0

I wish to examine assumptions for single regression using the built in diagnostics within R. The residuals vs fitted plot below seemly violated the assumption. There are 2 dummy coded variables within my model and 1 categorical variable (testing probability). The model only explains 8% of the variance in selection (DV). What improvements/Advice can be tell from the plot? Thanks.

Skn U
  • 1
  • The definition residual $=$ observed $-$ fitted implies that distinct observed values define parallel lines on this plot, in this case residual $= 0\ -$ fitted and residual $= 1\ -$ fitted. Nothing is really violated here; it is just that your data aren't well suited to the model you fit. – Nick Cox May 18 '21 at 15:23
  • For diagnosing a logistic regression, consider using DHARMa, see hints on logistic regression in https://cran.r-project.org/web/packages/DHARMa/vignettes/DHARMa.html#binomial-data – Florian Hartig Jun 24 '21 at 11:41

1 Answers1

1

Without wanting this to sound snotty, my advice would be not to use linear regression with a binary outcome variable. Use regression instead. (In R, that's: glm(<formula>, <data>, family=binomial). When you do use logistic regression, ignore these plots (see my answer to: Interpretation of plot (glm.model)).

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650