[suicidekings]

I've been playing around with my current NBA model over the last couple of days and I'm looking at injuries at the moment. Specifically, I want to automate the redistribution of minutes & other key stats for a matchup when a key player is out for whatever reason.

Taking one of tonight's games as an example, let's look at a SAS @ CHI matchup. Assume that Tony Parker is sick and won't play. He normally plays 32 minutes a game, and those minutes will obviously be redistributed among the rest of the team. My question is how to best go about redistributing those minutes.

Options:
1) Redistribute minutes among existing starter/bench players weighted by how many minutes they currently average (I'm currently using a general case of 5 starters and 5 bench players used in each game on average)
2) Weight the minutes by position. Should George Hill be receiving more minutes than Option (1) would give him because he is the backup PG for the Spurs?
3) Give the majority of the minutes to the other players as described in Option 1, with the remaining minutes given to an average replacement player derived from the remaining players on the Spurs bench that would otherwise not play.

I'm leaning towards the first option, however when the method is applied to injuries of less important players the model now tends to overestimate the team's edge. ie: MyDyess is out, but his 18 minutes gets redistributed to what are largely better offensive players, increasing the expected value of those minutes...

Does anyone have experience trying to model injury replacement in a similar way that could offer a suggestion as to which option is going to offer the best solution?

[Wrecktangle]

1) Redistributing minutes in almost any fashion for injuries doesn't work long term.

2) Releasing what does work "into the wild" is the last thing anyone who knows what does work would ever want to do.

[specialronnie29]

1) Redistributing minutes in almost any fashion for injuries doesn't work long term.

2) Releasing what does work "into the wild" is the last thing anyone who knows what does work would ever want to do.

1) Ignore everything wrecktangle ever says. seriously

[Justin7]

1) Redistributing minutes in almost any fashion for injuries doesn't work long term.

2) Releasing what does work "into the wild" is the last thing anyone who knows what does work would ever want to do.

I've had a lot of success with minutes redistributions.

I might try keeping your other 4 starters' minutes constant, and increasing the bench load.

[bolekblues]

great thread. i have also tried to model nba and have dealt with injuries, but rather manually lebelled each 'important' player with a certain number of points (deducted from the spread, or predicted MOV), ranging from 1 to 4,5-5. I did not really get too deep into the minutes redistribution because there is too much noise imo, u cannot precisly predict what kind of plan a coach has for replacing a player, esp in his first game out. maybe this will also be dependant on the opponent.

Definitely agree that option 1 seems best and as far as overstating team's strength i would suggest dealing only with players that have a real impact on a game. I am not saying mcdyess has no impact (because he plays really well), but if he's injured i guess you can increase some minutes for duncan / bonner / blair, or play scarcely used splitter.

Good luck SK

[roasthawg]

#1 is definitely easiest to implement... #2 seems the best bet of the three.

[That Foreign Guy]

I don't know how rigid positioning is in the NBA, but would a starting PG really play more minutes if the starting C was out?

If you have minutes played data you can see how minutes are actually distributed for injuries (query games where player x minutes = 0 and compare to minutes for player Y in that game to games where player X minutes > 10).

This should tell you how to distribute the replacement minutes.

[Pokerjoe]

Weighting the minutes can be complex. The problem with looking at past instances is that the sample would be small, and dependent somewhat upon the opponent that night, and the coach may flat out have been disappointed with the results and try a different tack this night. By the time the sample size is larger, you're no longer looking at a sub as much as you are the new starter.

Also, yes, sometimes a C being out might lead the PG to get more minutes. Sometimes a PG being out might mean the backup SF gets more minutes as the starting SF moves to SG and the SG plays more PG. Against team A but not team B. So it's pretty complex. Most coaches just want their best players on the court and don't obsess about nominal position. Lebron can play the point.

Also, bear in mind that you might be doing a lot of work trying to gain .1 points. Not that .1 points don't add up, just, keep in mind that it might be a LOT of work and it will then still only be MAYBE .1 points.

I'm not saying exactly how I do it, personally, because I might actually be doing it right, LOL, but I will say I don't think there's much merit to going crazy about it.