Basic Terminology 2 (Sections 3+4)
Experimental
error is caused by randomized nuisance variables, variations in the
experiment,
measurement errors, etc..
Within-group differences = experimental error
Between-group differences = experimental error + possibly treatment effects
between-group
differences expected value
------------------------------------ = 1 if there are
not treatment effects,
within-groups differences >
1 otherwise
since
(treatment effects)+(experimental error)
----------------------------------------------------
=
1 if there are not treatment effects
experimental error >
1 otherwise
Operationalization:
F = MSA
/ MSs/A
where
MSA = SSA / dfA
, MSs/A
= SSS/A / df s/A
If the treatment effect is 0, a probability distribution of F can be simulated and computed for a given number of treatments and subjects (or more precisely: for given degrees of freedom of the nominator and the denominator).