Little Crow Gunworks: Avg POI Is More Important Than Group Size. Thoughts?

[WeR0206]

This might be a bit controversial but he makes some good points. You need to look at both precision (mean radius/ES) and accuracy (API). To give some context his load development process he does use group size but its not the main focus.

He starts with testing 2-3 different powders with his desired bullet for the use case. For 308 he will use gordons reloading tool to find the charge weights for 50K to 62K and go in .2gr increments and load one shot at each increment. He will do this process for each powder then shoot 10-15 round groups (a single POA for each powder). The point of this is to see how temp/pressure sensitive each powder is.

Here he tested 3 powders:

He Id’d powder C as the best (look for group size < 2”) then repeated the same charges for powder C as step 1 but 3 shot groups for each increment instead 1.

The red numbers represent different temps. Look for 4-5 charge weights in a row that have similar API and low vertical stringing and the target in the middle will be your charge (see stars)

Here’s a nice AI summary of differences between mean radius and API:

 

[ShtrRdy]

That's pretty much how Optimal Charge Weight method works.

 

[patriotnation]

I have watched this whole series and it has confirmed much of what I know and do, while also learning a lot. It makes sense to me that before group size, you must consider POI first. After all, what good is that one tiny group if it is much farther away from your point of aim?

 

[ma smith]

This is statistics 101. You would only use the ES to "guess" the mean radius.

 

[WeR0206]

I have watched this whole series and it has confirmed much of what I know and do, while also learning a lot. It makes sense to me that before group size, you must consider POI first. After all, what good is that one tiny group if it is much farther away from your point of aim?

+1 on this. I think his series of reloading videos is fantastic so far. For example I had no idea gordons reloading tool even existed and its so helpful.

 

[patriotnation]

Yep, it convinced me to download it and learn to use it.

 

[memilanuk]

I'll admit, I'm probably - no, *definitely* - guilty of 'deciding' on the powder/bullet based on factors other than actual testing, and then trying to 'make' them shoot. In my defense, that was usually because a) I was shooting some combos that were pretty well mapped out by others (308/223 + Varget + pretty much any bullet, or 6CM / 6.5CM + H4350), or b) there were external 'requirements' (team sponsors, etc.).

That said, I am intrigued by both the way he uses GRT to filter/winnow the powder options, and doing the aggregate ladder test to explore the overall grouping tendency of the selected powders. I might have to tinker with that some myself.

The other bits about POI vs raw group size are, as others have pointed out, is basically OCW redux... but he does a good job covering how/why the POI moves and what to actually look for.

 

[richthe1]

Is finding a “node” a legit thing or no? I could be remembering wrong, but I feel like he says yes and Brian Litz says no?

 

[memilanuk]

All the videos are free for the watching on YT.

 

[WeR0206]

Part 14: deep dive on ladder testing

 

[Seymour Fish]

I'll admit, I'm probably - no, *definitely* - guilty of 'deciding' on the powder/bullet based on factors other than actual testing, and then trying to 'make' them shoot. In my defense, that was usually because a) I was shooting some combos that were pretty well mapped out by others (308/223 + Varget + pretty much any bullet, or 6CM / 6.5CM + H4350), or b) there were external 'requirements' (team sponsors, etc.).

That said, I am intrigued by both the way he uses GRT to filter/winnow the powder options, and doing the aggregate ladder test to explore the overall grouping tendency of the selected powders. I might have to tinker with that some myself.

The other bits about POI vs raw group size are, as others have pointed out, is basically OCW redux... but he does a good job covering how/why the POI moves and what to actually look for.

Monty, if it won’t show a solid node across at least a grain/grain and a half I’ll move to the next powder.

 

[richthe1]

Not statistically significant though, correct? Or am I missing something?

 

[memilanuk]

am I missing something?

Did you actually watch the video(s)?

 

[Seymour Fish]

Not statistically significant though, correct? Or am I missing something?

