Quote:
Originally Posted by Frequitude
How did you find your SD/z-score system worked for the draft?
|
Well there are a few bugs... Originally I had the goalies way too low. I figure multiplying their scores by 5 as there are 5 times as many skaters as goalies fixes it, and indeed it does, putting Hasek at +21.99, Kipper at +21.93, Heatley at +21.45. Brodeur though is +35.56! That's a 13.57 standard deviation between him and the next highest player! Also, as far as skaters go, Crosby is 7th, but I suspect that year to year improvement will justify having him as the top selection, although a case could be made for Brodeur's shutouts.
Another issue is that some stats, such as SHP and GWG, have a sample size problem, and are more likely to end up tied than others, such as assists, +/- and PIMs.
Then there's a bucketing issue, with the Cs LWs RWs and D. Originally I had seperate lists for Dmen and forwards, which gave the top defensemen and forwards similar scores. But then, if the defensemen have a lower average and standard deviation, a higher z-score will indicate less performance for a defenceman than it would for a forward. So I merged them into a skaters category. But that leads to another issue as I would have liked to cut down the list.
And the reason for that, is that really the guys who aren't getting drafted at all shouldn't be listed in my stats, because they're not in the game. So I wanted to remove guys and recalculate (which changes the order - leading to the possibility that the wrong guys would be cut out - maybe it would work best if you remove one guy at a time?). But I couldn't do that with a merged list of skaters. I guess the solution would be to do each position individually, make cuts, and then make the complete list, but that leads to the same error of making the wrong cuts. Actually, even better (but more effort) would be to make the combines skaters list, then cut guys by positions from the total list.
Another issue was making one file which is sortable on a season total basis and a per game basis (it's easy to make z-score formulas for both). Maybe if I knew how to sort by a middle column in Excel, or wasn't too lazy to make two files...
Anyway, I won't know how well it works until well into the season. For the most part though, I think it works fairly well. It looks pretty good in theory (although even in theory there are errors - that's why I just used it as a guideline - e.g. my Handzus pick which had NOTHING to do with my calculations). For the most part, players are ranked where you'd expect them to be so I see that also as an indication that it's an alright way to do things. It's easy to create custom rankings for a points pool, but I wanted some concrete stat for head-to-head, and the z-score system is what I came up with. Maybe for next year's pool, I'll have it perfected.
