Attribute Capability Analysis
Load Poisson sample data Load binomial sample dataAttribute capability analysis evaluates process quality when the result is recorded as a count rather than as a continuous measurement. Numiqo provides a Poisson capability analysis for defects per unit and a binomial capability analysis for defective units.
Poisson or binomial capability?
| Question | Poisson | Binomial |
|---|---|---|
| What is counted? | Individual defects | Defective units |
| Can one unit contribute more than one event? | Yes, one unit can contain several defects | No, every unit is classified once as acceptable or defective |
| Typical result | Defects per unit (DPU) | Proportion or percentage defective |
| Control chart | U chart | P chart |
Poisson capability analysis
Use Poisson capability when you count how many defects occur in each inspected sample. For example, a coated metal roll can contain several scratches, inclusions, or coating gaps. The sample size can be the number of inspected units or an exposure such as meters, square meters, hours, or pages.
The Poisson example contains the number of surface defects found during 30 production shifts and the corresponding square meters of coated sheet that were inspected. The inspected area varies between shifts. The example target is 0.025 defects per square meter.
Binomial capability analysis
Use binomial capability when every inspected unit has only two possible classifications: acceptable or defective. Count the defective units in each sample together with the total number of inspected units.
The binomial example contains defective sealed packages and the total packages inspected during 30 production shifts. The sample sizes vary between shifts. The example target is 3% defective, which is entered as 3 in the calculator.
Variable sample sizes
Both methods support a constant sample size or a column containing a different sample size for every row. Numiqo calculates the overall rate from the total count divided by the total sample size. This correctly gives larger samples more weight than smaller samples.
How to interpret the results
First inspect the U chart or P chart. An unstable process should be investigated before its capability is used to predict future performance. Then compare the observed rate and its 95% confidence interval with the maximum acceptable target. If the upper confidence bound is below the target, the data provide strong evidence that the process meets the target. If only the observed mean is below the target, more data may be required.
Return to the process capability calculator to analyze your own data.