Joe, If there is controversy in the stats world, it surely is not with respect to sample sizes for simple tests such as we do. Baysian statistics vs standard parametrics might yet be considered a "great" controversy, but we ain't there. Next, how do you know if your groups are not round? If one group is a 1.5" by 2.5" group unround? Could SDs give you a hint about that? You bet. But to know that a group is unround is something that can be dealt with statistically and it's not even that hard. - essentially, you would compare the mean displacement from the group's center in the horizontal dimension with the mean displacement in the verticle dimension. A conventional two-sample t-test (comparison of means) would suffice nicely. Since you have taught statistics - you know that distributions have infinite tails, and you know that the probably of finding a datum in any sample of size whatever from the extreme of either tail is totally dependent on sample size. Hence it is not a very useful measure. Also, having obtained a range or ES, you then procede to do what? You cannot use it in a t-test, or an Anova, or much of anything. You could use a bunch of range estimates in a formal statistical test, but the distributions of such tests are not known to me, and I have never seen anyone do such a thing. You will not find it in any stats text that I'm aware of. Think of it this way, an SD uses all the information contained in each and every shot to characterize a variance (square of the SD). Variance is a property that is very very well understood but statisticians. SEs use information from only 2 shots regardless of how many you shot. And, it tells you nothing about any distribution-defining parameter. Perhaps the best thing for you to do sometime, besides pulling out your old stats texts and looking at simple parametric comparsions of means (t-tests being the most obvious and simple, ANOVA being a possibility as well I suppose) would be to shoot 100 rounds or so, and plot SD for the first 2 shots, the first 3 shots, the first 4 shots, etc. up to 100 shots, and see what happens. It is not "crazy". FWIW, I have now written a program that will allow you to enter all the shots from a pair of targets and provides mean and SD estimates for the distances of each shot from the group center and also calculates a simple t-test comparison of means for a 1-tailed hypothesis testing (is my new load better than my old reference load?), estimating approximate confidence you can have that the two loads do indeed differ. I will eventually post that software and a tutorial, example and description of what and why this is done on a website. But right now, I'm going to go out and make a 100 shot comparison of Lapua Midas L and Wolf Match Extra ammo to decide once and for all, which is better. Adios, Brent
|