[Rasch] When local dependence is part of the construct

Mike Linacre mike at winsteps.com
Sun Jun 29 21:04:25 EST 2014

Thank you for your question about local dependency, Purya.

If the local dependency is roughly homogeneous throughout the data, then 
it reduces the stochastic background noise. Its effect is to spread the 
logit estimates a little wider. This is unlikely to cause problems in a 
practical situation. A parallel situation is an achievement test given 
under high-stakes and low-stakes conditions. The high-stakes data tends 
to be less noisy (more Guttman-like), so that the spread of the logit 
estimates is wider for the high-stakes data than for the low-stakes data.

If the local dependency makes groups of items cluster together, then 
each cluster of items can be summed into a super-item. These super-items 
are analyzed with the partial-credit model. If you do this, plot the 
ability estimates from the original dichotomous data against the 
estimates from the PCM analysis to see the effect of the local 
dependency on the ability estimates. Do tell us what you see in the 
plot, Purya.

Mike L.

On 6/29/2014 16:58 PM, Purya Baghaei wrote:
> Dear Colleagues,
> In a cloze test (a passage where every n^th word is deleted) items, 
> i.e. gaps within the text, are locally dependent. This is a violation 
> of the assumption of Rasch model and theoretically a cloze test cannot 
> be scaled with the RM.
> The point is that local dependence in this case is part of the 
> construct. That is, the ability to integrate information from 
> different parts of the text and using contextual clues to solve the 
> items are higher order language abilities that we want to measure with 
> a cloze test.
> Any ideas or comments on analyzing cloze test with the Rasch model?
> Regards
> Purya

Mike Linacre, mike at winsteps.com or winsteps1234 at gmail.com
Winsteps: www.winsteps.com/winsteps.htm - current version 3.81.0 - 
February 2014
Facets: www.winsteps.com/facets.htm - current version 3.71.4 - January 2014
Winsteps and Facets Tutorial PDFs at www.winsteps.com/tutorials.htm
Your comments and questions are welcomed.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20140629/99e403d5/attachment.html 

More information about the Rasch mailing list