Posted By Sean S

Do defensive stats tell us anything worth knowing?  Does having a defensive rating for players help us predict which teams will be most successful in keeping hits and runs off the scoreboard?  That is a question that has never been answered, and one that I will try to explore.  Previously, people have looked at how well defensive metrics correlate from year to year with themselves, and how well they correlate to other metrics.  Going a bit deeper, people have shown that the metrics do not correlate as well for players who switch teams as they do for players who stay put.  But what we really should want to know is how useful they are in predicting which teams keep hits and runs off the scoreboard.

The standard has to be team defensive efficiency record, a stat invented by Bill James that tells us simply how often a batted ball is turned into an out.  This stat is not the sole responsibility of the fielders, ballparks have a big effect on it, as do the types of batted balls allowed by pitchers.  Line drives usually become hits, popups are almost always outs.  This also has to be accounted for before we can attribute DER to fielders.

What I have done is take simple defensive projections of 4 systems, UZR and PM (John Dewan's) which are found on Fangraphs, my own TotalZone, and Zone rating runs, using the method of Chris Dial, and data from the excellent replacement level Yankees website.  The simple projection is a marcel-type where I weight 2008 as 5, 2007 as 4, 2006 as 3, and add in 1500 chances of league average for each player.  Since 2008 has a weight of 5, a player who got 300 chances in 2008 and nothing before that will have his rating regressed 50% to the mean.  These simple projections are prorated to the actual number of defensive innings for each team in the 2009 season.  Finally, the projected team defensive ratings are compared to a park and BIP adjusted figure for team runs, based on their DER.

The results are disappointing.  The correlation coefficients average only .02.  I won't tell you which system wins yet for a few reasons.  TotalZone is neither first nor last among the 4.

The first reason not to publish the results is that there is more work to do.  I think I need to design a more robust projection system for this, accounting for data at multiple positions.  For example, Franklin Gutierrez has a rating of close to +20 for 2006-2008, but as a right fielder since that is where he mostly played for Cleveland.  He played center last year for Seattle and was the defensive star of 2009.  But without data in center, right now I have all the projections expecting him to be average.  There is a simple solution, use the position adjustments to derive a CF projection from his RF data, but that will take a little time to program in and do it for all 4 defensive systems.

While I work on the data and towards a more robust test, I will invite anyone else who does defensive ratings to join the test.  You'll just need to provide the data for 2006 to 2008, along with a MLBAM ID for each player.  The format needed is like this, an example used is Adam Everett 2006:



I'll add any system to the mix as long as you provide me the data in the proper format.  In particular, I'd love to have Dave Pinto's PMR, Shane Jensen's safe, Peter Jensen's Big Zone, and Tango Tiger's Fan Scouting report, if the creators of those systems are willing.


7 Comment(s):
Jason VE said...
I could ignore it if baseball-reference wasn't using these numbers. The "bad data" costs Gwynn probably 40 runs just in 1989. That's pretty significant, and pushes him down about 10 spots in the WAR career leaderboard.
July 29, 2010 01:53:52
Sean said...
It's bad data. Those were some of the first years retrosheet had project scoresheet batted ball data. It looks like they had some trouble scoring consistently year to year in San Diego. I choose to use it for the league, for most teams the additional detail in the data gives you better ratings than not using them. And I will always use the same system for every team in the league, so I have to live with San Diego ratings that don't make sense. I think the Mets and Braves, to a lessor extent, have some data issues as well. In Gwynn's case, ignore the fluctuations. His career rating is OK.
July 28, 2010 06:42:47
Jason VE said...
How did Gwynn manage a TZ of -28 in '89, which is far worse than Manny's worse? What's more interesting is Gwynn followed up that TZ two years later with a +30, which is far higher than Griffey, Mays, or Clemente ever posted. His rate numbers were actually better in '89 - more putouts and assists per inning (he did split time between RF and CF in '89). He won gold gloves both seasons, one of which was the worst defensive season ever, the other the best.
July 27, 2010 12:55:57
Sean said...
Joe, or anyone else: rallymonkey (numeral for five) at comcast dot net. Thank you in advance.
July 26, 2010 10:09:32
joe arthur said...
I have a couple of the things you're looking for. I'll email you tonight - assuming your address has stayed the same since 2-3 years ago?
July 26, 2010 07:52:42
Sean said...
For the ones on Fangraphs I'm using Balls in Zone, because it's there. I'm scaling TZ chances down a bit so they are on the same scale as the others. Doesn't need to be perfect, any estimate of playing time could work here.
July 25, 2010 07:38:09
Colin Wyers said...
How are you defining "chances" for each system?
July 25, 2010 07:06:07
Leave a Comment:
Name: * Email: *
Home Page URL:
Comment: *
   char left.

Enter the text shown in the image on the left: *
 Remember Me?
* fields are requried


User Profile
Sean S
Maryland, USA


You have 390145 hits.

Latest Comments