Quote:
Originally Posted by GKrotkov
The way I solved this is by restricting analyses of individual fields to a select set of teams rather than all teams at the competition, which raised the average comparatively and reduced the overvaluing of absurdly consistent teams. I kept the ones with all teams analyzed, but honestly reality-checking the latter made me realize that restricting the number of teams for more specific fields could be helpful. For example, I did not include 1712's data in the high goal t-score calculations. This caused averages to change and thus some of the strange cardinal results you see in the final order sort. So, in the example of 708 and 1257: 708 has a higher average and higher standard error than 1257, so, with the low goal specific analysis the higher general average resulting from eliminating teams that aren't competitive low goalers makes standard error more important and thus 1257 does better relative to 708 in the low-goal specific one rather than the boulder volume one.
* at least, overvalued for my purposes in picklisting.
|
How did you determine the cutoffs for which teams to include for each analysis? I imagine it was pretty clear-cut for high goal scoring since that was an "either you can do it or you can't" ability for the most part, but where did you draw the line for low goal scorers? I know 25 was
good, but is their t-score so dominant compared to the rest of the teams because they were so good, or because there was a wider spread in low goal scoring ability, lowering the average compared to how well 25 was performing? (does that question make sense? Statistics really isn't my strong suit)
Also, is this data from just quals, just eliminations, or both?