I'll Post This Here - Hornady's Podcast #50. I thought it was one of their best, but some reloaders might not like what they see....

[mrcheesemoose]

You could remove a lot of the bias in measuring and aggregate in groups by changing the metric from group size center to center to the distance from the center of each impact to the point of aim.

If you had that metric, and I’d like to have some software that could measure it quickly, you could then take a simple average of every single shot’s distance from the point of them, and get a very statistically valid, practically useful average

Hornadys ballistic app gives you x and y coordinates for each shot with respect to your point of aim. It will give you the center of the group and how far off it is from the center of the target. I’ve taken these values and put them in a spreadsheet, then wrote some software that does CEP95 analysis on all the coordinates. I’ve also put some data in that spreadsheet to distinguish between different bullets, powders, powder charge, etc. so then I can shoot a 30 round group one day, then shoot another a couple days later, and as long as the combos are all the same I can put them through the software to get the CEP95 of the whole 60 shot group.

 

[supercorndogs]

An ammo manufacturer said we need 30 round groups. let me know when this changes the game.

 

[mrcheesemoose]

An ammo manufacturer said we need 30 round groups. let me know when this changes the game.

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

 

[beetroot]

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

 

[Niles Coyote]

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

Wouldn't you need 30 classes?

 

[supercorndogs]

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

They're

And I thought someone quoted me to let me know the game changed.

A 1.5" 3 shot group at 100y tells me a lot. 2 of them tells me all I need to know about a particular load.

 

[beetroot]

They're

And I thought someone quoted me to let me know the game changed.

A 1.5" 3 shot group at 100y tells me a lot. 2 of them tells me all I need to know about a particular load.

Nope, you need the next 27 to confirm your load is indeed a POS.

 

[diggler1833]

But see...if you shoot enough, you'll eventually F-up and pull a shot, thereby creating more validity in your data sample.

Joking aside; I did the multiple statistics and quantitative data and analysis courses in college too (where's my cookie). The short of it is that larger sample sizes will always give a closer approximate to a population. However, so many that are over quoting that here seem to forget that one does not have to continue to collect samples if the population fails to quickly meet his or her accuracy requirement. We are literally trying to manipulate the population through component testing.

^ "Hey, this one looks like shit after five shots."

Well, it's still going to look like shit after 35 then too...but now you've spent an extra hour and $40 to confirm that.

 

[spife7980]

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

Over educated and under intelligent.

 

[mrcheesemoose]

Gosh didn’t realize reloaders were such snowflakes

 

[mrcheesemoose]

But see...if you shoot enough, you'll eventually F-up and pull a shot, thereby creating more validity in your data sample.

Joking aside; I did the multiple statistics and quantitative data and analysis courses in college too (where's my cookie). The short of it is that larger sample sizes will always give a closer approximate to a population. However, so many that are over quoting that here seem to forget that one does not have to continue to collect samples if the population fails to quickly meet his or her accuracy requirement. We are literally trying to manipulate the population through component testing.

^ "Hey, this one looks like shit after five shots."

Well, it's still going to look like shit after 35 then too...but now you've spent an extra hour and $40 to confirm that.

That’s not exactly true. I’ve shot groups where the first 5 were crap, but the next 10 or 20 were much tighter. Turns out the first 5 just happened to be the worst 5. But you wouldn’t know that until you put more bullets through the barrel

 

[beetroot]

That’s not exactly true. I’ve shot groups where the first 5 were crap, but the next 10 or 20 were much tighter. Turns out the first 5 just happened to be the worst 5. But you wouldn’t know that until you put more bullets through the barrel

That does often happen, the a new load and rifle need to get to know each other first before they can perform at their best.

A bit of KY on the first few bullets can speed the process along.

 

[JAS-SH]

A 30-shot group is no more useful and sometimes worse than six 5-shot groups. Worse because its hard to measure shots that impact very close together. If shooting really accurately you might destroy the exact point of aim completely.

Take the six 5-shot groups and aggregate them. A little extra math but easier to accurately measure groups.

Hornady added a group analysis section to their app this year. Group Analysis can now be used for the app to calculate a very accurate zero angle solution. The short video below shows how to do that.

