If I had only a very limited time to do this, I might include
'month' as another effect, probably coded as 'sin' and 'cos' on an
annual cycle rather than as 12 individual Indicators. This would allow
you to explore not only main effects but interactions with plots.
Before I did that, however, I'd want to generate more plots of
data, residuals, and coefficients. For example,
qqnorm(resid(fm1Soy.nlme), datax=TRUE) displayed an S shape that
indicated inhomogeniety of variance. This suggests that there is
something else to be modeled in these data. I would next try plotting
residuals by 'month'. I'd also plot the averages and standard
deviations by 'month'. This might tell me if I only need to add a fixed
annual cycle, and how much of a Fourier series approximation to add. If
the standard deviations show a pattern, it suggests I need to model
heteroscedasticity. For that see '?varClasses' and the corresponding
information in a book by Pinheiro and Bates (2000), mentioned on that
help page. You can do 'anova' for any of these effects. [To test
changes in fixed effects, you will need to use method='ML', as discussed
in a book by Pinheiro and Bates (2000).]
Hope this helps.
Spencer
J S wrote:
Thanks. Here is a similar example from a book by Pinheiro and Bates
(2000, chapter 6):
library(nlme)
data(Soybean)
fm1Soy.lis <- nlsList( weight ~ SSlogis(Time, Asym, xmid, scal),
data = Soybean )
fm1Soy.nlme <- nlme( fm1Soy.lis )
*If we would like to make comparisons among the years we could just
simply involve years as a covariate, and later we could use L argument
to ANOVA to could compute contrasts. *
soyFix <- fixef( fm1Soy.nlme )
fm2Soy.nlme <- update( fm1Soy.nlme,
fixed = Asym + xmid + scal ~ Year,
start = c(soyFix[1], 0, 0, soyFix[2], 0, 0, soyFix[3], 0, 0) )
* *
*My question is: How can I compare variety of soybeans in a separate
month, i.e. if there was a difference in weight of soybeans F and P in
first month, …in twelve month?*
The dataset “Soybean”:
Plot Variety Year Time weight
1 1988F1 F 1988 14 0.106000
2 1988F1 F 1988 21 0.261000
3 1988F1 F 1988 28 0.666000
4 1988F1 F 1988 35 2.110000
5 1988F1 F 1988 42 3.560000
….
407 1990P8 P 1990 30 1.478330
408 1990P8 P 1990 37 2.601667
409 1990P8 P 1990 43 6.343330
410 1990P8 P 1990 51 6.131670
411 1990P8 P 1990 64 16.411700
412 1990P8 P 1990 79 16.946700
1) Involving months and variety as a covariates will probably
create too many parameters for the model?
2) Is it possible to use some test for comparisons, let’s say t
test? Perhaps not in case the data are dependent (i.e. previous
measurement is dependent on the next measurement, i.e. there is
temporal correlation (as in my study of Soil temperature)? What is an
alternative suggestion?
Thanks,
Julia
> Date: Fri, 4 Jul 2008 17:36:29 -0700
> From: [EMAIL PROTECTED]
> To: [EMAIL PROTECTED]
> CC: r-help@r-project.org
> Subject: Re: [R] Test for multiple comparisons: Nonlinear model,
autocorrelation?
>
> The question seems too general for me to offer specific suggestions.
>
> What problem are you trying to solve that you think 'multiple
> comparisons' will answer?
>
> Can you produce a similar problem that is completely self-contained
> example that eliminates complexity that may not be needed to understand
> your question (similar to the 'Auxiliary Problem' technique in "How to
> Solve It", http://en.wikipedia.org/wiki/How_to_Solve_It)? If you
can, it
> may lead you to a solution. If you get such an example but still can't
> see a solution, send that example to this list (following the advice in
> the posting guide http://www.R-project.org/posting-guide.html). The
> simpler the example, the more likely someone on this list will reply
> quickly with a useful suggestion.
>
> I know this doesn't solve your problem, but I hope it helps.
> Spencer
>
> J S wrote:
> > Dear R community,
> >
> > I have a nonlinear model describing average daily soil
temperature. What test should I use to compare differences in soil
temperature of the two studied vegetation types depending upon month?
> >
> > Building linear contrasts for the developed nonlinear model does
not help since this model does not include variable “Months” (only
“Days”).
> >
> > 1) Just a Student’s test is not probably an option because I would
violate an assumption of independency, since the daily soil
temperature observations have high autocorrelation. Or maybe I could
average the observations for each month and then use this test since I
have observations for a few years, and it might overcome the problem
of independency?
> >
> > 2) Should I develop a second nonlinear model with months instead
of days, but it would considerably increase a number of parameters in
the model...
> >
> > Or:
> > 3) ?
> >
> > Thanks for your help,
> > Julia
> > _________________________________________________________________
> > It’s a talkathon – but it’s not just talk.
> >
> > [[alternative HTML version deleted]]
> >
> >
> >
------------------------------------------------------------------------
> >
> > ______________________________________________
> > R-help@r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> > and provide commented, minimal, self-contained, reproducible code.
> >
------------------------------------------------------------------------
The i’m Talkaton. Can 30-days of conversation change the world? Find
out now.
<http://www.imtalkathon.com/?source=EML_WLH_Talkathon_ChangeWorld>
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
