ES or SD?

[bjld]

G'day all

What do you use in load development: standard deviation (SD), extreme spread (ES) or both?

[Tony Z]

edited 19/4/06

[Lynn Otto]

Hey Ben

Tony's response was sweet and short but realistic. When we're testing I'm only concerned with how tight a group I can get from a load without any outliers. We put it all through the chrony and Trev takes both the ES and SD plus velocity very serious but if its my rifle I get the final say and it's group size that counts every time. We've found some the loads with the best numbers just don't cut it at the target end of the business.

Lynn

[Simon C]

Gday Ben,

I look at ES for loads which have demonstrated tight groups then choose the one with the lowest ES across 10 - 15 rnds.

Just did this last W/E for one of my hunting rifles...but need to do it for 2 other projectiles I use in it It can be a real PITA but is worth it in the end

[bully_eye]

I don't have access to a chrony so have to use a target. I had the unfortunate experience when I rebarelled my old FO rifle, that I worked up a load at 100yd that didn't group at long range at all. Now at 100yd it was putting 5 shots into well under .5moa, usually around .3moa. However at 500m and beyond it flew all over the place. I still wouldn't take a load that DOESN'T group at 100yd into a match but when my new rifle is finished (or rather started first), I'm going to have to either take the arduous approach of shooting work up loads at long range with all the problems distance brings to the equation or borrowing a chrony so that I'm going to have a good chance of reasonable elevation out past 600m. I sometimes think nothing works well in a particular barrel unless the load is right, the shooter does their part.... and the sun, moon and all the planets are perfectly aligned. That will be the day I score my possible!
Michael

[bruce moulds]

hi,
if lynn & tony agree, there must be some truth in the matter, particularly if i agree too. the target tells no lies, even though condition can throw confusion into the equation bigtime.
trevor is known for his ability to get guns to waterline, after which it is up to the shooter.
there is a good pinned section on this subject on long range target website, i think under precision reloading.
the relevent qualityreference is by pengleman.
i am surprised that a statitician like ben would ask such a question.
bruce.

[RAVEN]

I use a chrony in conjunction with a target analysis program
Dont know how helpful this is but interesting all the same

Cheers
RB

[pjifl]

If you fire enough (a lot) shots then SD probably means more but of necessity we work with sparse data. Which can olften lead to erronious deductions.

Personally, I think eyeballing the groups is as good a method as any - especially with sparse data.

3 shot groups worry me and even 5 shots dont prove too much but are still very useful. Its very tempting to stop putting down shots as soon an the expected result is 'proved'

Its also worth recording shot order because falling or rising shot trends are invaluable knowledge to have but this data may often be better derived from actual shoots.

A related question. If one finds that elusive perfect load - how long does it remain relevant with a throat eroding for each shot. I simply dont know - maybe others can say something here.

Peter Smith.

[bjld]

I personally would prefer to use the 95% confidence interval over SD because it is so much easier to translate into probability, although the number of shots necessary to generate a realistic Gaussian distribution is prohibitively large. Thus I fall back onto SD, and 10 shots is sufficient to produce a meaningful SD. Dividing SD by the mean velocity produces the coefficient of variation and is probably the best way to compare loads. This allows meaningful comparisons between loads with very different muzzle velocities (such as different cartridges). I generally ignore ES.

[Lynn Otto]

LOL Ben

That was probably foriegn language to many people. I followed it but only because they forced me to do statistical analysis as a part of my uni degree. Take the most statistically sound route and you have a worn out barrel before you get to your first match, as you alluded. Do all the numbers, it will add confidence to your choice and we all know that believing in something is mighty powerful medicine. But at the end of the day make sure they are all smack in the middle of the ten ring regardless of the fancy numbers, because you won't be happy with anything less.

Lynn

[AlanF]

Ben

I seem to remember a lively discussion we had on this from Benchrest Central a year or two ago!

Here's my current thoughts. Because we shoot for score in F-class, I agree that SD is a better way of comparing test groups. Even if there is an odd flier, if a group has a lower SD it will usually outscore a smaller group with a higher SD. However, if we were shooting for groups (as in BR), then you need to be able to exclude loads which have fliers, because fliers have such a disastrous effect on group size. So it would stand to reason that ES is a more suitable type of indicator for reducing group size.

However you'll be forgiven for saying its not quite as simple as that!

Alan

[Tony Z]

edited 19/4/06

[bjld]

Alan F: SD can effectively predict ES (given sufficient sample size). If you shoot a group with one "velocity flier" (ie large ES), will it happen the next time you shoot a group with exactly the same load? Probably not. But the SD is likely to be more reproducible.

Tony Z: the best model for simulations is the one most like the real world, so you are correct - nothing can better 1000 yard load testing for shooting at 1000 yards, except that we cannot precisely define all the variables (eg wind - we all need more wind reading practice). Perhaps you only consider vertical dispersion? But doesn't a wind change from one side to the other cause vertical dispersion? You surely use a chronograph to measure your muzzle velocity (mv) in load development, at the very least to calculate your elevation for different ranges. Is that all you use it for? When you've generated the mv data why not use it?

[Tony Z]

edited 19/4/06