The eight out of control tests on a Shewhart chart
Updated: 2026-08-04 · ISKRIT
A point outside a control limit is only the first of eight signals. The other seven detect patterns that stay inside the limits, and they are what make a control chart more sensitive than a simple tolerance check.
Zones
The band between the centre line and each control limit is split into three equal zones, each one sigma wide. Zone C is nearest the centre line, zone B is the middle, zone A is the outermost, up against the limit. The tests are written in terms of these zones on both sides of the chart.
The eight tests
| Test | Pattern | Detects |
|---|---|---|
| 1 | one point beyond zone A | a large sudden shift or an outlier |
| 2 | nine points in a row on one side of the centre line | a small sustained shift |
| 3 | six points in a row rising or falling | drift, tool wear, reagent ageing |
| 4 | fourteen points alternating up and down | over-adjustment, or two alternating sources |
| 5 | two of three consecutive points in zone A or beyond | a moderate shift |
| 6 | four of five consecutive points in zone B or beyond | a small shift |
| 7 | fifteen points in a row in zone C | stratification, or limits computed too wide |
| 8 | eight points in a row outside zone C on both sides | mixture of two distributions |
The two counter-intuitive ones
Tests 7 and 8 flag data that looks too good and data that avoids the middle. Fifteen consecutive points hugging the centre line is not a well behaved process, it usually means the limits were calculated from a period that contained extra variation, or that subgroups were formed across sources that cancel each other. Test 8 is its mirror: points sit away from the centre on both sides, which is the signature of two processes plotted on one chart, two shifts, two instruments, two operators.
The cost of running all eight
Each test has its own false alarm probability, and they add up. Test 1 alone gives roughly one false signal in 370 points. Enabling all eight brings the combined rate to something closer to one in 90. That is a deliberate trade: more sensitivity to real shifts, more investigations that find nothing. A laboratory running daily control samples should decide which tests it applies and write that decision into its procedure, rather than switching them on and off after the fact.
Questions
Do the tests apply to range charts too? Test 1 always does; the run tests are normally applied to the averages chart, since the range distribution is skewed and the zone logic assumes symmetry.
What comes after a signal? Find the special cause. Recomputing limits to make the signal disappear is the one response that guarantees the chart stops working.
All eight are checked automatically, with the indices of the points involved, in the control chart tool.
ASTM D4253 and D4254: maximum and minimum index density
Relative density Dr and density index Id are not the same number
USCS classification step by step: ASTM D2487
Cu and Cc: what the two gradation coefficients actually tell you
Dual symbols in USCS: the 5 to 12 percent fines band
EN ISO 14688-1 against USCS: two different size ladders
AGS4 file validation: what the rules actually check
AGS4 syntax: the five rules that break a delivery
ASTM D2435 oedometer test: from the curve to Cc, mv and Eoed
Converting oedometer modulus to Young’s modulus
EN ISO 17892-5 and ASTM D2435: where the two diverge
ASTM D1196 plate load test: modulus and subgrade reaction
DIN 18134: Ev1, Ev2 and the compaction ratio
Ménard pressuremeter modulus: ASTM D4719 and EN ISO 22476-4
Shewhart control chart constants: A2, A3, B3, B4, D3, D4, d2, c4
Repeatability and reproducibility limits: where 2.77 comes from
How rebar cutting plans are optimised: the cutting stock problem
Cutting waste in rebar fabrication: what percentage is normal
Earthwork volume from survey points: grid, TIN and cross sections