Lawrence Joseph
Bayesian statistics
|
|
Please Note: Prof. Joseph has retired. These pages are left up in case they prove useful, but the pages and software will no longer be updated. All material and software is "as is" with no guarantees of functionality or correctness. |
|
binom_prop_misclass
Estimating binomial parameters from data subject to misclassification
Summary
This S-PLUS function computes the average coverage probabilities
of posterior credible sets of length l given prior information
on the sensitivity and specificity of a diagnostic test as well
as on the prevalence of the disease of the population. This
function accompanies the article titled:
Rahme E, Joseph L, and Gyorkos T. Bayesian Sample size determination for estimating binomial parameters from data subject to misclassification. Applied Statistics 2000;49(1):119-228. Basic setup
The basic setup is as follows: In planning a prevalence
study for a certain disease (or other similar misclassified
binomial data situation), an imperfect diagnostic test will
be used. Sample size estimates are required to estimate the
disease to an accuracy of l.
Prior information about the test is summarized by the following
beta densities for the test sensitivity and specificity:
sens ~ beta(sens.alpha, sens.beta) and spec ~ beta(spec.alpha, spec.beta) Prior information on the prevalence of the disease is also given by a beta density: prev ~ beta(prev.alpha, prev.beta) Given the above prior densities, as well as the desired length of the credible set (l), the function computes the average coverage probability over the predictive distribution of the data. In order to run the program, type: > avgcov(n, l, size, sens.alpha, sens.beta, spec.alpha, spec.beta, prev.alpha, prev.beta)where n is the sample size for which the average coverage is desired, l is the length of each credible interval, size is the size of the Monte Carlo simulation required for the calculations (larger valuse for size produce more accurate results at the expense of running time), and the rest of the parameters are beta prior parameters as defined above. The output will then be the average coverage probability for this input. Copyright
© Copyright Elham Rahme and Lawrence Joseph, 1997.
avgcov is a program written by Elham Rahme and Lawrence Joseph, at the Division of Clinical Epidemiology, Department of Medicine, Montreal General Hospital. This program is an implementation of the manuscript referenced below. You are free to use these programs, for non-commercial purposes only, under two conditions:
Please send any questions or comments to lawrence.joseph@mcgill.ca . |