LOESS and predict.gam

[janicew@bumc.bu.edu]

Hello,

I am using Splus version 6.2 for windows. I have a generalized additive model with a bivariate LOESS smoother.  I am trying to predict values for new data points using the predict.gam function.  Occasionally I seem to be getting very different predicted values for some of the data points in the original data set when I include them in a new data set, but I'm not getting any error messages. Here are my questions:

1) When using LOESS in a gam model, does predict.gam use the old and new predictor points combined to come up with the neighborhoods around the target point?  I think this is true, but I'm not sure why this is done.

2) Does predict.gam use the old and new predictor values to assign weights used in the LOESS? 

3) I thought that if predict.gam gave a prediction that was very discrepant from the predicted value in the original data, that I would get a warning or error message.  Is this not always true?

 

I can send an example where this problem occurred if it would be helpful.  Thanks in advance for any help on this issue.

Sincerely,

Janice Weinberg