Is a thousand poorly executed shots “ statistically significant”. Abandon your false god

 

[Mean_Man]

Hornady has a YT channel, where they cover this general topic in extreme depth, just throwing that out there.

 

[Csafisher]

Picking “nodes” based on 3 shot groups is a joke.

 

[Holliday]

Picking “nodes” based on 3 shot groups is a joke.

I shoot a 100 yard group then pitch my powder, bullets and rifle in the river and start over….But that’s me

 

[memilanuk]

Hornady has a YT channel, where they cover this general topic in extreme depth, just throwing that out there.

No, they really don't. Their load dev amounts to pick a cartridge, pick a powder, pick a bullet, set the seating depth to some random value then never touch it again, and run up the powder charge to where you think the velocity should be. If you call that 'in extreme depth'... good for you.

 

[Csafisher]

No, they really don't. Their load dev amounts to pick a cartridge, pick a powder, pick a bullet, set the seating depth to some random value then never touch it again, and run up the powder charge to where you think the velocity should be. If you call that 'in extreme depth'... good for you.

Based on their testing with large sample sizes in the context of PRS and hunting. Is that hard to comprehend?

 

[memilanuk]

Nope, not at all. Also not the subject of this thread. Moving on...

 

[WeR0206]

Picking “nodes” based on 3 shot groups is a joke.

If you have the time and money feel free to do bigger groups. You’re looking at this part of the process out of context. The goal of the 3 shot groups is to find a rough estimate (without spending a fortune) for the avg POI and vertical stringing of each charge. Once you find the series of 5 or so charges with consistent POI and low vertical stringing you can test with larger sample sizes of 10+ at each charge. At that point you can also play around with COAL.

 

[Csafisher]

If you have the time and money feel free to do bigger groups. You’re looking at this part of the process out of context. The goal of the 3 shot groups is to find a rough estimate (without spending a fortune) for the avg POI and vertical stringing of each charge. Once you find the series of 5 or so charges with consistent POI and low vertical stringing you can test with larger sample sizes of 10+ at each charge. At that point you can also play around with COAL.

Sure I understand. But you’re pissing into the wind because there is so much variance with 3 shot groups you can’t “actually” say which one is best. You could shoot these tests 5 times and come to 5 different conclusions.

 

[WeR0206]

Sure I understand. But you’re pissing into the wind because there is so much variance with 3 shot groups you can’t “actually” say which one is best. You could shoot these tests 5 times and come to 5 different conclusions.

I completely understand your point and I initially looked at it that way as well but you need to change how you look at the groups. If there was that much variability you wouldn’t see 4-5 groups in a row with similar API (essentially a 10-15 shot group of varying pressures). Is a 15 shot group big enough?

The point of these tests is to find a load that will have a similar POI and low vertical stringing regardless of temp/pressure swings (simulated by changing charge weights).

 

[Csafisher]

I completely understand your point and I initially looked at it that way as well but you need to change how you look at the groups. If there was that much variability you wouldn’t see 4-5 groups in a row with similar API (essentially a 10-15 shot group of varying pressures). Is a 15 shot group big enough?

The point of these tests is to find a load that will have a similar POI and low vertical stringing regardless of temp/pressure swings (simulated by changing charge weights).

OCW is a waste of time. It’s like reading tea leaves. For long range shooting your far better off spending time to get your SD down into the 6-10 range.

 

[WeR0206]

OCW is a waste of time. It’s like reading tea leaves. For long range shooting your far better off spending time to get your SD down into the 6-10 range.

Yes getting SD down is important (single base extruded powders with > 98% case fill, etc.) but how do you keep your given load shooting with a consistent POI etc when the temp goes up and down? Even the “temp stable” powders aren’t impervious to temp swings.

 

[Csafisher]

Yes getting SD down is important (single base extruded powders with > 98% case fill, etc.) but how do you keep your given load shooting with a consistent POI etc when the temp goes up and down? Even the “temp stable” powders aren’t impervious to temp swings.

