Statistical Quality Control and Process Capability

Introduction

Quality in a supply chain is not created only by final inspection. It is created by controlling the processes that transform inputs into products, services, information and deliveries. Statistical Quality Control (SQC) provides a disciplined way to distinguish ordinary, unavoidable variation from unusual variation that requires investigation. Process capability analysis then asks whether a stable process can consistently meet customer or engineering requirements. Acceptance sampling provides a practical decision method when inspecting every item is costly, destructive or impractical.

For an operations manager, these methods support evidence-based decisions about suppliers, production lines, warehouses, transport operations and service processes. They do not replace engineering knowledge or managerial judgement. Rather, they make variation visible and provide a common language for improvement.

Learning objectives

By the end of this chapter, the learner should be able to:

1. Explain common and special causes of variation.
2. Apply descriptive statistics and basic sampling methods to quality data.
3. Select and use check sheets, Pareto charts, histograms and cause-and-effect diagrams.
4. Construct and interpret appropriate control charts.
5. Calculate and interpret process capability indices.
6. Explain the logic and limitations of acceptance sampling.
7. Recommend suitable quality actions in a supply-chain context.

1. Variation and the statistical view of quality

A process is a connected set of activities that converts inputs into outputs. Inputs may include materials, machines, methods, people, measurements, information and environmental conditions. Even when the process is operated correctly, outputs are rarely identical. This natural variability is called variation.

Common and special causes

Common causes are numerous, small and inherent in the current process system. Examples include ordinary machine-to-machine differences, normal temperature fluctuation and routine variation in raw material properties. A process affected only by common causes is statistically stable, although it may still be incapable of meeting specifications.

Special causes are identifiable, unusual sources of variation. Examples include a broken tool, an incorrectly calibrated scale, a new operator using an unapproved method, a contaminated batch or a transport refrigeration failure. Special causes make the process unstable and should normally be investigated before capability is assessed.

The distinction is important. Treating common-cause variation as if it were a special cause creates tampering: frequent adjustments that increase variation. Ignoring special causes allows preventable failures to continue.

Specifications and control limits

A specification limit is an external requirement established by a customer, regulation, design authority or contract. The upper and lower specification limits are written USL and LSL. A control limit is calculated from process data and describes expected process behaviour when only common causes are present. Control limits are not customer requirements.

A process can therefore be:

• stable and capable;
• stable but incapable;
• unstable but apparently within specifications at a particular moment; or
• unstable and incapable.

Control charts assess stability; capability indices assess performance relative to specifications. These are related but different questions.

2. Descriptive statistics and sampling

Quality analysis begins with credible data. Before calculating statistics, define the unit of analysis, the quality characteristic, the measurement method, the sampling frame and the time period.

Measures of location and dispersion

For observations x₁, x₂, …, xₙ, the sample mean is:

Here x̄ is the sample mean, xᵢ is observation i and n is the sample size. The mean has the same units as the observations and is most informative when extreme values are not dominant.

The sample variance and sample standard deviation are:

The divisor n − 1 is used when the sample estimates the variability of a larger population. If the entire finite population is measured, a population standard deviation using divisor N may be appropriate. Standard deviation measures spread in the original units; variance is in squared units.

The median is the middle ordered value, while the range is max(xᵢ) − min(xᵢ). The interquartile range, Q₃ − Q₁, is useful when data are skewed or contain outliers. A histogram and time sequence should accompany summary statistics because the same mean and standard deviation can arise from very different distributions.

Sampling principles

A sample should represent the process and should not conceal time, shift, supplier, machine or product differences. Common designs include:

• Simple random sampling: each unit has a known, equal chance of selection.
• Systematic sampling: every kth unit is selected after a random start.
• Stratified sampling: the population is divided into groups such as supplier or shift, and each group is sampled.
• Rational subgroup sampling: units produced close together in time are grouped to reveal short-term process variation.

Convenience samples are easy but can be seriously biased. For control charts, rational subgroups should reflect the variation expected during routine production. For supplier comparison, sampling should cover relevant lots and operating conditions.

