APBRmetrics

Have you seen and compared your findings to Ken Shirley's, referenced here http://godismyjudgeok.com/DStats/APBRme ... rt=30.html

Here is a direct link to the poster http://www.nessis.org/nessis07/Kenny_Shirley.pdf

This is really neat. I think game simulation is really interesting, because, with specific enough info about the teams, you should theoretically be able to better predict interactivity effects, which is better for a game like basketball than assuming every +10 MOV team will play every +5 MOV team equally well.

Crow wrote:Have you seen and compared your findings to Ken Shirley's, referenced here http://godismyjudgeok.com/DStats/APBRme ... rt=30.html

Here is a direct link to the poster http://www.nessis.org/nessis07/Kenny_Shirley.pdf

I have seen the poster but have not seen that thread, thanks for that, I'll go through it.

Shirley is focused on win probabilities, and does some funky stuff with counting transitions and possessions. We are focusing more on predicting actual scores and box score stats.

nbacouchside wrote:This is really neat. I think game simulation is really interesting, because, with specific enough info about the teams, you should theoretically be able to better predict interactivity effects, which is better for a game like basketball than assuming every +10 MOV team will play every +5 MOV team equally well.

Thanks for the feedback. We are hoping that games between "extreme" teams (e.g. Houston vs. Indiana) will yield some interesting simulation results.

I'm surprised to hear that teams actually do simulations at all.

i'm surprised there are teams that don't. simulation is a great methodology for player/team evaluation, but then i'm biased...

The 2 teams who I hear use simulations may use static, non-start state dependent models. At least the public talk of those models didn't clearly articulate use of start states, to my recollection.

what's a start state? and who does Markov play for?...

bchaikin wrote: what's a start state? and who does Markov play for?...

Crow wrote:Have you seen and compared your findings to Ken Shirley's, referenced here http://godismyjudgeok.com/DStats/APBRme ... rt=30.html

Here is a direct link to the poster http://www.nessis.org/nessis07/Kenny_Shirley.pdf

Something that confuses about this approach is that the points scored in the previous possession is used in the start state. In almost all cases that really shouldn't matter.

Something that does matter and we have taken into account is the point difference in the game depending on the time left.

Do the transition matrices go down to the level of individual players? I would imagine sample size for some players/units is very small. Are there Bayesian techniques used to regularize the data?

Currently we are focusing on team level and not individual players. We hope later to model lineups, players, etc., but are not at that point yet.

Any updates on progress with the model? Anything going public?

Did you guys end up getting this off the ground?