I didn't watch any of Roger Federer's matches in his grass tune-up at Halle, but there was a general consensus from people who did that his age was showing and he probably isn't a favorite at Wimbledon. This stands in pretty stark contrast to AB, which only saw upside and a mini-surge in his grass-adjusted rank. If I had to pick who's more correct between the eyeballs and the computer, I'm inclined to believe the eyeballs for one simple reason: AB doesn't make any explicit adjustments for age-based decline.
Developing an aging curve for tennis player performance and coming up with rules for anticipated decline would be an interesting project. There are a bunch of ways to go about it, but it makes sense wherever possible to not reinvent the wheel if good approaches have been done before. Nate Silver's PECOTA and Kevin Pelton's SCHOENE are as good as any a place to start.
Both bring up a lot of parallels that are worth asking. Here are a couple off the top of my head:
- In basketball, different positions experience different rates of decline. Point guards who rely on a quick first step drop off faster than sharpshooters whose jump shots are usually the last thing to go. Is there a corollary in tennis, where different playing styles drop off at different rates? Would a player's serve drop off slower than their foot speed, meaning players like Andy Roddick and Ivo Karlovic might stay at a high form for longer?
- PECOTA does this particularly well, creating similarity scores between players to see how different types of baseball players decline. Baseball has an advantage in that it has a ton of metrics it can use to group players. Advanced Baseline only has one -- the player's rankings. What would be some meaningful metrics to start isolating a tennis player's playing style and separate them into groups?
- Does aging cause a decline equally across all surfaces? I would think the more rally-prone surfaces like clay would be more grueling at an older age.
- Basketball players can somewhat mitigate their production drop-off by reducing their minutes. We already see this with someone like Federer, who plays as few tournaments as possible these days. How much of a decline can tennis players stave off from limiting their schedule? In other words, what contributes more to their decline, age or mileage?
- How would the women's aging curves look different than the men's? There are even fewer comparisons to work with using other sports, since not only are there very few women's professional sports leagues, but also virtually none that have produced original findings on aging curves.
I would love to generate year-to-year rankings charts for all players to start getting some answers, but I don't have nearly enough historical data to reach reliable conclusions. Baseball and basketball have a timeframe of roughly 40 years to work with; AB only has about eight. Maybe I'll figure out a better method to get around the small data set, but in the meantime I'm probably stuck with Federer's chances being overrated this year.
One of the most interesting differences between the men's and the women's AB rankings is one I still can't explain all that well: predictions for women's matches tend to be more accurate than men's. (I already regret writing that a little, lest the same contingent that writes pieces like this will now accuse the women's game of being "too predictable.") The best I've been able to come up with so far is a higher correlation between scoring margin and future success, but I still don't know why scoring margin correlates better.
A couple of people have suggested serve dominance is less prominent in the women's game, which might explain the better correlation. This is an especially interesting question at Wimbledon, where serve dominance is at a premium. The implication is that scoring margin is good at capturing the rally component of tennis but not the serve component. While there is evidence that serve dominance is in fact less prominent in women's tennis, the connection to scoring margin deserves a little more thinking through. If you're exceptionally good at serving, don't you still win more games because you're harder to break? And wouldn't that be captured in scoring margin?
It sounds similar to saying scoring margin is flawed in basketball because it doesn't capture how good a team is at three-point shooting. The whole point of scoring margin is it coarsely captures the totality of a player's skill level, regardless of which particular components contribute the most.
That said, I fully believe there are players whose skill levels are not accurately captured by Advanced Baseline. Every ranking system has outliers that don't nicely fit the rules -- 2006 Gonzaga and 2009 Oklahoma are good examples for scoring margin outliers in college basketball. If serve dominance is indeed not captured well by AB, players with dominant serves (Serena Williams being the prime example) are good potential candidates for these kinds of outliers. That leads to an even scarier question for Wimbledon this year: As highly as AB rates Serena, is it possible she's still somehow underrated?