Day 28 (May 1)
Announcements
- Teaching evaluations
- There will be several questions
- Should take about 20 min
- The information you provide is really helpful
- Today is our last lecture
- Most everything project related is Due May 11.
- Thank you for the fun semester!
- I am looking forward to the projects/presntations
- A comment about the bumpy semester
- Questions from journals
- Statistical Analysis and the Illusion of Objectivity
- “I think it would be helpful to go more into depth about the application of decision theory.” (see this paper)
- “One concept I’m still trying to fully understand is when it makes sense to use Bayesian model averaging instead of simply selecting a single best model.However, it’s unclear to me what threshold or objective criterion—if any—should be used to decide whether to stick with just one model or to combine multiple models in an ensemble.”
- “On Bayesian model averaging, I am curious on the difference between this method and R2” (see paper here)
- “One thing I am wondering about has to do with the Bayesian model averaging. Why couldn’t we simply use a binomial distribution where we select each of our models with equal probability when generating our posterior for cases where we assume equal weights?” (see model based model selection in this mongraph)
- “I think one of the biggest things I’ve learned today and throughout this class is that you can make tons of decisions in model building, and the ‘best’ model is what fits your specific needs. You can decide whether bias in a model is worth the decrease in variance or not. Looking at the deer example, the statistically ‘best’ model only involved Sex, Age, and Age2. But since the reserve was specifically asking about the effect of Crop land, using that second ‘worse’ model is actually better for the situation.”
Spatio-temporal models
- What is covered today is from Ch. 28 (pgs. 501-515) BBM2L - Slides