The standard error of the sample mean, under independent observations, is:

where σ is the population standard deviation. In practice, s often replaces σ. The formula shows that increasing sample size reduces uncertainty, but with diminishing returns. Independence is an assumption; autocorrelation, clustering and changing process conditions require additional analysis.

Measurement-system quality

A process cannot be controlled using unreliable measurements. Operators should verify calibration, resolution, repeatability and reproducibility. Repeatability is variation when the same person measures the same item repeatedly. Reproducibility is variation among people or appraisal conditions. A measurement-system analysis should be completed before interpreting small process differences as operational problems.

3. The basic quality tools

Check sheets

A check sheet is a structured form for collecting data at the point of work. It may record defect type, location, time, supplier, machine or cause category. Effective check sheets define each category clearly, use consistent units and provide space for totals and comments. They convert observations into analyzable data and reduce reliance on memory.

Pareto charts

A Pareto chart ranks categories from largest to smallest, usually with a cumulative-percentage line. It implements the practical insight that a small number of categories often account for a large share of losses. The chart does not prove causation; it identifies where investigation may produce the greatest benefit.

For category i, its percentage contribution is:

where fᵢ is the frequency of category i. Categories should be mutually exclusive and collectively meaningful.

Histograms

A histogram groups numerical observations into intervals and displays their frequencies. It helps reveal centre, spread, skewness, multiple populations and possible outliers. A bimodal histogram may indicate two machines, suppliers or shifts combined in one dataset. Changing the number or width of bins can alter the appearance, so the analyst should examine the data by relevant strata as well.

Cause-and-effect diagrams

A cause-and-effect, or fishbone, diagram organizes possible causes of a defined effect. Typical manufacturing headings are the 6Ms: Manpower, Machine, Method, Material, Measurement and Mother Nature (environment). In logistics, headings might include people, technology, process, carrier, information and facility.

The diagram is a hypothesis-generation tool, not evidence that every listed cause is active. Suspected causes should be tested with observations, experiments, stratification or controlled comparisons.

Flowcharts and scatter plots

A flowchart exposes handoffs, rework loops and decision points. A scatter plot explores association between two variables, such as humidity and seal failures. Association does not establish causation; confounding variables and time order must be considered.

4. Control charts

A control chart displays a statistic in time order with a centre line and upper and lower control limits. A basic three-sigma chart has the form:

Here CL is the process centre line and σₛ is the standard deviation of the plotted statistic. The limits must be based on a stable baseline or a justified estimate of within-process variation. If the LCL is negative for a count or proportion chart, it is set to zero because such statistics cannot be negative.

A point outside a control limit is strong evidence of a special cause, subject to chart assumptions. Other signals include a sustained shift to one side of the centre line, a long trend, cycles, or an unusually long run near a limit. Exact run rules should be defined before analysis to avoid selectively finding signals.

Variable charts

For measurements such as weight, temperature or delivery time, common charts include:

• X̄–R chart: mean and range charts for small rational subgroups.
• X̄–s chart: mean and standard deviation charts, often for larger subgroups.
• Individuals–moving range chart: for one observation at a time when rational subgroups are unavailable.

For an X̄–R chart, if R̄ is the average subgroup range and x̄̄ is the grand mean, the conventional limits are:

A₂, D₃ and D₄ are control-chart constants determined by subgroup size and the assumed within-subgroup model. The constants must not be transferred between subgroup sizes. The range chart is examined first because an unstable spread undermines interpretation of the mean chart.

Attribute charts

Attribute data classify units or count events.

• p chart: fraction nonconforming when sample size may vary.
• np chart: number nonconforming when sample size is constant.
• c chart: number of defects when the area or opportunity is constant.
• u chart: defects per unit when the number of opportunities varies.

For a p chart with estimated average fraction p̄ and sample size n:

The limits are truncated to [0, 1]. The binomial model assumes suitable classification, independent units and a stable underlying fraction. A defect is not necessarily a defective unit: one unit may contain several defects.

