Seth
2010-Apr-11 23:27 UTC
[R] using varFunc in gnls to replicate Wedderburn-style variance?
Hi, I'm fitting a gnls on a response that varies from 0 to 1 using a basic logistic equation (1/(1+e^-bx)). I'm using gnls because this is spatial data with correlation in residuals well described by corSpher. Residual variance is compressed for fitted values near 0 and 1, and realtively constant in the 'middle'. Wedderburn suggested using (1-fitted value)*fitted value for nonlinear regression with data like mine (proportions), similar to what is expected for a binomial distribution. Is it possible to specify weight using varFunc so that the variance is a function of the fitted values? I've only seen examples for varFunc where covariates are used. Thanks, Seth -- View this message in context: http://n4.nabble.com/using-varFunc-in-gnls-to-replicate-Wedderburn-style-variance-tp1836537p1836537.html Sent from the R help mailing list archive at Nabble.com.