I haven’t noticed any issues. I try get a velocity in the cold and in the hot. And then input that into the kestrel.

 

[Emerson0311]

Yes getting SD down is important (single base extruded powders with > 98% case fill, etc.) but how do you keep your given load shooting with a consistent POI etc when the temp goes up and down? Even the “temp stable” powders aren’t impervious to temp swings.

Most ballistic calculator have the ability to set up a temp vs velocity table, or calculate a modifier.

 

[Niles Coyote]

No, they really don't. Their load dev amounts to pick a cartridge, pick a powder, pick a bullet, set the seating depth to some random value then never touch it again, and run up the powder charge to where you think the velocity should be. If you call that 'in extreme depth'... good for you.

Interesting... That's not at all what I have understood from their podcast and videos.

Hornady:
>Decide what your purpose is and pick a chambering and bullet that fits that need.
>Reasearch and use data available in print and online to choose a few powders that fit the burn rate needed for that chambering and have the attributes you will need for your purpose... that is if you don't already know what you want to use.
>Load charges low, mid and higher of each of the powders chosen. Pick the powder based on which shoots the best with all three loads tested in a single powder as this is the most likely suited for your chamber and bullet.
>Set seating based on data if available to begin with. *

It's hard to recall who I have heard say what beyond that as I am a consumer of information, but the above were all stated by Hornady.

*I do recall Jayden or Miles also speaking of doing seating depth testing in large, measured steps rather than small amounts.

I am 50% sure I heard them say doing a seating depth test before powder selection is an option, but I might be confusing that gem with a trusted friend who has a significate amount of lab testing time.

 

[diggler1833]

Load development, barrel cleaning, leveling your scope, etc... etc...

Chances are that whatever you're doing - even if it is giving you the results you desire - is "stupid" by the next guy's opinion, and you are a moron for continuing to do it NOT his way.

I just let all of these fancy new methods of reinventing the wheel go in one ear and out the other, as long as I'm happy with the consistency of my targets.

If you're starting from scratch: As far as picking who's advise to follow, it probably wouldn't hurt to follow whoever is winning all the matches in the discipline you're interested in... and do what they're doing.

 

[Gogga]

Load development, barrel cleaning, leveling your scope, etc... etc...

Chances are that whatever you're doing - even if it is giving you the results you desire - is "stupid" by the next guy's opinion, and you are a moron for continuing to do it NOT his way.

I just let all of these fancy new methods of reinventing the wheel go in one ear and out the other, as long as I'm happy with the consistency of my targets.

If you're starting from scratch: As far as picking who's advise to follow, it probably wouldn't hurt to follow whoever is winning all the matches in the discipline you're interested in... and do what they're doing.

Exactly, save yourself a lot of trouble by starting with what's winning. If you can improve on that you'll be very good, very quickly!

 

[Mean_Man]

No, they really don't. Their load dev amounts to pick a cartridge, pick a powder, pick a bullet, set the seating depth to some random value then never touch it again, and run up the powder charge to where you think the velocity should be. If you call that 'in extreme depth'... good for you.

We are clearly talking about different videos, I don't know where the hell you got that from.

Hit probability
Your groups are to small (sample size)
Let's talk load development
Let's talk mean radius
Let's talk load data
Ammunition demystified


There's others but that's all I'm going to bother to link to for now.

 

[Seymour Fish]

Picking “nodes” based on 3 shot groups is a joke.

Says the Chicken Sh-t Army

 

[Csafisher]

Says the Chicken Sh-t Army

Says pretty basic statistics….

 

[Seymour Fish]

OCW is a waste of time. It’s like reading tea leaves. For long range shooting you’re far better off spending time to get your SD down into the 6-10 range.

Lmfao !

 

[Seymour Fish]

I haven’t noticed any issues. I try get a velocity in the cold and in the hot. And then input that into the kestrel.

