R: Weighted Empirical Adaptive Variance Estimators for Time Series

Weighted Empirical Adaptive Variance Estimators for Time Series

Usage

weave.trunc(glm.obj, times, lag=NULL, ctrunc=4)
weave.smooth(glm.obj, times, lag=NULL, csmooth=4)

Arguments

`glm.obj`	A generalised linear model object
`times`	Vector of times at which responses were observed
`lag`	Maximum lag for variance estimator
`ctrunc`	Tuning constant for automatic choice of lag
`csmooth`	Tuning constant for automatic choice of lag

Description

Weighted Empirical Adaptive Variance Estimators (WEAVEs) give standard error estimates that are robust against model misspecification, autocorrelation and moderate non-stationarity in a time series. They are closely related to the information sandwich estimators used for gee models, but incorporate weights to allow them to work in a single time series without independent replicates. The weights are chosen based on the estimated autocorrelation function. There is usually no reason to specify the lag argument.

The estimators incorporate a small-sample bias correction and an approximate degrees of freedom for a t-distribution.

Value

`var`	estimated covariance matrix without bias correction
`bias.correction`	multiplicative bias correction
`df`	Approximate denominator degrees of freedom

Author(s)

Thomas Lumley

References

Lumley and Heagerty "Weighted Empirical Adaptive Variance Estimators for Correlated Data Regression", (in press) JRSSB