Conditional AIC Under Covariate Shift with Application to Small-Area Prediction

• Activity Number: 444
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Yuki Kawakubo* and Shonosuke Sugasawa and Tatsuya Kubokawa
• Companies: and The University of Tokyo and The University of Tokyo
• Keywords: Akaike information criterion ; conditional AIC ; covariate shift ; linear mixed model ; small area estimation ; variable selection

Abstract

We consider the problem of selecting explanatory variables of fixed effects in linear mixed models under covariate shift, which is the situation that the values of covariates in the predictive model are different from those in the observed model. We construct a variable selection criterion based on the conditional Akaike information introduced by Vaida and Blanchard (2005) and the proposed criterion is generalization of the conditional AIC in terms of covariate shift. We especially focus on covariate shift in small area prediction and show usefulness of the proposed criterion through empirical studies. This research is based on Kawakubo et al. (2015).

[References]

Kawakubo, Y., Sugasawa, S. and Kubokawa, T. (2015). Conditional AIC under covariate shift with application to small area prediction. arXiv preprint arXiv:1501.03889.

Vaida, F. and Blanchard, S. (2005). Conditional Akaike information for mixed-effects models. Biometrika 92, 351-370.