Those of you who have read my stuff over the years know that I have long been looking for some way to classify hitters into “families”.   I’ve tried a couple of other things to do this, but I have another idea here, and I think maybe this is it.

                        Each player has a unique, but not totally unique, ratio of doubles to triples to homers.   The most common extra base hit ratio is what I call a 6-1-3 ratio—about twice as many doubles as homers, and about three times as many homers as triples.   30 doubles, 5 triples, 15 homers.     Amos Otis, for example, had 374 doubles in his career, 193 home runs—about a 2-1 ratio—and had 66 triples.  That’s a 6-1-3 ratio (374-66-193).    Jay Bell had a 6-1-3 ratio (394-67-195), and Roy White (300-51-160), and Kirby Puckett (414-57-207).   It’s a common ratio.               

                        Over the last couple of years I have gotten into the habit, without really thinking it through, of categorizing players by this ratio. . ..I look at Paul Konerko and say, “Oh, that’s a 5-0-5 pattern”, or whatever.   It occurred to me recently that perhaps I could make this into a formal categorization process.

                        But first, the obligatory “Why are we doing this?” paragraphs.

                        Whenever you can identify players who have two things or three things or four things in common, it is extremely likely that they will also have other things in common.  If you identify the most-similar player to a player born in California, very often you will find another player born in California.   If you identify the most-similar player to a 6-foot-5-inch left-handed pitcher, very often you will find another 6-foot-5-inch lefty.   If you identify the most similar hitter to someone from the 1960s, very often you find someone else from the 1960s. 

                        Thus, if you can identify “families” of hitters, it is possible that you can learn things that you didn’t know by generalizing about the characteristics of the family.   If half the members of a family ground into a lot of double plays and you have no GIDP data about the other half, it is likely that they also will have grounded into a good many double plays.   If you observe that some members of a family age well, you might be able to predict that other members of the family, still in mid-career, will also age well.   If you find that some members of a family have difficulty making the jump from A ball to AA, you might anticipate that other members of the family might have the same problem.   If you notice that some members of a family are prone to a particular type of injury, you might want to do preventive training for other members of the family.   If one member of a family is arrested after being caught with a 14-year-old in his room, you might not wish to generalize from that, or even bring it up at the family re-union.     

                        Classification is central to the process of observation.   This is true in psychology, in chemistry, in zoology and quality control.  It is true in sabermetrics.  For years I have looked for some way to classify hitters.   Now I think that I have it.

                        OK, let’s move on.  The first thing we have to do is to figure, for each ten extra base hits that each player has, how many are doubles, how many are triples, how many are home runs—rounded off to the nearest integer, of course.   Well, sometimes you can’t round it off to the nearest integer and get ten.   Sometimes you get eleven, and sometimes you get nine.   Harmon Killebrew had 33% doubles, 3% triples, and 64.6% home runs.   If you round that off you get 3-0-6, whereas for Hal McRae you get 7-1-3, and for Joe Rudi, 6-1-4.  

                        We don’t want some players rounding off to nine or eleven, because we’re trying to put hitters into groups with similar hitters, and these nines and elevens tend to take players out of the groups.   The next thing we have to do, then, is to round the numbers off so that they add up to ten.   I would explain how we do this if I had any reason to believe that you would have trouble figuring it out on your own.

                        What if a player has 140 doubles, 30 triples, 30 homers?   That comes out to 7-1.500-1.500.   How do you break the tie?

                        In case of ties I assigned the overage to the lower category—that is, to triples rather than homers, and to doubles rather than triples or homers.   This player would be classified a 7-2-1.  

                        OK, we’ve got everybody’s extra base hits now grouped into a “ten-pattern”.   In theory there are 55 possible ten-patterns.  However, of those 55 possible ten-patterns, there are only 26 which have actually occurred, among players with 1000 or more plate appearances.   The 26 which have actually occurred are:

                        9-1-0

                        9-0-1

                        8-2-0

                        8-1-1

                        8-0-2

                        7-3-0

                        7-2-1

                        7-1-2

                        7-0-3

                        6-4-0

                        6-3-1

                        6-2-2

                        6-1-3

                        6-0-4

                        5-4-1

                        5-3-2

                        5-2-3

                        5-1-4

                        5-0-5

                        4-5-1

                        4-4-2

                        4-2-4

                        4-1-5

                        4-0-6

                        3-1-6

                        3-0-7

                        The only player in history who was 4-5-1 was Joe Visner (1885-1891), while the only two players who were 4-4-2 were Lew Whistler (1890-1893) and Jeff Stone (1983-1990).  All of them did it in careers that barely cleared the 1000-plate appearance mark.   These are not ratios that a player would likely sustain over a longer career.

                        OK, so we can sort players into those 26 categories, and, generally speaking, we can observe that some players within those categories do look very much alike.   There is an obvious problem.     Alex Rodriguez is 4-0-6, and so is Russell Branyan.   Can one really say that Russell Branyan and Alex Rodriguez are members of the same family of hitters?

