On New Procedures of Estimation for Binary Data

In developmental toxity studies, current methods divide animals equally among all treatment groups. New procedures are introduced for estimating correlated binary data. Instead of allocating an equal number to each treatment, observe clusters one at a time until a desired numer of clusters have a chosen number of responses or more. Dose levels, or treatments, known to have many responses would not need as many animals. This procedure could save animals but would not sacrifice any information. Focusing on exchangeable binary data, a new procedure for estimating the probability of a response is investigated. This alternate design is analyzed through a simulation study and applied to a clinical data set. Comparisons are made between past estimators and the new estimator given