To add to Michelle's comment regarding Kenward-Rogers degrees of freedom:  If you are on SAS/STAT12.1, try ddfm=kr2.  The algorithm has been updated and is much more stable.  However, I don't think it will help with convergence.  My take on the process flow is convergence to stable variance-covarince matrix, estimate parameters, apply denominator degrees of freedom for calculation of tests and intervals.  If you have not achieved convergence, the other stuff doesn't get done.  No reason not to apply the better estimate though, as you will get convergence eventually.  I really think the coalescing to doctor as the unit of analysis should get this done (and if not, then maybe doing this piecewise by timepoint to see what might be messing things up is in order). Steve Denham ... View more