A0827
Title: The adequate bootstrap: A new method for measuring model uncertainty
Authors: Toby Kenney - Dalhousie University (Canada) [presenting]
Hong Gu - Dalhousie University (Canada)
Abstract: In model adequacy testing, there is a fundamental disconnect between what is tested and what we would like to test. The usual approach is to test the null hypothesis ``Model M is the true model''. However, Model M is never the true model. This means that if we collect enough data, we will certainly reject the model adequacy test. The model might still be useful if it is close to the truth. We propose a new method for assessing this. The idea is to determine the sample size at which the model adequacy test would not be rejected, then perform inference based on this sample size. If the model is good, we will perform inference based on a large sample size, while if the model is not a good fit, we will perform inference based on a small sample size to incorporate the model uncertainty. The intuition is that if we had only taken a smaller sample, then we would not have rejected the model, so the inference we make under the model should be valid.
