Sadler W A, Smith M H
Comput Biomed Res. 1987 Feb;20(1):1-11. doi: 10.1016/0010-4809(87)90013-9.
A computer program is described for estimating a simple one or two parameter variance function using approximate conditional likelihood. The inverse of the fitted function provides a smooth weighting function for use in subsequent manipulation of the data. The program is designed for immunoassays, where both the raw experimental observations, and the final results from data reduction, are well known to exhibit nonuniformity of variance, but it may also have application in other areas where the observations follow approximately normal distributions and a weighting function is required. Five variance function forms are optionally available as subroutines, and if necessary, one or more of these can be incorporated into existing software with trivial modification. The test of fit of the assumed function provides an objective criterion for detecting outliers, and the program includes automatic outlier exclusion as an option.