Interpreting signals

When a chart signals, stop and identify the process condition at that time. Check maintenance records, material lots, operator assignment, software changes, measurement equipment and environmental conditions. Correcting the immediate cause is not enough; the organization should decide whether the cause can be prevented, detected earlier or incorporated into a revised standard process.

Do not recalculate limits merely to hide an out-of-control point. Limits may be revised after a documented process improvement, removal of verified special-cause data, and confirmation that the new process is stable.

5. Process capability

Capability analysis compares the natural variation of a stable process with specification limits. It should generally be performed only after the process is statistically stable and the measurement system is adequate.

The potential capability index is:

where σ is the within-process standard deviation. Cₚ measures the width of the specification interval relative to six standard deviations; it does not measure centring.

The centred capability index is:

where μ is the process mean. Cₚₖ is limited by the specification that is closer to the process mean. If the process is centred, Cₚ and Cₚₖ are equal; if it is off-centre, Cₚₖ is smaller.

For a one-sided specification, use the relevant one-sided index, such as (USL − μ)/(3σ) for an upper limit. Capability indices are dimensionless because numerator and denominator have the same units. They assume that σ represents the relevant stable process variation and that the distributional interpretation is appropriate. Strong skewness, mixtures or autocorrelation may make normal-based indices misleading; transformation, percentile methods or non-normal capability methods may be required.

Worked example

A filling process has specifications of 498 g to 502 g. A stable study estimates μ = 500.8 g and σ = 0.5 g.

The process spread is potentially adequate, but the mean is too close to the upper limit. Centring the process downward may improve performance, but the organization should confirm that the adjustment is technically safe and does not increase variation. Capability targets such as 1.33 or 1.67 are organizational or customer requirements, not universal laws.

6. Acceptance sampling

Acceptance sampling uses a sample from a lot to decide whether to accept or reject the lot. It is a disposition procedure, not a substitute for process control. A rejected lot may contain acceptable items, and an accepted lot may contain defects.

For an attributes plan, a single-sampling scheme is often described as (n, c): inspect n units and accept the lot if the number of nonconforming units is at most c. If the lot fraction nonconforming is p and the sample is treated as independent, the probability of acceptance is approximately:

Here C(n,x) is the number of combinations of n items taken x at a time. For finite lots sampled without replacement, the hypergeometric distribution is more appropriate.

The producer’s risk, α, is the probability of rejecting a lot of acceptable quality. The consumer’s risk, β, is the probability of accepting a lot of unacceptable quality. An operating characteristic (OC) curve shows Pₐ(p) across possible quality levels. A plan should be selected using explicit acceptable quality level (AQL), limiting quality level (LQL or LTPD), risks, lot size, inspection cost and consequences of failure—not simply by choosing a convenient sample size.

Inspection can be 100 percent, sampling-based or replaced partly by certification and process evidence. Destructive tests, expensive inspections and high-volume commodities often justify sampling, whereas safety-critical characteristics may require stronger controls, traceability and regulatory compliance.

7. Integrated application in supply-chain operations

Consider an illustrative case involving Northstar Cold Distribution, a regional distributor of pharmaceutical products. Customer complaints show that some deliveries arrive above the permitted temperature range.

Background and problem

The company records temperature excursions by route, carrier, vehicle and loading shift. Management initially proposes inspecting every delivery, but this would detect failures after they occur and would not explain their causes.

Analysis

A check sheet reveals 68 excursions in one month. A Pareto chart shows that 41% occur during loading, 29% during vehicle waiting time and the remainder during transit. A histogram of loading duration is bimodal, suggesting that two operating conditions are being combined. Stratification shows that the longer mode occurs mainly on the evening shift.

A cause-and-effect diagram identifies possible causes: delayed dock access, door-open time, sensor placement, inadequate pre-cooling, staffing and carrier procedures. A p chart is used weekly to monitor the fraction of deliveries with excursions. One point exceeds the upper control limit after a refrigeration-maintenance change, indicating a likely special cause.