Zero angle is very desirable if you are going to shoot in different conditions and from different POA elevation angles. Don't need to re-zero....

Here's the skinny:

 

[mrcheesemoose]

A 30-shot group is no more useful and sometimes worse than six 5-shot groups. Worse because its hard to measure shots that impact very close together. If shooting really accurately you might destroy the exact point of aim completely.

Take the six 5-shot groups and aggregate them. A little extra math but easier to accurately measure groups.

Hornady added a group analysis section to their app this year. Group Analysis can now be used for the app to calculate a very accurate zero angle solution. The short video below shows how to do that.

Zero angle is very desirable if you are going to shoot in different conditions and from different POA elevation angles. Don't need to re-zero....

Here's the skinny:

Yah when I mean 30 shot group, I mean 30 shot aggregate. The way I do it is to aggregate 3 10 shot groups.

 

[diggler1833]

I agg. If I shoot a good 5-shot group during development at 100, I'll repeat it just to see if it is a fluke (it has happened). If I repeat that group within a tenth, then I shoot it a third time at 300-385 yards (as far as I can go from my house). If it holds up again, I have a pretty solid confidence that when I go to take a shot at a coyote or whatever that I'll hit within my standard/capabilities of accuracy.

When I true my velocity at 500-600 the group usually tends to be within 1/4 MOA of what it was shooting at 300 (so, a 4th test essentially). I'll also note here that I only shoot load development at distance on calm days because some seem to overlook environmental factors regarding accuracy consistency.

If my first three shots at 100 are 1.5 MOA, I don't care if it just so happens that those would be the worst out of a 30 shot string...that load still fails to meet my expectations. I just save myself that hour and $40 trying to feel more comfortable about my sample size by moving on.

Extreme spread is just as important as standard deviation to some. You won't be competitive at a target match if 95% of your string is tight, but you always manage to blow an unaccounted for "7".

^ This is the argument for larger sample sizes, but it also works as an argument for abandoning a load quickly.

 

[secondofangle2]

View attachment 8221726

I care.

People uninitiated in statistics and data analysis are doing those things in everyday life based on traditions established eons ago by similarly uninitiated people.

Finally, we get sophisticated enough that professionals in the fields of stats and data analysis are bringing them to bear on our discipline which, insomuch as it involves measurement, is inextricably linked to stats, and "nobody cares"?

Sounds foolish to me, but to each his own.

 

[Marine52]

I care.

People uninitiated in statistics and data analysis are doing those things in everyday life based on traditions established eons ago by similarly uninitiated people.

Finally, we get sophisticated enough that professionals in the fields of stats and data analysis are bringing them to bear on our discipline which, insomuch as it involves measurement, is inextricably linked to stats, and "nobody cares"?

Sounds foolish to me, but to each his own.

No. The Hornady podcast was wrong on many levels. It was a display of individuals not familiar with correct statistical inference. A random variable (or random experiment) is a test in the sample space and maps the outcome to a number on the real line. So a single result of one experiment (5-shoot group) would generate one outcome on the real number line, say 0.5 MOA. A 50-shot group is one experiment and generates one number (observation) for example 0.75 MOA. IT IS NOT A LARGE SAMPLE — IT IS A SAMPLE SIZE EQUAL TO ONE. If you wanted to get statistically valid results, you would aim for a sample size of 30 or so. That would be thirty 50-shot groups and then you could generate meaningful test statistics like confidence intervals, t-tests, standard errors, etc.

if you have ever listened to Bryan Litz talk about his shot groups and test stats, you will notice that he talks about sample sizes of say 30 where he obtains thirty 5-shot groups, for example. This is correct sampling technique. He is then able to generate standard errors, confidence intervals and other meaningful test stats based on a sample size of 30. With thirty 5-shot groups you generate 30 observations

As someone mentioned above, If you were to take the 50 shots and then use the random variable of mapping each shot to a mean radius then you could generate a large sample of 50 and produce usable test statistics for population inference.

Also, the Hornady analysis ignores all prior information. But that’s a deeper topic and gets into Bayesian stats.

