I used SAS to analyze the data initially, since the data set was made up of
several files when I received it, and I'm still not very good at
manipulating data in R.
I have posted the data set from one location at the following address:
http://uwstudentweb.uwyo.edu/A/AKNISS/sxherb.txt
var=cultivar
trt=herbicide treatment
yield=response variable of interest
All plot# from 101 to 104 are rep 1, 201-204 rep 2, and 301 to 304 rep 3.
It was the only file that was in an easy format for R to read at the moment,
and was probably the most reliable trial of the two locations. I would like
to use power.anova.test() with this data set to plan next years study (to
get a sample size for each herb*var combination), but I'm not quite sure how
that is done for an interaction effect. Do I just use the MS for herb*var
as the between group variance and the MSE as the within group variance? Or
do I need to somehow include other variance parameters in the model?
The model for this location (split-block design):
yield = rep + herb + var + herb*var ## all are fixed effects
rep*herb = error term for herb
rep*var = error term for cultivar
residual = error term for herb*var
I hope this attempt at my question was a little more clear. I appreciate
any help that is offered.
Andrew Kniss
Assistant Research Scientist
University of Wyoming
Department of Plant Sciences
1000 E. Univesity Ave
Laramie, WY 82071 USA
akniss at uwyo.edu
-----Original Message-----
From: John Maindonald [mailto:john.maindonald at anu.edu.au]
Sent: Tuesday, February 22, 2005 3:37 PM
To: r-help at stat.math.ethz.ch
Cc: akniss at uwyo.edu
Subject: Re: R-help Digest, Vol 24, Issue 22
You need to give the model formula that gave your output.
There are two sources of variation (at least), within and
between locations; though it looks as though your analysis
may have tried to account for this (but if so, the terms are
not laid out in a way that makes for ready interpretation.
The design is such (two locations) that you do not have
much of a check that effects are consistent over locations.
You need to check whether results really are similar
for all cultivars and for all herbicides, so that it is
legitimate to pool as happens in the overall analysis.
If a herbicide:cultivar combination has little effect the
variability may be large, while if it has a dramatic effect
(kills everything!), there may be no variability to speak of.
John Maindonald.
On 22 Feb 2005, at 10:06 PM, r-help-request at stat.math.ethz.ch wrote:
> To: "'Bob Wheeler'" <bwheeler at echip.com>
> Cc: r-help at stat.math.ethz.ch
> Subject: RE: [R] power.anova.test for interaction effects
> Reply-To: akniss at uwyo.edu
>
>
> It's a rather complex model. A 37*4 factorial (37 cultivars[var]; 4
> herbicide treatments[trt]) with three replications[rep] was carried
> out at
> two locations[loc], with different randomizations within each rep at
> each
> location.
>
> Source DF Error Term MS
> Loc 1 Trt*rep(loc) 12314
> Rep(loc) 4 Trt*rep(loc) 1230.5
> Trt 3 Trt*rep(loc) 64.72
> Trt*loc 3 Trt*rep(loc) 33.42
> Trt*rep(loc) 12 Residual 76.78
> Var 36 Var*trt*loc 93.91
> Var*trt 108 Var*trt*loc 12.06
> Var*trt*loc 144 Residual 43.09
> Residual 575 NA 21.23
>
>
> -----Original Message-----
> From: Bob Wheeler [mailto:bwheeler at echip.com]
> Sent: Monday, February 21, 2005 4:33 PM
> To: akniss at uwyo.edu
> Cc: r-help at stat.math.ethz.ch
> Subject: Re: [R] power.anova.test for interaction effects
>
> Your F value is so low as to make me suspect your model. Where did the
> 144 denominator degrees of freedom come from?
>
John Maindonald email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473 fax : +61 2(6125)5549
Centre for Bioinformation Science, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.