Statistical Analysis of the Rokslide Drop Tests

[cubdriver84d]

I ran it on the pooled data to see if miss percentage was normally distributed in general. The pooled data sample size of 53 should return valid results with the shapiro-wilks. Welch’s does apply the normality assumption to both groups but there isn’t a great way to test for normality on really small sample sizes.

It isn’t perfect but for small sample sizes the assumption of normality is just that - an assumption. Since it appears to be normally distributed on the larger sample size it seems probable that the smaller samples should be normally distributed, there’s not a great way to confirm it. I appreciate the critique, it is a concern of mine ana well and if you have any good ideas on how to address it I’m all ears.

Making a binomial dataset and building a GLM will be my next step with the dataset. I’m also interested in using some group analysis software to compute the X and Y values of all of the shots fired compared to the X and Y values of the 10 shot proof groups with the same gun ammo to look into the changes in the variance of the Rayleigh distributions.

I’ll check out the coin package, permutation testing is a good idea I hadn’t thought of.

A while back I developed a Bayesian model to detect loss of zero and changes in group size. I coded all the MCMC in R and it uses recursive Bayes to update the probabilities of zero loss and change in group size as you shoot new groups. I didn't get as a far as pulling data from some of the softwares, just simulating data so far. Maybe I'll start working on it again. I have also been curious if the (bivariate) normal/Rayleigh distribution generally fit shot groups best, or if a multivariate t or something with fatter tails might fit better in some cases.