[Stocks]

I've been working on an NHL model the results are great the lines it gives me are where they should be but I don't have injuries built into yet and I'm not really sure what to do about it.

I've looked into advanced stats like fenwick and corsie and to be honest I'm not really a fan of them but some people are so it is was it is.

Anyway I was messing around with this today and here is what I came up with.

Goals x 2.5
Assists x 1
Shots x .25
Ice Time x .75
Face offs x .1

For face offs I multiply total face offs x win%

Once I get those numbers I get the average then I divide the average by the number of games played. Once I get that number I divide that by 18 number of players on a team not counting the goalie.

Here are some of the results I got for what a player is worst to his team.

Sidney Crosby 5.04%
Evgeni Malkin 4.28%
Steve Downie 1.69%
Kris Letang 3.04%

Patrice Bergeron 3.99%
Chris Kelly 2.16%
Zedeno Chara 3.59%

Sidney Crosby would be worth 5.04% so if the Penguins were 68% to win and Crosby wasn't playing I would make the Penguins 62.96% to win.

Here's the other thing I ran into. My model will adjust for Crosby not being in the line up on it's own fairly quickly so I can't just subtract 5.04% every game he doesn't play I'll have to lower the effect on the team every game he's not in the lineup.

I was thinking of subtracting .5% every game so lets just say Crosby was 5% to make it easier.

Game 1 5.0%
Game 2 4.5%
Game 3 4.0%
Game 4 3.5%
Game 5 3.0%
Game 6 2.5%
Game 7 2.0%
Game 8 1.5%
Game 9 1.0%
Game 10 0.5%
Game 11 0.0%

Here's the other thing I was thinking. What to do when Crosby comes back. After probably a dozen games the model will have fully leveled it self for Crosby not being in the lineup. So if he misses more then 11 games the first game back I would have to add 5% to the model or maybe less due to rust. Anyway I would basically have to do the same thing as if he was injured in reverse.

His first game back I would add 5% or less to the Penguins win%

Game 1 back add 5.0%
Game 2 back add 4.5%
Game 3 back add 4.0%
And so on.

If he missed just 1 game when he came back I would add 0.5% or 1% if he missed 2 games.

Thoughts or feedback?

[a4u2fear]

Without worry about injuries:

1) how did you come up with multipliers?
2) the best face off guys in the league are 56% and the worst are not much lower. seems like an irrelevant stat or one that could be left out. The difference is minimal
3) 5% for crosby seems low. Even though there are 18 or so other players, Crosbys impact is much more than the others. Other players assists and goals are much more correlated to crosbys time on the ice than others. I.e. Kunitz points are more likely a result of playing with Crosby than Kunitz generating those himself
4) you should be checking pittsburghs performance without Crosby and with Crosby to see if that makes sense with your estimates
5) your % don't necessarily have to match the line for the game, but do the differences in your calculated line vs Vegas show that your model is a better predictor than Vegas?

[Stocks]

Thanks for the response a4u2fear

I'll explain a little bit of my thought process.

1. The Multipliers were kind what I though they would be worth plus some trial and error. My goal for this was to have a top player like Crosby to be worth around 5%. Is that high or low I'm not sure but I think a lot of time people overestimate the impact of a single player. Either way a couple weeks of running this should tell me if I'm underestimating a players impact or not.

2. The faceoff stat has more to due with total number of faceoffs then it really win%. A good face off man is important like Patrice Bergeron of Boston has 504 total faceoffs, the next highest player on Boston only has 229 so Bergeron is a very key player for the Bruins.

3. 5% for Crosby I don't know maybe it should be more he does make the players around him better but players also make him better so it's kind of a 2 way street. A good scorer needs someone to get him the puck while a good passer needs someone to pass the puck to.

4. I agree checking Pittsburgh with and without Crosby would be a good idea and I might do that. I'm not sure if the sample size will be too small to get an accurate %.

5. The results my model is giving me so far are fairly close on a lot of games but the thing that I like is that it agrees with me on the lines I thought were off from my normal handicapping methods. A couple nights ago when Calgary played San Jose I thought the Vegas line of San Jose -200 was absolutely ridiculous, I figured that game as more of a coin flip and my model agreed actually said Calgary should of been a -115 favorite. I know one game don't mean shit but I've seen multiple examples just like this over the past couple weeks where I thought a line was way off and my model agreed.

