Life Sciences Seminar Series
|Name:||Jay Ver Hoef|
|Statistician. National Marine Mammal Laboratory, National Marine Fisheries Service, NOAA.|
|Title:||The Practical Benefits of Single Model Inference|
|Date:||Friday, 5 December 2014|
|Location:||Murie Life Science Bldg, Murie Auditorium.|
Multimodel inference accommodates uncertainty when selecting or averaging models, which seems logical and natural. However, there are costs associated with multimodel inferences, so they are not always appropriate or desirable. First, I present statistical inference in the big picture of data analysis and the deductive-inductive process of scientific discovery. Against this backdrop, I distinguish inferences for fixed states versus processes of nature. I show that inference for a fixed state of nature generally favors using a single model, which includes survey sampling methods, purely subjective Bayesian analyses, designed experiments, and cases where objectives define a highly aggregated process in nature. Multimodel inference is used primarily when modeling processes of nature. Sometimes the study objective is a model comparison or selection; naturally, multimodel inference methods should then be used. However, in other cases, a single model to meet an objective may be better. Researchers should consider model diagnostics, which are not easily implemented for multiple models. The analyst should recognize the importance of model diagnostics for discovering new features in the data in comparison to maintaining global inference probabilities. There is also additional cost, generally in terms of time, for multimodel inference. The decision to use multimodel inference methods, versus a single model, is an important one for management, as multimodel inference can take more time and resources when the objective may have been achieved adequately by a single model.
About the Speaker:
Jay Ver Hoef began his career as a statistician with the Alaska Department of Fish and Game, after receiving a co-major Ph.D. in statistics and ecology and evolutionary biology from Iowa State University. He now works as a statistician for a research lab, the National Marine Mammal Laboratory within the National Marine Fisheries Service, NOAA. Dr. Ver Hoef develops spatial statistical methods for animal abundance estimation and data on stream networks, and he consults on a wide variety of statistical methods related to marine mammals. Dr. Ver Hoef is a fellow of the American Statistical Association (ASA) and past-Chair of the Section on Statistics and the Environment of ASA, with over 100 publications, and he is a co-author of one book on spatial statistics.
Browse Life Sciences SeminarsThe fall 2014 faculty coordinators for this seminar series are Tamara Harms and Robert "Trey" Coker. The staff coordinator is Marie Thoms. Beginning in 2013, many of the seminars were recorded and can be viewed online. Speakers are listed in chronological order within academic years.
**Sept-Oct 2014: Problems with the university recording system have delayed the posting of seminar recordings. We hope to have this fixed by the end of October.Beginning in 1966 and continuing today, IAB hosts a weekly seminar for faculty, students, staff and the public during the academic year. The series attracts life scientists from Alaska and around the world.
If you wish to meet with a particular speaker, please contact the IAB director's office at 907-474-7649.
- 9/5/14 (Ken Cameron)
- 9/12/14 (Perry Barboza)
- 9/17/14 (John R. Speakman)
- 9/26/14 (Michael Harris)
- 10/3/14 (Heiko U. Wittmer)
- 10/10/14 (Barbara Taylor)
- 10/17/14 (Stacy Rasmus)
- 10/24/14 (Andrea Bersamin)
- 10/31/14 (Link Olson)
- 11/7/14 (Kerry L. Nicholson)
- 11/14/14 (David J. Meltzer)
- 11/21/14 (Teresa Hollingsworth)
- 12/5/14 (Jay Ver Hoef)