[R] still problems with predict!

[Sundar Dorai-Raj]

Anne wrote:

> Hi all,
> 
> I still have problems with the predict function by setting up the values on
> which I want to predict
> 
> ie:
> original df: p1 (193 obs) variates y x1 x2
> 
> rm(list=ls())
> x1<-rnorm(193)
> x2<-runif(193,-5,5)
> y<-rnorm(193)+x1+x2
> p1<-as.data.frame(cbind(y,x1,x2))
> p1
>              y         x1         x2
> 1   -0.6056448 -0.1113607 -0.5859728
> 2   -4.2841793 -1.0432688 -3.3116807
> ......
> 192 -1.3228239  1.0263013 -2.7801324
> 193  1.8736683  1.0480632  0.4746959
> 
> newdf<-data.frame(x1= seq(min( p1$x1),max( p1$x1),length=10),
>                               x2=rep(median( p1$x2),10) )
> pr<-predict(g<-lm(p1$y~p1$x1+p1$x2) ,newdf, se.fit = TRUE)
> 
> newdf
>            x1         x2
> 1  -2.3844149 -0.2594991
> 2  -1.8388635 -0.2594991
> ...
> 9   1.9799963 -0.2594991
> 10  2.5255477 -0.2594991
> 
> pr$fit
> 1   -0.6766906
> 2   -4.4198864
> .....
> 192 -1.6531906
> 193  1.6395442
> 
> so apparently the predict() function did not take up the new data.frame
> 
> 
> I looked up with conflicts() to see if I had masked objects in the search
> path potentially causing this problem
> but found none
> 
> 
> 

Hi Anne,

predict is working properly (though not as you expected). It's not 
evaluating your newdf because it has no columns called p1$x1 or p1$x2. 
Try this instead:

pr <- predict(g <- lm(y ~ x1 + x2, p1), newdf, se.fit = TRUE)

 > str(pr)
List of 4
  $ fit           : Named num [1:10] -2.365 -1.865 -1.366 -0.867 -0.367 ...
   ..- attr(*, "names")= chr [1:10] "1" "2" "3" "4" ...
  $ se.fit        : Named num [1:10] 0.1751 0.1424 0.1120 0.0868 0.0723 ...
   ..- attr(*, "names")= chr [1:10] "1" "2" "3" "4" ...
  $ df            : int 190
  $ residual.scale: num 0.987


--sundar