I have to admit I'm completely oblivious to that audience and whatever they are saying.
- Thread starter Taylorbok
- Start date
I have to admit I'm completely oblivious to that audience and whatever they are saying.
I think they moved on to a new PRS “god” since Scott moved out of country lol
Wouldn't surprise me.
It's why I left the FB precision rifle groups. That crowd is insufferably full of their idolation and worship of their chosen shooters and products. Way too much groupthink.
My personal views about tuners and load devs is that rather than change a bunch of variables and burn through components all the time, most shooters should just load using quality components, and find ways to improve the consistency of their loads. If you have consistent bullets, brass, headspace, neck tension, powder measure, etc you’ll have an accurate load as long as you have the right powder for the case you’re using. Doesn’t really matter if the seating depth is 10 or 25 thou. Doesn’t really matter if the charge is 30.0 or 29.8gr of varget, as long as it’s the same every time.
That’s true but it also takes the fun out of tinkering.
I got tired of the tinkering, and the time spent doing it, and I have stopped a lot of the things I think do not make a big enough difference but I know a lot of guys that enjoy the whole process.
My new load dev is basically the same. Do a ladder and find pressure. Now pick what speed you wanna run. Load 10 at that charge, if it produces acceptable group size and SD you are done. If the ES/SD is bad adjust powder up and down .5-1 gr and see if one of those is acceptable. If ES/SD is good but group is big try adjusting seating depth pick the smallest group and be done. Go shoot
I agree whole heartedly.
By all accounts I've seen or read, Scott is a great guy & a seriously good competitor which to me, is all the more shame that things turned out the way they did. No doubt, Scott quite innocently thought he was developing loads in a time & cost efficient manner, the principal of which may well have a good deal of merit, he just didn't realise that his method couldn't be quite so time & cost effective which, is the crux of this whole debate.
Unfortunately, there's been so much misconstrued in load development because of ignorance. From the significant testing I have conducted over the last 4 years, there is so little difference in load recipes, from a true statistical standpoint that, in reality, I don't believe that there is need to do much at all. My findings are that solid statistical testing reveals very little to no statistical difference when comparing all the test variables against one another. This is the perfect example of how the proper application of statistical analysis can work for us & not against us with regard to time & cost savings in load development. What I mean to say is that, my testing has revealed that in the case where high quality heavy barrels & match chambers, combined with high quality actions & stocks/chassis & etc are employed, the statistics revealed to me, (when adequate sample numbers are employed), it turns out that there is little to no difference between changes in variables tested against one another.
For example, after having conducted these statistically robust tests comparing variables &, realising that there is basically no real difference in group size, I have conducted another series of tests around one load recipe at a time &, found the results to line up the same, regardless of seating depth & charge weight.
The next most significant issue or, possibly a larger issue, in my opinion, is measuring dispersion with group ES. Groups taken on their own with no shot placement aggregation are misleading at best. The only easily attainable measure which is statistically robust is Mean Radius &, when this measurement is understood for what it truly represents, is the best representation of dispersion & the real world implications of dispersion.
From a high quality rifle & components, it is not unusual to measure 3/8" Mean Radius(MR) & much less, within a 10 shot group &, since the real-world goal is to achieve shot placements as close as possible to the POA, group ES measurement is of little consequence, particularly in light of the fact that group ES tells us nothing about the probability of individual shots landing within the measured MR of POA.
If we are going to pump rounds downrange to test our loads, it is far more advantageous to use those shots to align POI to POA because this is where the offset really matters. If the time & material is used to align POI with POA, the POA will always split group ES in half thereby rendering group ES misleading because for the majority of shots will disperse evenly around & within the measured MR 360 deg around the POA which, has the real world effect of roughly halving group ES measurements.
Not withstanding POA to POI alignment, the most important role of MR is in allocating statistical perspective to "flyers". Since MR takes these shots into perspective, it reveals the true significance or, in most cases, the true insignificance of the majority of assigned flyers.
In other words, if the proper testing is conducted & focussed on the real issues, you will see the real world truth & not be fooled by psychologically & visually misleading assumptions.
It’s amazing 99.9% of the hide cursed math before this thread.
2 shot groups
Pet loads
Magical bullet manipulation mid flight
Bigfoot
Lockness monster
Now a couple of guys talk stats and everyone has a doctorate in stats.
