[Rasch] computing outfit mean-squares for each rating category in eRm (PCM)

William Fisher william at livingcapitalmetrics.com
Wed Aug 10 02:41:43 AEST 2016


Good point, Rense. Continuing in that direction, why aren't instruments calibrated on samples chosen for that specific task before they are used in applications to compare outcomes? Experimental study samples are convenient, but they likely often do not provide the information on each possible score group, rating category, etc. needed for calibration. Furthermore, after an instrument has been calibrated, often more than once, why do users go on recalibrating it on their own samples, instead of checking published values against their own, reporting out the differences, if any emerge, and providing results in a standard unit comparable across applications? 

The latter is needed to follow through to networks of common metrics embodying the repeatedly emergent, self-organized autopoietic constructs we study. It's almost as though these independent forms of life have been jumping up and down and yelling at us for years, trying to get our attention and to be admitted as deserving a place in our world alongside meters, seconds, volts, grams, etc. It's oddly counterproductive, even schizophrenic, to aspire to the production of new knowledge while systematically ignoring the means within our grasp for fulfilling that aspiration. 

For more on this way of thinking, see the proceedings of the IMEKO TC-1, TC-7, TC-13 Joint Symposia of the last several years, and associated publications in Measurement and elsewhere. More info available by request. 

William Fisher

 > On August 9, 2016 at 4:54 AM Rense Lange <rense.lange at gmail.com> wrote:
> 
> 
> 
> I wonder why we don’t routinely use training and validation samples, as is standard in AI and Machine Learning - even when there are sufficient data to do so. That is, take part of your sample to fit (“train”) the model, and then use another (disjoint) subset to validate the model. If in this case the differences between models would persist in some meaningful fashion, this might be a good reason to prefer one version over the other. Depending on the application and purpose, one might select specific aspects of overall / local fit.
> 
> Rense Lange
> 
> On Aug 9, 2016, at 12:36 PM, Mike Linacre <mike at winsteps.com> wrote:
> 
> Alex,
> 
> you asked: "Should I consider other criteria when choosing between RSM and PCM?"
> 
> Please see www.rasch.org/rmt/rmt143k.htm  - Comparing "Partial Credit Models" (PCM) and "Rating Scale Models" (RSM)"
> 
> Mike L.
> ________________________________________
> Rasch mailing list
> email: Rasch at acer.edu.au
> web: https://mailinglist.acer.edu.au/mailman/options/rasch/rense.lange%40gmail.com
> 
> ________________________________________
> Rasch mailing list
> email: Rasch at acer.edu.au
> web: https://mailinglist.acer.edu.au/mailman/options/rasch/william%40livingcapitalmetrics.com


More information about the Rasch mailing list