[5] Complicating effects of highly correlated model variables on nonlinear least-squares estimates of unique parameter values and their statistical confidence intervals: Estimating basal secretion and neurohormone half-life by deconvolution analysis

We discuss the methodology for evaluating the impact of highly correlated model variables on nonlinear least-squares parameter estimates and statistical confidence intervals. We illustrate that the joint parameter contours for hormone half-life and basal secretion rates (determined for 24-hr serum LH concentration profiles) do not necessarily conform to predictions of asymptotic standard errors. Consequently, nonlinear curve fitting of neuroendocrine data should employ rigorous methods for error propagation (e.g., Monte Carlo perturbations) that do not depend on the assumptions of linearity of the fitting function and the statistical independence of fitted parameters as required for asymptotic error theory. Moreover, whenever possible, experimentally independent estimates of model parameters should be obtained prior to nonlinear least-squares curve fitting of data arising from highly correlated model variables.