Description
Plots a metric with statistically derived control limits to separate routine (common-cause) variation from signals of special causes that warrant investigation.
Shewhart's insight was that every process varies, and reacting to every fluctuation (tampering) makes things worse. Control charts set limits at about three sigma from the centre line, estimated from the process's own short-term variation (e.g. moving ranges in XmR charts). Points outside the limits, or patterns such as eight consecutive points on one side (Western Electric or Nelson rules), signal special causes. Healthcare quality improvement adopted SPC widely. For UX metrics programmes, SPC stops teams from narrating random week-to-week variation and directs attention to genuine signals, and run charts show whether improvements produced real process change.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Goal
- Monitor over time
- Stage
- Post-launch
- Data
- Quantitative
- Effort
- Hours
Steps
- Choose a metric measured regularly (e.g. weekly task success, daily error rate) and the right chart type (XmR, p-chart, u-chart).
- Establish a baseline period of at least 20-25 points to compute centre line and control limits.
- Plot ongoing data and apply detection rules (beyond limits, runs, trends).
- Investigate signals for special causes; do not react to points within limits.
- After a deliberate process change, recalculate limits for the new phase.
- Report charts instead of week-over-week percentage changes.
Sources
- Shewhart, Economic Control of Quality of Manufactured Product (1931)
- Wheeler, Understanding Variation: The Key to Managing Chaos (2nd ed., 2000)
- Provost & Murray, The Health Care Data Guide (2011)
Also known as
- SPC
- Shewhart charts
- Process behaviour charts
- XmR charts
Categories
- Analytics & Behaviour
- Statistics & Modelling