What Are Western Electric Rules — and Which Ones Should You Actually Enable?
The four Western Electric rules catch process shifts your control limits miss. But running all four on every characteristic creates noise. Here's how to decide which rules to turn on — and when.
Your Xbar-R chart has been running for three weeks. No points outside the control limits. Your quality engineer calls it “in control” and moves on.
Then a customer complaint arrives. The dimension was drifting for ten days before it crossed a spec limit — and the chart never flagged it.
That’s the gap the Western Electric rules exist to fill. They catch patterns inside the control limits that signal a process shift before you get a nonconformance. But they also generate false alarms. Run all four rules on every characteristic and your operators will start ignoring the alerts — which defeats the purpose.
This article explains what each rule detects, what it costs in false alarms, and how to decide which ones to enable for a given characteristic.
What the Western Electric rules are
The Western Electric rules (also called WECO rules) are four pattern-detection tests for control charts. They were codified by a committee of the Western Electric Company and published in the Statistical Quality Control Handbook in 1956. They’re designed to detect non-random patterns — evidence of a special cause — without waiting for a point to breach the 3σ control limits.
The rules work by dividing the control chart into three zones on each side of the centerline:
- Zone C: within 1σ of the centerline
- Zone B: between 1σ and 2σ
- Zone A: between 2σ and 3σ (just inside the control limits)
A process with only common-cause variation will distribute points across these zones in predictable proportions. The rules flag when that distribution breaks down.
The four rules, one at a time
Rule 1 (WE1): One point beyond ±3σ
The classic control chart signal. Any single point outside the upper or lower control limit.
What it detects: A large, sudden shift — a broken tool, a wrong material lot, a setup error, a measurement system failure.
False alarm rate: About 1 in every 370 points on a stable, normally distributed process. That’s the baseline most quality engineers are comfortable with.
When to use it: Always. This is the minimum viable SPC configuration. If you enable nothing else, enable WE1.
Rule 2 (WE2): 2 of 3 consecutive points beyond ±2σ on the same side
Two out of three consecutive points fall in Zone A or beyond, on the same side of the centerline.
What it detects: A moderate process shift — not large enough to push a single point outside the limits, but enough to consistently pull points toward one extreme. Think: a gradual fixture wear, a slow raw material drift, a temperature creep in a curing process.
False alarm rate: Roughly 1 in 155 points when evaluated alone. Combined with WE1, the pair generates a false alarm about every 130 points.
When to use it: For characteristics where moderate shifts matter — dimensions with tight tolerances, critical-to-function parameters on medical devices or automotive safety parts. The rule is sensitive enough to catch early drift without being so hair-trigger that it fires on normal process noise.
Rule 3 (WE3): 4 of 5 consecutive points beyond ±1σ on the same side
Four out of five consecutive points fall in Zone B or beyond, on the same side of the centerline.
What it detects: A small but sustained shift in the process mean. The kind of drift that takes a long time to produce a nonconformance but will eventually get there. Gradual tool wear on a turning operation. A slow change in ambient temperature affecting a dimensional measurement. A raw material property that’s trending within spec but moving.
False alarm rate: About 1 in 140 points on its own. This is where the combined false alarm rate starts to become a practical concern. Running WE1 + WE2 + WE3 together produces a false alarm roughly every 100 points.
When to use it: On high-volume processes where you have enough data to distinguish a real trend from noise, and where the cost of a missed small shift is high. Be cautious on short-run or low-volume processes — with fewer than 25–30 subgroups in your calibration window, WE3 will fire on patterns that are statistically meaningless.
When to think twice: If your process has natural autocorrelation — measurements that are correlated with the measurement before them — WE3 generates false alarms at a much higher rate than the theoretical 1-in-140. Autocorrelation is common in continuous processes (extrusion, coating, chemical batches). On those characteristics, WE3 and WE4 can become noise generators.
Rule 4 (WE4): 8 consecutive points on the same side of the centerline
Eight points in a row, all above or all below the centerline. No requirement about how far above or below — just consistent side.
What it detects: A sustained mean shift, even a small one. Also detects a reduction in process variation (a good thing, but still a change worth understanding). If your process mean has moved even slightly, the probability of eight consecutive points landing on the same side by chance is 1 in 256.
False alarm rate: About 1 in 256 points on its own. With all four rules running together, the combined false alarm rate rises to roughly 1 in 92 points — about four times the false alarm rate of WE1 alone, according to the Western Electric handbook.
