Pp process performance

Pp divides a two-sided specification width by six overall standard deviations in the conventional normal model. It uses spread alone. Report a location-sensitive index separately. Minitab: overall capability for normal analysis

In the worked example, limits 9.900–10.100 mm and overall SD 0.020 mm give Pp = 1.667. Moving the mean from 10.000 to 10.080 mm leaves Pp unchanged. Ppk falls to 0.333 because the upper-limit distance drops to 0.020 mm.

Formula and quantities

For the conventional six-sigma spread:

Pp = (USL − LSL) / (6 × s_overall)
s_overall = sqrt(sum((xi − x_mean)²) / (N − 1))

USL and LSL are the upper and lower specification limits. They must refer to the same characteristic and units as the measurements. Pp is dimensionless. The denominator uses all included individual observations about their overall mean; it is not the observed maximum minus minimum and not the standard error of the mean. A mean becomes more precisely estimated as sample size increases, but that does not make individual products less variable.

Minitab's normal capability analysis uses the sample standard deviation for its overall estimate by default and offers an unbiasing-constant option. Its within-subgroup estimate uses other methods. Record estimator settings when comparing reports from different software or configurations. Minitab: standard-deviation estimation

Minitab also permits changing the sigma-spread multiplier from its default of six. A result based on another multiplier is not directly comparable with the conventional index without identifying that change. Both specification limits are needed for Pp; for a genuinely one-sided requirement, use a corresponding one-sided measure rather than inventing a second limit. Minitab: Pp definition, formula settings

Distinguish Pp, Cp and Ppk

JMP distinguishes the within-subgroup C-family from the overall P-family and recommends examining differences between the estimates in the context of process stability. “Overall” describes the estimator and represented observations; calling it “long-term” does not make a short trial represent months of production. JMP: process capability and the P-family

For the normal model with the conventional multiplier:

Ppk = min((USL − x_mean)/(3 × s_overall),
          (x_mean − LSL)/(3 × s_overall))

Pp and Ppk coincide when the mean is at the specification midpoint. An approved process target may differ from that midpoint, so centering between limits and meeting the target are separate questions. Minitab: Pp and Ppk interpretation

Original reconciled example

Assume a dimension has LSL = 9.900 mm and USL = 10.100 mm, giving a 0.200 mm specification width. Suppose a defined dataset yields overall SD 0.020 mm:

six-SD spread = 6 × 0.020 = 0.120 mm
Pp = 0.200 / 0.120 = 1.667

If its mean is 10.000 mm, both distances to the limits are 0.100 mm, so Ppk = 0.100 / 0.060 = 1.667. If every reading instead shifts upward by 0.080 mm, the SD remains 0.020 mm and Pp remains 1.667. The mean becomes 10.080 mm; the upper-limit distance is only 0.020 mm, so Ppk = min(0.020/0.060, 0.180/0.060) = 0.333.

The unchanged Pp has not detected the location problem. These assumed summaries illustrate the arithmetic; they are not a validated capability study or a prediction of a particular defect rate. Shifting a dataset is also different from proving that a real process can be recentered without changing variation.

Establish what the sample represents

Define the part characteristic, measurement procedure, time window, production sources and inclusion rules. Preserve collection order and machine, cavity, tool, batch or shift identifiers where relevant. Do not combine unlike nominal dimensions into one distribution merely to increase sample size. A sample consisting only of accepted parts after sorting characterizes that selected population rather than the original unsorted output.

Minitab recommends enough continuous data over sufficient time to represent sources of variation and suggests trying to collect at least 100 observations for its normal capability analysis. It also recommends rational subgrouping where possible. This is software guidance, not a universal minimum imposed on every contract. Repeated measurements of one part cannot substitute for independent production observations. Minitab: data considerations

Assess stability using time-ordered information before projecting future performance. NIST describes stability in terms of constant mean, variance and distribution over time and uses plots or control charts to investigate it. Pp can still be calculated for an unstable historical dataset, but its number does not establish a predictable future process. Investigate shifts and mixtures rather than relying on a pooled histogram that hides chronology. NIST: assessing process stability

Distribution, uncertainty and acceptance limits

The six-SD ratio is computable for many datasets, but normal-tail interpretations require a suitable distribution model. Minitab's normal analysis calls for approximately normal data and directs users toward transformations or nonnormal analysis when appropriate. Counts and defect proportions need different analyses. Measurement error, rounding, truncation or an unsuitable pooled population can also distort the input distribution. Minitab: normal-analysis assumptions

Nonnormal software outputs may use different definitions. For example, Minitab's default nonnormal Pp uses specification limits transformed to standard-normal Z values. It is not simply the same raw-unit ratio applied without further assumptions. Identify the selected distribution, transformation and calculation method. Minitab: nonnormal overall capability

A sample Pp is an estimate. Minitab provides confidence bounds using a chi-square formulation for its normal calculation; report the confidence level and relevant assumptions when a threshold decision depends on estimation precision. A point estimate narrowly above a contractual benchmark is different evidence from a lower confidence bound above it. Minitab: Pp confidence bounds

Reporting and source scope

Record specification revision, units, sample size and dates, data-selection basis, mean, overall SD and estimator, multiplier, Pp, Ppk, uncertainty or confidence information, distribution assessment and stability evidence. State any applicable customer threshold separately; no Pp value is a universal product-release rule. Retain observed nonconformance counts rather than replacing them with the spread index.

Sources were reviewed November 1, 2026. The cited Minitab, JMP and NIST pages did not display article revision dates. Software-specific definitions and recommendations are identified as such; they are not presented as an automotive approval standard.

Inspection and quality documents