Roflmfao !

 

[Marine52]

Based on their testing with large sample sizes in the context of PRS and hunting. Is that hard to comprehend?

Large sample? When did they ever use a proper large sample?

 

[Csafisher]

Large sample? When did they ever use a proper large sample?

You’d have to watch their videos. Mentioned in several.

 

[Csafisher]

Lmfao !

Ok boomer. Stick to your fudd lore

 

[patriotnation]

That didn’t take long

 

[Marine52]

You’d have to watch their videos. Mentioned in several.

I watched their podcast accusing shooters of using groups that are too small. His knowledge of statistical methods appears nonexistent.

 

[WeR0206]

I watched their podcast accusing shooters of using groups that are too small. His knowledge of statistical methods appears nonexistent.

I got some good knowledge from the hornady episodes but to me the thing they completely ignored was they didn’t mention or look at average point of impact (Accuracy) all, they only really looked at extreme spread or mean radius, which are important but it only measures precision.

If you find a load that is accurate and precise then your are golden.

 

[Marine52]

I got some good knowledge from the hornady episodes but to me the thing they completely ignored was they didn’t mention or look at average point of impact (Accuracy) all, they only really looked at extreme spread or mean radius, which are important but it only measures precision.

If you find a load that is accurate and precise then your are golden.

Yes. Miles, in one statement, indicates a better stat would be mean radius but didn’t follow up.

The Hornady podcast on sample size 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 ES 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.

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.
However, Hornady chose not to do this. They shot one group and generated one number and falsely called this a large sample.

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

 

[WeR0206]

Yes. Miles, in one statement, indicates a better stat would be mean radius but didn’t follow up.

The Hornady podcast on sample size 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 ES 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
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.
However, Hornady chose not to do this. They shot one group and generated one number and falsely called this a large sample.

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

Great post. If you’re looking only at extreme spread it doesn’t matter how many shots are in your group, it’s still a sample size of one (the ES between two of the 50 shots).

When you think about it a lot of precision shooting boils down to understanding statistics, probabilities, etc.. What you really need to do is do enough testing so that you can fill in as much of the bell curve as possible (2 to 3 sigmas worth). If you can figure that out then you have a 95+% chance or whatever it is that your POI will be within a certain diameter. As you said with a normal distribution, a sample size of 30 would be considered adequate (this could be 30 three shot groups, etc).

 

[WeR0206]

Happt Fathers Day everyone!

Ep 15 seating depth testing. He notes that all of this testing isn’t to find the final load but only a contender that warrants further testing at long range 800+ yards.



Depth tests with lower speeds:

Higher speeds (he choose 43gr @ 2.247” to test at long range):

 

[stats_guy]

Yes. Miles, in one statement, indicates a better stat would be mean radius but didn’t follow up.

The Hornady podcast on sample size 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 ES 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.

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.
However, Hornady chose not to do this. They shot one group and generated one number and falsely called this a large sample.

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

This is a great point but I'd give Hornady the benefit of the doubt and believe that they were talking about measuring each shot as an observation (with an x and y component), so a 50-shot group is indeed a large sample that produces meaningful population inference for that particular rifle and ammo/load.

Comparing variances between groups with statistical rigor is where things get more complicated. A typical approach would be an F-test assuming a group is an isotropic bivariate normal distribution, but that becomes somewhat unstable at small sample sizes that we typically do load development with (say 5 shot groups). The highest level of rigor would be to use a Bayesian approach that incorporates priors from that particular rifle (something like a weakened prior from the previous group weighted at the sq. root of the no. of shots), generates posterior distributions of the precision parameter (say, Mean Radius) for each group being compared, and then compares those posterior distributions to determine if Group A is more accurate than Group B.

This is something I've been working on for a while, as part of a mobile app I'm making that is aimed at adding statistical rigor to load development and removing the need for guesswork.

