A few days ago I made a comment about the consistency of the Minnesota Timberwolves. The point I was trying to make was that given the past productivity of the players employed by Minnesota, one should expect this team to be hovering around the average mark. This means this team will win some and lose some.
When Kevin Garnett and Kevin McHale see this team win, though, they seem to think (judging by their quotes) that if the team could always play as well as it did when it won, it would win more frequently. In other words, from their perspective, the team loses because it is inconsistent.
I argued the opposite. The team loses fairly often because its players perform in a fashion consistent with past performance. Many of the players on the Timberwolves are below average, and consequently, we should expect this team to lose with some regularity. Not all the time. But more frequently than either Garnett or McHale want.
The Wisdom of Brian Goff
Okay, I restated the argument (badly again, I think). Here is a similar argument from Brian Goff, a fellow contributor to the Sports Economist.
ESPN replayed a little bit of Billy Donovan’s post-game press conference Sunday night. In response to an off-camera question that seemed to be about the Gators’ struggling to beat Purdue, Donovan offered remarks to the effect:
“The difference between teams like Purdue and us is not big. We’re a good basketball team, but so is Purdue. The margin for error in games like this on neutral courts is very small.”
His remarks were longer and more extensive but expressed, at least implicitly, two basic points that often escape the talking heads in the media. First, the average average performance level for the better teams in the tournament and those below them is not large. With all of the movement of young players to the NBA in recent years, very few teams have several juniors or seniors likely to play in the NBA. Most of the NBA-bound or NBA-impact players are freshmen or sophomores. With home court (or near home court) advantage removed in most NCAA games, this average performance difference moves even closer.
Second, team performance varies around this average level. That’s second nature to people in economics or statistics. Yet, sports media analysts frequently talk as if performance levels are fixed, or, at least, should be if coaches/players were really “focused” or some similar statements. Instead, variations in performance are going to occur for lots of reasons other than lack of preparation. Team-specific match-ups, player health, random bounces of the ball, officiating and other factors create variable performance.
Maybe nowhere do I see this lack of understanding more than when golf analysts talk about Tiger Woods. During some of Tiger’s winning streaks, some of these guys have seriously wondered whether anyone will ever beat him again. After his devastating performance in the 2000 U.S. Open, this kind of talk exploded as it again recently. In effect, the observers treated the upper end of his performance distribution as his average (seems to be a common occurrence among amateur golfers, also — I’m sure it has a cognitive science name).
For those of us who are in economic education, we should be careful not to undersell the value of fundamental ideas like this one or assume that the general point is widely grasped and easily applied to specific contexts.
An Attempt to Connect Two Stories
Both my story, and the argument Professor Goff offers, centers on the issue of sample size. Garnett and McHale observe their team win a game, or a small collection of games, and conclude that their team is really good (if it could play that well all the time). The larger data set – based on the player’s career performances – suggests something different. The larger data set gives us a better picture of what a player’s average, or expected performance, will be. And when we understand that picture, we see quite clearly why the Timberwolves will win a few games here and there, but are not likely to be consistent winners.
As we watch the NCAA Tournament it is the same story. The teams in the Sweet 16 tend to be among the best in college basketball. Tonight eight teams will offer us a performance, which may be better, the same, or worse than their average performance. If a better team performance worse than their long-run average, and a worse team performs better than their own long-run average, an upset will occur. This upset will tell us nothing about the nature of either team. Members of the media, though, will tell us stories about this upset as if something about the true nature of the players and teams could be inferred from one data point.
And Now I Will Be Inconsistent
All that being said, I am not sure life would be better if the media reported these events as Goff and I suggest. Do we really want to live in a world where the media tells us after each game “well, you can’t really draw an inference from this game. After all it is only a sample of one. Either team could have won this game and we do not know any more about the quality of these teams and players now then we did before the game was played. Basically, there is nothing to be learned here so let’s just get on with our lives.”
See, that wouldn’t make for interesting commentary at all. Yes, if you have some understanding of statistics it is a problem when people draw inferences where none can be drawn. But I am not sure I have a suggestion for what else the people in the media should say when the game is over. They have to say something, and the pure statistical answer is probably not going to appeal to very many people.
– DJ
Westy
March 23, 2007
This upset will tell us nothing about the nature of either team.
But wouldn’t it tell us at least that the lesser team was capable of beating the better team? Sure, it might be an upset (the other team would win 6 or 7 times out of 10), but many times there is a feeling (amongst the ‘media’) that the “better” team would win 10/10, and this has been proven wrong when the “lesser” team wins.
Also, I wonder, is there nothing to the notion of great teams fine-tuning their performance and performing at a higher level in crunch time? Is it fair to say that in the playoffs and tournament that a regression to their past mean performance is all that can be expected?
dberri
March 23, 2007
Westy,
I agree the upset tells us that it was possible for the lesser team to win. Hopefully we all knew that was possible, but it is true that often people present these games as if the outcome is a foregone conclusion.
I don’t think regression to the mean is what we see in the playoffs. What we see a random draw from a series of potential performances. The average in the series represents the team’s “true” nature. The random draw we observe may or may not be representative of that average.
Not sure I buy the idea that people can raise their level of performance systematically in the playoffs. At least, one of the stories we tell in The Wages of Wins is that NBA players do not do this.
Jason
March 25, 2007
“Average” players are rarely average because they consistently perform at a level consistent with average per -minute (or per game ) levels, but because their above average performances aren’t common enough or good enough to outweigh their below average performances. Some of this is influenced by variable abilities of opponents, some of this may be a product of the nature of performance. There are *always* some games better than others.
It becomes a bit circular at some point. Is the player/team “average” (or below average) because he/they are inconsistent, or is the average ability result in inconsistent performances?
At least from a statical standpoint, there’s more science supporting the latter since it’s unreal to think that in any endeavor where success breaks down to marginal rates of return over a small sample that there won’t be some distribution around the mean. Since there’s a left wall of failure (unable to score/rebound, etc.) and pratcally a right wall of human performance –total domination, but limited to the measured 48 minutes in an NBA game, 40 in the college game–the variance will fall in this realm. The shape of a bounded distribution changes as it approaches either wall as well, perhaps providing the illusion that a bad team could play much better if only all their games looked like the longer tail of better performance that will show up stochasticaly.
I’ve heard that team x or player y is inconsistent (perhaps a result of having such terrible names as player y and team x) but actual *measures* of consistency exist and they aren’t regularly reported. The variance of some team’s stats would tell us more about whether the team is actually more inconsistent than others or if it’s just fluctuation (coupled with human psychology recognizing some games but ignoring others). This still doesn’t necessarily mean that if they were less inconsistent they’d be better, but suggests that perpahs the way to be less inconsistent is to improve the overall performance and the distribution will follow.
None of this takes away from watching the games. It’s still what happens on the court/field/track that produces these data.
Westy
March 26, 2007
Great post, Jason.
And the more I think about it, I think this is what people really think (but cringe in accepting as reality); for instance, you hear so often that ‘so-and-so’ looks better on paper, but “let’s see them settle it on the court.” Or the much-used, “That’s why they play the game.” People recognize that results reflect only 1 measure. If Team A should beat Team B 7/10 times, in a one game series you have a 30% chance of a surprise. That result obviously is greatly decreased with a 5 or 7 game series.
And all that is probably why so much effort is put forth in practicing, coaching, game management, scouting, etc. All that effort is directed at producing a result on the positive side of average. And just maybe, the right strategy and preparation will help you do just that.