Choosing a run chart or control chart
Use a run chart to examine time-order patterns around a median; use a suitable control chart to compare observations with expected process variation.
Fit
When it matters, and when it does not
It matters when
- You measure a process repeatedly in time order.
- You need to separate a possible process change from ordinary variation.
It matters less when
- You only have one before and one after value. Consider a design with repeated observations. Read: Interrupted time series
- You need to estimate an intervention effect while allowing for trend. Consider segmented regression. Read: Segmented regression for interrupted time series
Before you start
What to check in your study
- The measure and its denominatorDistinguish continuous values, proportions and event rates. Chart choice depends on that distinction.
- A baseline periodDefine a period that can reasonably represent a stable process.
- Signal rules and planned responsesChoose rules before inspecting changes. Record interventions and data changes in time order.
- The definition and collection of the measure remain comparable.How to check: Check case mix, denominators and missing data before interpreting a signal.
- A control chart model fits the process.How to check: Check dependence, extra spread and baseline stability. Serial correlation can cause false signals.
Plain language
How it works
A run chart joins observations in time order and uses a median with stated rules for runs or crossings. It does not have calculated control limits. Rule sets differ in false-signal rates and sensitivity.
A control chart adds a center line and limits from a specified process model. Use a chart for individual continuous values, proportions, counts or rates as appropriate. For a p-chart, limits change when the denominator changes.
Control limits describe expected process behavior. They are neither confidence intervals for an average nor acceptable clinical thresholds. Both charts support monitoring; neither alone separates an intervention from other changes happening at the same time.
Worked example
A worked example
- Setting
- A clinic monitors a binary delay outcome each week.
- Question
- Are control limits the same as a performance target?
- Approach
- A stable baseline has 10 delays per 100 visits. Each future week also has 100 independent visits.
- Result
- The baseline is 10 in 100. A simple p-chart has approximate three-standard-deviation limits of 1 and 19 in 100.
- What it means
- A clinical target can differ from both limits. These illustrative limits assume binomial variation; they do not measure uncertainty in the baseline estimate.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Control limits describe process variation, not clinical targets or confidence intervals. A chart signal alone does not establish what caused a change.
- Calling a point outside a limit proof of improvement.Instead: Investigate the signal, its direction and possible data changes. The chart does not identify the cause.
- Moving the center line after every unusual point.Instead: Use a documented rule for establishing a new stable phase. Do not redraw limits merely to erase signals.
Inside Mallard
How Mallard uses this
Mallard lists Choosing a run chart or control chart under Core research concepts.
concept.choosing_run_vs_control, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Run charts revisited: a simulation study of run chart rules for detection of non-random variation in health care processes (opens in a new tab) · PLoS One · doi 10.1371/journal.pone.0113825
- Large-scale empirical optimisation of statistical control charts to detect clinically relevant increases in surgical site infection rates (opens in a new tab) · BMJ Qual Saf · doi 10.1136/bmjqs-2018-008976
- The run chart: a simple analytical tool for learning from variation in healthcare processes (opens in a new tab) · BMJ Qual Saf · doi 10.1136/bmjqs.2009.037895