[ Home ] [ Library ] [ Bookstore ] [ Contact ] [ Search ]
What does the term "invisible offense" mean? Does it refer to the spark that the intangibles of a veteran leader like Terry Pendelton brings to a club? Err...no.
Is "invisible offense" what inspired the New York media to dub Rey Ordonez with the monicker "The Magician"? Absolutely not.
Invisible offense isn't really invisible. It's just that it can be hard to see a player's overall offensive contribution to a lineup from a compilation of standard individual statistics. Invisible offense is what separates the likes of Frank Thomas and Barry Bonds from players like Joe Carter and Cecil Fielder even though their Triple Crown stats may not be remarkably different. Invisible offense is also what makes players like Ozzie Guillen, Luis Polonia, and Luis Sojo are so damaging to their own teams. I won't, however, ask you to take this on faith.
Consider the case of the New York Yankees in 1996 who felt that they needed to pump up their offense during the stretch run of the pennant race. Now if you were the Yankees' GM and had your choice of the following players to add as your DH which would you choose (solely on the basis of the listed single season performance)?
Player AB Runs Hits 2B 3B HR RBI OPS ----- -- ---- ---- -- -- -- --- --- A 645 115 184 44 7 15 105 .829 B 645 114 163 22 0 43 137 .839 C 725 151 147 0 0 146 283 1.014 D 384 129 5 0 0 0 27 .516
Have you made your choice yet? You've probably realized by now that this is a trick question, but please bear with me. Player D would be unlikely to ever reach the big leagues with that kind of offensive profile. I'm guessing that Bob Watson would have received the immediate axe from George Steinbrenner if he had traded for player C. I doubt that we'll ever see a player like player C, unless a 27 year old Rob Deer or Dave Kingman were allowed to play 162 consecutive games at Coors Field. If player D did exist, most GMs would be moving heaven and earth to acquire him.
How can such a choice be objectively made? One way would be to look at traditional stats like HR and RBI. Clearly looking at things this way, you'd order your preferences C,B,A,D. A more statistical approach would have the choice be made by OPS (on base plus slugging percentage). Again choosing by OPS would order the preferences C,B,A,D. These choices would be *dead wrong*, and I'll demonstrate why.
What is the ultimate goal of improving a team's offense? It is, obviously, to improve the amount of runs that the team scores. What is not so obvious is how a player's performance will affect team scoring. OPS is a convenient shorthand way to evaluate a player's performance. I doubt that any serious baseball stats guru would argue otherwise. There are methods like EQA and VORP that yield much better results, but I'm hoping to present a more visceral and perhaps more easily understood evaluation.
My method of choice is simulation. In what passes for my spare time over the last several years, I've developed a computer program which simulates and measures the performance of arbitrary lineups against average pitching. It has evolved from simply simulating singles, doubles, triples, home runs, and walks to handling things like errors, baserunning outs, GIDPs, strikeouts, stolen bases, baserunning ability, etc. The simulator is tuned to 1996 AL performance in terms of parameters like batting outs per game, team runs to team RBI ratio, bases empty/runners on base splits, etc. A detailed discussion of simulation issues is beyond the scope of this article.
The idea behind lineup simulation is that if an infinite number of seasons were simulated (with players performing statistically as they do in real life), the average runs scored by the lineup would be equal to the expected scoring of a real lineup over 162 games. This is *not* the same thing as team scoring because on a real team these players would not be getting every single plate appearance in every single game. It is, however, a good way to make apples to apples comparisons between lineups.
Since it's not possible to simulate an infinite number of seasons, it's necessary to make do with less. After about 100 seasons of 162 games (and perhaps even less), most of the individual and team stats tend to converge. The most important stat, team scoring, is not so easily pinned down. Why is this so? It's because scoring is so extremely sequence dependent. For instance, taken out of context, a home run followed by a walk is only half as valuable as a walk followed by a home run. I won't bore you with the details, but the variance in team scoring displayed by a typical major league lineup is large enough that ~30,000 seasons are required to bring the confidence level to 99% that the simulated average scoring is within 1 run of the theoretical expected scoring.
So what does my simulator say about the team scoring for the 4 Yankee lineups that use players A, B, C, and D as cleanup hitting DHs?
Player A: Max Runs = 1172, Min Runs = 772, Avg Runs = 981 Player B: Max Runs = 1147, Min Runs = 789, Avg Runs = 967 Player C: Max Runs = 1162, Min Runs = 808, Avg Runs = 975 Player D: Max Runs = 1198, Min Runs = 819, Avg Runs = 1004
These results certainly seem to be counterintuitive on the surface. Let's look a little deeper. Here's the lineup that's being tested:
AVG OBP SLG
--- --- ---
Raines/NYY .284 .383 .468
Boggs/NYY .311 .389 .389
O'Neill/NYY .302 .411 .474
Player A,B,C,D
B.Williams/NYY .305 .391 .535
T.Martinez/NYY .292 .364 .466
Jeter/NYY .314 .370 .430
Duncan/NYY .340 .352 .500
Girardi/NYY .294 .346 .374
First let's look at what's happening with the rest of the players in the 4 lineups. Their individual abilities are unaffected by the identity of the DH, but certain stats are team dependent.
