Control Chart Maker
Individuals and moving range, c and p charts from numbers you paste, with the Western Electric run rules switched on one at a time. A control chart separates routine variation from real signals, so you stop reacting to noise.
It cuts both ways. Every rule you switch on catches more real changes and also raises more false alarms, and this page tells you roughly how many to expect.
Nothing you type here leaves your browser. Every calculation runs in this page. The page itself counts visits, like every page on this site, but never what you enter.
1. Your data and the rules to apply
Leave blank to use every point. Set it to the stretch before a change, and later points are judged against the old limits.
Paste a column from a spreadsheet. Columns may be separated by tabs, commas or spaces. A text header on the first line is skipped. Do not use thousands separators.
2. The chart
3. Every point the rules flagged
The arithmetic, with your numbers
Shown so you can check it, or redo it in a spreadsheet. The spreadsheet formulas assume your data starts in cell A2.
What this does not tell you
- Every rule you switch on adds false alarms. On a stable process with normally distributed data, rule 1 alone flags about one point in 370 by chance. With all four rules on, the commonly quoted figure is roughly one in 90. These are approximations for the ideal case, not exact rates for your data. Switch on the rules you will act on, and expect a flag now and then from a process that has not changed.
- Limits calculated from data that contains special causes are too wide. An unusual point inflates the average moving range or the centre line, which widens the limits and hides other signals. If a flagged point has a known cause, fix the cause, then set the baseline to a stretch without it.
- A point inside the limits is not proof nothing changed. A small, steady shift can take dozens of points to show up, and the run rules exist because rule 1 is slow to catch it. No flag means no evidence of change, not evidence of no change.
- Individuals limits assume roughly normal, independent data. Skewed measurements such as cycle times or waiting times give too many flags on the long side and too few on the short side. Readings that drift together from one to the next, such as a tank temperature taken every minute, give limits that are too narrow.
- Control limits are not specification limits. The limits describe what the process does. The specification says what the customer needs. A process can be in control and still make every part out of tolerance, or be out of control and still make every part good.
- The c and p charts assume each defect or defective is independent. The c chart also assumes every sample is the same size. When defects arrive in clusters, such as a bad batch or a bad shift, the real spread is wider than the formula expects and the chart flags too often.
The rest of the Lean Six Sigma toolkit
- Process Capability Calculator: Cp, Cpk, Pp, Ppk: Cpk with its confidence interval, a Cpk against Ppk drift check and a normality test.
- Sigma Level Calculator: DPMO, yield and how sure you can be: DPMO and sigma level with the 95% interval, and whether a kaizen really cut defects.
- Takt Time and Line Balance Calculator: Takt time, line balance and the bottleneck, with stations that miss takt on slow cycles.
- OEE Calculator: Availability, Performance, Quality Losses: OEE with each loss in minutes, and the one fix worth most.