Solution and outcome

The company verifies sensor calibration, restores the approved maintenance procedure, introduces a pre-cooling checklist and records loading-to-departure time. A control chart confirms that the process becomes stable. Capability analysis of departure temperature is then performed against the permitted limits. Because the temperature distribution is skewed and bounded by operating conditions, the team supplements Cₚₖ with empirical percentiles and excursion rates.

Afterward, the organization changes carrier contracts to require electronic temperature records, assigns responsibility for dock delays and audits the process monthly. The key outcome is not merely a lower complaint rate; it is a controlled process with traceable evidence and a defined response to signals.

Learning points

1. Data must be stratified before conclusions are drawn.
2. Control charts identify process signals; they do not identify causes automatically.
3. Capability analysis is meaningful only after stability and measurement adequacy are established.
4. Acceptance inspection alone would not have prevented the excursions.
5. Quality methods should be connected to ownership, corrective action and supplier governance.

8. Implementation guidance and limitations

A practical SQC programme should:

1. Translate customer and regulatory requirements into measurable characteristics.
2. Validate the measurement system.
3. Map the process and identify critical control points.
4. Collect time-ordered, traceable data.
5. Use stratification and basic quality tools to locate patterns.
6. Establish control charts with justified sampling and rational subgroups.
7. Investigate special causes without tampering.
8. Assess capability only for a stable process.
9. Select acceptance plans using risks and economic consequences.
10. Document decisions, corrective actions and review dates.

Important limitations include non-random sampling, autocorrelation, changing specifications, mixtures of populations, small samples, poor measurement resolution and misuse of normality assumptions. Statistical significance is not the same as operational importance. A small defect rate may be unacceptable in a safety-critical application, while a larger rate may be tolerable in a low-risk service if agreed contractually.

Digital sensors, manufacturing-execution systems and transport platforms now enable near-real-time charts and automated alerts. Artificial intelligence can help classify images, detect unusual patterns and prioritize investigations. However, automated alarms can reproduce biased labels, generate false signals and obscure responsibility. Organizations should retain data lineage, access controls, human review, cybersecurity and audit trails. Quality decisions affecting safety, employment or supplier status require explainable criteria and documented governance.

Chapter Summary

Statistical Quality Control uses data to understand and manage process variation. Common causes belong to the current system; special causes are unusual and actionable. Descriptive statistics summarize centre and spread, while sound sampling and measurement systems establish trustworthy evidence. Check sheets, Pareto charts, histograms, cause-and-effect diagrams, flowcharts and scatter plots help collect, organize and investigate quality information.

Control charts distinguish statistical stability from random fluctuation and should be designed around the data type and rational subgroup. Specifications express external requirements, whereas control limits describe process behaviour. Process capability indices such as Cₚ and Cₚₖ compare stable process variation and centring with specifications, but their assumptions must be checked. Acceptance sampling makes lot decisions while explicitly balancing producer’s and consumer’s risks; it does not replace prevention or process control. Effective application links statistical signals to investigation, corrective action, supplier management, traceability and ethical governance.

Key Terms

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

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

Specification limit: A customer, design, contractual or regulatory requirement.

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

Statistical stability: A condition in which only common causes are evident.

Sample mean: The arithmetic average of observed sample values.

Standard deviation: A measure of dispersion in the original measurement units.

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

Check sheet: A structured form for consistent data collection.

Pareto chart: A descending bar chart ranking categories by frequency or impact.

Histogram: A frequency display for grouped numerical data.

Cause-and-effect diagram: A structured tool for organizing possible causes of an effect.

Control chart: A time-ordered chart with a centre line and control limits.

Process capability: The ability of a stable process to meet specifications.

Cₚ: Potential capability index based on specification width and process spread.

Cₚₖ: Capability index incorporating process centring.

Acceptance sampling: Sampling used to decide whether to accept or reject a lot.

Operating characteristic curve: A curve showing probability of lot acceptance at different quality levels.

Producer’s risk: The probability of rejecting acceptable quality.