 

[mrcheesemoose]

No. The Hornady podcast was wrong on many levels. It was a display of elementary statistical methods by individuals not familiar with correct statistical inference. A random variable (or random experiment) is a test in the sample space and maps it to a number on the real line. So ONE outcome of such an experiment (5-shoot group) would generate one outcome on the real number line, say 0.5 MOA. A 50-shot group is one experiment and generates one number (observation) for example 0.75 MOA. IT IS NOT A LARGE SAMPLE — IT IS A SAMPLE SIZE EQUAL TO ONE. if you wanted to get meaningful results, you would aim for a sample size of 30 or so. That would be 30 50-shot groups and then you could generate meaningful test statistics like confidence intervals, t-tests, standard errors, etc.

if you have ever listened to Bryan Litz talk about his shot groups and test stats, you will notice that he talks about sample sizes of say 30 where he obtains thirty 5-shot groups, for example. This is correct sampling technique. He is the able to generate standard errors, confidence intervals and other meaningful test stats based on a sample size of 30. With thirty 5-shot groups you generate 30 observations

As someone mentioned above, If you were to take the 50 shots and then use the random variable of mapping each shot to a mean radius then you could generate a large sample of 50 and produce usable test statistics for population inference.

Also, the Hornady analysis ignores all prior information. But that’s a deeper topic and gets into Bayesian stats.

This ^. 30 is considered the bare minimum sample size to get a decent representation of the true mean. If you considered each shot a sample, then the shot group of 30 would in fact be a sample size of 30. However, if you treated each seperate group as it’s own sample, then a group of thirty is a single sample. Most datasets are much larger then that, and much more indicative of the mean. I work with millions of records for some datasets, and when we have only 30 records, we try really hard to get more data, because anything below 100 is considered small in industry. Not to mention if your trying to fit a predictive model, such as a neural network or statistical model, you ideally want in the 1000s if you can get that. Now for getting group size, I still use a 30 shot group because it shrinks the confidence interval enough to where I’m ok with the error bounds, and I don’t have the money to afford larger sample sizes. But you have to know that your rifle may be worse by 5, or 10% then what your 30 shot group indicates. A 50 shot group would be better, 100 better then that, and I wish I could get this but a 500 shot group would be much much closer to the true mean.

 

[Tokay444]

I took 8 classes of statistics in my graduate and undergrad and I’m a professional data scientist. Their definitely right. Less then 30 won’t tell you much

They’re*

 

[Tokay444]

This ^. 30 is considered the bare minimum sample size to get a decent representation of the true mean. If you considered each shot a sample, then the shot group of 30 would in fact be a sample size of 30. However, if you treated each seperate group as it’s own sample, then a group of thirty is a single sample. Most datasets are much larger then that, and much more indicative of the mean. I work with millions of records for some datasets, and when we have only 30 records, we try really hard to get more data, because anything below 100 is considered small in industry. Not to mention if your trying to fit a predictive model, such as a neural network or statistical model, you ideally want in the 1000s if you can get that. Now for getting group size, I still use a 30 shot group because it shrinks the confidence interval enough to where I’m ok with the error bounds, and I don’t have the money to afford larger sample sizes. But you have to know that your rifle may be worse by 5, or 10% then what your 30 shot group indicates. A 50 shot group would be better, 100 better then that, and I wish I could get this but a 500 shot group would be much much closer to the true mean.

Than*
You’re*
I didn’t take 30 anything for me to realize this load was a winner. Not 30 shots. Not thirty samples. And since then, it lines up with my ballistic calculator 100% and has yet to fail to put a first round shot on target out to a mile, in a 6BR.

Here’s another load developed in less than 30 samples that I have extreme confidence in. Different gun. Different components. Different cartridge. 6.5Creed.

Those who say it cannot be done, should not interfere with those who are doing.

 

[SanPatHogger]

A 30-shot group is no more useful and sometimes worse than six 5-shot groups. Worse because its hard to measure shots that impact very close together. If shooting really accurately you might destroy the exact point of aim completely.

Take the six 5-shot groups and aggregate them. A little extra math but easier to accurately measure groups.

Hornady added a group analysis section to their app this year. Group Analysis can now be used for the app to calculate a very accurate zero angle solution. The short video below shows how to do that.

