6

I am trying to put the residuals from lm back into the original data.frame:

fit <- lm(y ~ x, data = mydata, weight = ind)
mydata$resid <- fit$resid

The second line would normally work if the residual has the same length as the number of rows of mydata. However, in my case, some of the elements of ind is NA. Therefore the residual length is usually less than the number of rows. Also fit$resid is a vector of "numeric" so there is no label for me to merge it back with the mydata data.frame. Is there an elegant way to achieve this?

1 Answers1

7

I think it should be pretty easy if ind is just a vector. 

sel <- which(!is.na(ind))
mydata$resid <- NA
mydata$resid[sel] <- fit$resid
John
  • 22,773
  • 6
  • 54
  • 82
  • 8
    As an alternative, the model can be fit with `lm(..., na.action=na.exclude)`. `residuals()` will then pad its output with `NA`s for the omitted cases - see [this answer](http://stats.stackexchange.com/a/11028/1909). – caracal Jun 12 '13 at 19:45
  • I thought this only worked for NA's in the data.frame, not in an array of weights that aren't in the data.frame. – John Jun 12 '13 at 20:12
  • Apparently, it works for weights, see [http://pastebin.com/0VUd68cz](http://pastebin.com/0VUd68cz). Although I didn't investigate this very thoroughly. – caracal Jun 12 '13 at 20:42