Possible additional developments in no particular order

 - Score-based confidence intervals for glms

 - More general HT estimator for sampling without replacement
 	- using joint inclusion probabilities
	- using population first-order inclusion probabilities
	- using Overton's approximation and sample inclusion probabilities

 - Survival curve estimation (the standard errors are the only 
	non-trivial part of this)

 - Calibration to estimating functions to reproduce the semiparametric
   estimators of Robins, Rotnitzky and Zhao.

 - Large data sets: design object with references to data stored in a 
   database. Linear regression, but probably no other regression models.

 - GEE. 

 - imputation

 - Parametric survival models (you can do these with svymle)

 - linear mixed models?
