Hi,

Kindly see the following link:

http://www.real-statistics.com/reliability/intraclass-correlation/

More precisely stated, the p-value expresses the probability of obtaining a correlation value - r (or ICC), as high as (or more than) what is computed from your sample data, given that your null hypothesis is true. If your null hypothesis is r = 0 (or ICC = 0), then you can imagine the likelihood of that being the case given your data by looking at the p-value...

ICCs are most frequently reported when it comes to inter-rater reliability. In that case the p-value is not so much of interest, because a very low ICC (say ICC = .3; indicative of low inter-rater agreement) could be significantly different from zero (small p-value).

Reporting confidence intervals or a standard error along with the ICC is much more informative in that situation.

Thank you all for your prompt responses and answers.

Dear Boris, SPSS actually gives the CIs, but one has to calculate the standard error of the mean, if applicable. Which one do you think is more informative and do they actually have a threshold for acceptable values or we only have to report them without interpretation?

Thanks again.

It is more convenient to report CIs.

Regarding the threshold: this will perhaps depend on your research question. If you have reference values from previous studies your CIs should cover the ICCs of those studies (of course, it would be also okay if your lower bound would be above those reference values).

If you did a rater-training prior to the rating process and set it as a goal to reach an ICC of, for example, .8, then the interval should be above or at least cover that value.

Thanks Boris. That's definitely helpful. I appreciate it.

one i am sure of is if p-value is less than alpha, u reject

and if p-value is greater than alpha,u accept