On Jan 1, 2014, at 10:16 AM, "Ben Bolker" <bbolker at gmail.com> wrote:
On 14-01-01 09:21 AM, Lenth, Russell V wrote:
Maybe I can help. But what's the situation? Do you already have pilot
data analyzed with one of those packages? If so, you can use that
analysis to estimate the variance component nets needed for the
sample-size calculation. If not, your first step really is to do such
a pilot study.
The next step is to figure out the power function for the test(s) of
interest. That depends on the intended model for the data and the
expected mean squares (at least under the usual uniformly wonderful
normality conditions). You also need to establish what size
difference or other effect is of interest to detect with a stated
power and significance level.
There is software that can help with the power calculations, but
let's get more details first on those other steps.
Russ Lenth
Seems like very good advice to me (although I'm not sure what
"variance component nets" means -- is that a typo?)
If you don't have pilot data but you know enough about your system to
have some idea what realistic variance components and effect sizes
should be (and if you don't, you probably don't know enough to design a
sensible experiment in any case!)
There are several packages dedicated to power analysis for mixed
models (pamm, longpower, nlmeU, odprism), mostly (all?) by simulation.
The simulation capabilities in the development version of lme4 make it
easier to roll your own power analysis.
Hi list Can someone illustrate me how is possible to calculate
sample size for linear mixed model (not repeated measure) build
with lme4 or nlme package Thank in advance happy Holidays Bonitta
Gianluca