sfTDist: 4.8: t-distribution Spending Function

Description

The function sfTDist() provides perhaps the maximum flexibility among spending functions provided in the gsDesign package. This function allows fitting of three points on a cumulative spending function curve; in this case, six parameters are specified indicating an x and a y coordinate for each of 3 points. Normally this function will be passed to gsDesign() in the parameter sfu for the upper bound or sfl for the lower bound to specify a spending function family for a design. In this case, the user does not need to know the calling sequence. The calling sequence is useful, however, when the user wishes to plot a spending function as demonstrated below in examples.

Usage

sfTDist(alpha, t, param)

Arguments

alpha

Real value $> 0$ and no more than 1. Normally, alpha=0.025 for one-sided Type I error specification or alpha=0.1 for Type II error specification. However, this could be set to 1 if for descriptive purposes you wish to see t

A vector of points with increasing values from 0 to 1, inclusive. Values of the proportion of sample size/information for which the spending function will be computed.

param

In the three-parameter specification, the first paramater (a) may be any real value, the second (b) any positive value, and the third parameter (df=degrees of freedom) any real value 1 or greater. When gsDesign() is called with a t-distribu

Value

An object of type spendfn. See spending functions for further details.

Details

The t-distribution spending function takes the form

f (t; α) = α F (a + b F^{- 1} (t))

where $F()$ is a cumulative t-distribution function with df degrees of freedom and $F^{-1}()$ is its inverse.

References

Jennison C and Turnbull BW (2000), Group Sequential Methods with Applications to Clinical Trials. Boca Raton: Chapman and Hall.

Examples

Run this code

# 3-parameter specification: a,  b,  df
sfTDist(1, 1:5/6, c(-1, 1.5, 4))$spend

# 5-parameter specification fits 2 points,  in this case
# the 1st 2 interims are at 25\% and 50\% of observations with
# cumulative error spending of 10\% and 20\%, respectively
# final parameter is df
sfTDist(1, 1:3/4, c(.25, .5, .1, .2, 4))$spend

# 6-parameter specification fits 3 points
# Interims are at 25\%. 50\% and 75\% of observations
# with cumulative spending of 10\%, 20\% and 50\%, respectively
# Note: not all 3 point combinations can be fit
sfTDist(1, 1:3/4, c(.25, .5, .75, .1, .2, .5))$spend

# Example of error message when the 3-points specified 
# in the 6-parameter version cannot be fit
try(sfTDist(1, 1:3/4, c(.25, .5, .75, .1, .2, .3))$errmsg)

# sfCauchy (sfTDist with 1 df) and sfNormal (sfTDist with infinite df)
# show the limits of what sfTdist can fit 
# for the third point are u3 from 0.344 to 0.6 when t3=0.75
sfNormal(1, 1:3/4, c(.25, .5, .1, .2))$spend[3]
sfCauchy(1, 1:3/4, c(.25, .5, .1, .2))$spend[3]

# plot a few t-distribution spending functions fitting
# t=0.25, .5 and u=0.1, 0.2
# to demonstrate the range of flexibility
t <- 0:100/100
plot(t, sfTDist(0.025, t, c(.25, .5, .1, .2, 1))$spend, 
    xlab="Proportion of final sample size", 
    ylab="Cumulative Type I error spending", 
    main="t-Distribution Spending Function Examples", type="l")
lines(t, sfTDist(0.025, t, c(.25, .5, .1, .2, 1.5))$spend, lty=2)
lines(t, sfTDist(0.025, t, c(.25, .5, .1, .2, 3))$spend, lty=3)
lines(t, sfTDist(0.025, t, c(.25, .5, .1, .2, 10))$spend, lty=4)
lines(t, sfTDist(0.025, t, c(.25, .5, .1, .2, 100))$spend, lty=5)
legend(x=c(.0, .3), y=.025*c(.7, 1), lty=1:5, 
    legend=c("df = 1", "df = 1.5", "df = 3", "df = 10", "df = 100"))

Run the code above in your browser using DataLab

Get 50% off unlimited learning