chngpt (version 2016.7-9)

chngptm: Estimate change point logistic model

Description

Estimate change point logistic model

Usage

chngptm (formula.1, formula.2, family, data, type=c("step","hinge","segmented","stegmented"),  est.method=c("default","smoothapprox","grid"),  var.type=c("none","robust","model","smooth","robusttruth","bootstrap","all"), aux.fit=NULL, test.inv.ci=TRUE, lb.quantile=.1, ub.quantile=.9, grid.size=500,  ci.bootstrap.size=500, alpha=0.05, save.boot=FALSE, m.out.of.n=FALSE, b.=-30, tol=1e-4, maxit=1e2, chngpt.init=NULL,  weights=NULL, verbose=FALSE, ...) 
"coef" (object, ...)
"vcov" (object, ...)
"print" (x, ...)
"summary" (object, ...)

Arguments

formula.1

The part of formula that is free of terms involving thresholded variables

formula.2

The part of formula that is only composed of thresholded variables

data

data frame.

type

step: flat before and after change point; hinge: flat before and slope after change point; segmented: slope before and after change point

est.method

string. Better leave it at NULL. grid: grid search; smoothapprox: smooth approximation

var.type

string. Different methods for estimating covariance matrix and constructing confidence intervals

aux.fit

a model fit object that is needed for model-robust estimation of covariance matrix

grid.size

integer.

test.inv.ci

Boolean, whether or not to find test-inversion confidence interval for threshold

ci.bootstrap.size

integer, number of bootstrap

alpha

double, norminal type I error rate

save.boot

Boolean, whether to save bootstrap samples

b.

a parameter controlling approximation of step function by logistic function in optimization

lb.quantile

lower bound of the search range for change point estimate

ub.quantile

upper bound of the search range for change point estimate

tol

numeric. Stopping criterion on the coefficient estimate.

maxit

integer. Maximum number of iterations in the outer loop of optimization.

chngpt.init

numeric. Initial value for the change point.

family

passed to glm

weights

passed to glm

verbose

Boolean.

x

chngptm fit object.

object

chngptm fit object.

...

arguments passed to glm or coxph

m.out.of.n

whether to perform m out of n bootstrap

Value

A an object of type chngptm with the following components

converged

Boolean

coefficients

vector. Estimated coefficients. The last element, named ".chngpt", is the estimated change point

test

htest. Max score test results

iter

integer. Number of iterations

Details

Without lb.quantile and ub.quantile, finite sample performance of estimator drops considerably! When est.method is smoothapprox, Newton-Raphson is done with initial values chosen by change point hypothesis testing. The testing procedure may be less subjective to finite sample volatility.

References

Fong, Y., Huang, Y., Gilbert, P. (2015) Estimation and hypothesis testing under four types of change point models using chngpt, submitted.

Pastor-Barriuso, R. and Guallar, E. and Coresh, J. (2003) Transition models for change-point estimation in logistic regression. Statistics in Medicine. 22:13141

Examples

dat=sim.chngpt("sigmoid4", type="step", n=200, seed=1, beta=1, alpha=-1, x.distr="norm", e.=4)

fit.1=chngptm(formula.1=y~z, formula.2=~x, family="binomial", dat, type="step", est.method="grid")
print(fit.1)
summary(fit.1)

## Not run: 
# # not run because otherwise the examples take >5s and that is a problem for R CMD check
# # has interaction
# 
# library(survival)
# test1 <- data.frame(time=c(4,3,1,1,2,2,3), 
#               status=c(1,1,1,0,1,1,0), 
#               x=c(0,2,1,1,1,0,0), 
#               sex=c(0,0,0,0,1,1,1)) 
# fit=chngptm(formula.1=Surv(time, status)~1, formula.2=~x, data=test1, family="coxph", type="step")
# 
# ## End(Not run)