Special cause variation is an unpredictable deviation in a process that results from a cause not normally part of that process. It stands apart from the routine, expected fluctuations that happen in any system. Think of it this way: if your commute normally takes between 25 and 35 minutes due to ordinary traffic patterns, that range is normal variation. But if a bridge collapse adds 90 minutes one day, that’s a special cause. Something specific, identifiable, and outside the usual system created the change.
The concept was developed by Walter Shewhart in the 1920s at Bell Labs and later expanded by W. Edwards Deming. It remains foundational in quality improvement across manufacturing, healthcare, software development, and virtually any field where people track process performance over time.
Common Cause vs. Special Cause Variation
Every process has two types of variation baked into it. Common cause variation is the random noise present in any stable process. It comes from the many small, interacting factors that are always at work: slight differences in materials, minor fluctuations in temperature, natural variation in human performance. No single factor dominates. The process is behaving as designed, and the results stay within a predictable range.
Special cause variation is fundamentally different. It comes from a specific, identifiable source that isn’t part of the process under normal conditions. A machine breaks down. A new, untrained operator takes over a workstation. A software update introduces a bug. A hospital changes its discharge protocol. These are discrete events that push results outside the expected range or create a noticeable pattern in the data.
The distinction matters because each type demands a completely different response. Common cause variation can only be reduced by redesigning the process itself. Special cause variation requires finding and addressing the specific thing that changed. Confusing the two leads to wasted effort at best and a worse process at worst.
How Control Charts Detect It
The primary tool for spotting special cause variation is a control chart, sometimes called a statistical process control (SPC) chart. A control chart plots data points over time against three reference lines: a center line representing the process average, an upper control limit, and a lower control limit. These control limits are typically set at three standard deviations above and below the average. When a process has only common cause variation, roughly 99.7% of data points will fall between these limits.
Three widely used rules signal that special cause variation is present:
- A single point outside the control limits. Any data point that lands above the upper limit or below the lower limit is an immediate signal. Something unusual happened at that moment.
- Eight consecutive points on the same side of the center line. Even if every point is within the control limits, having eight in a row all above (or all below) the average is extremely unlikely by chance alone. This pattern, called a “shift,” suggests the process has fundamentally changed.
- Six or more consecutive points moving in the same direction. Six data points in a row that are each higher than the last (or each lower) indicate a trend. Something is gradually pulling the process in one direction.
More advanced rule sets exist for finer detection. The Western Electric rules, for example, also flag situations like two out of three consecutive points falling beyond two standard deviations from the center, or four out of five points beyond one standard deviation. These catch subtler shifts that a single-point rule would miss. A full set of eight detection rules, known as the Nelson rules, adds patterns like fourteen points alternating up and down (suggesting the process is being over-corrected) or fifteen consecutive points clustered unusually close to the center line.
What It Looks Like in Practice
In manufacturing, special cause variation shows up when something concrete goes wrong or changes. A machine fails at one station on the production line and daily throughput drops noticeably. A forklift used to move materials between stations develops a mechanical problem and moves more slowly, creating a bottleneck that wouldn’t exist under normal operating conditions. These aren’t gradual trends or random bad days. They’re traceable to a specific event.
In healthcare, a hospital tracking the percentage of patient problem lists reviewed during admission might see a baseline hovering around 70%. After introducing a new protocol, data points start falling outside the upper control limit, eventually stabilizing around 90%. That jump from 70% to 90% is special cause variation triggered by a deliberate intervention. In one documented case, the improvement showed up through multiple signals simultaneously: individual points outside the control limits, six consecutive months of increasing values forming a trend, and eventually eight straight months above the old average, confirming the shift was real and sustained.
Not all special cause variation is bad. A quality improvement project that successfully reduces infection rates will show up as special cause variation on a control chart. The point is that something outside the normal process created the change, whether that something was a equipment failure or a carefully designed intervention.
What to Do When You Find It
When a control chart signals special cause variation, the first step is to use the time-ordered nature of the chart to pinpoint when the change happened. Because each data point corresponds to a specific time period, you can look back at what was different during that window: new staff, equipment changes, supply substitutions, policy updates, or environmental factors.
Once you’ve identified the cause, the response depends on whether the variation is desirable or not. If a machine failure caused a drop in output, you fix the machine and put safeguards in place to prevent recurrence. If a new training program caused an improvement in quality scores, you institutionalize the training. Either way, after identifying and addressing the special cause, recalculate the control limits and center line excluding the affected data points. This gives you a clean baseline going forward.
When the special cause represents a permanent, intentional change to the process, such as new equipment or a redesigned workflow, recalculate the limits using data from the new operating conditions. The old limits no longer reflect reality.
The Danger of Treating One Type as the Other
The most common and costly mistake in process management is reacting to common cause variation as though it were special cause. Deming called this “tampering.” Every process has natural ups and downs. If you adjust settings every time output dips slightly, you’re not fixing a problem. You’re adding variation to a system that was behaving normally. Each unnecessary adjustment compounds the last, and the process becomes less stable than it was before you started. Deming demonstrated this with his famous funnel experiment, showing that each of four increasingly aggressive adjustment strategies made outcomes progressively worse.
The reverse mistake is also damaging. If you ignore genuine special cause variation and treat an unusual result as just another random fluctuation, you miss the opportunity to find and fix a real problem, or to learn from a real improvement. A control chart prevents both errors by giving you an objective, statistical basis for deciding which type of variation you’re looking at, rather than relying on gut feeling or arbitrary thresholds.
In a stable process with only common cause variation, a false alarm (a point that appears to be special cause but isn’t) occurs roughly once every 370 data points on average. That low false-alarm rate is built into the three-standard-deviation limits and is what makes control charts reliable enough to act on.