I am an advocate for changes in the way that defensive baseball statistics are recorded and analyzed.  There are 2 main ideas; the first is a statistic called "defense shares," and access to the full data for that article is here.

The rest of this page concerns a manuscript published in the Baseball Research Journal, which is published by the Society of American Baseball Research, under the title "A manifesto for Defensive Baseball Statistics."  Below are links to some supporting information for an article that I have submitted detailing the changes I think should happen and the benefits of the new system.  Here is an abstract of the project:

Abstract:  Current defensive statistics are far less meaningful than batting or fielding statistics, mostly because the correct information is not collected.  Statistics that do exist tend either to be not revealing or incredibly complex.  I propose the collection of a single new piece of information that will be able to revolutionize popular understandings of defense and also allow complex analyses.  By simply assessing the difficulty of the play, it should be possible to measure Hit Saving Plays and Difficult Plays not Made.  Sample data are presented for the first 21 games of the 2004 Anaheim Angel season.

All fielders



First Base

Proposed Rules Changes and Reliability Coding Rules


A program of defense research

Things to do without new technology

            A first step is to isolate the demands of different positions.  If the point of statistics is to match the play on the field, it is important to begin by defining what happens on the field.  Although I have no objection to separate ratings for all nine positions, at a minimum players should be grouped into no fewer than five categories:  Catcher, firstbase, other infield positions, outfield, and pitcher.  It may seem surprising to rate pitchers separately, however, they do field a lot of balls.  Pitchers fielded 30 ground balls in my dataset and failed to make 25 difficult plays (on balls that passed back over the rubber or within the pitcher’s wingspan), roughly 3 chances per game.  They are closer to the hitter than any other defender, making their fielding job on a hard-hit ball much more difficult and thus warranting a separate category.

            A second step is to define the demands of each position, and based on that, to figure out which information is most relevant to each position.  Part of the work to be done is to disseminate the most relevant information that is already completed and feature it more prominently. 

            Pitchers and infielders have two basic jobs:  Stop the balls they can get to based on their positioning, and make good throws to the correct bases (of course, range is important but there are already useful measures for that).  The HSP and DNM are the most relevant information for these positions since they provide a vastly improved assessment of how good players are at stopping balls.  In addition, measuring throws would be useful.  Zone ratings might be of great use here.  In addition to providing information about range, the zones can usefully tell how far a throw had to travel to get to first base.  A shortstop that turns a ground ball into an out from the hole between second and third, as compared to one who records an out on a chopper that just gets over the pitcher, might or might not be showing more range depending on where he was originally positioned.  But he is definitively making a more difficult throw.  It would also be useful to differentiate assists on forced and unforced plays, and also count how many times a fielder decided to attempt an unforced out rather than make a throw to first.  Finally, assists that follow an initial force out (the turning of double plays) should be counted separately from assists that produce the initial out. 

            A catcher has five basic duties:  Block potential wild pitches, throw out baserunners, call a good game, block the plate, and field the balls they can get to.  Much useful information is already collected about catchers and most of it is readily available.  The team ERA when the catcher is playing (although there is some contest about how meaningful the figure is[i]) and stolen base percentages, for example, are easy to come by.  Counting the number of balls in the dirt in addition to wild pitches and passed balls would give easy percentages to compare across catchers.  Calling the event a “prevented wild pitch” might be handy.  Similarly, with replays readily available it should be possible to count the number of times there is an unforced play at the plate where the throw beats the runner and the number of times the catcher records a putout, known in the common parlance as successfully “blocking the plate.”  Combined with catcher’s ERA, stolen base percentage, HSP, and DNM, prevented wild pitches and successful blocks would yield a range of statistics that usefully evaluate almost all that a catcher is asked to do.

            For outfielders, HSP and DNM combined with zone ratings provide most useful information.  The other factor that is essential to outfield play is arm strength.  Information that can be had but needs to be more widely disseminated is the number of runners who had a chance to advance on a fly ball but decided not to.  For example, when a runner is on third with fewer than two outs and the left fielder catches a fly ball there are three possible outcomes:  The runner tags and is thrown out at home, the runner tags and is safe at home, or the runner holds.  Only the instance of an outfield assist is measured now.  Simple counts of the latter two occurrences combined with zone data would give a much more rounded view of outfield defense.  Similar information could be collected on plays that are hits and not outs.

            The key skill for a first baseman is to prevent an error by making an outstanding play on a bad throw.  If a “difficult catch” is defined as any ball in the dirt or that pulls the first baseman off the bag it would be possible to count the number of difficult throws a first baseman had to catch and how many they converted into outs.  Combined with HSP and DNM data, this single piece of additional information should allow fair statistical descriptions of first base play.

Things to do with new technology

            I have one final comment on range and arm strength:  Both are purely physical phenomena that can be measured directly as physical performances and do not require statistics.  With advances in digital broadcasting these direct measures might not be far away and could quickly supplant existing range calculations.  For infielders, range is a function of how far the defender must move to get to a ball and how fast it is traveling.  In a world with the radar guns, the K-zone, and the magic first-down line in football broadcasts, it is easy to imagine that software applied to moving digital video could soon measure the speed of the ball and the distance between where a fielder started and where they ended up.  Take pitching, for example.  If digital video takes a frame every 1/20th of a second, and it takes 7 frames for the ball to make it the 60 feet and six inches to the plate, it is only a few quick calculations to an MPH measurement.  A similar measure on the ball speed as it left the bat and a separate measure of how far the fielder had to travel to get to the ball would tell all that needed to be known about range and could quickly supplant all of our current guessing.

            In a similar way, range for outfielders is a combination of how fast the fielder is running to make the catch, how fast the ball is traveling, and the flight path of the ball (whether it is a fly ball or a line drive).  In slightly different language, you need to know how long the ball was in the air and how far the fielder had to run to get to where the ball landed.  Digital technology and analysis might soon allow this to be measured directly.  Finally, for all positions, arm strength is a function of how far the runner is from the base when the throw leaves the fielders hand and how fast the runner is traveling.  Again, these are figures that might become available with advances in technology.

            In short, the future of defense research involves three basic types of assessments.  The first are outcomes: The traditional putouts and assists, passed balls and wild pitches, and the sort of data this section has suggested (blocked pitches in the dirt and catches by the first baseman on throws that bounce).  The second are difficulty assessments, like the traditional error and the HSPs and DNMs this article has recommended.  These are inherently subjective measures, although the term “subjective” does not preclude a shared communal understanding that can be applied consistently to recurrent events on the field.  It simply means that the use of judgment is necessary in the evaluation of a play:  One can identify a strikeout by simply looking at the umpire, but to differentiate a hit from an error some decision must be made.  Those decisions can be made consistently.  A third class of assessments – the traditional five tools – are physical phenomena that can be measured directly.  Arm strength and range fall into this category, and advances in technology should soon be applied to these concepts and might end ongoing debates about these physical feats.


[i] James, Bill. “Underestimating the fog,” Baseball Research Journal, vol. 33, 2004, 29-33.