Let us begin now the long and arduous process of explaining how Win Shares and Loss Shares are calculated for fielders.   We will begin behind the mask.

As an overview, the way that Win Shares and Loss Shares are assigned to fielders is this:

a)  Each player is assigned an area of defensive responsibility.

b)  Each player is assigned credit for defensive Win Shares.

c)  The area of defensive responsibility beyond the Win Shares is Loss Shares.

We explained (a) earlier.    Today we start on (b), and we will start with catchers.

An overview of the catcher’s process.    The catcher’s defensive win shares are:

a) the team’s defensive area of responsibility,

b) times .154,

c) times a number close to the team’s defensive winning percentage,

d) modified by seven other factors.

The seven other factors are:

1. The team’s strikeouts,

2. The team’s walks,

3. The catcher’s errors,

4. The catcher’s assists,

5. The catcher’s passed balls,

6. The catcher’s stolen bases allowed (where data is available), and

7. The catcher’s runners thrown out (where data is available). 

To illustrate the process, I’ll run data for three catchers—Rick Ferrell, Mike Scioscia, and Don Slaught, and to start with we’ll look at the data for the season in which each player had the most plate appearances—Rick Ferrell in 1935, Mike Scioscia in 1985, Don Slaught in 1984.   We’ll begin by putting on record their offensive win and loss shares:

Rick Ferrell in 1935 had an offensive won-lost record of 10-10.   His career wins and losses, as a hitter, were 126-128.   Mike Scioscia:

Scioscia’s career won-lost record as a hitter was 102-89.   Scioscia in 1985 and Ferrell in 1935 have very similar offensive statistics—similar batting averages, on base percentages, and slugging percentages:

Each player created 77 runs.  The difference is this.  Rick Ferrell was playing in a league with a 4.46 ERA, and in one of the best hitter’s parks in the league, park run factor of 117.  Scioscia was playing in a league with an ERA of 3.59, and in the best pitcher’s park in baseball, park run factor of 82.   It took a lot more runs to win a game in Fenway Park in 1935 than it did in Dodger Stadium in 1985—therefore, creating the same number of runs, Scioscia created many more wins.  And Don Slaught:

Slaught was a good-hitting catcher, too, and for his career his offensive winning percentage is better than Ferrell’s and not too different from Scioscia’s.   His career batting wins and losses were 92-83.  He just didn’t really have a good year with the bat in 1984.

Like all decent and God-fearing people, I was appalled when Rick Ferrell was elected to the Hall of Fame in 1984.   It’s not that Rick Ferrell wasn’t a good player; it is that he so clearly was not a great player.    Probably even the committee that elected him did not imagine that he was a great player.   It is generally thought that, of the 12 voters who voted for Rick Ferrell for the Hall of Fame in 1984, at least 11 assumed they would be the only person to vote for him, and they were just doing a favor for good ole’ Rick.  But this is the nature of the Hall of Fame voting process.  If the horse gets out of the barn, he’s gone for good.

Anyway, Rick Ferrell actually had more plate appearances in 1933 than in 1935, but that was split between two teams, which complicates the process, so we’ll skip that.  

The defensive “area of responsibility” assigned to these three players is:

            Rick Ferrell in 1935 will have 8.11 Defensive Win Shares and Loss Shares.  How that is derived was explained in an earlier article. 

To figure how many of those will be wins, we start with the Defensive Win Shares and Loss Shares for these players’ teams, which are:

Again, the derivation of this figure was explained in an earlier article. . ..search “38.8”.  This figure we multiply by .154, as 15.4% of the responsibility for any team’s defensive success (or failure) is attributed to the catcher:

This we multiply by a figure based on the team’s defensive winning percentage.  The 1935 Red Sox have a defensive won-lost record, including pitching, of 152-80.   They had extremely good pitching, led by Lefty Grove and Wes Ferrell.    The 152-80 record (232 Win Shares and Loss Shares) is including pitching; the 86.02 defensive Win Shares and Loss Shares is NOT including pitching.

Anyway, we modify this figure by adding in 25 wins and 25 losses, making it read of 177-105.   This creates a Defensive Winning Percentage, for the 1935 Red Sox, of .629.   The parallel figure for the 1985 Dodgers is .529, and for the 1984 Royals, .547:

What we are doing here is creating a presumption, absent other evidence, that a good defensive team has a good defensive catcher.   One of the central problems of fielding analysis is that all teams look very much alike in the fielding statistics.  Good defensive teams and bad defensive teams, in the fielding stats, have very much the same kind of numbers, because of the circular design of fielding stats, which require that every team, good or bad, must eventually record three outs in every inning.   What we’re doing here is counter-acting that circular assumption by building in a different starting point for the evaluation of the fielder—a starting point that matches the known and actual end point (the success of the team being the end point.)   We add in the 25 wins and 25 losses before making the calculation as a kind of “leavening”, pushing the defensive winning percentages of the team back toward .500, upon the assumption that the analysis of the individual fielders will push them back out away from .500.  

