RE: SPC - independent data

["Paul Hollingworth" <ph@4gm.com>]

Like auto-correlation, the concept of independent data is a simple
one.
But as is so often the case, being simple does not make it easy to
define.

Would we not have to understand all the interacting forces in a causal
system before we could empirically claim 'independence'?

Alternatively, if we used a statistical method to calculate the
probable degree of independence, what theory would it rely on?

The fact that we can gain insight into process performance despite
non-independence of data it what is valuable.
And if we have reason to believe that data ARE auto-correlated then
with the right questions, we can gain even more insight.

Which brings me back to the point of SPC (setting aside the 'report
card' charts that QC systems sometimes require).
Shewhart talked of 'economic quality'. This calls for the right
questions. For example:
* To what extent might the future be like the past?
* Is some kind of action called for?
* Did the action taken have the expected beneficial effect?

IMHO we must not let the data distribution or causal system become an
excuse to avoid asking these important questions - whatever methods we
use.

PH