or left, but a "3" environment changes little if the same
treatments are used. Only if the usual environmental
index range is much higher or lower, or the range of the
index very narrow so that extrapolation is extreme, should
there be concern with the use of data from only one year.

Distribution of confidence intervals

 It appears from the modified stability analysis that
two recommendation domains exist relative to the use of
maize cultivars in the farmer-managed trial in Malawi. A
method of further examining the data to aid in making
specific recommendations is to use a distribution of
confidence intervals for the technologies and for each
partitioned recommendation domain. In this case there are
nine farms in the "poorer" environment (e < 2) and five in
the "better" environment (e > 2).
 The confidence interval is calculated for any level
of confidence, a, from

 S tgS / (n1/2)

where

 Y = the mean treatment yield,
 a = the level of confidence,
 S = [Zx2 / (n-1)]1/2, and
 ta = value from a "t" table

 In the example, for e > 2 and for the unfertilized
local variety, the yields are 2.2, 2.2, 1.9, 1.2, 1.8. As
always

 Zx2 = ZX2 (EX)2 / n
 = 17.97 17.30
 = 0.67

 S = (0.67 / 4)1/2
 = 0.409

So the confidence interval for any level a is