RE: posthoc step

From: Vicente.Casabo@uv.es Subject: RE: [NMusers] posthoc step Date: Thu, December 16, 2004 7:57 am Dear all very interesting discussion. I have enjoyed it very much. in Jerry's SAS code, if you are trying to estimate the Residual SD as he did, values of plus minus infinity are not weird at all, are obvious. should not you code a penalty function in that case? I do not use SAS, but in the excel file from Nick you just have to add the 2*LN of RUVsd to OLS (when you are estimating aleatory effects you required extended least squares). (If you recode the OLS in that way and add a restriction to RUVsd of being >1*10-8 to avoid negative or zero values for RUV) Using initial estimates of K from 10 to 1.8 solver converges to K=10 and SD=2.44 initial estimates K=0.1 and sd 0.2 lead to K 0.808 and sd0.4 initial estimates k=0.5 to 1.7 and sd=0.2 lead to K=0.94 and sd=0.12 initial estimates from K=1.8 and higher lead as I mentioned above to K=9.9988 and SD=2. Vicente G. Casabo Associate Professor University of Valencia Spain