Agriculture Reference
In-Depth Information
Miles stated that attempts should be made to eliminate real variation (systematic error) among labora-
tories. He identiied the causes of signiicant differences between or among germination tests as:
1. Chance alone - due to random sampling variation (this type of variation cannot be eliminated).
2. Poor equipment, including variation in environment within a germinator.
3. Poor method.
4. Poor technique.
5. Errors or inconsistency in distinguishing between normal and abnormal seedlings.
6. Fungi or bacteria.
7. Chemicals on the seed.
8. Inaccurate counting or recording.
9. Nonrandom selection of seeds to test.
10. Actual change in the percent germination between the tests.
Miles presented a series of nine tables showing one-sided and two-sided tolerances at different levels
of signiicance for use in comparing subsequent tests against the labeled germination. He showed different
tolerances for different sizes of samples ranging from two to twelve 100-seed replicates (200-1200 seeds).
Finally, he showed conidence limits (tolerances) for use by laboratories showing the limits by which future
tests on the same sample could be expected to vary.
AoSA and ISTA Germination Tolerances
Table 13J, AOSA rules shows current AOSA germination tolerances for comparing two 400-seed germina-
tion tests of the same or different submitted samples tested in the same or different laboratories (one-way
test at P=51). These are very similar to one-sided test comparisons of two 400 seed tests in Table G3,
Column A, B, and C presented by Miles (1963), a one-way test at the 5% probability level. Thus, they main-
tain experimental error found by Miles in the 1950s, including random sampling variation. These tolerances
employ a one-sided test for comparing results of a second germination test to a labeled rate, or irst test.
The use of these tolerances can be expected to lead to a Type-I incorrect decision (rejection of a lot that is
actually correctly labeled) 5% of the time.
Germination tolerances are applied by averaging the irst original and second germination test results
to give both an equal chance of being correct. The tolerance value applied is the number in the table oppo-
site to the average values of the two tests. If the difference between the irst and second test exceeds this
tabulated value, the lot is considered to be mislabeled, or out of tolerance.
For example, if a lot labeled as 96% germination was retested and found to germinate 88%, the tol-
erances applied would be 6 based on the average of the two tests of 92 (96% + 88%/2 = 92%). Since the
difference (96 - 88 = 8%) is greater than 6, the lot is considered mislabeled and the two test results are not
within tolerance.
Table 13.I AOSA (2010) (Table 13.5 in this topic) shows tolerances for determining excessive varia-
tion in germination among different replicates within a 400 or 200-seed test. The tolerances shown for
the average percent germination indicate how much difference is allowed between the highest and lowest
replicates. Unlike those for tests conducted in different laboratories, these tolerances recognize random
sampling variation only. They are similar (ISTA has only 4 replications but not 2) to the ISTA tolerances
found in Table 5.1 which were extracted from Column D of Table G1 of Miles. However, the use of these
tolerances denotes a willingness to make an incorrect decision (Type-I error) only 2.5% of the time where
a replicate is rejected that should actually be considered good.
Three tolerance tables for comparing germination tests are used by ISTA. The irst table gives maxi-
mum tolerated ranges between any of the four 100-seed replicates of a germination test (two-way test at
2.5% signiicance level). This is similar to AOSA tolerances shown in Table 13.5 for determining the maxi-
mum variation among different 100-seed replicates.
