[Rasch] (no subject)

Mike Linacre (RMT) rmt at rasch.org
Sun Oct 23 10:45:45 EST 2005


Dear Michael et al.

With numbers inserted into your statements, test-sample targeting, item 
S.D., sample S.D. and item structure (polytomy, dichotomy), then the 
expected contribution of person variance, item variance, and random 
variance to the observations can be computed.  The effects of data-model 
misfit can also be predicted.

If person variance=0 and item variance=0, then all variance in the data is 
"random" (this is tossing coins).

If person variance is huge, and item variance is huge, then nearly all 
variance in the data is predicted to be due to those variances.

OK?

MIke L.


At 09:53 PM 10/20/2005 -0400, Michael Lamport Commons wrote:

>Let us say there are two Rasch analysis of different data.  One has a very 
>tight participant standard deviation around the mean and the other has a 
>one that is about 5 times as large.  Both have the same item distribution.
>
>Could one infer any of the following:
>
>We would expect sampling effects to be smaller in the first case than the 
>second.
>
>We would say the most of the variability in the data was due to the items 
>in the first case but not in the second.




More information about the Rasch mailing list