Streak Study by Good Games and Bad Games
There are, of course, hundreds of studies of hot and cold hitters, most of which suggest that either (a) hitters do not get hot and cold, these are merely random groups of events, or (b) the extent to which hitters are hot and cold over the course of the year is so small that it has no practical impact. I don’t disagree with this; I have studied the same thing myself many times and found the same thing. However, even if we assume that hot streaks do not exist, this does not mean that they cannot be measured. Over the course of a year, a hitter will have stretches where he is effective and stretches where he is not so effective. There may nothing "real" behind it, but the divided outcomes can still be measured.
So I had this idea: suppose that we divide a hitter’s games evenly into "good" games and "bad" games, based on the hitter’s season. Let us suppose that the player plays 160 games; we rank the 160 games 1 to 160 in order of the number of runs that he has created in each game, and we divide those into his 80 best games and his 80 worst games. We then look at the sequence of games, simply as Good/Bad/Good/Bad/Bad/Good/Good/Good/Bad/Good, etc.
With an equal number of good and bad games, a player should, if he has no real tendency to get hot or cold, have an equal number of times when the pattern goes good/good as when the pattern goes good/bad. If he does have a "real" tendency to get hot and cold, then he should have more games that go good/good than games that form a random pattern. I decided to look at the "hot hand" phenomenon by studying those patterns.
1) I took a fairly large number of player/seasons, 637 player/seasons involving 78,806 games, although some of those seasons (and thus some of those games) turn out to be useless for the study,
2) I ranked all of the games in each season by the number of runs that the player had created in the game,
3) I divided each season evenly into "good" and "bad" games, see note of explanation below, and
4) I studied the good/bad sequences, trying to identify the streakiest and "unstreakiest" hitters in the study, and also to evaluate the overall question of "to what extent" do hitters form streaks?
If a player followed a good game with another good game, or if he followed a bad game with another bad game, that is marked as a +1, meaning that either way, it is an indication of streakiness in that particular hitter. Every time he followed a good game with a bad game or vice versa, I marked that as -1, meaning that that is a "non-streak" indicator. If he had THREE good games in a row, or three bad games in a row, I would mark that as +2. If he had three games in a row in which each one was different than the one before, that would be -2. Four games streaks would be marked as +3 or -3 or zero, and five-game streaks at +4 or -4 or zero. Six-game stretches of games would be marked as +5 or -5 or zero, and then that was the end of the point system; I didn’t continue to mark longer streaks.
Two of those little annoying questions that we have to work through are (1) what do you do with the center game when a player plays an odd number of games so they don’t break evenly, and (2) what do you do when there are ties as to the center game?
I should explain first: it doesn’t REALLY make any difference; it is one of those things which seems like it could make a difference if you have an imbalanced number of good and bad games, but actually it doesn’t. Even if, because of ties, you had to mark 55% of a player’s games as "good" or "bad"—which never happens—but even if that did happen, it wouldn’t make any real difference. If 50% of a player’s games are ‘good" and 50% are "bad", the expected number of two-game streaks is 50%--25% good/good, and 25% bad/bad. If 55% were good and only 45% were bad, the expected number of two-game streaks would increase by only 1%, to 50.5%--30.25% good/good, and 20.25% bad/bad. It’s not a real problem; it’s just a real annoying problem.
But anyway, this being skience and all, I’ve got to explain what I did. If a player played an odd number of games in the season, I always marked the center game as a "good" game. However, to balance the number of good and bad games in the aggregate, if a player had a tie in the center—that is, if he had multiple games with the same runs created and these were in the center of the chart—and if there were only two games which were "tied and in the center", then I would always mark both of them as "bad" games. If there were more than two games which were tied in the center, then I would always mark them the SAME—that is, if one 1-for-4 game with a single is a good game, then ALL of that player’s 1-for-4 games with a single would have to be good games, obviously—but I would mark the group whichever way left us closer to having a balance for the season. For the study as a whole, I wound up with 38,845 "good" games, and 38,961 "bad" games.
