Another good read from Kevin Pelton of Basketball Prospectus. He reassesses where the "advanced stats" movement in basketball is and where it's heading. It appears to be somewhat a response to the recent Bill Simmons piece where he talks about stats and sort of calls out the advanced stats community, specifically referring to the stat geeks at the message board Kevin Pelton runs. http://basketballprospectus.com/article.php?articleid=618 [rquoter] I've been thinking a lot lately about where NBA statistical analysis is headed, starting when I read Michael Lewis' feature on Shane Battier, Daryl Morey and the Houston Rockets. The MIT Sloan Sports Analytics Conference earlier this month provided plenty of thought-provoking conversation. The topic then returned to my attention this week after reading SlamOnline.com's interview with Roland Beech of 82games.com as well as Bill Simmons' treatise on the state of NBA statistics in the latest ESPN the Magazine. Ultimately, the discussion seems to center largely on the two questions that were the focal point of my columns five years ago: * Can we boil a basketball player's value down to a single number? * What better describes a player's value: his individual statistics or plus-minus data describing his impact on the team? If you would have told me five years ago that neither question would yet be answered to anyone's satisfaction, I would have been disappointed and somewhat surprised. What happened? Simmons notes one key explanation in his column: Teams got interested in this stuff. Five years ago, no NBA team employed an analyst on a full-time basis. Oliver was attempting to get just such a position, a process that culminated in his hiring by the Seattle SuperSonics that fall. While full-time statistical analysts remain the exception rather than the rule, at least nine teams use at least one person, with Morey's Rockets employing a full roster. ... In the context of that progress, let's return to those two key questions. Can we boil a basketball player's value down to a single number? I would answer this question by saying both "no" and "yes." Let's start with the "no." As Simmons points out, one of the biggest complexities for statistical analysts in basketball is the importance of fit. In some of his rare public work in the Journal of Qualitative Analysis of Sports, Oliver used the game of frescoball to illustrate how powerful fit can be in a team sport. This is especially problematic for those coming from the world of baseball, where you can reasonably look at a player's statistics and draw a safe conclusion about what his addition or loss will mean to a team. There's a desire for the same kind of tidy analysis of NBA trades or free-agent additions, but things are simply too complicated, as most recently illustrated by Elton Brand flopping this year in Philadelphia after people like me touted his signing. Everyone involved agrees that it would be ideal to use a dynamic method of player valuation that takes into account a player's teammates and his role. Figuring out how to turn this into reality has proven much more challenging, and this remains an area with the potential for huge advances in our understanding. ... What better describes a player's value: his individual statistics or plus-minus data describing his impact on the team? More than any other issue, I think this question divides the APBRmetrics community. There are zealots who believe individual statistics are entirely meaningless and a player's value can only be divined through adjusted plus-minus, others who are totally dubious of the inherent noisiness of plus-minus measures and even more scattered between those two extremes. I've struck a typically moderate position. As I noted in my follow-up to The New York Times' Battier story, I think there are elements of the game--primarily at the defensive end--that at this point can only be quantified using plus-minus data. At the same time, I take adjusted plus-minus from a single season or even combining multiple seasons with a shaker's worth of salt. The question raised by both Beech and Simmons is whether tracking additional statistics outside the box score can achieve this same goal in a more reliable fashion. I certainly figured five years ago that game tracking would have made strides by now, but even getting data on a regular basis for something as simple as charges drawn has proven difficult. It would be nice to see the league take the initiative to begin counting some of these categories on its own, but to get detailed, freely-available information may require a volunteer effort along the lines of Project Scoresheet or the Game Charting Project run by our counterparts at Football Outsiders. The more numbers we have at our disposal, the better, but I do see some problems with the "count everything" philosophy. The first is the issue of subjectivity. If we get to the point of trying to measure picks or assess the quality of a player's help defense, that is very different from counting deflections or charges. With a large crew of volunteers, subjective measurements create any number of potential headaches. Valuing these new statistics is another issue. Take the popular "deflections." Is a deflection as good as a half a steal? A quarter? Three-quarters? These questions are nearly impossible to answer from a logical perspective. In fact, the best way to value them might be using adjusted plus-minus in the same way as Dan Rosenbaum did in creating statistical plus-minus, bringing the two schools together. Ultimately, I feel like there are areas of the game that are simply too subtle to be quantified at the individual level, like a player's positioning. If being at the right place at the right time forces an opponent to pass the ball out for a more difficult contested shot, that's virtually impossible to track. Where it does show up is at the team level, which is why I feel that even with more detailed statistics at our disposal there will still be value to looking at plus-minus numbers. As compared to five years ago, I'm less certain now where APBRmetrics is headed. In part, the statistical revolution, no matter the sport, is a generational issue. It's no surprise that in addition to Morey, the league's other young GMs--like Oklahoma City's Sam Presti and Portland's Kevin Pritchard--tend to be especially interested in synthesizing scouting and statistical information. Especially as the teams with analytics departments continue to enjoy success, the demand for analysts at the team level figures to grow steadily. That will challenge the outside community to keep up. If we're to make more rapid progress in answering the important questions, the statistical community can't get caught up in the minutiae of pitting rating systems against each other. Innovation and ingenuity is needed. I'm optimistic, but now cautiously so. [/rquoter]
Good read. I guess I am leaning more to the +/- camp. Ultimately, how "good" a player is, should be measured in terms of how much he helps his team win, not how "well" he perform individually. Somehow, they need to come up with some ways to use the collective individual trackings to adjust +/- numbers. Given enough manpower to do the tracking, and enough computing power to churn the data, getting the numbers shouldn't be too difficult. It's assigning meaning and value to the numbers that's challenging.
