Run Chart: How to Make and Read One With Examples

By Linh Ngo · Updated Oct 10, 2026

A run chart is a line graph of one measure plotted in time order, with a median line through the middle. You read it with four simple rules (shift, trend, runs, astronomical point) that separate real change from ordinary ups and downs. It takes ten minutes to build, needs no statistics software, and tells you whether the change you made actually worked.

What Is a Run Chart?

A run chart plots a measure over time. Time goes on the horizontal axis in the order the data were collected. The measure goes on the vertical axis: a percentage, a count, minutes, dollars. The dots are connected with a line, and a median line runs through the middle as the center line.

That's the whole tool. And it's more useful than it looks, because most reports hide the one thing a team needs to see: what happened in what order.

Imagine two units that both report the same average cycle time after a process change. One unit was already improving before the change. The other improved for three weeks and then slid back. A bar chart of averages shows them as identical. A run chart shows them as completely different stories. Perla, Provost and Murray make this point in their paper on run charts as an analytical tool: data viewed over time give richer and more accurate conclusions than summary statistics that ignore time order.

Run charts sit inside the family of seven basic quality tools and feed straight into statistical process control. Think of them as the first, lightest step on that ladder.

Key Facts

• A run chart plots one measure in time order with the median as the center line, according to the IHI Open School run chart lesson.
• Perla, Provost and Murray's BMJ Quality & Safety paper sets out three probability-based rules (shift, trend, runs) at a 5% error level, plus a fourth, subjective rule for astronomical points.
• A shift is six or more consecutive points on one side of the median. A trend is five or more consecutive points all going up or all going down.
• The shift and run rules need more than 10 data points before they apply.
• A run chart can't tell you a process is stable. Only a control chart can.

Why Teams Use Run Charts

Three reasons keep showing up in practice.

They stop overreaction. People tend to over- or under-react to the latest data point and start tampering with a process that hasn't changed. The paper's authors describe exactly this habit in dashboards and board reports. A run chart gives the team an agreed way to say "that's noise" or "that's a signal."

They test changes. If you changed something on a Monday, the chart should show the data shift after Monday. If it doesn't, the change probably didn't do what you hoped.

They need almost nothing. No control limits, no standard deviations, no constants from a table. A spreadsheet and a median function will do. That makes them the right tool for teams that haven't used charts before, and for projects where you only have a few weeks of data.

Run Chart vs Control Chart

Both plot data over time. They answer different questions.

The BMJ paper is direct about the limit: run charts are built for early detection of improvement or decline, but they can't determine whether a process is stable in Shewhart's sense. Its authors even recommend avoiding the terms "special cause," "common cause," "stable" and "unstable" on a run chart, and saving them for control charts. Their example is a run chart showing no signals at all, where the same data on a control chart reveal special causes.

So the practical rule is simple. Start with a run chart. When you're ready to manage a process over the long haul, or when a run chart hints at something you need to confirm, move to a control chart.

The Four Run Chart Rules

These rules are what turn a squiggly line into evidence. Without agreed rules, words like "shift" and "trend" get used loosely to push a conversation along.

Rule 1: Shift

Six or more consecutive points all above or all below the median. Points that fall exactly on the median neither add to nor break a shift. You skip them and keep counting.

A shift means the process has settled at a new level. It's the signal you hope to see after an improvement.

Rule 2: Trend

Five or more consecutive points all going up or all going down. If two or more consecutive points have the same value, count only the first and ignore the repeats. Like values neither make nor break a trend.

A trend means the process is drifting steadily in one direction. Two points going up is not a trend, and neither is four.

Rule 3: Runs (too many or too few)

A run is a series of consecutive points on one side of the median. Count runs by counting how many times the connecting line crosses the median, then adding one.

If chance alone drives the data, points should cross the median with a certain regularity. Too few runs means the data are hanging on one side too long. Too many means they're zigzagging more than chance would produce, which can point to something systematic like a repeating cycle.

To judge, count the points that do not fall on the median and compare your run count to this table, which the BMJ paper bases on about a 5% risk of failing the run test for a random pattern:

The paper's full table covers every count from 10 to 60 points. If you have an odd count, use the row for that number from the original table.

Rule 4: Astronomical point

A point that's obviously, even blatantly, different from the rest, where everyone looking at the chart would agree it's unusual. This is the one subjective rule. The IHI lesson calls it an eyeball test, not a statistical determination on a run chart.

Don't confuse it with the highest or lowest point. Every run chart has those. An astronomical point is a value that makes you say "what happened there?" Check for a data error or a special event. If you want a statistical answer, put the data on a control chart.

How many points do you need?

The shift and run rules need more than 10 points before they apply, according to the BMJ paper. With very little data the median itself may not be useful. The authors note a run chart can still be handy with only three or four points (a patient tracking an annual test, say) as an early look at direction, just without the formal rules.

