One of the most pervasive beliefs among sports fans is a belief in "streaks". I cannot tell you the number of times I have heard sports commentators this week tell us that the Rockies have won 21 of their last 22 games. And this alone is
The Red Sox win, the Pats are kicking Colts ass... its a great time to be a Boston fan. The Celtics are actually looking good too. I'm thinking about putting some money on a betting line I found, if all 3 teams win the championship I get 10:1 (the patriots are a sure bet, the celtics could pull it off) What do you think?Go BOSOX!
J-My objection is to the misuse of statistical analysis.Why quote a flawed study when less flawed more acurate studies exist?Why do a statistical analysis on a situation where the basic assumptions that the math is based on don't fit?Why make far-reaching conclusions on analysis that don't apply to the situation being analyzed?(Am I ranting yet?)
I don't understand the objection. Sometimes when one side gets a streak it's because of luck, and they're no more likely to win next time because of it.
Sometimes it's because they have something special going for them that changed the odds.
How do you tell which it is, unless you get special knowledge?
Then maybe the streak ends. It's unusual for somebody to come up from behind unless they have something special going for them, that's new. Other things equal, if you get behind you're likely to stay behind. You can get a streak because you get some special advantage, and the streak ends when you lose the special advantage or somebody else gets a better one. Or you can get a streak by sheer random chance.
If people suppose it's something special more often than it is, how could we balance that?
The studies that show that there is such a thing as a hot-hand include Smith (2003), Frame et al (2003), and Dorsey-Palmateer and Smith (2004).These studies were done on horseshoes and bowling games where their is no defense on the guy who gets hot.One of the reasons studying basketball is not appropriate for the study of hot hands is that the guy who gets hot gets defended differently. This change in defense (Opposing coaches would often have their players foul me or double cover when I got hot, for example) makes the study completely different than studying a random phenomena.I wonder if Gilovich ever played basketball.Any game where the defense can change their stategy in response to someone getting hot is not a good game for the study of "randomness".
In most social betting venues, for example horse racing, there is a bias in favor of the long shot: that is the long shot's odds as set by the bettors are usually not quite as long as the actual odds (measured statistically afterwards). Apparently people are willing to pay extra for the bragging rights to having won on a long shot. However, the quoted odds usually do not get so out of whack with the actual odds as to be larger than the vigorish -- that would create an arbitrage opportunity.
So here's what I've never understood about the Tversky et al. study, and what makes me think that it's missing something. They start out by defining the problem thus:
"Such runs can be properly called streak shooting, however, only if their length or frequency exceeds what is expected on the basis of chance alone."
Wait a minute, why should THAT be the definition of "streak" shooting? Maybe streaks happen no more often than "chance" would predict, but that doesn't mean that streaks are themselves really just the equivalent of mere "chance." Anyone who has ever played basketball regularly knows that on some occasions, everything clicks: you've spent hours practicing, you're well-rested and on top of your game, you're feeling confident and invincible, and what do you know, your shots are nothing but net. This might occur no more often than would be predicted by chance, and sure, on every other occasion, a hit is no more likely after a hit than after a miss. But on this one occasion, it's not just chance at work -- it's a combination of skill, confidence, practice, and other factors that are allowing you to hit 15 in a row, and your teammates would be foolish to cut the streak short at 2 by letting everyone else have an equal chance to shoot.
Point taken on the bias, and maybe you aren't literally going to bet, but...The Red Sox are about 2-1 favorites most places and no team is ever a 2-1 favorite in the world series. Way too much randomness in that sample size.
Eliezer_Yudkowsky beat me to it. (I was assuming Justin_Wolfers meant 1.)
But I can probably kill some more joy with a remark like, "So, remind me real quick, what's the algorithm you're going to follow to place future bets that you believe will provide superior returns, and for which you will concede the error of your analysis if it doesn't?"
(1a) you think that others are betting on the streak and also (1b) more informed speculators haven't hoovered out all the free money already?
Or (2a) you think there's an actual anti-hot-hand, a true version of the gamblers fallacy for sports, and (2b) other speculators don't know this and haven't hoovered up the free money already?
The Red Sox win, the Pats are kicking Colts ass... its a great time to be a Boston fan. The Celtics are actually looking good too. I'm thinking about putting some money on a betting line I found, if all 3 teams win the championship I get 10:1 (the patriots are a sure bet, the celtics could pull it off) What do you think?Go BOSOX!
And now you look like a genius for this post, as the Rockies looked helpless for four games.
