Alternative accuracy testing


Wetside, Oregon
With the discussion on ladder testing, and other things, I wanted to share another way to test accuracy especially when you have questions about what is better... X or Y... Or Z?

It's called by any number of names, but usually it is design of experiments (doe), taguchi orthogonal array / matrix, or ANOVA.

Why should I care about it? It is a method used to test multiple variables simultaneously, with high confidence in the results, without having to try every single combination.

Example: I want to test out two different primers, what lube is best (lla or moly-bee), crimp or no crimp, different makes of brass (hxp vs win), and powder charge of bullseye powder, to see what's most accurate. Normally this is 2x2x2x2x2 = 32 different combinations... using taguchi, we can test this in 8 runs with a factorial array. This saves time, money, and the wear and tear.

I'll post more as I go along with my experiment. There's a few posts on the interwebs... Search taguchi reloading, or design of experiments reloading. I found this when I was working as an engineer and doing a lot of 6sigma stuff... Currently I do a bunch of statistical analysis... But I am far from an expert... More interested in sharing knowledge.



Last edited:


Active Member
Interesting method and article!
Have you ever tried to shoot the combination of quantitative/qualitative variables that were predicted to give the worst results? That would also be interesting, to evaluate the predictive model.


Wetside, Oregon
I have not! :) That would surely give great confidence in the study. But I think having it be a blind study where you have someone else shoot, so there's no undue influence might be another way to verify results.

I've used average group radius as a measurement of how well my reloads are working. I have done several studies during my engineering days, and we would confirm with full factorial analysis... kinda like what the article mentions.

I'll work on my experiment and share data over the next few weeks. Thanks!


Well-Known Member
We used DOE at work to optimize different production variables to get the desired
output quality. Pretty similar application, here the desired quality is a small group.

Interesting approach.



Well-Known Member
This method is one of several to find the most PROBABLE selection of variables that will produce the desired result of tight groups. It still requires a large effort to collect enough data points to be of any value. The author 'processed' many variable 'out' of the equation by loading techniques. The 'S' actually has 4-5 similar numbers so maybe not the best but several combinations that might be acceptable. Anova is 5th moment of variables in this case. Moments throw outliers away and is actually a method often used in what is called AI. IMHO, same a s Satterlee, won't find a good load unless a large number of data point is acquired. Would be interesting to apply it to the data in other thread. Cast bullets create another problem as seating depth does matter.


Well-Known Member
You know,when using machine equipment,you'll often hear..... sneaking up on a cut/measure... whatever.

Don't make the mistake thinking we're gonna keep cutting .002" off until the "mark" gets hit.

What we're doing is.... processing exactly how well everything is in tune. Meaning,I might still be .040" from the target dimension but,dialing in say,.007"(or any pre determined #) results in ___" practical measure. Which is a test of sorts,when we do actually try to hit the target.

And then there's also a spring cut.... this is the test that we're trying to account for "spring" or flex in the tooling.

Back to topic...


Well-Known Member
Cutting tungsten huh? off topic but reminds me of miss distance calc. We had a celbrationafter a film. Laser dot on front tire of a jeep, high speed camera caught the nose of the bird hitting the tire before the 'bang'. Defined a new miss distance.
Last edited:


Well-Known Member
I'm interested but still have ptsd from a semester of multi-variable calculus something like 25 years ago.


Well-Known Member
RPN is absolutely superior. I started with an HP45 in 1973 for grad school after getting my BS with
a VersaLog slide rule. HP25 replaced the 45 when one number key finally died, then the HP11s came along,
and I now keep about 4 of them running and a 32S like you'r Ian. On a given day, most likely using a
HP11C. Fine, fine machines. Super reliable.

Ah, partial differentials....I decided after my second grad school course in partial diffy Qs that I was
going to stop taking any more optional math, after a took a course in imaginary numbers. Untimately,
that stuff proved to be mostly a dead end for engineering, anyway. thank goodness. Second course
in partials hurt my head.

The fact that you could write a differential equation to accurately describe your problem, but that there
has never been any general strategy that works for solving "any" differential equation is what led to the
rise of the finite element analysis method. Yes, FEA is an approximation, but, given enough computer power and a fine
enough grid of elements, and enough time, the approximation can be arbitrarily accurate. Very powerful general purpose
tool which can be used for all geometries, unlike diff eqs. Then when explicit FEA came along to compliment implicit, in
the late 80s to early 90s, the contact and dynamic effects limitations were properly accounted for.
Initially, explicit took too much CPU power, but again, alternative numerical methods with successive approximations
has solved that nightmare. Of course, ever faster computers also helped a lot. The last computer I bought had
2048 CPUs.

Finite elements grew up as I became an engineer, and I grew up with it. It was an amazing ride, on
the cutting edge of computer technology from 1975 until a couple of years ago when I retired. Following
and to an extent blazing, the path beyond the dead end of diff EQs with big honking supercomputers.

Last edited:


Active Member
Good morning
When I find an alternative for a rifle I just might think there is some thing here.
But I am glad someone has the depth of mind to fathom this. I still am very happy to load 7 and let them rip though the target.