Learn · updated 2026-10-21
Which control chart should I use?
Two questions decide it. Are you measuring something (a length, a weight, a time) or counting something (defects, failed items)? And how many do you take per sample?
| Your data | Chart | Notes |
|---|---|---|
| Measurements, one at a time | I-MR | One reading per batch, shift or day. Guide |
| Measurements, subgroups of 2 to 9 | X̄-R | For example 5 parts measured every hour. |
| Measurements, subgroups of 10 or more | X̄-S | The standard deviation beats the range at this size. |
| Defective items (pass/fail), sample size varies | p | Plots the fraction defective, limits per sample. |
| Defective items (pass/fail), same sample size | np | Plots the number defective. |
| Defects (a unit can have several), same amount inspected | c | Scratches per panel, errors per form. |
| Defects, amount inspected varies | u | Plots defects per unit. |
Measuring
Measurements carry more information than counts, so measure when you can. With one value per time period use the I-MR chart. With rational subgroups (several parts made close together) use X̄-R up to 9 per subgroup and X̄-S from 10. The X̄-R limits use the constants A2, D3 and D4; for subgroups of 5, A2 = 0.577 and D4 = 2.114, as in NIST's table.
Counting
First decide what you count. A defective is a whole item that fails (pass or fail, binomial): use p or np. A defect is a flaw, and one item can have several (Poisson): use c or u. Then: if the sample size or inspection area changes from sample to sample, use the per-unit chart (p or u), whose limits move with each sample.
Common mistakes
- Averaging counts into an X̄ chart. Counts follow different distributions; use the attribute chart.
- Huge samples on a p chart. With thousands per sample the limits get so tight that everything looks out of control. A Laney p′ chart corrects for this (LimitLine Pro has it).
- Subgroups that mix sources. Five parts from five different machines in one subgroup hide the machine-to-machine difference inside the range.