Anyway as for the injury thing I think I'll try to find results of teams playing with and without players and I think I might increase the goal multiplier slightly.

[Stocks]

Here were the results from today. This does not account for several things like injuries, backup goalies and the Capitals and Islanders playing a home and home. I still have to look into that and see much the 2nd game is worth but I wouldn't doubt if its worth close to 10%.


Games My Win% - My Odds - Book Odds - Book Win% - Value

NY Rangers 55.40% -114 100 50.00% 5.40%
Philadelphia 44.60% 114 -110 52.40% -7.80%

NY Islanders 60.30% -152 100 50.00% 10.30%
Washington 39.70% 152 -110 52.40% -12.70%

Montreal 63.70% -175 -201 66.80% -3.10%
Buffalo 36.30% 175 180 35.70% 0.60%

Carolina 29.50% 239 215 31.80% -2.30%
Pittsburgh 70.50% -239 -241 70.70% -0.20%

Ottawa 53.70% -115 -105 51.20% 2.50%
Florida 46.30% 115 -105 51.20% -4.90%

Minnesota 50.50% -102 102 49.50% 1.00%
Dallas 49.50% 102 -113 53.10% -3.60%

Chicago 56.60% -130 100 50.00% 6.60%
Anaheim 43.40% 130 -110 52.40% -9.00%

Winnipeg 50.30% -101 148 40.30% 10.00%
Boston 49.70% 101 -164 62.10% -12.40%

Detroit 58.50% -141 -121 54.80% 3.70%
New Jersey 41.50% 141 110 47.60% -6.10%

Vancouver 66.30% -197 -140 58.30% 8.00%
Columbus 33.70% 197 127 44.10% -10.40%

Edmonton 30.00% 233 248 28.70% 1.30%
St Louis 70.00% -233 -280 73.70% -3.70%

[Stocks]

So I've been thinking about my injury solution and how to use it. I've decided to make goals worth more so I think Crosby is worth around 6.25% now.

Anyway I've been thinking what is Crosby 6.25% of? Is it winning? My original thought on this was just subtract a players value from the win% my model gives me but if I subtract 6.25% from 65% that's almost a 10% difference.

What about if I subtract the 6.25% from 100% and then multiply it by a teams winning % so it would be 100% - 6.25% =93.75% and then multiply that by the teams win % of 65% = 60.94%.

Sorry if this sounds retarded I have been drinking a fair bit today and apparently this is what I think about all day long now lol.

[geebo18]

It seems like you are only thinking about 1 side of the coin. If your model has Pittsburgh 59% vs. Columbus 41%, and you think Crosby is worth 6.25%, then Crosby being injured would not only hurt Pittsburgh's win percentage, but also help Columbus' win percentage.

So if you adjust Pittsburgh down by 59% - 6.25% = 52.75%, then I think you would re-normalize like so:
Pittsburgh = .5275 / (.5275 + .41) = .5627
Columbus = .41 / (.5275 + .41) = .4373
After all, there should be a 100% chance somebody wins, right?

[Stocks]

It seems like you are only thinking about 1 side of the coin. If your model has Pittsburgh 59% vs. Columbus 41%, and you think Crosby is worth 6.25%, then Crosby being injured would not only hurt Pittsburgh's win percentage, but also help Columbus' win percentage.

So if you adjust Pittsburgh down by 59% - 6.25% = 52.75%, then I think you would re-normalize like so:
Pittsburgh = .5275 / (.5275 + .41) = .5627
Columbus = .41 / (.5275 + .41) = .4373
After all, there should be a 100% chance somebody wins, right?

I didn't think about it like that but that's right. Anyway for my model I ended up just taking any injury% off the final win%. So if I had Pittsburgh 65% to beat Columbus and Crosby was out I just make Pittsburgh 59.1% and so far it seems to be matching up with fairly close to how much the books adjust their lines for injuries so I must be somewhat close lol.

By the way my injury formula up top don't work I had to divide all stats by games played apart from ice time. Then I squared ice time and divided it by a number and then weighted everything and averaged it and then I think I divided it by something else I'm not exactly sure what I did I'd have to check my model for the exact equation again.