| Robustness of group divisible treatment designs¡]GDTD¡^against missing data is investigated under two pattterns of missing observations. One is when observations of the control treatment pertaining to one block are missing, and the other is when all the observations in one block are missing. Two criteria of robustness are considered, the criterion of connectedness, and the criterion of A-efficiency. Robustness of general GDTD's are investigated, and the exact A-efficiencies of GDTD's when = 0 and = 0 are derived, respectively. |