rp.likelihood {rpanel}

Interactive inspection of one- or two-parameter likelihood surfaces

Description

This function plots a likelihood surface for a model with one or two parameters. It also creates a panel which allows the maximum likelihood estimate, a confidence region and other objects of interest to be added to the plot. For one-parameter models, the tkrplot package is required. For two-parameter models the rgl package is required.

Usage

rp.likelihood(loglik.fn, data, theta.low, theta.high, form = "log-likelihood", 
                  hscale = NA, vscale = hscale)

Arguments

loglik.fn

This should be either the name of a function, with arguments theta and data, or R code, in text form, which evaluates the log-likelihood function. The latter form allows simple R expressions such as sum(log(dexp(data, theta))) or sum(log(dgamma(data, theta[1], theta[2]))) to be used to define the log-likelihood.

data

an object which contains the data. This will be referred to in likelihood contributions.

theta.low

a vector of length one or two which defines the lower limit(s) of the parameter values for initial plotting.

theta.high

a vector of length one or two which defines the upper limit(s) of the parameter values for initial plotting.

form

a text variable which determines whether the likelihood or log-likelihood function is to be plotted. This applies only to one-parameter models. With two-parameter models, only the log-likelihood is plotted.

hscale, vscale

scaling parameters for the size of the plot when there is one covariate. The default values are 1 on Unix platforms and 1.4 on Windows platforms.

Details

The interactive controls allow a variety of aspects of the plots to be altered. This is intended to allow students and lecturers to explore likelihood surfaces in a manner which promotes an intuitive understanding of the concepts involved.

In the case of one parameter, the vertical axes of the (log-)likelihood plot can be clicked and grabbed to alter the plotting region interactively. This can be useful, in particular, in identifying the maximum likelihood estimator graphically.

Value

Nothing is returned.

References

rpanel: Statistical cartoons in R. MSOR Connections, 7, 3-7.

rpanel: Simple interactive controls for R functions using the tcltk package. Journal of Statistical Software, 17, issue 9.

Examples

## Not run: 
   rp.likelihood("sum(log(dexp(data, theta)))", aircond, 0.005, 0.03)
   rp.likelihood("sum(log(dgamma(data, theta[1], theta[2])))",
        aircond, c(0.3, 0.005), c(3, 0.06))

## End(Not run)