> From: rvarad...@jhmi.edu
> To: marchy...@hotmail.com; jmo...@student.canterbury.ac.nz;
> r-help@r-project.org
> Subject: RE: [R] Time-series analysis with treatment effects - statistical
> approach
> Date: Thu, 23 Jun 2011 02:59:19 +0000
>
> If you have any specific features of the time series of soil moisture, you
> could either model that or directly estimate it and test for differences in
> the 4 treatments. If you do not have any such specific considerations, you
> might want to consider some nonparametric approaches such as functional data
> analysis, in particular functional principal components analysis (fPCA)
> might be relevant. You could also consider semiparametric methods. For
> example, take a look at the "SemiPar" package.
>
> Ravi.
I guess just playing with it while waiting for other code to finish, I'd be
curious if you had
any controlled tests such as impulse response- what did treatment do when you
held
at constant temp and humidity and illumination in stll air after single burst
of rain?
If you were pursing the model approach, quick look suggests qualitatitve rather
than
just quantitative effects - in one case looks like linear or biphasic dry out
dynamics, others
seem to just fall off of cliff.
Objective of course matters too, if you are trying to sell this to farmers,
maybe a plot of
moisture for each treatment against control would help. I just did that after
averaging over sensors
and it may be a reasonable analysis for cost effectiveness if you can translate
moisture into
dollars. Now you would still need to put error bars on comparisons and use
words carefully etc
but that approach may be more important than getting at dynamics. I dunno.
Consider that in fact maybe all you care about is peaks, if too dry for one day
kills the crop then that is what you want to focus the analysis on etc etc etc.
> ________________________________________
> From: r-help-boun...@r-project.org [r-help-boun...@r-project.org] on behalf
> of Mike Marchywka [marchy...@hotmail.com]
> Sent: Wednesday, June 22, 2011 9:31 PM
> To: jmo...@student.canterbury.ac.nz; r-help@r-project.org
> Subject: Re: [R] Time-series analysis with treatment effects - statistical
> approach
>
> > Date: Wed, 22 Jun 2011 17:21:52 -0700
> > From: jmo...@student.canterbury.ac.nz
> > To: r-help@r-project.org
> > Subject: Re: [R] Time-series analysis with treatment effects - statistical
> > approach
> >
> > Hi Mike, here's a sample of my data so that you get an idea what I'm working
> > with.
>
> Thanks, data helps make statements easier to test :) I'm quite
> busy at moment but I will try to look during dead time.
>
> >
> > http://r.789695.n4.nabble.com/file/n3618615/SampleDataSet.txt
> > SampleDataSet.txt
> >
> > Also, I've uploaded an image showing a sample graph of daily soil moisture
> > by treatment. The legend shows IP, IP+, PP, PP+ which are the 4 treatments.
> > Also, I've included precipitation to show the soil moisture response to
> > precip.
>
> Personally I'd try to write a simple physical model or two and see which
> one(s)
> fit best. It shouldn't be too hard to find sources and sinks of water and
> write
> a differential equation with a few parameters. There are probably online
> lecture notes that cover this or related examples. You probably suspect a
> mode of action for the treatments, see if that is consistent with observed
> dyanmics.
> You may need to go get temperature and cloud data but it may or may not
> be worth it.
>
> >
> > http://r.789695.n4.nabble.com/file/n3618615/MeanWaterPrecipColour2ndSeasonOnly.jpeg
> >
> > I have used ANOVA previously, but I don't like it for 2 reasons. The first
> > is that I have to average away all of the interesting variation. But mainly,
>
> There are a number of assumptions that go into that to make it useful. If
> you are just drawing samples from populations of identical independent things
> great but here I would look at things related to non-stationary statistics of
> time series.
>
> > it becomes quite cumbersome to do a separate ANOVA for each day (700+ days)
> > or even each week (104 weeks).
>
> I discovered a way to do repetitive tasks that can be concisely specified
> using
> something called a computer. Writing loops is pretty easy, don't give up
> due to cumbersomeness. Also, you could try a few simple things like plotting
> difference charts ( plot treatment minus control for example).
>
> If you approach this purely empirically, there are time series packages
> and maybe the econ/quant financial analysts would have some thoughts
> that wouldn't be well known in your field.
>
>
> >
> > Thanks for your help,
> > -Justin
> >
- - - - - -
Mike Marchywka | V.P. Technology
415-264-8477
marchy...@phluant.com
Online Advertising and Analytics for Mobile
http://www.phluant.com
[[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.
