Statistical Tolerance Intervals per ISO 16269-6

🏭

Industry Use

Tolerance intervals in pharmaceutical validation

A statistical tolerance interval is a range [L, U] computed from a sample such that, with confidence γ, at least the proportion p of the sampled population lies inside it. It is distinct from a confidence interval (which brackets the mean) and a prediction interval (which brackets the next observation). Tolerance intervals answer the validation question: "Will at least 99% of future batches fall within specification?"

FDA's Process Validation guidance (2011) expects Stage 2 to provide statistical evidence that the process will consistently meet specifications; a tolerance interval sitting entirely inside the specification limits is direct evidence of exactly that. Typical uses: specification verification (tablet weight, assay), cleaning validation (surface results vs limits), potency and yield studies, and dissolution performance.

ISO 16269-6:2014 FDA Process Validation (2011), Stage 2 USP <1010>

∑

Calculation Method

Exact k-factors per ISO 16269-6:2014

Two-sided tolerance interval

[ x̄ − k·s , x̄ + k·s ]

k is determined so that the interval contains at least the proportion p of a normal population with confidence γ. ISO 16269-6 defines k by a coverage integral (Annex F; tabulated in Annex D) — there is no closed-form expression, and this calculator evaluates the integral exactly rather than using an approximation.

One-sided tolerance bound (upper shown; lower is symmetric)

( −∞ , x̄ + k·s ] with k = t′₁₋α(√n·u_p; n−1) / √n

The one-sided factor is an exact noncentral-t quantile (ISO 16269-6 clause A.5, eq. A.13; tabulated in Annex C). One-sided factors are smaller than two-sided ones — use them when only one limit matters (impurity: upper; yield or potency floor: lower).

Selected exact k-factors (γ = 95%), as computed by this calculator; values agree with the ISO 16269-6:2014 Annex C and Annex D tables to their printed precision.

n	Two-sided		One-sided
n	p = 95%	p = 99%	p = 95%	p = 99%
5	5.077	6.598	4.203	5.741
10	3.394	4.437	2.911	3.981
15	2.965	3.885	2.566	3.520
20	2.760	3.621	2.396	3.295
30	2.555	3.355	2.220	3.064

Published sources differ on some two-sided values because older references print approximations (Howe 1969, Wallis 1951) rather than the exact factor — e.g. for n = 20, p = 99%, γ = 95% you may see 3.604 or 3.615 in handbooks; the exact ISO 16269-6 value is 3.6210. This calculator computes the exact factor for any n, p, and γ.

📋

Variable Definitions

All input parameters and their meaning

Symbol	Variable	Units	Description
n	Sample size	count	Number of observations. ISO 16269-6 recommends n ≥ 10; the interval remains sample-sensitive below ~30.
x̄, s	Sample mean and SD	product units	Computed from the data; s uses the n−1 divisor.
p	Coverage proportion	—	Minimum fraction of the population the interval must contain (typically 0.90, 0.95, or 0.99).
γ	Confidence level	—	Confidence that the interval really contains the proportion p (typically 0.95).
k	Tolerance factor	—	Exact ISO 16269-6 factor — function of n, p, γ, and interval type. Two-sided k > one-sided k.
L, U	Tolerance limits	product units	L = x̄ − k·s and/or U = x̄ + k·s, per the interval type.
LSL, USL	Specification limits	product units	Optional; when supplied, the report states whether the tolerance interval sits entirely within them.

📖

Step-by-Step Tutorial

Worked example — API yield, process validation Stage 2

Scenario: An API synthesis process is in Stage 2 process validation. The yield specification is 90–110%. Twenty pivotal batches give x̄ = 97.2% and s = 1.8%. Compute a 99%/95% two-sided tolerance interval.

Tolerance factor. For n = 20, p = 0.99, γ = 0.95, two-sided:

Exact factor (ISO 16269-6, Annex D / Annex F)
k = 3.6210
Lower tolerance limit.

Calculation
L = 97.2 − 3.6210 × 1.8 = 97.2 − 6.518 = 90.68%
Upper tolerance limit.

Calculation
U = 97.2 + 3.6210 × 1.8 = 97.2 + 6.518 = 103.72%
Compare to specification.

Check
[90.68%, 103.72%] vs [90%, 110%]: 90.68 > 90 ✓ · 103.72 < 110 ✓
Conclusion. With 95% confidence, at least 99% of future batches will fall between 90.68% and 103.72% — entirely within specification. The process passes the Stage 2 statistical assessment; continue into Stage 3 monitoring with control charts. Note the margin at the lower limit is small (0.68%): a modest increase in s would break it, which is worth flagging to the review board.

✓ 99%/95% tolerance interval [90.68%, 103.72%] within specification [90%, 110%]

⚠️

Assumptions and Limitations

What the calculation presumes — and what it does not cover

Normality: the factors assume normally distributed data. Check skewness/kurtosis (reported by the calculator) or a normality test; for clearly non-normal data use a distribution-free (order-statistic) tolerance interval instead.
Random, stable sampling: the sample must represent a single stable population — data from a drifting or multi-modal process makes the interval meaningless.
Sample size: ISO 16269-6 recommends n ≥ 10; below that k is very large and the interval is dominated by sampling uncertainty.
Not a capability index: a tolerance interval inside spec supports capability but does not replace Cpk/Ppk metrics or ongoing Stage 3 monitoring.

📚

References

Primary sources for this guide

ISO 16269-6:2014, Statistical interpretation of data — Part 6: Determination of statistical tolerance intervals — clause A.5 and Annex C (one-sided factors); Annex D and Annex F (two-sided factors). The exact methods implemented by this calculator.
FDA, Process Validation: General Principles and Practices (2011) — Stage 2 statistical evidence.
USP General Chapter <1010>, Analytical Data — Interpretation and Treatment.
Hahn & Meeker, Statistical Intervals: A Guide for Practitioners (1991).
NIST/SEMATECH e-Handbook of Statistical Methods, §7.2.6.3 (tolerance intervals).