Zero angle is very desirable if you are going to shoot in different conditions and from different POA elevation angles. Don't need to re-zero....

Here's the skinny:

That is why you see some people shooting groups an inch or so from their aiming point. If I am shootin a 10 shot group with new 22 ammo I'll dial .5 right and just concentrate on shooting. Then look over at my group.

 

[357Max]

That is why you see some people shooting groups an inch or so from their aiming point. If I am shootin a 10 shot group with new 22 ammo I'll dial .5 right and just concentrate on shooting. Then look over at my group.

Yep I like these targets, just dial up .4-.5 tenths.

 

[beetroot]

I care.

People uninitiated in statistics and data analysis are doing those things in everyday life based on traditions established eons ago by similarly uninitiated people.

Finally, we get sophisticated enough that professionals in the fields of stats and data analysis are bringing them to bear on our discipline which, insomuch as it involves measurement, is inextricably linked to stats, and "nobody cares"?

Sounds foolish to me, but to each his own.

Than*
You’re*
I didn’t take 30 anything for me to realize this load was a winner. Not 30 shots. Not thirty samples. And since then, it lines up with my ballistic calculator 100% and has yet to fail to put a first round shot on target out to a mile, in a 6BR.
View attachment 8222246

Here’s another load developed in less than 30 samples that I have extreme confidence in. Different gun. Different components. Different cartridge. 6.5Creed.
View attachment 8222247
Those who say it cannot be done, should not interfere with those who are doing.

I have defintely had groups that looked like they were ok (sub MOA) during load development but after deciding on that load discovered further down the track that they weren't as good as I expected and must've gottten "lucky" during development.
But every time this has happened it's been were the groups were all pretty marginal to begin with.

For example my crappy 223 load I thought was around .75 MOA but in reality it's nearer to 1 MOA.
Would 30 round groups have let me know the true group size early on, YES.
Would they have helped me find that .75 MOA group I was looking for, probably not, I'm guessing all the groups would've been shite.

In Tokays above example he may haven gottten lucky with that .16 MOA group, but what is the likelihood that you were so lucky that it's actually a 1 MOA load? Pretty unlikely, especially if the groups before and after are a similar size, and the ES is good.

Perhaps all you stastically wizards could tell us the probability of shooting a .16 MOA group from a 1 MOA load?

In reality once you have done a charge test, you will load up a bunch of rounds at the choosen charge weight to verify that it is good at 100yards, then go and verify it at longer distance. By the time you are done with this process you've shot 30 odd rounds and gotten confirmation at distance as well, as well as verify the DOPE at distance.
So whilst 30 @ 100yards might be the minimum amount to be statisically certain that you know what your group size is, in reality you can find the same information in a far more productive way, without wasting reloading components.

With 22lr, yeah I'm going to fire the whole box when ammo testing as it's very easy to cherry pick groups with 22lr (hence the 6x5 threads) and the amount of variabiltiy in 22lr ammo is often painful.

 

[SanPatHogger]

Yep I like these targets, just dial up .4-.5 tenths.View attachment 8222383

Thank you. I'm about to print a few of these at work today.

 

[mrcheesemoose]

Than*
You’re*
I didn’t take 30 anything for me to realize this load was a winner. Not 30 shots. Not thirty samples. And since then, it lines up with my ballistic calculator 100% and has yet to fail to put a first round shot on target out to a mile, in a 6BR.
View attachment 8222246

Here’s another load developed in less than 30 samples that I have extreme confidence in. Different gun. Different components. Different cartridge. 6.5Creed.
View attachment 8222247
Those who say it cannot be done, should not interfere with those who are doing.

How do you know for sure it’s a winner? Could you repeat that group over and over and over again with that load and that gun? Can you repeat this 20, 30 or even 40 times? And also how do you know it’s the load, and not the barrel? I know in many of my larger groups I could cherry pick 5 shots that look exactly like that, yet the group as a whole was kinda meh. I wouldn’t have known that till I added more shots to the group, or until I had been having trouble with hitting things as accurately as I expected.

 

[Tokay444]

How do you know for sure it’s a winner? Could you repeat that group over and over and over again with that load and that gun? Can you repeat this 20, 30 or even 40 times?

Yes.

