Shewhart Control Charts for Variables

Introduction

Statistical process control (SPC) distinguishes common-cause variation, which is inherent in a stable process, from special-cause variation, which arises from identifiable and correctable disturbances. Shewhart control charts are among the principal tools for making this distinction over time. Variable charts are used when the quality characteristic is measured on a numerical scale, such as diameter, tensile strength, temperature, cycle time, or filling volume.

This chapter develops three central charting systems: the X̄–R chart, the X̄–S chart, and the individuals–moving range (I–MR) chart. The first two are designed for rational subgroups containing several observations collected under similar conditions. The I–MR chart is intended for individual observations, especially when subgrouping is impractical or production volume is low.

A control chart is not a specification chart. Control limits describe the expected behaviour of a process under statistical control; specification limits describe customer, engineering, or regulatory requirements. A process may be stable but incapable of meeting specifications, or apparently capable on average but unstable over time.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Explain the statistical logic of Shewhart charts for measured data.
2. Construct and interpret X̄–R, X̄–S, and I–MR charts.
3. Estimate process centre and variation from rational subgroups.
4. Select an appropriate chart according to subgroup size and process context.
5. Evaluate assumptions, identify special causes, and avoid inappropriate reactions.
6. Relate control-chart evidence to process capability and improvement decisions.

1. Statistical foundation of variable charts

Suppose a process characteristic X has a stable mean μ and standard deviation σ. If measurements are approximately normally distributed, individual observations are represented by X_i ∼ N(μ, σ²). The normality assumption is useful for deriving traditional three-sigma limits, but the deeper requirement is that observations used to estimate variation are generated by a stable and appropriately structured process.

For a subgroup of size n, the sample mean is:

where X̄_j is the mean of subgroup j and X_{ij} is observation i in subgroup j. If observations within a subgroup are independent with standard deviation σ, then the standard deviation of the subgroup mean is σ/√n. Thus, means are less variable than individual measurements when n > 1.

The range of subgroup j is:

The sample standard deviation is:

The range is simple and efficient for small subgroups, whereas S uses all observations and is generally preferable for larger subgroups. Control limits are conventionally placed at an estimated centre ± three estimated standard errors. Under stable normal conditions, approximately 99.73% of plotted statistics lie within these limits.

Rational subgrouping

A rational subgroup is a set of observations selected so that variation within the subgroup represents short-term, common-cause variation, while differences between subgroup statistics reveal changes over time. For example, five consecutive bottles from a filling machine may form a rational subgroup. Mixing observations from different machines, shifts, or raw-material lots may conceal meaningful changes and produce misleading limits.

Subgrouping is therefore a process-design decision, not merely a sampling convenience. The sampling interval should be short enough to detect important process shifts but long enough to be operationally feasible.

2. General structure of Shewhart limits

For a statistic Y with centre μ_Y and standard deviation σ_Y, the three-sigma limits are:

UCL and LCL denote the upper and lower control limits; CL denotes the centre line. In practice, μ_Y and σ_Y are estimated from Phase I data collected while the process is being investigated and stabilized. Phase II then uses the established limits for ongoing monitoring.

Limits are not normally recalculated after every new point. Recalculation can allow an unstable process to redefine its own limits and can mask deterioration. Any limit revision should follow documented investigation and approval.

3. The X̄–R chart

The X̄–R system consists of two linked charts. The X̄ chart monitors process location, while the R chart monitors within-subgroup dispersion. The R chart is interpreted first because the limits for the X̄ chart depend on the estimate of within-subgroup variation.

For k subgroups, each of size n:

The constants A₂, D₃, and D₄ depend on subgroup size and are tabulated statistical constants. The limits are:

For subgroup sizes commonly used in practice, D₃ may equal zero; a negative lower limit is replaced by zero because a range cannot be negative. The constants incorporate the expected relationship between the range and σ for normal observations. An approximate estimate of process standard deviation is:

where d₂ is the expected range of a sample of n standard-normal observations. This estimate is appropriate when the subgrouping and distributional assumptions are reasonable.

Worked example: X̄–R chart

An assembly process is sampled in four rational subgroups of n = 4. The subgroup means are 10.2, 10.5, 10.1, and 10.4 mm; the corresponding ranges are 0.8, 1.2, 0.6, and 1.0 mm. Therefore:

X̄̄ = (10.2 + 10.5 + 10.1 + 10.4)/4 = 10.30 mm

R̄ = (0.8 + 1.2 + 0.6 + 1.0)/4 = 0.90 mm

For n = 4, use A₂ = 0.729, D₃ = 0, and D₄ = 2.282.

