I recently happened upon a short memo-report I wrote back in 2007, for a consulting project. The project was about a medical device evaluation, where the device attempts to classify individuals into one of four disease states. The plan was to use the device on a number of subjects, and compare the classifications of the new device to those given by a "gold standard" method.
The problem may be viewed as one of estimating multinomial proportions. One of the main goals of the project was to help determine the necessary sample size.
I decided to post this report here because
It's a fairly smooth read and it might be interesting in its own right if someone is looking into a similar question.
It could be used as an example of the type of thing people would do in a statistical consulting project.
* It could be the basis of a "case study" type project if a student somewhere needs a practical situation involving categorical data analysis, sample size determination, and/or hypothesis testing.
Finally a disclaimer: I make no claims that the methods discussed in the report are the best way to handle the problem!
A screenshot of the opening section is given below. If you're interested you can download the full report pdf.