Degrees of freedom in Proc Mixed

[deleted_user]

The documentation on Proc Mixed is very thorough when it comes to describing methods for calculating denominator degrees of freedom (DDFM=CONTAIN, or BETWITHIN, or RESIDUAL, or SATTERTH, or KENWARDROGER), but it doesn't give any guidance on which method to use and when. Over the years I've used Proc Mixed in the pharmaceutical industry, this is all I've been able to gather on this topic:

- At a previous job, the conventional wisdom was DDFM=KENWARDOGER is preferable, but there was nothing concrete to support this statement.
- The FDA's guidance on individual bioequivalence (2001) provides sample Proc Mixed code which uses DDFM=SATTERTH. Is there anything more current?
- At one time, for _nonlinear_ mixed effects models, Doug Bates eschewed computing the denominator degrees of freedom altogether, and instead relied on bootstrapping for inference.

Any insights that anyone might have on this issue would be very welcome. Thanks!

ZardozJr

[deleted_user]

I'm not convinced that this belongs in this discussion group. The SAS forums should concentrate on the use of SAS to produce an outcome, and the areas in which the SAS procedure is not delivering the required result. I see it as an extension of the software manual that tells you how to use the software, not why a given model is selected. There are some exceptions, especially in the Stat area, but the selection of method and interpretation of results really belongs in a general statistical discussion group. I think that is where your question belongs. Faced with a similar dilemna, I would be looking for a stats discussion group. I know that University of Georgia runs a series of discussion list servers where you can read (listen) to the opinions of many statistically dexterous minds and I think that is where I'd go if I wanted an opinion based on statistical theory, and not on utility of a given software package.

A search for "stat-l" and "uga" will find you the address for the Georgia discussion list server. There are also other possible places you'll find by searching on "STAT" and "LIST" or "Discussion".

Having said that, if you're doing pharms work, then follow the path required by the FDA. While Satterthwaite may seem to you to be flawed or outdated, if the FDA want that for bioequivalence then that's what you give them.

Kind regards

David