I was perusing Bobby Gramacy's introductory slides for his Bayesian Inference course here at UCSB. On slide 11, he mentions the saying "All models are wrong, but some are useful." Googling for this phrase, I found a post by Andrew Gelman about this phrase where he reiterates the point that all models are wrong and the goal of posterior model checking is "to understand what aspects of the data are captured by the model and what aspects are not." My addition to this conversation focuses on the second half of the saying. The fact that all models are wrong should not discourage anybody from trying to model in the first place. Instead, modelers need to understand what scientific question is being asked and build a model to answer that question. A model will be useful if it can answer the question of interest.

blog comments powered by Disqus


30 September 2009