Even my calculator... [Bioanalytics]

posted by ElMaestro  – Denmark, 2009-01-23 12:12 (4468 d 18:22 ago) – Posting: # 3115
Views: 20,934

(edited by ElMaestro on 2009-01-23 12:58)


Ohlbe and HS you seem to be having great fun here.
Let me add, I know of a famous, fantastic, brilliant and heavenly piece of software for 2,2,2-BE calculation in which the programmer circumvented problems of this type by scaling all input data* before the model fitting and ANOVA takes place.

There is one potential drawback of doing that, however: In a model fit you need differentials of the dF/dX type, possibly also double differentials. In order to do that the value of the delta (d above) should be chosen VERY wisely. It should in theory be infinitely small, but computers of course cannot do that, so we try e.g. 10e-4 or 10e-6 or whatever. But if the scaling causes the numbers in the input dataset to be of a magnitude close to the delta then everything could be screwed up. The pros and hardliners use Al Gore Rythms with variable delta's in their matrices and differentials. The programmer behind the software mentioned above didn't do that kind of fancy stuff, mainly because of the log operation.
Would still perhaps be interesting to see if a BE-script in R or SAS can be challenged by datasets with very large or very small numbers (perhaps even worse: When the range of values in the input dataset is narrow). Since I don't have SAS I cannot do it all, but I might try a few R challenges one of these days.


*: All data divided by the highest data entry in the entire dataset. All values then between 0 and 1.

Complete thread:

 Admin contact
21,419 posts in 4,475 threads, 1,510 registered users;
online 16 (0 registered, 16 guests [including 4 identified bots]).
Forum time: Monday 07:35 CEST (Europe/Vienna)

There is one certainty in drug development
and statistics that one can depend on:
the data are always late.    Scott Patterson and Byron Jones

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