                        Of course not.  Stan Musial and Craig Paquette are both 5-1-4s.    We have to do something to recognize quality of contribution distinctions.  After experimenting with different ways to slice the loaf, I decided that the most obvious one worked the best:  OPS.  

                        Players with an OPS over .900 I marked “A”.

                        Players with an OPS of .8334 to .8999 I marked “B”.

                        Players with an OPS of .7667 to .8333 I marked “C”.

                        Players with an OPS of .7000 to .7666 I marked “D”.

                        Players with an OPS of .6334 to .6999 I marked “E”.

                        Players with an OPS of .5667 to .6333 I marked “F”.

                        Players with an OPS below .5666 I marked “G”.

                        Although many of the “Fs” and “Gs” disappeared later in the process, since there were rarely enough of them to form families.

            Anyway, with his very high OPS Alex Rodriguez was now classified as 406 A, while Russell Branyan was classified as 406 C.                  This would have given us, in theory, 182 “families” of players—26 times 7—except that about 40% of those were empty sets.  

The goal of the system is to put hitters into groups with similar hitters.   The only two players in history who have a ten-pattern of 3-0-7 are Mark McGwire and Harmon Killebrew.    McGwire, however, has an OPS over .900, which makes him 307 A, and Killebrew has an OPS just below .900, which makes him 307 B.  

                        Obviously Killebrew and McGwire are similar hitters and belong in the same family, but that only makes a family of two.   Our goal in this particular exercise is not to celebrate the uniqueness of Mark McGwire, but to place him in a group with those hitters to whom he is most similar.  There are other hitters who are fairly similar—Sammy Sosa, Hank Sauer, Rocky Colavito, Frank Howard, etc.—so why don’t we put him where they are?  Those guys are in group 406 B, so I moved Killebrew and McGwire to group 406 B.

                        The rule I made up in my head was that each player should be in a family of at least ten players; no groups of less than ten.   In practice, I had many, many players who were in groups of eight and nine, and only a manageable number in groups of seven or less, so I adjusted the rule to families of at least eight. 

                        In a few cases these “forced classifications” were debatable.    There are only six players in history who have a “natural code” of 406 A, so I tried to group the 406 As with the 406 Bs, and, since there were more 406 Bs than 406 As, I called them all 406 B.  But this put Alex Rodriguez (naturally a 406 A) in the same family of hitters with Harmon Killebrew (naturally a 307 B), which. . ..A-Rod and Killer are both great hitters, but not really that similar.   I wound up moving two of the 406 As (A-Rod and Ken Griffey Jr.) into group 415 A, a group which also includes Mel Ott, Willie Mays, Hank Aaron, Babe Ruth and Barry Bonds, and moved the other four 406 As (Jim Thome, Adam Dunn, Ryan Howard and Mike Piazza) into group 406 B, with Killebrew, McGwire and Rocky Colavito. 

                        It’s not a perfect process, and if you can find some better way to make those hard borderline sorts, well, that’s what separates knowledge from BS; knowledge is something you can build on.  Anyway, my next problem was large, sloppy groups.   Large families become ragged and ill-defined.   The largest family remaining at this point was 613 D, a group which included 173 players ranging from Khalil Greene to Don Mueller.    Don Mueller was a 1950s outfielder who never struck out and would typically hit around .300 with less than ten homers a year.   How does he wind up in a group with Khalil Greene?

                        He doesn’t belong with him, of course, and I decided to break up the large families into groups of 80 or less. . ..actually 81 or less, since the one family which had 81 seems fairly cohesive, and I didn’t see that I needed to split it up.   I decided to further divide the too-large families by the ratio of base hits to secondary bases, secondary bases being defined as extra bases on hits, plus walks, plus stolen bases.    I divided 613 D into three groups—613 D1, which contains Khalil Greene along with players like Gary Redus, Oddibe McDowell, Lee Mazzilli, Lloyd Moseby, Tony Phillips, Tommy Harper, Vince DiMaggio, Ken Henderson and Devon White, 613 D2, which contains players like Emil Brown, Ed Charles, Gabe Kapler, Jose Cardenal and Claudell Washington, and 613 D3, which contains guys who hit .300 sometimes like Don Mueller, Jimmie Piersall, Bobby Avila, Cleon Jones and Vic Power.   613 D1 has an average batting average of .254 but with a good many strikeouts and walks and a few homers, while 613 D3 has an average batting average of .274 with few walks, fewer strikeouts, fewer stolen bases and slightly fewer home runs.  

                        Of course, smaller groups also tend to become ragged sometimes; that is, smaller families of players sometimes contain players that one would not tend to think of as similar.    One more problem.   As I was writing up the system, explaining who goes where, I became aware that the exceptionally high-walk players, like Ed Yost, Gene Tenace and Mickey Tettleton, stuck out like sore thumbs in this system as I had established it.   They just didn’t fit; their exceptional high-walk totals caused them to be classified, given the idiosyncrasies of this system, with players whose batting averages were 50 points higher and who also had more power.   I had to create two special groups, ending with a “W” code, to put some of the high-walk players into families together.

