Spatial variability and efficiency of treatment mean comparisons in an experiment with fodder pea using modern statistical methods.

Janusz Gołaszewski

Abstract

It is typical of breeding experimentation to conduct experiments on large breeding material tested on small plots with a limited number of replications. Under such conditions, observations made on adjacent plots are biased by the effect of autocorrelation and fertility trends. The actual treatment effects can be masked and the capability of the breeder to detect true treatment differences is impaired. This paper deals with the problem of the estimation of effects of spatial variability and their impact on the efficiency of treatment comparisons. The considerations are based on the results from a breeding experiment with 25 treatments of fodder pea arranged according to the partially balanced incomplete block design (IBD) with 4 replications. Plant height and seed yield were analysed with the conventional statistical method ANOVA, the nearest neighbour analysis (NNA) and kriging. Eventually, the efficiency of the neoclassical methods relative to the completely randomised design (CRD) and randomised block design (RBD) was calculated. The estimation of the treatment effect on plant height was accomplished most efficiently with the NNA, whereas the efficiency of the alternative methods in the estimation of seed yield was comparable to the efficiency of the RBD.