Statistical tools for setting in-process acceptance criteria.

Establishing meaningful and reasonable acceptance criteria for process validation or continual monitoring is crucial to approval and successful manufacturing. The limits should be based on statistical analysis of historical data when possible. The control limits of "mean +/- 3 standard deviations" is one industry standard. However, the limits may be artificially constraining if the standard deviation does not reflect the true variance of the process. Under-estimation of process variance is common with small data sets. This paper presents three methods for correcting underestimated variance, allowing the setting of acceptance criteria that are slightly larger than +/- 3 standard deviations. These limits are more meaningful in that they account for true process variability and will signal process deviations due only to a specific cause.