🎯 Quick Answer
What: A control chart shows a metric in time order, with a center line and two control limits. The control limits are calculated from the process's own data.
How to read a control chart: Look for three signals. One point outside the control limits. Eight points in a row on one side of the center line. Six points in a row going up, or going down.
No signal: The ups and downs are normal variation. The process has not changed, so do not adjust it.
The payoff: A slow change shows up in weeks, not months, while it is still small and easy to fix.
Last week's article named one process owner and put the metric on a one-page control plan. Now the process owner posts that metric every week.
The next question is how to read it. A metric on a board is only useful if the person looking at it can tell when something has changed.
That is the job of a control chart. In DMAIC, the step-by-step method used in Lean Six Sigma, it is one of the main tools of Control, the last phase. Here is how to read a control chart, step by step. It starts with a run chart. Then it adds control limits and the three signals. It ends with an example and four common problems.
Why a monthly number is not enough
Most reports show a metric once a month. One number a month cannot show a slow change.
Here is an example. A plastics plant tracked the share of molded parts it rejected. Each month, the number was a little higher than the month before. But it was only a little, so nobody saw a problem.
When the process owner plotted the same data by week, the change was clear. The reject rate had gone up six weeks in a row. One mold was slowly wearing. After the worn part was replaced, the reject rate came back down.
The data had been there the whole time. It only needed to be plotted in time order, with enough points to show the pattern.
Start with a run chart
A run chart is the simplest way to watch a metric over time.
Choose the metric. Use the one on the control plan, with its operational definition from the project's data collection plan. If people measure it by hand, check the measurement system first.
Plot it in time order. One point for each shift, day or week. Time goes left to right.
Use at least 20 points. Fewer points cannot show a pattern clearly.
Draw a center line. On a run chart, this is usually the median: half the points are above it and half are below.
A run chart already shows a lot. You can see a climb, a drop or a sudden jump. What it cannot tell you is how far a point can move before it is unusual. For that, you need control limits.
From a run chart to a control chart
A control chart is a run chart with two more lines.
The center line is the average of the points.
The upper control limit and the lower control limit are drawn above and below the center line. They are calculated from the process's own data, about three standard deviations from the center line.
The control limits show how much the metric moves up and down when nothing has changed. This is called normal variation. Most points stay between the limits.
Calculate the limits from at least 20 points collected when the process was running as it should. After an improvement project, use data from after the improvement, not from before it, because the process has changed.
How to read a control chart: the three signals
A process always moves up and down a little. The question is whether it has changed. Look for three signals.
1. One point outside the control limits. When nothing has changed, a point outside the limits is very rare. Something happened that week, shift or day. Look for the cause while the details are still easy to find.
2. Eight points in a row on one side of the center line. Every point can be inside the limits and this is still a signal. It means the process has shifted to a new level. Some courses and some software use seven or nine in a row instead of eight. Pick one rule and use it every time.
3. Six points in a row going up, or going down. Each point is higher than the one before, or lower. This is a trend. Something is changing a little at a time. For example, a part may be slowly wearing.
If you see none of these, the process has not changed. The ups and downs are normal variation. Do not adjust the process because one week was higher than the last. Adjusting a process that has not changed usually makes it vary more.
An example: the steel wire plant
A plant that makes steel wire tracked the wire's diameter on a control chart. Each shift, an operator measured a sample of wire and added the average to the chart. The process owner was the line supervisor.
For months, every point stayed inside the control limits. Then most new points were just above the center line. They were not far above it, and they were well inside the limits.
At the Monday meeting, someone said the chart looked fine, because nothing was outside the limits.
The line supervisor counted the points. Eight in a row were above the center line. That is the second signal: the process had shifted.
The team checked what had changed. The plant had started using drawing dies from a new supplier, and the new dies made the wire slightly thicker. The team worked with the supplier to correct the die size, and the points moved back around the center line.
No wire had gone out of specification. The shift was found while it was still small.
Control limits are not specification limits
This is the most common confusion about control charts.
Control limits come from the process. They show what the process normally does.
Specification limits come from the customer or the design. They show what the product must be.
A process can be inside its control limits and still make product outside the specification. That process is stable, but it is not capable. A process can also show a signal while every part is still inside the specification, as the wire plant did.
Do not draw the specification limits on a control chart in place of the control limits. They answer different questions. Comparing the process to the specification is the job of a process capability study.
Four common problems
Too few points. Five or six points cannot show a pattern. Collect at least 20 before you calculate control limits.
New limits every month. If the limits are recalculated every month, a slow change becomes part of the new limits and is never seen. Keep the limits fixed. Recalculate them only after a deliberate change to the process.
Adjusting the process for every point. A point that is higher than last week but inside the limits is normal variation. Adjusting the process for it adds variation. Act on signals only.
Limits from before the improvement. A project changes the process. Limits calculated from the project's baseline describe the old process. Calculate new limits from data collected after the improvement.
Drawing the chart
You do not need special software to start. A run chart can be drawn by hand on graph paper or on a board next to the line. Excel can plot the points and draw the center line and the limits.
Statistical software such as Minitab, JMP, SigmaXL or SPC XL works out the control limits for you. It also marks the signals on the chart. The method is the same in all of them. If you know how to read a control chart, you can read one from any of these tools.
One note: a control chart is for watching a process that is running. Whether a before-and-after difference from a project is real is a different question, with its own test. This article covers that question.
When you see a signal
When a signal shows, the process owner looks for the cause. Start with what changed: a new supplier, a new setting, a new person, a different shift or a different material.
The control plan's reaction plan lists the first things to check, and who to call. A later article covers the reaction plan in full.
What leaders can do
If you review metrics every month, ask to see them as control charts.
Ask for the chart, not this month's number. The last 20 points or more, in time order, with the center line and the control limits.
Ask one question first: is there a signal? If there is, ask what the process owner found.
If there is no signal, do not ask for an explanation. The ups and downs are normal variation. Go to the next metric.
Keep the limits fixed unless the process was changed on purpose.
I write one of these every week, based on what actually happens on projects rather than what the textbook says should happen. You can join the newsletter here.
FAQ
How do you read a control chart?
Plot the metric in time order with a center line and control limits. Then look for three signals. One point outside the control limits. Eight points in a row on one side of the center line. Six points in a row going up or going down. If none of these appear, the variation is normal and the process has not changed.
What is the difference between a run chart and a control chart?
A run chart shows a metric in time order with a center line, usually the median. A control chart adds an upper and a lower control limit, calculated from the process's own data. The control limits show how far the metric normally moves, so you can tell a signal from normal variation.
How are control limits calculated?
Control limits are calculated from the process's own data, about three standard deviations above and below the center line. Use at least 20 points collected while the process was running as it should. Statistical software such as Minitab, JMP, SigmaXL or SPC XL does the calculation, and Excel can too.
What is the difference between control limits and specification limits?
Control limits come from the process and show what it normally does. Specification limits come from the customer or the design and show what the product must be. A process can be inside its control limits and still make product outside the specification.
How many points do you need for a control chart?
At least 20 points to calculate the control limits. Once the limits are set, add a new point every shift, day or week, and keep the limits fixed unless the process is changed on purpose.
Plot the metric every week. Then look for the three signals.