Anyway, the “8.33” above is not actually the starting-point figure for Rick Ferrell, but the starting-point figure for all the catchers on the 1935 Boston Red Sox.   This figure we multiply by the percentage of the team’s innings that this player caught.   We don’t know how many innings Ferrell actually caught, but we know that the catchers for the 1935 Boston Red Sox among them had 626 putouts, 94 assists and 14 errors, and the team played 1376 innings in the field.   Ferrell had 520 putouts, 79 assists and 1 error.  We can estimate, based on this, that Ferrell probably caught about 1,143 innings.  For Scioscia and Slaught, due to the godsend of Retrosheet, we have actual data:

These, then, are the starting points for these three catchers—6.92 Win Shares for Ferrell, 4.94 for Scioscia, 4.55 for Slaught.

The first thing we adjust for, from there, is the teams’ strikeouts by pitchers.   If the team’s pitchers strike out a lot of hitters, we give some minute fraction of the credit for this to the catchers.   The 1935 Red Sox pitchers struck out 470 batters, which was 23 below the league average, which becomes -27 if you adjust for innings pitched.   Ferrell caught about 83% of the team’s innings, so he gets about 83% of these. ..if you watch the decimals it works out to -22.41 for Ferrell, +66.10 for Scioscia, and  -69.80 for Slaught.  

For each 1000 strikeouts that the catcher is above or below the league norm, we credit him with 1.00 extra win shares as a fielder:

Then we do the same with walks.   For each 500 walks that the team is below the league average, we credit the catcher with 1.00 additional fielding win shares.   All three of these catchers worked for teams that had relatively good control, and thus all three of these catchers will be helped by this adjustment.   We figure that, in his innings, Ferrell’s pitchers were about 45.43 walks better than average, Scioscia’s were 58.74 better, and Slaught’s were 56.20 better.   We credit the catcher with one additional fielding win share for each 500 walks:

We count hit batsmen the same as walks.

These are very small credits; those things don’t really change the evaluations very much.   The first thing that does change things, sometimes, is errors.  

First of all, we figure the league non-strikeout error percentage for catchers.   The error percentage is the complement of the fielding percentage.   If the fielding percentage is .990, the error percentage is .010; if the fielding percentage is .880, the error percentage is .120.  

Before we figure the league fielding percentage for catchers, however, we take out the strikeouts.   Official fielding statistics somewhat mystifyingly credit a catcher with a putout whenever there is a strikeout.   The overwhelming majority of catcher’s putouts are strikeouts, and there are very few errors that occur on those plays, so that’s really a misleading inclusion.   “PO NK” is catcher’s putouts, not including strikeouts.  In the American League in 1935 there were 1,131 PO NK, 729 catcher’s assists, and 97 catcher’s errors.   This makes a non-strikeout error percentage of .050 (actually, .04957.)  

Using that number, we calculate the “expected errors” for each catcher, also removing his pitcher’s (estimated) strikeouts from his record.   Rick Ferrell in 1935 had 520 putouts, of which we estimate that 390 were strikeouts and 130 were something else.   (Catchers in that era had many, many more putouts that were not strikeouts than they have now.)   Given 130 PO NK, 79 assists and 13 errors, he had 222 plays not involving a strikeout.   With an expected error rate of .04957, that makes 10.98 expected errors.

He actually had 13 errors, so that’s 2.02 more errors than expected.   For each 16 errors above expectation, the catcher loses one win share; an error is thus 31 times more powerful in this system than a walk, and 62 times more powerful than a strikeout, because we assume that strikeouts and walks are basically the work of the pitcher, and only to some small extent a reflection on the work of the catcher.   Anyway, Scioscia and Slaught also had relative poor years in terms of making errors.   Scioscia also made 13 errors (+2.08), and Slaught made 11 (+2.71).    This lowers our defensive evaluations for all three players:

Back at the beginning of this, I gave you this overview:

The catcher’s defensive win shares are:

a) the team’s defensive area of responsibility,

b) times .154,

c) times a number close to the team’s defensive winning percentage,

d) modified by seven other factors.

