[R] Outlier Detection for timeseries
gunter.berton at gene.com
Mon Feb 16 18:47:38 CET 2009
Danger: More careful thought required.
"Outlier....s" (Title of a TECHNOMETRICS paper of a couple of decades ago)
are an artificial construct: there is NO SUCH THING in the abstract. They
exist only wrt to a model. So there is no such thing as software that "tells
whether the changes are considered ...". Rather, you must consider
alternative "suitable" models, examine their fits, scientific implications,
interpretation, etc. Frequently, several models will fit essentially equally
well, but different subsets of the data will appear "unusual" (I no longer
use the word "outlier" because of the intimation that there is an objective
statistical meaning to this term, which there is not) for each.
Statistical algorithms cannot replace careful thinking. Sorry about that.
-- Bert Gunter
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Pele
Sent: Saturday, February 14, 2009 5:16 AM
To: r-help at r-project.org
Subject: Re: [R] Outlier Detection for timeseries
I am doing cross correlation analysis and I am trying to find a outlier
detection function in R that can detect changes in the level of the response
series that are not accounted for by the estimated model. Something that
tells whether the changes are considered Additive Outliers, Level Shifts, or
Temporary Changes... The output in the original not is what SAS produces and
I was looking for something similar.. R is very new to me (4 weeks) hence
still feeling my way around...
> Hello R users,
> Can someone tell if there is a package in R that can do outlier detection
> that give outputs simiilar to what I got from SAS below.
> Many thanks in advance for any help!
> Outlier Details
> Chi- Prob>
> Obs Time ID Type Estimate
> Square ChiSq
> 12 12.000000 Additive 2792544.6
> 186.13 <.0001
> 13 13.000000 Additive 954302.1
> 21.23 <.0001
> 15 15.000000 Shift 63539.3
> 9.06 0.0026
View this message in context:
Sent from the R help mailing list archive at Nabble.com.
R-help at r-project.org mailing list
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
More information about the R-help