Runs Scored for the 5 players batting ahead of the DH:
Player A Player B Player C Player D
-------- -------- -------- --------
Duncan/NYY 92 92 89 95
Girardi/NYY 83 82 78 85
Raines/NYY 141 138 129 140
Boggs/NYY 111 108 102 108
O'Neill/NYY 115 114 117 115
--- --- --- ---
SubTotal: 542 534 515 543
Player A,B,C,D 115 114 151 129
--- --- --- ---
Total: 657 648 666 672
This still seems odd...Player D, who can't hit a lick, has more runs scored by the five guys in front of him than any other lineup. Player C, who hit more than twice as many HRs in 162 games than Babe Ruth or Roger Maris ever did, has the fewest runs scored by the (same) five players hitting in front of him! His 151 runs scored (148 scored on his own HRs) aren't even enough to close the gap totally! Hmm... Maybe this has soemthing to do with invisible offense. We'll have to come back to this later.
RBIs for the 5 players batting behind the DH:
Player A Player B Player C Player D
-------- -------- -------- --------
B.Williams/NYY 140 123 77 174
T.Martinez/NYY 122 106 71 149
Jeter/NYY 92 85 62 110
Duncan/NYY 103 103 90 113
Girardi/NYY 73 73 69 76
--- --- --- ---
SubTotal: 530 490 369 622
Player A,B,C,D 105 137 283 27
--- --- --- ---
Total: 635 627 652 649
What's going on here with these wild fluctuations in individual RBI totals? These players aren't slugging any differently from one case to another. In player D's lineup, Bernie Williams looks positively Ruthian in his RBI totals, but in player C's lineup he looks like a pesky banjo hitter. In the other two lineups, he looks like a normal (for 1996-level offense) slugger. Let's look more closely at some of Bernie Williams' team dependent stats in the four lineups. We'll look at RONB (runners on base), RISP (runners in scoring position), PAWR (plate appearances with runners), and LDO (number of innings lead off).
Bernie Williams:
Player A Player B Player C Player D
-------- -------- -------- --------
RONB 651 659 290 882
RISP 388 373 170 523
PAWR 407 439 187 502
LDO 155 161 189 126
There are several things going on here. Player C is never getting on base. Furthermore, his frequent HRs are clearing the bases just before Bernie Williams comes to the plate. Player C also frequently makes outs. Many of these outs are third outs which cause Williams to lead off the next inning. As a result, Williams is starved for RBI opportunities. Player C is an "RBI vulture". His slugging ability allows him to dry up the RBI opportunities for the hitters behind him (which in and of itself is *not* a bad thing), and his weak on base ability does not replenish those opportunities (which is a bad thing). On the other hand, Player D is getting on base more often than not and not making outs. Player C is also not driving in a whole lot of runs. The result is a feast for the players hitting behind him in the lineup.
Think about it. Player C, with his 151 runs and 283 RBIs, helps his lineup to score 975 runs. Player D with his 129 runs and 27 RBIs, helps his lineup to score 1004 runs. The five players hitting behind Player C only get 369 RBIs, while the same five players hitting behind Player C get 622 RBIs. This is an extreme example, but it directly illustrates why the stathead community is so adamant about not using runs and RBIs to measure individual performance.
We still have yet to identify any invisible offense. Let's continue our search...
Player A Player B Player C Player D
AB BB* AB BB* AB BB* AB BB*
-- --- -- --- -- --- -- ---
Raines/NYY 692 109 688 109 674 107 705 111
Boggs/NYY 695 88 691 88 677 86 708 90
O'Neill/NYY 645 120 641 120 628 117 657 123
Player A,B,C,D 645 103 645 99 725 4 384 379
B.Williams/NYY 639 91 636 90 624 84 652 96
T.Martinez/NYY 641 73 637 72 623 69 655 75
Jeter/NYY 639 57 635 56 620 54 653 58
Duncan/NYY 665 13 661 13 645 12 681 13
Girardi/NYY 611 49 607 49 592 47 625 50
--- --- --- --- --- --- -- ---
Total: 5872 703 5841 696 5808 580 5720 995
PAs: 6575 6537 6388 6715
What's this? Player D's lineup gets 327 more plate appearances than Player C's lineup. That's 36 more complete trips through the lineup! It hardly seems fair. A typical game will have about 39 plate appearances. Considering that each lineup only played 162 games, there's a whole lot more opportunities for run scoring in player D's lineup. Perhaps we're beginning to shed some light on this concept of invisible offense.
[ Home ] [ Library ] [ Bookstore ] [ Contact ] [ Search ]
Last Updated: Contact webmaster@stathead.com for corrections or problems
Copyright 1997-2001 by Keith Woolner. All included authors retain the copyrights to their original works.