Agriculture Reference
In-Depth Information
Sources of Variation
Test results may vary because of two basic types of variability. The irst is
random sampling error
which
is unavoidable and is due to chance alone. Examples of this type of variation are: the random distribution
of contaminants in a sample such as weed seeds or other crops in a purity sample; and the distribution of
non-viable seeds in a viability test such as germination or TZ. This kind of error is predictable and follows
well-established patterns, and therefore can be easily estimated by statistical analysis. The second type of
variation is caused by
systematic
error and can usually be traced to procedures, materials and/or equip-
ment failures. Both systematic and nonsystematic errors are collectively known in seed testing literature as
experimental error
. Systematic errors are less predictable than that caused by random sampling error (also
called random sampling variability) and may result from inaccuracies in the testing process such as those
caused by poorly-calibrated instruments, counting errors, impure reagents, and differences in analyst expe-
rience and qualiication as well as by the conditions under which the tests are conducted.
If human errors such as misreading instruments, recording wrong values, incorrect experimental setup
or calculation mistakes are detected, the test or experiment should be repeated and the test data should not
be analyzed. Such errors should not be considered as experimental errors, and erroneous measurements
should not be included in the data analysis. In other words, detected human errors should not be a part of the
experimental error. The mean number of replications (such as the mean of four replications in a germination
test) is the best estimate of a variable (in this case, germination). The standard deviation of the mean [i.e.,
the dispersion of values (e.g., replications) from their mean] indicates the accuracy of the estimate.
Results of purity tests and noxious weed seed examinations may vary because of differences in analyst
ability and experience in inding and recognizing common and noxious weed seeds or in classifying inert
matter. Germination test results may vary due to differences in the temperature under which the tests are
conducted as well as from the inluence of different media used. Though seed testing results will always
relect a level of variability consistent with the seed testing at any given time, such variation can and should
be minimized by careful calibration of equipment and proper training of analysts.
Experimental error can cause seed testing results to vary beyond the limits due to chance (random
sampling variation) alone. Most, but not all, ISTA and AOSA tolerances recognize the existence of such
variation as part of seed testing and account for it through wider tolerance limits than would be necessary
for random sampling variation alone.
Precision measures how closely two or more measurements agree with each other, while accuracy
relects how “correct” a measured value is, or how close it is to the true value. The precision of measure-
ments is subject to random sampling error or variation and can be improved by increasing the sample size
and/or replications. Random sampling error affects the
precision
of measurements and to a lesser extent
their
accuracy
. Systematic errors have consistent effects on measurements and result in either a systematic
increase or decrease in the value of measurements. Systematic errors affect the accuracy of measurements
but not their precision, and cannot easily be analyzed by statistical analysis. Although generally dificult
to detect, once identiied they can be eliminated or reduced by reining the measurement techniques. Miles
(1963) reported that the increased precision reduces the tolerances for the same probability. Although
increasing sample size or number of replications will result in increasing precision, the gain in precision
and the decrease on the tolerance is less per extra unit of work.
Seed Lot Heterogeneity
The discussion of variability presumes that samples to be tested are obtained from seed lots that are homo-
geneous for all quality factors for which tests will be performed. Homogeneous seed lots are those in
which quality aspects such as inert matter, weed seeds and other incidental contamination are uniformly
distributed throughout the lot. If a lot is completely homogeneous, all samples, regardless of size, should
result in similar estimates of quality within the limits of random sampling error, assuming other sources
