Coming up in September, the Statistics Department will be hosting two lectures by van Eeden invited lecturer Professor Roger Cooke (Chauncey Starr Chair for Risk Analysis, Resources for the Future, USA; and Department of Mathematics, Technische Universiteit Delft, Netherlands).
The lectures will be on Tuesday, September 22nd and on Thursday, September 24th. There will be small receptions before each lecture in the lobby of the building where the lecture is being held. Lecture times, locations, and abstracts are below.
Date: Tuesday, September 22
Time: 11am - 12pm
Location: Lecture Theatre 102, Michael Smith Labs (2185 East Mall)
---------------
Title: Vine Regression
Abstract: Vine regression is a parametric non-linear method that does not rely on the Gaussian error assumption. Regression is considered as conditional expectation after fitting a multivariate distribution.
Regular Vine theory provides a rich class of N-dimensional densities with arbitrary one-dimensional margins and arbitrary N-choose-2 (conditional) bivariate copulas. For ordinal data, "Gaussian smoothing" uses regular vines to build a tractable density that emulates the empirical rank correlation structure. Conditional expectations and variances can be computed in the emulating density. The effect of a given regressor on a variable of interest is computed as the expected difference of two regression functions, where (a) the regression is conditional on all regressors, and (b) the regression is conditional on all regressors with the given regressor augmented by one unit. In one stroke this eliminates issues like confounders, interactions, collinearity, transformations, heteroscedasticity, marginality etc.
Vine regression is illustrated by computing the effect of duration of breast feeding on IQ using the National Longitudinal Study of Youth. The talk will give a brief introduction to copulas and vines.
***************************************************************************************
Date: Thursday, September 24, 2015
Time: 4pm - 5pm
Location: Room 4192, Earth Sciences Building (2207 Main Mall)
---------------
Title: Structured Expert Judgment and Invasive Species in the Great Lakes
Abstract: Structured Expert Judgment (SEJ) treats experts as statistical hypotheses. Experts quantify their uncertainty for variables from the field for which true values are known post hoc, in addition to the variables of interest. Statistical accuracy and informativeness scores are used to construct "performance based combinations", which in turn can be evaluated and compared with other combinations schemes. Many examples of SEJ are in the TU Delft expert judgment data base.
SEJ has recently been employed to assess the impact of an Asian Carp invasion into Lake Erie. A general introduction to SEJ, including in- and out-of-sample validation will be given, followed by an in depth discussion of the Erie application.
A reference for expert judgment is the following book. Cooke R. (1991), Experts in Uncertainty; Opinion and Subjective Probability in Science, Oxford University Press; ISBN 0-19-506465-8.