hmm, must not have hit post earlier, so here is a redo.

harryn, an extension to your factorial analysis for data analysis is "Design of Experiments" or DoE http://en.wikipedia.org/wiki/Design_of_experiments (saddly, the wikipedia article is of a lesser quality than most and does not do the process full justice) DoE is a tool to help plan experiments with multiple interconnected variables without doing a full sweep of every possible combination. DoE uses statistics to tease out influencing coeficients between variables and can evaluate how each variable affects other variables, even when multiple things are changed with each sample. As mentioned earlier, the math is a bear, but programs like minitab are realy the only way to tackle this.

Unlike factorial analysis, which is primairly an analysis tool, DoE is used to determine which variables are changed on each sample and statistically reduces the amount of samples needed. The real magic is that DoE can take the influencing coeficients it generates uses them to spit out an optimized formula automatically.

Aside from academic use, I have used this to sort out a year long build/tweak/redesign cycle on a new spring clutch design. The design was broken into 13 variables and I believe we ran 50 or so samples (it may have been a little more, but I'm not sure, either way it was not a lot of samples for 13 varriables). Not surprisingly, none of the samples worked, but that wasn't the point. Once the computer crunched the data, we built up the "optimal" result and it worked perfectly. This was a problem that dozens of engineers could not solve with their best efforts in measuring, tweaking, designing, and guesing, but a 2 day experiment solved everything.

Now the drawbacks, most DoE trials use linear interpolation and assume straight line relationships betweeh the high and low test points used for each varriable. 3 and 4 point per variable tests are possible, but the required samples begin to increase dramatically. Surprisingly, this is rarely a problem, but EG may be a bit unique. The other problem is the requirement to break the test down into individual independant variables. While this may not be possible for some of the agregate tests, the "other" factors like epoxy, and addins are certainly a candidate.

Finally, the most obvious problem is actually creating and testing 15-20 or more unique sample formulations (preferrably with 5 individual specimens each)

If anyone is thinking of an "extensive" test procedure, this is something to look at because you can get a lot more bang for your experimenting buck by correctly choosing which variables to change with each test.

John