Process Capability Index Calculator — Cp, Cpk, Pp and Ppk

The process capability index calculator above compares what a process actually produces against what its specification demands. It returns Cp, the ratio of the specification width to the process spread; Cpk, the same idea but penalised for how far off-centre the process runs; and the Pp and Ppk pair, which repeat both calculations using long-term rather than short-term variation. It also converts the result into an expected number of defective parts per million and a sigma level, so the abstract index has a consequence attached to it.

Arb Digital publishes this page because capability indices are quoted constantly and understood rarely. A supplier reporting "Cpk 1.33" is making a specific statistical claim that depends on assumptions almost nobody checks: that the process is stable, that the measurements are roughly normal, and that the sigma being used is the right one for the index being named. When those assumptions fail, the index is not merely imprecise — it is meaningless, and reporting it anyway gives false confidence to everyone downstream.

What This Process Capability Index Calculator Does

It takes two specification limits, a process mean and two standard deviations, and produces the standard family of capability and performance indices. Cp is the specification width divided by six standard deviations of process spread. Cpk is the smaller of the two one-sided indices, Cpu and Cpl, which measure the distance from the mean to each limit in units of three standard deviations. Pp and Ppk are computed identically but with the overall standard deviation substituted for the within-subgroup one.

This is the opposite direction of travel from our control chart limits calculator, and the distinction is worth stating plainly. Control limits are derived from the process itself and describe what the process does when nothing unusual is happening. Specification limits come from the customer or the design drawing and describe what the process is required to do. Capability indices compare the two. A process can be perfectly in control and completely incapable, or capable on paper while wildly unstable — the two questions are independent, and you must answer the control question first.

How to Use It

1. Confirm the process is stable first. Plot the data on a control chart. If there are out-of-control signals, stop — the indices below will describe a process that does not have a single stable distribution to describe.
2. Enter the specification limits. These come from the drawing or the customer requirement, never from the data. Switch to one-sided if only one limit exists.
3. Enter the process mean. Use the grand average of your measurements.
4. Enter both standard deviations. Within-subgroup sigma drives Cp and Cpk; the overall sample standard deviation drives Pp and Ppk. If you only have one figure, enter it in both fields and note that the Cp/Pp distinction has been lost.
5. Read Cp against Cpk, then Cpk against Ppk. The first gap tells you about centring. The second tells you about drift between subgroups.

The Formula and How It's Calculated

Cp = (USL − LSL) ÷ 6σ_within. Cpu = (USL − μ) ÷ 3σ_within and Cpl = (μ − LSL) ÷ 3σ_within, with Cpk being the smaller of the two. Pp, Ppu, Ppl and Ppk use exactly the same expressions with σ_overall in place of σ_within.

Take the defaults. The specification runs from 9.5 to 10.5, a width of 1.0. With a within-subgroup sigma of 0.12, six sigma is 0.72, so Cp = 1.0 ÷ 0.72 = 1.3889. The process mean is 10.1, which is 0.1 above the midpoint of 10.0. Cpu = (10.5 − 10.1) ÷ 0.36 = 1.1111 and Cpl = (10.1 − 9.5) ÷ 0.36 = 1.6667, so Cpk is 1.1111 — the upper limit is the binding constraint. With the overall sigma of 0.15, Pp = 1.0 ÷ 0.90 = 1.1111 and Ppk = 0.4 ÷ 0.45 = 0.8889.

The defect estimate assumes a normal distribution and adds both tails. The upper limit sits (10.5 − 10.1) ÷ 0.12 = 3.33 standard deviations above the mean, which leaves about 429 parts per million above it. The lower limit sits 5.0 standard deviations below, contributing well under one part per million. So the expected defect rate is roughly 429 ppm, essentially all of it at the top end. The NIST/SEMATECH e-Handbook section "What is Process Capability?" sets out these estimators and their confidence intervals in full.

Cp Ignores Centring. Cpk Does Not.

This is the single most important thing to understand about the pair. Cp asks only whether the process spread would fit inside the specification width. It never looks at where the process actually sits. A process could be running entirely outside its tolerance and still report a Cp of 2.0, because the distribution is narrow enough to fit — it just is not in the right place. Cp is best read as "potential capability", meaning the capability you would have if you centred the process perfectly and changed nothing else.

Cpk closes that hole by measuring the distance to the nearer limit and taking the pessimistic answer. When the process is exactly centred, Cpu and Cpl are equal and Cpk equals Cp. As the mean drifts, the nearer index falls and Cpk falls with it while Cp stays fixed. So the ratio between the two is a direct measure of how much capability is being lost purely to being off target. In the defaults, Cpk is 80 per cent of Cp; a fifth of the available capability is being wasted on a 0.1-unit offset.

That framing turns the two numbers into a diagnosis rather than a score. A high Cp with a much lower Cpk means the process is precise but mis-aimed, and a setpoint adjustment might recover most of the gap in an afternoon. A low Cp means the spread itself is too wide for the tolerance, and no amount of centring will fix it — you need less variation, which usually means changing equipment, method, or material. Confusing the two leads to teams spending months chasing variation when the machine was simply set a fraction high.

