[Rasch] How to deal with displacement/item parameter drift over time?

Rense Lange rense.lange at gmail.com
Fri Feb 23 14:12:54 AEDT 2018


Stuart,

I dont think that the difficulty should be kept constant (point 3) - after all aren’t you assuming that some items changed and the rest stayed constant? this will change their sum (mean).

I would recalibrate all data in one run (winsteps / facets can do millions), track the suspect items - this would need ‘year’ as a facet. Then i would create a year dependent copy of the shifting items (e.g. q12-2009, q12-2010, ....) and rerun. 

I think you might find little effect unless all item changes go in the same direction and/or are quite large (>> 0.5 logits) without counter changes ...

Rense Lange

Sent from my iPhone

On Feb 23, 2018, at 05:54, Stuart Luppescu <lupp at uchicago.edu> wrote:

Fellow Raschies, We have been using a set of questionnaire measures,
some since 1994. Some of the item difficulties and thresholds were
determined a way long time ago, and have been used as anchors for
measure scoring for some items since the beginning[1]. 

We have recently decided to deal with some item parameter drift that
has arisen over time. In order to preserve comparability with previous
years' results, we can't just recalibrate every measure as if from
scratch. My proposal on how to deal with this, which I would appreciate
feedback on, is as follows.
1) Identify a subset of items with large displacements and misfit
2) Let those items float, but keep the others anchored; recalibrate and
get new item difficulty values
3) Adjust all the item difficulties (add a translation constant) so the
mean item difficulty with the recalibrated items is the same as the
original mean item difficulty.

Does this sound reasonable? Will the resulting person measures be
comparable to the old ones?

Thanks for the advice.

[1] At various times, items have been added or removed, so the mean
item difficulties are not always 0. 
-- 
Stuart Luppescu
Chief Psychometrician (ret.)
UChicago Consortium on School Research
http://consortium.uchicago.edu

lupp at uchicago.edu
________________________________________
Rasch mailing list
email: Rasch at acer.edu.au
web: https://mailinglist.acer.edu.au/mailman/options/rasch/rense.lange%40gmail.com


More information about the Rasch mailing list