2 shot groups
Pet loads
Magical bullet manipulation mid flight
Bigfoot
Lockness monster
Now a couple of guys talk stats and everyone has a doctorate in stats.
Agreed…all the worlds problems could be solved if half as much time was spent on them…I mean statistically speaking.
Not sure if you're referring to my comment or not.
All I'm saying is that I do & have used text book statistical analysis in my testing & proven to myself while using robust analysis, for the most part, there is no significant difference in results between most loads.
As a consequence, I now choose a load configured from GRT, test for pressure then, spend 20 to 40 shots aligning POI to POA. Done.
1000%
Stats would turn off social media and forums immediately
By pure stats 50% of the population is below average intelligence
That means 50% of the members of the hide should not be allow to post…just sayin’
If you look a few pages back, I modeled POI with data Miles provided.
Go to a FB shooting group. 80% can’t construct a proper sentence let alone read a research paper from end to end and actually understand what is being communicated.
As it has been proven, to my satisfaction at least, your statement is more or less correct, at least from the perspective of hindsight. Not so much when abrogated from prior tested proof & applied across the board because it sounds good.
So in the past I spent a shit ton of time testing and shooting and tweaking loads and torque settings and every other thing you can think of…I’ve shot thousands of rounds doing it…BUT unlike all the smart guys here(and no disrespect intended because there are some really smart guys here)I did not do it with computers and graphs and Exel sheets and 5000 round groups.
I shot over a chronograph and seen what happened on target, took in what worked and the fastest way for me to get there, threw out all the stuff I thought was a waste of time and ran with it.
I do the Saterlee thing and have for a long, long time. I find the velocity/charge I’m looking for and adjust seating and done, that simple.
I find 90% of what I’m looking for as I’m playing with the first 100 rounds through the barrel.
I’ll admit it…I am.Haha, you may well be.
Hmmm, now that I think about it.............................
Yea, but he never clarified and in several videos even confirmed that he was looking for "flat spots" in velocity, as well as "positive compensation" at 100yds. With single shots of each powder charge.
So, I agree that he didn't want all the hoopla that came from it, but he definitely thought/thinks that one shot per ladders were showing him usable data.
I’m not 100% sure as I do not follow him but I don’t think he shot ladders he just shot to find velocity and flat spots.
As I said above I shoot 1 shot looking for velocities but I am not looking for nodes or flat spots.
If you’re shooting increasing or decreasing charge weights, you’re shooting a ladder.
Last edited:
Never thought of it like that as I don’t shoot at a target at this point but technically you are right.
Yes, that is basically part of the Satterlee method.
Depending on the level of resolution required, 1 shot per velocity measurement is neither statistically or practically relevant & that, is the basis of our argument.
If you want to be statistically relevant, I would say that no less than 10 shots per charge, depending on the resolution required &, it the resolution required which makes all the difference. eg/ If you want to discern the difference between 0.2 grain increments, a very high sample number is required whereas, if all you need to do is differentiate between 1.0 grain increments, a much smaller sample number may suffice, to the point where maybe a single measurement per charge is adequate. Again, it depends very much on how the information will be used as to the sample number required.
With regard to the accepted use of the Satterlee method, 1 shot (sample) per measurement is grossly inadequate to adequately represent the expected/required resolution deeded.
In stating the above, the required resolution could, IMHO, arguably be considered to be met with as little as three shots per charge weight which, delivers vastly more relevant data than one shot measurements &, whether or not strictly statistically robust, could be argued to deliver, at least some modicum of reality to the results.
Last edited:
Check out his video with the Lily’s. He points to the target he shot and talks about seeing compensation.
Not to be argumentative but, it is guys seeing some difference in the group & then ignorantly assigning a cause for the dispersion they see, simply because it seems reasonable to them. This kind of assumption is the result of poorly reconciled or inadequate data which leaves very little room for measured & substantive cause & effect to become obvious.
In all of this, I've never been an advocate for the unconditional necessity for hundreds of samples in order to produce a practical data resolution however, there's no getting around the fact that there is indeed a viable minimum with regard to any particular testing which will yield useful results.
In my own testing of late & after undertaking a good deal of high sample testing, I see no problem with a significant reduction in the sample numbers required to produce useful results. Agreement on some kind of happy medium would go a long way toward reducing testing to be more realistic with cost in time & consumables as well as gaining useful data with relation to the realistic resolution required.