 

[mrcheesemoose]

Yes.

Pics?

 

[Tokay444]

 

[Tokay444]

Pics?

Anything else?

 

[mrcheesemoose]

Anything else?

First couple pics don’t really say much, but the last pic with the larger group is pretty good. Though what does the larger group with more shots compare to your other loads? How do you know this is because it’s the load, or the barrel?

 

[Tokay444]

First couple pics don’t really say much, but the last pic with the larger group is pretty good. Though what does the larger group with more shots compare to your other loads? How do you know this is because it’s the load, or the barrel?

That’s a Kraft drill target. 1 shot from 4 positions, 3 times.
I don’t know how you differentiate from something being, “the load” or, “the barrel”. Everything is the whole system. Including the shooter.
If a given load doesn’t work in a given system, but a different load does, is it the load, or the system?

 

[SanPatHogger]

View attachment 8225600
View attachment 8225612
View attachment 8225609
View attachment 8225610View attachment 8225611

I love the black diamond. I did that on a 1 inch orange dot one time. Best cold bore shot I've ever made. Then I ruined it by shooting again....

 

[Tokay444]

Clearly I’m leaving a lot of precision and accuracy on the table by not taking a statistical approach to reloading. Or, I can just read groups and don’t need to waste the ammo.

 

[mrcheesemoose]

That’s a Kraft drill target. 1 shot from 4 positions, 3 times.
I don’t know how you differentiate from something being, “the load” or, “the barrel”. Everything is the whole system. Including the shooter.
If a given load doesn’t work in a given system, but a different load does, is it the load, or the system?

You absolutely can tell what parts of the system are causing the issues. You just have to change the variable and test. So if your testing if the load is good or not, you change the load and do testing between it and another load and compare. For statistical comparisons where your trying to see if a certain part is causing a certain value, there’s so many tests you can do to see if it’s statistically significant to tell. T test, p test, confidence interval overlap, etc etc. Engineers and data scientists do this all the time with complex systems so they can diagnose what’s causing what to go wrong, or right. A system only performs as good as it’s individual parts working good all together. Thing is all these tests require a minimum of 30 data points per variable change to tell if there’s any known change. And some of them require hundreds of data points to tell. Which is why the military does acceptance testing for a certain load or rifle with hundreds of rounds per group. With tests like these, you can eliminate other variables causing inaccuracy, such as the shooter or the barrel as long as you keep those variables constant or close to constant. (Same shooter, same barrel, or even put the gun in a vise). And even then some of these tests are robust against external variable changes as long as they don’t change too much. I can shoot 5 shot groups between two loads, and if I compare the two 5 shot groups, I won’t be able to tell if their different. That’s just how statistics works. People can say their able to tell if a 5 or even 10 shot group is a good load, but you can’t really tell until you’ve done serious testing. Again just how statistics works. You might get lucky and get a good load, or have a good barrel that works with a lot of loads anyway, but until you’ve shot that load enough, you just don’t know for sure if that’s true or not.

 

[timintx]

You absolutely can tell what parts of the system are causing the issues. You just have to change the variable and test. So if your testing if the load is good or not, you change the load and do testing between it and another load and compare. For statistical comparisons where your trying to see if a certain part is causing a certain value, there’s so many tests you can do to see if it’s statistically significant to tell. T test, p test, confidence interval overlap, etc etc. Engineers and data scientists do this all the time with complex systems so they can diagnose what’s causing what to go wrong, or right. A system only performs as good as it’s individual parts working good all together. Thing is all these tests require a minimum of 30 data points per variable change to tell if there’s any known change. And some of them require hundreds of data points to tell. Which is why the military does acceptance testing for a certain load or rifle with hundreds of rounds per group. With tests like these, you can eliminate other variables causing inaccuracy, such as the shooter or the barrel as long as you keep those variables constant or close to constant. (Same shooter, same barrel, or even put the gun in a vise). And even then some of these tests are robust against external variable changes as long as they don’t change too much. I can shoot 5 shot groups between two loads, and if I compare the two 5 shot groups, I won’t be able to tell if their different. That’s just how statistics works. People can say their able to tell if a 5 or even 10 shot group is a good load, but you can’t really tell until you’ve done serious testing. Again just how statistics works. You might get lucky and get a good load, or have a good barrel that works with a lot of loads anyway, but until you’ve shot that load enough, you just don’t know for sure if that’s true or not.