The problem is I don't see how +/- tell us how much the player helped the team win. At least not the simple +/- we see on the box scores. To me they tell me how the team did when the player was on the floor, but not how much they impacted the team while on the floor. Despite having a player having a positive +/- score, it might just be a case of the team winning in spite of him instead of winning because of him. If we want to see the impact they had, tracking individual stats seems much more useful to me.
Come up with a scoring system that can equate everything to 1 number. Then you take the individual numbers and add the +/- number.
Everybody knows the straight +/- number is pretty much useless. That's why I said they'd got to find a way to use individual numbers (and not just the boxscore stuff) to adjust, plus accounting for teammates and opponents etc.
Good read, thanks for posting. On a side note: I agree with Simmons. I wish there was some sort of "super assist" type of category since there's obviously a huge difference between a catch-and-shoot type assist and an assist that directly leads to a score, lay up or dunk.
Sorry can you elaborate? Are you saying something like: Brooks came off a screen from Scola to score a 3, Brooks gets +1.5, Scola gets +0.5, Batter gets +0.2, man guarding Brooks gets -1.5... etc?
I am not an expert. So I don't really know. The basic concept of +/- is to measure how your team does when you are on the court. But you have to know whether it's you or your teammates or your opponent that makes your team succeed/fail while you are there. So you have to know what you do too, such as how much you touch the ball, what you are doing on offense (spacing the floor, cutting, drawing double team, setting picks, etc.) and on defense (stopping your man, denying your man the ball, boxing out, help defense, taking charge, getting loose ball, etc.) Correlating these things with +/- together with your teammates' and your opponents' would bring a better picture. Like I said, tracking those things aren't difficult, given enough manpower and computing power. Figuring out correlating values is the hardest.
Like you said tracking the individual stats are easy to do, so why not just do that without trying to correlate +/- to the individual stats. The attractiveness of +/- to me is that it gives an objective number to how effective a particular line-up is, but if applying it to an individual is going to require making subjective correlations as to how much each player contributed on every single play I might as well just read a traditional scouting report. The downside of only tracking individual stats is that it will probably never give a nice single number determining the value of the player. But at least it paints an objective picture of how the individual is performing, unlike any application of +/- I can think of atm.
Because without the +/- component(s), you do not know whether if those individual stats really help your team, and if they do, how much. Tracking individual stats, whatever stats, is no difference from just looking at a number on the boxscore. It doesn't mean anything until you know how to translate that number into how much it contributes to outscoring the opponent. For example, you can track how many picks a player sets. But how much does that help the team? Only when you can correlate the individual stats with +/-, can you truly understand the value of a player.
I agree having a nice single +/- number to sum up the value of a player would be very nice, kinda like what win shares or VORP is to baseball. I just don't think it is possible for basketball. And in absence of that having a lot of individual stats tracked is better than nothing. Tracking individual stats without +/- may not "mean anything" in terms of determining the exact value of a player, but they can still be useful without doing that. For example when deciding to pull the trigger on a trade between two players who play at the same position, the GM doesn't need to determine the exact value of the players involved relative to the rest of the league, he only needs to decide which of the two is more valuable. If player A and B are very similar in their defensive and playmaking stat categories, but player B's stats show him to be the more efficient scorer overall, then stats can be used to make an argument on whether to pull the trigger or reject the trade. It won't always be so simple, and subjective decisions will have to be made by the GM in the end, but that doesn't make the stats without +/- meaningless.
You have a valid point. But it can only apply to comparing similar players. You cannot compare players that play a very different game. How do you compare the value of a defense only player like Chuck Hayes and a shoot only guy like Steve Novak? How much does each of these players contribute to winning? And in the case of MVP discussion, how do you know if Chris Paul is more valuable than Dwight Howard when their contributions to their respective teams are totally different?
^ I agree there will be many cases where making a comparison won't be easy. I don't think stats in basketball will ever get to the point where we point to a single stat and say, "Chris Paul is the MVP" without watching a single game, like you arguably can for baseball. (From what I've heard from Morey's interviews, I get the feeling he feels this way too?) Still wouldn't you rather have more stats tracked outside of the traditional boxscore to challenge/confirm the opinions you gained of a player from watching him play? To summarise, I think our differences in opinion are: - I don't think a good +/- system ranking individuals will ever be able to be developed in the future, you do - I have lower expectations of what stats will be able to do in basketball than you
Actually, I don't. Or to be more precise, I am not sure. What I've been saying is a response to the article where it says there are generally two camps, the +/- camp and the individual stats camp. Philosophically, I am more with the +/- camp in terms of how to define "good" in basketball. But you have to correlate the basic +/- concept with a lot of data on what exactly the player does on court (individual stats), rather than just blind +/- stuff. Fine. I think even if there is no such thing as a single overall stat, having several different objective measurements is always better than "just watch the game."
