42! SASophylistic power oracle [R for BE/BA]

posted by d_labes  – Berlin, Germany, 2009-09-29 13:37  – Posting: # 4267
Views: 26,007

Dear guys,

let me take part in your hairsplitting party :-D .
Hope you have already waited for me.

Here are the results of the SAS Proc Power.
Code can be found here, of course appropriately adapted.

                 The POWER Procedure
        Equivalence Test for Paired Mean Ratio

               Fixed Scenario Elements

       Distribution                   Lognormal
       Method                             Exact
       Lower Equivalence Bound              0.8
       Upper Equivalence Bound             1.25
       Alpha                               0.05
       Geometric Mean Ratio                   1
       Coefficient of Variation           0.075
       Correlation                            0

              Computed Power
            Index    Pairs    Power

               1        3    0.604
               2        4    0.867
               3        5    0.968
               4        6    0.993

The power is given with 3 decimals only by default. No simple way to change.

Seems the power is calculated in this case totally wrong!
This is astonishing enough, because the same Proc Power code can reproduce most of the sample sizes given in Diletti et.al., at least for power 80%.
Ok, Proc Power is called experimental by SAS, usually means the user shall do the testing and report bugs. Others call such software beta.

My own rolled code (relying on Owen's Q-functions, undocumented but available in SAS) gives:
        Coeff. of
        variation   'true'  lower BE  upper BE        achieved   required
 alpha     (%)      µ(T/R)    bound    bound     N     power      power

  0.05     7.5        1       0.8      1.25      4     0.729014    0.7
  0.05     7.5        1       0.8      1.25      6     0.986992    0.7
  0.05     7.5        1       0.8      1.25      8     0.999577    0.7



Complete thread:

 Mix view
Bioequivalence and Bioavailability Forum |  Admin contact
19,486 posts in 4,135 threads, 1,334 registered users;
online 6 (0 registered, 6 guests [including 4 identified bots]).
Forum time (Europe/Vienna): 23:18 CEST

If you don’t know anything about computers,
just remember that they are machines that do exactly what you tell them
but often surprise you in the result.    Richard Dawkins

BEBAC Ing. Helmut Schütz