Intraclass correlation coefficient (ICC) with macro intracc from SAS

[Celine_France]

Good evening,

on the SAS support, I've found the macro intracc which calculates the intraclass correlation coefficient.
In books, it is written that this value is between 0 and 1. However, with some data (where judges give very different marks) and using the macro, I've found negative values.
Is there an error in the macro ? Shall I changed these values to 0 ?

Thanks a lot,

Céline

[Celine_France]

I've forgotten to precise that I work on difference of marks, which means that I can have negative values.
May it have an influence on the calculation of the ICC ?

Thanks a lot,

Céline

[statsplank]

Hi Celine_France,

No, it doesn't matter if your data values are negative.

In general, there could be situations leading to negative estimates of ICC. It depends on how variance components used to calculate ICC are estimated. I am not familiar with this macro so I can only guess that the method of moments is applied.

[Celine_France]

Thanks a lot Statsplank !

I've done some researches about this topic and if ICC is negative, it is because intra-group variance is higher than inter-group variance.

Some documents explain that if ICC is negative, we can conclude to no agreement between the raters. Do you think that I can conclude to that in my case ?

By advance, thanks a lot,

[statsplank]

I would think so... Although you should probably discuss it with somebody who has more experience in this area.

All the best to you.