Agriculture Reference
In-Depth Information
Tolerances may represent either one- or two-sided tests. A two-sided test hypothesizes that a test value
is compared to two critical values. If it exceeds the larger or falls below the smaller, the lot is considered
inconsistent with the labeled value and should be rejected. A Type-I error is committed if the lot meets the
labeled criteria, but the test would lead to its rejection. The tolerance represents the range of possible label
values the test result does not contradict, the range between the upper and lower tolerance. As long as the
test result falls between the tolerances, the label (lot) can not be rejected. The Type-I error rate of the signii -
cance test translates into the probability with which the range between upper and lower tolerance does not
include the target value. For 5% two-sided tolerances, one should expect, in independent repetitions of the
test from the same lot that in 2.5% of the cases, the true value is below the tolerance limits, while in 2.5%
it is above the tolerance limits (Fig. 13.1).
Seed testing tolerances can also be derived from one-sided signii cance tests which specify the upper or
lower extent of label conformity (Fig. 13.1). One-sided limits are appropriate for “at-most” characteristics
where the lot should be rejected if it exceeds a label value (noxious weeds) or for “at-least” characteristics
such as germination percentage, where testing should ensure that the germination percentage of the seed lot
is equal to at least the labeled amount. It is not considered incorrect labeling if the germination percentage
is higher than that labeled or the occurrence of noxious weed seeds is less than labeled, although this may
entail i nancial loss for the seed producer. For precision planting, a higher than labeled germination rate
may be of concern as well. For 5% one-sided tolerances, one should expect, in independent repeated tests
from the same lot that in 5% of the repetitions, the true value to be below (lower side) or above (higher side)
the tolerance limits.
Most seed testing tolerances have been calculated to account for both systematic and random sampling
variation. Most of the present ISTA and AOSA tolerances are based on work done in the mid-1950s and
early 1960s by Miles and presented in the handbook of tolerances and measures of precision for seed test-
ing (Miles, 1963). This was the i rst attempt to make a comprehensive compilation of statistical procedures
using voluminous seed testing referee data and to develop tolerances for a wide range of testing procedures
Figure 13.1. The values for which we can reject the null hypothesis (Ho) are located in two-tailed (upper)
or in one-tailed (lower) tests of the probability distribution curve. In tolerance tables, the two-tailed (or
two-way) test is used to decide if one (t) test is poorer or better than another test; whereas the one-tailed (or
one-way) test is used to decide if a second test is poorer than a label or a i rst test.
