When you collect velocity data, how many shots do you consider relevant?

Usually 5 but sometimes 10, or sometimes 2 groups of 5. Mine has a printer and I often staple the printer tape to the target. I would think that to be really scientific, the more the better. But, we have to live within reason.

I normally do 5 or 6 (mostly in revolvers). If you have so much variability in a load that you need more than 5 shots to get a reliable mean you probably need to work on improving the load.

I like to get 6-10 well-controlled rounds thru the Chrony ... o'course groups trump numbers.

From the standpoint of statistics a statistically valid sample size is usually held to be 30. Most of us do not care to know the velocity to that degree of confidence.
What many are interested in is the variability of the velocity.
You have to understand what is important or significant to you then shoot the sample and perform the calculations.
Anyone can do it but it takes about a semester long statistics class to gain the understanding of what is a reasonable goal.

I normally do 5 or 10. I feel this is enough to know if I am in the ballpark. If I see a large extreme spread with just 5 shots, more shots will only make it worse, even if the standard deviation may improve. More will always be better, statistically, but some bullets can be expensive.

i normally use ten shots when i test a load. i ignore sd and solely rely on es (extreme spread) when evaluating the numbers.

any es below 25 for a pistol round is excellent in my book. fifty is probably average and is acceptable to me unless i am making up an accuracy load. in that case, i don't stop until the es number is less than 25 (or as close as i can cause 25 is hard to get to for pistol rounds).

also, i always set the chrony 12 feet away from the muzzle so the readings are consistent and so the powder blast doesn't affect the reading. i know you don't have this problem since you use the labradar, but thought i'd throw that one in for those of us that don't use that fancy gadget.

luck,

murf

10 shots for velocity is the usual standard for gun-writers for handguns. It's often a pretty good estimate.

The old story that you need 30 samples is almost always false. That comes from the assumptions associated with doing statistical tests based on the normal distribution. That path assumes that you know the true mean, and it takes about 25-30 samples to get close enough for the test to work reliably. But nobody does their statistical tests based on the normal distribution anymore. We practically always use the T distribution, which works just fine for small samples. Plus, in the case of estimating MV, we aren't doing a statistical test, we are just estimating a parameter. So forget 30.

The number of samples you need to estimate mean MV depends on how precise an answer you need, and how much the MV varies. For most common ammunition, around 5 rounds is usually enough. That will get you within about plus or minus 25 FPS most of the time. If you need more precision than that, quadrupling your sample size cuts your error in half. So going to 20 rounds cuts your error to about plus or minus 12.5 FPS.

Estimating standard deviation is more difficult. An estimate of SD based on 5 samples is not very precise at all.

Range ("extreme spread") is a good estimate of variation if samples are small. For a sample of 5, it's 90% as good as SD. If you're using 5 shots, you might as well use range instead of SD.

I never test less than 10. When I am on my final testing to narrow down the final couple of candidates, it is usually 25 rounds.

I load 10 of each charge. With my optical chrono you may not get a reading for one or two, so 5 rounds just isn't enough. With your shiny new Labradar I'm sure the capture rate is better. I probably should have bought one..

460Shooter said:

What do you consider to be a large extreme spread? I was shooting Federal HSTs.

Click to expand...

Depends on what I am shooting and the intended purpose of the ammo I am working up.

For most of my pistol ammo, that I will never shoot past 25 yards, and mostly at 10 yards, an extreme spread of 75 would be acceptable as long as it is accurate. Accuracy of 1" or better at 10 yards is fine. I am primarily interested in the extreme spread for knowing that my ammo will reliably cycle the gun.

For rifle plinking loads (100 yards or less) I like an extreme spread of 50 or less, but accuracy will trump this. For precision rifle, past 100 yards, as low as I can get.

460Shooter said:

I missed a few rounds because I got a little too rambunctious.

Click to expand...

That beats getting a little too rambunctious and shooting the chrono.

Just because you don't like to expend the effort to test does not make it invalid. Only the person doing the testing can decide what is reasonable for his situation.
Go back to your statistics book.
You use a T distribution because you think it is better for small sample sizes. But you chose a small sample size to reduce the amount of work. That is circular logic. But is the student T distribution accurate enough or do you use it out of sloth? There is nothing to prevent you from estimating the standard deviation except the lack of effort.