In a perfect categorization system,

a)      players who were very similar would always be in the same family, and

b)      players who were not similar would never be in the same family.

This isn’t a perfect system, and it doesn’t meet either of those criteria perfectly.  Because it relies on drawing lines between groups of players, and because those lines have to be arbitrarily determined, it does sometimes separate players who seem very similar.   Because it doesn’t look at every facet of the player’s skills, it sometimes puts together players who seem very dissimilar.   It isn’t a perfect system.   But I have been wanting to have a way to classify hitters into families for a long time, and this is the best thing I’ve come up with.

                        I wound up with 96 families of hitters, which include all non-pitchers in history with 1000 or more plate appearances.  

The 96 Families

                        In a separate article (Appendix to 96 Families), I’ll list all players in history and the family they are in—in fact, I’ll list them twice, once by the family and once alphabetically.   But here I wanted to offer a few more general observations about the families of hitters.

                        The system is intended to be sort of intuitive.  Who would be in family 811 C, for example? 

                        Well, let’s see. … 811 means singles hitters, obviously, since the home run number is very low—eight doubles for each homer.   That means five homers a season or less.  C indicates hitters who have an OPS around .800, and .800 is a high OPS for a singles hitter, so 811 C would be high-average singles hitters.  

                        Freddy Sanchez is 811 C—the only contemporary player who is 811 C.   Billy Herman was 811 C, and Lou Boudreau, Joe Sewell.    Wade Boggs is in group 811 C; Boggs has a natural code of 811 B, but is the only player in history who would be 811 B, so the family 811 B were his nearest relatives.   No one in history is or would naturally be 811 A.   Singles hitters don’t have an OPS over .900.   So Freddy Sanchez is classed in a family with Billy Herman, Lou Boudreau, Joe Sewell, Wade Boggs and a handful of other guys; it’s a small family, eight players. 

                        721 C are the same type of guys, only they run well enough to hit a few more triples (Rod Carew, Frankie Frisch, Lyman Bostock, Ben Chapman), and 721 B are the similar hitters only with even higher averages or a bit more power (Paul Waner, Nap Lajoie, Pete Browning, Heinie Manush.)   Tris Speaker and Ed Delahanty, who would naturally be classified 721 A, are included in 721 B because they would otherwise be a family of two.

                        The chart below summarizes the 96 families.   The players listed here are generally the best players in the family, or the most recognizable names, and they tend to be a little better than the center of the group.   Each group represents a range of ability; these players represent the top end of the range.   I was trying to use the most recognizable names to characterize the skills of the group, and the most recognizable are more recognizable because they were better.

                        For a lot of the families, a player might be a regular if he was an infielder or a catcher, but a part-time player if he was an outfielder.  Thus, group 514 D2 includes Gary Gaetti, Bill Freehan, Earl Battey and Benito Santiago—regulars and even minor stars—but also includes many guys who were about the same as hitters, but weren’t true regulars because they were outfielders (Rip Repulski, Pedro Feliz, Jerry Martin, Glenn Braggs, Pedro Munoz, Matt Mieske, Carmelo Castillo.)    I’m just pointing this out to remind you that it is a classification of HITTERS, not PLAYERS. 

                        I’ll make notes here about the families that had an OPS over .800.   More understanding of the families can be garnered from studying the rosters in the companion article, Appendix to 96 families.   The categories below are the code of the family (Family), the number of players in the family (#), the number of players in the family who are in the Hall of Fame (Hall), the average batting average for the players in the family (B Avg), the average On Base Percentage of the family members (OBA), and the average Slugging Percentage (Slug).   And a few examples of the most prominent players who are in the family.  

Post Script:

                        After filing this article three ways occurred to me that I could have done this better.   First, I should have added to the definition of a “family” that a family of players should have been defined by as many shared characteristics as can be identified for a suitable group.      Second, consistent with that change, I should have changed the parameters of a family from eight to eighty players to eight to twenty players—thus adding additional identifying patterns to the larger groups to break them down into tighter “families”.    In other words, take the 49 players in 811 F and break them down into three groups based on strikeout/walk ratios and/or speed, thus making more compact, more unified families.  

                        Third, I should have “named:” the families after their most prominent players, as I have done informally before, rather than allowing the families to be defined by codes.   If I had identified group 703 F as the Buck Martinez/Bob Melvin family, for example, the average fan would have been much more able to relate to that, more able to understand what I was doing.  

                        If I had made these changes it would have taken me another two or three days to finish the research.   I didn’t do these things, I think, in part because I hadn’t posted anything for a long time, and I was feeling some pressure to get something up.  That’s not a good reason, but I think that’s what happened.