Of course, if the Rockies had a hot hand, it sure didn't benefit from lying between the ground and an @ss for eight days. Good scheduling job, MLB.
-- mobile the Rockies apologist
If you are actually going to make the bet, how much of your bankroll are you going to bet and why?
Aren't those who believe in streaks balanced by those who believe that a team is due? We used to joke about this where I once worked.
J-My objection is to the misuse of statistical analysis.Why quote a flawed study when less flawed more acurate studies exist?Why do a statistical analysis on a situation where the basic assumptions that the math is based on don't fit?Why make far-reaching conclusions on analysis that don't apply to the situation being analyzed?(Am I ranting yet?)
I don't understand the objection. Sometimes when one side gets a streak it's because of luck, and they're no more likely to win next time because of it.
Sometimes it's because they have something special going for them that changed the odds.
How do you tell which it is, unless you get special knowledge?
Then maybe the streak ends. It's unusual for somebody to come up from behind unless they have something special going for them, that's new. Other things equal, if you get behind you're likely to stay behind. You can get a streak because you get some special advantage, and the streak ends when you lose the special advantage or somebody else gets a better one. Or you can get a streak by sheer random chance.
If people suppose it's something special more often than it is, how could we balance that?
The studies that show that there is such a thing as a hot-hand include Smith (2003), Frame et al (2003), and Dorsey-Palmateer and Smith (2004).These studies were done on horseshoes and bowling games where their is no defense on the guy who gets hot.One of the reasons studying basketball is not appropriate for the study of hot hands is that the guy who gets hot gets defended differently. This change in defense (Opposing coaches would often have their players foul me or double cover when I got hot, for example) makes the study completely different than studying a random phenomena.I wonder if Gilovich ever played basketball.Any game where the defense can change their stategy in response to someone getting hot is not a good game for the study of "randomness".
In most social betting venues, for example horse racing, there is a bias in favor of the long shot: that is the long shot's odds as set by the bettors are usually not quite as long as the actual odds (measured statistically afterwards). Apparently people are willing to pay extra for the bragging rights to having won on a long shot. However, the quoted odds usually do not get so out of whack with the actual odds as to be larger than the vigorish -- that would create an arbitrage opportunity.
What about the bias of rooting interest? There are a lot more Red Sox fans around the country who want to bet on their team than Rockies fans.
The Yanks were something like 2:1 over the Marlins a few years ago.
I think Justin conveniently ignored the biases towards favorites and the market rate in order to push his points about hot streaks and Beantown love.
Baseball Prospectus has the series at 60/40 (see URL).
Tonight's game in itself is going off at 2:1 (Beckett v Francis).
So here's what I've never understood about the Tversky et al. study, and what makes me think that it's missing something. They start out by defining the problem thus:
"Such runs can be properly called streak shooting, however, only if their length or frequency exceeds what is expected on the basis of chance alone."
Wait a minute, why should THAT be the definition of "streak" shooting? Maybe streaks happen no more often than "chance" would predict, but that doesn't mean that streaks are themselves really just the equivalent of mere "chance." Anyone who has ever played basketball regularly knows that on some occasions, everything clicks: you've spent hours practicing, you're well-rested and on top of your game, you're feeling confident and invincible, and what do you know, your shots are nothing but net. This might occur no more often than would be predicted by chance, and sure, on every other occasion, a hit is no more likely after a hit than after a miss. But on this one occasion, it's not just chance at work -- it's a combination of skill, confidence, practice, and other factors that are allowing you to hit 15 in a row, and your teammates would be foolish to cut the streak short at 2 by letting everyone else have an equal chance to shoot.
Point taken on the bias, and maybe you aren't literally going to bet, but...The Red Sox are about 2-1 favorites most places and no team is ever a 2-1 favorite in the world series. Way too much randomness in that sample size.
Eliezer_Yudkowsky beat me to it. (I was assuming Justin_Wolfers meant 1.)
But I can probably kill some more joy with a remark like, "So, remind me real quick, what's the algorithm you're going to follow to place future bets that you believe will provide superior returns, and for which you will concede the error of your analysis if it doesn't?"
Actually, I guess I am pretty much a killjoy.
Not to be a killjoy, but...
You're betting against the Rockies because:
(1a) you think that others are betting on the streak and also (1b) more informed speculators haven't hoovered out all the free money already?
Or (2a) you think there's an actual anti-hot-hand, a true version of the gamblers fallacy for sports, and (2b) other speculators don't know this and haven't hoovered up the free money already?