Statistical Process Control: Using Control Charts in Practice

Every process produces variation. Cycle times differ from one unit to the next. Dimensions fall at slightly different points along a scale. Defect counts shift from week to week. The question is not whether variation exists — it always does — but whether the variation you are seeing is the normal background noise of the process or a signal that something has changed. Statistical Process Control (SPC) provides a rigorous framework for answering that question. Developed by Walter Shewhart and later refined by W. Edwards Deming, SPC monitors process output over time and distinguishes variation that is inherent in the process from variation caused by something specific and identifiable.

The control chart: anatomy and logic

The control chart — also called a Shewhart chart — plots process measurements over time against three reference lines: the centre line (the process average), the upper control limit (UCL), and the lower control limit (LCL). Control limits are set at three standard deviations above and below the mean, calculated from the process data itself. This is a critical distinction: control limits are not specification limits. Specification limits are set by the customer or design — they define what is acceptable. Control limits are set by the process — they define what the process naturally produces. A process can be in statistical control and still produce output outside specification. Those are separate questions.

Common cause versus special cause variation

The fundamental distinction in SPC is between two types of variation. Common cause variation — also called random or inherent variation — is the result of the many small factors always present in the process: minor equipment wear, ambient temperature, small differences in raw material lots, operator technique within normal range. It is stable, predictable, and the result of the system itself. Removing common cause variation requires changing the process design. Special cause variation arises from something specific: a tool that has broken, a new operator who has not been trained, a batch of raw material from a different supplier. It is unpredictable and signals that something outside the normal process has occurred. Special cause variation requires investigation and correction at the source. Treating common cause variation as if it were special cause — adjusting a process every time it produces a result away from the average — actually increases variation. Deming called this "tampering."

The Western Electric rules

A point outside the control limits is the most obvious signal of special cause variation, but patterns within the limits are equally diagnostic. The Western Electric rules define eight tests for non-random patterns: a single point beyond three sigma; two of three consecutive points beyond two sigma on the same side; four of five beyond one sigma on the same side; eight consecutive points on the same side of the centre line; six points trending in one direction; fourteen alternating up and down; fifteen within one sigma of the centre line (a signal of artificially reduced variation); and eight consecutive points on either side with none within one sigma. Most SPC software applies these tests automatically. Understanding what each signals helps direct the investigation correctly — a single outlier calls for a different response than a sustained shift or a trend.

Selecting the right control chart

1. XBar-R and XBar-S charts are used for continuous data in subgroups. XBar-R suits small subgroup sizes (two to eight); XBar-S is preferred for nine or more, where the range understates spread.

2. I-MR (Individuals and Moving Range) charts are used when data arrive one point at a time — common in process industries and administrative processes where natural subgrouping does not exist.

3. p-charts and np-charts handle attribute data: the proportion of nonconforming items (p-chart, variable sample size) or the count of nonconforming items (np-chart, constant sample size).

4. c-charts and u-charts count defects (nonconformities) rather than defective units. Use c-charts when the inspection area is constant; u-charts when it varies.

Implementing SPC without a dedicated statistician

The most common barrier to SPC adoption is the belief that it requires statistical expertise to implement. It does not — at least not at the outset. Start by identifying one or two key process outputs that matter: a cycle time, a defect rate, a dimensional characteristic. Collect data consistently over time, using consistent measurement methods. Most SPC software handles the statistical calculations automatically; your role is to understand what the charts are telling you and respond appropriately when a signal appears. The discipline is in the response protocol: when the chart signals, investigate immediately, identify the cause, correct it if it is adverse, replicate it if it is beneficial, and document what you found. A control chart that generates signals nobody investigates is not SPC — it is chart-making. The value is in the disciplined response to what the chart tells you.

If your organisation is building a quality management system or integrating SPC into an existing Lean Six Sigma programme, XNM's strategic advisory practice can help you design a measurement and monitoring approach that is rigorous without being burdensome to operate.