Talk:Intraclass correlation

Error in the article[edit]

In the ANOVA model, the ICC is not bounded below by zero, it is bounded below by 1/(1-r) in the case of equally replicated groups. 128.123.198.142 (talk) 23:04, 26 February 2013 (UTC)[reply]

That error in the model should be corrected. I see one "meaning" properly described at

http://www-personal.umich.edu/~gonzo/course/new.html

You get a similar dependency any time the scores for every subject/group sum to (nearly) the same total. This arises in nature, and the negative ICC might be your clue to it. RichardFloyd (talk) 07:46, 11 February 2016 (UTC)[reply]

Needs a clearer definition[edit]

This article needs to be more accessible to a general audience. In addition to what has already been written, it needs a simlified, clear definition with at least one really clear and concrete example. —Preceding unsigned comment added by 173.52.5.43 (talk) 02:23, 15 December 2008 (UTC)[reply]

Suggestion to review change[edit]

The article used the word ingressingly, which to my knowledge is not a word, or is so arcane as to be meaningless to most readers. The sentence affected beings "as the number of groups grow" and I put in the word exponentially. However, I do not know matrix math so someone should review that this is correct 198.190.230.61 (talk) 14:46, 19 November 2007 (UTC)[reply]

SPSS[edit]

I don't think it's entirely correct to say that the ICC can be calculated from SPSS output. SPSS contains (potentially dubious) algorithms for the calculation of ICCs and this can be viewed in the output. —Preceding unsigned comment added by 99.233.88.19 (talk) 23:49, 17 April 2008 (UTC)[reply]

What do you mean potentially dubious? One thing I will say is that the output is confusing, but the calculations should be sound as I believe they are based on the work by McGraw and Wong's 1996 paper. Do you have any evidence to support this? 194.83.142.156 (talk) 11:31, 6 March 2009 (UTC)[reply]

Perhaps 99.233.88.19 didn't click on the statistics button within the scale section and turn on ICCs. If you don't do this you just get Cronbach's alpha which is not an ICC. An example of how to do this is here http://www.uvm.edu/~dhowell/StatPages/More_Stuff/icc/icc.html. I'm sure SPSS have used the correct algorithms as the software is used in research and industry. 194.83.139.34 (talk) 10:14, 10 March 2011 (UTC)[reply]

Error "Early definition" section[edit]

In the definition for more than 2 values per group the article says "As the number of groups grows, the number of cross-product terms in this expression grows rapidly." I think this should read "As the number of values per group grows, the number of cross-product terms in this expression grows rapidly.", because the article has been talking about N groups all the time, only the number of values within groups is growing here.

Any comments?

Arnoldma (talk) 12:00, 11 March 2011 (UTC)[reply]

I believe you are correct. I have changed it. Skbkekas (talk) 15:48, 11 March 2011 (UTC)[reply]

Plots captions[edit]

It'd be useful to know what method was used to compute the ICC coefficients shown in both plots at the beginning of the article. — Preceding unsigned comment added by 140.105.70.136 (talk) 15:38, 4 December 2012 (UTC)[reply]

The answer to this appears to be in the image page which has matlab code describing how this was done. Specifically the code contains:

def ICC(X):
   K = X.shape[1]
   m = X.mean(1)-X.mean()
   A = (m**2).mean()/X.var()
   A *= K/(K-1)
   A -= 1/(K-1)
   return A

I'm not sure exactly what type of ICC this is 194.83.141.96 (talk) 16:54, 20 December 2012 (UTC)[reply]