Fun with data

When all is said and done with the IDPA match, all the scores are tallied up and put in to a giant spreadsheet that gets put up in various places. Being the recreational statistician that I am, I always want to toy with the data, so this time I did.

So, for the first item, I took a number of different shooters’ scores and put them side-by-side on a good old stacked-bar graph to give a pretty good impression of where things went wrong:

I even managed to put the best shooter (David Bramble) right next to the worst (myself). Whoops. Anyways, I think this is actually a pretty good indicator of where things went wrong. Things just broke down on stages 3 and 4, and they were complex enough that my score was a real problem. On average, 2.3 times as long as the average shooter on a given stage, but on Stage 3, that jumped to 3.4 times the average time. That stage was when I really started figuring things out. And it shows. My times, relative to the average, improved significantly on the last two stages.

But the real question is: what happened to Brian on Stage 4? Well, a mixture of weapon malfunctions and hits on non-threats ended up giving him the worst overall score for that stage of anyone in the match, by far:

Name	Overall	1	2	3	4	5	6
Brian Buczkowski	175.85	18.71	19.76	23.47	98.69	8.76	6.46
Average	124.15	19.28	23.62	32.76	30.694	10.19	7.58

He was consistently below average time on every stage but that one. So what happens if we find the ratio of his time versus average time on every other stage? Well, that number is .94, meaning that he was just a hair faster than average overall. (Looking at the stats, a dismal Stage 1 ruined that ratio which would otherwise be closer to .85.) So multiply the average Stage 4 time by .94, and we get the new Stage 4 time:

Name	Overall	1	2	3	4	5	6
Brian Buczkowski	106.07	18.71	19.76	23.47	28.1	8.76	6.46
Average	124.15	19.28	23.62	32.76	30.694	10.19	7.58

Wow, that’s a heck of a difference. Eyeballing the list of overall times, that bumps him up from about 20th percentile to 60th (where 50th percentile is the average score).

I probably could have gotten more accurate answers if I remembered how to do normal distributions (i.e. not just using a simple average where a single outlier–me–can slant the outcome heavily). I’ll have to see if I can figure it out for next month. Until then, may your R-squareds be indistinguishable from 1.