I excluded from the study all seasons in which a player played in less than 100 games, and there were a few other seasons that I had to exclude from the data because the format was unworkable. That would happen if the player got into 100 games, but was used mostly as a pinch hitter, pinch runner or defensive substitute. A pinch runner/defensive substitute has a large number of "games" in which he is 0-for-0 without a plate appearance. These games invariably occupy the center of his good game/bad game chart, and it is impossible to say whether they are "good games" or "bad games" as a hitter. They’re neither. A pinch hitter will have a lot of games in which he goes 0 for 1, and those also will wind up in the center of his chart, if pinch hitting was his regular gig. You obviously can’t call those "good" games, and if you call them bad games he will wind up with a 30/70 split good/bad games, which is not useful for the study.
There were 657 player/seasons in my data, but the exclusions cut it to 487 "usable" seasons. In the aggregate, the players having "good" games hit .441 with a .507 on base percentage, .736 slugging. The players having "bad" games hit .095 with a .156 on base percentage, .104 slugging. Per 162 games, the players having good games averaged 257 hits, 45 doubles, 10 triples, 36 homers, 127 RBI, 83 walks, 16 stolen bases, only 6 GIDP. The players having "bad" games averaged 52 hits, 4 doubles, no triples, no homers, 22 RBI, 39 walks, 4 stolen bases and 18 GIDP.
We are now into the category of "marginalia associated with the study" rather than "actual results of the study", but of the 38,961 "bad games" in the study, there are 16 games in which a player hit a home run, but still wound up with what we identified as a bad game. In all 16 of those, no exceptions, the player who hit the home run/had a bad day also grounded into a double play. There are 32 games in the data in which a player hit a triple but had a bad day, but almost 1,000 games in which a player hit a double but had a bad day. There are 154 games in the study (about 4/10th of 1% of the bad games) in which a player had two hits, but it still counts as a bad day, but there are no games in which a player hit two doubles or had two extra base hits, but it’s still a bad day. There are no games in which a player had three hits, but it’s still a bad day.
I didn’t count runs scored/RBI as contributing to the good day/bad day classification; of course I could have, I just chose not to. There are nine games in the study in which a player scored three runs, but still had a bad day, no games in which a player scored 4 runs but had a bad day. There are, however, two games in which a player drove in four runs, but still had a bad day. When I found those I thought I would use them to explain/explore the arbitrary dimensions of the study; of course, if you drive in four runs, it’s not really a bad game. But when I looked at them individually. . .no, that’s really a bad game, or at the least it is not really a good game. Rico Petrocelli on June 10, 1970, hit a grand slam home run, but it still counts as a bad day. (a) The Red Sox lost the game, (b) the Red Sox were never ahead in the game, (c) it was an extra inning game in which Petrocelli went 1-for-6, and (d) he grounded into a double play and struck out. Not a good game—1-for-6 with a double play and you lose the game, never have the lead. Not a game you star on your resume.
The other one was Earl Averill on June 5, 1933. Averill (a) was credited with an RBI on a double play ball, because the scoring rules were different then, (b) drove in a run with a fly ball, (c) hit a two-run single, and (d) his team also lost the game.
"Empty set" 1-for-4 games are a nuisance. A player goes 1-for-4 without a walk, an extra base hit, a stolen base or a double play ball to break the ties. Everybody has games like that, but for some players, like Bill Mazeroski and Dick Groat, those games always wind up in the center of their chart. Sometimes a player has 15 of them in a season—not often, but occasionally—squatting right in the center of his chart; you have to count them out to see which side of the line they fall on, and then you have to make sure that he doesn’t have 55% of his games on one side of the line. . . .it’s a pain, but that’s research.
After shucking down to 487 player/seasons I had 68,696 game lines in the data—34,280 "good" games and 34,416 "bad" games. The streakiest season in the data was either by Maury Wills in 1970 or Lou Brock in 1967, depending on how you interpret the data. The fact that two incorrigible base stealers are at the top of the list will no doubt incline some of you to believe that base stealers are streakier, but no; that’s just a coincidence. Anyway, Wills in 1970 had a "streak total", as explained above, of 192 and a "non-streak total" of 62, making him +130 in his measured streak tendency. Brock in 1967 had a streak total of 253 but a non-streak total of 127, making him +126. Brock has the highest streak total in the study (253), while Wills has the highest streak/non streak score (+130).
