Empirical likelihood inference for non-randomized pretest-posttest studies with missing data