That Is all fine and dandy if the gun , barrel break in , throat erosion , shooter strain , wind conditions were a constant ,But in reality these variables are not . The statistical analysis is no doubt valid but the 30 shot sample sizes to confirm a suspected bad load are unrealistic for a common shooter . And to assume that a shooter stops at a few groups to confirm is a assumption and is just not happening like that. Almost every shooter I know will always continuously retest and retest until they are satisfied . Another problem I see here is no one person could shoot a thirty shot group without change occurring during that group . Either shooter strain or barrel temp or wind will skew that outcome . The trick in the civilian world where resources are limited by barrel life , ammo cost , and time is to find a good load in the least amount of time and shots . All while gaining confidence in our system . Just because our tuning is done with small samples does not mean the result is wrong , it just means it is correct at that moment . If I were to to confirm my load with large samples at the beginning it does not mean the gun will shoot that way for the life of the gun. We must always be prepared to change and correct . That just takes experience and new shooters need to take that to heart . If you do a load work up and use large sample size before the barrel is even broke in and think you are done , you are in for a let down . Point being set up the rifle first with small samples and only then confirm with large samples . Life is much easier that way .

TiminTx

 

[carbonbased]

I’m not a reloader. But I assume many get into reloading to save money, among other things.

However, regardless of being a reloader or not, I always get this feeling that many people cannot stand the slow drip-drip cost of shooting ammunition.

They might buy a $3k rifle, a $70k pickup, but that slow subtraction of their money (ammo or gas) just drives them wild.

My brother was that was with gas. He decided to sell his perfectly good pickup and buy a $70k Tesla in order to save money on gas!

He might break even when he’s 70. Meanwhile he can’t drive off-road to hunt!

 

[Seymour Fish]

First couple pics don’t really say much, but the last pic with the larger group is pretty good. Though what does the larger group with more shots compare to your other loads? How do you know this is because it’s the load, or the barrel?

An innate lack of decernment does not gift you with the entitlement to pronounce judgments, while your extremely narrow education prevent sseeing the forest for the trees. The load and barrel are part of a System, the concept of which you fail to grasp. In a man of your claimed caliber this ignorance appears unlikely unless it is willful, leading to the statistical probability that you are an attention-seeking troll.

 

[mrcheesemoose]

An innate lack of decernment does not gift you with the entitlement to pronounce judgments, while your extremely narrow education prevent sseeing the forest for the trees. The load and barrel are part of a System, the concept of which you fail to grasp. In a man of your claimed caliber this ignorance appears unlikely unless it is willful, leading to the statistical probability that you are an attention-seeking troll.

Huh more snowflakes

 

[mrcheesemoose]

That Is all fine and dandy if the gun , barrel break in , throat erosion , shooter strain , wind conditions were a constant ,But in reality these variables are not . The statistical analysis is no doubt valid but the 30 shot sample sizes to confirm a suspected bad load are unrealistic for a common shooter . And to assume that a shooter stops at a few groups to confirm is a assumption and is just not happening like that. Almost every shooter I know will always continuously retest and retest until they are satisfied . Another problem I see here is no one person could shoot a thirty shot group without change occurring during that group . Either shooter strain or barrel temp or wind will skew that outcome . The trick in the civilian world where resources are limited by barrel life , ammo cost , and time is to find a good load in the least amount of time and shots . All while gaining confidence in our system . Just because our tuning is done with small samples does not mean the result is wrong , it just means it is correct at that moment . If I were to to confirm my load with large samples at the beginning it does not mean the gun will shoot that way for the life of the gun. We must always be prepared to change and correct . That just takes experience and new shooters need to take that to heart . If you do a load work up and use large sample size before the barrel is even broke in and think you are done , you are in for a let down . Point being set up the rifle first with small samples and only then confirm with large samples . Life is much easier that way .

