lm: eval(parse(text=)) works on one side y/x but not on both?

   lm(freeny)

On Sun, Nov 29, 2009 at 8:49 AM, Matthieu Stigler 
<matthieu.stigler at gmail.com <mailto:matthieu.stigler at gmail.com>> wrote:

    Hi

    My goal is to do a (multiple) regression, just knowing that my Y
    variables will be the say k first variables of a matrix/data
    frame. I thought I should do it with eval(parse)) but encounter a
    strange problem.

    See:
    lm(y~.-y, data=freeny) #that's what I want to do in the one
    equation case
    #Problem is I don't know name of the variable... only that it is
    the first one...
    #so idea is to just take first name
    a<-colnames(freeny)
    #and then use eval(parse(text=a[1]))

    #it works if I replace y on either the left or right side:
    lm(eval(parse(text=a[1]))~.-y, data=freeny) #does the same
    lm(y~.-eval(parse(text=a[1])), data=freeny)

    #but not if I do this call twice:
    lm(eval(parse(text=a[1]))~.-eval(parse(text=a[1])), data=freeny)
    #variable I wanted to remove (y) ist still there

    Do you understand why I can call eval(parse) only once? Should I
    try a update workaround? Or have idea of any other solution? Maybe
    there is something much simpler I'm missing:-(

    Thanks a lot!!!

    Matthieu Stigler

Try this:

ix <- 1:2
lm(as.matrix(freeny[ix]) ~., freeny[-ix])

On Sun, Nov 29, 2009 at 8:56 AM, Matthieu Stigler 
<matthieu.stigler at gmail.com <mailto:matthieu.stigler at gmail.com>> wrote:

    Thanks for answering so fast!!


          lm(freeny)

    :-)
    Ok that's working for the one equation case :-) Was example case...

    But now I want to have not only first column of freeny on the left
    but both first? And I don't know their names a priori...

    Thanks!


        On Sun, Nov 29, 2009 at 8:49 AM, Matthieu Stigler
        <matthieu.stigler at gmail.com
        <mailto:matthieu.stigler at gmail.com>
        <mailto:matthieu.stigler at gmail.com
        <mailto:matthieu.stigler at gmail.com>>> wrote:

           Hi

           My goal is to do a (multiple) regression, just knowing that
        my Y
           variables will be the say k first variables of a matrix/data
           frame. I thought I should do it with eval(parse)) but
        encounter a
           strange problem.

           See:
           lm(y~.-y, data=freeny) #that's what I want to do in the one
           equation case
           #Problem is I don't know name of the variable... only that
        it is
           the first one...
           #so idea is to just take first name
           a<-colnames(freeny)
           #and then use eval(parse(text=a[1]))

           #it works if I replace y on either the left or right side:
           lm(eval(parse(text=a[1]))~.-y, data=freeny) #does the same
           lm(y~.-eval(parse(text=a[1])), data=freeny)

           #but not if I do this call twice:
           lm(eval(parse(text=a[1]))~.-eval(parse(text=a[1])),
        data=freeny)
           #variable I wanted to remove (y) ist still there

           Do you understand why I can call eval(parse) only once?
        Should I
           try a update workaround? Or have idea of any other
        solution? Maybe
           there is something much simpler I'm missing:-(

           Thanks a lot!!!

           Matthieu Stigler

On Sun, Nov 29, 2009 at 9:16 AM, Gabor Grothendieck
<ggrothendieck at gmail.com> wrote:

By the way, if you really do want to create the formula anyways then:

?? ix <- 1:2
?? left <- paste(names(freeny)[ix], collapse = ",")
?? fo <- as.formula(paste("cbind(", left, ") ~ ."))
?? lm(fo, freeny)

or possibly replace last line with:

?? eval(substitute(lm(fo, freeny))

which will cause the formula to appear in the lm output.

This last line should have been:

This last line should have been:

?eval(substitute(lm(fo, freeny), list(fo = fo)))