A survey analysis package has recently been added to R. It
has been written by Thomas Lumley of the University of
Washington. He has been very helpful in adapting the package
following our suggestions.
The survey package seems to be convenient from the point of
view of an experienced R user. We have developed the
exemplars with the version current in the first half of 2004.
The current version is 3.0 (May 2005). Details of
improvements from earlier versions can be found at the NEWS
link from the suvey package home
page.Further updates are expected.
The new version 3.2 is eaven more improved, easier to use,
and includes generalised calibration methods.
The software covers a wide range of survey analysis
methods. In particular: - mean, quantiles, variance, tables,
- generalised linear models (e.g. linear regression, logistic
regression, Poisson models, etc.)
- proportional hazards models
There is also a new set of procedures for non-response
weighting, but we have not had time to investigate them yet.
These methods are implemented with standard errors from
- Taylor linearisation in functions such as svydesign ,svymean and svyreg
- or replication methods such as balanced repeated
replication and the jackknife are available through
functions such as svyrepdesign, svrepmean and svrepreg
In earlier versions these were seperate functions. More
recent versions (2.9 and later) use the same functions for
both methods and R recognises which one to use from the
properties of the design you have declared.
(Click here for more info on
standard errors when designing surveys.)
The main source os Thomas Lumley's home
Manual pages for the survey package are available while
you work and can be read online here. A paper describing the package is
Lumley T. (2004) "Analysis of complex survey samples" Journal
of Statistical Software 9(8) available here