Statistical variation of the Dillon powder thrower

[Nature Boy]

I was doing some loading this morning and decided to record the data on what the Dillon throws

Powder was H4895
Scale was an A&D FX 120i (0.02g resolution)
Population = 15
High = 26.06g
Low = 25.74g
Sample SD = 0.0979
Mean AVG = 25.89g

99% confidence level = 25.83g - 25.96g

 

[Laphroaig]

Thanks for posting that.

A fellow shooter remarked that he thought his Dillon was more accurate than his Harrell. Don't know the load specifics but that's an impressive recommendation nonetheless.

 

[WelshShooter]

Surprised that the powder thrower is that consistent with extruded stick powder.

What software you using to calculate confidence limits? Minitab, R, JMP or something else?

 

[Rule3]

Been "100" years since I took statistics. Is a sample of 15 a valid sample?
What about other powders? one powder does not a test make
I gather the target was 26 grains

Not doubting the accuracy.

 

[Howland]

Excel will calculate SD. Mean -3SD to Mean +3SD gets 99% confidence.

 

[Nature Boy]

Excel will calculate SD. Mean -3SD to Mean +3SD gets 99% confidence.

That’s what I used

 

[Nature Boy]

Been "100" years since I took statistics. Is a sample of 15 a valid sample?
What about other powders? one powder does not a test make
I gather the target was 26 grains

Not doubting the accuracy.

My purpose for doing it was for no other reason than curiosity. My target was a few grains under the weight I was trickling up to

 

[rfwobbly]

Great report !

What I would like to ask is how many "drops" were made before measurement started? IOW, was this the first 15 drops after filling the hopper, or did you drop (say for instance) 20 loads before you started recording.

I've always loved the Dillon measures and think they are highly innovative.

 

[Nature Boy]

What I would like to ask is how many "drops" were made before measurement started? IOW, was this the first 15 drops after filling the hopper, or did you drop (say for instance) 20 loads before you started recording.

I already had powder in the hopper from a previous session and it was about half full. The first thrown was the first measured (25.96g)

 

[Telekinesis]

Interesting numbers - much more accurate than I was expecting.

Been "100" years since I took statistics. Is a sample of 15 a valid sample?
What about other powders? one powder does not a test make
I gather the target was 26 grains

Not doubting the accuracy.

Technically speaking you'll need a minimum of 30 samples to be able to assume your distribution is normal (thus allowing you to use the mean and standard deviation to calculate your confidence interval). There are of course formulas to determine exactly what a sample size needs to be, but 30 is a useful approximation.

 

[rfwobbly]

I already had powder in the hopper from a previous session and it was about half full. The first thrown was the first measured (25.96g).

Wow, then those really are significant numbers, and an astounding SD.

You said you recorded this in MS Excel. Excel will also allow you to graph the results. A plot of drop number vs. weight would show if most of the deviation came in the first 5 throws while the measure "settled in". That would be an interesting plot to see.

 

[Typetwelve]

Like I always start with, I"m new to reloading, so my equipment is brand new stuff.

With that said, I've ran a good 2000 or so loads of 38 and 44 special through my 550. I'm loading very light 38 special loads with 3.1g of titegroup. The most variance I've seen from the thrower is possibly a +.1gr throw. In lots of 10 throws, I've yet to see anything over 32 grains total. At times, it will be 31 exactly. I've yet to see any light loads.

With 44 special, I'm using 6.0 grains of Unique, the variance with it is even smaller than titegroup. Pulling multiple runs of 10x throws, I've yet to see anything much over 60 grains. I believe that most I've seen was 60.3. I will say that unlike the Titegroup, the Unique can sometimes throw under, but we're talking a -.1 max. I believe I saw one load at 5.9.

My stuff may wear out and begin to drift more, but so far I've very pleased with it.

 

[Olon]

Been "100" years since I took statistics. Is a sample of 15 a valid sample?
What about other powders? one powder does not a test make
I gather the target was 26 grains

Not doubting the accuracy.

I wondered the same.

Currently taking a statistics course so I'm not the expert but in order to use a normal distribution and standard z scores my instructor set an arbitrary minimum sample size of 30.

But this is real life and the stakes aren't very high so I guess it's acceptable for me. Now if we're claiming this is superior to some other powder measure a larger sample would be in order I think.

 

[ATLDave]

IIRC, most of the standard sample size guidelines and a great many of the statistical tools assumes a standard distribution (bell curve).

If you're dealing with a phenomenon with a non-normal distribution, some of the stat tools don't work the same.

Let's say that we had a very good powder measure that worked well 99% of the time and gave normally-distributed results for roughly 99 out of 100 times. But approximately 1% of the time, powder bridging occurred and the drop weight was off-the-charts low... A graphed curve of results would have a bell-curve centered around the desired drop weight... and then a spike well off to the left. A sample size of 30 wouldn't be enough to consistently catch that dynamic, much less accurately characterize/quantify it.

