Individuals-Moving Range Chart: Hold the Gain After the Project Team Leaves

The Individuals-Moving Range Chart monitors individual continuous measurements for stability. You use it when you get one reading per period rather than a subgroup of several. One delivery time per day. One batch yield per run. One turnaround per shift. In the tool's own terms, it handles n=1 data, draws control limits, and flags special causes.

It belongs in the Control phase of RDMAICS. By that point you have already made the change. The question is no longer whether the change works. The question is whether the new level holds once attention moves elsewhere.

That is the honest value of a control chart. It separates the ordinary wobble of a process from a genuine shift. Without it, people react to every bad day and ignore slow drift. With it, you have a rule that says when to act and when to leave the process alone.

The chart pairs two views. The individuals chart tracks the level. The moving range chart tracks the variation between consecutive readings. Both need to be stable before you can claim the gain is locked in.

The sample data shipped with the tool follows a food delivery operation. The case is fictional, and the numbers exist only to show how the tool behaves.

The team has just finished an improvement project on delivery time. They want to confirm the new level is steady. So they record one delivery time figure per day for twenty days, alternating between the lunch shift and the dinner shift as the stratum.

The readings sit in a tight band for most of the month. Day 1 is 29. Day 5 is 27. Day 15 is 31. Then Day 20 arrives at 42. That single value is the whole story of this worked example.

Nineteen days of quiet followed by one loud day is exactly the pattern a control chart is built to interpret. The chart tells you whether 42 is bad luck or a broken process.

The input table has four columns. Observation or period carries the label, here Day 1 through Day 20. Observations in time order carries the measurement itself and is the required field. Group or stratum carries Lunch shift or Dinner shift. Series or scenario carries Delivery time.

Time order is not optional. The moving range is the difference between each reading and the one before it. If you sort the rows by value, or load them out of sequence, every moving range becomes meaningless and so does the chart built on top of it.

The group and series columns do not change the maths in this run. They give you a way to describe where each reading came from, which matters later when a signal appears and you need to ask who was working and what was running.

The tool also lets you set a baseline. If you leave it blank, limits come from all the data. That is what happened here, and the tool states it plainly under the chart.

The tool reports twenty points. The centre line sits at 29.8. The upper control limit is 35.819 and the lower control limit is 23.781. Sigma is 2.0063. Out of control count is one.

Those limits come from the data, not from a target. They describe what this process actually does when nothing unusual happens. Eighteen of the twenty readings fall comfortably inside the band. Day 20, at 42, sits well above the upper limit of 35.819.

The signals panel names the rule. Point 20, value 42, rule one, one point beyond three sigma. The tool's plain explanation is that a single reading this far out is almost never chance.

The moving range chart below the individuals chart shows the step between consecutive days. For most of the series those steps are small. The jump into Day 20 is large and shows up there too. The tool applies all eight Nelson rules to the main chart and tests the moving range chart for points outside its limits only, because moving range values are related to each other by construction. An AI review is available if you want the result explained in prose.

A handover is the moment a project is most at risk. The improvement team disbands. The daily owner inherits a process and no way to tell good noise from bad news. The chart is what you hand over.

With the limits in this example, the standing instruction is short. Plot the day's delivery time. If it falls between 23.781 and 35.819 and no pattern rule fires, do nothing. If it falls outside, or a rule fires, start an investigation the same day.

On Day 20 that instruction triggers. The investigation is not about the number. It is about what was different. The stratum column says Day 20 was a dinner shift. That is one thread to pull. Staffing, volume, weather, a system outage, a new driver. The chart does not answer the question, but it tells you the question is worth asking and roughly when the answer lives.

The other decision the chart supports is the quieter one. Nineteen days produced no signal. That is evidence the new level is holding. It is also permission to stop tinkering.

The first mistake is confusing control limits with specification limits. The numbers 23.781 and 35.819 describe behaviour. They say nothing about what a customer will accept. A process can be perfectly stable and still miss the target every day. Keep the two conversations separate.

The second is leaving a known special cause inside the limit calculation. Here the limits were calculated from all twenty points, including Day 20. That inflates sigma and widens the band, which makes future signals harder to detect. Once you understand Day 20 and fix the cause, consider setting a baseline from the earlier period and judging later data against it. The tool says explicitly that limits came from all twenty points.

The third is recalculating limits every time new data arrives. Limits that chase the data cannot detect a shift, because the shift becomes the new normal. Fix the limits during a stable period and hold them.

The fourth is reacting to every point inside the band. A reading of 27 is not better than a reading of 31 in any meaningful sense. Both are ordinary variation around 29.8. Chasing them wastes effort and often makes the process worse.

The fifth is plotting individuals data when you actually have subgroups. If you take five readings a day, an I-MR chart throws away information.

The chart detects a signal. It does not find a cause. The tool flagged point 20 at 42 and named the rule. It cannot tell you why that day was different. That work is yours, and the chart only narrows where to look.

The chart does not judge whether 29.8 is good. A centre line is a description. Whether an average of 29.8 meets a commitment is a question for the business, not for the control limits.

The limits here rest on twenty points, and one of them is a special cause. That is a thin and contaminated base. Limits firm up as you add more stable data. Treat early limits as provisional and revisit them once you have a clean period.

The chart also assumes the readings are in genuine time order and come from the same process. Mixing shifts, sites or products into one series can hide real differences or create fake ones. The stratum column exists so you can check. It does not check for you.

Finally, no chart sustains anything by itself. Somebody has to plot the point, read the rule and act. The Individuals-Moving Range Chart makes that routine small enough to survive after the project team moves on. The discipline to run it is still a human decision.