[Rasch] Advice needed on analysis

matt.schulz at act.org matt.schulz at act.org
Fri Dec 2 08:35:00 EST 2005

OK.  I'll bite.

I would analyze the standard and essential versions separately, then equate
on the 20 items.  I would look at the stability of the linking items
(analytically and graphically via a paired comparison plot), with
particular attention to the possibility that items representing certain
areas of content or skill, or position in the instrument, were relatively
harder or easier on one version than on the other.

Assuming you can link the two forms satisfactorily, you can simply do a
t-test for independent samples (or one-way ANOVA) to assess  differences in
person measures between standard and essential versions.

If "time" is a variable that can be carried along in the person-ID field,
you can use the person output files as input in an analysis of the
relationship between time and person measures (regression, ANOVA,
whatever).  You may or may not want to do this analysis separately by
version.  You might want to disentangle the effects of time and version on
person measures with a general linear models analysis treating version as a
class variable and time as parametric.

Also, if you discover any interesting patterns in your initial study of
paired item calibrations, you might want to see if they are a function of
time, version, or measure.  This may involve creating samples matched
separately on time and total score on the linking items and looking at
paired comparison plots for all combinations.  Some fancy analyses of
residuals are also a possibility.

Matthew Schulz

More information about the Rasch mailing list