A specification tells you what product is acceptable. A control chart tells you whether the process behavior has changed. The two questions are related, but they are not interchangeable—and the difference is exactly why a process can show a high Cpk while its X̄ chart still demands investigation.
This guide builds both charts from a complete 25-subgroup sample. Within-subgroup ranges remain stable, but the process mean shifts after subgroup 15. The result is deliberately useful: the R chart passes its immediate check while the X̄ chart exposes a second operating state.
One subgroup produces two signals
R chart
Is the short-term variation inside each subgroup consistent?
Each point is the largest minus the smallest value in one subgroup. A high range can indicate a short-term special cause, measurement issue, material difference, or inconsistent process condition.
Interpret this chart first.X̄ chart
Are subgroup averages consistent with one process location?
Each point is the mean of one subgroup. The chart can reveal shifts, trends, cycles, and unusual means that remain hidden when all measurements are pooled into one histogram.
Interpret after variation is understood.If the R chart is unstable, the estimate of within-subgroup variation used to construct the X̄ limits is not reliable as one common baseline. Investigate the variation signal before treating an X̄ result as the final process-location conclusion.
Rational subgrouping determines what the chart can see
A rational subgroup should contain observations produced under conditions expected to be as similar as practical. The variation within a subgroup estimates short-term common-cause variation. Differences between subgroup averages then reveal changes across time, setups, lots, tools, or other meaningful production intervals.
Subgroup choice is an engineering model of the process. Five measurements from five different machines do not estimate the same short-term variation as five consecutive pieces from one machine. Both arrangements can be calculated, but the limits answer different questions.
Calculate the limits from subgroup means and ranges
For each subgroup i, calculate its mean X̄i and range Ri. Then average the subgroup means to obtain the grand mean X̄̄ and average the subgroup ranges to obtain R̄.
Ri = max(xij) − min(xij)
R chart CL = R̄
LCLX̄ = X̄̄ − A2 × R̄A2 depends on constant subgroup size n.
LCLR = D3 × R̄D3 and D4 also depend on subgroup size n.
| Subgroup size n | A2 | D3 | D4 |
|---|---|---|---|
| 2 | 1.880 | 0 | 3.267 |
| 3 | 1.023 | 0 | 2.575 |
| 4 | 0.729 | 0 | 2.282 |
| 5 · used here | 0.577 | 0 | 2.115 |
| 6 | 0.483 | 0 | 2.004 |
| 7 | 0.419 | 0.076 | 1.924 |
| 8 | 0.373 | 0.136 | 1.864 |
| 9 | 0.337 | 0.184 | 1.816 |
| 10 | 0.308 | 0.223 | 1.777 |
The range method is commonly used for small subgroups. As subgroup size grows, an X̄–S chart generally uses the subgroup standard deviation more efficiently. When subgroup size is one, an individuals-and-moving-range approach is a different chart—not an X̄–R chart with missing values.
Worked example: the mean changes while the range remains stable
The prepared dataset contains 25 subgroups of five diameter measurements. Subgroups 1–15 operate near 10.000 mm. Subgroups 16–25 operate near 10.035 mm. The within-subgroup patterns and ranges remain nearly unchanged.
View all 125 measurements and subgroup statistics
| SG | x1 | x2 | x3 | x4 | x5 | X̄ | R |
|---|
LCL = 10.01448 − 0.577 × 0.03360 = 9.99509
LCL = 0 × 0.03360 = 0
The product charts make the shift visible
The charts below are generated from the same 125 readings using the X̄–R graph treatment implemented in the Mechatrovich SPC workflow. Hover over each point for subgroup values and signal labels.
X̄–R Stability Report — Diameter
25 rational subgroups · n = 5 · A2 = 0.577 · D3 = 0 · D4 = 2.115
R Chart / Within-Subgroup Variation
Read first
X̄ Chart / Subgroup Average
Read after the R chart
The R chart remains within its calculated limits. The X̄ chart shows two separated bands, eight means above the UCL, and a sustained side-of-center pattern. The pooled process is not one stable location.
All subgroup ranges lie between 0.032 and 0.035 mm, far below the R-chart UCL. No single subgroup shows unusually large within-subgroup spread.
