[Rasch] Validity in Rasch

Svend Kreiner S.Kreiner at biostat.ku.dk
Fri Mar 25 21:57:50 EST 2011


I agree completely with Mike's classification. Statistically 
unacceptable fit may indicate that there can be a measurement problem, 
but not that there actually is a problem. I would say, that we have a 
substantially-good-enough fit if measurement is robust to the model 
errors. If there, for instance, is DIF for one out of 30 items I would 
expect that measurement by the Rasch model is relatively robust. If 
there is DIF among 20 out of 30 items, I would expect that measurement 
is not robust. In both cases, however, I would attempt to assess the 
robustness of measurement to the misfit before I decide what to 
conclude. I would never assume that my expectations are correct.

Concerning overfitting data, e.g. data where infits are much closer to 1 
than expected by the random component of Rasch models, I would in most 
cases do the same as Mike. It happens, but not very often.

Svend

Den 24-03-2011 06:02, Mike Linacre skrev:
>
> Svend writes: What we may disagree about is what to do when items do 
> not fit the model.
>
> Here, an extra category of fit of the data to the model is helpful:
>
> 1. Statistically-acceptable fit
> 2. Substantively-good-enough fit
> 3. Unacceptable fit
>
> The criteria for good-enough fit depend on the situation in the same 
> way as industrial-quality-control tolerance limits depend on their 
> situation. So, before rejecting an item (or a person, a rater, ...), 
> we need to decide "what amount of misfit is acceptable?" For instance, 
> in rated performances, we may decide that observations that overfit 
> the Rasch model (too predictable) are always acceptable.
>
> Mike L.
>
> Mike Linacre
> Editor, Rasch Measurement Transactions
> rmt at rasch.org _www.rasch.org/rmt/_ <http://www.rasch.org/rmt/> Latest 
> RMT:  24:4 Spring 2011
>
>
> _______________________________________________
> Rasch mailing list
> Rasch at acer.edu.au
> Unsubscribe: https://mailinglist.acer.edu.au/mailman/options/rasch/s.kreiner%40biostat.ku.dk

-- 
Svend Kreiner
Professor
Department of Biostatistics
University of Copenhagen

Øster farimagsgade 5, entr. B
P.O. Box 2099
DK-1014 Copenhagen K, Denmark

Email: S.Kreiner at biostat.ku.dk
Phone: (+45) 35 32 75 97

Fax: (+45) 35 32 79 07

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20110325/06302b7c/attachment.html 


More information about the Rasch mailing list