Basktball Game simulation

[leerichardson]

Hello,
For one of my classes, I am creating a project that simulates a basketball game, using Monte Carlo methods. I am hoping to incorporate as many recently developed statistics as possible. Anyone have any ideas/suggestions for this? My eventual goal would be to make projections for the entire year, but I figured I should start with a possession, then game, and work my way outwards.

[Mike G]

At this site -- http://www.basketball-reference.com/teams/CLE/2013.html
check out the Advanced stats. Among these are Offensive and Defensive Rebound Percent (ORB%, DRB%), and these tell you the % of rebounds players tend to gather. If the 10 players on the floor total more than 100, then everyone will get less than normal. If they total 110, everyone gets 100/110 of their average rate.

Ask questions here as they come up. Don't do too much questionable work without checking in here; otherwise, you could be re-inventing the wheel.

[leerichardson]

Thanks! Yeah I got confused yesterday when I was reading the http://www.basketballprospectus.com/art ... cleid=2433 schoene projection page. What they do is use players individual %'s and normalize them so they equal the right # of possessions. My original idea was just to use team rebounding (offensive and defensive), team to%, percentage of possessions that end in 2's, 3's, and ft's to assign probabilities of what "event" would happen on each possession. then once you have the possession event, use player usg rates (if the event was a shot) to see which player will use the shot. Then, use the player's individual shooting percentages to determine when he makes of misses, then use the rebound %'s (team) if the shot is missed to see which team gets the next possession.

I guess the downside of that is that you need a team % first before you can start the game. But maybe we could just use the player's individual percentages pre-game(or possession) to determine these team percentages, then just run though the simulation once they're established.

Anyways, that's where I am so far. Let me know if anything sounds flawed or wrong. My project first draft is due on August 9th, so I have aobut 3-4 weeks to get something together.

[Crow]

viewtopic.php?f=2&t=65&hilit=simulation

Or try to talk with Bob Chaikin or Ben Falk (here or elsewhere) who have created game simulations.


memberlist.php?mode=viewprofile&u=145
memberlist.php?mode=viewprofile&u=109

[leerichardson]

Thank you!! That thread really helps. Our prof wants us to take a minimalist approach. Staring as simply as possible and then adding functionality as we go. I'm not quite sure what she's expecting, but she loves our project lay-out, in large part due to the fact that the just read "Signal and the Noise", i'm posturing.

[Crow]

I am glad it was of some use to you.

[leerichardson]

We now have a game that can take Inputs from Team Data (for both teams, these are just the variables):
double teamB2ps = .5;
double teamB3ps = .25;
double teamBFt = .1;
double teamBTurn = .25;
double teamB2pPerc = .5;
double teamB3pPerc = .33;
double teamBFtPerc = .8;
double teamBOReb = .3;
double teamBDReb = .8;

( http://www.leerichardson09.com/blog.html You can look at all our code here)

For both teams, then loops through a set number of possessions for each team, then can churn out a final score for each team. Where we have hit our wall is how to now incorporate player data into the simulation. We could either scrap our current possession function in favor of a function that keys strictly on usage rate, and using individual players data to determine the outcome of each possession, after the usage rate determines who uses the possession. Or we could try and use individual player statistics to come up with the above listed variables. Or just something I haven't thought of yet. I really want to try and use RAPM as well, but I think that's getting ahead of myself. Thanks to anyone who reads this.

[Crow]

You'd probably really benefit from talking to Bob, Ben or the guy who took over the simulator that Ben started (For the life of me I can't recall the full name of the game even though I once played it.) That game did a lot of what you are contemplating.

Here is one thread that might be of some general interest.
viewtopic.php?f=2&t=146&p=289&hilit=ffs#p289

[Crow]

here is the site I was referring to above:

http://www.xohoops.com/

not sure yet if it is still active.

[leerichardson]

Cool, yeah I wonder if these guys http://www.whatifsports.com/locker/ bought them out or something, XO sports looks down. Thanks for the post, right now we are simulating a game based solely on player data (usage rates triggering the possession, instead of events). Once we have that working, were going to run the two simulations a bunch of times and compare the results. Maybe the variance on the player data only simulation is much higher than the first one, or something.

[TheOddOne]

[...] XO sports looks down.

The site seems to have some quirks, but a few leagues are active: http://www.xohoops.com/Leagues/XOD/League/Teams

[nbacouchside]

does bob still sell his full simulator or is that over since he works with the heat now?

[bchaikin]

he does not... unfortunately...

[nbacouchside]

he does not... unfortunately...

bummer. it sounds really neat & i'd be curious to see how certain teams match up in it now. aaaand i'm not savvy enough to build something similar of my own.

[leerichardson]

I think to see how realistic our simulations are, we need to compare it to actual seasons results. Is there a good data source anyone knows of the get a log of single game scores throughout the year, and number of possessions in each games. Then I could have a distribution of real game results to compare with a distribution of simulated game results, and see how they compare. That would be a good way of measuring how realistic our game simulator is, I think. Also let me know if that logic/methodology sounds wrong.