Braveheart! [R for BE/BA]

posted by Helmut Homepage – Vienna, Austria, 2020-07-13 14:16 (32 d 12:47 ago) – Posting: # 21685
Views: 4,932

Hi ElMaestro,

» The solution I posted is a total winner. :-D:-D:-D Where do I collect my medal? :-D:-D:-D

Hey, it looks great indeed! However, that’s only the foundation of the house. Walls, a door, windows, and a roof would be nice.

» Will be happy to learn of performance on other datasets, …

Try my code with
path <- "https://raw.githubusercontent.com/Helmut01/replicateBE/master/inst/extdata/DS04.csv"
That’s an  incomplete  balanced data set (Cmax of Table II).1

PHX (convergence after 8 iterations without warnings, -2REML 292.14, AIC 314.14):
lambda(1,1) 0.43651691
lambda(1,2) 0.63844444
lambda(2,2) 0.41685382
s2wR        0.31426995
s2wT        0.00718607

Manual:
lambda(1,1)^2=var_bR=0.1905470

My master’s code:
  Component  Estimate     Initial
      var_T 0.5885621 0.591211271
     var_bR 0.1905466 0.008845665
     var_wR 0.3142704 0.318270251
 covar_bTbR 0.2786906 0.300028468

Not that bad. ;-)

» … in particular of differences between results or non-convergence (which is often something related to instability caused by the initial guess being too far from the optimum).

See this post with links to John’s nasty datasets. Will see whether I still have the one which failed completely both in SAS and PHX.

» A slight change this morning, for the initial guesses, less clumsy and more likely (in the nonrestricted fashion ;-) ) to reflect the optimal solution:

Will check that later.2


  1. Patterson SD, Jones B. Viewpoint: observations on scaled average bioequivalence. Pharm Stat. 2012;11(1):1–7. doi:10.1002/pst.498.
  2. Edit: This time a really imbalanced one (1 subject in RRT and 1 in RTR removed):
    https://bebac.at/downloads/ds02a.csv
    Gives:
    Maximum         : 70.14324
    Evaluations     : 269
    Convergence code: 0
      Component   Estimate    Initial
          var_T 0.06800845 0.06571035
         var_bR 0.03821576 0.03199571
         var_wR 0.01065516 0.01009356
     covar_bTbR 0.04412955 0.04885303
    PE: 1.031329

    In PHX (Warning: Newton's algorithm converged with modified Hessian. Output is suspect.):
    s2wR 0.010655215
    PE: 1.0313293


    With your modified code and cosmetic tweaks:
    Nelder-Mead optimization (39 seconds)
      Final       70.14324
      Evaluations 283
      Successful completion.

      Component   Estimate    Initial
          var_T 0.06800851 0.06571035
         var_bR 0.03821582 0.03750818
         var_wR 0.01065521 0.01009356
     covar_bTbR 0.04412958 0.05160927

          Parameter     Value
                swR 0.1032241
               CVwR    10.35%
     Point estimate  1.031329


    [image]

Dif-tor heh smusma 🖖
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

Activity
 Admin contact
21,026 posts in 4,382 threads, 1,460 registered users;
online 16 (0 registered, 16 guests [including 12 identified bots]).
Forum time: Saturday 03:04 CEST (Europe/Vienna)

It has yet to be proven
that intelligence has any survival value.    Arthur C. Clarke

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5