## Criterion / Action

### Grubb’s *G* and Cochran’s *C* statistics

If *G* (Grubb) and *C* (Cochran) are:

- Less than or equal to their 5% critical value –
*accept* - Exceeding their 5% critical value but smaller than their 1% critical value – straggler (*)
- Exceeding their 1% critical value – “outlier” (**)

### Mandel *h*-statistics

If a measured datum has an *h*-statistic:

- Has a different character from all other data
- Is very distinct in value from the other data
- Depends systematically on the level

then the measurements made by that laboratory should be examined in, in particular.

### Mandel *k*-statistics

If a datum has a *k*-statistic which is:

- larger – this indicates that the lab has a worse repeatability
- smaller, indicating that the lab may have rounded their values too much, e.g.

then the measurements made by that laboratory should be examined in, in particular.

### Chart

Action must be taken if two consecutive results on a x-chart lie outside the warning limit.

### Range (*R*-) Chart

Action must be taken if two consecutive results on a R-chart lie outside the warning limit.

### T-test

If *t *<*t _{max}* of the Student t-test, the new measurements, combined older, where the new “pooled” average is calculated as:

where

*k, j*are the number of measurements: new (

*n*) resp. old (

*o*).

### F-test

If *F* <*F _{max}* of the F-test, the new measurements can be combined with the older, where the new “pooled” standard deviation is calculated as:

### Process capability (mean) (standdev)

When making a process capability test based on the mean:

If *M* <± 15%: Process is accepted

When making a process capability test based on the standard deviation:

If *C _{pk }*<1.0: Process must be improved immediately

If *C _{pk}* is 1.0 – 1.3: Process (just) meets specification.

If *C _{pk}*> 1.0: Process is accepted

________________________________________________________________________________________

## Leave a Reply