I have always used good old-fashioned Pearson's r to establish test-retest
reliability; i.e. correlate first administration scores with second
administration scores. This student threw a reference at me arguing that
Pearson's, which he called an *inter*class correlation, was inappropriate
for this. The author recommended repeated measures ANOVA, and then
provided a formula for estimating R from the ANOVA results. This he called
an *intra*class correlation. The formula was:
R = (MS - MS )/ MS
Subjects Residual Subjects
To add to the confusion, the student showed me an SPSS print-out with the
repeated measures ANOVA and a Cronbach's alpha. I dismissed the Cronbach's
as irrelevant, given that internal consistency was not the issue, but when
I calculated the R using the above formula by hand, it turned out to be the
same as the Cronbach's.
I have never seen this stuff before. Is this approach common? Is it
valid? What does it all mean? Any info appreciated.
**************************************************************************
* John Reece, PhD *
* Department of Psychology & Intellectual Disability Studies *
* Royal Melbourne Institute of Technology *
* Bundoora Campus *
* PO Box 71 Phone: 03-468-2512 *
* Bundoora Victoria 3083 Fax: 03-468-2303 *
* AUSTRALIA Internet/Aarnet: reece@eos.xx.rmit.edu.au *
**************************************************************************