Consumer’s risk: The probability of accepting unacceptable quality.

Review Questions

Short Answer Questions

1. Distinguish between common-cause and special-cause variation.
2. Why are specifications different from control limits?
3. What information does a Pareto chart provide?
4. Why should a process normally be stable before capability is calculated?
5. What is the purpose of an acceptance-sampling plan?

Long Answer Questions

1. Explain how check sheets, histograms, Pareto charts and cause-and-effect diagrams can be used as a connected investigation sequence.
2. Describe the design and interpretation of an X̄–R control-chart system.
3. Explain the assumptions and interpretation of Cₚ and Cₚₖ.
4. Compare variable control charts with attribute control charts, including suitable supply-chain examples.
5. Discuss the advantages and limitations of acceptance sampling compared with prevention and 100 percent inspection.

Analytical / Application Questions

1. A process is stable with USL = 105, LSL = 95, μ = 103 and σ = 1. Calculate Cₚ and Cₚₖ, then recommend an improvement priority.
2. A warehouse defect Pareto chart shows that picking errors account for 52% of all errors, but the category combines three different picking zones. Explain how you would redesign the data collection and investigation.
3. A control chart shows a sudden out-of-control signal immediately after a supplier change. Design a step-by-step investigation that separates supplier effects from measurement and process effects.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a control limit from a specification limit?

A. A control limit is set by the customer, while a specification limit is calculated from process data.

B. A control limit describes expected process behaviour, while a specification limit states an external requirement.

C. Both terms describe exactly the same boundary.

D. A specification limit is used only for attribute data.

Correct Answer: B. A control limit describes expected process behaviour, while a specification limit states an external requirement.

Question 2

A process has USL = 20, LSL = 10 and σ = 1. What is Cₚ?

A. 0.83

B. 1.00

C. 1.67

D. 3.33

Correct Answer: C. 1.67

Question 3

Which tool ranks defect categories from largest to smallest and often includes a cumulative-percentage line?

A. Histogram

B. Pareto chart

C. Scatter plot

D. Flowchart

Correct Answer: B. Pareto chart

Question 4

What should normally be established before calculating process capability?

A. A stable process and an adequate measurement system

B. A larger specification range

C. A 100 percent inspection plan

D. A cause-and-effect diagram with no data

Correct Answer: A. A stable process and an adequate measurement system

Question 5

Which chart is generally appropriate for the fraction of nonconforming units when sample size varies?

A. p chart

B. c chart

C. np chart

D. X̄ chart

Correct Answer: A. p chart

Question 6

In an acceptance plan (n, c), what does c represent?

A. The number of lots inspected

B. The maximum number of nonconforming units allowed for acceptance

C. The process standard deviation

D. The customer complaint rate

Correct Answer: B. The maximum number of nonconforming units allowed for acceptance

Question 7

If Cₚ is 1.33 but Cₚₖ is 0.80, what is the most likely interpretation?

A. The process has no variation.

B. The process is centred exactly between the specifications.

C. The process spread may be adequate, but the process is poorly centred.

D. The measurement system is necessarily invalid.

Correct Answer: C. The process spread may be adequate, but the process is poorly centred.

Question 8

What is the primary purpose of a check sheet?

A. To calculate control limits automatically

B. To collect observations consistently at the point of work

C. To prove that one cause produces a defect

D. To replace sampling design

Correct Answer: B. To collect observations consistently at the point of work

Question 9

Which situation is most likely a special cause?

A. Normal small differences among routine production cycles

B. Ordinary variation in material properties within the established process

C. A broken tool causing an abrupt change in dimensions

D. Stable variation around the process mean

Correct Answer: C. A broken tool causing an abrupt change in dimensions

Question 10

What does consumer’s risk mean in acceptance sampling?

A. Rejecting a lot that is actually acceptable

B. Accepting a lot that is actually unacceptable

C. The cost of measuring every item

D. The risk that a control chart has too many points

Correct Answer: B. Accepting a lot that is actually unacceptable