Learning from variation: run charts, SPC and interpretation

• Explain why time order matters in improvement measurement.
• Construct and interpret a basic run chart with a stable operational definition.
• Distinguish common-cause from special-cause variation conceptually.
• Explain the difference between targets/specification limits and statistically derived control limits.
• Avoid overreacting to single points, short baselines and misleading before/after comparisons.

A pre/post percentage collapses the story. Two averages can look different because of random variation, changing case mix, seasonal demand or a temporary event. Plotting data in time order allows teams to see level, direction, stability and the relationship between changes and performance. Annotate interventions and contextual events.

A run chart plots a measure over time, usually with a centreline such as the median calculated from a suitable baseline or stable period. The chart helps teams look for non-random patterns such as sustained shifts or trends. Exact run-chart rules vary by method; teams should use one recognised rule set consistently rather than inventing patterns after seeing the data.

Common-cause variation is the expected variation produced by the current system. Special-cause variation suggests that something unusual has affected the process. The management response differs: reacting to every common-cause fluctuation can create instability, while ignoring a special cause can miss an important learning opportunity.

Statistical process control charts add a centreline and statistically derived control limits appropriate to the type of data. The limits describe the expected range of common-cause variation if the process is stable. They are not performance targets. Choice of chart depends on whether data are counts, proportions, rates or continuous measurements and on denominator behaviour; specialist support is sensible when the choice is uncertain.

A target states desired performance; a control limit is calculated from process variation. A process can be stable but consistently fail its target, meaning the system itself needs redesign. A process can also meet the target on average but be unstable, meaning unpredictable special causes remain.

Do not keep recalculating the centreline simply because recent results look better. Establish a baseline using a defensible period, then apply recognised rules for when a sustained change justifies a new centreline. Very short baselines create false confidence. If the process is highly unstable, investigate special causes before treating the average as “normal”.

Weekly timely-summary completion is 62, 65, 61, 64, 63, 62% for six weeks. After a small workflow test, performance becomes 74, 78, 80, 79, 82, 81%. A time plot shows a sustained change far more clearly than comparing one baseline week with one post-intervention week. The team should still examine context, denominator stability and balancing measures before attributing causation.

Avoid truncated axes that exaggerate small changes, mixed denominators, cherry-picked start dates, cumulative charts that hide current performance, and averages that combine very different sites or populations. A chart should make the process easier to understand, not merely make improvement look impressive.

First confirm the measure definition and denominator are stable. Second inspect the time series and annotations. Third ask whether a recognised non-random pattern is present. Fourth explore contextual explanations and possible special causes. Fifth decide whether to test, adapt, investigate or implement. The chart supports judgement; it does not replace it.

• IHI - Establishing measures
• NHS England - Quality management systems
• HQIP - Guide to QI methods