The seven other factors are:

1. The team’s strikeouts,

2. The team’s walks,

3. The catcher’s errors,

4. The catcher’s assists,

5. The catcher’s passed balls,

6. The catcher’s stolen bases allowed (where data is available), and

7. The catcher’s runners thrown out (where data is available). 

We have moved past point “3” there.   We still have to deal with the catcher’s assists, passed balls, and the stolen base data. 

In modern baseball, catcher’s assists result primarily from stolen base attempts.  Historically, this was much less true.   In the past there were many more bunt attempts, often fielded by the catcher, and there were also many more balls that were chopped in front of the plate and died there.   Batters used heavier bats that generated less bat speed, and the grass was higher, and one result of those things was that catchers had to make more plays.

An unfortunate circumstance here is that catcher’s assists are inversely correlated to wins.   Catchers on bad teams have more assists than catchers on good teams do—and it is not a trivial difference.   Here, let me check a decade for illustration. . ..from 1950 through 1960 there were 38 teams that had winning percentages of .580 or better.   The catchers on those teams averaged 76 assists.  There were also 38 teams that had winning percentages of .420 or worse.  The catchers on those teams averaged 84 assists.

Now, let me ask you:  who had better catchers, the bad teams or the good teams?  The good teams, of course.   So getting these things back to EQUAL doesn’t solve our problem.   Getting them back to equal assumes that the bad teams were as good as the good teams.   We don’t want them to be equal; we want the good teams to come out ahead.

To do that, I invented a new category:  Wins + Catchers’ assists.   In the 1950-1960 example, the good teams averaged 95 wins and 76 assists, so they’re at 171.  The bad teams averaged 58 wins and 84 assists, so they’re at 142.   We don’t KNOW how much better the catchers on the good teams are than the catchers on the bad teams, but we can very safely assume that they are better.  171-142 is a reasonable and modest ratio.   We’ll call it W + CA. . ..Wins + Catcher’s Assists.

In the American League in 1935 there were 1335 W + CA (606 wins, 729 catcher’s assists).   The Red Sox’ share of the league innings pitched was .1261 (1376 / 10914).   The Red Sox thus have an expectation of 168 W + CA (1335 * .1261).   Their actual total is 172—four above expectation.  Rick Ferrell catches 83% of the team’s innings, so he gets 83% of the credit for that.   He’s now—saving a few extra decimals—3.1 assists above expectations.  Scioscia is 18.9 assists above expectation—a very large number for this category—and Slaught is 5.6 assists UNDER expectation.  

For each 40 assists above or below expectation, the player receives one Win Share.   This makes the numbers for these players as follows:

That sounded awfully complicated, so let me try to simplify it a little bit.   What we’re asking is, “Did this catcher have more assists than you would expect him to have, based on the league, the won-lost record of his team, and the number of innings he caught?   The rest of this business was asked and answered in the pursuit of that question.

Moving on now to passed balls.   This one couldn’t be any more simple.

Take the league passed balls.

Divide by the league innings pitched.

Multiply by the catcher’s innings caught.

That’s the player’s expected passed balls.    For each 200 passed balls that the player is better than the league average, credit him with one win share.

We’re basically ignoring passed balls—not quite, but pretty much.  We’re putting such a low weight on them that they don’t do much.   The reason for this is the knuckle ball problem.   The catchers who are best at catching a difficult pitch get assigned to catch the toughest pitchers, thus wind up with the most passed balls, which is backward.   It makes the data unreliable, so we give it an extremely low weight. 

Rick Ferrell in 1935 had 8 passed balls against an expectation of 8.06, so that would give him. . .let’s see, +.06 divided by 200, I guess that’s +.0003.   Scioscia had 5 passed balls versus and expectation of 10.87, so that +.029.   Slaught had 2 passed balls versus and expectation of 7.36, so that’s +.027.   This makes the new estimates:

Both Slaught and Ferrell later had intense knuckle ball experiences, so they’ll be happy we’re not putting much weight on that.  

OK, that leaves us runners thrown out stealing.  This is a really, really complicated process, and if I try to explain every widget here we’ll be here forever, so I may rush through some of this a little quickly. . .I’ll do my best to make it clear. 

We begin with each team’s Estimated Runners on First Base.   The formula for estimated runners on first base is:

Hits allowed minus Home Runs Allowed, times .781

            (assuming that 78.1% of non-home runs are singles)

Plus walks allowed

Plus hit batsmen

Minus Wild Pitches

Minus Balks

Minus Passed Balls

Later on, in figuring Fielding Wins and Losses for players at other positions, we will use a different version of Estimated Runners on First Base which also removes stolen bases and caught stealing.   At the moment, however, we still have stolen bases and caught stealing in the data.  

