Control Chart

1. Treating a common cause of variation as if it were a special cause. This error occurs when operators adjust a stable machine or process in response to normal, random fluctuations. Such overadjustment, also known as tampering, increases overall process variability and destabilizes output.
2. Treating a special cause of variation as if it were a common cause. This error occurs when operators dismiss an unusual shift, drift, or spike as normal background variation, missing the opportunity to identify an assignable root cause and prevent recurring defects.

By establishing objective mathematical boundaries, the control chart provides clear rules for when to intervene and when to let a process run without modification. It enables operational teams to maintain stability before attempting process capability improvements.

Because three standard deviations encompass 99.73% of observations in a normal distribution, roughly 0.3% of observations fall outside three-sigma limits purely by random chance. Consequently, a single data point beyond either limit provides strong statistical evidence of an assignable cause.

To detect non-random patterns within the control limits, practitioners apply decision rules. The most common framework is the Western Electric rules, introduced in 1956. The three most widely used rules include:

• Single point beyond a control limit: An individual measurement exceeding the UCL or falling below the LCL.
• Eight consecutive points on one side of the center line: The mathematical probability of eight consecutive points falling on one side of the median by pure chance is less than 1%, signaling a persistent shift in the process average.
• Six consecutive points trending in one direction: Six points in a row continuously increasing or continuously decreasing signal a systematic drift, such as tool wear or temperature change.

When any of these conditions occurs, the chart signals that the process is out of control, prompting immediate root cause analysis.

Manufacturing

A computer numerical control (CNC) lathe turns metal shafts with a target diameter of 10.00 mm. Based on initial baseline sampling, the center line is established at 10.00 mm, with an LCL of 9.94 mm and a UCL of 10.06 mm. The operator measures samples of five consecutive shafts every hour and plots the sample average on an X-bar chart.

During normal operation, points fluctuate randomly between 9.96 mm and 10.04 mm. During a morning shift, the operator observes eight consecutive parts measuring between 10.01 mm and 10.04 mm. Although none of the parts violate the 10.06 mm upper limit or customer tolerances, the control chart signals an assignable cause under the Western Electric run rule. The operator pauses the machine and discovers that thermal expansion in the lathe spindle has shifted the baseline offset. Correcting the coolant flow returns the process average to 10.00 mm before defective parts are produced.

Healthcare

A hospital quality department monitors inpatient falls on a medical-surgical unit. The department tracks monthly falls using a u-chart normalized to patient exposure. After establishing a stable baseline of 0.8 falls per 1,000 patient days within three-sigma limits, the unit implements a nonslip flooring protocol.

Following the implementation, the team records monthly fall rates over the next year. When the fall rate drops to 0.3 falls per 1,000 patient days and remains there for eight consecutive months, the chart confirms a statistically significant improvement rather than random month-to-month variation. The hospital then calculates revised control limits based on the new 0.3 mean, locking in the improved standard.

Transactional and administrative processes

A customer support center monitors average call handle times weekly using an I-MR chart, with a stable historical baseline of 4.2 minutes and an upper limit of 5.5 minutes. The center uses this chart to evaluate the stability of customer service workflows.

Following a major software release, the average handle time spikes to 5.8 minutes, crossing the upper control limit. Because this single point exceeds the 5.5-minute limit, it signals a special cause rather than normal variation. Investigation reveals that a navigation change in the internal billing interface added four redundant clicks per transaction. The software team restores the previous interface workflow, bringing the handle time back below the upper control limit.

How do control limits differ from specification limits on a control chart?

Control limits are calculated entirely from observed process performance data and show what the system is currently delivering under normal conditions. Specification limits are determined by product designers, engineers, or customers to define acceptable tolerance thresholds for a part or service. A process can operate in complete statistical control within its control limits while still failing to satisfy external engineering specifications.

What patterns indicate a process is out of control without points exceeding the limits?

Under the Western Electric rules, an out-of-control condition occurs when eight consecutive points fall on one side of the center line, indicating a shift in the process average. A steady sequence of six consecutive points continuously increasing or decreasing also signals an assignable cause, such as tool wear or temperature drift. Both non-random patterns require immediate root cause analysis even though every measurement remains inside the control limits.

How do variable data control charts differ from attribute data control charts?

Variable data charts track continuous physical measurements such as length, weight, or duration. Attribute data charts track discrete counts, such as pass-fail outcomes, defective unit proportions, or the number of defects found per inspection unit. Choosing between variable and attribute charts depends entirely on whether data is measured on a continuous numerical scale or counted in discrete categories.

How does a control chart prevent operational tampering?

Tampering occurs when operators treat normal common-cause variation as an assignable defect and adjust a stable machine or workflow. Making adjustments in response to standard, random fluctuations increases overall process variability and destabilizes output. Control charts establish objective mathematical thresholds so operators let stable processes run and intervene only when genuine special-cause variation appears.

When should an I-MR chart be used instead of an X-bar and R chart?

An Individual and Moving Range (I-MR) chart is used when continuous measurements are collected one unit at a time rather than in subgroups. An X-bar and Range (X-bar and R) chart is used when continuous data is gathered in small, consistent rational subgroups of 2 to 9 units per sample. The I-MR chart calculates dispersion from the difference between consecutive units, whereas the X-bar and R chart evaluates dispersion using the range within each subgroup.