Thread: Obtaining more information from your groups - overlaying groups

Originally Posted by gimp

The sample size required to have the power to detect whether a load is "better" confidently depends on how much better. If you're trying to detect a 10% improvement, you'd need a very large sample size. 300 rounds? A 25-50% improvement should be visible with 10-20 shots. I'd recommend doing your testing at 100m to remove the variable of wind as much as possible, you'll note your 20rd group is slightly wider than tall - maybe wind, maybe recoil, maybe trigger control, maybe random chance.

It is more sensible to identify your requirements first - then simply select a load that meets those and move on.

If you do decide to try identify a load that is "better", you'll notice that looking at the mean radius rather than the group size will be much more helpful. With the same load you have a 40% variation in group size between your 2 groups. The mean radius measurements are within 5% of each other.

With that kind of variability group-to-group, even with 10rd groups which is much better than most people use, group extreme spread (size) is clearly next-to-useless for informing any decisions about whether a particular load is better than another - unless there's an absolutely massive difference. Like a factor of 4 or more

Originally Posted by gimp

If you do decide to try identity a load that is "better", you'll notice that looking at the mean radius rather than the group size will be much more helpful. With the same load you have a 40% variation in group size between your 2 groups. The mean radius measurements are within 5% of each other.

With that kind of variability group-to-group, even with 10rd groups which is much better than most people use, group extreme spread (size) is clearly next-to-useless for informing any decisions about whether a particular load is better than another - unless there's an absolutely massive difference. Like a factor of 4 or more

That's helpful to know. Makes sense of some folk suggesting larger rather than smaller changes (to a load), which I guess gives you greater 'signal to noise' in a smaller sample size.