackspider.com/>

More information about the Rasch mailing list