Evaluation of statistical methodologies for estimation of median effective dose.

A comparison is made of statistical techniques currently in use for analysis of quantal assay data. The discussion focuses on the choice of the functional model to use in the description of the tolerance distribution for the estimation of the median effective dose and in the contrast between exact and graphical methods of estimation for a given choice of model. The data used to illustrate these focal points are derived from test data from biological assays performed at the University of Pittsburgh. A criterion for making a choice among functional models based on the significance of the dose-response relationship and the length of the fiducial limits for the median effective dose is proposed. It is also shown that graphical methods can lead to results similar to those arrived at through exact calculations where the data fit model assumptions but can lead to erroneous conclusions where this is not the case.