"4 sigma" and "3,400 DPMO" describe the same process, but they answer different questions. Sigma level puts a defect rate on a single memorable scale that's easy to compare across projects and years. DPMO (and the DPU and DPO it's built from) is the arithmetic underneath that scale, normalized so a six-opportunity part and a one-opportunity part can be compared honestly. Rolled throughput yield answers a third, related question: given a process with several steps, what's the real probability one unit gets through all of them without a defect, a rework loop, or a scrap tag?
None of these numbers are hard to calculate. What trips teams up is skipping the definitions—what counts as a unit, a defect, an opportunity, a "pass"—and then being surprised when the resulting metric doesn't mean what they expected.
Why These Metrics Matter
One Scale, Any Product
Sigma level lets a stamped bracket and a populated circuit board be compared on the same axis, once opportunities are counted honestly.
Complexity-Adjusted, Not Just Defect Rate
DPMO accounts for how many ways a unit can fail, so a complex assembly isn't unfairly flattered next to a simple part.
Exposes the Hidden Factory
RTY reveals the rework and scrap that a healthy-looking final inspection pass rate quietly absorbs before anyone sees it.
A Common Improvement Language
"We moved from 3 sigma to 4 sigma" means the same thing in every functional area, which is exactly why Six Sigma adopted it as a project scorecard.
Core Terms
| Term | Meaning |
|---|---|
| Unit | One completed item, transaction, or instance being evaluated — a part, an order, an invoice. |
| Defect | Any failure to meet a specification, requirement, or customer expectation. One unit can carry more than one defect. |
| Opportunity | One meaningful chance for a defect to occur on a unit — a critical dimension, a required field, a solder joint. |
| DPU (Defects Per Unit) | Total defects divided by total units. Ignores opportunity count entirely. |
| DPO (Defects Per Opportunity) | Total defects divided by total opportunities (units × opportunities per unit). |
| DPMO | DPO expressed per million opportunities — DPO × 1,000,000. |
| First-pass yield | The fraction of units that clear a single step correctly the first time, with no rework. |
| Rolled throughput yield (RTY) | The product of every step's first-pass yield — the probability a unit clears the entire process first-time-right. |
The DPU → DPO → DPMO Ladder
These three numbers describe the same underlying defect count from three different angles, each one solving a limitation of the last.
DPU is the simplest and the easiest to explain, but it treats a part with one critical dimension the same as a part with ten. DPO fixes that by dividing by total opportunities instead of total units, so complexity is accounted for. DPMO just rescales DPO into a million-opportunity denominator, because "0.003125 defects per opportunity" is a clumsier number to put on a scoreboard than "3,125 DPMO."
From DPMO to Sigma Level
Sigma level converts a defect probability into a Z-score on the standard normal distribution — effectively asking, "how many standard deviations would a spec limit need to sit from the process mean to produce this defect rate?" A higher sigma level means a lower defect rate, on a scale that gets dramatically harder to climb the further up it goes.
| Sigma level (short-term) | DPMO (with 1.5σ shift) | Yield |
|---|---|---|
| 2σ | 308,537 | ~69% |
| 3σ | 66,807 | ~93% |
| 4σ | 6,210 | ~99.4% |
| 5σ | 233 | ~99.98% |
| 6σ | 3.4 | ~99.9997% |
Notice the last two columns don't shrink at a constant rate — each additional sigma level demands roughly an order-of-magnitude improvement in DPMO. That's why moving from 3σ to 4σ is achievable with focused process control, while 5σ to 6σ typically requires a fundamentally more capable process, not just tighter monitoring of the existing one.
The 1.5-Sigma Shift: Long-Term vs. Short-Term
Every sigma-level table has two versions hiding inside it, and confusing them is one of the most common sources of disagreement about what a process "really" is.
Long-Term Sigma
Calculated directly from the observed defect rate over the actual data collection period. This is what the process really did.
Short-Term Sigma
Long-term sigma plus a conventional 1.5σ allowance for the drift a process accumulates beyond what a shorter sampling window can see.
The 1.5σ shift is a convention from Motorola's original Six Sigma work, not a law of nature — it's an assumption that long-term process means drift by about 1.5 standard deviations relative to short-term capability studies. It's also the entire reason "6σ" is famously quoted as 3.4 DPMO: a true, unshifted 6σ process would be almost immeasurably good (roughly 2 defects per billion), but the shifted long-term figure everyone actually quotes assumes the process has drifted to what is really a 4.5σ long-term performance level.
When a sigma level appears without saying which one it is, assume short-term (shifted) unless told otherwise — it's the more commonly quoted convention and the one used in the familiar benchmark table above.
