To use the code load it with the source() command. All the functions take a glm object as the main argument and return either a covariance matrix or a list containing a covariance matrix, bias correction and approximate degrees of freedom. Complete on-line documentation in HTML, S-PLUS or R format is available from:
http://www.biostat.washington.edu/~thomas/weave.html

jack.glm(glm.object,groups),infjack.glm(glm.object,groups)

Variance estimates for independence working models fitted to clustered data. The first is based on the one-step jackknife, the second is the GEE sandwich estimator based on the infinitesimal jackknife.

newey.west.glm(glm.object,times,lag,clag)

Newey-West estimator for estimating equations fitted to time seriesdata. The truncation lag (lag=) optimally increases as the cube root of sample size, you can specify the multiplier (clag=) instead of lag=.

kernelvar.glm<-function(glm.obj,times, lag=NULL,clag=NULL, kernel=c("tukey","bartlett","parzen"))

Estimator based on specified weight function: "bartlett" gives Newey-West estimator, other two are better. If clag= is specified the truncation lag is clag*n^a where a=1/3 for "bartlett", 1/5 for the other kernels.

weightvar.glm<-function(glm.obj,times,weights)

Roll your own. Specify a vector of weights for lags 0:M weave.trunc(glm.obj,times, lag=NULL, ctrunc=4)weighted empirical adaptive variance estimate with weights (n*rho^2>ctrunc) or with weights 1 up to lag=lag, 0 further away weave.smooth(glm.obj,times,csmooth=1) weighted empirical adaptive variance estimate with weightsmin(1,csmooth*N*rho^2)

lele.glm(glm.obj,times,lag)

One-step version of Lele's estimating equation jackknife for time series.

xeffect.glm(glm.obj,g1,g2)

Sandwich standard errors for a crossed design where observations are correlated if they share the same value of g1 or of g2. Unbalanced or incomplete designs are fine.

contact address:

T. Lumley
Department of Biostatistics
University of Washington
Box 357232
Seattle
WA 98195-7232
USA