This year, against all odds, the Cubs won the title, the Warriors blew a 3-1 lead in the finals, and the 15-1 Panthers lost in the Super Bowl. While everybody was going bananas about these upsets, nobody was talking about the fact that the winning teams were all 1-seeds, the top ranked team on their side of the bracket. (For those wondering, this year's Stanley Cup champion Penguins were a 2-seed.) Why was everybody so surprised?
That got me thinking -- in some sports, it seems like the highest-ranked teams always make it to the championship. In others, it seems like anything could happen. So I was curious to see in which sports are playoff seedings most predictive of the eventual champion?
I scraped playoff and seed data (from sports-reference) for the four "major" American sports: basketball, football, baseball, and hockey, going back to the 1980s. At first glance, it looks like top-seeded NBA teams have done quite well in the past, whereas the MLB and NHL look to be more of a crapshoot:
I can think of two main reasons why 1-seeds might do better is some sports than others. The first is the sport itself. Some sports are more random than others -- one lucky goal can win a hockey game, whereas one lucky basket doesn't sway the chances of winning as much. The second is the playoff structure. Sports have different rules that govern byes, home field advantage, strength of schedule, and other factors in the playoffs. For example, the top two seeds get a first-round bye in the NFL, the only league that awards byes in the playoffs.
Anyways, I was curious to look a little more in depth at how all seeds performed in the playoffs so I drew up some Sankey charts for each sport. Hover over the connecting lines to see the numbers.
Not all sports have the same sized brackets. For example, only four MLB teams from each league make the playoffs, while 8 make it from the NBA. One implication of this is that the top-ranked NBA teams have to play more games than the top-ranked MLB teams to win the championship. Because each additional game is a chance for an upset, we would expect that top-ranked teams will win less as playoff brackets get bigger. Controlling for the different playoff bracket sizes will give us a more clear understanding of which sports favor their top-ranked teams the most, rather than which bracket sizes are favorable.
To control for the different bracket sizes, I calculate a metric that I call Win Percentage Above Random (WPAR):
This tells us how often a given seed wins the championship compared to how often a random team would if they had to play the same number of games. So, for the NBA where there are four rounds, we give a random team a 50% chance of winning each game and a (.50)4, or 6.25%, chance of winning it all. Emperically, the top NBA seed wins 70% of the time, so their WPAR is 63.75%. This gives due credit to teams that have to play more games by making the "penalty" term in the WPAR equation smaller as the number of rounds increases. Here are the first charts you saw, but recalculated using WPAR instead of simple win percentage:
Under WPAR, top-seeded NHL teams out-perform top-seeded MLB teams, a switch from when we used simple win percentage!
I acknowledge that calculating win percentage above a 50/50 match might not be the correct counterfactual, but it gets us closer. Top NBA/NHL seeds, who have to play an extra game, win that game about 90% of the time instead of 50% as I assume in WPAR. This means WPAR gives NBA/NHL teams too much credit and over-penalizes NFL/MLB teams. Even so, NFL 1-seeds have the second best WPAR after NBA 1-seeds, and MLB and NHL fall somewhere thereafter.
In the end, it looks like seeding in baseball matters the least. This is interesting because MLB seeding (not to be confused with baseball seeding), which is based on an 162-game season should be more accurate than NFL seeding, which is based on a 16-game season. But it's not. And that's why you play the game.
* Okay, so actually 10 teams make it from the MLB, but that's a recent change. I use data from the 1994-2011 MLB seasons when 8 teams made the playoffs. More generally, the playoff systems for all of these sports are constantly changing, which makes this sort of analysis and data scraping particularly tricky.