In probability and statistics, Student's t-distribution (or simply the t-distribution) is any member of a family of continuous probability distributions that arises when estimating the mean of a normally distributed population in situations where the sample size is small and population standard deviation is unknown.

denton said:

The old story that you need 30 samples is almost always false. That comes from the assumptions associated with doing statistical tests based on the normal distribution. That path assumes that you know the true mean, and it takes about 25-30 samples to get close enough for the test to work reliably. But nobody does their statistical tests based on the normal distribution anymore. We practically always use the T distribution, which works just fine for small samples. Plus, in the case of estimating MV, we aren't doing a statistical test, we are just estimating a parameter. So forget 30.

The number of samples you need to estimate mean MV depends on how precise an answer you need, and how much the MV varies. For most common ammunition, around 5 rounds is usually enough. That will get you within about plus or minus 25 FPS most of the time. If you need more precision than that, quadrupling your sample size cuts your error in half. So going to 20 rounds cuts your error to about plus or minus 12.5 FPS.

Estimating standard deviation is more difficult. An estimate of SD based on 5 samples is not very precise at all.

Range ("extreme spread") is a good estimate of variation if samples are small. For a sample of 5, it's 90% as good as SD. If you're using 5 shots, you might as well use range instead of SD.

Click to expand...

I like multiples of 10 myself. Makes the math simpler .

Writers have deadlines and expenses to consider. I will guess that more time at the range obtaining data means less time at the keyboard. Plus, if you are paying for each shot, getting a 10 shot sample (naturally) costs 33.3% less than a more complete 30 shot sample.

Obviously getting more data (that’s accurate!) is more valuable then less, but if 10 shots allows for a reasonably accurate sample then start there .

Stay safe!

Ireload2 said:

Just because you don't like to expend the effort to test does not make it invalid. Only the person doing the testing can decide what is reasonable for his situation.
Go back to your statistics book.
You use a T distribution because you think it is better for small sample sizes. But you chose a small sample size to reduce the amount of work. That is circular logic. But is the student T distribution accurate enough or do you use it out of sloth? There is nothing to prevent you from estimating the standard deviation except the lack of effort.


In probability and statistics, Student's t-distribution (or simply the t-distribution) is any member of a family of continuous probability distributions that arises when estimating the mean of a normally distributed population in situations where the sample size is small and population standard deviation is unknown.

Click to expand...

Let me see if I can do a little better job of explaining myself....

Estimating the mean does not involve assumption of any particular distribution at all. Getting the Confidence Interval (how precise your estimate is) doesn't involve assumption of any particular distribution of the raw data, either. Your data can come from a Normal Distribution, an F Distribution, a Uniform Distribution, or a Maxwellian Distribution. It doesn't matter.

If you are sampling two groups to see if they come from different populations, you can use the Normal Distribution (Z score) or you can use the T Distribution (T score). The Normal Distribution assumes that you have the true population mean, and if you have 30 samples that assumption is usually well enough met. The T Distribution assumes that you only have an estimate of the true population mean based on a sample. Nobody beyond someone in elementary statistics ever does the Z score thing. Everybody with any experience doing that kind of test uses the T Test. The T Test works regardless of sample size. When the sample is large, T and Z become the same. But, of course, that's all irrelevant because that's not what we're doing when measuring mean MV.

You can run as many cartridges as you like, and get any precision that suits your purpose. But to halve your error, you have to quadruple your sample size. That gets out of hand in a hurry, and most people don't need an answer to within plus or minus .5%.

SAAMI makes a conservative assumption that the standard deviation of MV is 4% of MV. It's almost always better than that, but that's a safe assumption for some of the things they are doing. So if you use their number and shoot 5 shots, your Standard Error will be .04/(square root 5) = 1.8%. Want a better estimate? Shoot 10 shots. Your Standard Error is now .04/(square root 10) = 1.2%, not that big an improvement. Well then, double your sample again and go to 20 shots. Now your Standard Error is .04/(square root 20) = .9%. You've quadrupled your sample, and halved your error. Adding more samples really isn't buying you much more information.

I do 5 for velocity when I'm working up a load. When I hit a sweet spot that groups real nice, I'll do about 20 then save the avg into my ballistics calculator.

If you're my father-in-law, the correct answer is "all of them"...

From the standpoint of statistics a statistically valid sample size is usually held to be 30

Click to expand...

Uhmmm....no. There is no blanket valid sample size without a consideration of the variability of the sample and the level of statistical power you desire to achieve.

Sample size: how "accurately" do you want to measure the true mean. The means we get will always be the SAMPLE mean, although the sample mean of 1000 shots will most always be closer to the true mean than a sample mean of 5 shots.