The 1.5 Sigma Shift: Why Six Sigma Means 3.4 DPMO

A Six Sigma table reports 3.4 defects per million opportunities at a sigma level of 6.0. The two-sided tail of a centered normal distribution at the same sigma level is 0.002 defects per million. Both calculations are mathematically valid, but they describe different process conditions.

The difference comes from the 1.5 sigma shift convention. Standard Six Sigma tables begin with a short-term capability figure, subtract 1.5 sigma to account for possible long-term process drift, and report the one-sided tail beyond the nearest specification limit. A direct normal-distribution calculation assumes that the process remains centered and reports both tails.

At 6.0 sigma, those conventions produce results that differ by 1,722 times. That difference becomes operationally significant when a short-term capability study is quoted as a forecast of long-term production performance.

What the 1.5 sigma shift assumes

A sigma level expresses the distance between the process mean and a specification limit in units of process standard deviation. For capability results reported through Cpk, the relationship is:

Z = 3 × Cpk

A Cpk of 2.00 therefore corresponds to a short-term sigma level of 6.0. If the process is centered on nominal and remains there, each specification limit is 6.0 standard deviations from the mean. The resulting two-sided defect rate is 0.002 DPMO.

The conventional Six Sigma table applies an additional assumption. It allows the process mean to move by 1.5 standard deviations during long-term operation. Once that shift is applied, the nearest specification limit is only 4.5 standard deviations from the shifted mean. The table then reports the one-sided tail beyond that limit, which is 3.4 DPMO.

Figure 1 shows the assumption directly. The solid blue curve represents a process centered on nominal, with specification limits at minus 6 and plus 6 sigma from its original center. The dashed red curve has moved 1.5 sigma toward the upper specification limit. Its nearest limit is therefore 4.5 sigma from the new mean.

The table value does not come from observed drift in the process being evaluated. It comes from applying the standard 1.5 sigma allowance. That allowance has a practical basis, since process means can move as tooling, materials, settings, environmental conditions, and operating practices change. The reporting problem occurs when the allowance remains unstated.

Two defect rates associated with one sigma level

The following table shows the two conventions across the commonly reported range. The sigma table column applies Z - 1.5 and uses a one-sided tail. The centered column uses the two-sided tail of a process operating on nominal.

Figure 2 shows how the difference increases across the table. At 3.0 sigma, the sigma table result is 25 times the centered result. At 4.0 sigma, it is 98 times the centered result. By 6.0 sigma, the ratio has reached 1,722 times.

The curves separate continuously because moving the mean 1.5 standard deviations has a larger relative effect as the original tail probability becomes smaller. This is why the disagreement becomes most visible near the top of a sigma table, where both results may be described informally as “Six Sigma performance” even though their numerical meanings differ substantially.

How much of the quoted DPMO comes from the shift

The 1.5 sigma shift is constant on the Z axis, so its effect can appear to be a flat penalty. Its contribution to the quoted DPMO grows as process capability increases. At higher sigma levels, nearly all of the quoted defect rate comes from the shift convention, while progressively less comes from the centered distribution observed by the study.

At a sigma level of 3.0, the assumed shift accounts for 95.96% of the quoted rate. At 6.0, it accounts for 99.94%: only 0.002 of the quoted 3.4 DPMO is attributable to the observed distribution.

The shift convention has a practical basis, since processes drift over time. Reaching a short-term Z of 6.0 also represents genuine process capability. The reporting issue is that the convention is usually left unstated, even when it accounts for almost the entire quoted defect rate. A capability study provides the measured quantities needed to separate these interpretations.

A capability study example

Consider a capability study based on 30 subgroups of 5, using within-subgroup variation, which produces a Cpk of 1.33. The corresponding short-term sigma level is:

Z = 3 × 1.33 = 3.99

Interpreted as a centered process, that result corresponds to 66 PPM. Interpreted through the conventional sigma table, with the 1.5 sigma shift applied, it corresponds to 6,387 PPM. The second figure is 97 times the first.

These values differ slightly from the 4.0 sigma row in the table. The table gives 63.3 centered DPMO and 6,209.7 shifted DPMO at exactly 4.0 sigma. A reported Cpk of 1.33 is a rounded capability value, and multiplying it by 3 gives 3.99, so its calculated tail areas should remain based on 3.99.

This distinction matters when the study is used to estimate production over a longer period. The study measures short-term variation from the sampled subgroups. It does not measure every source of movement that may affect the process mean over the production period being forecast. Applying the shift adds an allowance for that movement, while the centered calculation assumes that the sampled process distribution continues to operate on nominal.

Neither result is a direct measurement of future defects. Each is a model-based conversion of the same short-term capability result under a different assumption about centering.

How to identify which number is being reported

The first question is whether the defect rate came from a short-term capability study or from observed long-term performance.

If the source is Cpk, the sigma level normally comes from Z = 3 × Cpk, using an estimate of short-term within-subgroup variation. A DPMO value taken from a conventional Six Sigma table then includes the 1.5 sigma shift, even when the capability report never observed that amount of drift.

If the source is actual defect data collected over the operating period, the result is observed long-term performance. Converting that result back into a sigma level requires identifying whether the conversion uses a shifted one-sided convention or a centered two-sided convention. The sigma label alone does not provide that information.

A complete report should therefore identify the source statistic, the variation estimate, the centering assumption, the tail convention, and whether the 1.5 sigma shift was applied. The sigma level calculator can perform the conversions while keeping the shifted and centered results separate.

What this changes in plant reporting

A plant can appear to move from 66 PPM to 6,387 PPM with no change in equipment, material, settings, operators, or measured output. The apparent change can result entirely from switching between a centered interpretation and the shifted sigma-table convention.

That reporting error affects decisions about containment, improvement priorities, customer risk, and expected yield. It can also create a false comparison between a capability study and production defect history, since the capability result is based on a distribution model while the history records defects that actually occurred.

The required control is explicit labeling. A short-term sigma level should be identified as short-term. A sigma-table DPMO should state that it includes the 1.5 sigma shift and a one-sided tail. A centered estimate should state that it uses both tails and assumes operation on nominal. Observed long-term defects should remain identified as observed performance.

With those definitions included, 3.4 DPMO and 0.002 DPMO can coexist in the same analysis without contradiction. They answer different technical questions and depend on different assumptions about process behavior over time.