2
$\begingroup$

Is there any advice on how to deal with outliers in a generalized additive model?

I have been following the book "Generalized Additive Models : An Introduction with R, Second Edition" by Simon Wood but can't find advice on how to deal with outliers of random effects.

I am building a GAM in R using the mgcv package to model jobs over time, with a regional random effect. However I have one region which behaves differently to the rest and I am unsure how to deal with it. I want to explain the effect of it so I believe I should not remove it?

Improve this question
$\endgroup$
2
  • $\begingroup$ In what sense does the region behave differently? If you just mean the estimated value of the random intercept for that one region is a long way from the others, then it might not matter at all; if you're just interested in estimating the region-specific deviations from the overall response mean then you should be fine. If you want to do some testing on the entire random effect "smooth", then you might need to check the assumptions of the test for the ranef term in the output from summary.gam() in Wood (2013). $\endgroup$
    – Gavin Simpson
    Commented Jul 14, 2020 at 15:24
  • $\begingroup$ ...which does assume large sample normality of the individual random effect estimates (from what I can tell in the apper). $\endgroup$
    – Gavin Simpson
    Commented Jul 14, 2020 at 15:24

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.