2

Can anyone advise how to interpret this resulting residuals vs fitted graph and what may be wrong with the model?

enter image description here

Nikki
  • 21
  • 1
  • 6
    [Searching for "diagonal" and "residual"](http://stats.stackexchange.com/search?q=diagonal+residual) gives several other posts on this site for this phenomenon: [in ANOVA](http://stats.stackexchange.com/q/132686/22228), [in Poisson regression](http://stats.stackexchange.com/q/25068/22228) and in [multiple regression](http://stats.stackexchange.com/q/33165/22228). I think yours is really a duplicate of the latter, though I like how clearly your graphs show the phenomenon! – Silverfish Sep 17 '16 at 11:59
  • Not the answer, but always prefer software in which the axis labels can be lifted to the periphery, leaving just the axes themselves. – Nick Cox Sep 17 '16 at 14:00

1 Answers1

3

This is a well known phenomenon of regressing integer valued outcomes on a continuous predictor. Aside from that, there is surprisingly little that can be revealed from any "interpretation" of this residual plot.

Example:

> x <- rexp(1000)
> y <- rpois(1000, x)
> plot(y ~ x)

enter image description here

AdamO
  • 52,330
  • 5
  • 104
  • 209