We develop two local quasi-likelihood imputation estimators for mean in a generalized varying-ciefticient model when response variables are missing at random.