mets (version 0.2.4)

easy.binomial.twostage: Fits two-stage binomial for describing depdendence in binomial data using marginals that are on logistic form using the binomial.twostage funcion, but call is different and easier and the data manipulation is build into the function. Useful in particular for family design data.

Description

If clusters contain more than two times, the algoritm uses a compososite likelihood based on the pairwise bivariate models.

Usage

easy.binomial.twostage(margbin = NULL,
    data = sys.parent(), score.method = "nlminb",
    response = "response", id = "id", Nit = 60, detail = 0,
    silent = 1, weights = NULL, control = list(),
    theta = NULL, theta.formula = NULL, desnames = NULL,
    deshelp = 0, var.link = 1, iid = 1, step = 0.5,
    model = "plackett", marginal.p = NULL, strata = NULL,
    max.clust = NULL, se.clusters = NULL)

Arguments

margbin

Marginal binomial model

data

data frame

response

name of response variable in data frame

id

name of cluster variable in data frame

score.method

Scoring method

Nit

Number of iterations

detail

Detail for more output for iterations

silent

Debug information

weights

Weights for log-likelihood, can be used for each type of outcome in 2x2 tables.

control

Optimization arguments

theta

Starting values for variance components

theta.formula

design for depedence, either formula or design function

desnames

names for dependence parameters

deshelp

if 1 then prints out some data sets that are used, on on which the design function operates

var.link

Link function for variance

iid

Calculate i.i.d. decomposition

step

Step size

model

model

marginal.p

vector of marginal probabilities

strata

strata for fitting

max.clust

max clusters

se.clusters

clusters for iid decomposition for roubst standard errors

Details

The reported standard errors are based on the estimated information from the likelihood assuming that the marginals are known. This gives correct standard errors in the case of the plackett distribution (OR model for dependence), but incorrect for the clayton-oakes types model. The OR model is often known as the ALR model, but our fitting procedures gives correct standard errors and is quite a bit quicker.

Examples

data(twinstut)
twinstut0 <- subset(twinstut, tvparnr<2300000)
twinstut <- twinstut0
theta.des <- model.matrix( ~-1+factor(zyg),data=twinstut0)
margbin <- glm(stutter~factor(sex)+age,data=twinstut0,family=binomial())
bin <- binomial.twostage(margbin,data=twinstut0,
		         clusters=twinstut0$tvparnr,theta.des=theta.des,detail=0,
	                 score.method="fisher.scoring")
summary(bin)

twinstut0$cage <- scale(twinstut0$age)
theta.des <- model.matrix( ~-1+factor(zyg)+cage,data=twinstut0)
bina <- binomial.twostage(margbin,data=twinstut0,
		         clusters=twinstut0$tvparnr,theta.des=theta.des,detail=0,
	                 score.method="fisher.scoring")
summary(bina)

theta.des <- model.matrix( ~-1+factor(zyg)+factor(zyg)*cage,data=twinstut0)
bina <- binomial.twostage(margbin,data=twinstut0,
		         clusters=twinstut0$tvparnr,theta.des=theta.des,detail=0,
	                 score.method="fisher.scoring")
summary(bina)

twinstut0$binstut <- (twinstut0$stutter=="yes")*1
out <- easy.binomial.twostage(stutter~factor(sex)+age,data=twinstut0,
                              response="binstut",id="tvparnr",
			      theta.formula=~-1+factor(zyg1),
                              score.method="fisher.scoring")
summary(out)

desfs <- function(x,num1="zyg1",namesdes=c("mz","dz","os"))
    c(x[num1]=="dz",x[num1]=="mz",x[num1]=="os")*1

out3 <- easy.binomial.twostage(binstut~factor(sex)+age,
                               data=twinstut0, response="binstut",id="tvparnr",
                               score.method="fisher.scoring",
                               theta.formula=desfs,desnames=c("mz","dz","os"))
summary(out3)