TiminTx

That’s why you use the statistical methods that are robust to variability change that I mentioned. Every system has variability in variables. Realistically you can almost never remove that variability in the other variables, just minimize it as much as you can. Engineers and data scientists have been dealing with these exact problems, and come up with solutions to them, for decades. Things wear out, and they have to deal with that. They have created statistical proven solutions to that problem that accounts for and even identifies wear on these systems in their data. A small data set doesn’t really tell you much about your system and what the wear is and how good the load is etc. you need a minimum threshold to understand that. You can shoot five shot groups all day, but really your not getting that much info from them.

 

[TheOfficeT-Rex]

A small data set doesn’t really tell you much about your system and what the wear is and how good the load is etc. you need a minimum threshold to understand that. You can shoot five shot groups all day, but really your not getting that much info from them.

Could you post a couple pictures of your 30 round groups for a good load? TIA.

 

[mrcheesemoose]

Could you post a couple pictures of your 30 round groups for a good load? TIA.

Sure. Two of these are pretty good groups and the barrel I used was a pricey custom barrel. The other two look more like what I see out of standard priced barrels I’ve used.

 

[TheOfficeT-Rex]

Kudos for actually shooting those, how many 30 round groups does it take you to do load development?

 

[mrcheesemoose]

Kudos for actually shooting those, how many 30 round groups does it take you to do load development?

depends. I do a lot of research and modeling with GRT and quick load first so I can get a full understanding of velocity based on powder. Then I’ll do a ladder to check if that velocity model is accurate or not. I then load 30 to the target velocity and shoot a group. Most of the time I can get groups that are within my acceptance criteria first try. But sometimes it’s not that great and I may drop the speeds once or twice before trying a new powder bullet combo in the barrel depending on how bad the group is. I’ve definitely seen that the combo has a much larger effect on group size then powder amount and oal. (Not that those don’t have an effect, just not nearly as much as the combo). The powders and bullets that don’t do so well I use in other combos or just for practice. Best case scenario I have something I’m very confident in with 40 shots. Worst case scenario it may be about 100 shots to either find something I like or throw out that combo.

 

[Tokay444]

You absolutely can tell what parts of the system are causing the issues. You just have to change the variable and test. So if your testing if the load is good or not, you change the load and do testing between it and another load and compare. For statistical comparisons where your trying to see if a certain part is causing a certain value, there’s so many tests you can do to see if it’s statistically significant to tell. T test, p test, confidence interval overlap, etc etc. Engineers and data scientists do this all the time with complex systems so they can diagnose what’s causing what to go wrong, or right. A system only performs as good as it’s individual parts working good all together. Thing is all these tests require a minimum of 30 data points per variable change to tell if there’s any known change. And some of them require hundreds of data points to tell. Which is why the military does acceptance testing for a certain load or rifle with hundreds of rounds per group. With tests like these, you can eliminate other variables causing inaccuracy, such as the shooter or the barrel as long as you keep those variables constant or close to constant. (Same shooter, same barrel, or even put the gun in a vise). And even then some of these tests are robust against external variable changes as long as they don’t change too much. I can shoot 5 shot groups between two loads, and if I compare the two 5 shot groups, I won’t be able to tell if their different. That’s just how statistics works. People can say their able to tell if a 5 or even 10 shot group is a good load, but you can’t really tell until you’ve done serious testing. Again just how statistics works. You might get lucky and get a good load, or have a good barrel that works with a lot of loads anyway, but until you’ve shot that load enough, you just don’t know for sure if that’s true or not.

None of my good loads required a minimum 30 data points. I just know how to read them.

 

[Tokay444]

Sure. Two of these are pretty good groups and the barrel I used was a pricey custom barrel. The other two look more like what I see out of standard priced barrels I’ve used.

Are these at 400 yards?

 

[timintx]

That’s why you use the statistical methods that are robust to variability change that I mentioned. Every system has variability in variables. Realistically you can almost never remove that variability in the other variables, just minimize it as much as you can. Engineers and data scientists have been dealing with these exact problems, and come up with solutions to them, for decades. Things wear out, and they have to deal with that. They have created statistical proven solutions to that problem that accounts for and even identifies wear on these systems in their data. A small data set doesn’t really tell you much about your system and what the wear is and how good the load is etc. you need a minimum threshold to understand that. You can shoot five shot groups all day, but really your not getting that much info from them.

