berwin@maths.uwa.edu.au
2005-May-06 10:37 UTC
[Rd] But in add1, presumably in add1.lm (PR#7842)
xDear all,
while investigating some data, I noticed a strange behaviour of add1
which is reproduced below using a data set from a recommended
package.
Essentially, it seems the inclusion of an interaction term by add1 is
only correctly handled if the order of the variables in the
interaction coincide with the order in the formula that is given as
the scope. (I have a vague memory that such a bug was already once
posted some time ago, but could not find anything in the bug report
repository.)
I noticed this problem on R 2.1.0, but it also exists in all other R
version installed on my machine (R 1.8.1, R 1.9.1, R 2.0.0 and R
2.0.1).
Cheers,
Berwin
> library(MASS)
> fmb <- lm(Days ~ .^4, quine)
> fm0 <- lm(Days ~ Eth + Sex, quine)
> add1(fm0, formula(fmb), test="F")
Single term additions
Model:
Days ~ Eth + Sex
Df Sum of Sq RSS AIC F value Pr(F)
<none> 35045 806
Age 3 2119 32926 803 3.0030 0.03263
Lrn 1 162 34882 808 0.6605 0.41773
Eth:Sex 1 141 34904 808 0.5724 0.45057
> fm0 <- lm(Days ~ Sex + Eth, quine)
> add1(fm0, formula(fmb), test="F")
Single term additions
Model:
Days ~ Sex + Eth
Df Sum of Sq RSS AIC F value Pr(F)
<none> 35045 806
Age 3 2119 32926 803 3.0030 0.03263
Lrn 1 162 34882 808 0.6605 0.41773
Eth:Sex 0 0 35045 806
> fm0 <- lm(Days ~ Age + Eth, quine)
> add1(fm0, formula(fmb), test="F")
Single term additions
Model:
Days ~ Age + Eth
Df Sum of Sq RSS AIC F value Pr(F)
<none> 33049 802
Sex 1 123 32926 803 0.5226 0.4710
Lrn 1 598 32450 801 2.5815 0.1104
Eth:Age 0 0 33049 802
> fm0 <- lm(Days ~Eth + Age , quine)
> add1(fm0, formula(fmb), test="F")
Single term additions
Model:
Days ~ Eth + Age
Df Sum of Sq RSS AIC F value Pr(F)
<none> 33049 802
Sex 1 123 32926 803 0.5226 0.470950
Lrn 1 598 32450 801 2.5815 0.110369
Eth:Age 3 2837 30212 795 4.3194 0.006047
--please do not edit the information below--
Version:
platform = i686-pc-linux-gnu
arch = i686
os = linux-gnu
system = i686, linux-gnu
status =
major = 2
minor = 1.0
year = 2005
month = 04
day = 18
language = R
Search Path:
.GlobalEnv, package:methods, package:stats, package:graphics,
package:grDevices, package:utils, package:datasets, Autoloads, package:base
ripley@stats.ox.ac.uk
2005-May-06 11:56 UTC
[Rd] But in add1, presumably in add1.lm (PR#7842)
The problem is in the way terms() reorders interactions in forming the term labels. If asked to fit ~ Age + Eth + Eth:Age it actually fits ~ Age + Eth + Age:Eth It is not clear to me why R does this (S does not) nor if it is safe to change that, but it would be better to fix the root cause if possible as it occurs in several places. In particular, it affects all known add1 methods and can affect drop1 too. If used consistently, there is a no problem: here your upper scope does not (strictly) include your lower model. On Fri, 6 May 2005 berwin@maths.uwa.edu.au wrote:> xDear all, > > while investigating some data, I noticed a strange behaviour of add1 > which is reproduced below using a data set from a recommended > package. > > Essentially, it seems the inclusion of an interaction term by add1 is > only correctly handled if the order of the variables in the > interaction coincide with the order in the formula that is given as > the scope. (I have a vague memory that such a bug was already once > posted some time ago, but could not find anything in the bug report > repository.) > > I noticed this problem on R 2.1.0, but it also exists in all other R > version installed on my machine (R 1.8.1, R 1.9.1, R 2.0.0 and R > 2.0.1). > > Cheers, > > Berwin > >> library(MASS) >> fmb <- lm(Days ~ .^4, quine) > >> fm0 <- lm(Days ~ Eth + Sex, quine) >> add1(fm0, formula(fmb), test="F") > Single term additions > > Model: > Days ~ Eth + Sex > Df Sum of Sq RSS AIC F value Pr(F) > <none> 35045 806 > Age 3 2119 32926 803 3.0030 0.03263 > Lrn 1 162 34882 808 0.6605 0.41773 > Eth:Sex 1 141 34904 808 0.5724 0.45057 > >> fm0 <- lm(Days ~ Sex + Eth, quine) >> add1(fm0, formula(fmb), test="F") > Single term additions > > Model: > Days ~ Sex + Eth > Df Sum of Sq RSS AIC F value Pr(F) > <none> 35045 806 > Age 3 2119 32926 803 3.0030 0.03263 > Lrn 1 162 34882 808 0.6605 0.41773 > Eth:Sex 0 0 35045 806 > >> fm0 <- lm(Days ~ Age + Eth, quine) >> add1(fm0, formula(fmb), test="F") > Single term additions > > Model: > Days ~ Age + Eth > Df Sum of Sq RSS AIC F value Pr(F) > <none> 33049 802 > Sex 1 123 32926 803 0.5226 0.4710 > Lrn 1 598 32450 801 2.5815 0.1104 > Eth:Age 0 0 33049 802 > >> fm0 <- lm(Days ~Eth + Age , quine) >> add1(fm0, formula(fmb), test="F") > Single term additions > > Model: > Days ~ Eth + Age > Df Sum of Sq RSS AIC F value Pr(F) > <none> 33049 802 > Sex 1 123 32926 803 0.5226 0.470950 > Lrn 1 598 32450 801 2.5815 0.110369 > Eth:Age 3 2837 30212 795 4.3194 0.006047 > > --please do not edit the information below-- > > Version: > platform = i686-pc-linux-gnu > arch = i686 > os = linux-gnu > system = i686, linux-gnu > status > major = 2 > minor = 1.0 > year = 2005 > month = 04 > day = 18 > language = R > > Search Path: > .GlobalEnv, package:methods, package:stats, package:graphics, package:grDevices, package:utils, package:datasets, Autoloads, package:base > > ______________________________________________ > R-devel@stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-devel > >-- Brian D. Ripley, ripley@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