Sample size calculation for ICC

[Margaret]

Hello

I am currently considering the calculation of an intra-class
correlation coefficient (ICC) for a two-way mixed model in which the
data are
on a Likert scale (from 1 to 6) and subjects are a random effect whilst
assessors are a fixed effect. I would therefore be interested to
receive suggestions as to the most suitable formula(e) for performing
an
associated sample size calculation based on statistical power. I am
particularly concerned that the assumptions for the sample size
calculation
should be in agreement with the sort of data that I have, as I am
already aware that there are a variety of formulae for predicting
sample
size calculations for ICCs based on assumptions relating to Normality.

I look forward to receiving some useful suggestions.

Many thanks in advance for your kind consideration of this request.

Margaret

[damian mcentegart]

you're probably aware of the paper by Walter in statistics in Medicine
in 1998. Vol 17, 111-120 1998. This deals with normally distributed
data for one way anova. I look forward to seeing advances on this for
your situation

damian

[BXC (Bendix Carstensen)]

I am inclined at this point to point out that computing power these days
is such that formulae for sample size, power and precision are obsolete:
Simulate a large number of datasets, analyse each of them and derive the
relevant power/precision for the various scenarios (in casu, sample
sizes).

Yes, it requires that your computer work while you are at lunch, and it
has the terrifying prerequsite that you have to be very explicit about
your assumptions. The former I consider irrelevant, the latter an
advantage.

Best,
Bendix
----------------------
Bendix Carstensen
Senior Statistician
Steno Diabetes Center
Niels Steensens Vej 2
DK-2820 Gentofte
Denmark
tel: +45 44 43 87 38
mob: +45 30 75 87 38
fax: +45 44 43 73 13
b...@steno.dk
www.biostat.ku.dk/~bxc
----------------------

[Margaret]

Dear Bendix

You clearly have considerable experience in an area which I have not
even dabbled in as yet.

Any suggestions on how to get started would be appreciated.

Thank you for replying.

Regards

Margaret

[Margaret]

Dear Damian

Thank you for kindly taking the time to reply.

For your interest, I found the book 'Health Measurement Scales' by
Streiner and Norman a useful reference for predicting sample size in
terms of the standard error of the reliability coefficient.

Best wishes

Margaret