Monday, March 18, 2013

Another Defensive Study

The good folks at Distant Past LLC have developed an tool that creates player rating scores based upon an owners league database.  The theory behind the ratings is intriguing:

  • They ran a whole bunch of sims for a select bunch of players
  • All the rest of the players in each sim were neutralized so that they were exactly equivalent to the average player per era.
  • Each player was simmed for 100,000 seasons and the results aggregated into a database.
  • Multiple regressions were then against this dataset and determined weights for each aspect of player performance, including running and fielding statistics.
Here is a note of theirs on accuracy
The multiple regressions were excellent in accuracy but excellent means 97% of the variation was explained by the model. That means 3% of the variation was unexplained. After such a large number of sims it was surprising that 3% remained. However, all tests run using these ratings indicate significant advantage compared to evaluations made from real-life statistics or using the in-built DMB draft tool.

I am honestly not sure if this method is any better than the previous ones we've used.  To refresh memories, previous methods chose to put all players in a neutral database and run a whole bunch of sims.  Then, change  the defensive ratings of one player, run a whole bunch more sims, and compare the results.

The Distant Past ratings seem to make more sense (you'll see in a moment) but the methodology is behind the scenes so we can't poke whole in it.

The scoring method is in terms of "runs per game contributed, so it's the number of runs contributed for each player per 100 games.  Today we're talking defense only.

I don't have a great way to present the data, so'll summarize and then let you play with a spreadsheet on you're own:

  • The basic rule is that range is much more important than error rates.  This is especially true for outfielders.
  • Each jump in Range doubles the amount of runs contributed.  Here is the table for Infield ranges where the error rate is constant at 100

  • Outfield is slightly more important than infield (yellow means I had to estimate the exact 'runs contributed since we didn't have players in the database with that exact Rg/Error rate)

  • At the elite ends (Ex/25) outfield remains slightly more important, but as you go down the scale slightly (Ex/50 and Ex/75) middle infielder contributions drop measurably drop off.

  • The scoring for outfield arm is alarming:

  • Catcher Arms have minimal impact.  +3 for Ex, +2 to Vg, -2 for Fr, -3 for Pr
Here's a compilation of the data.  Make of what you will.  Defense continues to matter.

No comments:

Post a Comment