No standard distribution of bullet dispersion?? 5 shot groups useless?

[norcan]

Interesting, but how do I use that information? Give up shooting and take up golf?

(Wouldn't he have been better shortening the range and trying to fix some variables?)

I don’t think anyone’s suggesting giving up shooting. It’s valuable information for understanding what’s a waste of time and ammunition, and what isn’t.

 

[pob]

I don’t think anyone’s suggesting giving up shooting. It’s valuable information for understanding what’s a waste of time and ammunition, and what isn’t.

I think I disagree.
He doesn't really tell what isn't a waste of ammunition, does he? Which was my poorly constructed point.

 

[Utectok]

I'm not sure I agree with his logic. Why add the second SD/curve, what is that supposed to represent? Lots of input variables create one factor which is shot placement (i.e. the distance from the aim point regardless of angle). The way he adds the second SD and thus the 2nd distribution was like having two rifles firing simultaneously at the same POI. The way he sets two curves which stay on each side of the POI makes zero sense and generally speaking that is not how bimodal distributions generate over time.

I don’t know how this all works but I think he is suggesting a non standard deviation is the pattern of one rifle. I’d be interested in what’s done wrong?

 

[billgatese30]

I don’t know how this all works but I think he is suggesting a non standard deviation is the pattern of one rifle. I’d be interested in what’s done wrong?

I've done quite a bit of stats over the years, but generally speaking you don't assume there will be a non standard distribution unless there is a particular reason. I can't see why he assumed it would be a chi distribution. "typically" a one sided distribution like this (i.e. where you can't go below zero), would fit some kind of Poisson distribution. Chi is more often used (in my humble experience) where the different input factors can effect each other as well as the final output. In his example, all of the input variables are independent and do not necessarily have an effect on each other.

Really he would have been better off shooting the targets and tracking the impacts, then plotting the distribution rather than assuming the distribution first. Also, there are tests to best fit distributions to the data to see which is most appropriate, then using that he could calculate power vs. sample size for selecting the number of shots required in a group to use for future testing.

I'm not saying he's wrong (I'm not clever enough for that), but its certainly a very different way to approach a stats problem.