Efficient Load Development Using Statistical Design Options

[Brians356]

... The limited data points would be a problem if you had discontinuous sudden spikes, highly nonlinear behavior, ...

So best not to start at bullet jammed and work out, or start jumping a bit but working in to jammed? A series of substantial jumps (as the OP chose) would be best for linearity.
-

 

[michaelbz]

I would do one DOE at/near the jam and another away from the jam and then compare the two results. I personally would have used finer jumps. We have a sinusodal vibrational system in both cases, and the results curve looked sinusoidal to me. By non-linear I really mean random, you got to use the right curve. The large jumps are for cost and time, if you know there are no higher frequency nodes you are fine.

 

[CharlieNC]

You do need some notion of the behavior to decide on how many levels and how fine increments to use. A key point is with the DOE you can efficiently map out the entire space (that is COMBINATIONS of the variables) whereas the one at a time approach will never get you there. In my example the combination of high charge with a long jump was very bad for groups. This is one additional dimension which is easily learned using this approach. As mentioned this methodology has been used for many years, esp since WWII, and I'm surprised I've not seen more shooting applications.

 

[BoydAllen]

There is an article in the Berger reloading manual ( that is also on their web site if you look around) that describes a method for finding the best seating depth for VLD bullets. It is entirely different than anything that I have seen published, and it is from an excellent source. I called Berger and asked if the method works for non VLD bullets and the technical resource person that I spoke with told me that it does. The reason that I mention this is that your method may have merit, but the results can only be as good as the assumptions you base your test on. Also, I see a lot of testing that is flawed because shooters dismiss the need for wind flags for dealing with that variable.

(added later) Link to article:http://www.bergerbullets.com/vld-making-shoot/


If I may, let me tell a story that relates to this. A friend who shoots short range benchrest, and who builds rifles built a special purpose varmint rifle, in .222 specifically to shoot Nosler no lead ballistic tips. He worked up a load using Xterminator with his bullets seated for a strong touch of the rifling, then because of the nature of the rest that he uses in the field, and his desire to reduce muzzle flip so that he could more easily see his hits, he put a muzzle break on the barrel. This threw him completely out of tune. Not wanting to change his charge weight because he liked the velocity and case fill that it gave him, he decided to experiment with seating depth. He loaded up two shots each starting at the bullets touching the rifling, and increased jump by .010 for each subsequent test. He was using wind flags and shooting off of a concrete bench. Unlike what many, including myself, would have done, he tested to amounts of jump that would have seemed ridiculous. At a jump of .080 one hole cut the other, and subsequent testing verified the accuracy of the load.

Read the article that I referred you to, and think about how your assumptions may have resulted in a less than optimal result. Prior to learning what I have told you about I would have made the same assumptions.

In the case of my friend, his before brake powder charge test probably involved a total of perhaps 20 shots if that, and the one that first gave him an indication that he had the right seating depth for use with the brake would have been 16. I do not think that saving shots to totals below these numbers is important. His method has worked well for a number of calibers and bullets. I use a method that takes even fewer, but I like his better.

Finally, I wonder why no testing was done with bullets seated various to various depths into the rifling.

 

[dcali]

This is more or less what I do, although I tend to use more data points because of the uncertainty in so few a number of rounds - the data is pretty noisy as it is, and with just a handful of two shot groups you're really pushing the statistics. But yeah, it works.

 

[dcali]

This also reminds me of a method I learned in grad school called the Golden Section Search. It's an efficient way of discovering the low points in your plot without having to plot all the points. It's been rattling around in my head as a way that may actually work, but I've never attempted to use it (or something similar) for shooting, partly out of laziness and partly because you have to load at the range to do it, and I'm not really set up for that.

 

[spclark]

I'm all in for ways that might shorten the journey to the most accurate load despite being higher math-challenged. (I'm gonna work on how that surface plot gets described, it's not quite clear to me but it's been a rough week.)

I do very much appreciate your both taking the time to do the work evident in your PDF as well as making it available here for the rest of us.

Counting myself among those who avoid jamming at all costs (even that of accuracy, if necessary!) I'm gratified to see your results kinda support my sense that jumping 0.020" (+/- 0.010") is a good place to start with load development.

I frequently leave it at that, seeing as how my 1 MOA target's X-ring is 'big enough' to allow good scores despite my 2 MOA hold... most days.

 

[CharlieNC]

To clarrify a few points, for my purpose this DOE was to combine what has been historically done as two separate separate ladder tests which were aimed at finding the least sensitive poi. If on the other hand the goal were to find the smallest group sizes then more shots per trial could have been done in a similar fashion. Finer or coarser increments are at the discretion of the shooter, and not etched into stone. There are many, many other test designs which are possible to suit exact circumstances and this was but one example. To optimize group sizes I would choose 10 shots and use the standard deviation, not extreme spread; but thats another story.