When validating a linear mixed model, I get the attached plot.
It seems that errors are largest when fitted values are highest. However, there is no clear funnel pattern, but an envelope that opens up to a funnel only when fitted values get really high.
I am not sure whether this means heteroscedasticity or any other violation of model assumptions, or is it ok to go on with this model.
Graphs should only be used to 'eye-ball' the data. Have you tried running any tests?
https://cran.r-project.org/web/packages/olsrr/vignettes/heteroskedasticity.html
Thank you very much Samantha Curle .
I am not sure the tests proposed in the link are appropriate for linear mixed effects models.
In addition, everywhere I read it is recommended to use graphical tools rather than tests (e.g. Zuur et al 2009_Mixed Effects Models and Extensions in Ecology with R).
I suggest you try a transformed response variable, for example ln y or sqrt y or other such forms pertaining to Box and Cox family of transformations and see if the corresponding plots improve w.r.t the structure less ness of the residuals.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation