Win Shares: Finding Clutch Performers

[viriya]

I'm done with the first iteration of this and wanted to share some results.

I thought it would be interesting to focus on the 2 semifinals that just finished to start things off - one a close game that went to the last over and the other more one-sided in nature. I've bolded the players with the significant win share contributions.

NZ vs SA:

Player: WS
Grant Elliott: 0.56
Morne Morkel: 0.49
AB de Villiers: 0.28
Brendon McCullum: 0.22
Faf du Plessis: 0.11
Daniel Vettori: 0.06
David Miller: 0.04
Corey Anderson: -0.01
Quinton de Kock: -0.01
Rilee Rossouw: -0.02
Jean-Paul Duminy: -0.03
Luke Ronchi: -0.04
Hashim Amla: -0.04
Tim Southee: -0.05
Martin Guptill: -0.05
Trent Boult: -0.07
Matt Henry: -0.08
Kane Williamson: -0.10
Ross Taylor: -0.13
Imran Tahir: -0.19
Vernon Philander: -0.32
Dale Steyn: -0.90

To make sure there is no confusion, the 0.56 attributed to Elliott means that he increased NZ's win odds by 56%. The -0.9 for Steyn means he decreased SA's win odds by 90%. A +0.5 WS basically implies that a player single-handedly won the game for the team, while a number as massive as -0.9 means that a player single-handedly lost the game for his team when the team was in a winning position.

Since this was a close game, the players who performed at the crunch get significant win shares. Morkel bowled a great over to get SA back into the game, Steyn lost SA the game when they were heavy favorites to win it in the last over. Elliott and Vettori pulled NZ through in the last overs which impacted their win shares significantly. Early on McCullum's assault moved the odds massively which is shown in his win share contribution, with Philander and Steyn bearing the brunt on the negative side. Even though Taylor and Guptill made decent contributions with the bat, the time they got out affected NZ winning chances so this is reflected in their negative win shares.

One thing to note is that even though you would think that the winning team would have a cumulative win shares value of +0.5 with the losing team having -0.5, due to extras playing a part, this is not actually the case. NZ's chase benefited from 6 byes and 2 leg byes and 5 wides - which are negatively attributed to the bowler but no batsman gets credit as expected.

Aus vs Ind:

Player: WS
Steven Smith: 0.26
Mitchell Johnson: 0.16
Josh Hazlewood: 0.12
Rohit Sharma: 0.02
Shikhar Dhawan: 0.02
Shane Watson: 0.02
Aaron Finch: 0.02
Mohammed Shami: 0.01
Glenn Maxwell: 0.01
Brad Haddin: 0.01
David Warner: 0.00
Mitchell Starc: 0.00
MS Dhoni: -0.02
Suresh Raina: -0.02
Michael Clarke: -0.04
Ravindra Jadeja: -0.04
Ajinkya Rahane: -0.06
Mohit Sharma: -0.07
James Faulkner: -0.09
Ravichandran Ashwin: -0.09
Virat Kohli: -0.13
Umesh Yadav: -0.13

Steve Smith was the clear matchwinner with his effort with the bat, with Mitch Johnson (with bat and ball) and Hazlewood making major contributions. India lost the game collectively, and while Dhawan + Rohit had a decent chance, they have minimal win shares because India's win odds plummeted due to the time they got out. Kohli's wicket was match-turning which is reflected in his negative win share value.

Next steps:
- Run this for the past decade and generate career win shares and best/worst win share performances.
- Incorporate fielding events and attribute win shares accurately. Attribute run outs, great catches to the fielder, and possibly implement negative win shares for dropped catches (what if scenario) - finally a solid way to value a player's fielding ability in terms of how it affects the match result.
- Update odds calculator to take momentum into account and weigh recent historical matches higher (important since ODI rule changes have affected the game significantly)
- Resolve any remaining bugs

Any comments/suggestions/criticism appreciated.

 

[hendrix]

Harsh for Corey Anderson to have the same share as Quinton De Kock.

 

[viriya]

Harsh for Corey Anderson to have the same share as Quinton De Kock.

His batting WS was 0.03 but it was canceled by his bowling WS of -0.036 since he got smacked, and his wickets were relatively meaningless in terms of match impact.

 

[hendrix]

Fair

 

[viriya]

His batting WS was 0.03 but it was canceled by his bowling WS of -0.036 since he got smacked, and his wickets were relatively meaningless in terms of match impact.

To go a little bit deeper, his battingWS was a match-winning 0.44 just before he got out because he was winning NZ the match, but as soon as he got out SA got back into the game and the odds pretty much flipped, so his battingWS plummetted to 0.03. Just setting up the chase is not enough - if a batsman gets out at the wrong time it can make a big difference. This is similar to Dhawan's case where he threw it away at a crucial time after setting India up.

 

[Daemon]

What would've Aswin's share been had he not batted?

 

[AndyZaltzHair]

Finch's innings probably deserves more winning share

 

[weldone]

...The shares attributed would probably need to be adjusted based on team strengths - so a +20% win shares vs Bangladesh isn't equated to be the same as vs Australia.

Any ideas/suggestions/comments appreciated.

Actually I should clarify something - if done properly it won't be +100% for the winning team and -100% for the losing. It would be dependent on what the team odds were at the start of the game.

So for the SL vs Ban game coming up tomorrow, SL starts with 80% win odds, so if SL end up winning, SL players will only be able to gain +20% win shares total. Bangladesh however, if they pull off a win can game +80% win shares (-80% for SL players in that case).

Doing both of these will create double-counting. [i.e. There is no need to do the part mentioned in part from first post, if you're doing what's mentioned in the 2nd post.]

Overall, a good idea to start with - provided you go with the 2nd option.

