Throwing out highest & lowest shots when calculating ES?

[Cproflow]

I'm not even close to saying I'm versed in statistics...but I was wondering if anyone made a practice of discarding their highest & lowest chronographed velocities when determining their ES? This would be from at least a statistically significant population (at least 15...better to have 35+). I always seem to have a doozey or 2 in there--likely from an aspect of reloading, be it charge weight or a neck tension thing. It seems to falsely give me a greater than desired ES (30-40), when otherwise things looked better (15-25). I'm not sure if it is 'cheating', or an accepted practice when evaluating a string of data points--basically getting rid of the outliers.

The SDs sometimes do not bear out these larger ES values--meaning I can get desirable SDs (10-12), and groups to match...even if the ES is in the 40s.

I may be able to drive down the ES by babysitting other aspects of loading (eg. neck turning, which I don't do...but I do anneal). Even without doing that now, I've been please by and large with the body of data I collect--minus those outliers

Am I way off base here?

 

[TimK]

It's done. I tend to not trust my chrono, so I regularly discard high and low readings.

 

[hrfunk]

I'm starting to do this more and more. It seems like I'll routinely have 7 or 8 shots out of a 10 shot string that are within about 10 fps of each other. Then there will be the odd 2 or three that completely skew the results. All of a sudden my ES of 10 goes to 90 as I have one shot 45 fps below the others and another one high by the same amount (obviously, the high and low aren't usually exactly the same, but you get the idea). So I've started looking at the trend in the MAJORITY of shots. I still record the entire string with the high, low, average, ES and SD in my notes, but then I break the figures down further, throw out the high and the low as anomolys, and recalculate the remaining figures. I think this is giving me a better idea of how my ammunition is performing, and I'm also seeing greater uniformity lot to lot, and across a progression of loads (as in initial load development).

HRF

 

[insight3b]

No I keep all data

 

[Bill Larson]

hobbyists chronographs are nortoriously inaccurate.I`ve used pacts,chronys.... now have an Ohler 35P and a magna..??...Bayonet style.All have given suspisious numbers at one time or another......they do have value as to general speeds....
I have competed in 2 nat`l "F" class events...(finished in the middle)...shoot LR (600-1200 yds.) once a week weather permitting.
talked at length with the current world champions....some don`t own chronographs.....
.the proof is in the target at the longest range you can shoot at.....
I use 5 shot [email protected] determine which load is best.

 

[RussW1911]

hobbyists chronographs are nortoriously inaccurate.I`ve used pacts,chronys.... now have an Ohler 35P and a magna..??...Bayonet style.All have given suspisious numbers at one time or another......they do have value as to general speeds....
I have competed in 2 nat`l "F" class events...(finished in the middle)...shoot LR (600-1200 yds.) once a week weather permitting.
talked at length with the current world champions....some don`t own chronographs.....
.the proof is in the target at the longest range you can shoot at.....
I use 5 shot [email protected] determine which load is best.

Ding ding ding..... Mr. Bill Larson for the win.

Reading through and up to his post I was going to post exactly the same thing. Unless you own a multi million dollar Ballistic Laboratory, proof is in the groups at range. I run the 35P for initial testing. Then I pay a licensed lab to do in depth analysis on my ammo before anything is taken for fact.

It doesn't matter what you are using, if it can be purchased by the hobbyist, it will give you erroneous numbers. I have or the company has pretty much owned them all. It's a fact, plain and simple.

The sad part is people spend some money, be it $49 all the way up to $2000 for these and expect them to work...... T'aint gonna happen that way.

 

[jbailey]

ES is based on two shots, and, in a statistical sense, is a near meaningless number, especially with the crappy chronos we have (see above posts on that). Standard Deviation is built on EVERY shot in the group and as such is so much more robust. It does everything that ES does and more. Want to know the MV of the outlayers? Easy with SD: just multiple it by 2x and you capture the MV of the 95% fastest and slowest shot expected. Need more of an outliner? multiple SD by 3x and you get the 99.7% fastest and slowest.

This is a much better method than randomly throwing out numbers...