In summation, I think it reasonable to start from the premise that the Satterlee Method, as currently undertaken by most reloaders is, of so low a data resolution that it is virtually useless & thus misleading. Since even the most miserly data number required to attain a true average with min, max & average is 3 data points per charge weight, I therefore propose that 3 samples per charge weight may well be a balanced place to start.
Last edited:
The issue with 3 or 5 shots, even though it would be well inside most people's "reasonable" amount of shots, is once you run any type of simulation, its all but useless to do much except eliminate "bad" loads.
So, the community would only be agreeing to a minimum number of rounds in s sample size that doesn't really do anything for us.
Run a random number generator for 1000 rounds with a chosen average velocity and something like 7sd. Then randomly break those into 3 shot samples. You'll be all over the place. Unfortunately.
And the same goes for shot group dispersion.
For example, I just did the above. 1000 round sample with a 2850 avg velocity and 7fps SD. Then broke it down in 333 three shot groups.
76 groups were 3sd or less, and 52 groups were 10sd or higher. So, around 38% chance that you will have and SD that looks very low or looks very high when you shoot 3 rounds of ammunition that has an actual 7sd.
So, you'd have to attach something else to the commonly agreed upon 3 round sample. Something like "if you want a single digit SD, you can't eliminate a 3 shot sample that has an SD of 12 or less." Which, IMO, people would lose their minds over.
76 groups were 3sd or less, and 52 groups were 10sd or higher. So, around 38% chance that you will have and SD that looks very low or looks very high when you shoot 3 rounds of ammunition that has an actual 7sd.
So, you'd have to attach something else to the commonly agreed upon 3 round sample. Something like "if you want a single digit SD, you can't eliminate a 3 shot sample that has an SD of 12 or less." Which, IMO, people would lose their minds over.
Well yes & no.
With particular regard to the Satterlee method, I think there is ample room to widen the intervals between charge weights which would significantly reduce the resolution required in order to see what they are looking for. Since most can't relate to probability factors anyhow, I can't see why a 60% or 70% probability factor wouldn't serve as a vast improvement over the current situation. Not that I've seen evidence for velocity flats pots myself but, in the event where there was an actual flat spot within their ladder test, there'd be far more probability of picking up on it than with the current shenanigans.
Last edited:
Additionally with regard to group ES, not withstanding the fact that group ES should be referred to as a secondary tell tail only, there are charts available formulated by Grubbs which serve as very reliable multiplication factors specifically designed to compensate where low or substandard sample numbers are used. From memory, I think they range from resolutions as low 2 samples up to 20 samples or more. They are corrected for Mean Radius, CEP, ES, SD of radial impact dispersal as well as several other commonly used measurement calculations.
If they really want to know the truth, there are a number of statistically verified methods available, without having to use hundreds of samples.
If they really want to know the truth, there are a number of statistically verified methods available, without having to use hundreds of samples.
I'll defer to someone with multiple degrees in the subject like @JB.IC comment on why using methods people think can extrapolate small sample sizes.....really don't.
He might be tired of posting the same thing, so I'm sure you can find his posts on this. It's almost impossible to take 3 and 5 shot samples and extrapolate them to anything usable.
The TLDR of it......it's just not that simple. Or we'd all already be doing that.
I don't believe the tables Grubbs formulated were intended to be direct replacements for robust sampling. The were however, derived from the 10,000 shot samples conducted at the proving ground over some years & are what I believe to be extrapolations of averages of multiple SDs when broken up into the applicable group sample numbers. During Grubbs examinations of the data, he found that multiplication factors could indeed be calculated which can be used to derive close approximations within a calculated & acceptable error tolerance.
From my understanding, the US & other militaries around the world have accepted these tables as deriving results within error boundaries due to their use for small arms accuracy & repeatability assurance trials. As I understand, the current accepted testing regimen is 4 or 5 groups of 7 shots.
I believe the Chinese settled on 4 groups of 8 shots after their own extremely extensive statistical analysis program which I believe was done in the late 90's early 2000's.
One of the major goals of these comprehensive statistical analysis was to determine the most reliable but cost & material effective methods of testing the veracity of small arms as well as artillery & etc in proving accuracy & repeatability.
As previously stated, the application determines the necessary resolution which determines the minimum data points required.
