[Rasch] Fwd: Reliability, separation and strata

David Andrich david.andrich at uwa.edu.au
Wed Mar 1 23:35:56 AEDT 2017

Fabio. I do not think you should use only one method no matter what - it depends on what question is you want to answer. It is a pity that the same word "reliability" , which implies something good if you have it, is used for multiple indices. The traditional reliability answers a question as to how well the test you have has separated the people you have. Sometimes that is useful, for example, if you have a low reliability because the people do not cover a wide range, then you know that you will have low correlations with other variables. Sometimes you might want to relate the values to other variables, and then the question arises what to do about - deliberately get a sample that has a higher spread, increase the number of items, and so on. However, the traditional reliability should be used only after you are convinced that you do not have artificial floor and ceiling effects in the data. Like all indices, the assumptions behind them need to be considered before interpreting their values.
Hope that helps

David Andrich, BSc MEd W.Aust., PhD Chic, FASSA
Chapple Professor david.andrich at uwa.edu.au
Graduate School of Education
The University of Western Australia
M428, 35 Stirling Highway, Crawley,
Western AUSTRALIA, 6009
Telephone: +61 8 6488 1085;   Fax: +61 8 6488 1052
Pearson Psychometric Laboratory

From: Rasch [rasch-bounces at acer.edu.au] On Behalf Of Fabio La Porta [fabiolaporta at mail.com]
Sent: Wednesday, 1 March 2017 8:12 PM
To: rasch at acer.edu.au
Subject: Re: [Rasch] Fwd: Reliability, separation and strata


Considering that Ben's method makes the calculation of reliability indipendent from the sample distribution, I think that it is much closer to the Rasch philosophy than the typical  methods for  calculating reliability employed by Rasch softwares, all of which assume a normally distributed sample.

 I wonder why it's being used so little.

Definitely, I will use only this from now on.

Thank you very much.

Il giorno mer 1 mar 2017 alle 11:14 Mike Linacre <mike at winsteps.com<mailto:mike at winsteps.com>> ha scritto:

Ben Wright's method does not use the sample of persons in its computation. The distribution of the sample of persons is irrelevant to
his method.

Mike L.

On 3/1/2017 7:35 PM, Fabio La Porta wrote:
> Just the last question: the Ben Wright's method can work also on a
> distribution of persons approaching normality?
> I am asking this because I like very much the possibility to define
> graphically the ranges of the statistically different levels of
> performance along the measurement continuum.
> On the other hand, I have always find it difficult to draw strata on
> the measurement continuum, as they have, almost always, decimal
> values, making the whole process rather inaccurate.

Rasch mailing list
email: Rasch at acer.edu.au<mailto:Rasch at acer.edu.au>
web: https://mailinglist.acer.edu.au/mailman/options/rasch/fabiolaporta%40mail.com

More information about the Rasch mailing list