So we divide the team’s estimated runners on first base by the league’s estimated runners on first base, and we multiply this by the league stolen bases allowed.  For the Boston Red Sox in 1935, this is 1537/12398 times 480.   It works out to 60.   The American League in 1935 had 480 stolen bases with 8 teams; we expect the Red Sox to allow one-eighth of them, which is 60.   It doesn’t always work out that evenly, but it’s usually close.

The problem, again, is that bad teams allow more stolen bases than good teams do.  Part of the problem is that they have more runners on base—but most of it isn’t.  Most of the problem is that good teams are usually ahead, and bad teams are usually behind.   You run more when you’re ahead than you do when you’re behind, therefore good teams allow fewer stolen bases attempts than bad teams.  

So we have to adjust the data, again, for that.   We multiply the figure above by

Games Played plus Losses

Divided by

1.5 times games played

This creates a figure which is less than 1.000 for a good team, and greater than 1.000 for a bad team.

In this case, we’re actually giving a break to the catcher on the BAD team.   It’s the catcher on the bad team who is helped by this adjustment.   He allows more stolen bases, but he is expected to allow more stolen bases.   In any case, the 1935 Boston Red Sox played 154 games and had 75 losses, so this ratio for them is 229/231.  We multiply their expected stolen bases allowed—60—by 229/231.   That makes 59.    We expect them to allow 59 stolen bases. 

The Red Sox in 1935 actually allowed 49 stolen bases—10 less than expected.   For each 40 bases that they allow above or below expectation, we credit (or debit) them with one Win Share.    Ten stolen bases below expectation is +0.25 Win Shares, for the team’s catchers.

The bigger thing, however, is how many runners they throw out.  The formula here is different if we have stolen base data for the catcher, or if we don’t.  If we DON’T have stolen base data for the catcher, but we do have it for the team, the process is:

1)  Find the league percentage of base stealers that were thrown out,

2)  Multiply that by the number of stolen base attempts against the team,

3)  Subtract that from the number of runners thrown out by the team,

4)  Multiply that by the percentage of the team’s innings the catcher worked, and

5)  Credit the catcher with one Win Share for each 20 runners thrown out.

In the American League in 1935, 329 of 809 base stealers were thrown out, or 41%.

The Boston Red Sox had 104 stolen base attempts against them, so they could have expected to have 42 runners caught stealing.

They actually had 55 runners caught stealing, which makes them +13.

For the Boston team in 1935, the data is actually +10.12 in stolen bases allowed, and +12.71 for runners caught stealing.   It’s really positive data.   If you divide the first number by 40 and the second by 20 and add them together, you get +0.89.

Rick Ferrell caught 83% of the team’s innings, so he gets 83% of that.   It comes out to +0.74 Win Shares—a big item, but then, cutting off the running game is a big item for a catcher.   It’s what he is supposed to do.

For a catcher who has individual stolen base data, the first part of that is the same, but the second part is different.   For Scioscia in 1984, the stolen bases allowed vs. expectation—based on the team data—is

League Stolen Bases Allowed (1636)

Times Los Angeles Estimated Runners on First Base (1344)

Divided by league Estimated Runners on First Base (17747)

            This equals 124

Multiplied by Dodger Games + Losses (229)

Divided by 1.5 times Dodger Games (243)

                        This equals 117, actually 116.76

            The Dodgers in 1985 are expected to allow 117 stolen bases.

            Subtract their actual stolen bases allowed (99)

            Divide by 40

                        This equals .44.

            Scioscia caught 78.6% of the Dodgers’ innings, so he gets 78.6% of that, which is .35.   Scioscia gets 0.35 Win Shares for the fact that the Dodgers in 1985 allowed fewer stolen bases than expected.

            However, he also gets credit for runners caught stealing, and this is based on his INDIVIDUAL data, without respect to the team.

            Take the league throw-out percentage (716/2352)

            Multiply that by the number of stolen base attempts against Scioscia (121)

            And you have the number of runners that he is expected to throw out (36.84).

            Subtract that from the number he actually did throw out (43)

                        That makes 6.16

            Divide by 20

            And that’s .31.   Add those two together--.35 and .31—and that’s Scioscia’s credit for base stealing, +.66.

            For Don Slaught, 1984:

League Stolen Bases Allowed (1304)

Times KC Estimated Runners on First Base (1425)

Divided by league Estimated Runners on First Base (21149)

            This equals 88

Multiplied by Royal Games + Losses (240)

Divided by 1.5 times Royal Games (243)

                        This equals 87, actually 86.78

            The Royals in 1984 are expected to allow 87 stolen bases.

