Six Sigma Calculator

What this Six Sigma calculator does

Six Sigma turns a defect count into a single number describing how capable a process is. The scale runs from about 1 sigma, which is a process producing defects most of the time, up to 6 sigma, which is the famous target of 3.4 defects per million opportunities.

This calculator does two jobs, selected from the dropdown at the top.

Sigma takes your defects, units and opportunities per unit and returns the sigma level, DPMO, yield, and the proportion of units carrying at least one defect.

Sample Size takes a standard deviation, a maximum acceptable error and a confidence level, and returns how many samples you need.

Worth saying plainly: the sigma conversion here is done properly, using the inverse normal distribution rather than a lookup table, and it includes the 1.5 sigma shift that the methodology requires. Fed the canonical DPMO figures it returns exactly 1, 2, 3, 4, 5 and 6 sigma. A surprising number of free calculators get that last part wrong.

Everything runs in your browser. Nothing typed here is stored or sent anywhere.

How to use it

Sigma mode

1. Defects. Total defects found. Note that this counts defects, not defective units: one unit with three faults contributes three.
2. Number of Units. How many items you inspected.
3. Opportunities per Unit. How many distinct ways a single unit could be defective. This is the judgement call and it is worth reading about before you pick a number.

Sample Size mode

1. Standard Deviation. The variability of the measurement, from historical data or a pilot sample.
2. Maximum Error. How far from the true mean you are willing to be, in the same units.
3. Confidence Level. How sure you want to be that the true value falls within that error.

Press Calculate. Press Reset to clear it.

Leaving defects blank in sigma mode is treated as zero defects, which returns 6 sigma. That is correct arithmetic and worth noticing, because a blank field and a perfect process produce the same answer.

DPMO, and why opportunities matter

Everything runs through one intermediate figure.

DPMO = (defects ÷ (units × opportunities per unit)) × 1,000,000

Defects per million opportunities. The denominator is total opportunities across everything you inspected, and the million is there to give a number big enough to talk about.

The opportunities count is the part that decides your answer and the part nobody standardised. An opportunity is a distinct chance for a defect to occur. A soldered board with 200 joints has 200 opportunities. A form with 12 fields that could each be filled in wrongly has 12.

And here is the problem. Counting opportunities generously makes the process look better, because it enlarges the denominator. The same 25 defects across 1,200 units gives a very different sigma level depending on whether you call it 4 opportunities per unit or 40.

The discipline that keeps this honest is to count only opportunities that are genuinely independent and genuinely inspected. If you are not actually checking all 200 solder joints, you cannot claim 200 opportunities. If two supposed opportunities always fail together, they are one.

The practical consequence: sigma levels are comparable within an organisation that counts opportunities consistently, and much less comparable between organisations that do not. When someone quotes a sigma level at you, asking how they counted opportunities is a fair question.

A worked example

You inspect 1,200 units, each with 4 opportunities for a defect, and find 25 defects.

Total opportunities: 1,200 × 4 = 4,800

DPMO: (25 ÷ 4,800) × 1,000,000 = 5,208.3

Yield per opportunity: 100% − 0.5208% = 99.4792%

Sigma level: 4.06

Unit yield: 0.994792 to the power of 4 = 97.93%

So about 4.06 sigma, which is a fairly typical figure for a competent manufacturing process and well short of the six sigma target.

The last line is the one worth pausing on. Your yield per opportunity is 99.48 percent, which sounds excellent. But each unit has four chances to fail, so the proportion of units leaving with at least one defect is 2.07 percent, roughly four times the per-opportunity defect rate. That gap has a section of its own.

The sigma table

The canonical relationship between sigma level and defect rate, which this calculator reproduces exactly:

Read down the DPMO column and the scale reveals itself. Each step is not a fixed increment, it is roughly an order of magnitude. Going from 3 to 4 sigma cuts defects by about ten times. Going from 5 to 6 cuts them by nearly seventy times.

Which explains two things about Six Sigma programmes. Early improvements come relatively easily, because moving from 3 to 4 sigma means fixing obvious problems. And the last stretch is brutally hard, because eliminating 229 defects per million to get from 5 to 6 requires controlling sources of variation that most processes cannot even measure.

It also explains why most real processes sit between 3 and 4 sigma, and why a genuine 6 sigma process is rare enough to be worth talking about.

The 1.5 sigma shift, and why 3.4 is the famous number

Here is something that puzzles people who check the arithmetic themselves.

Look up 3.4 defects per million in a standard normal table and you get about 4.5 sigma, not 6. The two do not match, and it is not an error in the table.

The reason is a convention built into the methodology. Six Sigma assumes that a process which is perfectly centred in the short term will drift over the long term, as tooling wears, materials vary, shifts change and conditions move. The standard allowance for that drift is 1.5 sigma.

So the sigma level quoted is the short-term capability plus 1.5:

Sigma level = inverse normal of the yield + 1.5

Which makes the two versions line up:

So a 6 sigma process is really one whose long-run defect rate is 3.4 per million, allowing for the fact that it will not stay perfectly centred. This calculator applies the shift, which is why its figures match the published table.

