condGEE {condGEE} | R Documentation |
Solves for the mean parameters (\theta), the variance parameter (\sigma^2), and their asymptotic variance in a conditional GEE for recurrent event gap times, as described by Clement, D. Y. and Strawderman, R. L. (2009) Biostatistics 10, 451–467. Makes a parametric assumption for the length of the censored gap time, and assumes gap times within subject are conditionally uncorrelated.
condGEE(data, start, mu.fn=MU, mu.d=MU.d, var.fn=V, k1=K1.norm, k2=K2.norm, robust=TRUE, asymp.var=TRUE, maxiter=100, rtol=1e-6, atol=1e-8, ctol=1e-8, useFortran=TRUE)
data |
matrix of data with one row for each gap time; the first column should be a subject ID, the second column the gap time, the third column a completeness indicator equal to 1 if the gap time is complete and 0 if the gap time is censored, and the remaining columns the covariates for use in the mean and variance functions |
start |
vector containing initial guesses for the unknown parameter vector |
mu.fn |
the specification for the mean of the gap time; the default is a linear combination of the covariates; the function should take two arguments (\theta, and a matrix of covariates with each row corresponding to one gap time) and it should return a vector of means |
mu.d |
the derivative of mu.fn with respect to the parameter vector; the default corresponds to a linear mean function
|
var.fn |
the specification for V^2, where the variance of the gap time is \sigma^2 V^2; the default is a vector of ones; the function should take two arguments (\theta, and a matrix of covariates with each row corresponding to one gap time) and it should return a vector of variances |
k1 |
the function to solve for the conditional mean length of the censored gap times; its sole argument should be the vector of standardized (i.e.\ (Y-\mu)/(\sigma V)) censored gap times; the default assumes the standardized censored gap times follow a standard normal distribution, but K1.t3 and K1.exp are also provided in the package - they assume a standardized t with 3 degrees of freedom and an exponential with mean 0 and variance 1 respectively
|
k2 |
the function to solve for the conditional mean length of the square of the censored gap times; its sole argument should be the vector of standardized (i.e.\ (Y-\mu)/(\sigma V)) censored gap times; the default assumes the standardized censored gap times follow a standard normal distribution, but K2.t3 and K2.exp are also provided in the package - they assume a standardized t with 3 degrees of freedom and an exponential with mean 0 and variance 1 respectively
|
robust |
logical, if FALSE , the mean and variance parameters are solved for simultaneously, increasing efficiency, but decreasing the leeway to misguess start and still find the root of the GEE
|
asymp.var |
logical, if FALSE , the function returns NULL for the asymptotic variance matrix
|
maxiter |
see multiroot ; maximal number of iterations allowed
|
rtol |
see multiroot ; relative error tolerance
|
atol |
see multiroot ; absolute error tolerance
|
ctol |
see multiroot ; if between two iterations, the maximal change in the variable values is less than this amount, then it is assumed that the root is found
|
useFortran |
see multiroot ; logical, if FALSE , then an R implementation of Newton-Raphson is used
|
Uses the function multiroot
in the rootSolve
package to solve the conditional GEE. As in multiroot
, there is no guarantee of finding the root.
A monotone increasing transformation can be applied to the observed gap times before calling condGEE
.
When robust=TRUE
, \theta and \sigma^2 are solved for in an alternating fashion until convergence. Note that the estimating equation for the mean parameters depends on \sigma^2 through the censored gap time.
a list containing:
eta |
the parameter estimate (\theta^T,\sigma^2)^T |
a.var |
an estimate of the asymptotic variance matrix of the eta estimator |
David Clement <dyc24@cornell.edu>
Clement, D. Y. and Strawderman, R. L. 2009 Biostatistics 10, 451–467.
data(asthma) demo(asthmaExample)