Attribute control charts: p, np, c, u
Charts for defectives and defects, varying sample sizes, and Laney's correction for very large samples.
Charts for defectives (binomial)
The p chart's limits get narrower for larger samples, so limits step up and down when sample sizes vary. The np chart needs a constant sample size.
Charts for defects (Poisson)
A negative lower limit is set to zero.
Laney p' and u' charts
With very large samples (thousands of units), ordinary p chart limits become extremely tight and nearly every point signals, because real-world proportions vary more than the binomial model predicts (overdispersion). Laney's charts (2002) widen the limits using the moving range of standardized points. Use them when the ordinary chart flags most points.
G and T charts
For rare events, counting events per period gives mostly zeros. Instead plot the number of opportunities (G chart) or the time (T chart) between events. A long run means improvement.
Questions and answers
How big should samples be for a p chart?
Large enough to see a few defectives in most samples. A common rule is ; with 1% defective that means samples of at least 500.
Should I use a p chart or a u chart for "defective boards per day"?
Defective boards (pass/fail per board) use a p chart. Defects counted on boards (several per board possible) use a u chart.
