M. Uzik et A. Zofajova, REDUCTION OF EXPERIMENTAL ERROR BY THE ME THOD OF THE NEAREST NEIGHBORS AT RED-CLOVER (TRIFOLIUM-PRATENSE L), Rostlinna vyroba, 39(10), 1993, pp. 865-876
Results of eight trials conducted with red clover to compare the effic
iency of the classical method (AV) and the method of nearest neighbors
(NNA). Each cut was considered as the basic trial (ZP), so the total
number of basic trials was 45 (Tab. I). Derived trials (OP) were compo
sed from the basic trials by the combination of the number of replicat
ions from two to four. The total number of basic + derived trials was
196 (Tab. II). Trials were evaluated by the two-way analysis of varian
ce (AV) and by the method of nearest neighbors (NNA) Anoft (Schwarzbac
h, 1984). Trials were characterized by the ratio of variance of the tr
end between and inside the blocks to MSe from two-way AV (Tab. Il). To
evaluate the efficiency of the method Anoft (Tab. III) the minimum di
fference md in % (md AV : md Anoft) was compared, the probability of t
he selection of four the most yielding varieties (Tab. IV) from basic
trials and derived trials according to arithmetic mean (Px) and accord
ing to adjusted mean (Pa). The probability in the selection from the s
ame trial (Pm) by the mentioned methods (Tab. V) and selective differe
nces of four varieties at the selection according to (x) over bar and
according to adjusted sequence, but by nonadjusted value (x) over bar(
A) (Tab. VII). Neither 10 % of trials had variances of the trend betwe
en and inside the blocks higher than the double of MSe (Tab II). The N
NA method, due to the reduction of error was reducing md value more wi
th lower number of replications. With six replications it was by 10 %,
with four replications by 15 %, with three and two replications by 30
to 52 % (Tab. III). The probability of the selection of common most y
ielding varieties from independent trials at two replications was low
for both the methods (0.293 and 0.331), this was higher in the depende
nt trials according to the number of common replications and was of ri
sing trend, though without significant differences between the methods
(Tab. IV). The difference between average values of four most yieldin
g varieties with the selection according to (x) over bar and (x) over
bar(A) (of adjusted sequence) was higher than 5 % solely in the trials
with two replications and their percentage from the total number of t
rials was 10 to 30 % (Tab. VII). The NNA method reduced the minimum di
fference compared with the classical AV method progressively with redu
ction of the number of replications, though despite it, the estimation
of the yielding capacity of genotypes solely on the basis of two repl
ications can lead to the selection of less yielding genotypes, as the
probability of selection of common genotypes from independent trials w
ith two to four replications was identically low for both the methods
of evaluation.