In article <[EMAIL PROTECTED]>,
Steve <[EMAIL PROTECTED]> wrote:
>The domain of interest is Psychology/Behavioral Sciences. The problem is
>that many researchers will examine data for group mean differences across
>factors that CANNOT be randomly assigned. This seems to violate a basic
>assumption of ANOVA- random assignment to treatment groups.

Not necessarily.  Randomization can be restricted (e.g. blocked designs).

If absolutely *all* the assumptions of normality, linearity, form of
the mean, etc. etc. hold, then the distributional results for the ANOVA
hold even if you don't randomize.  Randomization protects you against
failures of these assumptions.  (The quotation usually attributed to
George Box that 'All models are wrong, but some are useful' is worth
remembering.)

You might find it interesting to read some of the early fundamental
literature in experimantal design - e.g. Cox, Cochran & Cox, Kempthorne.
I think that Kempthorne's book has a section where he derives the
standard distributional results for one-way ANOVA as an approximation
to the distribution under randomization only.

>EXAMPLE (purely bogus- I'm making this up on the fly):

Any ANOVA model with all fixed effects corresponds directly to a
regression model, so use of MR does not guarantee anything.
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to