June 02, 2004
More Fun With Simulations
This post from yesterday elicited requests for a better simulation. That's something I wanted to do also. So with the new simulation, you the user are able to set the number of teams, the number of games teams play against each other, and whether random intrinsic winning percentages are assigned (if not, every team is intrinsically .500). Just fill the the form and click enter. If you leave everything blank, you'll get the same simulation as yesterday.
Posted by David Pinto at
09:30 PM
|
Standings
|
TrackBack (0)
Do we really think teams have intrinsic winning percentages that fixed in some kind of higher realm of Truth?
It's very Platonic ... very idealist.
http://sourceforge.net/projects/baseballsim/ let's you simulate games batter by batter. It's pretty good, no sacs, no stolen bases, but double plays are included.
I ran various cardinal lineups through and it's pretty easy to see how various orders are affected. (http://www.go-cardinals.org/archives/000466.html)
This is fun. I have an idea.... can you arrange these teams into a divisional system? I'd like to see just how things would turn out if say... teams 1,2,3,4 and 5 were always in the same division...would one randomly be better than the others? I wondered this after refreshing your projects over and over and yesterday finding team #3 to be over .500 8 out of 10 times. So I just think it would be fun to display the teams in divisions.
Yes, I believe teams have intrinsic winning percentages. When we do pre-season predictions, that's essentially what we're trying to say. We look at a team, look at the players and their pasts, and say, I think this team will win X number of games. That means we believe this team intrinsically has a Y winning percentage.
Very cool stuff, I could play with it all day. Isn't that sad!
David, I think what we do is try to pick a normal distribution of winning percentages centered on the mean expected intrinsic percentage, with the mean weighted by the probabilities of breakout years, comeback years, collapses, injuries, etc, ad infinitum. So if everything breaks right, as it seems to have for (say) the Braves last year, they surprise everybody and end up in the right tail of their expected W% distribution.
Dave,
Did you ever simulate to see how the intrinisic winning percentages play out over a whole 162 game season? I'm just curious how a team's actual winning percentage can typically vary from an intrinsic winning % over a 162 game stretch.
Thanks,
Rob
Teams DO have intrinsic win%, but:
a - we don't know what it is
b - it changes every day
Therefore, if you want to give NYY a .600 true talent win%, you are really saying
a - it's .600, with a standard deviation of .010 (or something); that's because we don't really know NYY's true talent, and .600 is an estimate, and we are 95% sure it's between .580 and .620
b - each day, it's .600, with a standard deviation of .025 (or something); that's because players are human, and each game has a different lineup, especially when you consider pitchers
However, this shouldn't really affect the general point David is trying to make.
If you are interested, I did a Shockwave app based on this little thread at
http://www.outskirts.com/director/archives/000573.html
It shows how after multiple seasons things will 'average out' to 0.500 (or whatever the intrinsic winning percentage is). If anyone uses Macromedia Director and wants to play with it, feel free to download the source.