Simulation Reboot Part 1

Work is currently being done on the next iteration of the regular season simulation.  Last year a 5 part series of posts attempted to explain how this simulation worked.   Simulation in the past two years relied on all games from 1970 – present; around 100K games.  The next iteration of this simulation will use the daily snapshots back to 1920 or around 200K games.

Due to major differences in the way teams were managed between then and now there are some issues that need to be resolved.  This set of posts will highlight those issues and some of the decisions being made to address them.  This is after all a log book.

Relief was a big vector for error and how this was solved will be covered later.  Today will be a short, perhaps interesting post about starting pitching between then and now that affects the way the simulator treats starters.

As explained in part 5, there are two kinds of combo pairs; lineup -> starter and lineup -> relief. Each of these pairs is assigned an integer between -6 and +6.  A tier combo of -6 is the worst lineup facing the best pitcher or relief squad; +6 the opposite.

Each game consists of 4 pairs; an away ls, lr, and a home ls,lr.  The simulator will look back into the past and for lineup->starter grab the number of innings pitched and number of earned runs scored for both home and away.  It calculates who won that simulated game, counts it, and does it again one million times per simulation.  At the end it tabulates wins and losses and that’s the estimated probability for the current game.

There is a rather large difference between modern baseball and legacy baseball.  Right now studies are being made to show those differences.  Below is a table showing percentages for starters who pitched complete 9 innings.

Tier Inn 1950 1960 2000 2010
-6 9 0.459 0.413 0.088 0.062
-5 9 0.442 0.392 0.069 0.052
-4 9 0.397 0.350 0.047 0.034
-3 9 0.361 0.315 0.043 0.029
-2 9 0.337 0.288 0.039 0.024
-1 9 0.302 0.270 0.039 0.021
0 9 0.312 0.239 0.030 0.016
1 9 0.280 0.221 0.020 0.018
2 9 0.288 0.208 0.024 0.017
3 9 0.259 0.191 0.027 0.015
4 9 0.261 0.215 0.023 0.023
5 9 0.226 0.167 0.028 0.019
6 9 0.252 0.194 0.027 0.024
TOT 9 0.315 0.257 0.033 0.021

The columns represent a decade of games for the 1950s, 1960s, 2000s, and 2010s.  A -6 Tie combor is the worst lineup vs. best starter, +6 tier combo is the best lineup vs. worst starter.  The TOT row is  average across all tier combos for that decade.

in 1950s  A starter pitched 9 innings in 31.5% of all games.  That dropped to a little more than 1/4 in 1960s.  In modern baseball for which this simulator is supposed to handicap, it’s down to 2.1%.  Even though Tier combo -6 is almost 3x that at 6.2%, it’s still extremely rare.  Managers want to conserve wear and tear on their pitchers’s arms.  Sports medicine wasn’t as sophisticated back in the 1950s.

One would expect more complete games as Tier combos go negative where starters exceed lineups and that’s what we’re seeing.  The 0 Tier combo row represents even steven between lineup and starter.  One would expect that percentage should come close to the overall average and it does.  It’s a little off in modern baseball but that could be due to smaller sample size.

There is a table like this for each inning but the 9th is most interesting.  It is still a work in progress resolving 1950s data with modern baseball.  The more data we have the more accurate the simulation will be.

That’s all for this tidbit into the simulator.  This model cannot start simulating until May when there’s at least a month of baseball in the books and there’s still some problems with that. More on that later.

I had planned to cover spring training but decided to wait until opening day when we’ll do playoff horse race based upon 3 year splits of current roster.  Then we’ll cover new guys for both White Sox and Cubs.   Since the Cubs may not be on TV at my local tavern or any tavern around here we may be forced to follow the White Sox this season.  We’ll see.   Until then ….