Nonparametric logistic regression

Usage

sm.logit(x, y, N=rep(1, length(y)), h, ngrid=25, eval.points, 
         add=F, display="estimate", xlab, ylab, pch=1, col=2, ...)

Arguments

x vector of the covariate values
y vector of the response values; they must be nonnegative integers.
h the smoothing parameter; it must be positive.
N a vector containing the binomial denominators. If missing, it is assumed to contain all 1's.
ngrid the number of points where the regression curve must be estimated (only used if eval.points is not given).
eval.points the vector of points on the x axis where the regression must be estimated. If the parameter eval.points is not given, this vector is chosen to be formed by ngrid equally spaced points between min(x) and max(x).
add if graphical output is produced, this parameter controls whether a new plot is created, or graphical output is added to the existing one.
display controls the type of graphical output; possible values are "estimate" (default), "se", `"none".
xlab label of the x-axis
ylab label of the y-axis
pch plotting character of the raw observed frequency.
col colour used for plotting curves and points
... additional graphical parameters

Description

This function estimates the regression curve using the local likelihood approach for a vector of binomial observations and an associated vector of covariate values.

Details

see Sections 3.4 and 5.4 of the reference below.

Value

A list containing vectors with the evalutation points, the corresponding probability estimates, the linear predictors, the upper and lower points of the variability bands (on the probability scale) and the standard errors on the linear predictor scale.

Side Effects

graphical output will be produced, depending on the value of the display parameter, unless this is set to "none".

References

Bowman, A.W. and Azzalini, A. (1997). Applied Smoothing Techniques for Data Analysis: the Kernel Approach with S-Plus Illustrations. Oxford University Press, Oxford.

See Also

sm.logit.bootstrap, sm.poisson, sm.poisson.bootstrap

Examples

# the next example assumes that all binomial denominators are 1's
sm.logit(dose, failure, h=0.5)
# in the next example, (some of) dose levels are replicated 
sm.logit(dose, failure, n.trials, h=0.5)


[Package Contents]