ABSTRACT
To study stress tolerance of a set of breeding lines, it is necessary to evaluate the lines under stressed
conditions and control conditions without stress. Thus, the evaluation involves two factors, i.e. stress treatment
(with levels ‘control’ and ‘stressed’) and genotypes. There are many valid experimental designs for factorial experiments,
which involve randomization of stress treatment-by-genotype combinations. Conducting trials laid out according to such
standard designs may be difficult, however, due to potential neighbour effects between plots receiving different stress
treatments. This frequently leads plant breeders to assess stress treatments in completely separated trials. For example,
one trial may be conducted for the control treatment and one trial for the stress treatment. This approach limits the type
of inferences that are available. In this paper it will be shown that inferences based on between-trial information are not
possible. Inference based on within-trial information is feasible, however, as will be illustrated using an example.
The practical implication is that a genotype-specific stress tolerance index cannot be estimated with this approach,
but that a relative comparison of the genotypes’ stress tolerance is possible.
