There are more control charts than most people need, and the choice looks intimidating until you realise it comes down to one question asked twice. Pick the wrong chart and your limits are wrong; pick the right one and the rest is routine. Here is the decision path.
First question: what kind of data do you have?
Everything forks here.
- Variables data is measured on a continuous scale — length, weight, temperature, diameter. Anything you read off an instrument in units.
- Attributes data is counted — pass or fail, number of defects, present or absent. Anything you tally rather than measure.
Variables data carries more information per point, so its charts detect change faster with smaller samples. If you can measure rather than count, do.
Variables charts: match the chart to your subgroup size
Variables charts always come as a pair — one chart for the average (where the process is centred) and one for the spread (how much it varies). Always read the spread chart first: if the spread is unstable, the limits on the averages chart are not valid.
- I-MR (Individuals and Moving Range) — use when you measure one item at a time (n = 1): slow, expensive or batch processes where a single reading per period is natural.
- X̄-R (average and range) — the workhorse, for small subgroups of 2 to 9 consecutive parts. The range is a simple, robust measure of spread at these sizes.
- X̄-S (average and standard deviation) — for larger subgroups of 10 or more, where the standard deviation is a more efficient spread measure than the range.
In practice most shop-floor charting uses X̄-R with subgroups of four or five.
Attributes charts: defectives or defects?
If you are counting, ask a second question: are you counting whole items that fail, or individual flaws?
- A defective is a whole item that fails — it passes or it does not. Use a p-chart when your sample size varies, or an np-chart when it is constant.
- A defect is a single nonconformity, and one item can carry several — think blemishes on a painted panel. Use a c-chart when the unit of inspection is constant, or a u-chart when it varies.
So the four attribute charts are really just two questions: defectives versus defects, and constant versus varying sample size.
The decision in one line
Measured data, one at a time goes to I-MR. Measured in small subgroups goes to X̄-R. Measured in large subgroups goes to X̄-S. Counting failed items goes to p or np. Counting flaws goes to c or u.
Before you plot anything: rational subgrouping
The right chart still needs the right data. The principle of rational subgrouping is to form subgroups so that the variation within a subgroup is only common-cause noise, and special causes show up as differences between subgroups. In practice: sample consecutive parts from one stream over a short interval, keep one machine or cavity per chart, and collect at least 20 to 25 subgroups before you trust your limits. Mix two streams into one subgroup and you inflate the limits until the chart can no longer see real problems.
Get the data structure right and the chart choice is mechanical. Get it wrong and even the correct chart will lie to you.
The SPC Control Chart Toolkit has a tab for each of these charts and a Start-Here guide that walks the same decision — pick the tab, enter your data, and the control limits and chart build themselves. New to reading the output? Start with Control charts explained.