 

[weldone]

Let's say Bangladesh is chasing 250. They are on 150/4 after 35 overs. After 36 overs they become 151/5. Obviously there will be a reasonable amount of shift in odds (say -5% against Bangladesh). Let's say Shakib was out in the 36th over, he faced 3 balls, Mushfiqur Rahim faced 2 balls and scored the only run of the over, the new batsman faced 1 ball and didn't score a run. How will you divide the -5% among Shakib, Rahim and the new batsman?

 

[viriya]

Doing both of these will create double-counting. [i.e. There is no need to do the part mentioned in part from first post, if you're doing what's mentioned in the 2nd post.]

Overall, a good idea to start with - provided you go with the 2nd option.

Yes the plan was to go with what's on the 2nd post. Although currently I don't - every match starts with 50-50 odds. This is an adjustment for the future. One thing wrong in the 2nd post is that ignoring extras it would be +0.5 for the winning team and -0.5 for the losing.

 

[viriya]

Let's say Bangladesh is chasing 250. They are on 150/4 after 35 overs. After 36 overs they become 151/5. Obviously there will be a reasonable amount of shift in odds (say -5% against Bangladesh). Let's say Shakib was out in the 36th over, he faced 3 balls, Mushfiqur Rahim faced 2 balls and scored the only run of the over, the new batsman faced 1 ball and didn't score a run. How will you divide the -5% among Shakib, Rahim and the new batsman?

Good question. This was one I had to grapple with because a simple run % scored * win share change doesn't work here.

First, Shakib is attributed the majority of the win share drop by evaluating the odds of the Bangla scoreline when he gets out (150/5 after say 35.3 overs). The starting Ban odds are adjusted accordingly (-4.8% say).

Then the win share change is calculated from that position (150/5) to the end of the over (151/5). Say that comes out to be -0.2%, since just 1 run scored is not good enough late into the match. Since the win share change is negative and not positive, attributing that in terms of runs scored by each batsman makes no sense since the guy who went scoreless would benefit. So for negative changes, I look at balls faced. So in this case, Mushfique will get attributed 2/3 of the -0.2% and the new batsman 1/3.

 

[viriya]

What would've Aswin's share been had he not batted?

The same. India already had a 0% chance to win when he came in to bat so he didn't affect the odds. His bowling effort seems decent but he needed to get a wicket before the 30 over mark to actually make an impact - his Maxwell wicket was a little too late to affect odds. Basically he could go for not much in the 30th over, but just the fact that Australia didn't lose a wicket there could change odds by +2% in their favor - which would be attributed negatively to him. Of course we know that the time he got Maxwell was still important because the latter was liable to increase the Aus score by +25-30 possibly, but Maxwell is just considered another wicket and there is no way to know his potential impact in the game except that he was the 3rd batsman out when the Aussie score was already a healthy 232/2.

 

[viriya]

Finch's innings probably deserves more winning share

He plays his part by helping build the innings, but the change in the win odds is attributed mainly to Smith during the time Finch is in the crease because when he faced most of an over he tended to stagnate - possibly lowering Aus's win odds for that over. tbf I would've expected something close to 0.05 for him myself.

 

[AndyZaltzHair]

It looked like that at that moment but going by how most of other batsmen failed in both the teams, his innings comes out to be more important after the whole game. But in all fairness this is another argument

 

[Furball]

Plot twist: 'clutch' doesn't exist.

 

[weldone]

Yes the plan was to go with what's on the 2nd post. Although currently I don't - every match starts with 50-50 odds. This is an adjustment for the future. One thing wrong in the 2nd post is that ignoring extras it would be +0.5 for the winning team and -0.5 for the losing.

You should do that now. Starting every match from 50-50 odds doesn't make sense, given that you have a much better alternative which is not that hard to implement.

 

[weldone]

Good question. This was one I had to grapple with because a simple run % scored * win share change doesn't work here.

First, Shakib is attributed the majority of the win share drop by evaluating the odds of the Bangla scoreline when he gets out (150/5 after say 35.3 overs). The starting Ban odds are adjusted accordingly (-4.8% say).

Then the win share change is calculated from that position (150/5) to the end of the over (151/5). Say that comes out to be -0.2%, since just 1 run scored is not good enough late into the match. Since the win share change is negative and not positive, attributing that in terms of runs scored by each batsman makes no sense since the guy who went scoreless would benefit. So for negative changes, I look at balls faced. So in this case, Mushfique will get attributed 2/3 of the -0.2% and the new batsman 1/3.

Broadly that sounds fine - needs fine-tuning later on. [e.g. what if Mushfiq takes the single on first ball before Shakib faces 3 balls? Shakib shouldn't get credit for that single]

Edit: Doing calculation after every ball (in stead of every over) will solve this problem, if that's possible.

 

[weldone]

I'm interested to see aggregated results over a few matches - because although some people are saying that finishers get benefitted from these, I'm interested to see if in the long-run that kinda offsets (because finishers get large penalties for failed chases too).

 

[viriya]

Broadly that sounds fine - needs fine-tuning later on. [e.g. what if Mushfiq takes the single on first ball before Shakib faces 3 balls? Shakib shouldn't get credit for that single]

Edit: Doing calculation after every ball (in stead of every over) will solve this problem, if that's possible.

He doesn't he only gets credit for the runs he makes in that over. It doesn't matter when Mushfiqur gets that single.

 

[viriya]

I'm interested to see aggregated results over a few matches - because although some people are saying that finishers get benefitted from these, I'm interested to see if in the long-run that kinda offsets (because finishers get large penalties for failed chases too).

Yea Corey getting almost zilch for the match makes me think finishers won't overly benefit. I have a feeling opening batsmen might though since it's easier to push odds from 50% -> 65% than 65% -> 80% generally speaking.