How to Make a Run Chart: Step by Step

Step 1: Pick one measure

Choose something that matters and can be counted the same way every time. "Invoice approval time in days" works. "Finance efficiency" doesn't. Write down exactly how it's measured, including what's counted and what's excluded, so next month's number means the same thing.

Step 2: Choose the time unit

Match the interval to how often the process runs and how fast you want feedback. Daily for a busy service desk, weekly for invoice approvals, per-batch for production. Keep the interval fixed.

Step 3: Collect at least 12 data points

Aim for more than 10 so the rules can apply. If the process is slow, say monthly reports, you may need to start with fewer and use the chart for direction only.

Step 4: Plot the points in time order

Time on the horizontal axis, earliest on the left. Put the measure on the vertical axis, label both, and connect the dots.

Step 5: Calculate and draw the median

Sort the values and take the middle one. With an even count, average the two middle values. Draw it as a horizontal line across the whole chart. Many teams use a spreadsheet function for this, but it's worth doing by hand once so you know what it is.

Step 6: Set a baseline and freeze the median

The BMJ paper recommends building the initial median from baseline data. If the baseline shows no signals, extend (freeze) that median into the future so new data can't pull the line toward themselves. Changes then stand out clearly against the baseline.

Step 7: Apply the four rules

Check shifts, trends, runs and astronomical points in that order. Mark any signals directly on the chart. Annotate the dates of any process changes so readers can connect cause and effect.

Step 8: Decide what to do

A signal after a change is evidence the change worked (or failed). A signal with no known cause is a prompt to investigate. No signal means don't react. Then keep plotting. Run charts earn their keep through repetition inside a PDCA cycle, where each test of change gets its own annotation.

A Worked Example: Invoice Approval Time

This example is illustrative. A finance team tracks the average days to approve a vendor invoice, one point per week.

Weeks 1 to 12 (baseline): 6.1, 5.8, 6.4, 5.9, 6.6, 6.0, 5.7, 6.3, 6.2, 5.9, 6.5, 6.0

The baseline median is 6.05 days. The line crosses the median 9 times, giving 10 runs across 12 points. The table says too few is under 3 and too many is over 11, so 10 runs is fine. No signals. The process is bouncing around 6 days by chance, which is exactly the situation where you freeze the median and test a change.

In week 13, the team introduces a rule: invoices under a set value are auto-approved if the purchase order matches.

Weeks 13 to 24: 5.6, 5.4, 5.7, 5.3, 5.5, 5.2, 5.4, 5.1, 5.3, 5.0, 5.2, 5.1

Every one of the 12 new points sits below the frozen median of 6.05. Counting back, week 12 (6.0) is also below it, so there are 13 consecutive points under the median. Rule 1 asks for six. That's a clear shift.

Notice what the team can and can't say. They can say the process settled at a lower level after the change. They can't say it's "stable" in the control chart sense, and they shouldn't claim savings beyond what the days show. If they want to monitor the new level indefinitely, the next step is a control chart built on the post-change data.

Run Chart Examples by Function

The same chart works anywhere a number comes out of a repeating process. These are illustrative scenarios.

In each case the chart answers the same two questions: did the data move, and did it move after the change?

Run Chart Template

Copy this table into a spreadsheet and fill it in. Add a line chart of the Value column, then add the median as a second, flat series.

And a small checklist to keep beside it:

Do's and Don'ts

Do

• Plot data in the order it happened, with a clear time axis.
• Write down what you're measuring and how.
• Freeze the baseline median before testing a change.
• Annotate changes, holidays and unusual events right on the chart.
• Use the same rules every time so everyone reads the chart the same way.
• Move to a control chart when you need to prove stability.

Don't

• React to one point. A single high or low value is not a signal.
• Use "stable" or "special cause" language on a run chart.
• Count points on the median toward a shift or a run.
• Treat the highest or lowest point as an astronomical point by default.
• Apply the median rules when half or more of your points sit at an extreme value like 0% or 100%. The BMJ paper notes the median becomes the extreme value itself, which makes the rules useless.
• Change the median every week. It defeats the purpose of a baseline.

Where Run Charts Fit With Other Quality Tools

A run chart rarely works alone.

• A check sheet is often where the raw counts come from before you plot them over time.
• A Pareto analysis tells you which problem to chart first.
• A histogram shows the shape of the same data but throws away time order, which is exactly what the run chart keeps.
• A scatter diagram helps when you suspect another variable is driving a shift you've spotted.

Where to Go from Here

Pick one measure you already report every week. Pull the last twelve values, find the median, and draw the line. If nothing else, you'll see for the first time whether that number has actually moved or whether your team has been reacting to noise.

When a run chart starts giving you answers you want to rely on, the natural next step is a control chart inside a wider statistical process control practice.