Statistical process control (SPC) is a method for monitoring, controlling, and improving a process by applying ongoing statistical analysis to data taken from that process over time. You sample the process at planned intervals, calculate a statistic from each sample, and plot it in time order against control limits calculated from the process itself. Those limits separate common cause variation, which is inherent to the process, from special cause variation, which indicates that something has changed. Because the analysis happens while the process is still running, statistical process control lets you act on a change before nonconforming product reaches the customer. The core tool of SPC is the control chart, developed by Walter A. Shewhart in 1924.
Where Statistical Process Control Came From
In May 1924, Walter A. Shewhart sketched the first control chart in a one-page memo written for Western Electric’s inspection engineering group, the organization that became Bell Telephone Laboratories. With it came the idea that a process could be in a state of statistical control.
SPC gained wide usage during World War II, when the military applied it in munitions and weapons facilities. Wartime demand forced a search for a more efficient way to monitor product quality without compromising safety, and statistical process control filled that need.
Later in the century Japanese manufacturers adopted the method, and it remains in daily use there. Today SPC is a standard quality tool across automotive, aerospace, medical device, food, pharmaceutical, and electronics manufacturing, and it is still one of the fundamental building blocks of production. The American Society for Quality maintains a useful reference overview of statistical process control and its core tools.
Every Process Varies: The Question SPC Answers
Every process you run produces variation. No machine, no operator, and no measurement system delivers the exact same result twice.
The real question is not whether your process varies, because it always will. The question is whether that variation is behaving in a predictable way, or whether something has entered the process that should not be there. Statistical process control answers that question, and it answers it while the process is still running, not after the shipment has left the building.
What Statistical Process Control Actually Is
SPC is the disciplined practice of sampling a process, summarizing each sample with a statistic, and plotting that statistic in time order against limits derived from the process itself. The pattern of the plotted points tells you whether the process is behaving consistently.
Nothing about that definition is specific to manufacturing. Any repeatable process that produces measurable output — cycle time, fill weight, error rate, turnaround time — can be monitored the same way.
The Building Blocks of an SPC System
A working SPC system is built in a fixed order, and skipping a step undermines everything after it:
- Define the measurement system — for example, deciding that a part diameter will be measured with a caliper.
- Validate that measurement system through measurement system analysis (MSA), so you know the numbers reflect the process rather than the gage.
- Define the sampling interval and subgroup scheme. This may be a customer requirement or a judgment call, such as recording the first five parts of every production batch.
- Monitor those measurements on a control chart, which is the engine of statistical process control.
The control chart is a time-ordered plot with three reference lines. The center line is the average of the plotted statistic, and above and below it sit the upper and lower control limits. Those limits are conventionally placed three standard deviations of the plotted statistic — not of the individual readings — above and below the center line.
Where Control Limits Come From
The critical point, and the one most often misunderstood, is the origin of those limits. Control limits are calculated from the process data itself. They are not targets, and they are not specification limits, which come from your customer or your designer.
Nobody chooses control limits. They are the mathematical description of what this process does when only common cause variation is acting on it. Once established from a stable baseline, they are held fixed and recalculated only when the process itself has genuinely changed — a new machine, a new material, a validated improvement.
The control chart is the voice of the process. It reports what the process is actually capable of, which is why process capability analysis is only valid once the chart shows statistical control.
Why Three Sigma, Rather Than Two or Four?
Shewhart chose three sigma as an economic balance, not as a statistical law. He weighed the cost of investigating problems that do not exist against the cost of missing problems that do — the same reasoning preserved in ISO 7870-2, the international standard for Shewhart control charts.
In a stable process whose plotted statistic is approximately normally distributed, three-sigma limits cover 99.73 percent of the plotted points. That leaves a false alarm rate of about 0.27 percent, roughly one point in every 370. That rate is low enough that when a point does fall outside the limits, investigating it is worth the effort, and nearly a century of application has confirmed the balance holds up in practice. The NIST/SEMATECH e-Handbook of Statistical Methods works through the probability arguments behind this choice in detail.
Types of Control Charts
The chart you use depends on the type of data you collect and how you group your samples.
