SPC Basics: Common vs Special Cause

SPC Basics: Common vs Special Cause (CSSGB BoK VI.A.1 — Analyze)

Quick Answer: Statistical process control (SPC) monitors a process with control charts so you can tell common-cause variation (stable system noise—improve the system) from special-cause variation (assignable events—find and remove the cause). Points beyond control limits or non-random patterns flag special causes; random scatter inside limits means the process is in statistical control.

Objectives of SPC

Walter Shewhart’s control chart idea is the backbone of the Control phase and of ongoing process management. SPC is not about “checking every unit against the print.” It is about predictability:

1. Detect change quickly — know when the process has shifted or become unstable.
2. Avoid over-adjustment — do not chase every wiggle that is only common cause (tampering increases variation).
3. Guide the right improvement type — special causes get local investigation; common-cause problems need process redesign, standard work, or capability projects.
4. Sustain gains after DMAIC so the improved process does not silently drift.
5. Communicate with data — one chart often ends debates about “whether quality got worse.”

In control means the process is stable and predictable within its current common-cause system. It does not automatically mean the process meets specifications (that is capability—BoK III.F).

Continuous vs Discrete Data

Chart families split first on what you measure.

Continuous data take values on a continuum (or a fine measurement scale). You estimate mean and dispersion and chart both (location chart + range/s chart).

Discrete data for SPC usually means:

• Defectives — each item is conforming or nonconforming (binomial setting) → p or np charts.
• Defects — countable flaws that can appear more than once per unit or per area (Poisson-like setting) → c or u charts.

Exam tip: “Percent defective” or “fraction nonconforming” → attribute (p/np). “Scratches per panel” or “defects per unit” → c/u. “Diameter in mm” → variables chart.

If you can measure the characteristic continuously, variables charts are usually more informative for the same sample size than go/no-go attribute charts.

Common-Cause vs Special-Cause Variation

Deming and Shewhart framed management action around two kinds of variation.

Common-cause variation

• Always present in the current process as designed and operated
• Many small, ordinary sources (material lot micro-variation, ambient conditions, operator micro-differences within standard method)
• Produces a stable, random pattern on a well-designed control chart
• Improving common-cause performance requires changing the system (method, equipment, design, training standard)—not hunting a one-off “who messed up?”

Special-cause variation (assignable cause)

• Not part of the usual system noise
• Linked to a specific event, shift, tool, batch, or method change
• Produces signals on the control chart
• First response: investigate, contain if needed, remove or control the cause, then restore stability

Over-control (tampering): adjusting the process for every common-cause wiggle adds extra variation. Under-control: ignoring special-cause signals lets problems grow. Analyze-level skill is matching the reaction to the type of variation.

How Variation Type Is Deduced from Control Charts

A control chart plots a statistic (individual, subgroup average, proportion, count) in time order with:

• Centerline (CL) — usually the process average of the plotted statistic
• Upper control limit (UCL) and lower control limit (LCL) — typically set at about ±3 standard errors of the plotted statistic (Shewhart 3-sigma limits), calculated from process data, not from specs

Signals of special-cause variation

Classic rules (Western Electric / Nelson-style patterns; exact set varies by organization—know the logic):

1. One point beyond UCL or LCL — strongest single-point special-cause signal
2. Run on one side of the centerline — e.g., 7–9 consecutive points above or below CL (a sustained shift)
3. Trend — e.g., 6–7 consecutive increasing or decreasing points
4. Two of three consecutive points in the outer third (beyond 2σ from CL on the same side)
5. Cycles or systematic patterns — alternating high/low, shift-to-shift zigzags that are not random
6. Hug the centerline unusually tightly — sometimes a calculation error, stratified subgrouping, or reduced variation worth confirming

If none of these appear and points bounce randomly inside the limits, treat the process as in statistical control: only common-cause variation is acting at a detectable level.

Worked interpretation sketch

An X-bar chart for fill weight runs for 30 subgroups. Points scatter around the centerline with none outside UCL/LCL and no long runs. Conclusion: common-cause system—predictable at the current mean and within-subgroup variation. If capability is poor, launch a system project (reduce σ or retarget μ), not a witch hunt for each subgroup.

Next week, three consecutive X-bar points sit above the UCL after a new supplier lot arrives. Conclusion: special-cause signal—quarantine product if needed, check the lot, measurement, and setup; do not average the bad points into “the new normal” until the cause is understood.

Control Limits vs Specification Limits (Again)

A point can be inside control limits and outside specs (stable but incapable). A point can be outside control limits yet still inside specs (unstable but currently “lucky”). Analyze both questions separately.

Practical SPC Workflow

1. Define the CTQ and operational definition; confirm measurement adequacy.
2. Choose data type and subgrouping (next sections).
3. Collect time-ordered data; compute CL and control limits from a baseline in control (or recalculate carefully after process change).
4. Plot ongoing points; apply rules; react to special causes.
5. When stable, use capability analysis if specs matter; improve the common-cause system if needed.
6. Keep the chart live in Control so gains stick.

Bottom Line for VI.A.1

SPC’s objective is stable, predictable processes through statistical separation of variation types. Continuous vs discrete data determine chart families. Common cause = inherent system noise → improve the system. Special cause = assignable signal on the chart → investigate and remove. Master reading the chart so you analyze variation correctly instead of guessing.