R Graphical Manual

Browse All

Last data update: 2014.03.03

R: Temporal Process Regression

tpr	R Documentation

Temporal Process Regression

Description

Regression for temporal process responses and time-independent covariate. Some covariates have time-varying coefficients while others have time-independent coefficients.

Usage

tpr(y, delta, x, xtv=list(), z, ztv=list(), w, tis,
    family = poisson(),
    evstr = list(link = 5, v = 3),
    alpha = NULL, theta = NULL,
    tidx = 1:length(tis),
    kernstr = list(kern=1, poly=1, band=range(tis)/50),
    control = list(maxit=25, tol=0.0001, smooth=0, intsmooth=0))

Arguments

`y`	Response, a list of "lgtdl" objects.
`delta`	Data availability indicator, a list of "lgtdl" objects.
`x`	Covariate matrix for time-varying coefficients.
`xtv`	A list of list of "lgtdl" for time-varying covariates with time-varying coefficients.
`z`	NOT READY YET; Covariate matrix for time-independent coefficients.
`ztv`	NOT READY YET; A list of list of "lgtdl" for time-varying covariates with time-independent coefficients.
`w`	Weight vector with the same length of `tis`.
`tis`	A vector of time points at which the model is to be fitted.
`family`	Specification of the response distribution; see `family` for `glm`; this argument is used in getting initial estimates.
`evstr`	A list of two named components, link function and variance function. link: 1 = identity, 2 = logit, 3 = probit, 4 = cloglog, 5 = log; v: 1 = gaussian, 2 = binomial, 3 = poisson
`alpha`	A matrix supplying initial values of alpha.
`theta`	A numeric vector supplying initial values of theta.
`tidx`	indices for time points used to get initial values.
`kernstr`	A list of two names components: kern: 1 = Epanechnikov, 2 = triangular, 0 = uniform; band: bandwidth
`control`	A list of named components: maxit: maximum number of iterations; tol: tolerance level of iterations. smooth: 1 = smoothing; 0 = no smoothing.

Details

This rapper function can be made more user-friendly in the future. For example, evstr can be determined from the family argument.

Value

An object of class "tpr":

`tis`	same as the input argument
`alpha`	estimate of time-varying coefficients
`beta`	estimate of time-independent coefficients
`valpha`	a matrix of variance of alpha at tis
`vbeta`	a matrix of variance of beta at tis
`niter`	the number of iterations used
`infAlpha`	a list of influence functions for alpha
`infBeta`	a matrix of influence functions for beta

Author(s)

Jun Yan <jyan@stat.uconn.edu>

References

Fine, Yan, and Kosorok (2004). Temporal Process Regression. Biometrika.

Yan and Huang (2009). Partly Functional Temporal Process Regression with Semiparametric Profile Estimating Functions. Biometrics.