Subgrouping Mistakes Lead to False Signals on Control Charts — A Five-Step Method for Rational Subgrouping and Sampling Strategy
1. Daily Alarms, but No Issues Found on the Production Line
A plastic injection molding company implemented an X̄-R control chart for a critical dimension of a housing component, making the quality control process appear very stringent: every shift, 5 parts produced during the shift were selected and sent for measurement, forming a subgroup for recording.
After three months of operation, a strange phenomenon emerged — the control chart reported 4 to 5 abnormalities each week. When the production line was stopped to investigate, issues with the mold, material, and parameters were rarely found. Conversely, on the day when the dimension actually started to drift, the control chart remained quiet until the entire batch of nonconforming products reached the assembly line and was discovered.
The quality manager reviewed the original records and found the problem at the most fundamental step: the 5 samples were not continuously produced parts but were "picked" from different times and different machines within the shift. A subgroup contained parts produced in the morning with mold 1, in the afternoon with mold 2, and even with a different batch of material in between. The control limits derived from such data were both wide and skewed, leading to frequent false alarms and missed detections.
There are many reasons why control charts can fail, but the most overlooked and fatal one is incorrect subgrouping and sampling strategy. The tool itself is not the issue; the problem lies in the way the data is organized.
2. What Does "Rational" in Rational Subgrouping Mean?
To understand rational subgrouping, remember the foundation of statistical process control: process variation is divided into two categories.
- Within-group variation: the differences within the same subgroup, representing common causes, which are inevitable random fluctuations in daily operations.
- Between-group variation: the differences between subgroups, representing special causes, such as shift changes, mold changes, material changes, parameter adjustments, and equipment wear.
The control chart manages these two types of variation: it uses within-group variation to estimate the width of the control limits and the position changes of subgroup points to detect between-group variation. Therefore, the principle of rational subgrouping is to maximize the representation of common causes within a subgroup and leave special causes to be reflected between subgroups.
Breaking this principle can lead to two consequences. If a subgroup includes parts from different mold changes or material batches, the within-group variation is artificially inflated, making the control limits wider and masking true abnormalities, leading to missed detections. Conversely, if a subgroup is sampled at the edge of an abnormal period, the within-group range may be falsely small, making the control limits too narrow and causing frequent false alarms. False alarms erode trust on the production floor, while missed detections erode customer trust, both pointing to the same root cause.
The judgment is straightforward: when you receive a set of data, ask yourself — were these parts produced in the shortest possible time, on the same equipment, with the same mold, from the same batch of material, by the same operator, and under the same parameter settings? If any of these six "same" conditions are not met, the subgroup is not valid.
3. Five-Step Method: From Sources of Variation to Sampling Plan
Step 1: List the sources of variation. Identify all factors that can cause changes in the characteristic of the process: shifts, equipment, molds and tooling, material batches, parameter settings, operators, and environmental temperature. This list determines how subgroups should be formed. For example, in injection molding, parts produced continuously with the same mold, the same batch of material, and within the same shift are considered the same condition; a mold change should be treated as a new condition.
Step 2: Form subgroups in chronological order. Sample parts that are produced consecutively, with n typically being 4 to 5 pieces. Do not combine parts from different times or different machines into one subgroup. If multiple machines are running on a line, the correct approach is not to mix them into one chart but to create separate charts for each machine or to use the machine as a stratification variable before analyzing whether the process is in control.
Step 3: Determine the sample size. If n is too small (e.g., n=2), the within-group variation estimate is unstable, and the control limits fluctuate significantly. If n is too large (10 or more), the slow drift within the group is averaged out, reducing sensitivity. The typical range is 4 to 5. For processes with long cycle times and automatic online measurements, use an I-MR (Individuals-Moving Range) chart, where "rationality" is ensured by the moving range and the sampling interval, rather than by forming a subgroup.
Step 4: Set the sampling frequency. The frequency should match two factors: how quickly the process changes and the cost of missed detections. The principle is that no undetectable changes should occur between two sampling points. After a change in type, material, or shift, or after equipment maintenance, an additional subgroup must be sampled immediately. High-risk periods such as night shifts, new employees, and start-stop operations should have increased sampling frequency. The frequency can be relaxed when the process stabilizes, but it should not be canceled. More frequent sampling is not always better; if it is too frequent to be practical, it is as good as not being done.
Step 5: Perform a subgroup validation. After plotting the data, first check the average range R̄ and the shape of the range chart. If R̄ is significantly large and the range chart frequently shows points close to the upper limit, the primary suspect is not a change in the process itself but the mixing of different conditions within the subgroup. Conversely, if the control limits are unrealistically narrow and alarms are too frequent, re-examine the subgrouping. Correct subgrouping ensures that the control limits are meaningful and that the rules for detecting abnormalities are valuable.
4. Four Common Mistakes in Subgrouping
- Arbitrary subgrouping. At the end of the shift, parts are randomly picked to form a subgroup, or data is added to the chart whenever a sufficient number of parts are produced. This is the most common mistake and has the most significant consequences — the control limits lose their physical meaning.
- Combining across shifts and batches. Parts from two different shifts or material batches are combined into the same subgroup, artificially inflating within-group variation and prematurely "digesting" special causes.
- Subgrouping by work order or inspection batch. Work orders and inspection batches are management units, not rational subgroups. When an inspection batch spans several days of production, grouping by batch is equivalent to compressing the entire period's drift into the within-group variation.
- Overly large subgroups. To make the data "more sufficient," n is increased to a dozen or more pieces. The within-group range is averaged out, significantly reducing sensitivity, and placing an unbearable measurement burden on the production floor.
5. Three Key Sentences to Implement the Principle
First, identify the conditions that change in this process and decide how to form subgroups; second, a subgroup should only contain parts produced under the same conditions, leaving special causes to be reflected between subgroups; third, the sampling frequency should serve the purpose of "timely detection of changes," and additional sampling must be conducted after a change in type, material, or shift.
A control chart is not just a tool for filling in data; its resolution is determined by the subgrouping method. When subgroups are correctly formed, the chart can reveal true issues.
The resolution of a control chart is determined by the subgrouping method. When subgroups are correctly formed, the chart can reveal true issues.
Knowledge code: 6.3.1
Version: v20260914
Author: QTank QTank is dedicated to providing systematic knowledge, methodologies, and practical tools for quality management professionals, helping companies continuously improve their quality capabilities.