Statistics is "real world" enough to have a lot more "yeah buts" than most branches of math!

 

[RugerOldArmy]

I like Dillion equipment, and have 7 powder measures for my 550.

I'm not surprised by your results.

But I'd be very surprised if throwing charges of IMR-4064 came close. It is like trying to meter twigs.

Are H-4895 grains shorter than IMR-4895 grains?

 

[Sniper66]

I'm getting a headache.......

 

[RandyP]

I was promised there would be no math! …..lol

 

[ATLDave]

I was promised there would be no math! …..lol

 

[Nature Boy]

Now if we're claiming this is superior to some other powder measure a larger sample would be in order I think.

No intention of making any claims. Just simple curiosity and thought I’d share the data with others who might also be anal retentive types like yours truly

 

[Nature Boy]

Wow, then those really are significant numbers, and an astounding SD.

You said you recorded this in MS Excel. Excel will also allow you to graph the results. A plot of drop number vs. weight would show if most of the deviation came in the first 5 throws while the measure "settled in". That would be an interesting plot to see.

I’ll do that and post the graph. Better yet, when I get ready to load up some for a match coming up this weekend I’ll run the numbers again for 75 rounds.

As I was saying on the original post, I was using my new Auto Trickler to finish off the charge. I thought about running the stats on that but of the 15 only 4 deviated from the 30.60g target weight. 3 were 30.58g and 1 was 30.62. Variation would be ridiculously low.

 

[Olon]

No intention of making any claims. Just simple curiosity and thought I’d share the data with others who might also be anal retentive types like yours truly

Yep I didn't see your post as making claims, just throwing out my $.02 in the hypothetical situation of drawing conclusions based off of it. I've thought of doing this with the length of my soft point rifle bullets so I can get a good gage of the consistency of my seating die. The difference between the variance of COAL and that of the bullet length would be a decent evaluation of how precise the seating is unless I'm mistaken. Assuming all my brass was trimmed to the same thousandth of an inch...

Props to you for actually doing it, whereas I've only ''thought about it''

Thanks for posting.

 

[Nature Boy]

I've thought of doing this with the length of my soft point rifle bullets so I can get a good gage of the consistency of my seating die

If you don’t have it, get the Hornady bullet comparator kit and measure off the ogive instead of the OAL. That takes a lot of variation out of it and you can get consistent measurements for seating depth. The only time I’m measung OAL now is when I’m trying to fit the magazine

https://shop.brownells.com/reloadin...t7vx4SundmpKubA-bEL7-lCeyL7AYfc0aAsqTEALw_wcB

 

[WelshShooter]

That’s what I used

Ah I see what you mean. To be pedantic, 99.7% of your population falls in the +/-3 sigma window so you gained an extra 0.7%

IIRC, most of the standard sample size guidelines and a great many of the statistical tools assumes a standard distribution (bell curve).

If you're dealing with a phenomenon with a non-normal distribution, some of the stat tools don't work the same.

Let's say that we had a very good powder measure that worked well 99% of the time and gave normally-distributed results for roughly 99 out of 100 times. But approximately 1% of the time, powder bridging occurred and the drop weight was off-the-charts low... A graphed curve of results would have a bell-curve centered around the desired drop weight... and then a spike well off to the left. A sample size of 30 wouldn't be enough to consistently catch that dynamic, much less accurately characterize/quantify it.

Statistics is "real world" enough to have a lot more "yeah buts" than most branches of math!

If you have a non-normal distribution you may analyse the median value rather than the mean using a non-parametric test.

With regards to the powder bridging comment, this can be detected by the user. Whenever I've used Vihtavuori stick powders in a powder hopper such as the RCBS Uniflow, I can tell that I will get a different mass of powder when the handle binds up. Therefore, I dump this powder back in and only throw charges when the handle doesn't bind. Going back to the bell curve, you can remove outliers if you have an assignable cause, in this case it could be the user "feeling" a difference.

The biggest question people always ask is "how many samples do I need?". From a statistics point of view, it all depends on what you are trying to achieve.

 

[ATLDave]

Going back to the bell curve, you can remove outliers if you have an assignable cause, in this case it could be the user "feeling" a difference.

Yeah, although once you get into "removing outliers," the human judgment on what constitutes an outlier becomes its own source of error. Your specific example would leave the metrics based on a population of powder throws that didn't feel weird to the user. Might be the best that can be done, but it's hardly characterizing the entire performance profile of the powder measure/press/powder combination... and what feels odd to the user might vary from day to day!

 

[Nature Boy]

The biggest question people always ask is "how many samples do I need?". From a statistics point of view, it all depends on what you are trying to achieve.

If I was going to do this for any serious purpose it would be to see if a load with a wide accuracy window could be successfully loaded in progressive mode, without weighing each charge, and still stay in tune.

In that case I’d increase the sample size then shoot it to verify. The proof would be less in the statistics (only to verify the equipment was capable) and more on the paper down range.