I have a quick question about the function gam.
We are using the GAM function to model time series data, with the code:
gam(formula = y ~ s(x1)+*, family=quasi(log,mu))
We have serial correlation in the residuals. It's been suggested to us that the
effect of serial correlation on the standard errors of the parameter estimates
can be corrected by fitting a quasi-likelihood model. We understand that the
effect of overdispersion is adequately corrected by using quasi-likelihood
estimation, but are unsure about correction for serial correlation.
Does the S+ code above provide the robust estimates of SE(beta) from data with
serially correlated residuals?
ta
Doug
dougl@nrhs.health.nsw.gov.au
|