K-fold cross-validation based frequentist model averaging for linear models with nonignorable missing responses

A frequentist model averaging method based on K-fold cross-validation is proposed for linear models with nonignorable missing responses. When all the candidate models are misspecified, the asymptotic optimality of the proposed method is established in the sense that it can asymptotically achieve the minimum squared loss. If the set of the candidate models contains the correct ones, the proposed method is also shown to be selection consistent in the sense that the sum of the model averaging weights assigned to the correct candidate models asymptotically converges to one. Numerical simulations were conducted to demonstrate the performance of the proposed method, and as an illustration the proposed method was applied to analyze a diabetes dataset.