Let’s forget the Group-by-Treatment interaction, please! [Regulatives / Guidelines]

posted by ElMaestro  – Denmark, 2017-05-01 18:19 (3328 d 13:23 ago) – Posting: # 17286
Views: 34,928

Hi Hötzi and Mittyri,

this thread is interesting and confusing to me.
May I ask or comment for clarification:

M: "Is it possible to prove that with sims?" - what is it you want to prove? Can you formulate it plain and simple? Sims are totally possible, I just need to figure out the equations, as well as have a purpose.:-D

H: "It should be noted that in rare cases (e.g., extremely unbalanced sequences) the fixed effects model gives no solution and the mixed effects model has to be used." - a realistic linear model will have a single analytical solution unless you make a specification error. Imbalance would not affect that, please describe where/how you came a cross a fit which failed with the lm.

M+H: FDA are also fitting subject as fixed even when using the random statement in PROC GLM. Some of them just have not realised it :-)

H: "(...) seemingly ~⅒ of studies show a significant group-by-treatment interaction. " - this is expected by chance. You apply a 10% significance level. By chance 10% will then be significant.
(and by the way: Which denominator in F did you apply; within or between?)
—
Pass or fail!
ElMaestro

Complete thread:

UA Flag
Activity
 Admin contact
23,653 posts in 4,991 threads, 1,571 registered users;
152 visitors (0 registered, 152 guests [including 30 identified bots]).
Forum time: 07:43 CEST (Europe/Vienna)

Outside his own ever-narrowing field of specialization,
a scientist is a layman.
What members of an academy of science have in common
is a certain form of semiparasitic living.    Erwin Chargaff

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