|
Activity Number:
|
424
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 12, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #300568 |
|
Title:
|
Multicriteria Inference for Process Models: Structural and Parametric Inference for a Stochastic Model of Feline Hematopoiesis
|
|
Author(s):
|
Joel H. Reynolds*+ and Daniela Golinelli
|
|
Companies:
|
U. S. Fish & Wildlife Service and RAND Corporation
|
|
Address:
|
1011 E. Tudor Rd., MS 221, Anchorage, AK, 99503,
|
|
Keywords:
|
Pareto optimization ; goodness-of-fit ; stem cells ; hidden Markov chain ; model structure
|
|
Abstract:
|
Stochastic process models live at the intersection of statistical and mechanistic modeling. Their assessment and calibration raise important questions regarding the appropriateness of statistical inference methods when one cannot implicitly assume a correct model specification. Such methods make parameter inference the primary objective and structural inference secondary, exactly the opposite of the early stages of process modeling. A stochastic model of feline hematoepiesis is used to compare inferences from the method of moments to those from a goodness-of-fit-based structural inference method, the Pareto Optimal Model Assessment Cycle (POMAC). Traditional statistical methods conduct parameter inference and then informally assess model structure adequacy at the selected parameterization. POMAC uses multicriteria optimization to directly assess model structure adequacy, with parameter inference a byproduct upon achieving an adequate model structure. The example model appears adequate with regard to the selected assessment criteria. Simulations based on the POMAC-based parameter estimates more closely mimic the experimental observations.
|
