help with R

Berton,

Firstly, thanks for your comments.

To address the what you first said, plotting the 3690-element vector is what
is causing R to hang.
Well, amongst the things you are not telling us are your OS and version of 
R and what commands you are using.

The only relevant thing I know of is that NT-based versions of Windows 
were causing R to take a long time (but not hang completely) when plotting 
longish series in 2.1.0, but that was worked around in 2.1.1, and even 
there n=3690 would only just begin to show up the Windows bug.
 Rather than lose everything I've entered by hand each
interactive run I've switched to using a batch script, which I can now load
and run at prompt.  Using sink("filename.txt") I'm able to save the output
to study.

My usage of dummy variables is to identify seasonality on the daily level
over
15 years with 246 days per year.  I need to identify the day each month
when an (expected) event is occuring.  The date the event occurs does
not occur on necessarily the same day each month.  I don't know of
another method that could identify these statistically significant seasonal
events using R.  Dummy variables with a LM is the only method I have
experience with using R.  If you or anyone has suggestions on what other
methods to use I would appreciate some suggestions.  Using 245 dummy
variables is quite awkward.

I see lag() can be used to build a first- or multi-order differenced time
series to extract any underlying trend in a time series.

Using STL() might be promising.  It appears to be similar to other methods
I've used with MINITAB but called something different.

Nor an ARIMA nor a BSM is really what I need as I'm not focused on
performing predictions or modeling of the (possibly non-normal) properties
of the residuals.

Thanks.  All your advice is greatly appreciated.

Ed

      "A man is not old until regrets take the place of dreams."
                    Actor John Barrymore

From: Berton Gunter <gunter.berton at gene.com>
To: "'Ed Wang'" <eymw at hotmail.com>, <r-help at stat.math.ethz.ch>
Subject: RE: [R] help with R
Date: Tue, 29 Nov 2005 11:05:02 -0800

You're not telling us something or there's a problem with your R build: a
3960 element vectors of integer is tiny and will not cause R to crash.

Regarding your regression model. You do **not** need dummy variables in R.
Please read the docs (e.g. AN INTRODUCTION TO R) and help files on lm() and
factor() to see how to do linear modeling in R. lag() and diff() may also be
relevant. OTOH, R has many better ways to model time series and seasonality,
both in base R and numerous add-on packages. Try help.search('time series')
and RSiteSearch('time series')

-- Bert Gunter
Genentech Non-Clinical Statistics
South San Francisco, CA

"The business of the statistician is to catalyze the scientific learning
process."  - George E. P. Box

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595
Morning,

I've downloaded the precompiled R 2.1.1 version and am using Windows XP
on my office workstation.  As mentioned previously, I've resorted to batch
jobs to avoid the hanging that occurs when I try to plot the 3690 length
vector of data.  If it's warranted, I can do a build from the source and 
change
specific parameters in the makefile if people feel it is warranted.

Based on Berton's suggestion to look at the range of packages available
I think stl() might be as appropriate a package to use to identify all three
components of the time series data I have: underlying trend, seasonality
over a full year period (periodicity of one year, or 246 days in my case),
and residual (which I have no expectation that it will necessarily be
~N(0,\sigma^2)).

For the following dataset (15 years, 246 days/year => 3690 days of data)
what reasonal parameters for running stl() would folks suggest?  I've not
had any luck with getting stl() to return any useful information.  It 
continues
to stop with the statement

        series is not periodic or has less than two periods

using

stl(zHO, s.window=1, s.degree=2, l.window=246)

or the obvious ways I might try running stl() (i.e. plot(stl(zHO))).  It's
possible I've not properly specified the length of expected periodicity as
a parameter (246 days in my case).

All suggestions are welcome!  I'm trying to avoid going back and fitting
a linear model with 245 dummy variables.

Thanks.

Ed
Morning,

I've downloaded the precompiled R 2.1.1 version and am using Windows XP
on my office workstation.  As mentioned previously, I've resorted to batch
jobs to avoid the hanging that occurs when I try to plot the 3690 length
vector of data.  If it's warranted, I can do a build from the source and 
change
specific parameters in the makefile if people feel it is warranted.

Based on Berton's suggestion to look at the range of packages available
I think stl() might be as appropriate a package to use to identify all three
components of the time series data I have: underlying trend, seasonality
over a full year period (periodicity of one year, or 246 days in my case),
and residual (which I have no expectation that it will necessarily be
~N(0,\sigma^2)).

For the following dataset (15 years, 246 days/year => 3690 days of data)
what reasonal parameters for running stl() would folks suggest?  I've not
had any luck with getting stl() to return any useful information.  It 
continues
to stop with the statement

        series is not periodic or has less than two periods

using

stl(zHO, s.window=1, s.degree=2, l.window=246)

or the obvious ways I might try running stl() (i.e. plot(stl(zHO))).  It's
possible I've not properly specified the length of expected periodicity as
a parameter (246 days in my case).
In creating the timeseries 9ts) object you need
myts <-  ts(mydata,    ....,   frequency=246)

Kjetil
All suggestions are welcome!  I'm trying to avoid going back and fitting
a linear model with 245 dummy variables.

Thanks.

Ed

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html