Estimation of the response-error relationship in immunoassay.

Abstract

Estimation of the response-error relationship in immunoassay provides a weighting function for the main analysis, and may in general be essential to ensure statistically valid data reduction. In this study we generated 50,000 sets of simulated radioimmunoassay response data with a computer, using five response-error functional forms that are commonly assumed. Parameters were estimated by three least-squares regression methods and three that are modifications of a maximum-likelihood method. Two likelihood estimators that require significantly different computing times were shown to be virtually indistinguishable, statistically more efficient than least-squares estimators, and-in contrast to least-squares estimators-to guarantee positive predicted variances in the range of the data.