Skip to content

Confidence intervals in GAMM4

1 message · Highland Statistics Ltd

#
Hello,
Here is an alternative (and probably the only) approach:

1. Write your smoother as X * beta + Z * b (e.g. using an O'Sullivan spline)
2. Add more covariates to your predictor function (if needed)
3. Add random effects to the predictor function, and also an 
auto-regressive correlation structure on the residuals in the predictor 
function.
4. Put the whole thing in JAGS and let it run for a while

Plenty of papers are available for step 1...see for example:

ON SEMIPARAMETRIC REGRESSION WITH O?SULLIVAN PENALIZED SPLINES
M. P. WAND AND J. T. ORMEROD

They also provide R code for such smoothers. No need to dive into the 
underlying stats.


Steps 3 & 4 are described in our upcoming book
'Beginner's Guide to GLM & GLMM with R"
Zuur, Hilbe, Ieno

available next week

Other smoother options are described in 'A Beginner's Guide to GAM', 
Zuur (2012)
Or in Wood (2006), or Ruppert et al. (2003). Plus a whole bunch of 
papers from Wand.


Alain