Portrait of a reliever: Firpo Marberry, 1925by Brandon Isleib
September 24, 2009
It's been a wonderful year for historyish baseballish types like myself. Why? Retrosheet now has daily box scores of the entire 1920s. The site doesn’t have play-by-play data through all of it, but just having the box scores is amazing enough.
The 1920s were the petri dish for station-to-station offenses and a few other things, but of primary interest is relief pitching. Prior to 1925, only eight pitchers had accumulated 20 or more relief appearances with zero starts in a season, Lou North’s 40 for the 1921 Cardinals setting the highwater mark. Five of the eight seasons were from John McGraw’s Giants, and another was Allan Russell of the world champion Senators of 1924. In ’25, McGraw sent Jack Wisner to relieve 25 times without starting, but our man Firpo Marberry broke all barriers by relieving 55 times without starting—a record that would stand until 1939. If you flip casually through record books, the number jumps off the page, anachronistic in about every way possible.
The problem for years has been that the number looks so modern in context that the performance end is obscured. Was Marberry really the first “modern” reliever? Does that 15-save total mean he was a closer? Fortunately, we now have the box scores and the measurements to piece the answer together. Using gmLI, or Game Leverage Index, a Tangotiger-made, Fangraphs-maintained metric that ranks the importance of the situation in which a reliever entered—the strength of the fire Marberry came to fight—we’ll get a full handle on what happened.
More backgroundRising offenses and logic dictated that relief pitching would come eventually; as with Lou North and the cadre of Giants, it had been coming for the last several years. It was Marberry who broke it wide open, and with good reason. The 1915 Chicago Whales aside, the 1924 Senators were the first team to get at least 15 wins from each of two starters who were 35 or older, in their case Walter Johnson and George Mogridge. This still is rare; counting the Whales, it’s happened 17 times, the last being in New York in 2006 (Randy Johnson/Mike Mussina for the Yankees and Tom Glavine/Steve Trachsel for the Mets). The Senators followed up their historic campaign by trading for Stan Coveleski, 34 (his baseball age would be 35 in 1925) and coming off a down year in Cleveland. This left them with three rather old stars, and it was unique enough that they had succeeded with two. If the Senators repeated, it would be more reliant on aging starters than any previous winner.
Speaking broadly, aside from McGraw it was player-managers who most advanced the state of roster maneuverings, in several cases taking themselves out of the game to do so. Tris Speaker and Senators’ second baseman-manager Bucky Harris were among the foremost users of pinch-runners and defensive substitutions. Perhaps it was because they were from a newer generation of managers, or because they as players were more likely to react to game states in making decisions (the difference in a field marshal versus the general back at headquarters), but on balance the player-managers found the most uses for their bench.
Without having read specifically on Harris’s thoughts on Marberry, my guess is that having Johnson, Coveleski, and Mogridge in his opening day rotation cemented a need for a good exclusive reliever, making Marberry the pitcher version of what Harris already was doing through his lineup. Whatever the thought, it worked; until Glavine and Greg Maddux hooked up on the 2001-02 Braves, the 1924-25 Senators were the only team to get back-to-back seasons of 15 wins from two pitchers 35 or older, and only one of three to get 20 wins in a season from both of them (the others were the 1956 Indians with Bob Lemon and Early Wynn, and the 2002 Diamondbacks with Randy Johnson and Curt Schilling).
If you’re getting the sense that Marberry wasn’t used quite as closerly as superficial inspection would reveal, consider yourself prescient. But let’s get to the actual figures for verification.
gmLI isn’t just a dwarf
Simply put, gmLI looks at the game state to determine how leveraged a reliever was as he entered the game. Through the box scores, I was able to get half of the game states to a certainty. That half usually was because the previous pitcher had been replaced with a pinch hitter, indicating that Marberry started the inning he came in. The other half I had to determine as much as possible via deduction. All the box scores gave how many batters each pitcher faced, which was crucial in narrowing down a few things.
For example, in Marberry’s first game of 1925, he was brought in to face pinch hitter Wally Schang after reliever Vean Gregg had faced three batters and gotten one of them out. So we know there was one out when Marberry entered, and at least two runners on. What we don’t know from that is whether the run Gregg gave up had already scored when Marberry entered or whether it was an inherited runner Marberry let score (inherited runners in box scores are the bane of this project). But looking at Schang’s pinch-hitting line, which was a single that scored two runs, we know that not only Gregg’s run but one of Stan Coveleski’s scored. Until then, we didn’t even know that Coveleski had started the eighth and failed to record an out. As this was the only hit Marberry gave up the whole game, and as Marberry allowed no runs himself, we know therefore that he entered an 8-5 game made 8-7 by Schang’s single, meaning the only thing we don’t know is whether there were two or three runners on. While that changes the leverage index significantly, it’s a lot more narrowed down than a cursory look at the box score provides.
In the other appearances, the ones in which he entered the game mid-inning, we don’t know exactly how many base runners there were, but thanks to the lack of specialization in the era, we can assume with confidence that there was at least one, eliminating the bases empty scenario. This is verified in Marberry’s season by multiple instances of entering the game with only one or two outs left in a close game, and how many times Marberry entered the game after the previous batter walked (or like on June 27, when the pitcher had just singled). By implication, Harris sent Marberry in at the first sign his old starters “lost their stuff.”
For every scenario where runners are unknown, I took the median for presumed leverage; this ordinarily would presume runners on first and third. It’s not perfect, but it gives a range and a practical upper limit on Marberry’s leverage. I call the median PgmLI (Projected gmLI) for ease of discussion, usually giving the low end as well. For reference, in Tangotiger’s system less than 0.8 is low leverage, 0.8-1.5 is medium, 1.6-2.9 is high, and 3.0 and above is very high.
