[Rasch] Re: Rasch Digest, Vol 35, Issue 12

Thomas Salzberger Thomas.Salzberger at wu-wien.ac.at
Wed Jul 2 06:11:44 EST 2008


I am not sure whether you got a response already. 
I had to sort out my internet access, so I was offline a copule of days.

>1. For each item (or person), Rasch provides m-1 
>thresholds for the m response categories.  The 
>arithmetic mean of the m-1 thresholds for each 
>item(person) is then the item (or person) 
>location or commonly called item (or person) measure.  Am I right?

Yes, the mean of the thresholds (the tau 
parameters) is the (overall) item location 
(delta). However, for persons there is just one 
parameter, so no need to calculate a mean.

>2. If we call the item measure as "item 
>endorsability", how do we call the person measure?
I personally call them simply person measures. 
You may call it attitude or measure of attitude 
(if it attitude what you are measuring) or whatever it is.

>3. What is your idea about the infit and outfit for each measure?

Infit and outfit are provided by winsteps. I am 
not familiar with winsteps. The outfit is said to 
be equivalent to RUMM's fit residual. As it is 
the case with all fit statistics, it is not so 
much one statistic that counts but you should use 
multiple fit/misfit indicators.

>4. Relative to the instrument, how do we 
>interpret the item and person separation and reliability?
First, note that separation is defiend 
differently in RUMM and WINSTEPS, respectively.
In RUMM, person separation essentially is 
reliability using the Rasch measures. It is what 
it is in classical test theory. It tells you how 
well the instrument discriminates between the 
persons in the sample. Its sample dependence 
makes it much less relevant in Rasch analysis 
compared to classical test theory. In RUMM, it is 
used as a (rough) indication of the power of the 
test of fit (excellent, very good etc. the 
classification solely sepends on the person 
separation index). In practice, many factors 
influence the person separation. If person 
separation is very low, there are major problems 
in the data, in most cases, the items do not form 
a unidimensional scale but represent more or less 
unique ideas. In summary, the fit statistics and 
the correspondence of what you expected from the 
items and the actual item location estimates 
count more than a high person separation.

Liebe Gruesse
Thomas Salzberger
Dr. Thomas Salzberger
Österreichische Gesellschaft für Absatzwirtschaft
c/o Department of Marketing, Institute for Marketing Management
University of Economics and Business Administration (WU-Wien)
Augasse 2-6 / A-1090 Wien / Austria, Europe
Email: Thomas.Salzberger at wu-wien.ac.at
Tel: (+43-1-) 31336 4406
Personal Fax: (+43-1-) 31336 90 4406
Fax Institute: (+43-1-)  31336 732
WWW Personal: http://www2.wu-wien.ac.at/marketing/user/salzberger/
WWW Institute: http://www2.wu-wien.ac.at/marketing/
WWW The Matilda Bay Club: http://www.matildabayclub.net 

More information about the Rasch mailing list