The 1.5 figure is a convention rather than a law, and it has been argued about since it was introduced. If your process is genuinely stable and you have the data to show it, a smaller shift is more honest. The important thing is knowing which convention a quoted sigma level used, because the difference between short-term and long-term sigma is a full 1.5 levels, and that is the gap between a good process and a world class one on paper.

Why the unit figure is worse than the opportunity figure

This is the section most worth reading if you are reporting quality to anyone.

Your yield per opportunity looks good because it is measured against a very large denominator. But a customer does not receive an opportunity, they receive a unit, and a unit fails if any of its opportunities fail.

If each opportunity has a probability p of being defect free, and there are n of them, then the chance the whole unit is clean is p to the power of n. That is what the calculator reports.

On our example, 99.4792 percent per opportunity across 4 opportunities gives a unit yield of 97.93 percent. So one unit in fifty carries a defect, from a process that reports a defect rate of half of one percent.

Watch what happens as complexity rises, holding the per-opportunity yield fixed at 99.4792 percent:

At 200 opportunities per unit, a process with an excellent per-opportunity rate delivers a clean unit only a third of the time.

This is the real argument for Six Sigma in complex products, and it is more persuasive than any of the branding around it. A modern car or aircraft has millions of opportunities. At 3 sigma per opportunity, nothing would ever leave the factory working. The reason to chase very low defect rates per opportunity is that complexity multiplies them mercilessly.

It is also an argument for reducing opportunities where you can. Simplifying a design does more for unit yield than tightening a process, because it changes the exponent rather than the base.

The sample size mode

Switch the dropdown to Sample Size and the tool answers a different question: how many measurements do you need to estimate a process mean within a stated tolerance?

n = (z × standard deviation ÷ maximum error)²

Where z is the two-sided normal value for your confidence level. The result is rounded up, because you cannot take four fifths of a sample.

The z values behind the dropdown:

Two things fall out of that formula and both are worth internalising before you plan a study.

Precision is expensive. The error term is squared, so halving your acceptable error quadruples the sample size. Going from an error of 1 to an error of 0.5 at 95 percent confidence takes you from 97 samples to 385.

Confidence is comparatively cheap. Moving from 90 to 95 percent confidence costs you about 40 percent more samples. Moving from 95 to 99 costs about 70 percent more. Meaningful, but nothing like the cost of tightening precision.

So when a sampling plan comes back impossibly large, the lever to pull is almost always the tolerance rather than the confidence level. And if the standard deviation is the problem, reducing process variation shrinks the sample requirement quadratically too, which is a rather pleasing argument for fixing the process before measuring it more.

The formula assumes you know the standard deviation and are estimating a mean from a roughly normal distribution. For small samples where the standard deviation is itself estimated, a t distribution is more correct and gives a slightly larger number.

Questions people ask

How do I calculate DPMO?

Divide defects by total opportunities, then multiply by a million. Total opportunities is units times opportunities per unit.

Why is 6 sigma 3.4 defects per million and not 2 per billion?

Because of the 1.5 sigma shift built into the methodology to allow for long-run process drift. Without it, 6 sigma would be about 2 defects per billion. See the section above.

How do I count opportunities per unit?

Count distinct, independent ways a unit can be defective, and only ones you actually inspect. Counting generously inflates your sigma level without improving anything.

Do I count defects or defective units?

Defects. One unit with three faults contributes three. If you count defective units instead, your DPMO will be understated.

What sigma level is normal?

Most real processes sit between 3 and 4 sigma. Six sigma is rare and hard, because each step up the scale is roughly an order of magnitude fewer defects.

Why is my unit yield lower than my opportunity yield?

Because a unit needs every one of its opportunities to be clean. Raise the per-opportunity yield to the power of the number of opportunities. With 4 opportunities at 99.48 percent, unit yield is 97.93 percent.

How do I halve my margin of error?

Take four times as many samples. The error term is squared in the formula, so precision gets expensive quickly.

I left defects blank and got 6 sigma. Is that right?

A blank is treated as zero defects, and zero defects is 6 sigma. Correct arithmetic, but check you meant to leave it empty.

References

A note on sourcing. The sigma level conversion used here is the inverse of the standard normal cumulative distribution applied to the yield, plus the conventional 1.5 sigma long-term shift, which is what makes 3.4 defects per million correspond to a 6 sigma process. The sample size formula for estimating a mean, and the underlying normal distribution theory, are set out in the NIST and SEMATECH engineering statistics handbook, which is freely available.

1. National Institute of Standards and Technology and SEMATECH, e-Handbook of Statistical Methods, NIST Handbook 151. https://www.itl.nist.gov/div898/handbook/
2. National Institute of Standards and Technology, NIST/SEMATECH Engineering Statistics Handbook. https://www.nist.gov/programs-projects/nistsematech-engineering-statistics-handbook
3. Acklam, P. J., An Algorithm for Computing the Inverse Normal Cumulative Distribution Function, rational approximation used for the normal quantile in this calculator.