[R] ploting a linear model

Ott Toomet siim at obs.ee
Thu May 16 10:33:52 CEST 2002

On 15 May 2002, Ernesto Jardim wrote:

  |I have the result of a linear model "lm98" and I want to plot it to
  |analyse the regression but I'm getting the folowing error message
  |> lm98<-lm(log(U+1)~Bq+Gi+Lv+Tb+Bq:Gi+Bq:Lv+Bq:Tb+Gi:Lv+Gi:Tb+Lv:Bq)
  |> plot(lm98)
  |Error in plot.window(xlim, ylim, log, asp, ...) : 
  |        need finite ylim values
  |In addition: Warning message: 
  |NaNs produced in: sqrt(1 - hii) 

You should try to describe and investigate the problem better.  As you can
see, the problem arises in plot.lm(). Typing plot.lm (note -- without
parentheses) on command prompt, you can find a part of the plotting routine
which include ,,hii'':

        hii <- lm.influence(x)$hat
        w <- weights(x)
        r.w <- if (is.null(w)) 
        else (sqrt(w) * r)[w != 0]
        rs <- r.w/(s * sqrt(1 - hii))

?lm.influence says that 

     A list containing the following components: 
     hat: a vector containing the diagonal of the `hat'' matrix.

Hat matrix (at least it is common notion in econometrics) is

H <- X %*% solve( t(X) %*% X) %*% t(X)

Your problem is associated with the fact that any of the diagonal elements
is greater than 1.  This is probably due to numerical problems, caused by
near-singular matrices.  

You should carefully look at your data.  Have you checked for
multicollinearity?  Does the results from summary(lm(...)) seem reasonable? 
What do say specification tests?

Perhaps it helps



r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch

More information about the R-help mailing list