Re: Anderson's "gum-pigment ratio test" (fwd)

Wayde Allen (allen@boulder.nist.gov)
Tue, 09 Jun 1998 07:40:08 -0600 (MDT)

On Mon, 8 Jun 1998, Jeffrey D. Mathias wrote:

> >From experience as a research physicist for eleven years, be very
> careful when using ANY of the statistical approaches.

I don't think anyone is suggesting otherwise. My whole point was that one
has to think carefully about ones testing methodology, and the conclusions
drawn.

> The assumptions
> and boundary conditions are critical. I say this for the benefit of
> anyone wishing to experiment with any new process.

Exactly.

> The best method is as Judy suggests. Investigate one parameter at a
> time.

But like I said before, don't ignore testing for correlation between
variables. This is a very common mistake.

> The multitude of variations can be reduced by making some careful
> assumptions to eliminate variables.

Yes, and that is exactly what the statisticians mean by the term
"experimental design".

> For example, include all the
> ingredients necessary to make a gum print. It's nonsense to leave
> something out.

Noone said otherwise.

- Wayde
(wallen@boulder.nist.gov)