The R-chart limits are:

UCL_R = 2.282 × 0.90 = 2.054 mm and LCL_R = 0.

The X̄-chart limits are:

UCL_X̄ = 10.30 + (0.729 × 0.90) = 10.956 mm

LCL_X̄ = 10.30 − (0.729 × 0.90) = 9.644 mm.

All four ranges and means lie within these preliminary limits. This does not prove that the process is capable or that the assumptions are valid; it only indicates no three-sigma signal in this small dataset.

4. The X̄–S chart

The X̄–S chart is preferred when subgroup sizes are moderately large or when a more efficient estimate of within-subgroup variation is desired. The subgroup standard deviations are averaged:

The standard chart constants are c₄, A₃, B₃, and B₄. For subgroup size n, c₄ corrects the small-sample bias of the sample standard deviation. An unbiased estimate is:

The S-chart limits are:

The X̄-chart limits are:

The constants are functions of n. As with the X̄–R chart, the dispersion chart should be assessed before the location chart. If the S chart shows instability, interpretation of X̄-chart signals is weakened because the standard error estimate is unreliable.

X̄–R versus X̄–S

Neither chart is automatically superior. The choice must reflect subgroup size, data quality, measurement resolution, and the process mechanism.

5. Individuals–moving range chart

When only one observation is available at each sampling time, subgroup charts cannot be formed. The individuals–moving range chart uses the individual value chart and a moving range chart. For consecutive observations:

The usual moving range has width 2. If there are m moving ranges:

Using d₂ = 1.128 for a moving range of two observations, the short-term standard deviation is estimated as:

The individuals-chart limits are:

Equivalently, because 3/1.128 ≈ 2.66:

The moving-range limits are:

These formulas assume that consecutive observations are meaningfully comparable and that the process does not contain strong autocorrelation. If observations are serially correlated, moving ranges may underestimate or overestimate the relevant variation. Time-series modelling or a carefully designed subgroup strategy may then be more appropriate.

Worked I–MR example

Suppose a chemical concentration is recorded once per batch. Ten observations have a mean of 50.20 units and an average moving range of 0.80 units. Then:

σ̂ = 0.80/1.128 = 0.709 units

The individual-chart limits are:

UCL_I = 50.20 + 3(0.709) = 52.327 units

LCL_I = 50.20 − 3(0.709) = 48.073 units.

The moving-range upper limit is 3.267 × 0.80 = 2.614 units. A moving range above this value signals an unusually large change between consecutive batches, while an individual value outside its limits signals an unusual level.

6. Interpreting control charts

A point beyond a control limit is strong evidence of a special cause, but points inside the limits do not guarantee control. Nonrandom patterns can also signal instability. Common supplementary rules include:

• one point beyond a three-sigma limit;
• two of three successive points beyond the same two-sigma zone;
• four of five successive points beyond the same one-sigma zone;
• eight successive points on one side of the centre line;
• six successive points steadily increasing or decreasing.

Rules should be defined before routine monitoring, applied consistently, and selected with awareness of false-alarm rates. Adding many rules increases sensitivity but also increases the chance of false signals. A signal is an invitation to investigate, not automatic proof of a particular cause.

A disciplined response is to preserve evidence, verify the measurement system, identify changes in materials, equipment, methods, personnel, environment, and software, and determine whether affected product requires containment. Operators should not simply adjust the process toward the centre line after every fluctuation; such tampering can increase variation.

7. Assumptions and diagnostic evaluation

Stability and independence

Control limits estimate future behaviour from past data. If the process changes systematically, historical limits may be inappropriate. Autocorrelation is especially important in continuous processes, temperatures, pressures, and cycle-time data. A chart may show apparent runs because adjacent observations are related rather than because the process has shifted.

Distributional shape

The three-sigma rationale is most direct for approximately normal data. Moderate nonnormality may be tolerable, especially for subgroup means, but severe skewness, mixtures, truncation, or outliers require investigation. Transformations, distribution-free methods, or model-based approaches may be appropriate. Control limits should not be changed merely to make an inconvenient signal disappear.

Measurement-system adequacy

A noisy or unstable measurement system can create false signals, while coarse resolution can conceal real changes. Before relying on a chart, assess calibration, repeatability, reproducibility, bias, stability, and resolution. Measurement-system analysis is therefore a prerequisite, not an optional afterthought.

Constant subgroup size

The standard formulas assume a constant subgroup size. If subgroup sizes vary, use constants corresponding to each n, use software designed for variable sample sizes, or redesign sampling. Substituting a single arbitrary constant can produce incorrect limits.

