[LON-CAPA-users] KR-21

Peter Riegler lon-capa-users@mail.lon-capa.org
Fri, 22 Feb 2008 21:32:43 +0100


I have run across the Kuder-Richardson statistic in loncapa. Does 
anybody have experience with that?

As I understand it is a measure of reliabilty of a test.

I have computed this statistic for two assessments. In both tests scores 
  per test item are either 0 or 1.

Assessment 1: 22 test items
fall 2006 kr-21=0.82
spring 2007 kr-21=0.81
fall 2007 kr-21=0.82

Assessment 2: 27 test items
fall 2007 kr-21=0.73
spring 2008 kr-21=0.74
The last assessment has been retaken after two weeks of additional 
training, resulting in kr-21=0.76 (with better overall results).

What do these data tell me?
Is assessment 1 considerably more reliable than assessment 2?
An what does it mean that an assessment is less reliable?
I guess that low reliability means that one will likely get a different 
distribution of scores if one repeats the assessment.
If I did my algebra correctly, kr-21=1 for coin flipping. The maximum 
value of kr-21 is k/(k-1) ~ 1. So, if I want to have reliable 
assessments I ask my students to flip a coin???


Peter Riegler
Fachhochschule Braunschweig/Wolfenb├╝ttel
Salzdahlumer Str. 46/48, 38302 Wolfenb├╝ttel
Tel. +49 5331 939 6314, Fax. +49 5331 939 6002
email: p.riegler@fh-wolfenbuettel.de