Boot.CI-methods {EffectiveDose}R Documentation

Bootstrap confidence intervals for ED levels

Description

gives the Bootstrap confidence interval for the effective dose level.

Usage

Boot.CI(fitprob, alpha, ...)

Arguments

fitprob an object either of class "list", "locfit", or "locpoly".
alpha alpha-levels where the bootstrap confidence intervals for effecitve dose levels is to be estimated. Missing values are not accepted.
... further parameters see details.

Details

The confidence level of the bootstrap interval can be fixed through the argument "level". The default is level=0.95.

The argument "bandwidth" is a single smoothing parameter of class "numeric" to evaluate the effective dose level. If missing a kind of Silverman's rule of thumb is used.

With the argument "R", the number of bootstrap replications can be determined. By default, the procedure uses R=100.

The argument "N" specifies the gridsize to approximate the integral to obtain the effective dose level. However, do not use too large values for N. By default N=101 is used.

The argument "mono" specifies if the effective dose level is assumed to be monotone increasing or decreasing. The default value is mono="increasing", whereas in toxicology applications mono="decreasing" has to be used in order to get reasonable results.

The argument "type" can be used if the data consists of continuous measurements and the user still wants to the quantiles. Default is type="cont". To obtain .5 quantile for continuous data use type="prob".

Bootstrap confidence intervals might give inaccurate lower and upper bounds of the confidence interval, if the underlying dose response curve "p" is rather flat. The function Boot.CI gives a warning if the estimated derivative of the dose response curve is smaller than 0.1 in absolute values.

Value

An object of class ED.Boot.CI.

Author(s)

Regine Scheder Regine.Scheder@rub.de

See Also

ED, ED.Boot.CI.

Examples

#Same model as discussed for the function ED
ybin=function(x){
        n=length(x)
        y=numeric(n)
        p=pnorm(x, mean=0.5, sd=0.5)
        for(i in 1:n){
        y[i]=rbinom(1,1,prob=p[i])
        }
        return(y)
}

x=seq(0,1,length.out=50)
y=ybin(x)

fit=locfit(y~lp(x,deg=1, h=0.1, nn=0))
fit2=locpoly(x,y, degree=1, bandwidth=0.1)

#For each usage of the function Boot.CI a Bootstrap confidence interval is computed

Boot.CI(list(x,y), alpha=0.5)
Boot.CI(fit,alpha=0.5)
Boot.CI(fit2,alpha=0.5)

#In this example, it is not useful to compute the Bootstrap confidence intervals close to the boundaries, since the function p is very flat
ybin=function(x){
        n=length(x)
        y=numeric(n)
        p=pnorm(x, mean=0.5, sd=0.05)
        for(i in 1:n){
        y[i]=rbinom(1,1,prob=p[i])
        }
        return(y)
}
#The following plot shows how flat the function p is at the bounds.
plot(x,pnorm(x, mean=0.5, sd=0.05), type="l")

x=seq(0,1,length.out=50)
y=ybin(x)

fit=locfit(y~lp(x,deg=1, h=0.1, nn=0))
fit2=locpoly(x,y, degree=1, bandwidth=0.1)

#For each usage of the function Boot.CI a Bootstrap confidence interval is computed and the upper and lower limits are returned followed by a warning.

Boot.CI(list(x,y), alpha=0.2)
Boot.CI(fit,alpha=0.2)
Boot.CI(fit2,alpha=0.2)


[Package EffectiveDose version 1.0-6 Index]