[Rasch] logistic regression vs UCON

Stuart Luppescu s-luppescu at uchicago.edu
Wed Feb 3 09:22:48 EST 2010

On Tue, 2010-02-02 at 16:01 -0600, Rense Lange wrote:
> I have a situation where I already have person's location estimate,
> and where an additional question with an unknown location estimate is
> answered as well. So, across a sample I have a continuous variable
> together with, in the case of binary items, 0 or 1's, and I want to
> estimate the item's location.
> I don't want to re-estimate everything simultaneously (e.g., using
> Winsteps), but I want to do something incremental. My first thought is
> to logistic regression, and using simulated data I found that this
> works quite well (i.e. the recovered item locations resemble the ones
> used to generate the data).

Why not just use Winsteps with the item difficulties anchored?

> My problem is that (a) I'd like to use something simpler (and more
> efficient to compute), and (b) I'd like a methods that also works for
> rating scale type data. Perhaps, I should just implement UCON?

I have SAS code for implementing UCON. It's for dichotomous data but you
could easily modify it. Let me know if you want me to send you a copy.

Stuart Luppescu -=- slu .at. ccsr.uchicago.edu        
University of Chicago -=- CCSR 
才文と智奈美の父 -=-    Kernel 2.6.30-gentoo-r5                
Willow: Everything seems normal. Not a snake, not a
 wasp. Cordelia: Yup. School can open tomorrow.
 Xander: Explain to me again how that's a good
 thing? Cordelia: I'm drawing a blank. 

Please consider the environment before you print

More information about the Rasch mailing list