[Rasch] Misfitting persons

Nicholas Reynolds nicholasaustinreynolds at gmail.com
Sun Dec 3 21:02:11 AEDT 2017


Hi members

I want advice about a scale with several misfitting items. I am not sure
how to modify the data and report on the findings. Could someone point me
in the right direction?

The scale is a self-report questionnaire of addiction

·         there are 13 items

·         there are four response options (strongly disagree to agree)

·         I have chosen partial credit paramaterization in RUMM2030

·         My sample size is 560; there is no sign of mistargeting and the
PSI is .85



The data are misfitting

·         The scale has four reverse worded items, all of which are
misfitting

·         The items function well-enough for a majority of the sample. But
a substantial minority of people (5-10%) responded unexpectedly to them

·         Standard residuals for these people are generally > 3 and
sometimes > 4.

·         The unexpected response behaviour is concentrated at the extreme
thresholds of these items. This is consistent with the idea that reverse
wording has caused participants to respond unexpectedly to them

I am reluctant to remove the misfitting items because they measure
important aspects of the latent attribute. Instead, I deleted about 40
participants (7% of the sample), which resolved the item misfit.

Having deleted the participants, I don't know whether to interpret the item
order without them, or to delete the items so that data fit the model, or
if there is an alternative approach I might take?

How would you recommend modifying the data and reporting on these findings?
I'd appreciate any advice or reading material.

Of course, when reporting the findings, I'll suggest that the problematic
items re-worded in future, given that I am reasonably sure about the cause
of misfit in this context.

Regards

Nick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20171203/4a19bb98/attachment.html>


More information about the Rasch mailing list