A process is in control when all of its variation comes from normal, predictable sources and none from unusual or external disruptions. On a control chart, this looks like data points scattered randomly within the upper and lower control limits, with no unusual patterns. The moment a point falls outside those limits, or the data form a suspicious pattern, the process is considered out of control.
The Two Types of Variation
Every process has variation. A machine filling bottles won’t put exactly 500 mL in every single one. The question is whether that variation is normal or whether something has gone wrong.
Common cause variation is the natural background noise of a process. It comes from dozens of small, random factors: slight temperature fluctuations, minor differences in raw materials, tiny vibrations in equipment. This type of variation is inherent and expected. A process experiencing only common cause variation is, by definition, in statistical control. It’s predictable, and you can describe its behavior with a stable average and a consistent spread.
Special cause variation comes from something identifiable and external: a worn-out tool, a new batch of defective material, an untrained operator, a power surge. These causes push the process outside its normal behavior. When special cause variation is present, the process is out of control because its output is no longer predictable.
The entire goal of statistical process control is to separate these two types. You leave common cause variation alone (it requires systemic changes to reduce) and hunt down special causes to eliminate them.
How Control Limits Work
Control charts use a center line (the process average) and two boundaries called the upper control limit (UCL) and lower control limit (LCL). These limits are typically set at three standard deviations above and below the center line. That three-sigma range captures about 99.7% of data points when a process is running normally, so a point landing outside that range is a strong signal that something unusual happened.
The simplest version of this calculation for an average chart uses the formula: UCL equals the overall average plus a factor (called A₂) multiplied by the average range of your samples, and the LCL equals the overall average minus that same value. The specifics change depending on sample size and chart type, but the core idea stays the same: measure the process’s natural spread, then draw boundaries at three standard deviations.
These limits are not the same as specification limits. Specifications come from customer requirements or engineering tolerances. Control limits come from the process itself. A process can be perfectly in control, with all points inside the control limits, while still producing output that fails to meet specifications. That distinction matters.
In Control vs. Capable
Being in control means the process is stable and predictable. Being capable means it can consistently meet your quality requirements. These are two separate questions, and you need to answer them in order.
A stable process that produces too much variation to fit within specification limits is in control but not capable. Think of a dart thrower who always hits the same general area of the board (consistent, predictable) but that area is six inches left of the bullseye. The throws are stable, but they don’t meet the target. Capability indices like Cp and Cpk quantify how well a stable process fits within specification limits. You can only meaningfully calculate capability after you’ve confirmed the process is in control, because an unstable process gives you unreliable capability numbers.
Pattern Rules That Signal Trouble
A single point outside the control limits is the most obvious out-of-control signal, but it’s not the only one. Several sets of rules exist to catch subtler problems. The most widely used are the Western Electric rules and Nelson rules, both of which divide the control chart into zones based on one, two, and three standard deviations from the center line.
Western Electric Rules
These rules, developed at Bell Telephone Laboratories, flag a process as out of control if any of the following occur:
- One point beyond 3 sigma: Any single data point above the UCL or below the LCL.
- Two out of three points beyond 2 sigma: Two of the last three consecutive points fall more than two standard deviations from the center, on the same side.
- Four out of five points beyond 1 sigma: Four of the last five points fall more than one standard deviation from the center, on the same side.
- Eight consecutive points on one side: Eight points in a row all above or all below the center line.
Each rule targets a different type of process shift. The first catches sudden, dramatic changes. The last three catch gradual drifts that might not trigger a single out-of-limits point but still indicate the process has shifted.
Nelson Rules
Nelson rules expand on the Western Electric set with eight tests. Beyond the four above, they add detection for:
- Trends: Six or more points in a row continuously increasing or decreasing, suggesting a drifting process (a wearing tool, for example).
- Alternating patterns: Fourteen or more points alternating up and down, which suggests two different sources are influencing the process in turns (such as two alternating machines or shifts).
- Hugging the center: Fifteen consecutive points all within one standard deviation of the mean. This sounds good but actually suggests the data is too uniform, possibly indicating mixed data streams or incorrect control limit calculations.
- Avoiding the center: Eight points in a row with none within one standard deviation of the mean, with points on both sides. This bimodal pattern often means two distinct processes are being charted together.
Reading a Control Chart
When you look at a control chart for a process in control, you should see data points bouncing randomly above and below the center line. Roughly two-thirds of the points should fall within one standard deviation of the center. Points near the control limits should be rare. There should be no obvious trends, cycles, or clusters.
A shift, where eight or more points line up on one side of the center, means the process average has moved. A trend of six or more points steadily climbing or falling means something is gradually changing, like a tool wearing down or a chemical concentration slowly drifting. Cycling patterns, where values swing up and down in a regular rhythm, often point to environmental factors like temperature changes between day and night shifts.
The key insight is that random-looking data is actually good news on a control chart. It means nothing unusual is acting on the process. Patterns that look orderly or structured are the warning signs.
What Happens When a Process Goes Out of Control
When a control chart signals an out-of-control condition, organizations typically follow an out-of-control action plan (OCAP). This is a predefined set of steps so operators don’t have to improvise under pressure.
The typical sequence starts with stopping or isolating the affected output to prevent defective product from moving forward. Next, an operator or engineer investigates to identify the special cause. This might involve checking equipment settings, inspecting raw materials, reviewing recent changes in personnel or procedures, or examining environmental conditions. Once the root cause is found, corrective action is taken: replacing a worn part, recalibrating equipment, retraining an operator, or adjusting a supplier process.
After the fix, the process is monitored to confirm it has returned to a stable state. Some industries, particularly semiconductor and pharmaceutical manufacturing, use automated monitoring systems that trigger these action plans in real time, reducing the window between a process shift and the corrective response.
The distinction between reacting to special causes and trying to “fix” common cause variation is critical. Tampering with a stable process because of normal random variation, sometimes called overadjustment, actually increases variation and makes things worse. Control charts exist precisely to tell you when to act and when to leave things alone.