I run a bam with timeseries data and a response variable defined as cbind(positive, negative), family = quasibinomial (because of overdispersion), a number of interactions, and some random effects. When I use gam.check(model, old.style = TRUE) to get some diagnostic plots I run into some weird patterns.
Especially the one plotting residuals vs. linear predictors is quite striking. Now I am wondering how to find out which datapoints might be the ones causing the pattern in order to understand better what is happening. Plotting the single predictors vs the residuals hasn´t helped me so far. Any clues?!
Thanks for any help.