Control Chart

Every repeated process varies: a packaging line never fills every container at precisely the same amount, and call times differ from one customer to another. A control chart helps managers avoid treating every movement as an emergency.

Plot observations in time order, add a centre line for the process reference and upper and lower limits estimated from suitable historical data, and then compare new observations with the established pattern. ASQ describes control charts as a way to distinguish consistent process variation from special-cause signals, and some charts use runs or other defined patterns.

A signal is a prompt to investigate, not proof that a particular machine or employee caused the change, so check measurement, timing, materials and operating conditions before making an adjustment. Routine variation calls for a different response: if stable but unacceptable output persists, a manager may need to improve the underlying system, because repeatedly turning a machine knob after ordinary fluctuations can make results worse.

Choose a chart suited to the measure, since individual observations, subgroup averages, counts of defects and proportions defective are not interchangeable. NIST's individuals-chart guidance estimates variation using moving ranges of successive observations, so its specific formula is not simply the mean plus or minus three times the raw standard deviation for every chart, and any shortcut needs that caveat.

A subgroup chart may track both an average and within-group spread, so sample several items under similar conditions and monitor whether either statistic changes, because poor subgroup design can hide a shift. The data must be collected consistently, as a change of instrument, measurement location or sampling interval can look like a process signal, so record those changes to keep the investigation from misleading.

Control limits come from process behaviour under a chosen chart method, whereas specification limits come from a customer's requirement or design, so a process can be stable yet consistently make products outside specifications. Likewise, a process can meet a specification in a sample while showing instability that threatens future performance, since statistical control and customer acceptability answer different questions and both should stay visible.

Start with enough relevant baseline observations: ASQ advises recalculating conditional limits once a stable series of at least 20 sequential points exists, and an initial short run is a provisional estimate, not a final truth. Do not recalculate limits after every bad point to make alarms disappear; investigate special causes first and rebase the limits only when a real sustained process change is understood and the method calls for it.

Some data are autocorrelated or seasonal, since yesterday's call volume may predict today's and Mondays may differ from Sundays, so an ordinary chart on mixed patterns can signal too often or miss real shifts. Use a sensible sampling frequency, because hourly fill-weight measurements might reveal equipment drift where a weekly average may hide it, and define who reviews signals and records evidence, since an alarm that no one investigates provides little value.

For an invented filling line, imagine a stable reference around 500 millilitres and limits calculated for the chosen chart, so a future point beyond the upper limit prompts a check of calibration and equipment. Managers can apply the method outside manufacturing, such as a clinic charting appointment delays or a service team tracking daily errors, provided the measure is defined consistently, the chart fits its distribution and the plot is shared with annotations for maintenance, staffing changes or product switches; the goal is to respond to meaningful process change while leaving routine noise alone, because a control chart is a decision aid, not a guarantee of quality or a replacement for professional judgment.