Within-Subgroup Versus Overall Sigma: Why Four Indices Exist

Cp and Cpk use short-term variation, estimated from the spread inside rational subgroups collected close together in time. That figure captures the process's inherent, moment-to-moment noise. Pp and Ppk use the ordinary standard deviation of every measurement in the study, which also picks up any drift, shift or tool wear that happened between subgroups. The overall figure is therefore almost always the larger of the two, which is why Pp and Ppk are almost always lower than Cp and Cpk.

The gap between the two pairs is informative on its own. If Cpk and Ppk are close, the process is behaving consistently across the whole study period and short-term noise is essentially all the variation there is. If Ppk is much lower, something is moving between subgroups — a shift change, a batch of material, an ambient temperature swing — and that between-subgroup movement is where the improvement lies. Our standard deviation calculator gives you the overall figure from raw data, while the within-subgroup estimate normally comes off the control chart via the R-bar over d2 route.

Some organisations reserve Cp and Cpk for a validated, stable process and use Pp and Ppk for an initial study before stability has been demonstrated. Others use the pairs to mean short-term and long-term regardless of stage. Both conventions exist, so when someone quotes a capability index, ask which sigma produced it before you compare it to anything. ASQ's process capability resource covers the terminology and the study designs behind each.

Why an Index on an Unstable Process Is Meaningless

Every capability index is a statement about a single, fixed probability distribution: this is where it is centred, this is how wide it is, here is how much of it falls outside the limits. A process that is out of statistical control does not have one distribution. Its mean or its spread is moving, so the mean and standard deviation you computed are averages over a moving target, and the number you get describes a process that existed only in aggregate, over that window, and will not repeat.

The practical failure mode is predictable. You compute Cpk 1.4 from a month of data that happened to contain a high period and a low period that averaged out. The specification looks comfortably met. Next month the process sits in the high period alone and produces scrap continuously, and nothing in the index warned you, because the index had already averaged the warning away. This is why every serious treatment of capability puts stability first and calculation second, and why the calculator above will happily give you an answer that you should not use.

The same logic applies to normality. The defect-rate conversion assumes measurements follow a normal distribution, which fails badly for naturally bounded characteristics like flatness, concentricity or time-to-complete, where the distribution is skewed against a hard floor of zero. On skewed data the ppm estimate can be wrong by an order of magnitude in either direction, and a transformation or a non-normal capability method is required. The normal distribution calculator and the skewness and kurtosis calculator will tell you quickly whether the assumption is plausible.

Sigma Level, the 1.5 Shift, and What the Conventions Mean

Capability is often restated as a sigma level. The short-term sigma level is simply the distance from the mean to the nearer specification limit measured in standard deviations, which equals 3 × Cpk. The tool also reports a long-term sigma level that adds 1.5 to that figure. That 1.5 is a convention adopted in the six sigma tradition to allow for a process mean that drifts over time, and it is a convention, not a measurement — it is the reason a "six sigma" process is quoted at 3.4 defects per million rather than the two per billion that six true standard deviations would give.

Common industry practice treats Cpk of 1.33 as a routine requirement and 1.67 as a demanding one, with 1.0 meaning the specification limits sit exactly three standard deviations from the mean. Those figures are conventions that vary by sector and by customer, not universal thresholds, and this page does not tell you which one your process should meet. What the calculator can do is show you exactly how many parts per million each level implies, which is a far more useful basis for a conversation than a bare index. The z-score calculator performs the same tail conversion for any single value and limit.

Common Mistakes to Avoid

• Using control limits as specification limits — control limits come from the process, specification limits come from the requirement. Feeding one into the other guarantees a capability index near 1.0 that means nothing at all.
• Computing capability on an out-of-control process — the mean and sigma you use assume a single stable distribution. Without stability there is no distribution for the index to describe.
• Mixing the two sigmas — pairing the within-subgroup sigma with the Pp label, or the overall sigma with the Cp label, produces a number that is not the index you named.
• Quoting Cp alone — Cp cannot see a mis-centred process, so a good Cp with no Cpk beside it tells the reader nothing about whether parts are conforming.
• Trusting the ppm figure on skewed data — the defect estimate is a normal-distribution tail. On bounded or skewed characteristics it can be out by a factor of ten.

Related Free Tools From Arb Digital

Set the chart limits that establish stability with the control chart limits calculator, summarise a measurement run with the descriptive statistics calculator, check how much of a distribution sits within a given number of standard deviations using the empirical rule calculator, or attach an interval to an estimate with the confidence interval calculator. The full free online tools hub lists every quality and statistics calculator we publish.

Frequently Asked Questions

This page explains a statistical calculation for educational purposes only. It is not engineering, regulatory or compliance advice, and capability figures for a controlled process should be validated against your own measurement system analysis and quality plan.