> Dutifully running this test one day with fresh paper, on a whim I tried it
> on preshrunk paper. Oops, the result was completely different. I
> understood in a trice that any variable -- length of soak, order of coat,
> etc. -- would also be completely different. Later it dawned on me that the
> dichromated emulsion also changes the paper, ...
Not surprising. You've run into a very common misconception about the
so-called scientific method. Namely that one should only change a single
variable at a time. That is only partly true since such an approach gives
you no information about correlation effects between your test variables.
For instance if you want to test your process for the effect of:
volume of gum in emulsion
volume of dichromate sensitizer in emulsion
volume of pigment in emulsion
type of paper
thickness of coating layer
and that you choose to test only two values of each of these 5 variables.
You would have to run 32 tests to get a complete data set. Adding a sixth
variable increases this to 64 separate tests! If you allow for more than
two possible values for each variable the numbers get even bigger. It
doesn't take long before one simply can't perform all of the experiments
needed. The solution to this dilemma is through statitical experimental
design where several variables are changed simultaneously. Experimental
design is a dicipline in itself so I'll quit while I'm ahead here.
- Wayde
(wallen@boulder.nist.gov)