Subgroups 16, 17, 19, 20, 21, 22, 24, and 25 exceed the upper control limit. The visible level change begins at subgroup 16 and persists through subgroup 25.
Check setup, adjustment, tool change, material lot, machine state, program revision, operator, temperature, or another recorded event at that boundary. The chart locates the change; process evidence identifies the cause.
Why a high Cpk does not cancel the control-chart signal
If the same 125 measurements are compared with specifications of 9.800 and 10.200 mm using one overall sample standard deviation, the pooled Cpk is approximately 2.92. That sounds excellent—but it combines two distinct process levels.
Capable-looking, but not statistically stable
The specification is wide relative to both operating levels, so no measurement is close to a limit. The X̄ chart still proves that the process location changed. Capability and stability answer different questions; the high index cannot explain or approve the shift.
After the cause is identified, separate the operating states and decide which one represents the intended process. Establish a defensible baseline before making a capability claim. If the higher level is an approved permanent improvement, new limits may eventually be justified. If it is an assignable cause, correct it and retain limits that represent the controlled process.
Control-limit violations are only the first signal rule
Additional rules increase sensitivity, but they also increase false alarms. Choose and document the rule set before reviewing the chart. Applying every available rule after seeing the data encourages overreaction and makes the monitoring plan difficult to audit.
A signal starts an investigation; it does not name the cause. Do not delete the point, recalculate the limits, and continue without evidence. Record what changed, whether the cause is assignable and preventable, what action was taken, and whether subsequent data confirm the result.
Phase I limits and Phase II monitoring serve different jobs
Phase I · establish the baseline
Use historical or preliminary subgroups to find special causes, understand process structure, and estimate limits from a period judged representative. Removing a subgroup requires technical evidence, not merely an inconvenient signal.
Phase II · monitor the process
Apply established limits to future subgroups. Do not continuously recalculate limits after every new point; moving the reference with the process can hide the change the chart is meant to detect.
The 25-subgroup sample here is a Phase I illustration. Because the shift is built into the baseline data, the first conclusion is not “publish these limits.” The first conclusion is “investigate and separate the two states.” Limits should be re-estimated only after the intended controlled state is defined.
A practical X̄–R review sequence
Use continuous data, constant small subgroup size, and a defensible rational-subgroup definition.
Check MSA evidence, units, rounding, subgroup IDs, timestamps, missing values, and collection sequence.
Retain the raw values so any signal can be traced to the original measurements.
Investigate unusual within-subgroup variation before accepting one pooled variation estimate.
Look beyond limit violations to sustained shifts, trends, or other documented patterns.
Stratify by machine, cavity, tool, lot, material, operator, setup, temperature, maintenance, and other known sources.
Only then estimate process capability using data and a sigma method that represent the intended state.
Common X̄–R mistakes
- Using specification limits as control limits.
- Creating subgroups from convenient reporting buckets rather than process logic.
- Reading the X̄ chart before resolving an unstable R chart.
- Changing subgroup size while using one constant set.
- Using X̄–R for individual measurements or very large subgroups without reviewing chart suitability.
- Sorting or aggregating away the production sequence.
- Deleting signals and recalculating limits without technical evidence.
- Recalculating limits so frequently that the baseline follows the shift.
- Using a high Cpk as evidence that a control-chart signal does not matter.
Continue from the chart to the workflow
Use the free X̄–R calculator for a small subgroup check, or return to the capability guide when the process has a defensible stable state. The complete SPC toolkit adds full subgroup tables, signal evaluation, logs, saved studies, and report output.
For a full offline subgroup workflow with both control charts, extended rule checks, a signal log, and reporting, continue to the Mechatrovich X̄–R SPC Toolkit.
References and method sources
- NIST/SEMATECH e-Handbook, Shewhart X̄ and R and S Control Charts — formulas, constants, chart selection, and range-method efficiency.
- NIST, What are Variables Control Charts? — control limits versus specifications, baseline-sample context, recalculation guidance, and Western Electric rules.
- AIAG & VDA SPC Manual, 1st Edition (2026) — current automotive-industry framework for SPC implementation, monitoring, performance, and capability.