Title: Frequentist model averaging in regression models

Speaker: Dr. Shaobo  Jin(金少博),   Uppsala University

Time:  3:00pm, 2018-11-21

Location:   The 4^th Floor, Sir RRShaw Building, Yuquan Campus, Zhejiang University

Abstract: In many applications of regression models, randomness due to model selection is commonly ignored in post-model selection inference, which leads to too optimistic confidence intervals. When the main focus is prediction, model selection uncertainty is also often overlooked. In order to account for the model selection uncertainty, least-squares or likelihood-based frequentist model averaging has been recently proposed. Instead of choosing the optimal candidate model, a weighted average of candidate models is preferred. In this talk, frequentist model averaging in regression models will be briefly reviewed. In the linear regression context, we can show that the confidence interval from model averaging is asymptotically equivalent to the confidence interval from the full model. The asymptotic-based finite-sample confidence intervals are equivalent to that from the full model if the parameter of interest is a linear function of the regression coefficients. We can also show that this equivalence holds for prediction intervals. We can also apply the principle of frequentist model averaging to covariance analysis-based latent regression models. Some current work will be briefly discussed.

All are welcome!

Contact Person: Zhonggen Su, suzhonggen@zju.edu.cn