Measurement & Chart Guide
Start from the question you are trying to answer.
The right chart is the one that fits the question and the data — not the fanciest one. This guide maps common operational questions to the chart that usually fits, shows what each one looks like, and explains how to read genuine signal from normal variation.
Completed transfers / week
Signal: shift after changeWhy this matters to leaders
The biggest measurement mistake in operations is reacting to every up-and-down as if it meant something. Control charts separate the normal noise of a stable process from a real signal that deserves action — so you stop chasing ghosts and start managing change.
Which chart should I use?
Start with the kind of question you are asking. This maps the question to the chart family that usually fits.
| If you are examining… | Chart that usually fits |
|---|---|
| Individual observations over time, or one calculated result per period | I-chart / XmR |
| Early learning with limited data | Run chart |
| A percent where the numerator is a subset of the eligible denominator (each case yes/no) | P-chart |
| Repeatable events per exposure or opportunity (a rate) | U-chart |
| Event counts where the opportunity is reasonably stable | C-chart |
| Multiple observations deliberately grouped into subgroups (average + variation) | X-bar & S chart |
| Which categories account for the most | Pareto chart |
The charts, and what they show
Each chart answers a different question. Here is what to reach for, with healthcare examples and the trap to avoid.
Run chart
Over timeBegin here when you are starting data collection or testing a change. The centerline is the median and no control limits are drawn.
Examples
- · Weekly transfer request volume
- · Time to first dose during a small pilot
Watch for: Move to a control chart once you have enough stable data to set limits.
I-chart / XmR
Over timeDaily ED boarding hours
Signal: shiftValues over time, either way: individual observations plotted in order, or one calculated result (average, median, total, percentage, or other statistic) for each period. A summary is one calculated result per period — it does not have to be a sum.
Examples
- · Daily ED boarding hours
- · Each patient's length of stay plotted in order
- · Monthly median LOS — one calculated result per month
Watch for: Not ideal for comparing unordered categories (providers, units) — the moving range depends on the order of the data.
P-chart
ProportionsProportions or percentages where the numerator is a subset of the eligible denominator and each case is yes/no (met / not met). The denominator may change between periods.
Examples
- · ED admission rate by provider
- · Transfer acceptance rate by referring hospital
Watch for: Very small denominators make proportions unstable; very large ones make limits too tight to be useful.
U-chart
Event ratesRepeatable events counted against a unit of exposure or opportunity (a rate). A case can have more than one event, and the exposure may change between periods.
Examples
- · Falls per 1,000 patient days
- · Defects per 100 transfer requests
Watch for: If the opportunity is stable, a C-chart may be simpler. If each case is a yes/no subset of the eligible group, use a P-chart instead.
C-chart
Defect countsEvent or defect counts when the opportunity is reasonably stable from period to period.
Examples
- · Processing errors when each person handles a similar caseload
Watch for: If the opportunity changes, changing volume alone does not decide the chart: collect a denominator and use a P-chart (eligible yes/no cases) or a U-chart (repeatable events per exposure).
X-bar & S chart
ThroughputWhen you deliberately organize multiple observations into rational subgroups and want to monitor both the subgroup average and the within-subgroup variation. An average by itself does not require X-bar/S — patient-level data can still be plotted on an I-chart.
Examples
- · Arrival-to-provider time grouped into daily subgroups
- · Order-to-discharge time grouped by unit
Watch for: Plotting individual observations, or one calculated result per period, is an I-chart. Reach for X-bar/S only to monitor the variation within each subgroup.
Pareto chart
CategoriesRanking categories by contribution to find the few that account for most of the volume, delay, or impact.
Examples
- · Denial reasons by frequency
- · Causes of transfer delay
Watch for: Pareto compares categories at a point in time — pair it with a time-series chart to see whether they are changing.
Reading variation: signal vs. noise
Every process varies. The question is whether a point is part of the normal, expected variation of a stable process (common cause) or a genuine signal that something changed (special cause).
- Don't overreact to noise.A single high or low point inside the control limits usually means nothing. Reacting to it — “why was last week bad?” — wastes energy and often makes things worse.
- Do act on a signal. A point outside the limits, or a sustained run on one side of the centerline, is evidence that the process has shifted. That is when you investigate or — if you made the change — confirm it worked.
- Annotate what changed. Mark interventions on the chart so a shift can be connected to the action that caused it.
Use charts in operating reviews, not as dashboard theater
A wall of green-and-red tiles tells you nothing about whether you are improving. A few well-chosen charts, read for signal, turn an operating review into a learning conversation.
Deep dives for each chart type are on the way. This guide is grounded in improvement science and analytic guidance from The Health Care Data Guide: Learning from Data for Improvement(Provost & Murray), adapted for patient flow and hospital operations.