Sometimes, once in a while, you do studies that completely fail.   Actually, most of my studies fail.  I would guess that, of all the studies I have done in my half-century of baseball research, 80 to 85% have completely failed, meaning I didn’t learn anything at all from doing the research, and another 5 to 10% have mostly failed, meaning that I picked up something but didn’t get to what I was trying to get to.  

            I’ll explain what I was trying to do, and then I’ll explain why it failed, which is not my usual practice but it seems better this time, and then I’ll report the data.   What I was trying to do was see whether teams were most successful, in retrospect, when they used a "quick hook" strategy or a "slow hook" strategy.   We’re in the World Series, and the often-insightful-but-always-annoying John Smoltz said that its always been the case that you get to the bullpen quicker in the World Series.   I thought about that, and I thought "well, that’s true, isn’t it?", but then I thought "Does that strategy actually make sense?  WHY should you go to the bullpen quicker in the World Series than you would in the regular season?   Does that actually increase your chance of winning?" 

            So then it occurred to me to ask "Do teams win more often when they go to the bullpen quickly (Quick Hook) or when they stay with the starter (Slow Hook)?"   The answer I got. . .I’ll share this with you, although it is not convincing and really doesn’t mean anything, but the answer I got it "Neither."  You have the highest winning percentage in the center of the chart, neither quick nor slow; you’re a little over .500 with a Quick Hook and well under .500 with a Slow Hook, but you’re best off being neither quick nor slow.  

            I was trying to use the pitcher game logs.  I have a big file which has almost every start by a pitcher since 1921, the great majority of starts 1921 to 1949 and almost every one 1950 to 2018, which is data that comes from Retrosheet.org.   (Which turned out to be the wrong data set to be using, but that’s another story.)   There were 329,994 pitcher starts in the file, but I had to eliminate 52 games in which starting pitcher was removed before facing a batter (due to an injury, I would presume) and 629 games which ended in a tie—data which only makes sense if there was a game in which one starting pitcher was removed due to injury AND it ended in a tie, but let’s move on. 

If a pitcher gives up 10 runs in six innings before he is pulled out of the game, obviously that’s a Slow Hook, and if he is pulled out of the game after 5 shutout innings, obviously that was a Quick Hook.   But how do you systematically sort that into Quick Hooks and Slow Hooks? 

            Start with a "Hook Score".  I figured the Hook Score for a game as:

              3 times the number of outs recorded by the starting pitcher, plus

              2 for each hit allowed, plus

              5 for each run allowed, plus

              3 for each EARNED run allowed, so that an earned run actually counts 8 altogether, and

              3 for each walk surrendered. 

            It was three for a walk, honestly, because that made the average score for all the games in my data 100, and I like things to average 100.   That wasn’t why the study failed.   Anyway, the slowest hook in my data (essentially 330,000 games) was for Bob Smith of the Boston Braves on May 17, 1927.   He pitched 22 innings, giving up 20 hits and 4 runs, 3 of them earned, and walked 9 batters.  That’s a Hook Score of 294.  Usually you pull the starting pitcher out no later than the 20^th inning.   Second-highest was Ted Lyons on May 24, 1929; 21 innings, 24 hits allowed, 6 runs, 6 earned runs, two walks.  That’s a Hook Score of 291.   That ruined Lyons’ arm, so that he won only 22 games the next year and had to retire just 17 years later. 

            Anyway, the study yielded unconvincing results because of the way that innings and hits allowed are mixed in the formula. . .not the RATIO in which they are mixed; just the FACT that they are mixed.   I’ve used formulas like that before, for example for separating "Quick Hook" managers from "Slow Hook" managers, and that approach work fine for some tasks.  It just creates problems for this particular study.  

            Obviously the standards have changed over time (we no longer allow starting pitchers to pitch more than 20 innings in a game), so the distinction between Quick Hooks, Normal Usage and Slow Hooks has to be adjusted for each year.  The averages actually haven’t changed as much as I would have expected.   The average Hook Score in 1921 was 114.  It stayed in that range until 1940, touching 116 a couple of times.  It declined a little bit during World War II, but was still at 110 in 1944.  It declined by about one point a year after the war, dropping to just over 100  (100.6) in 1957.    From then, it didn’t really change until 1980; the averages dropped as low as 96 in the mid-1960s because fewer runs were scored and runs are part of the system, but the standards for when take the pitcher out of the game didn’t really change.  The numbers began to edge downward after 1980, but very slowly; from 2002 to 2006 the numbers were 96.20, 96.20, 96.50, 96.60, and 96.42.   

