[Rasch] DIF Question

Looveer, Juho Juho.Looveer at det.nsw.edu.au
Fri Mar 16 23:31:07 EST 2007

Not so familiar with Winsteps.
However, have you considered that the DIF may not be consistent or uniform?  

Mantel-Haenszel, like many other earlier ways of assessing DIF, assumes that the DIF is uniform, and thus that one number can describe the DIF entirely.  This is not true when non-uniform DIF is present.  This can most easily be seen from the DIF curves (as produced by RUMM, also other software), which will cross or behave in unexpected ways.  There may be simple crossing DIF (i.e. one group performs "better" than the other for one range of abilities, but the reverse occurs for the other); or the two groups may perform at similar levels for some ability levels, but differently at other ranges.

Apart from more erudite comments from people more experienced with Winsteps, this may be something to investigate.



Dr Juho Looveer
Manager, Data Collection
Planning and Innovation
NSW Department of Education and Training
Level 5, 35 Bridge St Sydney
GPO Box 33
Sydney NSW 2001
work phone: 956 18192
work fax: 956 18055
Juho.Looveer at det.nsw.edu.au 

-----Original Message-----
From: rasch-bounces at acer.edu.au on behalf of Eric Wong
Sent: Fri 16/03/2007 8:50 PM
To: rasch at acer.edu.au
Subject: [Rasch] DIF Question
Dear all,

I used WINSTEP to run DIF on 12 (ordinal) unidimensionality items

DIF class specification is: DIF=@SMOKE



| PERSON   DIF   DIF   PERSON   DIF   DIF      DIF    JOINT                MantelHanzl ITEM                                                  |

| CLASS  MEASURE S.E.  CLASS  MEASURE S.E.  CONTRAST  S.E.   t  d.f. Prob. Prob.  Size Number  Name                                     


| 1         .30   .16  2         .77   .09      -.47   .18 -2.56 376 .0108 .1653  -.24      3 Had angina that affected life                  |

| 2         .77   .09  1         .30   .16       .47   .18  2.56 376 .0108 .1653   .24      3 Had angina that affected life                  |

The Mantel-Haenszel DIF has probability .1653 and size .24 logits, and 
The Winsteps-Rasch equivalent is DIF Contrast = .47 logits, probability = .0108.

Any idea why the p-values and the DIF sizes disagree?  What conclusion could I make?

Many thanks!


This message is intended for the addressee named and may contain
privileged information or confidential information or both. If you
are not the intended recipient please delete it and notify the sender.

More information about the Rasch mailing list