Iteratively Reweighted Least Squares (IRWLS)
On the previous page all of the weighting schemes involved variance as a function of the value of the observed data point, except for the equal weight scheme. This is computationally efficient since each weight only needs to be calculated once at the beginning of the optimization step. It also ties the weight to the variance of the observed data points. However, when a weighting scheme is applied to a series of data points, data points with very low values may be given more emphasis than is appropriate. An alternative weighting has been used to try to overcome this disadvantage. This method is the iteratively reweighted least squares method. Effectively, this is identical to the methods on the previous pages except that observed data is replaced with calculated data. Thus, the weight is recalculated during each phase of the optimization process. Thus very low observed data points would not have the emphasis on the overall analysis. However, it is possible that the optimization may drive calculated values low, giving these points more emphasis and potentially distorting the final analysis.
Weighting Schemes Based on IRWLS
Equation 13.6.1 Variance Proportional to Calculated Data
Equation 13.6.2 Variance Proportional to the Square of Calculated Value
Equation 13.6.3 Variance as a Function of Calculated Value
Equation 13.6.4 Variance as a Function of Calculated Value with an Assay Sensitivity Term
Equation 13.6.5 Variance as a Function of Calculated Value with Time Information
References
- Peck, C.C., Sheiner, L.B. and Nichols, A.I. 1984 "The Problem of Choosing Weights in Nonlinear Regression Analysis of Pharmacokinetic Data", Drug Metabolism Reviews, 15, pp 133-148
return to the Course index
This page was last modified: Sunday, 28th Jul 2024 at 4:53 pm
Privacy Statement - 25 May 2018
iBook and pdf versions of this material and other PK material is available
Copyright © 2001-2022 David W. A. Bourne (david@boomer.org)