dropterm() is a tool for model building, not primarily for significance
testing.
As the name suggests, it tells you what the effect would be were you to
"drop" each *accessible* "term" in the model as it currently
stands. By
default it displays the effect on AIC of dropping each term, in turn,
from the model. If you request them, though, it can also give you test
statistics and significance probabilities.
If there is an "A:B" interaction in the model, the main effects
"A" or
"B", if present, are not considered until a decision has been made on
including "A:B". The meaning of "A:B" in a model is not
absolute: it is
conditional on which main effect terms you have there as well. This is
one reason why the process is ordered in this way, but the main reason
is the so-called 'marginality' issue.
If you do ask for test statistics and significance probabilities, you
get almost a SAS-style "Type III" anova table, with the important
restriction noted above: you will not get main effect terms shown along
with interactions.
If you want the full SAS, uh, version, there are at least two
possibilities. 1. Use SAS. 2. Use John Fox's Anova() function from the
'car' package, along with his excellent book, which should explain how
to avoid shooting yourself in the foot over this.
(This difference of opinion on what should sensibly be done, by the way,
predates R by a long shot. My first exposure to it were with the very
acrimonious disputes between Nelder and Kempthorne in the mid 70's. It
has remained a cross-Atlantic dispute pretty well ever since, with the
latest shot being the paper by Lee and Nelder in 2004. Curiously, the
origin of the software can almost be determined by the view taken on
this issue, with Genstat going one way and SAS, SPSS, ... the other.
S-PLUS was a late comer...but I digress!)
Bill Venables.
Bill Venables
CSIRO Laboratories
PO Box 120, Cleveland, 4163
AUSTRALIA
Office Phone (email preferred): +61 7 3826 7251
Fax (if absolutely necessary): +61 7 3826 7304
Mobile: +61 4 8819 4402
Home Phone: +61 7 3286 7700
mailto:Bill.Venables at csiro.au
http://www.cmis.csiro.au/bill.venables/
-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org]
On Behalf Of DaniWells
Sent: Sunday, 10 February 2008 11:40 PM
To: r-help at r-project.org
Subject: [R] Do I need to use dropterm()??
Hello,
I'm having some difficulty understanding the useage of the
"dropterm()"
function in the MASS library. What exactly does it do? I'm very new to
R, so
any pointers would be very helpful. I've read many definitions of what
dropterm() does, but none seem to stick in my mind or click with me.
I've coded everything fine for an interaction that runs as follows: two
sets
of data (one for North aspect, one for Southern Aspect) and have a
logscale
on the x axis, with survival on the y. After calculating my anova
results i
have all significant results (ie aspect = sig, logscale of sunlight sig,
and aspect:llight = sig).
When i have all significant results in my ANOVA table, do i need
dropterm(),
or is that just to remove insignificant terms?
Many thanks,
Dani
--
View this message in context:
http://www.nabble.com/Do-I-need-to-use-dropterm%28%29---tp15396151p15396
151.html
Sent from the R help mailing list archive at Nabble.com.
______________________________________________
R-help at 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.