Message-ID: <alpine.LRH.2.01.1001191011040.13781@hymn11.u.washington.edu>
Date: 2010-01-19T18:11:04Z
From: Thomas Lumley
Subject: Sampling theory
In-Reply-To: <Pine.LNX.4.64.1001191753280.13845@egon.stats.ucl.ac.uk>
On Tue, 19 Jan 2010, Christian Hennig wrote:
> are there any R-packages for computations required in sampling theury (such
> as confidence intervals under random, stratified, cluster sampling; I'd be
> partoculary interested in confidence intervals for the population variance,
> which is difficult enough to find even in books)?
>
Yes, these are in the survey package, for fairly general designs, using linearization or replicate weights.
I don't know how good the confidence intervals for the variance are. One of the disadvantages of implementing survey estimators in a general way is that you lose the opportunity to use bias corrections that are only available for simple cases.
The forthcoming version 3.19 (later this week) has nicer output for the population variance, but the computations are still the same.
-thomas
Thomas Lumley Assoc. Professor, Biostatistics
tlumley at u.washington.edu University of Washington, Seattle