SD: overrated...?

[Jay Christopherson]

CharlieNC,



I guess I am asking, what can cause a load to increase from 1 SD to two or three SD?

Bear with me, I normally just shoot live varmints at extended ranges here in SW Idaho instead of paper.

Inconsistent loading practices, mostly. Also, potentially factors like barrel or external temps - for example, in current temperatures around here, it takes a long time for my barrel to warm up to the point where it produces consistent numbers (and accuracy). Which, obviously will affect my SD numbers if you look at it in the short term. If I look back over my data, I can see that in 60º weather, it only takes a couple shots before the numbers level out. Right now, in 20º weather, it's been taking a lot more shots before the numbers level out. You have to consider those, but it's still valuable data if you shoot matches at one extreme or the other.

Now that I am gathering enough data across a long period of time, I can look back at my numbers for 20º last year and consider whether my numbers are consistent with those. Same goes for warmer temps. All of which tells me whether controllable factors (like my loading practices) are changing or not.

EDIT: All of this long winded stuff I posted is really to make the point that you can't judge your SD based on a couple of strings. You can't look at String 1 (20) with an SD of 4 and String 2 (20) with an SD of 12 and make a very good causation analysis. It's suggestive, sure. But ultimately too small of a sample to be really useful by itself. Like any data, the longer the history, the more relevant SD analysis based on it becomes, which takes me back to how much I love my Labradar with it's non-intrusive setup.

 

[fredhorace77]

That said, I still hold that only the number of shots for record should be considered, in terms of the significance of ES and SD data.

I guess I am asking, what can cause a load to increase from 1 SD to two or three SD?

First paragraph....you answered your own question. What everyone is saying is....yes only the number of shots should count. And as the number of shots increase so does SD.

Second paragraph....discrepancies in tolerances of brass, bullet, primer, powder manufacturering.

 

[savagedasher]

Your group will never get better after the first shot. It can only get worse.

One shot I don't call a group But I can say I never had a group get SMALLER then my Second Shot . That is the reason I shoot Two .
Larry

 

[JDMock]

Your group will never get better after the first shot. It can only get worse.

One shot is not a group, so after 2 shots one's group cannot get better.

 

[Toby Bradshaw]

And as the number of shots increase so does SD.

Nope. ES increases with sample size (assuming a normal distribution). SD converges on a single value.

Generate numbers from a normal (Gaussian) distribution and subsample them with increasing sample sizes to see for yourself.

Basically, increasing sample size means that you're more likely to generate a value in the tails of the distribution (increasing ES). But the shape of a Gaussian distribution is specified by its mean (average) and SD. I.e., SD is a parameter that we are trying to estimate from our sampling. Parameter estimation improves with increasing sample size.

ES is not a parameter, it is just a function of mean, SD, and sample size.

 

[Rust]

The SD number. I used to be more anal retentive about SD numbers but I feel much better now.

My primary rifle is a 6.5 X 284 which has been fully developed, with a good load that has provided single digit ES numbers across every barrel thus far. I don't have a lot of time on my hands, there isn't a top smith in the area, and money is tight. Given that a barrel lasts no more than 1,200 rounds I'm not wasting any barrel life looking for the preferred large sample SD. Any additional load development is done when swapping barrels or starting a new lot number of a component.

My initial 1,000 yard load development happens at 100 yards, There isn't a conveniently located 1,000 yard range in my area. I observe group development (size, shape, POI vs POA) and find when the groups stabilize for typically three consecutive loads, that is usually where ES tightens up too. A small group does not mean low velocity spreads and I've seen a lot of loads that shoot tight at 100 I wouldn't think about shooting at 1,000. And if I'm getting single digit ES on top of tight groups SD will take care of itself.

.223 and .308 barrels last much longer but once development is complete it's much the same.

Past a certain point local range knowledge is more important. What happens with no wind, wind from the right or wind from the left depending on terrain features, shooting off concrete, dirt or grass, sun vs. overcast, anything which can affect actual shooting. Record all of that religiously.

 

[Jammer Six]

Brian does not understand statistics.

A group of two could have either a small SD or a large SD.

Reliability is not a property of standard deviation.

 

[GaryD]

Maybe I remember correctly from stats is 1 SD means 95% 2 SD is 99% and 3 SD is 99.9%. So if I load 10,000 rounds and shoot 10 rounds and get an SD of 20 I should expect 95% of the rounds to be +\- 20 fps of the average fps, 99% to +\- 40 fps of the average and 99.9% of the rounds to be +/- 60 fps. However with the small sample set there is low confidence in the SD number.

