09-23-2011 10:37 AM
does somebody know how the npar1way procedure handles tied values for example using the hl statement to calculate Moses intervals.
The treatment difference has some tied values, I know npar1way does something to them, but unfortunately this is not documented (or I didn't find it).
I appreciate any help, thanks in advance.
09-23-2011 10:41 AM
I know that average ranksums are used instead of absolute ranks, but i'd need more exact information, how handling of tied values influences the formula (e.g. of the Moses intervals)
09-26-2011 04:52 AM
The documentation of the Hodges-Lehmann estimation of location shift (and Moses intervals) is quite incomplete regarding the formula for the variance if tied values occur. There is only one formula (C-alpha = ...) documented, this is for untied values only. My question is, what is npar1way doing exactly if there are tied values?
09-26-2011 08:42 AM
Does the following provide an explanation that is sufficient for your needs?:
09-26-2011 06:50 PM
From what I'm reading cccording to the documentation and formula's there is NO correction for tied values.
You can test this by running for some data that has no ties and then adding in some tied values around the mid range (this shouldn't change the CL but will change the median) from what i'm seeing.
Hopefully that makes some sense.