So how EXACTLY do YOU find a satisfactory load? Not your loading procedure but how do you establish a load is good? How many rounds do you shoot until you say this is a solid load?
Also, I am not sure how shooting one of these helps anything.
Maybe someone can elucidate.
Thank you, in advance.
Maybe someone can elucidate.
Thank you, in advance.
You put a target in each rung and when your slow bullets hit the higher targets and your fast bullets hit the lower ones, you’ve complete a positive compensation ladder.
Without getting too deep into weeds and typing a novel. I don't set out with the goal of "the best I can get." For example, I'm not going to do all this testing like people do and then try to pick out the lowest SD or best group. Because I know that it's a very good chance what you pick won't be the best. It just appeared to be the best at the time.
So, I will instead define the worst I'm willing to accept. Let's say I decide I want at worst a 7sd. I'll run sims and such with 3 and 5 shot strings and see about where the cutoff is for somewhere around 80-90% chance I'll be at my minimum SD. Then when I shoot my 3 or 5 shot strings over chrono, I can eliminate the combinations that miss that mark.
What this does, is it ensures that I didn't eliminate a load because it happened to not shoot well in my small sample size. Which tons of people do, and they don't know it. Because they never go back and test the "bad" loads.
After that's done, I'll have X amount of samples that passed my minimum threshold. Since you have to pick a place to start, I'll normally start with the sample that produced the smallest numbers. As I said, you have to pick a place to start, so I just start there. I load a more significant size, 10-30 rounds, depending. If at any point, the SD goes over the threshold I set above in the initial testing, I stop and move to the next.
Once I finish, I'll have only a handful or less charge weights that made it to the "finish" line, and I'll have a fairly significant amount of sample size. I'll have the first 3-5 shots I used initially. And then I'll have the other 10-30 shots. So, I'll normally have a minimum of 15rnds of a charge weight in a handful of charge weights that were acceptable via the limits I set above. I can then decide which to use based on that data.
I do a similar test for seating depth and such.
Sounds like a lot, but it's not that much more than most people use with their 3-5 shot strings. I just move to my stage two with many more charge weights to test, and I use larger shot strings in phase 2.
The main difference is, I'm not looking for the "best" because the odds of finding the best with low sample sizes is not good. I'm looking for "at least this good" and hopefully I get better. But at worst, I'll get minimally acceptable.
As I've said earlier, when you have guys like Cortina or F Class John, they stick with a single cartridge for years at a time. So their loads they are currently using are products of much, much larger sample sizes. They just post the low shot string samples in their videos. If you were to give them a brand new, never used cartridge, they very likely wouldn't find the "best" within the first barrel or three. They would however, progressively work their way into amazing ammo as they always do.
That part is always left out of people's consideration when they reference these guys. People think they can take their 22-250, do a little bit of testing, and find ammo combinations that shoot as good as Cortina's 7mm XYZ (whatever he's been shooting for a while at the time).
For example, let's say I was going to use 3 shot strings, and I wanted a 7sd at worst. I'd likely take anything that has around a 10sd or less to phase 2. Maybe 11 or 12sd, but probably 11 at the most.
Where people looking for the "best" SD possible might, through luck of the draw, see a lot of 5sd or less in their 3 shot samples.....and toss a lot of charge weights in the garbage that have a 7-12sd. And they would be throwing a lot of charge weights away that may very well be good enough or even the "best."
As if we run a sim of 1000rnds with a real 7sd, you can easily have 50 or more of those 3 shot strings that have an SD of 10 or more. That's a 15% chance. Which is quite high as far as odds go.
Posting to add, once I move into the second stage with larger sample sizes, I will rerun sims.
I.E. if I'm going to load 10rnds, I'll now run sims on 13 or 15rnds(the 3 from the initial test, or 5 if I used 5, plus 10 more) and find my threshold.
Once you go from small sample sizes to significantly larger sample sizes, the thresholds get smaller. Instead of say 11sd and below, I'll now be somewhere around 8 or 9sd as the threshold.
So, if I'm shooting my string and it goes too high to recover, say it goes up to 15sd or something and its unlikely it will recover to 9 or better, I just stop and move to the next.
And you can also tighten up the "worst" you'll take. You could decide that now you'll only take 6 or 5sd as your target goal instead of 7. Run sims with that goal, and then come up with your threshold for failure.