 

[WeR0206]

This is a great point but I'd give Hornady the benefit of the doubt and believe that they were talking about measuring each shot as an observation (with an x and y component), so a 50-shot group is indeed a large sample that produces meaningful population inference for that particular rifle and ammo/load.

Comparing variances between groups with statistical rigor is where things get more complicated. A typical approach would be an F-test assuming a group is an isotropic bivariate normal distribution, but that becomes somewhat unstable at small sample sizes that we typically do load development with (say 5 shot groups). The highest level of rigor would be to use a Bayesian approach that incorporates priors from that particular rifle (something like a weakened prior from the previous group weighted at the sq. root of the no. of shots), generates posterior distributions of the precision parameter (say, Mean Radius) for each group being compared, and then compares those posterior distributions to determine if Group A is more accurate than Group B.

This is something I've been working on for a while, as part of a mobile app I'm making that is aimed at adding statistical rigor to load development and removing the need for guesswork.

If you ever get that app going definitely let us know, I'd be down to try it out.

 

[RegionRat]

What individuals with limited resources can do, versus what big defense can do, means that the vast majority of folks are forced to make decisions on scant data.

That means taking some short-cuts and accepting some risks is an economic reality.

That said, LCGW isn't naive and has a pretty good grasp on the risks he is taking with small sample methods. He has gained the admiration and respect of several professional ballistics and weapons folks and is very good about taking feedback.

The LCGW videos should be judged in part by the final results and taken as a whole.

It is easy to criticize sample sizes and load development methods, but it is another thing to be willing to pay for them when you don't agree.

If we are to make progress (in the hobby) and accept that there is always some risk with these small sample methods, then what he is showing should be judged on his results.

He is doing a pretty good job of no-nonsense work without wasting many shots when the varied rifles and chamberings are taken into account.

After all, most of the rifles and calibers he is showing are not F-Class or PRS rigs that follow a narrow pattern with a good history.

He never claims an uncertainty is good with a given 3 shot group, but he is skilled in spotting trends of several sequential three shot groups and is typically running a decent quality rifle.

At times I wish he wouldn't work with low quality brass, but unless I am sending him a check to cover it, I have to shut up and stay in the bleachers and cheer for him instead of whine about that brass or the sample sizes. YMMV

 

[flogxal]

I found the most valuable part of his longer multi-part reloading series was the way he used GRT to help guide reloaders. Maybe that's because I was already using GRT when I found his channel, but everything he describes GRT as doing, I found the same results at my end.

I'm still not keen on his resizing with industrial force method but then my resizing isn't great and maybe it's my 98 lb weakling press operation. If I were planning on that kind of pressure/striking, I would create a much sturdier press/bench setup and I'd be attaching hydraulics and a timing system to do the work for me. Meaning, his method seems for a heavier set of tools.

 

[RegionRat]

Many folks tend to size too slowly, and that opens the door for a different kind of friction problem.

If you have ever seen the speed of the machines in a factory, they are not moving slow. By that same token, they are not going as hard as LCGW either.

What it takes is a pretty good velocity into the stop without slowing down and getting into what is called a "stick-slip" friction problem. The really hard slam into the stop isn't really required, but if that is what it takes to avoid the friction regime change for you, then slam away.

You can practice with cheap brass and check your headspace. The problem shows up as variation in that shoulder datum length. It is better to go a little faster than it is to go a little slower and accidentally get into stick-slip.

Make sure you don't get sloppy with case lube, or you also introduce problems. More is not always better, and too skinny can also be bad.

Brass is magical in terms of how well it is suited to cartridge use. It has a shear storage, and shear loss term built into the modulus, as well as a damping term. When you cold work it, you want to be as consistent as possible with your speed while avoiding that stick-slip which is like dozens of starts and stops. YMMV

 

[stats_guy]

If you ever get that app going definitely let us know, I'd be down to try it out.

Will do - it's a work in progress but here's a sneak peek at what the group variance comparison feature will look like: