Switch to qgam for quantile regressions

[florianhartig]

Idea: switch to https://cran.r-project.org/web/packages/qgam/index.html for calculating quantile regressions in plotResiduals

[florianhartig]

An issue is that qgam sometimes throws errors for small data sizes. Example code

testData = createData(sampleSize = 50, family = gaussian())
fittedModel <- lm(observedResponse ~ Environment1, data = testData)

res = simulateResiduals(fittedModel)
plot(res) # current quantile plot

library(qgam)
dat=data.frame(res =  res$scaledResiduals , pred = predict(fittedModel))
fit <- qgam(res ~ s(pred),  data =dat, qu = 0.5)
# tends to throw wrror for small n

[mfasiolo]

Hi Florian,

The new version of qgam on https://github.com/mfasiolo/qgam should avoid this.
Let me know if you encounter further problems.

Matteo

[florianhartig]

Yes, this works - any idea when this will go to CRAN?

[florianhartig]

Matteo, I'm just playing around with that - your package is awesome, especially that one gets CIs and p-values on the splines, this will be a great addition to DHARMa because I can directly read of p-values for the residual tests!

[florianhartig]

A version of DHARMa based on qgam is now available in the branch qgam-switch. These are the old plots

These are the new plots

There is also a new testQuantiles function. This can be tried via installing

devtools::install_github("florianhartig/DHARMa/DHARMa", ref = "qgam-switch")

[florianhartig]

This is basically implemented in 0.3 dev so I will close this