3.4. A test-hold program must be in place where harvest cannot occur until results are
confirmed as negative, along with a corresponding procedure to prevent harvest and
deal for positive results. The coordination of sampling and harvesting must be carefully
planned and choreographed. If a test is performed, it CANNOT be released until it has
cleared.
3.5. The sampling program must also take into account and minimize normally occurring
events where sampling cannot be performed:
3.5.1. Inclement weather (heavy rain, lightning, extreme heat)
3.5.2. Field activity (irrigation, fertilizing, pesticide spraying)
3.5.3. Employee limitations (more samples needed than can be collected in one day,
sick or injured employees, employees on vacation or at training, vehicle
maintenance or other equipment issues). Employee limitations need to be reviewed
thoughtfully: for instance, are there enough samplers to get samples completed
within the timeframe? Too many fields and not enough samplers mean samples
are not completed in time to match planned harvest dates.
4. Considerations in developing a sampling plan (Russ Flowers, Dan Morse)
4.1. Importance of a threshold for acceptance
There are 4 quantities that are needed for the calculation of sample size:
•
The contamination rate (P) that is considered acceptable. The level at which one
would not want to take action. This is called the Acceptable Quality Level (AQL)
•
The error rate for acting when the lot is at the AQL. This is the Type 1 error or
Producer’s risk. A typical Type 1 error is 5%.
•
The contamination rate that is considered unacceptable. The level at which one
would want to take action. This is called the Rejectable Quality Level (RQL)
•
The error rate for not acting when the lot is at the RQL. This is the Type 2 error or
Consumer’s risk. A typical Type 2 error is 10%.
With a very low contamination rate, it may not be possible to provide a sample size that is
both of reasonable size and able to distinguish between the current contamination rate and
something that is a couple orders of magnitude higher.
4.2. Statistical confidence based on sample numbers
Using Type 1 error rate of 5% and a Type 2 error rate of 10%, with 60 samples, the “AQL” is
0.09% and the “RQL” is 3.8%. AQL and RQL are in quotes because these are probably not
at the levels wanted, but are the performance characteristics of a sampling plan that accepts
with no positives and rejects with 1 positive. So here, if 0.09% of the individual samples
were contaminated, 5% of the time the lot would be rejected. If 3.8% of the individual
samples were contaminated, 10% of the time the lot would be accepted.
If 0.2% of the time a group of 60 is positive, and this is an indication of the mean level of
contamination, the estimate is that 0.0033% of individuals are contaminated. If this was
called the AQL, the sampling plan generated would have this fail 5% of the time instead of