            Subtract their actual stolen bases allowed (94)

            Divide by 40

                        This equals negative .18.

            Slaught caught 69.9% of Kansas City’s innings, so he gets 69.9% of that, which is negative .13.   Slaught loses 0.13 Win Shares for the fact that Kansas City in 1984 allowed more stolen bases than expected.

            However, he also gets credit for runners caught stealing, and this is based on his INDIVIDUAL data, without respect to the team.

            Take the league throw-out percentage (738/2042)

            Multiply that by the number of stolen base attempts against Slaught (103)

            And you have the number of runners that he is expected to throw out (37.23).

            Subtract that from the number he actually did throw out (38)

                        That makes 0.77

            Divide by 20

                        This equals positive .04.

            So Slaught is -.13, +.04, and on balance he is -.09.  Adjusting now the data for all three players:

Some of you are objecting to the fact that we treated stolen bases, caught stealing and catcher’s assists as three different elements, when in reality they are all part of the same puzzle.  They are, and we did.   It’s called “hedging your bets” or “not putting all of your eggs in one basket.” 

They are all part of the same, but the data is all different; we have different data for different eras of baseball history.   We treated them as three separate objects, but we gave a fairly light weight to each one.   Assuming that

a)  A stolen base allowed is one-fourth of a run,

b)  Ten Runs are a Win, and

c)  A Win is three Win Shares,

We would estimate the value of one stolen base at one-thirteenth of a Win Share.   We credited it only as one-fortieth of a Win Share.  

Assuming that

a)  A caught stealing is one-third of a run,

b)  Nine Runs are a Win, and

c)  A Win is three Win Shares

We would estimate the value of a runner caught stealing at one-ninth of a Win Share.   We credited it only as one-twentieth of a Win Share.   That leaves room for things to be double-counted, if indeed they are being double-counted.

I have always believed that it is best to look at the data from as many different angles as possible.    If a catcher is credited with a lot of assists, that could mean many different things.   But if a catcher is credited with a good number of assists, a low number of stolen bases allowed and a high percentage of runners caught stealing, we can be confident of what that means.    We want to look at—and try to account for—all of the data that we can.

This is the end of the process of figuring Win Shares for these catchers in the field.   We now compare this to the areas of defensive responsibility assigned to these players, which were 8.11 for Ferrell, 6.69 for Scioscia, and 6.26 for Slaught:

We can now add these players fielding Win Shares and Lost Chairs to their Batting Wins and Losses, which we gave long ago, to find their year-by-year Win Shares and Loss Shares.  Rick Ferrell:

Ferrell’s a winning player.   We’ve got won-lost records for him of 14-9, 15-9, 14 and 11, 15-10, 17-10 and 15-8.   That’s a good player. 

Mike Scioscia:

And Don Slaught:

Slaught also had some good years.  When he hit .345 in 1992, his won-loss contribution was 12-2.    In 1990 it was 10-3.    That’s a player you want on your team.

Retrosheet uses a comprehensive player metric headed “BFW”, which I think stands for “Batting and Fielding Wins”.   They show Mike Scioscia at +18.2, Rick Ferrell at +11.5, and Don Slaught at +4.1.   Our calculations appear to be essentially consistent with those numbers.   We have Scioscia at 59 wins over .500, which—remembering that each win is three Win Shares—would make him +19.7.  Ferrell, at 37 games over .500, would be at +12.3, and Slaught, at 14 games over .500, would be at +4.7.   Essentially, we have exactly the same relationship among these three players as does that method.  

On the other hand:

1)  Ferrell’s selection to the Hall of Fame does not appear to be reasonable.   We like for a Hall of Famer to have 300 Win Shares or be 100 games over .500, or preferably both.   We’ll cut Ferrell a little slack because he’s a catcher, but the fact is that he’s not close, on any standard.  He was nowhere near being a Hall of Fame player.

2)  We have Scioscia as a better player than Ferrell, by a good margin.

3)  We have Scioscia in 1985 at a won-lost contribution of 22-3, which one would think would draw some attention in the MVP voting.   He was mentioned in the voting, but barely.   Basically, we think Scioscia was a lot better than the MVP voters did.

As hitters and ignoring career length, we would rank these players as follows:

Not really much difference.   And as fielders:

Thus, we see Scioscia as a better hitter than Ferrell and a better fielder, albeit in a shorter career.   But Scioscia isn’t a Hall of Famer, either—as a player.  He might be a Hall of Fame manager.