| This study was designed to explore whether we can use well understood and documented statistical methods to simultaneously decide which model, among statistical population models, best describes an animal population time series process. Specifically, I brought into question the utility of information criteria (IC forms) for making such decisions.;Ultimately, I wished to distinguish whether animal time series could be appropriately classified as density dependent or density independent, and whether the actual form of density-related process model may have generated a time series. These issues have deeply seated historical precedence in ecology and have a number of contemporary applications, such as population viability analysis.;In a comparative approach, 5 IC forms prevalent in the literature were used for this study. These IC forms were the Akaike IC, Hurvich and Tsai's IC, consistent IC, Bayesian IC, and Hannan and Quinn's IC. All forms estimate twice the negentropy of statistical models based on the likelihood of data. Each IC form can make simultaneous estimations of the negentropy a multitude of statistical models and do not require that models are related.;It was my contention that the question of how well IC forms could draw distinctions between models based on their negentropy was an empirical question. Forming a basis for an empirical approach were Monte Carlo methods. Using 117 animal time series, parameters were estimated under 6 population models; 5 parametric and 1 nonparametric forms. Parameters were used to generate new time series of lengths 8, 17, 35, 50, and 100. Likelihood calculations were made for each simulated time series and negentropy calculations were made with each IC form using each likelihood calculation. In all, 23,400 bootstrap iterations were carried out at each time series length and the estimated results of model selection, based on negentropy calculations, were compared to the known origin of the (simulated) time series on which they were based.;Results were tallied and measures were formed to evaluate the performance of each IC form. Performance measures were density dependent and independent model family selection, model family selection, rejection, and identification, a measure of how concordant model selection and rejection were, and the bias that may be shown by an IC form to overfit or underfit a model. All IC forms performed above expectation in ;I conclude that the BIC was the most consistently useful IC form for studies of this nature. I then used the BIC to classify 121 time series as to model family and model and explore associations with ecological variables of taxonomic classification, trophic classification, climatic source, and migratory status. The results were greatly heuristic but show that in virtually every category roughly 70-75% of the time series were identified as density dependent. |
