Statistical modeling has evolved around building increasingly more complex models, even though it is common knowledge among statisticians that an optimal model size usually exists for any given data set. Having overly complex models leads to imprecise parameter estimates and tends to increase the subjective role of the modeler, which can distort the perceived characteristics of the system under investigation. One approach for controlling the tendency of contemporary models to increase in complexity and subjectivity is to use model selection criteria that account for these factors. The initial task of this thesis was to review existing model selection criteria. The second task involved testing the effectiveness of several model selection criteria. The Stock Synthesis program, which is often used on the U.S. west coast to assess the status of exploited marine fish stocks, was used for this evaluation because of its ability to handle multiple data sets and mimic highly complex population dynamics. In the review of existing model selection criteria the Akaike Information Criterion (AIC) and Schwarz's Bayesian Information Criterion (BIC) were identified as the criteria that most completely satisfied the fundamental principles of model selection: goodness-of-fit, parsimony, and objectivity. Their ability to select the correct model form and produce accurate parameter estimates was evaluated in Monte Carlo experiments with the Stock Synthesis program and were compared to a simple maximum log-likelihood criterion. The maximum log-likelihood criterion surprisingly outperformed both AIC and BIC in several of the experiments. / Graduation date: 1999
| Identifer | oai:union.ndltd.org:ORGSU/oai:ir.library.oregonstate.edu:1957/33648 |
| Date | 04 June 1998 |
| Creators | Helu, Siosaia Langitoto |
| Contributors | Sampson, David B. |
| Source Sets | Oregon State University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
Page generated in 0.014 seconds