TABLE OF CONTENTS

estimation/updatedispmarg2 [ Functions ]

NAME

    updatedispmarg2 --- Marginal estimators for the dispersion parameters

FUNCTION

Compute dispersion parameter estimates using a marginal method, based on the moments of the event indicators delta under the auxiliary Poisson model.

SYNOPSIS

1134 updatedispmarg2 <- function(m, Ji, datamat, datamatd, alphars, alpharsd, as, asd, 
1135                         betahat, betadhat, fixzero, univariate = FALSE)

INPUTS

    m          number of clusters
    Ji         cluster sizes
    datamat    data matrix generated by makedatamat for event 1
    datamatd   data matrix generated by makedatamat for event 2
    alphars    matrix of baseline hazard parameters for event 1
    alpharsd   matrix of baseline hazard parameters for event 2
    as         matrix of discretization breakpoints for event 1
    asd        matrix of discretization breakpoints for event 2
    betahat    regression coefficient estimates for event 1
    betadhat   regression coefficient estimates for event 2
    fixzero    list of estimators that should be fixed at zero (testing only)
    univariate boolean indicator whether a univariate model is being fit

OUTPUTS

    sigma2hat      cluster-level dispersion for process 1
    sigma2dhat     cluster-level dispersion for process 2
    nu2hat             subject-level dispersion for process 1
    nu2dhat            subject-level dispersion for process 2
    thetahat       subject-level covariance

SOURCE

1138 {
1139     # The function getdispmarg3 makes the FORTRAN calls that do most of the work
1140     # out1 contains only terms involving event 1
1141     # out2 only the event 2, and out3 contains the cross-terms
1142     out1 <- getdispmarg3(m, Ji, datamat, datamat, alphars, alphars, 
1143                     as, as, betahat, betahat, TRUE)
1144     out2 <- getdispmarg3(m, Ji, datamatd, datamatd, alpharsd, alpharsd, 
1145                     asd, asd, betadhat, betadhat, TRUE)
1146     out3 <- getdispmarg3(m, Ji, datamat, datamatd, alphars, alpharsd, 
1147                     as, asd, betahat, betadhat, FALSE)
1148 
1149     # Extract the estimated dispersion parameters, and truncate from below
1150     sigma2hat <- max(out1$sig2, 1e-4)
1151     sigma2dhat <- max(out2$sig2, 1e-4)
1152     # Fix parameters at 0 if desired
1153     if(!is.null(fixzero)){
1154         if("sigma2hat"%in%fixzero) sigma2hat <- 0
1155         if("sigma2dhat"%in%fixzero) sigma2dhat <- 0
1156     }
1157     nu2hat <- max(out1$sig2nu2 - sigma2hat, 1e-4)
1158     nu2dhat <- max(out2$sig2nu2 - sigma2dhat, 1e-4)
1159     thetahat <- out3$sig2nu2
1160     # Fix parameters at 0 if desired
1161     if(!is.null(fixzero)){
1162         if("nu2hat"%in%fixzero) nu2hat <- 0
1163         if("nu2dhat"%in%fixzero) nu2dhat <- 0
1164         if("thetahat"%in%fixzero) thetahat <- 0
1165     }
1166     # The marginal estimators do not guarantee that the esimated covariance matrix
1167     # is valid. This code truncates the estimate of thetahat if needed
1168     if(!univariate)
1169         badtheta <- try(thetahat / sqrt(nu2hat * nu2dhat) > 1 | thetahat / 
1170             sqrt(nu2hat * nu2dhat)< -1)
1171     else badtheta <- FALSE
1172     if(inherits(badtheta, "try - error") | is.na(badtheta)) {browser(); return(0)}
1173     if(badtheta) thetahat <- try(sign(thetahat) * sqrt(nu2hat * nu2dhat))
1174     if(inherits(thetahat, "try - error")) return(0)
1175     
1176     # Prepare for output
1177     return(list(sigma2hat = sigma2hat,
1178                 sigma2dhat = sigma2dhat,
1179                 nu2hat = nu2hat,
1180                 nu2dhat = nu2dhat,
1181                 thetahat = thetahat))
1182 }