Calibrating Expectations

In statistical modeling, assessing how closely the real data and the assumed model match is important to understand the validity of inferences drawn from the model. Reading diagnostic plots to correctly detect real issues with model assumptions and understand when patterns of results might not be unexpected if the model assumptions are true (which is never known) can be difficult for students as well as experienced researchers. This talk discusses enhanced and interactive diagnostic plots and a way to easily employ a calibration approach that involves simulating new observations from the estimated model (where all assumptions are true) to compare to the observed results. This approach leads to a new way to calibrate a numeric model diagnostic, Cook's Distance, which is a measure of influence of observations on a model. With some modifications, this same idea leads to a new approach to assessing influence of observations for dimension-reduction techniques such as multidimensional scaling and other mapping techniques. This talk presents results from a sabbatical in 2022-2023.  As an application, we consider how duality between curves and points can be used to construct nonlinear Kakeya sets.

refreshments will be available after the presentation. 

All are welcome to attend. This Statistics Seminar is co-sponsored by MSU Student Chapter of the ASA and the Montana Chapter of the ASA. 

Webex virtual link:   Meeting link:
https://montana.webex.com/montana/j.php?MTID=mc49fcb2f1c5e20804dc97a1124d43a5e

Meeting number: 2632 565 1607 and Mtg. Password: MwNuM33GRs3