[Rasch] Model Standard Errors

Jun Kim ambcarpet at googlemail.com
Fri Jan 23 03:32:14 EST 2009

I utilised FACETS to capture the characteristics of written
performance of learners on ten linguistic items. FACETS provided a
developmental pattern of learners and linguistic items from the
person/item fit statistic report. Since idiosyncratic performance of
learners at particular proficiency levels is one of the main issues in
my study, I examined the bias/interaction report, too. The
bias/interaction report suggested a great number of biased
interactions between learners and items and most of the interactions
were suggested to have no statistical significance.

My concern is that the magnitude of model standard error in the
bias/interaction report is relatively big across interactions. How do
I interprete that? What should I do if the model s.e. is big?

Some suggested I could run FACETS again without misfit learners and
items. However, what I'm dealing with is not a test item, but a
performance of individual learners on a particular linguistic item and
every single misfit learner and item itself, I think, is evidence for
idiosyncratic development of learner langauge.

Would anyone please help me?

PS - I would like to ask this qustion directly to Linacre, but I
failed to obtain his active email address.

More information about the Rasch mailing list