Re: Mixed 2-way ANOVA

On Thu, 29 Sep 2005 08:34:38 -0400, Bruce Weaver
<bweaver@xxxxxxxxxxxx> wrote:

> Corky wrote:
> > Hi thanks for all of the responses. It does seem to be a 3x2 and not
> > 4x2 design. The particpants will have to put into the groups depending
> > on criteria rather than being randomly selected as one of the factors
> > is height, unless i put some of them on the rack!!
> >
> >
> >>In any case, for complicated designs like this, it is usually wise to
> >>reduce things to as simple a contrast as possible (e.g., a t-test), and
> >>to obtain a sample size estimate for that contrast.
> >
> >
> > How would this be possible?
> >
> > Many thanks
> >
>
>
> What is/are the critical comparison(s) you want to make? Can you boil
> each of them down to a comparison of two means?

To push the question one step further -- It *looks* as if
you must be interested in the Pre-Post comparison, and
Height is included as one sort of control.

a) For the overall comparison, you have a paired t-test
of Pre-Post: What is the size of change? What is the expected
standard deviation of change? That gets a simple power
analysis.

b) What is the role of height, in the design or among the
hypotheses? Does Height mess up the change scores?
Is this a 'scaling' issue, where percent-change would not
show the same problem? Is it reasonable, instead of factors,
to include Height as a continuous covariate?

c) If you are interested in *differences* in change between
the groups with different heights, that is a tougher power
analysis. You would need to estimate the variances of
change scores, and estimate the *expected* differences (or
the differences that would be interesting to find).


--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.