How does a control chart show common versus special cause?

A control chart plots the measurement over time with a center line and control limits set about three standard deviations out. Random points within the limits indicate common cause variation. A point beyond a limit, or a non-random pattern like a run or trend, signals a special cause that something specific changed.

What is tampering in process control?

Tampering is adjusting a stable process in response to common cause variation, treating ordinary random noise as if it were a special cause. Deming's funnel experiment showed that adjusting point to point adds variation rather than removing it, so tampering makes a capable process worse.

Who coined the terms common cause and special cause?

Walter Shewhart introduced the distinction between chance causes and assignable causes of variation in the 1920s at Bell Labs and invented the control chart to separate them. W. Edwards Deming later renamed them common cause and special cause and applied the idea to management.

What percentage of variation is common cause?

Deming estimated in Out of the Crisis that about 94% of the problems and improvement potential in a system come from common causes, which are the responsibility of management, and about 6% from special causes. The exact split varies by process, but most variation typically lives in the system, not with individual workers.

Should operators or management fix each kind of variation?

Special causes are usually fixable at the floor level because the assignable cause is within an operator's reach, such as swapping a broken tool. Common cause variation lives in the system, so reducing it requires management to change materials, equipment, or conditions. Blaming workers for common cause variation does not reduce it.

Common Cause vs. Special Cause Variation

Q: What is the difference between common cause and special cause variation?

Common cause variation is the inherent, random variation a stable process shows from many small sources, and it is predictable in aggregate. Special cause variation comes from a specific, identifiable disturbance outside the normal process, such as a broken tool or a bad lot, and it can be traced and eliminated.

The one distinction that decides whether you should adjust a process or leave it alone.

Common cause variation is the natural, random noise built into a stable process from many small sources. Special cause variation comes from a specific, identifiable disturbance that is not part of the normal process. A control chart is what tells them apart.

This one distinction is the whole foundation of statistical process control. Get it wrong and you do real damage: you adjust a machine that was fine and make it worse, or you shrug off a genuine signal because it looked like ordinary scatter. Getting it right tells you exactly two things, when to leave a process alone and when to go find out what changed.

What is the difference between common cause and special cause variation?

Common cause is the variation a process shows when nothing special is happening; special cause is the variation that appears when something specific does. Every process wobbles a little, run to run, even when everything is nominal, slightly different material, a half-degree of ambient temperature, the normal play in a fixture, the small differences between one operator's motions and the next. Those many tiny, random inputs add up to common cause variation. It is stable and predictable in the aggregate: you cannot say what the next single reading will be, but you can say the range it will fall in.

Special cause variation is different in kind, not just size. It has a specific, findable source that is not part of the normal process: a tool that broke, a new lot of out-of-spec material, a setting bumped during changeover, a gauge that drifted. Because it has an assignable cause, you can hunt it down and remove it, and once you do, it is gone. Common cause you cannot chase one incident at a time, because there is no single incident, only the sum of the system's ordinary behavior.

Where did these terms come from?

From Walter Shewhart in the 1920s and W. Edwards Deming after him. Shewhart, working at Bell Labs, drew the line between what he called chance causes and assignable causes of variation, and invented the control chart to separate them economically. Deming carried the ideas into management and renamed them: common cause for Shewhart's chance causes, special cause for his assignable ones. The rename mattered because Deming's point was managerial, not just statistical.

Deming's estimate, from Out of the Crisis is the number worth remembering: in his experience, about 94% of the trouble and the improvement potential in a system belongs to the system itself, common causes, which only management can change, and roughly 6% to special causes. If that split is even close to right, then blaming individual workers for what is really common-cause variation is not just unfair, it is aimed at the wrong 6% and guaranteed to miss the 94%.

Random points inside the limits are the process breathing. The point outside is a signal something changed.

How does a control chart tell them apart?

By drawing control limits from the process's own data and watching whether points stay inside them and scatter randomly. A control chart plots the measurement over time with a center line at the process mean and upper and lower control limits, conventionally set three standard deviations out. As long as points fall randomly within those limits, the chart says the process is in statistical control, you are looking at common cause variation, and there is nothing to chase.

A special cause shows up as a signal the chart is built to catch: a point beyond a control limit, or a non-random pattern like a run of points on one side of the center line, a steady trend, or a sudden shift. Those patterns are statistically unlikely if only common causes are at work, so when one appears, the chart is telling you something specific changed. Crucially, control limits are not the customer's tolerance, they come from the process, which is a distinction worth its own read on control limits versus specification limits. The control chart answers one question: has something changed? It does not answer whether the parts are good enough; that is a separate question.

