HVI Replicate & Outlier Checker

The formula

This is the expression the tool evaluates. Every term is named underneath, with the unit it must be supplied in.

How the result is derived

Step by step, from the values you type to the figure on screen.

1. The 7 inputs are read from the form on every keystroke: Number of Replicates, Replicate Mean, Replicate Standard Deviation, Suspect Reading, Check Cotton Assigned Value, Allowed Bias on Check Cotton and Method Repeatability SD.
2. Each value is checked against the accepted range in the input table below. A value outside its range stops the calculation rather than producing a misleading figure — the results blank out and a message appears.
3. The validated values are substituted into the expression above, which resolves Grubbs Statistic G together with every supporting figure in one pass — no value is carried over from a previous entry.
4. The supporting outputs — Maximum G Possible at This n, G as a Share of That Ceiling, Standard Error of the Mean, Replicate CV, Check Cotton Drift vs Allowed Bias and Replicate SD vs Method Repeatability — come from the same pass, so they always describe the same case as the headline figure.
5. Results are rounded for display only. The full-precision value is used throughout the chain, so reading a rounded intermediate figure back into the tool by hand can shift the last digit.

What each input means

Where to read each value on the floor, the unit it must be in, and the range the tool accepts.

What the tool returns

The headline figure and every supporting value it is built from.

Worked example

The tool loads with this case already solved — the Grubbs Statistic G shown above is its answer. Change one value and the difference from this baseline is the sensitivity of the result to that variable.

How to use it

1. Work through the input groups in order — Replicate Set and Instrument & Method. The defaults are a realistic case, so you can change one value at a time and watch what moves.
2. There is no calculate button. Every figure recalculates as you type or drag, which is what makes this usable for a what-if sweep rather than a single answer.
3. Read Grubbs Statistic G in the dark results panel — that is the headline figure, expressed in σ.
4. Check the supporting rows underneath (Maximum G Possible at This n, G as a Share of That Ceiling, Standard Error of the Mean, Replicate CV, Check Cotton Drift vs Allowed Bias and Replicate SD vs Method Repeatability) before acting on the headline — they are where an implausible input usually shows itself first.
5. Reset to defaults returns every field to the reference case, which is the quickest way to check whether a surprising result came from the tool or from an input you had changed earlier.

Where this is used

• Process planning — establishing Grubbs Statistic G before a trial is booked, so machine time and material in Fiber Testing, Bale Management & Laboratory Sampling are committed against a calculated figure rather than an estimate.
• Costing and quotation — Grubbs Statistic G is an input to the cost sheet, and quoting from a worked number rather than a remembered one is what keeps a margin intact.
• Troubleshooting — when the floor result drifts from plan, entering the measured values (starting with Number of Replicates) shows how much of the gap in Grubbs Statistic G each variable explains.
• Teaching and study — the accepted ranges bracket normal Fiber Testing, Bale Management & Laboratory Sampling practice, so moving one variable at a time shows the shape of the relationship rather than a single answer.

Assumptions and limits

• G is a statistic, not a verdict: it must be compared against the tabulated Grubbs critical value for the replicate count and significance level in use — at n = 8 that is about 2.13 at 5% and 2.27 at 1%. The ceiling shown is a hard algebraic limit, (n−1)/√n, and no single value in a set can exceed it; a computed G above the ceiling means the suspect reading was not part of the set that produced the mean and standard deviation, which is itself a finding. Grubbs assumes an otherwise normal population and tests one outlier only, so applying it repeatedly to strip successive extremes progressively destroys the very distribution it relies on. Above all, a statistical outlier is not a licence to delete data. A reading may be extreme because the specimen genuinely was — contamination, a seed-coat fragment, a different growth — and discarding it because it is inconvenient converts a measurement into an opinion. Record every excluded value and the physical reason for excluding it.
• Every input is bounded to the range normal practice occupies (Number of Replicates 3 to 200, Replicate Mean 0.001 to 100000 units and Replicate Standard Deviation 0.0001 to 10000 units, and so on for the rest). Those bounds are guard rails against typing errors, not a claim that the formula fails one unit outside them.
• The calculation is deterministic: the same inputs always give the same result. It carries no allowance for machine condition, operator skill, ambient conditions or lot-to-lot material variation unless an input above explicitly represents one.
• Nothing is sent anywhere. The maths runs in your browser, so the numbers you type never leave the page.

Questions people ask