Model selection uncertainty and parameter estimation of non linear growth models

Abstract

The study is to discuss the application of nonlinear growth models to measure the growth data and the selection of best model for growth prediction among the compiting candidate models. Six nonlinear growth first functions were first fitted to the South African population data. The nonlinear distribution functions were fitted using iterative method, so that the process is repeated optimized using a predefined stopping rule. The method requires specification of the starting values of the parameters to be estimated, making it more difficult than the linear models. The second objective is to explain and illustrate a method, which interface information theory and mathematical statistics for selection of an estimated best approximate model. An approximating AIC weight is proposed instead of raw AIC or BIC for model selection for the non-nested candidate models. For the population growth forecasts, it was found that the empirical distributions performed well as traditional times series polynomial models