X̄-R Chart: How to Hold the Gain After the Project Team Leaves

The X̄-R Chart monitors the mean and range of subgrouped continuous data with subgroup sizes between 2 and 10. It detects shifts in the process centre and shifts in process variation at the same time. The mean chart answers one question. Has the process moved. The range chart answers another. Has the process become more or less consistent.

You reach for it in the Control phase of RDMAICS. By that point you have made a change and you believe it worked. The risk is no longer whether the improvement works. The risk is whether it lasts once the project team moves on and attention moves elsewhere.

A control chart is the handover document that keeps working. It gives the process owner a running answer to a simple question. Is the process still behaving the way it behaved when we signed off. That is the whole purpose of the tool in this phase.

Use it when your measure is continuous, when you can collect several readings close together in time, and when you want to separate the ordinary noise of the process from a genuine change.

Everything below is built on the sample data that loads with the tool. The case is fictional. It is there so you can see how the inputs, the subgrouping and the result fit together before you load your own numbers.

The scenario is order fulfilment time at a food service operation. One measurement is recorded per day for 25 days, labelled Day 1 through Day 25. Each day is tagged as either Lunch shift or Dinner shift, and the whole series is labelled Five-order subgroups.

The values sit in a narrow band for most of the series. Day 1 is 29. Day 4 is 27. Day 11 is 31. Day 24 is 28. Then Day 25 is 41, far above anything that came before it.

That single large value is the interesting part of the case. It is exactly the kind of event a process owner will face after handover. The chart exists to tell you what to do about it, and in this instance the answer is not what you might expect.

The tool takes four things. An observation or period label, the measurement, an optional group or stratum, and an optional series or scenario. The measurements column carries a note worth reading twice. Measurements are in time order, and they are grouped in the order supplied.

That note matters. You do not tell the tool which reading belongs in which subgroup. You set the subgroup size and the tool reads down the list in order. In the sample data the subgroup size is 5. So Days 1 to 5 form the first subgroup, Days 6 to 10 form the second, and so on through the 25 rows.

Twenty-five measurements at a subgroup size of 5 gives you 5 subgroups. The result panel confirms this with POINTS 5. Each plotted point is a subgroup, not a day. If you change the subgroup size you change the number of points and you change the limits.

The Group or stratum column holds Lunch shift and Dinner shift alternately. It is useful context to carry alongside the numbers. It does not drive the subgrouping. The order of the rows does that.

The result panel reports a centre of 29.56 for the subgroup mean chart, an upper limit of 32.445 and a lower limit of 26.675. Sigma is 2.1496. Out of control count is 0.

Read the centre first. At 29.56 it is the average of all the measurements in the period. Then read the distance to the limits. The band runs roughly from 26.675 to 32.445. That band is not a target or a specification. It is what this process does when nothing unusual is happening.

The subgroup mean plot is scaled from 28.5 to 32.3, and the subgroup range plot from 0 to 16, with its own upper and lower control lines. The range chart is the one people skip. Do not skip it. A stable mean with a widening range still means the process has changed.

Now the surprise. Day 25 is 41, yet the out of control count is 0 and the signals note says no rule was triggered. The reason is subgrouping. Day 25 sits inside the fifth subgroup with four other readings, and its effect on that subgroup mean is diluted. Averaging is what makes the mean chart sensitive to sustained shifts and insensitive to a single spike.

The signals note adds a line every process owner should memorise. The process is in statistical control over this period, which means it is predictable, not that it is good enough.

The result panel notes that limits were calculated from all 5 points, and that you can set a baseline above to judge later data against an earlier period. This is the feature that turns the chart into a control mechanism rather than a description.

During handover you freeze a baseline. You take the period that represents the improved process and you lock its centre and limits. From then on, new data is judged against that frozen reference. If the process drifts back towards the old behaviour, the new points will fall outside limits that were set when the process was good.

Without a baseline, the limits move as you add data. A slow drift quietly recalculates the limits and the chart keeps reporting control while the process gets worse. That is the single most common way a control chart fails to control anything.

So the sequence is simple. Collect the post improvement period. Confirm it is in control. Freeze it as the baseline. Hand the chart to the process owner with a written rule for what to do when a point falls outside.

A control chart is a decision tool, not a report. It tells you when to act and, just as importantly, when to leave the process alone. The second use saves more time than the first.

With the sample data in front of you, the decision is to do nothing to the process. No rule was triggered. Adjusting a process that is in control adds variation. You would make the output worse while believing you were improving it.

The Day 25 value of 41 still deserves attention, but as an investigation rather than a process adjustment. You would ask what happened that day, check whether the reading was recorded correctly, and note the finding. If similar values start appearing, subsequent subgroup means will move and the chart will say so.

The tool also offers an AI review that explains the result in plain language. Use it as a second read on what the chart is showing, not as a replacement for the process owner's judgement about the work itself.

Most X̄-R charts fail for setup reasons, not statistical ones. The errors below show up again and again in Control phase handovers.

Each one is easy to avoid if you check it before you freeze the baseline.

The X̄-R Chart will not tell you whether the process is acceptable. In the sample data the process is in control, and the tool says so plainly while also saying that predictable is not the same as good enough. Capability against a customer requirement is a separate question and needs a separate tool.

It will not tell you why something happened. The chart flags that behaviour has changed. The explanation comes from the people who run the process. A chart with no investigation routine behind it is just a picture.

It will not catch everything. With a subgroup size of 5 and only 5 points, the chart is built to detect sustained shifts in the centre and in the spread. A one off spike like Day 25 can pass without a signal. If single extreme values are what you care about, a chart built for individual readings is the better fit.

It will not maintain itself. Someone has to keep entering data, keep the time order intact, and act when a rule fires. The chart holds the gain only while a named owner keeps using it. That commitment, not the statistics, is what makes the Control phase work.