Testing the Ewing Theory
July 18, 2007 4 Comments
I have to be honest: I know as much about basketball as I know about nuclear physics. A little bit, but enough that if you put me in charge of either a nuclear reactor or a basketball team, there would be a catastrophe. Worse, I would swear to you that I knew what I was doing right up to the part where things started exploding.
But, then there’s the easy-to-understand Ewing Theory, coined by ESPN’s ever-delightful Bill Simmons and named after former Knicks’ center Patrick Ewing. Simmons credits this particular theory to a friend who noticed that whenever Ewing wasn’t playing, the team that he was on at the time seemed to actually play better. Odd for a man who was usually the undisputed “best player on the team” by far. After all, how could subtracting such a player make the team better? Still, Simmons put together a pretty impressive list of other times that the Ewing Theory seems to have worked in the past across a number of sports, most of them single games. After all, the ’88 Dodgers won the World Series, despite having lost Kirk Gibson to an injury (although I’m told that he did have one pinch at-bat).
There are plenty of things that people believe about sports (and life) that simply aren’t true. It makes no sense that taking away a team’s best player would do anything other than make them worse. If you cherry pick your examples, you can make a (false) argument that the theory is valid. And there will be plenty of examples from which to choose. After all, taking away the star player doesn’t reduce the team to complete impotence. Suppose that your team is involved in a big game, and right before the game, the star player decides to retire and take up competitive poker instead. Let’s say that your chances of winning the game were 60% beforehand, and now they are 40%. Well, that means that 4 times out of 10, the Ewing Theory “works”. All that’s left is to forget about the six times that you lost, and you have a nice happy illusion.
Or is it an illusion? The idea behind the Ewing Theory is that a basketball team (or any team) can become so dependent on one star, but his removal can have the effect of having other team members “step up.” It has a certain logic to it. In basketball, where five players must interact with each other to achieve their goal (put the ball in our basket, keep it out of theirs), I could actually see it working. If too much of the focus is on the star player (the “Ewing”, if you will), the other players may actually not get a chance to show off some of the talents that they have. Taking away the Ewing means that they get a chance to use them, and the talents which they are able to uncover are actually better, as a whole, than those of the team before the Ewing decided to take up poker. I have no idea how exactly that would work in basketball, but the one guy who never gets to shoot, even though he’s really good at it, might get a chance to show that he can. In baseball, it seems a little less obvious how this might work. Perhaps players get moved around the lineup and asked to do things (hit for power instead of contact) that they were really quite good at, but never had the chance to do.
The question of whether or not the Ewing Theory actually works in baseball is one that can be answered with a few pokes around the data. The key in investigating something like the Ewing theory is not to cherry pick examples. We need a very large data set, and we need to look at all games in which our “Ewings” were absent. In general, good players don’t get a lot of rest (although they may get hurt), so we might be looking at just a handful of games per Ewing to investigate. So, I’m going to be looking at the years 1980-2006 (thanks to the magic of Retrosheet game logs) to ensure that I have a big enough sample size. I eliminated 1981 because the season was dramatically shortened due to the strike.
Let’s identify some Ewings. First off, in baseball, the only way to really look at the Ewing Theory is with hitters. In baseball, an injured ace starter would really only affect a team in one out of five games (the ones that he was slated to pitch). But hitters play every day. The Ewing Theory also stipulates that a player must be a superstar and that the team is basically “his”. So, let’s isolate the hitters who are really good, say the top 30 hitters in baseball as ranked by OPS within the year in question (with a 400 AB minimum). This eliminates the Dmitri Youngs of the world who win the McDonalds’ Chef of the Year Award as the best hitters on bad teams (actually, Young was a Ewing on the 2003 Tigers!). Now, a good hitter who has two or three other superstars around him probably won’t be missed as much. But, if he’s the only superstar level hitter, he really qualifies as a Ewing. So, if a team has two players in the top 30 for a year, neither one counts as a Ewing for that year. To give you an idea of who’s left, in 2006, there were nine players who fit this definition of a Ewing: Aramis Ramirez, Carlos Guillen, Miguel Cabrera, Lance Berkman, Vladimir Guerrero, Frank Thomas, Ryan Howard, Jason Bay, and Albert Pujols. Really good hitters without someone else on their team to back them up. Overall, I came up with 275 Ewings over the 25 years under study.
For those curious, the most often “Ewingized” players (is that even a word?) over the course of the last few decades were Frank Thomas and Sammy Sosa (7 times each), followed by Mike Schmidt (6), and Carlos Delgado, Vlad Guerrero, Brian Giles, and Fred McGriff (each with 5).
Thanks to the magic of Retrosheet game logs, we have the starting lineup for every game played in baseball during the time in question, and the score. I assumed that if a player was not in the starting lineup, he did not play. That’s, of course, flawed, as he might have pinch hit, but it’s close enough for government work.
Once I’ve identified my Ewings, it’s easy enough to identify games in which they started and those in which they didn’t. I took the team’s overall W% when the Ewing started. Suppose that a team had a .600 win percentage with the Ewing in the lineup in whatever year I’m looking at. Then, let’s suppose that in the three games where he didn’t play, the team won two and lost one. For those two wins, that’s .4 wins above what we would expect from them. For the loss, they get a .6 win debit (or .6 losses, however you want to look at it.) If teams really do compensate in some way when losing a star player, we would expect that adding up all of these numbers over all teams and all years would produces a zero (if they break even) or a positive number if they somehow manage to do even better. If teams do actually suffer from missing their Ewing, the sum would be negative.
The rest is just crunching the numbers. So, what do the numbers say? Over the course of the last 26 years (excluding 1981), teams were 137 wins below what they would be expected to do when their Ewing was missing. Teams really do get worse when you take away their best player. Over the course of the seasons in questions, these teams combined for a record 19209 – 18883 with their Ewing in the lineup (a winning percentage of .504), and a 2680-3001 record without (.472).
Now, that’s not to say that there weren’t teams that didn’t get better after the loss of their star. The most extreme example was the 2004 Los Angeles California Angels of Anaheim California which is near Los Angeles. That year, the Angels were a .530 team in the 115 games where Vlad was in the starting lineup and a .660 team in the 47 games when he wasn’t. That’s a total of 6 extra wins without Vlad over what they would have been expected to win if they had just held steady to what they did with Vlad. The 1987 Brewers, on the other hand, really missed Paul Molitor (in the year of his 39 game hit streak… I was at game number 35 in the fourth row! Best seats I’ve ever had for a baseball game.) They were a .647 team when he was in the lineup and .348 when he wasn’t, for a total of 13 extra losses without Molitor. A few Ewings never gave their teams a chance to test the theory. Cal Ripken, in 1991, was in the middle of an MVP season, but… well, perhaps you’re familiar with his work attendance rates during that time period. In all, 99 teams got better without their Ewing, 3 performed the same, and 161 were worse off. At least in baseball, the Ewing Theory doesn’t work.
The magic of the Ewing theory is in the small sample size. Overall, teams lost about a three percent chance of winning by having their Ewing amputated, but still that’s a 47% chance of winning. Over the course of a season, a reduction of 3% in win chances is about 5 wins. But, if all you care about if one game, then you’ve got a sporting chance of having your theory “proved” right. It’s just that’s not the way to bet.