Sponsoring Section/Society: COPPS
Session Slot: 4:00- 5:50 Wednesday
Estimated Audience Size: xx-xxx
AudioVisual Request: xxx
Session Title: Fisher Lecture
Theme Session: Yes/No
Applied Session: Yes/No
Session Organizer: TBN
Address:
Phone:
Fax:
Email:
Session Timing: 110 minutes total (Sorry about format):
110 minutes total...please allocate Opening Remarks by Chair - 5 or 10 minutes Speaker - 30 or 45 minutes Discussant - 10 minutes (or none) Discussant - 10 minutes (or none)
Session Chair: TBN
Address:
Phone:
Fax:
Email:
1. Logicist Statistics: Modeling and Inference
Dempster, Arthur P., Harvard University
Address: Dept of Stat, Science Center Harvard Univ Cambridge, MA 02138-2901
Phone: 617-495-5601
Fax:
Email: dempster@stat.harvard.edu
Abstract: An alternative view of the practice of statistics is presented in this lecture, stressing roles for both formal and informal reasoning about specific situations. Whereas most expositions of our subject put a dominating focus on procedures for data gathering, data analysis, and inference that emphasizes their mathematical, computational, and reporting aspects, the approach called here logicist stresses that procedures are also packaged forms of reasoning, (i.e., ``logic'' in a broad sense), and insists that for applications such logic is the sine qua non of statistical analysis. Formal statistical reasoning depends on formal models that may be (i) ``empirical'' based on data analysis, or (ii) ``stochastic'' representing repetitive phenomena such as sampling, measurement, or temporal-spatial variation, or (iii) ``predictive'' aimed at probabilistic reasoning about unknowns. These types are strongly overlapping. The basic inferential mechanisms are postdictive interpretation of model-dependent probabilities associated with observed outcomes, as with significance tests or likelihood ratio interpretations, and prospective interpretation of probabilities of unknown factual assertions computed from fusion of data and other information, as with interpretation of Bayesian or belief function posteriors. Procedures for model assessment from data, including model testing, selection, and sensitivity are important, but are only half the story of model construction. Independence judgments, and prior understanding and knowledge, are also fundamental. The logicist attitude to probability and to statistical science originates in the scientific outlook of R. A. Fisher, and is sharply different from narrower ``frequentist'' or ``Bayesian'' schools which neither individually nor jointly support enough of inferential logic to carry us forward into a future of burgeoning technological complexities.
Address:
Phone:
Fax:
Email:
Address:
Phone:
Fax:
Email:
List of speakers who are nonmembers: None