0

I am performing a multiple regression with GAM.I found the significant predictors using the stepAIC function and now I am performing a model checking with mod.check().

This is what a get:

m2.k1 <- gam(Totden ~ s(T, k=12) + s(S, k=12) + s(Si, k=12), data = train.data)
gam.check(m2.k1)

Method: GCV   Optimizer: magic
Smoothing parameter selection converged after 7 iterations.
The RMS GCV score gradient at convergence was 105455022 .
The Hessian was positive definite.
Model rank =  34 / 34 

Basis dimension (k) checking results. Low p-value (k-index<1) may
indicate that k is too low, especially if edf is close to k'.

         k'   edf k-index p-value
s(T)  11.00  4.75    0.97    0.32
s(S)  11.00  7.41    1.06    0.74
s(Si) 11.00  3.72    0.92    0.10

As the p-values were quite high I thought that the model was working fine. Then, I looked at the plots and now I can't understand if my model is performing right or not, if it is acceptable or if I have to reject it..any advice would be of a great help and very appreciated.

Here are my plots from mod.check

Franc
  • 1
  • Residuals look skewly and heavy-tailed distributed; you may want to try out a transformation of your Totden variable. – Christian Hennig Oct 08 '20 at 12:13
  • Now it seems much better! Thank you very much! Can you recommend me readings about these kind of "problem solving"? – Franc Oct 08 '20 at 12:52
  • 1
    Unfortunately not, but I expect good books on regression and statistical modelling to say such things. – Christian Hennig Oct 08 '20 at 14:46
  • 1
    What are you planning on using the model for? There are quite a few things I would suggest you change but it would be helpful to know whether you just want to predict from the model or whether you are interested in the smooth effects themselves, for way of developing understanding or theory about the system you are modelling? – Gavin Simpson Oct 08 '20 at 15:36
  • The idea is to quantify the influence of environmental parameters on the abundance of different phytoplankton groups. So, this is the reason why I am interested in using this model. I am also interested in what is behind the model, but just for my own knowledge. Any suggestion would be would be very appreciated! – Franc Oct 12 '20 at 07:28

0 Answers0