When to use it: On processes with a long, stable run history where you want maximum sensitivity to mean shifts. Also useful when you’re monitoring a characteristic post-corrective action — you want to know quickly if the fix held.
When to think twice: Same autocorrelation caveat as WE3. And on any process where the centerline itself shifts legitimately between setups (different operators, different fixtures, different material lots), WE4 will fire constantly. If your process is inherently multi-modal, WE4 is the wrong tool.
The false alarm math
Here’s the practical summary, sourced from the Western Electric handbook and Wikipedia’s treatment of the rules:
| Rules enabled | Approx. false alarm rate |
|---|---|
| WE1 only | 1 in 370 points |
| WE1 + WE2 | ~1 in 130 points |
| WE1 + WE2 + WE3 | ~1 in 100 points |
| All four (WE1–WE4) | ~1 in 92 points |
These rates assume a stable, normally distributed process. Real processes are neither perfectly stable nor perfectly normal. Skewed distributions, autocorrelation, and non-normal tails all inflate the false alarm rate — sometimes dramatically.
The implication: on a high-volume process running 500 subgroups a month with all four rules enabled, you should expect roughly five to six false alarms per month from random variation alone. If your team investigates every alert, that’s five to six wasted investigations. If they start ignoring alerts, you’ve broken your SPC system.
Three scenarios and what to enable
Scenario 1: High-volume, stable process, tight tolerance
A 200-person automotive stamping plant monitoring a critical hole diameter on a safety bracket. Running 50 subgroups per shift, three shifts per day. The characteristic feeds a PPAP submission with a Cpk requirement of 1.67.
Recommended: Enable all four rules. The high volume means false alarms are manageable in absolute terms (a few per week), and the PPAP context means you want maximum sensitivity to drift. WE3 and WE4 will catch the gradual tool wear that would otherwise erode your Cpk before the next capability study.
Scenario 2: Short-run, low-volume process
A 40-person medical device contract manufacturer running a machined titanium component. Lot sizes of 25 parts. The SPC chart has 20–30 subgroups before a new lot resets the process.
Recommended: Enable WE1 and WE2. Skip WE3 and WE4. With only 20–30 subgroups, the pattern-detection rules don’t have enough data to distinguish a real trend from noise. WE3 needs at least 5 consecutive subgroups to fire; WE4 needs 8. On a 25-part lot, a single WE4 trip consumes a third of your data. The false alarm rate in practice will be much higher than the theoretical figures.
Scenario 3: Continuous process with autocorrelation
A pharmaceutical contract manufacturer monitoring fill weight on a liquid filling line. Measurements are taken every 30 seconds. Adjacent measurements are correlated — a slightly heavy fill is likely to be followed by another slightly heavy fill before the system self-corrects.
Recommended: Enable WE1 only, or WE1 + WE2 with caution. Autocorrelation inflates the false alarm rate for WE3 and WE4 significantly. If you need trend detection on an autocorrelated process, consider a CUSUM or EWMA chart instead — they’re designed for it. The Western Electric rules assume independence between observations.
A note on regulated industries
If you’re operating under ISO 13485:2016, 21 CFR Part 820 (the FDA’s Quality Management System Regulation), or IATF 16949, your SPC configuration is part of your quality system documentation. The rules you enable — and why — should be defensible to an auditor.
That doesn’t mean you must run all four rules. It means you should have a rationale. “We enable WE1 and WE2 on this characteristic because it’s a short-run process and WE3/WE4 generate noise that operators have learned to ignore” is a defensible rationale. “We enabled all four rules because that’s the default” is not — especially if your investigation records show a pattern of closed-with-no-cause alerts.
Document your rule selection in your control plan or SPC procedure. If you change the configuration, document why.
What to do today
Pick one characteristic you’re currently monitoring with SPC. Pull the last 30 days of violation records. Count how many alerts were closed with “no assignable cause found.”
If that number is more than 20% of total alerts, you have a false alarm problem. Start by disabling WE3 and WE4 on that characteristic and see if the signal-to-noise ratio improves over the next month.
If you have zero alerts in 30 days on a high-volume characteristic, you may have the opposite problem — all four rules disabled, or a calibration window so wide that the control limits are too loose to catch anything. Check your rules_enabled configuration and your calibration window subgroup count.
The goal is a chart your operators trust. A chart that cries wolf gets ignored. A chart that never speaks misses the shifts it was built to catch.
Opwise’s SPC module lets quality engineers configure Western Electric rules 1–4 independently per characteristic, directly from the control plan. Rule changes are logged with a timestamp and user, so your audit trail reflects the current configuration — not just the default.