 

[Heman]

When I used my chronograph I would get the same issue having a group of close velocities and that one outlier. (Side note, I'm a HS Physics Teacher and took a lot of stat classes) So I would run a T test on the value and if if fell outside I would drop it. Then I would just look at the velocity spread from shot to shot instead of the entire range. So long as that was good I usually considered the load good. However this was early on. Had some loads with great numbers but shot like crap on target but assumed that was me. Well when went to a match they shot like crap. So instead of looking at number I just started shooting and letting the target tell me. Loads with least vertical shot the best in the match. Only had 100yds to test but was good enough for 1000yd matches. Now I have access to longer areas to test so started doing the ladder test then shoot teh best for group and if it holds well I go with it. I haven't pulled my chrony out for a couple of years CED M2 as the on target testing method as done well to pick my best load and then Quickload does a pretty good job letting me know what the velocity is.

 

[memilanuk]

Minor quibble... its not multiply SD by 2x to get a 95% indicator of ES... more like 4x, as it's plus or minus 2 SD (Wikipedia)

That aside... I agree with the principle that throwing out the low and the high when using 'extreme spread', which already skews numbers artificially by relying on what by definition are 'outliers' or flukes, the least likely to be repeated... is somewhat hilarious.

 

[jbailey]

Minor quibble... its not multiply SD by 2x to get a 95% indicator of ES... more like 4x, as it's plus or minus 2 SD (Wikipedia)

That aside... I agree with the principle that throwing out the low and the high when using 'extreme spread', which already skews numbers artificially by relying on what by definition are 'outliers' or flukes, the least likely to be repeated... is somewhat hilarious.

Right, sorry, I should have been clearer on SD use: it is 2xSD + the average velocity to get you the MV of the 95% fastest shot and -2xSD + the average velocity to get the MV of the 95% slowest shot. SD assumes shots are distributed across a 'bell curve model' and has positive (faster) and negative (slower) plots.

For example, say you recorded the following data from your chrono:

Shot # MV (fps)
1 2600
2 2590
3 2670
4 2650
5 2580
6 2610
7 2590
8 2620
9 2630
10 2610

That gets you an average of 2615 and SD of 28 (per MS Excel spreadsheet). You can now calculate that 95% of your shots are expected within the 2670fps (on high end, 2615+2*28) and 2558fps (on low end, 2615-2*28) range. You can then run your ballistic program and see how big a target you can hit at various ranges using these highs and lows.

For the record, I agree with the guys saying "look at the targets" not the numbers. I'm just presenting a more valid math process than the ES concept the OP asked about.

 

[memilanuk]

For the record, I agree with the guys saying "look at the targets" not the numbers. I'm just presenting a more valid math process than the ES concept the OP asked about.

+1... I don't have the utmost faith in consumer chronos, particularly those that susceptible to lighting and/or alignment variations. I do think that going to the line with a load that appears to shoot well but gives absolutely horrid numbers would not make me very comfortable... I'd be wanting to hunt down where those outliers were coming from. But once those get pruned down... the increased sample size usually necessary to make any valid statistical analysis between load 'A' and load 'B' if they are at all close starts to negate some of the theoretical benefit to 'running the numbers'.

 

[Deadshot2]

I pay less attention to SD or ES, more carefully watching the MAD (Mean Absolute Deviation). This is more meaningful, especially in larger sample sizes as it shows the deviation from the average rather than just the Extreme's.

This keeps me from condemning a load just because I had one or two "strays". I then focus more on case prep, including volume matching.

 

[memilanuk]

Interesting... did some digging on the 'Net and mean (or average) absolute deviation does seem to have some pros over standard deviation for data that may contain a measurable percentage of error (like readings from consumer-grade chronos) and for non-Gaussian (i.e. real-world) sets. It's supposed to be less affected by outliers than standard deviation... which is certainly better than extreme spread in the first place. If anything, the only downsides, with modern computer programs doing all the heavy lifting and number-crunching for us these days, would be finding a ready-made solution without having to cobble it together yourself, and that comparing numbers with others (who would mostly be using SD) would be tough. Aside from that... it does look like a better solution, at least from the little I've read.

Holes in the paper downrange are still the final arbiter, though

 

[Twoboxer]

Each of us has to find what works. And if one's choice is to throw out the high and low because the remaining numbers more closely correlate to what happens on the target . . . that's perfectly appropriate. And, if it matters lol, that's fine by me.

