This is a ready-to-use form for documenting why a control chart’s subgroups are built the way they are. Complete it before the chart goes live, replace every <<FILL: ...>> placeholder, and attach it to the parameter’s monitoring record or CPV plan. A worked filled specimen follows. The design decision recorded here is what determines whether the chart can see the signal it exists to catch; a technically correct chart built on the wrong subgroup composition is not a working chart.
Form control
| Field | Format | Required | Completed by | When |
|---|---|---|---|---|
| Form number | Text | Yes | Document control | On issue |
| Parameter / characteristic | Text | Yes | Process owner / statistician | Before the chart goes live |
| Chart type | Text | Yes | Process owner / statistician | Before the chart goes live |
| Date | Date | Yes | Preparer | At completion |
Instructions
- Name the specific source of variation this chart needs to catch before deciding anything about sample size or frequency; the rest of this form follows from that answer.
- State what the subgroup composition holds constant and what it lets vary, and confirm the source of variation named in step 1 falls in the “lets vary, between subgroups” column, not the “holds constant, within subgroup” column.
- Where more than one source of variation matters, decide whether one chart can honestly serve both or whether separate charts are needed, and say which.
- Complete the worst-case check before sign-off: if the named source of variation actually occurred, would this design show it up between subgroups, or would it be absorbed inside a subgroup and stay invisible?
1. Header
| Field | Entry |
|---|---|
| Parameter / characteristic | <<FILL>> |
| Process step | <<FILL>> |
| Chart type selected | <<FILL: X-bar/R, X-bar/S, I-MR, p, np, c, u>> |
| Governing SOP | <<FILL: control chart program SOP-ID>> |
2. Source of variation targeted
| Field | Entry |
|---|---|
| Specific source of variation this chart is meant to catch | <<FILL: e.g. shift-to-shift difference, head-to-head difference on a multi-head filler, lot-to-lot raw material difference, day-to-day drift>> |
| Why this source matters for this parameter | <<FILL: link to the risk assessment, prior deviation history, or process knowledge that makes this the source worth watching>> |
| Other sources of variation considered and not targeted by this chart | <<FILL: state them, and whether a separate chart covers each>> |
3. Subgroup composition
| Field | Entry |
|---|---|
| Subgroup size (n) | <<FILL>> |
| What is held constant within a subgroup | <<FILL: e.g. same head, same operator, same 10-minute window>> |
| What is allowed to vary between subgroups | <<FILL: e.g. different shifts, different raw material lots>> |
| Confirmation the targeted source (section 2) falls in the “varies between subgroups” column | <<FILL: Yes, confirmed / No, redesign needed>> |
| Sampling frequency | <<FILL>> |
| Rationale for that frequency | <<FILL: e.g. matched to batch cadence, matched to the shortest interval a real shift could occur and go undetected otherwise>> |
4. Sample size rationale
| Field | Entry |
|---|---|
| Sample size within each subgroup | <<FILL>> |
| Basis (statistical power, historical convention, destructive-test cost constraint, regulatory minimum) | <<FILL>> |
| Number of subgroups planned for the baseline | <<FILL: commonly 20 to 30>> |
| If below the common baseline range, state why and how the limits will be labeled | <<FILL: e.g. low-volume product, limits will be provisional until N more subgroups accumulate>> |
5. Worst-case check
| Question | Answer |
|---|---|
| If the targeted source of variation (section 2) occurred exactly as described, would this subgroup design show it as a between-subgroup signal? | <<FILL: Yes/No, with reasoning>> |
| Could this design instead absorb that variation inside a subgroup and hide it (a false sense of stability)? | <<FILL: Yes/No, with reasoning>> |
| Is there a competing source of variation that this design could mistake for the targeted one? | <<FILL: e.g. a shift change that coincides with a raw material lot change, confounding the two>> |
If the answer to the first question is anything other than a confident “yes,” the subgroup design is not ready to go live; return to section 3 and redesign before proceeding to sign-off.
6. Acceptance criteria
- The targeted source of variation is named specifically, not generically (“normal process variation” is not an answer).
- The subgroup composition is shown to place that source between subgroups, not within them.
- Sample size and frequency are each justified, not defaulted to a round number with no stated reason.
- The worst-case check in section 5 is completed and answered “yes” before the chart goes live.
- Any confounding source identified in section 5 is either resolved by the design or explicitly accepted and stated as a limitation.
7. Signatures
| Role | Name | Signature | Date |
|---|---|---|---|
| Preparer (process owner / statistician) | <<FILL>> | ||
| Reviewer (statistician / quality engineer) | <<FILL>> | ||
| QA | <<FILL>> |
8. References
Shewhart, W. A., the original rational subgrouping concept underlying Western Electric and Nelson rule interpretation. FDA, Process Validation: General Principles and Practices (2011), Stage 3. See statistics in quality: Cpk and control charts for the reasoning behind rational subgrouping and worked chart examples.
Confirm the current version of each reference before issue.
Filled specimen
Parameter: fill weight on a two-head filling line, X-bar/R chart, n = 4 per subgroup.
Section 2: Targeted source of variation is a head-to-head difference (one head drifting out of adjustment while the other stays centered), based on a prior deviation where head 2 drifted for three shifts before detection because sampling had been pulling all four units from head 1 only. Other sources considered: shift-to-shift drift, covered separately by a weekly shift-comparison report, not by this chart.
Section 3: Subgroup size 4: two units from head 1, two from head 2, pulled within the same 5-minute window so within-subgroup variation reflects only short-term common cause, not a time-separated shift. What is held constant: the time window (5 minutes). What varies between subgroups: which head each unit came from is fixed at 2-and-2 every subgroup, so the average difference between heads shows up as a component of the subgroup’s own range rather than as a between-subgroup signal. On review, this was identified as a design flaw or confirmed as the section 5 worst-case check shows below.
Section 5, worst-case check: If head 2 drifted out of adjustment while head 1 stayed centered, would this design show it as a between-subgroup signal? No. Pulling 2-and-2 every subgroup means a head 2 drift shows up as increased within-subgroup range (wider spread inside each subgroup, since the two heads now disagree), not as a shift in the subgroup average, and range-based rule violations are far less intuitive for operators to act on than a mean-shift signal. Could this design hide the targeted source? Partially: a moderate drift would widen the R chart rather than shift the X-bar chart, which is a real signal but a different one than intended. Competing source: none identified beyond the head-mixing effect itself.
Resulting action: The design was revised to run two separate X-bar/R charts, one per head, each subgrouped from that head alone (n = 4, all four units from the same head per subgroup). This lets a head-specific drift appear as a straightforward mean shift on that head’s own chart, which is what the original deviation needed and what the mixed-head design could not reliably provide. The mixed-head design was retained only as a secondary combined-average trend, not as the primary monitoring chart.
Common findings this form prevents
- A subgroup built for sampling convenience that averages away the exact signal the chart exists to catch.
- No documented reason for the sample size or frequency chosen.
- A chart that has been running for months on a subgroup design nobody checked against a worst-case scenario.
- Two competing sources of variation (a shift change and a lot change that happen together) silently confounded in one chart’s signal.
How to adapt this form
- Set the form number, parameter, and chart type in the header.
- Name the real source of variation from your own deviation history or risk assessment, not a generic placeholder.
- Run the worst-case check honestly; if it fails, redesign the subgroup before the chart goes live, as shown in the specimen.
- Attach this form to the parameter’s monitoring record, CPV plan, or the control limit establishment worksheet for the same parameter.