How to include an externally defined NULL value in lm

a <- if (TRUE) . ~ . + freeny.x[,4] else . ~ .
fo <- update(freey.y ~ freeny.x[, -4], a)
fo

Matthieu wants to manipulate the *formula* --- i.e. control
what predictors go into the formula --- rather than to add something
to one of the predictors (or not, as the case maybe).

I.e. what is wanted is that if ``condition'' is TRUE then the
formula should be

       freeny.y ~ freeny.x[,-4] + freeny.x[,4]

and otherwise it should be

       freeny.y ~ freeny.x[,-4]

The subtlety is that `` y ~ x + NULL'' gives the model ``y = beta_0 +
beta_1 * x + E''
Whereas ``y ~ x + a'' gives the model ``y = beta_0 + beta_1 * x +
beta_2 * a + E''.
Irrespective of what the object a happens to be --- or even of
whether it exists.
(This is a bit of an oversimplification I think, but close enuff.)

Then when lm() tries to fit y = beta_0 + beta_1 * x + beta_2 * a + E
it gets all in a dither when a is NULL, because the model does not then
make sense.

I'm afraid I don't know how to get around the problem using the NULL
concept.
Wiser heads than mine may be able to suggest a way.

There are however different approaches to the problem, ***not***
using NULL.

E.g.

       fff <- if(connie) as.formula(freeny.y ~ freeny.x[,-4] + freeny.x[,
4]) else as.formula(freeny.y ~ freeny.x[,-4])
       fit <- lm(fff)

or (more sensibly?)

       fff <- if(connie) as.formula(freeny.y ~ freeny.x) else as.formula
(freeny.y ~ freeny.x[,-4])
       fit <- lm(fff)

More sensibly still:

       fff <- if(connie) y ~ . else y ~ . - market.potential
       fit <- lm(fff,data=freeny)

Where ``connie'' is some condition (with value either TRUE or FALSE).

Formulae can also be constructed as text strings (e.g. using
``paste'') and then converted
to actual formulae using as.formula().

HTH.

       cheers,

               Rolf Turner




On 4/03/2008, at 11:34 AM, <markleeds at verizon.net> wrote:

From: Matthieu Stigler <stigler3 at etu.unige.ch>
Date: 2008/03/03 Mon PM 04:07:09 CST
To: r-help at r-project.org
Subject: [R] How to include an externally defined NULL value in lm

maybe you should a<-0 unless there's special
behavior of NULL that's unknownst to me ?

Hello!

I would love to be able to include an external variable  to a lm
call, I
mean something:

if(TRUE)
   a<-freeny.x[,4]
else
   a<-NULL

lm(freeny.y~freeny.x[,-4] +a)

but it does not work with a<-NULL, whereas

lm(freeny.y~freeny.x[,-4] +NULL)

I don't understand why and did not find an answer in the
manuals... do
you see it? Any idea?

Thanks!!

######################################################################
Attention:\ This e-mail message is privileged and confid...{{dropped:9}}