On injuries

[vzografos]

Hello,

so I want to improve my predictions a little bit by working on the injury information. And for this I would like to ask for some suggestions on how do you normally deal with this (i.e. for prediction).

-First off. What do you think is a good website for getting up-to-date injury information for every team or every upcoming game? I am using http://www.donbest.com/nba/injuries/ and this one https://www.cbssports.com/nba/injuries/. Would you suggest something better? As for the past injuries I use prosportstransactions.com

-Second. How would you deal with uncertain/game-time-decision injuries? Do you write the players off (the lineup/roster for the day) or wait until the last minute?

-Three. A player coming off an injury. Obviously they might not play at their best (based on the injury type, time off etc). Do you manually adjust his expected performance or do you use some past trained model of performance before and after injury? If manual how would you do it? Can you give some example?
If using a model, can you give some details about it? OR some model you have seen online that might be interesting.

Thanks

[rainmantrail]

I use the same source for past injury data. For up-to-the-minute injury data, any of those sources you listed are probably just as good. Most of them likely share the same API feeds on the back end, so whichever site you find easiest to scrape from is the one I'd go with unless you see significant differences on when players are reporting.

As far as dealing with players whose statuses are TBD, you might want to lay off those games (I assume you're betting based on your posts). Some big bettors are likely to have insider information on those games, and unless you can find out whether or not key players are going to play for a given game, I'd be leery of placing a wager on that game if I were you.

For players coming off injuries, it really depends on what the injury is and how serious it was. I would build a multi-level model and treat specific injuries as fixed effects (e.g. - how does coming back from a torn ACL affect performance). That way, you can place a specific value on that type of injury. Same thing I do for NFL when a team loses a key position, like the starting center being out or something like that. You can estimate the impact of losing your starting center against the point spread by using a multi-level framework with fixed effects (and random effects).

[vzografos]

For players coming off injuries, it really depends on what the injury is and how serious it was. I would build a multi-level model and treat specific injuries as fixed effects (e.g. - how does coming back from a torn ACL affect performance). That way, you can place a specific value on that type of injury. Same thing I do for NFL when a team loses a key position, like the starting center being out or something like that. You can estimate the impact of losing your starting center against the point spread by using a multi-level framework with fixed effects (and random effects).

Yes that is in my todo list as well. We can train a model for how much the individual boxscore stats are affected based on the type of injury and how long before they recover. That should be easy to acomplish. If only I had more time