Estimation and comparison of parameters in stochastic growth models for barn owls

Abstract

Alternative methods for parameter estimation and the incorporation of stochasticity into growth models are investigated and compared to the commonly used sampling error model in which error terms are simply added-on to the integrated form of the growth equation. A process error model, in which the process of growth is assumed to have stochastic variation or error incorporated within it, was found to be more appropriate for use with nonlinear estimation procedures based on a minimization of sigma ei2. The process error model tended to minimize and/or eliminate the autocorrelation of residuals, which were characteristic of the sampling error model. These analyses further suggest that while the commonly used sampling error growth model may indeed provide unbiased parameter estimates, the estimated variances of such estimates are likely to be unwarrantedly low, thus raising questions as to the validity of any statistical comparisons based on such analyses. The procedure is illustrated with growth data from captive-reared sibling nestling barn owls, using the Richards' growth curve. These analyses suggest that both growth rate and growth form are subject to a higher degree of genetic control than is asymptotic weight which showed a greater tendency to vary according to the hatching order of the nestlings.