I am analyzing a proportion $y \in [0,1]$. I will eventually be conducting analysis using zero-and-one inflated beta regression.

Before I perform that analysis, I would like to quantify the reliability this measurement. 75 raters rated the same subject on two different days. I was wanting to calculate the ICC.

Does anyone have a suggestion on how to do this for $y \in [0,1]$? I could build an appropriate beta model to calculate the components for an ICC calculation... but there are 4 parameters for a zero-and-one inflated beta model. I am not sure how I could combine these elements to compute an ICC. Is there perhaps a non-parametric way calculate the ICC?


1 Answer 1


What you're trying to do is summarize how close 75 numbers (in the range $[0, 1]$) are to each other. So, the standard deviation or the mean absolute error from the median would make sense. You could also report extrema or far quantiles (say, the 5th and 95th percentile).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.