An individuals and moving range (I-MR) chart is the control chart you use when you can only take one measurement at a time, instead of subgroups. The individuals chart tracks each reading; the moving range chart tracks the difference between consecutive readings, which stands in for the process spread.
Most control-chart theory assumes you collect small subgroups, say five parts every hour, and plot their average and range. Plenty of real processes do not cooperate. A batch reactor produces one homogeneous result per batch. A destructive test consumes the part it measures. A slow, expensive, or low-volume operation gives you one data point per shift. In all of these, rational subgroups do not exist, and the I-MR chart is the tool built for exactly that case.
This guide covers when to reach for I-MR instead of an X-bar and R chart, how to compute the limits with the moving range, the constants that make it work, and how to read the pair without fooling yourself. It sits alongside the broader guides to control charts and statistical process control.
When should you use an I-MR chart?
Use an I-MR chart when rational subgrouping is impossible or meaningless, so you are left with a stream of single values over time. The common triggers:
- Low volume or long cycles. You make a few parts a day, or one batch per shift. Waiting to accumulate a subgroup would delay the signal past the point of usefulness.
- Homogeneous batches. A tank of coating, a heat of steel, a reactor batch: measuring five samples from the same batch tells you about your sampling, not about batch-to-batch variation, which is what you care about. One value per batch is the right subgroup.
- Destructive or costly testing. Tensile tests, burst tests, and lab assays consume or expense the sample. You take as few as the process allows, often one.
- Accounting-style metrics. Daily scrap dollars, energy per unit, cycle time per job: naturally one number per period.
If you can form a genuine subgroup of like parts made under the same conditions, an X-bar and R chart is more sensitive and you should prefer it. I-MR is what you use when nature refuses to hand you a subgroup.
How does the moving range replace the subgroup range?
This is the trick that makes the chart work. On a subgroup chart, the range within each subgroup estimates the short-term process spread. With individuals you have no within-subgroup range, so you manufacture one: the moving range is the absolute difference between each reading and the one before it.
Concretely, for readings x1, x2, x3, the moving ranges are MR2 = |x2 - x1|, MR3 = |x3 - x2|, and so on. The first point has no moving range. Each moving range uses two consecutive points, so the effective subgroup size is n = 2, and the average moving range (MR-bar) becomes your estimate of short-term variation. The reason you use consecutive points is subtle but important: consecutive readings are the closest thing you have to "same conditions," so their difference captures the noise you want to separate from real shifts.
How do you calculate I-MR control limits?
Because the moving range fixes the subgroup size at 2, the control-chart constants are fixed too. You do not look anything up beyond these three values, which come straight from the standard control-chart constant tables for n = 2:
| Constant (n = 2) | Value | Used for |
|---|---|---|
| d2 | 1.128 | Estimating sigma: sigma-hat = MR-bar / 1.128 |
| D4 | 3.267 | Upper limit of the moving range chart |
| D3 | 0 | Lower limit of the moving range chart (so LCL = 0) |
From those, the limits fall out. Let X-bar be the average of the individual readings and MR-bar the average of the moving ranges. The individuals chart uses:
- Center line = X-bar
- UCL = X-bar + 3 × (MR-bar / 1.128), which simplifies to X-bar + 2.66 × MR-bar
- LCL = X-bar − 2.66 × MR-bar
The moving range chart uses:
- Center line = MR-bar
- UCL = 3.267 × MR-bar (that is D4)
- LCL = 0 (D3 is zero for n = 2, so a moving range chart has no meaningful lower limit)
The 2.66 factor is just 3 divided by 1.128; it is why you will see the individuals limits written as X-bar plus or minus 2.66 times MR-bar. The logic mirrors any Shewhart chart: three standard deviations either side of the center, with sigma estimated from the average moving range rather than from a subgroup range. This is the same three-sigma reasoning described in the NIST/SEMATECH e-Handbook.
How to build and read an I-MR chart in 7 steps
- Collect at least 20 to 25 readings in time order. The limits are only as trustworthy as the baseline. Fewer than about 20 points and MR-bar is a shaky estimate; the chart will mislead you.
- Compute the moving ranges. Take the absolute difference of each consecutive pair. You will have one fewer moving range than you have readings.
- Compute X-bar and MR-bar. Average the individuals for X-bar, average the moving ranges for MR-bar. These are your center lines.
- Set the limits. Individuals: X-bar plus or minus 2.66 times MR-bar. Moving range: upper limit 3.267 times MR-bar, lower limit zero. Nothing to look up beyond the constants.
- Check the moving range chart first. If the MR chart is out of control, your estimate of variation is corrupted, which means the individuals limits are wrong. Fix or explain the MR signals before you trust the individuals chart.
- Plot the individuals and apply the run rules. Look for points beyond the limits, but also for runs of eight on one side of center, trends of six or seven rising or falling, and other Western Electric or Nelson patterns. On individuals charts these run rules matter more than usual, because single points are less sensitive to small shifts than subgroup averages.
- Investigate signals, then recalculate only with cause. A signal means find the assignable cause, not erase the point. Recompute limits only after you have removed a documented special cause or the process has genuinely changed. Do not re-baseline just because a point looks inconvenient.
What are the limitations of the I-MR chart?
Two honest ones. First, individuals charts are less sensitive than subgroup charts. A single reading carries the full process variation, so a small shift can hide inside normal scatter for longer than it would on an X-bar chart. That is why the supplementary run rules earn their keep here. Second, I-MR assumes the individual values are roughly normal and, importantly, independent. If consecutive readings are autocorrelated, common when a process has momentum, like a slowly heating tank, the moving range underestimates the true spread and the limits come out too tight, producing false alarms. When you see that, the fix is a different chart or a time-series model, not a tighter reaction to every point.
Neither limitation makes I-MR wrong; they make it a tool with a job. When subgroups exist, prefer them. When they do not, I-MR is the honest choice, and pairing it with a process capability study tells you not just whether the process is stable but whether stable is good enough.
I-MR facts worth pinning down
A few reference points, straight from the standards and handbooks:
- For a two-point moving range, the bias-correction constant d2 equals 1.128, so the process standard deviation is estimated as MR-bar divided by 1.128 (NIST/SEMATECH e-Handbook of Statistical Methods).
- The moving range chart upper limit uses D4 = 3.267 and has no lower limit, because D3 = 0 at n = 2 (ASQ control chart guidance).
- Shewhart control limits are placed at three standard deviations from the center line, the convention that gives roughly a 0.27 percent false-alarm rate on a stable, normal process (ASQ).
Where the I-MR chart fits with the rest of quality
The I-MR chart is the low-volume workhorse of SPC. It is what you hand an operator running a batch process or a destructive test so in-process readings still get a real control chart instead of a gut call. It reads the same output your machine monitoring and manufacturing analytics already collect, and it feeds the same capability conversation once the process proves stable. Before building one, a quick measurement system analysis is worth the hour: on an individuals chart, gauge noise and process variation land in the same moving range, so a bad gauge inflates your limits directly.
The math is not the hard part; keeping a clean, time-ordered stream of readings is. Harmony captures each reading at the point of measurement, computes the chart automatically, and flags a run or a trend the moment it appears, on top of the systems and gauges you already run, no rip-and-replace. See how that looks on a real floor in our CLS case study or explore the capture and search features that turn one-at-a-time readings into a live control chart.