Events login

Main Events Calendar

Warning Icon This event is in the past.
April 12, 2019 | 2:30 p.m. - 3:30 p.m.
Category: Seminar
Location: Old Main #0121
Cost: Free
Audience: Academic Staff, Alumni, Community, Current Graduate Students, Current Undergraduate Students, Faculty, Parents, Prospective Students, Staff

The Earth and Environmental Science Seminar Series presents application of quantitative methods for analyzing complex environmental and ecological data: Model Selection – Information Criteria or LASSO? with Dr. Song Qian, PhD, Department of Environmental Sciences | University of Toledo.

Friday April 12th at 2:30pm

Old Main Room 0121 (in tbe basement near the mineral museum)

Reception and refreshments to precede 40 – 50 min seminar followed by time for questions

Abstract: Model selection has been a difficult and confusing topic in environmental and ecological data analysis. In 2014 the leading Ecological Society of America journal Ecology published a special forum on the topic. At issue is the question of whether to use the null hypothesis testing (p-value of a slope) or an information criterion (particularly AIC) as the main tool for selecting important predictors when using regression analysis. A selected (or statistically significant) variable is often seen as ecologically "important", which often implies a causal link between the variable and the response. Just as the attraction of the frequentist concept of null hypothesis testing (NHT) is fatal to the scientific community, the lure of using AIC as a universal variable selection tool is irresistible to many ecologists. However, the use of NHT and AIC is mostly unhelpful to scientists. In this presentation, I will discuss why NHT is not suitable for scientific research and should be replaced by a suitable shrinkage estimator. I present LASSO as an estimation-oriented alternative (also a shrinkage estimator) to AIC and discuss its use in exploratory data analysis. Bayesian implementation of LASSO is also discussed. I will present two examples: exploring shoreline characteristics linked to nearshore fish community in Lake Erie and (2) relevant stream flow metrics for assessing stream ecosystem in the Southeastern US.

Biography: Dr. Song Qian is an Associate Professor in the Department of Environmental Sciences at the University of Toledo. He has a Ph.D. in Environmental Science and an MS in Bayesian Statistics, both from Duke University. Through his Ph.D. research (modeling phosphorous retention in the Everglades wetlands), he developed several non-parametric Bayesian regression methods. His Ph.D. dissertation work was the first to introduce the Markov chain Monte Carlo to environmental scientists. At University of Toledo, he continues his work in the applications of Bayesian statistics in environmental and ecological studies, including the evaluation of the effectiveness of agricultural conservation practices in reducing nutrient runoff, and risk assessments of (1) invasive fish establishing in Lake Erie and (2) cyanobacteria toxin contamination of drinking water. His textbook Environmental and Ecological Statistics with R (currently in its second edition) was translated into Japanese and Chinese. He, with two of his former Ph.D. students, is currently writing a Bayesian textbook to be published in late 2019/early 2020.


For more information about this event, please contact Shirley Papuga at 313-577-2506.