So you are saying that you picked one powder charge / seating depth based on calculations and shot 30 for confirmation and then changed 2-3 more times and done ? Forgive my math here but proper loading procedures require powder charge work up with at least 5 shot groups and more importantly to know where you start to pressure and to watch the groups grow and shrink [50 rounds]. Then I do this 2 more times for a total of 150 rounds. Then a seating depth test 8 differing depths in .003 increments. The depth with the level velocities and small groups wins and then it gets reconfirmed and implemented. 40 rounds for the seating depth test . So basically it took 220 rounds to tune the gun with no hiccups and that is when the barrel is broke in as well. So from a statistics standpoint using standard loading procedures as I have outlined : 240 rounds for the seating dept test and then 300 rounds for powder charge work up. So at 540 rounds you are tuned with a high confidence level? How do you expect to do it in 100 rounds?

Timintx

 

[TheOfficeT-Rex]

I understood it as a 10 round ladder for velocity for the powder model, then using software for picking charge weight and shooting 30 to see if it “meets standards”.

Sounds more like “guess and check” than “load development “ to me. Which is fine, but recognize you’re not playing the same game.

 

[timintx]

I understood it as a 10 round ladder for velocity for the powder model, then using software for picking charge weight and shooting 30 to see if it “meets standards”.

Sounds more like “guess and check” than “load development “ to me. Which is fine, but recognize you’re not playing the same game.

It appears that way , god only knows how many good combos that were never explored. I want to explore the whole spectrum and find the absolute best combo .The only way is to try them all and it also helps to find the width of accuracy windows .

Timintx

 

[diggler1833]

Well, 20 rounds into my newest box of 130gr ELD-M, and I'm already at an ES of .011" in bullet base to ogive. It wasn't just one or two with a significant bell curve, I had 3-4 at or within .001 of each extreme.

Of course, I was just using a Hornady comparator - so maybe that was off (probably is, but I'm adding a little sarcasm).

I hate changing lots.

 

[mrcheesemoose]

So you are saying that you picked one powder charge / seating depth based on calculations and shot 30 for confirmation and then changed 2-3 more times and done ? Forgive my math here but proper loading procedures require powder charge work up with at least 5 shot groups and more importantly to know where you start to pressure and to watch the groups grow and shrink [50 rounds]. Then I do this 2 more times for a total of 150 rounds. Then a seating depth test 8 differing depths in .003 increments. The depth with the level velocities and small groups wins and then it gets reconfirmed and implemented. 40 rounds for the seating depth test . So basically it took 220 rounds to tune the gun with no hiccups and that is when the barrel is broke in as well. So from a statistics standpoint using standard loading procedures as I have outlined : 240 rounds for the seating dept test and then 300 rounds for powder charge work up. So at 540 rounds you are tuned with a high confidence level? How do you expect to do it in 100 rounds?

Timintx

So I did do exactly that awhile ago. And I ended noticing when I plotted these group sizes and velocity distributions, that the 5 round groups were just noise. I didn’t see any trends, and I didn’t see any meaningful outcomes from the data. So even though there were so many different 5 round group loads, it didn’t really help, and actually ended up leading me down the wrong path to accuracy in some cases. So I decided to run some tests with a “known load” I had come up with using the 5 round groups, using larger groups like 30 round groups, and what I was finding was a couple things. First, the groups that I had thought were “accurate” were either exactly the same or slightly worse then the actual best group I had tested with using 30 round groups. And by exactly the same I mean that almost every group I tested using seating depth and powder charge for this particular caliber bullet barrel powder combination, we’re all statistically exactly the same. Second, that small changes in seating depth or powder amount changed things so little that the noise produced by other variables completely outweighed any benefit that those two things provided. And finally after going through like 2000 rounds for these tests I decided that that was an absurd amount of testing, for minute changes, that was completely unnecessary. If your doing benchrest or f class and you want to squeek every ounce of accuracy out of your rifle, then yah the small changes matter and the testing helps. But if your me, and I do prs sometimes, then that doesn’t really matter enough to spend that much money for testing. And I can use less bullets, and get similar outcomes as long as I use enough rounds per group