s-news
[Top] [All Lists]

Pointwise variances in GAM

To: "'s-news@lists.biostat.wustl.edu'" <s-news@lists.biostat.wustl.edu>
Subject: Pointwise variances in GAM
From: "Macneill, Stephanie J" <s.macneill@imperial.ac.uk>
Date: Tue, 18 Feb 2003 16:08:53 -0000
Dear All,

I am new to generalized additive models in Splus and was wondering if
someone could offer me some advice on a problem I have run into. I recently
ran the following model in Splus 6.0 :

gam.model1<-gam(formula = var1 ~ var2 + var3 + var4 + lo(var5, span = 0.5),
family = binomial(link = logit),data = minimodel, na.action = na.exclude,
control = list(epsilon = 0.001, bf.epsilon = 0.001, maxit = 50, bf.maxit =
10, trace = F))

When I tried to look at the pointwise variances in GAM using gam.model1$var,
I found three points in my dataset (n=492) where the pointwise variance was
negative. I have re-run the model with span=0.6, 0.7, 0.8 and 0.9; each time
I find a few negative variances. This seems impossible to me and I was
wondering if someone could explain how this could occur and what it means.
Any and all advice would be sincerely appreciated.

Kind regards,

Stephanie MacNeill

<Prev in Thread] Current Thread [Next in Thread>
  • Pointwise variances in GAM, Macneill, Stephanie J <=