8. Selecting the appropriate chart

Chart selection also depends on the decision horizon. A chart for machine adjustment may require frequent small subgroups; a chart for monthly supplier performance may appropriately use individual observations, provided the data-generating context remains comparable.

9. Control limits and specification limits

Control limits are calculated from process data; specification limits come from external requirements. Capability indices should be calculated only after statistical control is reasonably established. For a stable approximately normal process with lower and upper specification limits LSL and USL:

C_p measures potential capability if centred; C_pk also reflects displacement from the specification centre. These indices do not repair instability, nonnormality, or an inadequate measurement system.

10. Illustrative case: pharmaceutical fill-volume monitoring

Background

A packaging line fills sterile vials nominally at 10.0 mL. Every 30 minutes, five consecutive vials are measured using a validated gravimetric method. The quality team initially considers an X̄–R chart because n = 5 and rapid operator calculation is desirable.

Problem

Several complaints concern underfilled vials. A preliminary chart shows that subgroup means are generally within limits, but one range is unusually large. Investigation finds that the five vials in that subgroup were not consecutive: three came from the beginning and two from the end of a batch after a nozzle adjustment.

Analysis

The original subgrouping mixed two operating conditions. The large range reflected a within-subgroup process change rather than ordinary short-term variation. The team redesigns sampling so that each subgroup consists of five consecutive vials, records nozzle adjustments, and verifies the measurement system.

Approach and solution

A new Phase I dataset is collected after equipment stabilization. The R chart is reviewed first. One out-of-control range is traced to a partially blocked nozzle and removed only with documented technical justification. Limits are then recalculated. In Phase II, an X̄-chart shift follows a change in product viscosity; the batch is contained, viscosity is confirmed, and the process input is corrected.

Outcome

The revised chart distinguishes ordinary filling variation from nozzle and material-related causes. The process is subsequently evaluated against fill-volume specifications using capability indices only after a stable period is demonstrated.

Learning points

1. Rational subgrouping determines what variation the chart can detect.
2. The dispersion chart must be interpreted before the mean chart.
3. Investigation should target process causes, not merely chart points.
4. Limit revision requires evidence and traceability.
5. SPC supports, but does not replace, product disposition and regulatory controls.

11. Technology, governance, and contemporary practice

Modern SPC software automatically calculates limits, applies supplementary rules, and links signals to production records. These capabilities improve speed but do not replace statistical judgement. Automated systems should preserve raw data, time stamps, subgroup definitions, revisions, user actions, and audit trails. In regulated environments, electronic records and electronic signatures must comply with applicable quality-system and data-integrity requirements.

Machine-learning systems may help detect nonlinear patterns or combine process variables, but an opaque prediction is not equivalent to a validated Shewhart signal. A responsible implementation defines the target, validates the algorithm, controls false alarms, protects confidential production data, and retains human accountability for investigation and disposition.

Best-practice checklist

• Define the quality characteristic, sampling frequency, and rational subgroup before collecting data.
• Verify measurement-system adequacy.
• Use a documented Phase I study to establish stable limits.
• Inspect the dispersion chart before interpreting location signals.
• Record causes and actions for every investigated signal.
• Avoid routine recalculation and unjustified deletion of points.
• Separate control limits from specification limits.
• Assess autocorrelation, nonnormality, and changing process conditions.
• Train operators to investigate signals rather than tamper with the process.
• Review chart effectiveness periodically against business and customer risk.

Chapter Summary

Shewhart variable charts use measured data to distinguish common-cause variation from special-cause variation. The X̄–R chart is suited to small rational subgroups and estimates dispersion through ranges. The X̄–S chart uses subgroup standard deviations and is generally more efficient for larger subgroups. The I–MR chart is appropriate when observations occur singly, using consecutive moving ranges to estimate short-term variation.

Correct chart construction requires rational subgrouping, valid estimates of centre and dispersion, and appropriate chart constants. The dispersion chart is assessed before the location chart. Signals include points beyond limits and nonrandom patterns, but every signal requires investigation rather than automatic adjustment. Assumptions concerning stability, independence, distribution, subgroup structure, and measurement quality must be evaluated. Finally, control limits describe process behaviour, whereas specification limits describe requirements; capability analysis is meaningful only after stability has been established.

Key Terms

Control chart: A time-ordered graphical method for monitoring process behaviour.

Common cause: A source of variation inherent in the current process system.

Special cause: An unusual, identifiable source of process variation.

Rational subgroup: A group of observations selected to represent short-term process conditions.

X̄ chart: A chart for monitoring subgroup means.

R chart: A chart for monitoring subgroup ranges.

