Rich Ulrich <[EMAIL PROTECTED]> wrote in article =
<[EMAIL PROTECTED]>...
>  - what to do -
> Approximate your design and power calculation with
> the nearest, likely, paired t-test.
>=20
> Or, look at the contrast of interest, and compute
> the power based on the simple ANOVA on the contrast;
> you probably want to dummy up some data for this,=20
> if you don't have good pilot data on hand.

Definitely reccommended for beginners.

> Having said that, I will say that you should probably
> not use any power analysis program that does not
> have ample documentation, including sample problems --
> plug in numbers to match previous results,
> and match (with *some*  accuracy)  your design.

I'd suggest GPower as there is some documentation out there and it is =
free.

> The problem has been asked before, of course.
> I don't remember being impressed by Repeated Measures
> reported (even) for expensive programs, though there=20
> was a recommendation or two (I think I recall).

The problem with repeateed measures seems to be estimating
the additional parameter(s). Unless you have good sources for
estimates (e.g., pilot data) it will involve a lot more hand-waving than
independent measures.

Thom
.
.
=================================================================
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