April/MayMarberry was used sparingly until mid-May; the Schang game above was his first of the season and has a PgmLI of 2.4 (low 2.2). For the next two weeks, he was used only three times, with his highest gmLI at 0.6. His season didn’t hit top gear until a four-game series in Cleveland, when he pitched three times. Although only the first game of that series was even medium leverage, he gave up no runs over those games, saving the first one; he was used in increasingly tough situations after that. His May PgmLI was 0.99 (low 0.72, high 1.45, but the high only comes if the bases are loaded every time) – decent leverage at best. He also gave up five runs in one of the highest-leverage situations, entering the top of the eighth in a tie game versus the Red Sox (gmLI of 1.9).
JuneIn the first 10 days of June, Marberry pitched five times, all in situations with PgmLI of at least 1.0, and saved three of them, giving up only two runs and that in a game he saved anyway. For the rest of the month, the only runs he gave up were in a mid-month series against the Browns. To be fair, he was brought in twice to face Baby Doll Jacobson and once to face George Sisler, so he wasn’t up against patsies.
The game in that series in which he didn’t give up a run was a two-out, runner-on-first situation in the ninth inning with a one-run lead, good enough for a 2.9 gmLI. Balancing that leverage out, is a save he earned on June 23 while preserving to a seven-run lead. Still, his June PgmLI was 1.35 (low 1.11, high 1.79), so it’s clear that Marberry was more a fireman than before.
JulyJune was mostly at home; almost all of July was on the road, and it featured some close games. Of Marberry’s 13 appearances, either six or seven came in one-run games. (The one in question is July 2, where Marberry entered the seventh inning with two outs while the Red Sox apparently were batting around. The hits Marberry gave up in that inning scored an undetermined amount of inherited runners. I've never hated the inherited runners concept before this project, but now it's quite annoying. Make it stop.)
It wasn’t a great month for Marberry—his ERA was 5.96—but he came through a number of times. Marberry saved both ends of a doubleheader against the Yankees on July 26, entering the bottom of the ninth with a three-run lead in the first (gmLI of 1.0) and the bottom of the 11th with a one-run lead in the second (gmLI of 3.6). Mind you, three days earlier he had allowed six runs trying to protect a one-run lead, so it wasn’t all fun and games. But his leverage was coming much closer to a modern closer: July’s PgmLI was 1.7 (low 1.15, high 2.24).
AugustThough Marberry wasn’t all that helpful in July, his performance in August (3-1, 1.83, five saves) was critical in getting the Senators to first place and keeping them there against an upstart A’s team. Aside from an easy one-out save on August 19 and a mop-up role on August 22 in an 11-5 game that became 20-5 by the end, all his gmLIs were at least 1.0, peaking for the month and season on August 6 while preserving a one-run lead with at least one runner on and one out in the top of the ninth (PgmLI 5.7, low 3.9, high 7.3). Games like this gave him a 2.05 mark on the month (1.62 low, 2.88 high). This is premium closer range and not typically sustained over a season; nobody this year has a gmLI above 2.01, and only three relievers—Jonathan Papelbon, Joakim Soria, and J.P. Howell—were at 2.05 or better this August. Even 1.62 comfortably is within closer territory. If at any point a 1925 Senator could look like a modern closer, it was here, and Marberry’s success in the month was important from any angle.
His one loss was an extra-inning affair in which he gave up a single run; this was a bit unfair, perhaps, but he still deserved a loss for the month. It’s games like this that make me excited for old box scores: August 21 against the Tigers. The game had been scoreless until the eleventh, when Stan Coveleski put Al Wingo on, either via walk or triple. With the much more formidable Heinie Manush, Harry Heilmann, and Lu Blue due up with no outs, Harris called on Marberry to get the job done. He walked all three, ending the game without getting the blame in the box score that he so richly deserved. He entered a high-leverage situation (anywhere from 2.1 to 3.1) and messed it all up. Not to be mean to his memory, but this is my favorite game of his in 1925, since it’s a box score that reveals an unusual play-by-play when you dig into it.
(By the way, facing three batters and walking all three has happened just 123 times since 1954, with none from 1966 to 1968.)
September and conclusionThe A’s were fading quickly by the end of August, with the Senators sitting on a three-and-a-half-game cushion going into the stretch drive. Perhaps because of this, Marberry was used in only three games, all from September 17 to 20, and all protecting leads of at least three runs.
All told, Marberry’s PgmLI for 1925 is 1.49, with 1.14 as the low and 2.07 as the unrealistic high. 1.49 is a fairly normal gmLI for, say, a closer on a team with few opportunities. Francisco Rodriguez’s gmLI is 1.47 this year; Phil Hughes is at 1.54. While I do think my projection method is a bit generous, some of that is given back by Marberry’s mop-up usage in April and September. Taking only the games from May 16 to August 30, Marberry’s PgmLI is 1.59, with a low of 1.23 and a high of 2.15. This is more representative of how Marberry was used over the bulk of the season, and making that estimate more conservative would bring it to around 1.4, which is about where I would put him after examining the data.
So it’s most accurate to say that Marberry’s use was somewhere in between setup man and closer. The best comparison that springs to mind is Mel Rojas setting up John Wetteland in Montreal, where Rojas would get about a dozen saves just from pitching so often. (Rojas’s gmLI in that period was around 1.5.) You may have looked at Marberry’s stats and come to that conclusion yourself, but it’s always best when you can put numbers and evidence to a perception, especially when discussing a reliever of 75 years ago.
References and Resources
Retrosheet, specifically Firpo Marberry's daily page;
Tangotiger's Leverage IndexTangotiger's Leverage Index and how he got there.
The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org".
Brandon Isleib is a lawyer and writes about stuff sometimes. He can be reached via the electronic mails.