S chart: A chart for monitoring subgroup standard deviations.

Individuals chart: A chart for monitoring single observations.

Moving range: The absolute difference between consecutive observations.

Centre line: The estimated central value of a chart statistic.

Control limit: A statistically estimated boundary for expected process behaviour.

Specification limit: An externally defined acceptance boundary.

Phase I: The retrospective stage used to understand and stabilize a process.

Phase II: Prospective monitoring using established control limits.

Process capability: The relationship between stable process variation and specifications.

C_p: A potential capability index assuming centring.

C_pk: A capability index incorporating process centring.

Autocorrelation: Dependence between observations ordered in time.

Measurement-system analysis: Evaluation of measurement error and adequacy.

Tampering: Unnecessary adjustment of a stable process in response to random variation.

Review Questions

Short Answer Questions

1. What is the purpose of a Shewhart control chart?
2. Define a rational subgroup.
3. Why is the R or S chart examined before the X̄ chart?
4. What does a moving range measure?
5. Distinguish control limits from specification limits.

Long Answer Questions

1. Explain the construction and interpretation of an X̄–R chart, including the roles of A₂, D₃, and D₄.
2. Compare the X̄–R and X̄–S charts with respect to subgroup size, efficiency, and practical use.
3. Derive the logic of the I–MR chart and discuss its assumptions.
4. Explain how poor rational subgrouping can conceal or create apparent process problems.
5. Discuss the relationship between statistical control, process capability, and customer specifications.

Analytical / Application Questions

1. A process is sampled in subgroups of three, but observations are collected from different machines and shifts. Evaluate whether the subgrouping is appropriate and propose a redesign.
2. An I–MR chart shows no points beyond its limits, but the moving-range chart shows a long sequence of very small ranges. What process or data issues should be investigated?
3. A stable process has C_p = 1.50 but C_pk = 0.82. Interpret the results and recommend an improvement priority.

Multiple Choice Questions with Answers

Question 1

Which chart is primarily used to monitor within-subgroup dispersion when small rational subgroups are collected?

A. X̄ chart

B. R chart

C. Individuals chart

D. p chart

Correct Answer: B. R chart

Question 2

For an X̄–R chart, the usual estimate of process standard deviation is:

A. R̄/d₂

B. d₂/R̄

C. X̄̄/A₂

D. S̄/c₄

Correct Answer: A. R̄/d₂

Question 3

Which chart is most appropriate when only one comparable observation is available at each time point?

A. X̄–R chart

B. X̄–S chart

C. Individuals–moving range chart

D. np chart

Correct Answer: C. Individuals–moving range chart

Question 4

A rational subgroup should primarily contain observations that:

A. Come from unrelated processes

B. Represent short-term conditions of the process

C. Are selected only after a defect occurs

D. Mix several shifts deliberately

Correct Answer: B. Represent short-term conditions of the process

Question 5

For an X̄–R chart with n = 4, if X̄̄ = 10.30 and R̄ = 0.90, using A₂ = 0.729, the upper X̄ limit is approximately:

A. 9.644

B. 10.300

C. 10.956

D. 12.582

Correct Answer: C. 10.956

Question 6

The usual estimate of standard deviation for an I–MR chart is based on:

A. The grand mean divided by n

B. The average moving range divided by 1.128

C. The maximum observation divided by 3

D. The specification width divided by 6

Correct Answer: B. The average moving range divided by 1.128

Question 7

Why is the dispersion chart normally interpreted before the location chart?

A. Location charts cannot be plotted first

B. The location limits depend on the estimated within-subgroup variation

C. Dispersion is always more important than location

D. Mean values are not measured data

Correct Answer: B. The location limits depend on the estimated within-subgroup variation

Question 8

Which statement correctly distinguishes control and specification limits?

A. Both are always calculated from customer requirements

B. Control limits describe process behaviour; specification limits describe requirements

C. Specification limits are recalculated after every subgroup

D. Control limits determine whether a product meets every design requirement

Correct Answer: B. Control limits describe process behaviour; specification limits describe requirements

Question 9

If C_p is high but C_pk is substantially lower, the most likely implication is that the process:

A. Has no measurable variation

B. Is centred perfectly

C. Has potential capability but is poorly centred

D. Must be statistically unstable

Correct Answer: C. Has potential capability but is poorly centred

Question 10

A point beyond a control limit should normally be treated as:

A. Automatic proof that all product is defective

B. A reason to investigate a possible special cause

C. Evidence that the control limits must immediately be widened

D. A reason to delete the observation

Correct Answer: B. A reason to investigate a possible special cause