Abstract: Robustness of incomplete block designs against missing data is investigated. Necessary and sufficient conditions for robustness of an arbitrary incomplete block design are derived under two patterns of missing observations. Simple sufficient conditions are also provided. Some classes of designs that are robust are identified. The efficiency of the residual design is evaluated for certain group-divisible designs when all observations in a block are lost. Finally, a lower bound to the efficiency of the residual design is obtained when a single observation is missing in an arbitrary incomplete block design.
Key words and phrases: Criteria for robustness, conditions for robustness, robust designs, efficiency.