The actual yields from multisite variety trials are usually analysed a
s a two-way table (rows = varieties, columns = sites) of mean yields.
Recently, Yen and Hamblin (1994) proposed the variance of relative yie
ld across sites as a useful stability parameter. In this paper, it is
shown that this stability parameter measures the variability of the de
viations from a reference line with the slope equal to variety mean/to
tal mean.