And lastly, this all assumes that it's a cartridge for myself or a customer that requires or wants full on load development.
If you're gonna go to say a PRS match shooting 2moa targets off barricades, you can literally take your choice of 6mm, look in the loading section for a common charge weight and seating depth, load it, and you won't be missing because of your ammo.
If you're looking to be competitive at 1k F Open, different story.
If you're gonna go to say a PRS match shooting 2moa targets off barricades, you can literally take your choice of 6mm, look in the loading section for a common charge weight and seating depth, load it, and you won't be missing because of your ammo.
If you're looking to be competitive at 1k F Open, different story.
I don’t follow the work of Grubbs and never once heard his name mentioned
in any of the 60+ credit hours of statistics.
You can estimate population SD off ES for low sample sizes. But it’s an “on average” method which means you still need to shoot multiple groups to make the estimation work properly. Around 25 groups would be adequate. So, it’s not really a small set of observations anymore.
The problem with groups sizes is that it’s at a minimum two dimensions in Euclidean space at 100yds, which I think is enough, and three dimensions at larger distances. So not only is there variation in the x & y axes but also z. Probably getting into 6DOF if someone wants to get complicated. Thats out of my knowledge though. If one really wants to keep track of the group dispersion, then they have to use a statistic that maintains the memory of the POI.
^^^ This right here.
In my limited knowledge and not having reloaded, this makes the most sense. Many people are going to shoot a small group or two. Take the best one at the time and think it is the best for all time. And when they run into a larger group down the road, they will blame everything but the sample size. Jayden Quinlan in the group size episode spoke to this tendency to take the blame for ammo problems and rifle problems. People will shoot a flier and say "that was me."
How was it "you"? Did you jerk sideways? No? Did the rifle fall off the rear bag or did one of the bipod legs collapse? No? Then how was it you?
Like you said, the more practical perspective is to define what is acceptable error, which means bigger sample size and have the majority of your hits within that margin of error and then you know how the rifle is acting and what you need to do in the wind. I watch the Texas Plinking challenge and most of the guys get good elevation but there are variances from bullet to bullet. Where they really have problems is the wind. That being said, if you already have the parameters defined of the load you are working with, you know better now whether it is best to stay with same hold left or right, for example. Or just favor left or right.
Because what happens if someone accepts a small size and thinks that is the best and it turns out not to be? Granted, that is hypothetical in my mind.
Last edited:
Here's a link to an easy to read extract from Geoff Kolbe.
Just copy the URL as text & paste in google. It should go straight to the page on Geoff's site
When I was studying statistical analysis of rifle accuracy, I couldn't avoid Grubbs work. Just saying.
With reference to Grubbs tables, Grubbs was using the measured SD values to determine the multiplication factors of different sample numbers. No estimating involved as I understand it.
I'm also interested in this 3rd axis you refer to. I've not seen it mentioned in reference to 2 dimensional targets with regard to statistical analysis of rifle accuracy.
Yea, there's just too many things that can go "wrong" to be eliminating ammo the way most do.
I just ran a sample of 1k shots with a 7sd and a 5sd and used 3 shot samples for SD. The 5sd ammo had 252 groups with a 5sd or less. The 7sd had 151 groups with a 5sd or less. With the way most people look for the "best," that's a lot of overlap which results in a very high % chance of low sample sizes causing you to believe the worse ammo is actually better than the best ammo.
Change it to a 10sd and you still have around 100 of those three shot groups that will register in at 5sd or less.
So, you can have literally twice the SD, and still at a very high % of the time, have data from 3 shot groups showing as good or better SD than the twice as good ammo.
Last edited:
I've repeated Satterlee tests for 35 cycles and Jeff Siewert has done 100 cycles.
When you plot them out it's a gigantic mess of criss-crossed lines. The individual Satterlee plot bounces around within the true average +/- ES for any given charge weight and any "node" that appears in an individual test is a "pattern" that your mind is tricking you into believing in.
I don’t want to sound like a snob, but anyone who has studied statistical theory at the graduate level would have came to similar conclusions in a short amount of time as what’s mentioned in that link. I even showed a round about application of it except I made it multivariate distribution and I’ve never read his work. I guess he might have been the first to apply it to firearms?
Edit: I acknowledge I’ve had the benefit of standing on the shoulders of giants so it’s not really fair to compared now VS then.
Last edited:
