Here's my list with some values from my very preliminary pcs-ripoff type model
1) Connor 67.4%
2) Roslovic 58.0%
3) Vesalainen 55.6%
4) Samberg 8.9% (USHS can't be modelled well)
5) Niku 20.0%
6) Poolman 20.0%
7) Spacek 21.0%
8) Kovacevic 20.2%
9) Comrie
10) Lemieux 22.3%
11) Foley 18.6%
12) Gawanke 16.5%
13) DeLeo 21.1%
14) Appleton 14.4%
15) Harkins 12.3%
16) Green 10.3%
17) Stanley 8.3%
18) Berdin
Of course, it still has a lot of work, so take the numbers as very rough approximates, not specifically correct.
My own specific list, I will make some changes here and there. I may raise Harkins, for example...
Thanks!
I like the top of the list.
I think Harkins is a bit low. One variable that I've been considering with Harkins is his offensive production relative to his team. Though he's not been a high scorer, he was easily the top producer on his low-scoring team.
I also think that Foley should be a bit higher. I'd rank him higher than Spacek and perhaps Lemieux. Again, I think he gets a bit penalized for playing on a low-scoring team (Providence) that tends to spread playing time around a lot.
Interesting that both Green and Stanley are low, which I agree with at this point. One is a small, skilled D and the other is a large, not skilled D.
As you've pointed out in the past, the probability of NHL success is also patterned substantially by opportunities, which has a number of franchise-level variables including draft position (teams tend to give higher picks more opportunities), prospect pipeline, developmental status of the organization, age structure of the roster, depth chart, etc. I would bet that if a model could incorporate a lot of those variables it would substantially improve the performance of predictive models, but that seems a pretty tough task, since you'd need to use time-dependent models.