Fitting response curves for radioimmunoassays or immunoradiometric assays.

The paper describes non-linear regression methods for the evaluation of radioimmunoassay or immunoradiometric assay data. The underlying model is an overdispersed Poisson process with four regression line parameters and one parameter related to the overdispersion of the variance. A generalized least-squares algorithm is described for the parameter estimation of non-contaminated data. In the presence of outliers in Y-direction, the results are improved by a winsorized version of the generalized least-squares method.