Robustness of Neighbour Balanced Complete Block Designs for Correlated Observations

Abstract

Neighbour balanced designs, permitting the estimation of direct and neighbor effects of treatments, are used when the treatment applied to the experimental plot may also affect the response on neighbouring plots. The assumption in the classical neighbor balanced block design is that the fertility associated with plots in a block is constant. However, the observations in a block may be mutually dependent on correlated through some systematic pattern of environmental variations. In this article, t he performance of neighbor balanced complete block designs has been investigated when the observations within a block are correlated. The A-efficiency for direct and neighbor effects of the treatments under autoregressive and nearest neighbor correlation structure is computed and tabulated for some cases. It is observed that efficiency is quite high and it is concluded that the designs a e robust under correlated observations with loss of efficiency being small.