            So we learned something here; we learned THAT.  There actually is a change hidden in the averages.   In the 1950s good starting pitchers were allowed to pitch 10, 11, 12 innings in a game if that was required, more sometimes, but also, it was surprisingly common in that era for a #3 starter or a #4 starter to be jerked out of the game in the second or third inning with the score still 2-0 or 2-1 or 3-1.  After 1970 that didn’t really happen, but in 1958 a so-so pitcher would give up a couple of hits in the first inning, give up a homer in the second inning, walk a couple of batters in the third, and the manager would pull him out of the game, saying that "Johnny just didn’t have his good stuff today."  Comparing 2006 to 1960, there were many fewer complete games by starters—but also many fewer early in-game removals.  

            Beginning in 2007, the averages started to drop steadily: 94-95 in the years 2006-2011, then. . .well:

            The point is that, while the numbers have not yet changed as much as I might have guessed, a game that represents a Quick Hook in 1926 might be Slow Hook in 2018.    Let me check if that is literally true. . .yes, it is.  In 2018 a score of 93 counts as a Slow Hook, whereas in 1926, 1929 or 1930 a score of 106 counts as a Quick Hook.  That may be a contributing factor to the failure of the study; not sure whether it is or isn’t.

            Anyway, each SEASON was then divided into three groups:  Quick Hooks, Slow Hooks, and Captain Hooks.   As nearly as possible each season had one-third Quick Hooks, one-third Slow Hooks, and one-third starts in which the starting pitcher was neither pulled quickly nor left in a long time.   80% of complete games or more would be considered Slow Hooks, although some complete games in the past were in the neither/nor category. 

            At this point we had 294 Groups of Games—Quick Hooks, Slow Hooks, and Normal length pitching performances (3 groups) in each season from 1921 to 2018 (98 seasons.)  The goal was determine whether teams were more successful when they had a Quick Hook or a Slow Hook.  However, it was necessary to control for offensive support, since obviously offensive support influences how long you leave the pitcher in game.  If you’re up 11-0 and the pitcher gives up a two-run homer, you don’t worry too much about that; if it’s 3-0 and he gives up a two-run homer, you probably take him out.   I thus had to divide the 294 Groups of Games in nine additional groups, by offensive support, Runs Scored by the offense in the game (0 runs, 1 run, 2 runs. . .8 runs or more.  You obviously would not distinguish between 12 and 13 runs of support, since, after a certain point, the winning percentage is going to be near 1.000 whether you use a Quick Hook, a Slow Hook, or a Sewing Needle.)

            So that gave us 2,664 groups of games—1926, Quick Hook, 7 runs offensive support; 1964, Slow Hook, 5 runs of offensive support; 2017, Central Group, 3 runs of offensive support, etc.    Because the study was so large, there were still about 1200 pitcher starts (average) in each group. 

            The study shows that if the team scores 7 runs or more, then you’re more likely to win with a Slow Hook:

            Whereas if you score 1 to 6 runs, you’re more likely to win with a Quick Hook:

            Whereas if your team scores zero runs, you’re really unlikely to win no matter what you do:

            This is, in a sense, something baseball people have always known—that if it’s a low-run game, you need to act quickly to keep things under control.  That’s commonplace wisdom. 

            In the aggregate, teams are most successful in the games in which they have neither a quick hook nor a slow hook, but better off with a quick hook than a slow hook. 

            But this doesn’t really mean anything, because there is an obvious confusion of cause and effect here—obvious now that I have done the study; I didn’t see it in advance.   Suppose that it’s the 7^th inning, score 4-2, two men on base; the pitcher stays in the game and gives up a three-run bomb; then all of a sudden it was a slow hook.   But if it is the same situation and he gets out of the inning, then it ISN’T a slow hook; if he is taken out after the inning it might still be a Quick Hook (in some eras.)  It isn’t that the Quick Hook helps you win; it is that failure causes your game to be sorted as a Slow Hook.

            So the study fails.  What I would have had to do to make it work is to isolate sets of game situations BEFORE the decision was made to remove or not remove the starter.   But I don’t have a data set that is suitable to run that study, so I’ll have to leave it to somebody who has the skill set to do that. 

            Thanks.  I appreciate the positive feedback I got the last time I reported failed results.