Charts for Variable Data
Variable data is anything measured on a continuous scale, such as diameter, weight, pressure, or time.
| Chart | Use when | What is plotted |
|---|---|---|
| I-MR | Measurements arrive one at a time and rational subgrouping is not possible | Each individual value plus the moving range between consecutive values |
| X-bar and R | Small subgroups, typically 4 or 5 units | Subgroup average and subgroup range |
| X-bar and S | Larger subgroups, generally 9 or 10 units and above | Subgroup average and subgroup standard deviation |
The I-MR chart is the default for slow processes, batch chemistry, and any output where one reading is all you get. Where subgrouping is possible, X-bar and R charts are the standard choice.
The switch from R to S happens because the range uses only the largest and smallest values in a subgroup. As subgroup size grows, the standard deviation becomes the more efficient estimate of spread, which is why larger subgroups call for an X-bar and S chart instead.
Charts for Attribute Data
Attribute data classifies units as conforming or nonconforming, or counts defects on a unit.
| Chart | What it tracks | Condition |
|---|---|---|
| p chart | Proportion of nonconforming units | Subgroup size may vary |
| np chart | Number of nonconforming units | Subgroup size is constant |
| c chart | Count of defects per inspection unit | Inspection unit is constant in size |
| u chart | Defects per unit | Inspection unit varies in size |
The distinction that trips people up on attribute control charts is nonconforming units versus defects. A single unit can carry several defects, so p and np count units while c and u count occurrences. If you are unsure which chart applies, the full control chart selection logic walks through it step by step.
How to Read a Control Chart: Signals, Not Points
Whichever chart you use, reading it comes down to recognizing signals. A point outside a control limit is the most obvious signal, but it is not the only one, and it is often not the earliest one.
The Western Electric rules, and the closely related Nelson rules, define patterns that indicate a process is no longer behaving randomly even when every point sits inside the limits.
| Pattern | What it usually means |
|---|---|
| One point beyond a control limit | A special cause has acted on the process |
| Eight consecutive points on one side of the center line (nine under the Nelson rules) | The process mean has shifted |
| Six consecutive points steadily increasing or decreasing | A trend, often tool wear or gradual fouling |
| Repeating cycles | A systematic input such as shift changes or ambient conditions |
| Points hugging the center line unusually tightly | Faulty subgrouping, over-adjustment, or edited data |
Every additional rule you activate raises the overall false alarm rate, because each rule carries its own probability of firing on a stable process. Most organizations get the best results from the out-of-limit rule, the run rule, and the trend rule, leaving the rest off unless a specific process history justifies them.
An out-of-control signal is a statement that the process changed, not that the part is bad. A process can be statistically out of control while every unit still sits inside specification. Your job is to run a root cause analysis on what changed, document it, and act on the cause.
Frequently Asked Questions
Why does statistical process control (SPC) matter?
SPC gives you an objective rule for distinguishing noise from signal, and that single capability is what makes prevention possible instead of detection. It reduces scrap and rework, it shortens the time between a process change and its discovery, and it produces the objective evidence auditors ask for.
What is statistical process control in simple terms?
Statistical process control is a way of checking, with data and while production is running, whether a process is behaving the way it normally does. You measure samples over time and plot them on a control chart with limits calculated from the process. If a point breaks a limit or forms an unusual pattern, something has changed and needs investigation.
What is the difference between control limits and specification limits?
Control limits are calculated from the process data and describe what the process actually does; specification limits are set by the customer or designer and describe what the product must be. The two are unrelated in origin and must never be drawn on the same axis as if they were interchangeable. A process can sit comfortably inside specification while being statistically out of control, and the reverse is also possible.
Which control chart should I use?
Choose the chart by data type and subgroup size: I-MR for individual measurements, X-bar and R for subgroups of about four or five, and X-bar and S for subgroups of roughly ten or more. For attribute data, use p or np for nonconforming units and c or u for defect counts, depending on whether the subgroup or inspection unit stays constant. The data you already collect usually decides the chart for you.
Is statistical process control part of Six Sigma?
Yes — SPC is one of the core tools of Six Sigma, used mainly in the Measure and Control phases of DMAIC. Control charts establish the process baseline early in a project and then hold the improvement in place once changes are implemented. SPC predates Six Sigma by decades, however, and works perfectly well on its own.
Does “in control” mean the process is producing good parts?
No — in control means the process is predictable, not that it meets specification. A stable process can consistently produce out-of-tolerance parts, which is why control charts are paired with capability indices such as Cp and Cpk. Stability is the prerequisite; capability is the separate question of whether that stable output fits the tolerance.