Monday, April 18, 2011 14h-16h
All models are wrong, some of them are useful (George Box)
Ok, but which ones? (the crowd)
Model selection is an integral part of statistical inference, as stated in an article by S. Buckland in 1997. Inference techniques everyone is using are conditional on the specification of a statistical model. More often than not, the model is not postulated from prior knowledge or belief, but instead selected from a set of candidate models based on its fit to the data. In such cases, using the selected model for inference just as if it were the truth is misleading.
A set of techniques have been developed in the past decade to include the so-called model-selection uncertainty into statistical inference. They involve weighting models with an appropriate criterion (e.g. AIC) and then using all candidate models, instead of just one, for inference (model-averaging, or multimodel inference, techniques).
This workshop will aim at applying multimodel inference in R, with the help of package glmulti (get glmulti from the CRAN). The focus will be on practicing and observing the benefits one expect from multimodel inference, rather than providing a full introduction to glmulti. More on the package can be found in this article. The key theoretical concepts underlying the methods will also be introduced along the way.
Some familiarity with linear models and maximum likelihood methods will be helpful. That being said, as with all of the workshops in this series, we strive to foster a collaborative learning environment. We expect there to be something for participants of all levels of prior knowledge, and you might be surprised by how much you can learn in a short time from your peers!
NOTE: If you are using your own computer, in order to be able to load the rJava library required by glmulti, you need to first install Java JDK. You can do so by going here.
Select the Java Platform JDK (the one on the left) and follow the instructions. Once you have done this, everything should work fine.