neither the size of the sample nor the method of sample selection are in
accordance with the rules for statistical sampling techniques. As a result
the extrapolation of planned sampling has no validity and the ODR may not be
compared with the MADR, the MADR being defined as the defect rate in the
population above which the contractor's quality control is considered
Changes in the DR from month to month, calculated on the basis of the total
population, gives an indication of whether the contractor's quality control
is getting better or worse, providing the same sampling procedures are
carried out each month.
4-340 Random Sampling Random sampling is a surveillance method based on
probability and statistical theory. Key elements of random sampling are as
(a) Assesment of the contractor's performance for the entire
population of services is based on the results observed from a study of only
a portion or sample of the total population. Each occurrence in the sample
is classified as either conforming to a performance requirement or as
nonconforming. The percentage of nonconforming occurrences or defects in
the sample is an estimate of the percentage of defects in the total
population. This process is termed extrapolation.
(b) The size of the sample is specified for a given total population to
achieve a predetermined level of statistical accuracy in accordance with
standardized, population - sample size, tables (See Appendix F, Tables 1
(c) The population - sample size tables are included as an attachment
to the contract documents, providing the contractor with an opportunity to
assess the methodology involved. This approach also prevents later disputes
concerning the correctness of the tables, the contractor having assentedto
the sample sizes by the terms of the contract.
(d) The sample of work occurrences selected for evaluation must be
selected by a random process in which each occurrence has equal dance for
(e) Once an evaluation schedule is established, it must be followed
through the surveillance period (i.e., monthly schedules).
(f) When the random sampling is being used to extrapolate defects to
the entire population, surveillance data gathered by other methods (i.e.,
validated customer complaints and incidental inspections) cannot be combined
with data gathered by random sampling.
(g) The defect rate observed in the sample is adjusted by an
"Adjustment Factor" to ensure the statistical validity of the calculations.
A table of adjustment factors is shown in Appendix F, Table 3.