BuzzerBeater Forums

i personnally expect portland to lose, but greg oden will still get 30 + points.

just to clear things up: this game is already done with a 101 - 86 win for portland ;)
here´s the boxscore if you ar interested (http://www.nba.com/games/20081128/NOHPOR/boxscore.html)

hm, I guess we BB-managers have the advantage that the world of BB is after all slightly more deterministic than reality.

This i think is an interesting question. First just think about the nba... what do you think the variance is in the nba? I mean its not like you can ever play the same EXACT game many times in a row... so you cant know what the real variance of results are... This is relevant to the task I have before me in so far as whether I should care about my results exactly matching the results from the NBA... or should my view be that the average result is reasonable... and the actual NBA results might simply be a particular result picked from the distribution around that average.

Whats further the more and more dimensions that I consider measuring the results and evaluating how the simulation matches reality.. the more i should expect one of those dimensions to be away from the actual mean. I dont know if that's a clear description...

Let me try again in the context of buzzerbeater. Last season we made a change to the game engine which effectively reduced the variance of the results... Now there was some great debate amongst us as to whether that was appropriate, and how we should evaluate whether or not the variance was too small or too great. It seems some users really would love there to be zero variance, and the "better" team to win everytime.... Now of course I dont know how you define better other than running it through the simulation and if the simulation is non deterministic than that sort of loses meaning unless you mean run it a million times and pick the median result... If we did that would it be a better game? would it be more realistic? are the answers to both those questions actually different? or the same?

My contention is that the game should have some variance, and the actual result should deviate from the median result some percentage of the time. I think that is both more realistic and a more interesting game to play... but from reading the forums sometimes i get the feeling that the userbase does not agree and would rather there be no variance.

there should certainly be variance, but what form it takes is the really interesting problem in my mind. players having particular skills in the various aspects of the game is a set starting point, and even in real life there's probably a way to measure that (statistics). from there BB has built layers that have added effects (tactics, opponents tactics, stamina, home court, etc), and these, again, are things we see the effects of in the NBA. the results of these are also measurable, in some ways, through statistics.

the thing that's building as a question in my mind is how the BB simulations of NBA games are taking game shape into account. are they at all?

and that got me thinking... isn't game shape (or form in HT) one of those things that should define a lot of the variance? it certainly does in real sport. perhaps game shape in BB should be less known and predictable...? a manager can put a player on the court expecting them to play well based on what they knew about the previous season or game or practice's performance, but should they really know? like in the NBA, couldn't a player that's been playing well totally flop one night and blow a game? or vice versa?

i'm not trying to make a suggestion - just posting the things i was wondering about when i considered how you'd judge game shape in the NBA.

at present i have actually completely ignored the question of game shape.. and set it equal and average for all players.

I guess in my mind I was thinking of using it as a way to model the effects of having back to back road games... or such effects.. and maybe have them be differential on age.. for instance Shaq plays better with more rest.

One could even conceive of trying to model each player individually with respect to game shape.

Honestly though I see game shape as a more minor component of real life variance. For the vast majority of times when a player has a bad night its not because he is tired per say... thats not to say his poor performance was a fluke of statistics.. meaning if he had taken the same shots and boxed out people with the same vigor he might have gotten many more points and rebounds. However, I dont know if there is any hope in being able to model those fluxuations.. beyond considering the specific opponent he is facing on that night.

For instance.. a follower of New Orleans might have concluded that David West had a poor Game Shape versus Portland, or Elton Brand had a poor game shape versus LA. However, my simulation suggests that you only needed to consider the defensive ratings of his opponent to explain much of their poor performance, as I didnt module their ratings or game shape for either game.

Game Shape i think serves a very different and important role in buzzerbeater the game because it is one more strategic decision managers have to make... do i play my best players a lot this game and reap benefit now at the cost of them being tired later.. or do I save their energy for later. I think you need only look at the substitution patterns in the NBA to tell that this is not a decision which NBA coaches and managers make on a regular basis, except with a few players such as Shaq. Certaintly not to the extent Buzzerbeater managers do or MLB baseball managers do.

Last edited by BB-Forrest at 12/5/2008 4:58:19 PM

that makes a lot of sense.

have you guys considered trying to model something on a player by player basis that does intentionally introduce variance? i mean like something in the background that managers can't see that effects the player's performance somewhat randomly. like a way to model a player being "hot" or "cold"? for BB, not for the NBA simulation obviously.

it ok

can i see james poseys skills?

please?