OTOH, if one reports their ES and SD without reporting the discard of the high and the low . . . I think that's inappropriate and would drive folks crazy trying to achieve the unachievable.

ETA: Personally I think its inappropriate as a routine practice, but is a reasonable tool to determine why the target looks different than expected.

 

[ChrisGarrett]

Minor quibble... its not multiply SD by 2x to get a 95% indicator of ES... more like 4x, as it's plus or minus 2 SD (Wikipedia)

That aside... I agree with the principle that throwing out the low and the high when using 'extreme spread', which already skews numbers artificially by relying on what by definition are 'outliers' or flukes, the least likely to be repeated... is somewhat hilarious.

I don't know if I'd call the lowest velocity and highest velocity in a 10 shot string, flukes, they're just the high and the low.

2590

2645

2599

2608

2632

2594

2618

2611

2596

2621

In that tight example, I don't know if the low and high shots are flukes, per se, so I'd keep them.

Now, if there were a 2450, or a 2800 in there, sure...maybe one could classify them as flukes, but not in every situation.

Chris

 

[damoncali]

The rule of thumb is never throw out data without a good reason. In other words, if you found out one of your rounds was mislabeled or something like that. Never just because. This is how we end up with people wondering why their factory Rem 700 isn't shooting 1/4" groups.

 

[Fuzzball]

Some people do throw out the high and low shots to help things sound better than they are. Ditto some toss shots outside the size groups they wish to see, "on average." Neither practice helps the quality of the reloads but it sure helps in conversations at the range.

 

[easternncnative]

I pay less attention to SD or ES, more carefully watching the MAD (Mean Absolute Deviation). This is more meaningful, especially in larger sample sizes as it shows the deviation from the average rather than just the Extreme's.

This keeps me from condemning a load just because I had one or two "strays". I then focus more on case prep, including volume matching.

Could you provide a reference to the descriptive statistic "MAD (Mean Absolute Deviation)" ? It's one with which I'm not familiar. In the mean time, I'll begin googling. Thanks!

 

[memilanuk]

Could you provide a reference to the descriptive statistic "MAD (Mean Absolute Deviation)"

From the NIST Handbook of Statistical Methods... 1.3.5.6 'Measures of Scale' (note: this reference calls it 'average' absolute deviation (AAD) and 'MAD' is 'median' absolute deviation)

This measure does not square the distance from the mean, so it is less affected by extreme observations than are the variance and standard deviation.

</pre>Revisiting a 90-year-old debate: the advantages of the mean deviation

On the other hand, it is possible to argue that the mean deviation is preferable and that, since Fisher, we have taken a wrong turn in our analytic history. The mean deviation is actually more efficient than the standard deviation in the realistic situation where some of the measurements are in error, more efficient for distributions other than perfect normal, closely related to a number of other useful analytical techniques, and easier to understand.

 

[easternncnative]

Thank you sir for providing those sources. My last course in statistics was taken while in graduate school (mid-1970s). Alas, chronic exposure to ethanol coupled with the normal aging process of the brain has apparently erradicated those brain cells in which my knowledge of statistical analyses were once stored.

From the NIST Handbook of Statistical Methods... 1.3.5.6 'Measures of Scale' (note: this reference calls it 'average' absolute deviation (AAD) and 'MAD' is 'median' absolute deviation)


</pre>Revisiting a 90-year-old debate: the advantages of the mean deviation

 

[memilanuk]

Mine was a little more recent, if probably not as rigorous as a graduate-level course... luckily my stats teacher was cool with letting me use the mounds of notes (data) I had sitting around the loading room as a starting point for a lot of the assignments. Made things *much* more interesting...

 

[Deadshot2]

I don't know if I'd call the lowest velocity and highest velocity in a 10 shot string, flukes, they're just the high and the low.

2590

2645

2599

2608

2632

2594

2618

2611

2596

2621

In that tight example, I don't know if the low and high shots are flukes, per se, so I'd keep them.

Now, if there were a 2450, or a 2800 in there, sure...maybe one could classify them as flukes, but not in every situation.

Chris

Take each of these numbers and then subtract them from the average velocity. Disregard any +/1 signs and it cause certain speeds to stand out.

One thing for sure, statistics can be the most controversial of all topics. Hard to get a consensus, if not impossible as to what they mean to the big picture.

Targets are the real "truth-sayers".