27 Day 27 (April 29)

27.1 Announcements

  • Peer review portion of the class project has been canceled.

    • If you have completed this or want to complete this, lets talk.

  • Send me () and Aidan () an email to request a 30 min time slot between May 1 and May 9 to give your final presentation. When you email us, please give 3 dates/time that work for you.

  • Teaching evaluations

    • There will be several questions
    • Should take about 20 min
    • The information you provide is really helpful

  • Questions from journals

    • “These lectures make me realize that I am learning a lot of the basics, things that maybe I should know but I don’t, it makes me wonder if I got as much as I could from past classes…”
    • Not agreeing with my opinion related to model selection/averaging…(Paper by Jay Ver Hoef and presentation)
    • “These classes make me realize that I am learning a lot of the basics, things that maybe I should know but I don’t, it makes me wonder if I got as much as I could from past classes…”
    • “I am still trying to understand a few things: How does regularization affect bias and variance, and what happens if the regularization penalty is too high or too low?”
    • “The idea of averaging across models, rather than selecting just one based on criteria like AIC, feels more robust, especially when competing models yield different practical recommendations.”

27.2 Model selection/comparison

  • What is covered today is selected material from Chs. 13 - 15 of BBM2L.

  • If you have more than one model for a given dataset/problem how do you determine which one(s) to use for prediction and inference?

    • Good resources (here and here)
    • Diversity of approaches from Hooten and Hobbs (2015)

  • Predictive performance metrics

    • Information criteria vs. scoring functions
    • Important characteristics of a predictive distribution (example using Day 14 notes; maximize the sharpness of the predictive distributions, subject to calibration)
    • Good resource (here)

  • Live example using DIC R code

  • Live example using regularization R code

  • Live example using Bayesian model averaging R code

  • Cool new stuff!

27.3 Spatio-temporal models

  • What is covered today is from Ch. 28 (pgs. 501-515) BBM2L - Slides