What is tampering, and why does it make things worse?

Tampering is reacting to common cause variation as if it were a special cause, adjusting a stable process in response to ordinary noise. It is the single most common and most expensive mistake in process control, and it feels like diligence, which is what makes it dangerous. An operator sees a reading a little high, nudges the setting down; the next reading, which was going to be lower anyway, now lands too low, so they nudge back up. Every adjustment is a response to randomness, and every adjustment injects a new offset the process did not have before.

Deming demonstrated this with his funnel experiment: dropping a marble through a funnel at a target, then moving the funnel to "correct" for where the last marble landed, reliably produces a wider scatter than leaving the funnel dead still. The lesson is exact and counterintuitive: when a process is in control, adjusting it point to point adds variation. The right move on common cause is to leave the individual readings alone and instead work on the system to shrink the spread. Reacting to every wobble is how a capable process gets turned into a scrap generator.

Chasing common-cause noise with adjustments (tampering) makes the spread bigger, not smaller. Deming proved it with a funnel.

How should you respond to each kind of variation?

The chart routes the response, and the two responses are almost opposites. Run it the same way every time so nobody has to guess.

Read the control chart first, before touching anything. Do not react to a single number in isolation; react to what the chart says about the pattern.
If the process is in control, do not adjust point to point. The variation is common cause. Chasing it is tampering. Leave the settings alone.
To reduce common cause variation, change the system. Better material consistency, tighter fixturing, reduced temperature swings, error-proofing, structural changes that shrink the whole spread. This is management-owned work, usually a root cause analysis or improvement project, not a knob turn.
If the chart shows a signal, treat it as special cause. Something specific changed for that point or run. Investigate promptly while the trail is warm.
Find and remove the assignable cause. The broken tool, the bad lot, the bumped setting. Fix that specific thing, then confirm the chart returns to random scatter.
Standardize what you learn. A recurring special cause deserves a permanent fix so it stops reappearing, and a persistent common-cause problem deserves a system change, both are candidates for a verified corrective action.

The chart routes the response. Treat common cause like special cause and you tamper; treat special cause like common cause and you ignore a real problem.

Who owns the fix, the operator or management?

Special causes are usually fixable at the floor level; common causes almost always require management to change the system. This is the practical heart of Deming's point. An operator can spot a special-cause signal, stop, and swap the broken tool or reject the bad lot, the assignable cause lives inside their reach. But an operator cannot fix the fact that the incoming material varies lot to lot, or that the shop has no temperature control, or that two machines are simply built to different tolerances. Those are common causes baked into the system, and only management controls the budget and decisions that change them.

That is why blaming workers for common-cause variation backfires. Pressure and retraining cannot move variation that lives in the system, so the defects keep coming and morale drops for nothing. The productive division of labor: give operators clear rules and the authority to act on special-cause signals, and hold management accountable for the system changes that shrink common-cause spread. Both need the same thing to do their job, the variation has to be visible, which is exactly what a live chart on the floor provides.

The evidence behind the split

Deming estimated that about 94% of the problems and improvement potential in a system come from common causes (the system, management's responsibility) and about 6% from special causes (W. Edwards Deming Institute, Knowledge of Variation).
Shewhart's control chart, developed at Bell Labs in the 1920s, was designed to distinguish the two kinds of variation while economically balancing the risk of over- and under-reacting (NIST/SEMATECH e-Handbook, control charts).
Reacting to common cause as though it were special cause, tampering, adds variation to a stable process rather than removing it, the core lesson of Deming's funnel experiment (ASQ, Control Chart).

What are common examples of each?

Grounding the idea in the kind of thing you see on a floor makes the call faster in the moment:

Common cause: the normal ±0.02 mm swing in a turned diameter from ordinary tool and material variation; the day-to-day scatter in fill weight from a well-behaved filler; small differences between shifts running the same standard work.
Special cause: a diameter that jumps when an insert chips; fill weight that shifts the moment a new resin lot loads; a burr that appears only after a fixture is bumped in changeover; a reading that drifts because a gauge lost calibration.

The tell is usually in the pattern, not the single value: common cause is the quiet, random hum inside the limits, while special cause arrives as a jump, a trend, or a run that the chart flags. None of this works if the data lives in a binder, though, you cannot see a trend you never plotted. Capturing checks and readings digitally at the station is what turns raw numbers into a live signal a team can actually act on, which is the gap Harmony's quality intelligence is built to close; our CLS case study shows one plant doing exactly that.