RE: posthoc step

From: "Ludden, Thomas (MYD)" luddent@iconus.com Subject: RE: [NMusers] posthoc step Date: Fri, December 10, 2004 11:48 am Nick, Jerry, Leonid, Steve, Yaning et al. The objective function value for this problem is rather ugly (see tabulation below). It is generally flat around a K value of 10 but there is a minor maximum at about K=7.2-7.25. Some gradient search algorithms will have trouble with this. The Solver in Excel (obviously not the Gold Standard) gives a K estimate of 9.78 with initial estimates of 7.25 up to 10. With initial estimates of .1 to 7.2, the K estimate is 0.944. The initial estimates for NONMEM's POSTHOC ETA search are not readily accessed so I have not tested NONMEM itself but I have been in contact with Stuart Beal regarding this question. Jerry, would you please try the SAS procedure with an initial estimate of 10 with and without estimation of the residual variance? It would be interesting to see if SAS's search procedure can get past the minor maximum. initial initial Final Final K est OFV K est OFV 10 448.0646546 9.781 448.06 9 448.1637275 9.781 448.06 8 448.5035677 9.781 448.06 7.5 448.6452926 9.781 448.06 7.25 448.6697797 9.781 448.06 7.2 448.6687872 0.944 21.603 7.1 448.6595588 7 448.6393969 0.944 21.603 6.5 448.3096192 6 447.3663721 0.944 21.603 5.5 445.3349748 5 441.4403283 4.5 434.4249225 4 422.270891 3.5 401.8019287 3 368.1922873 2.5 314.6254852 2 233.1745355 1.5 121.66397 1 23.59852696 0.944 21.603 0.9 23.0073974 0.8 39.55006733 0.7 83.27305134 0.6 170.024752 0.5 325.154084 0.4 589.7760326 0.3 1031.4946 0.1 2973.924151 0.944 21.603