Wills in 1970 was 37 years old; he played 132 games, hit .270 with no homers, 28 stolen bases, fair number of walks. Wills opened the season (first three games) in an 0-for-11 slump, then had a very peculiar 11-game hitting streak which included four straight "bad" games, four straight Empty 1-for-4s, in which he also stole no bases, was caught stealing twice and grounded into a double play. Twenty-two games into the season he was hitting .217 with 4 RBI and a .253 slugging percentage. In the next 14 games he hit .442, lifting his average to .288. It was that kind of year; he was hot or cold all year.
Now that I think about it this is not that dramatic a story; since Wills didn’t hit any home runs he never had those 7-game stretches where he hit 6 homers and drove in 16 runs, like some hitters do. Wills followed a good game with a good game or a bad game with a bad game 86 times with 131 two-game stretches in the season, which is a very high percentage; that’s all I am really saying.
Lou Brock’s 1967 season is somewhat more interesting; Brock had 206 hits although he hit only .299, and his 206 hits included 32 doubles, 13 triples and 21 homers, plus he stole 52 bases. These were big numbers in the run-starved 1960s. He finished 7th in the league in the MVP voting, and was 8th in baseball reference WAR. He hit .414 in the World Series, scoring 8 runs in the 7 games, and going 7-for-7 stealing bases. He had a great World Series, a very good season, and, as we see now, a very streaky season. On April 15 and April 16, against two different teams, he had two of the best games of his career, getting four hits in each game, include two homers in each game; in the two games total he scored 7 runs and drove in 8, went 8-for-11 and stole a base. That was near the start of a ten-game stretch in which he hit .500 and created 19 runs. In the twelve games immediately following that ten-game stretch he created less than two runs total, hitting .185 and scoring only two runs. Not very long after that (May 19 to May 29) he had a 10-game stretch in which he created 13 runs, and he stayed hot for a week after that. Then he had an absolutely awful 24-game stretch (June 18 to July 9) in which he created only two runs total, hitting .167 without a single walk, and going 3 for 6 as a base stealer. From July 24 to August 10 he had a 17-game hitting streak during which he averaged almost two hits per game and scored more than a run a game, followed immediately by ten games in which he created an estimated 1.499 runs. He stayed in that slump until August 30, then hit .371 in September with 6 homers, 16 RBI, 26 runs scored in 26 games and 12 stolen bases. I don’t know if this sounds unusual, since all hitters go through slumps and streaks every year, but take my word for it; his slumps and streaks that year were quite unusual.
The counterpoints to Wills and Brock, in my study, were (1) Eddie Mathews, 1953, (2) George Scott, 1966, and (3) Ken McMullen, 1971. That’s an even less dramatic story; those guys just went bad-game-good-game-bad-game-good-game most of the year. Mathews had a "positive streak total" of 88 for the season, a negative streak total of 252, for a balance of negative 164—a larger negative balance than Brock and Wills had positive. Scott was 93-215, negative 122, and McMullen was 98-218, negative 120.
On balance, the average player in this study had a positive streak balance of (+) 7.46. The average player in the study played 141 games, with an average of positive streak score of 139, and a negative streak score of 131.5.
Of course, a positive streak balance is not proof of streakiness; it merely indicates that there is some clustering of good games and bad games within a player’s season. There are 34,683 cases within the data in which a player followed a good game with a good game or a bad game with a bad game, and 33,626 cases in which the two games did not match—50.8% "matches", 49.2% "non-matches".
Among the elements that could cause a player’s good games and bad games to cluster, independent of "streakiness" or confidence issues, are:
1) The home field advantage,
2) Injuries,
3) Having some games in hitter’s parks and some in pitcher’s parks,
4) Playing a group of games against good teams and against weak teams,
5) Weather variables. Batters hit much better in 90 degree heat than in 60 degree heat, which would certainly create some clustering of types of games.
To what extent the clustering which is observed in this study is a result of these factors is an unknown, and a question capable of further research.
For whatever it may contribute to that research, players in this study had 17,449 "good" games when they were at home, and 16,924 "bad’ games; that is, they had a .5076 "good game percentage" when they were in their home park. On the road, they had 16,831 good games, and 17,492 bad games, for a good game percentage of .4904. This data would not contribute at all to the clustering of good games and bad games; in fact, because there are slightly more "bad games" than "good games" in the data, the expected matches of consecutive games (based on these percentages) would be slightly LESS than 50%.