WHEN ZERO IS NOT ZERO
by Jim Dodd
The zero tolerance requirements for the Starlink testing brings attention to the statistical problems related to all testing of seed involving sampling. Of course if every seed within a seedlot could be tested, results could be absolute. Because that would use all seed in the lot, sampling needs to be done to best represent the seedlot. As carefully as a company tries to accurately reflect the seed within a seedlot, the sample size still limits the probability that test results of zero presence of a GMO or outcross, for that matter, is actually present in the seedlot.
Using the statistical model presented by the USDA (http://www.usda.gov/gipsa) one can enter the sample size and minimum standard to determine the probability that the GMO is at or above the minimum standard. The table below illustrates the results as applied to different sample sizes.
Probability of Actual Percent OUTCROSS or GMO
Than Standard if Test Result Is Zero*
*assumes that test method accurately detects the outcross or GMO
*assumes that sampling method was valid
Even using a 2400 kernel sample size there is some chance (.08%) that there could be at least 0.2% GMO present in the seedlot. This explains why a seed company should be reluctant to declare 100% assurance that a seedlot is non GMO even if the results are negative.
This table also relates to growout sample sizes. PSR uses a 400 kernel sample size for growouts because of this statistical advantage. A 100 kernel electrophoresis or field growout, even if accurately tested, may show zero outcrosses, but there remains a 36.6% probability that there is at least 1% outcrosses in the seedlot. A zero outcross from a 400 kernel sample has only a 1.8% probability of having 1% outcrosses in the seedlot.