Why RTY Multiplies Instead of Averaging
A process with five steps running at 98%, 96%, 94%, 97%, and 99% first-pass yield looks healthy at a glance — the average step yield is a respectable 96.8%. But a unit has to survive every step in sequence, so the correct combined probability is the product of the five yields, not their average.
98% × 96% × 94% × 97% × 99% works out to roughly 84.8%, more than 12 points below the naive average. Multiplying several numbers just under 100% compounds faster than intuition expects, and the gap between the average and the true RTY grows with every additional step — which is exactly why a long process with many "acceptable" stations can still ship a surprisingly low fraction of units clean on the first try.
Worked Example: A Month at Ridgeline
Ridgeline Precision Machining runs the same bore-machining cell used throughout the SPC, Process Capability, and MSA guides. Over one month the cell produces 8,400 finished housings, each checked against six critical characteristics on the print: bore diameter, two length dimensions, thread pitch diameter, surface finish, and concentricity. Quality logs 189 defects across the month.
Step 1: DPU, DPO, and DPMO
- DPU = 189 / 8,400 = 0.0225 defects per unit.
- Total opportunities = 8,400 units × 6 characteristics = 50,400.
- DPO = 189 / 50,400 = 0.00375.
- DPMO = 0.00375 × 1,000,000 = 3,750.
Step 2: Sigma Level
Converting a DPMO of 3,750 into a Z-score gives a long-term sigma level of about 2.67σ. Adding the conventional 1.5σ shift gives a short-term sigma level of about 4.17σ — a capable, above-average process, consistent with the healthy Cpk of 1.21 found for this same cell in the Process Capability guide, but still a meaningful distance from the 6σ frontier.
Step 3: Per-Unit Yield Becomes an RTY Input
The unit-level yield for this step — 1 − DPU = 1 − 0.0225 = 97.75% — is the machining cell's first-pass yield. That number now becomes one link in a longer chain: the housing still has to clear rough turning, deburr, wash, and final inspection before it ships.
| Step | First-pass yield | Cumulative yield |
|---|---|---|
| Rough Turn | 99.2% | 99.2% |
| Finish Turn / Bore (this guide's cell) | 97.8% | 97.0% |
| Deburr | 99.5% | 96.5% |
| Wash | 99.8% | 96.3% |
| Final Inspection | 99.0% | 95.4% |
RTY = 0.992 × 0.978 × 0.995 × 0.998 × 0.990 = 95.4%. Of the 8,400 housings that start the month, roughly 8,012 are expected to clear all five steps first-time-right, with about 388 falling out somewhere along the way to rework, scrap, or repeat inspection — even though the weakest single step, Finish Turn/Bore, still clears a respectable 97.8% on its own.
How to Interpret the Results
DPMO Needs a Stable Opportunity Count
A DPMO trend is only meaningful if the opportunity definition hasn't changed underneath it from one reporting period to the next.
Short-Term vs. Long-Term Isn't Optional Detail
Quoting a sigma level without saying which one invites a comparison against the wrong benchmark.
RTY Below the Weakest Step Yield Is Impossible
If a calculated RTY comes out higher than the worst individual step yield, a data entry or units error is the more likely explanation than good news.
Final Yield Is Not RTY
A strong final-inspection pass rate can coexist with a mediocre RTY when rework absorbs the difference upstream and never gets counted.
Common Mistakes
Inflating or Shrinking Opportunity Counts
Adjusting the opportunity count to make DPMO look better defeats the entire purpose of a normalized metric.
Averaging Step Yields Instead of Multiplying
The average always overstates the true combined yield, and the gap grows with every additional step in the chain.
Counting Reworked Units as First-Pass Good
A unit that needed a second pass through a step was not a first-pass success, even if it eventually shipped clean.
Treating Sigma Level as a Precise Physical Property
Sigma level is a convenient rescaling of a defect rate, not a direct physical measurement — small sample sizes make it noisy at the tails.
Quick Reference
Setup Checklist
- Unit, defect, and opportunity definitions are documented and held constant.
- Every sigma level quoted states whether it is short-term (shifted) or long-term.
- RTY step yields reflect first-pass results only, with rework excluded.
- Opportunity counts per unit are reviewed on a fixed schedule, not adjusted reactively.
Using the Result
- Use DPMO and sigma level to compare dissimilar products on one consistent scale.
- Use RTY to find where a multi-step process is losing more than local yields suggest.
- Target the weakest RTY step first, but expect gains from the whole chain, not one station.
- Re-baseline DPMO and RTY after any process, inspection, or definition change.
Sources and Further Reading
- Mikel Harry and Richard Schroeder, Six Sigma: The Breakthrough Management Strategy.
- Forrest Breyfogle III, Implementing Six Sigma: Smarter Solutions Using Statistical Methods.
- ASQ Certified Six Sigma Black Belt Body of Knowledge.
- ASQ Certified Quality Engineer Body of Knowledge.