[Stoves] Laboratory Comparison - but industrial standards for testing first!

[Robert Penn Taylor]

Charlie,

I agree with what you say in your full post, and in particular the parts 
about normal distributions and design of experiments (quoted below for 
emphasis). I think a basic understanding of design of experiments could 
alleviate some significant issues I've seen in data presented at ETHOS 
and elsewhere. Taguchi arrays are pretty easy to use and could open up 
the data from these experiments to ANOVA and simple response surfaces. I 
suspect the interaction terms are significant in many of the stove 
processes, rendering the "hold everything constant but one variable" 
methodology questionable.

-Penn Taylor


Charlie Sellers wrote:
> ... 
> Certainly a huge part that I see is missing is the use of the concept
> of a "normal distribution" - people are citing the standard deviation
> of 3 points, throwing out points arbitrarily, and worse without ever
> taking all the accumulated data on a stove and checking to see
> whether the distribution is such that ANY of the data can be trusted
> (Step 3).  Try not to average data or throw any away - you are losing
> information when you do this.
> 
 > ...
> 
> Thanks for tolerating an outside view.  Finger pointing about
> favoritism and mis-testing goes away completely when the testing is
> accepted by all - and concepts like guage R&R, normal distribution,
> and design of experiments (to examine the importance of different
> variables, as well as their interaction) are what it takes to get
> there.
> 
> Charlie Rogue Stover