If you were to shoot 1000 round there is a chance the SD will go up or down but the confidence in the SD will alway increase with a increase in data points.

I am assuming remember those numbers from a class 20 years ago

Gary

 

[bigfish28]

SD and ES are both methods to evaluate the quality (precision) of reloads. Sd can be used to predict the MV of your loads. 1 SD plus and minus will be 63% of your shots. So if SD is 10 for +/- 1 SD your MV will be lower - MV - 10, upper = MV + 10. This can be used with trajectory tools to changes in flight path. At close ranges these can be small.
SD of +/- 2 covers 95% of your ammos MV. Again you can predict the impact for 95% of the rounds. Above +/- 3 SD occurs 6 times out of a million.
ES is does not know where in this range the ammo is so it cannot be used for prediction.
Twenty of preferably 30 shots are needed to determine an accurate SD.

 

[Toby Bradshaw]

So if I load 10,000 rounds and shoot 10 rounds and get an SD of 20 I should expect 95% of the rounds to be +\- 20 fps of the average fps, 99% to +\- 40 fps of the average and 99.9% of the rounds to be +/- 60 fps. However with the small sample set there is low confidence in the SD number.

With a sample size of 10 the chances are about 50-50 of estimating SD within 15% of the true value (Greenwood and Sandomire 1950). You can be 90% sure that your SD estimate is within 35% of the true value. So any prediction of ES distribution based on an SD estimated from 10 shots is going to be suspect.

To be 90% sure that your SD estimate is within 10% of the true value requires a sample size of more than 100.

All of this assumes that the chronograph itself is perfectly accurate (no instrument error), and that the distribution of muzzle velocities is normal.

You're better off just looking at the targets.

 

[rardoin]

Nope. ES increases with sample size (assuming a normal distribution). SD converges on a single value.

Generate numbers from a normal (Gaussian) distribution and subsample them with increasing sample sizes to see for yourself.

Basically, increasing sample size means that you're more likely to generate a value in the tails of the distribution (increasing ES). But the shape of a Gaussian distribution is specified by its mean (average) and SD. I.e., SD is a parameter that we are trying to estimate from our sampling. Parameter estimation improves with increasing sample size.

ES is not a parameter, it is just a function of mean, SD, and sample size.

Great explanation Toby. During the course of obtaining one of my Doctoral degrees I had to take an in depth course on statistical analysis, as I had to justify the analysis of data presented during the defense of my thesis and in publication of my research findings. I had/still have a very loose grip on statistics. Thank God my NIH grant was flush enough to hire a statistician to handle that aspect as this dumb old coonass could not comprehend half of what she was explaining!

 

[Jammer Six]

You're better off just looking at the targets.

Winner.

The one thing that hasn't been mentioned here is that the SD can indicate a non-normal curve. It might, it might not, your shooting curve may be normal, it may not, for whatever reason that day.

That makes SD interesting, once you have a thousand data points, but not much more than that.

It's a tool, and it doesn't do the things you guys are trying to do with it, and a change that is a true improvement in a load will not necessarily show up as a smaller SD.

 

[memilanuk]

Maybe I remember correctly from stats is 1 SD means 95% 2 SD is 99% and 3 SD is 99.9%.

Close.

+/- 1 SD covers ~68%
+/- 2 SD covers 95%
+/- 3 SD covers 99.7%

Source

I am assuming remember those numbers from a class 20 years ago

Pretty close, considering!

 

[memilanuk]

More than you ever wanted to know about statistics and shooting...

Ballistipedia

 

[GaryD]

Thanks for the correction milanuk

 

[CharlieNC]

Nope. ES increases with sample size (assuming a normal distribution). SD converges on a single value.

Generate numbers from a normal (Gaussian) distribution and subsample them with increasing sample sizes to see for yourself.

Basically, increasing sample size means that you're more likely to generate a value in the tails of the distribution (increasing ES). But the shape of a Gaussian distribution is specified by its mean (average) and SD. I.e., SD is a parameter that we are trying to estimate from our sampling. Parameter estimation improves with increasing sample size.

ES is not a parameter, it is just a function of mean, SD, and sample size.

Toby that is an excellent point. Because of this it is meaningful, to collect historical SD results and "average" them to obtain a useful overall SD. But ES is not nearly as friendly in this regard and must always be adjusted based on sample size. And this is why people are misled by a low ES when shooting 5 shots, and the real world performance of say 20 shots is much worse on the target.