Actually, terminology here makes it a little more confusing, because you're actually talking about acceleration, not speed. Which car is faster? That depends...are we talking about acceleration, or top speed? Can Car B even go faster than 109 mph? This reflects my point. Sometimes doing some quick math to try and "normalize" one thing to another may be flawed, if in doing so you inadvertantly throw out the window a pretty relevant fact. In this example, that fact could be that Car B just doesn't go any faster. It may accelerate better, but it tops out at a lower number. In basketball, by normalizing out pace, you are getting rid of a characteristic that happens to be an important component of the variable you're trying to measure. Maybe team B simply isn't capable of performing at a pace that would allow it to hold its opponent's to fewer PPG?
I'm more than convinced that good offense leads to good defense, and vice-versa. All I'm doing is pointing out a flaw in the per possession metric. Doesn't mean that metric isn't still enormously useful....it just is something to consider. Regardless of the fact that the Rockets still lead the league on a per possesion basis defensively, Bill Walton saying the Rockets are only good defensively because they play at a slower pace makes absolutely no sense. The pace they choose is a factor in why they are good defensively. Of course the Rockets wouldn't be as good defensively if they played at the Suns pace. That's why they don't.
Good offense could lead to good defense, and vice versa, but not necessarily. Looking at per-game stats, there's actually a significant negative correlation ("good offense" tends to lead to "bad defense"). That's obviously nonsensical. The per-possession stats, on the other hand, indicate almost no correlation. As I've argued, that makes much more sense. Offense and defense are two different things and shouldn't be obscured if you want a reasonable analysis. But that's not a flaw. Consider it this way. We can safely say that the best teams generally have the best winning% (perhaps, if you like, you could adjust for strength of schedule). But is that "flawed" because a team's winning% might change if they played at a different pace? Does that make winning% a flawed statistic? Per possession stats do hide the pace a team plays at. But does knowing that pace change our evaluation of their defense (in terms of their defensive ranking, not simply their style of play)? If so, why? If not, then where's the flaw?
Durvasa, very nice work indeed! What did you use to calculate the off/poss, etc, ie. month by month averages or similar? Also, in trying to see whether off/poss or def/poss is a better winning strategy, it appears these correlation numbers seem to jump around quite a bit from season to season. Thoughts?
I used the season data from basketball-reference.com (e.g. 2006 season). I calculate total possessions using the following very accurate estimator: poss = FGA + 0.4*FTA + TO - 1.07*(OffReb%)*(FGA-FGM) The correlation numbers to win% jump around a bit. I don't have a good explanation for that. For 2002 to 2006 seasons, you have: offense: [ .765, .785, .557, .723, .587] defense: [.673, .575, .642, .658, .663] Correlation for offensive efficiency to winning% dips quite a bit from the 2003 season to 2004. That in itself doesn't necessarily mean that offense became less important. For instance, it could have to do with the extent to which good offensive teams were ALSO good defensive teams. For example, in 2003 the Mavericks were the best offensive team in the league, but they were also the 9th best team in defense. Thus, they ended up winning 60 games. In 2004, they were still the best offensive team (and by a wider margin, too), but their defense dipped all the way to 26th. Consequentially, they *only* won 52 games that year.
Well, I just don't know what to say anymore . (1) Terminology or not the mathematical principle is the same described in the car example. No smoke and mirrors going on here. (2) This isn't quick math to "throw stuff out". Normalization means you're comparing things based on an equal basis of measure. It's the basic tenet for any type of comparison in ANY field of knowledge that exists. If you don't measure using an equivalent metric / scale you are comparing apples and oranges. (3) Lastly we aren't "getting rid of pace" whatsoever. I've said it about what seems to be a dozen times above already. The reason it looks like it is because pace is now factored in as a rate of change and you can't see it anymore but it's embedded in the calculation. The idea is not to "get rid of pace" but to provide a basis for apples to apples comparison by measuring two things relative to the same scale of measure - in this case a hypothetical scale of pace. I'm done trying to explain. Got plenty of other things I can do. If you don't believe me go find someone who is knowledgeable in math & statistics . Cheers!
Yeah the Suns are an awesome defensive team. They just gave up 72 points in 1 half to the Grizzlies. The Grizzlies scored like 65 in a whole game against the Bulls.
about to post that myself. i know walton goes WAY OUT there sometimes (okay maybe all the time), but to say that phoenix plays better defense than houston is one of the most absurd comments made in a very long time. they gave up 72 points IN A SINGLE HALF!
Fascinating stuff, durvasa. The offensive/defensive correlations especially are some interesting stuff - interesting to note that offense seems to have a slightly higher correlation with winning than defense over the past 5 years, but not the case this year. One thing that looks interesting to me is the defensive correlation with winning seems to have a lower deviation of the correlation correlations. Of course it's only a 4.5 year look, but my guess is that would play out over history - empirically, offense has a tendency to be more volatile with guys having bad shooting nights whereas a defensive effort can be pretty consistent. Jay - I don't think control of pace is completely "factored out". A slower paced team that is sound defensively will be able to limit transition and fast break opportunities, which are generally going to have much higher scoring percentages.
Here's an application of the "per possession" concept. I'll use the same estimator of possessions mentioned above (it's not quite as accurate for a single game, but it still comes pretty close). In the first half, it looks like the Suns had about 53 possessions, and the Grizzlies had about 54 possessions. These teams usually get about 47 or 48 possessions in 24 minutes, but you'd expect them to get a lot more possessions when facing eachother (since neither team is looking to try the slow the other one down). So, the Suns scored 143.6 pts/100poss while the Grizzlies scored 131 pts/100poss. That's either great offense, or pathetic defense.
You don't normalize what you're measuring, though. no, you don't get rid of it at all. you normalize it. you know this. I know this. I just think it can hurt the analysis, as well as help it. Normalizing for possessions helps you understand how much each team, individually, allows on a per possession basis. But it CLEARLY isn't the whole picture, because not all teams allow the same # of possessions per game. If I told you team A allowed 100 points per possession, and team B allowed 98 points per possession, and both teams scored 99 points per possesion, I'd have no idea which team is better than the other. And I'd only have part of the picture on which team is better defensively. It's not that hard a concept. Per possession stats fail to describe a team's overall performance...that's painfully obvious. Or, to put it more succinctly, normalization refers to the division of multiple sets of data by a common variable in order to negate that variable's effect on the data, thus allowing underlying characteristics of the data sets to be compared. you haven't "gotten rid" of pace, but you have negated its effect by normalizing for it. Thanks. I'll just use my own statistical major/math background...though school was a long time ago, and statistics was only 1 of my 3 majors.
Actually NIKEstrad, I think an even closer estimator is using the pythagorean method that uses both offensive and defensive efficiency in the estimation of winning pct. The thread was about rating defenses so I stayed away from discussing it. Here's the equation. durvasa - if you get a chance I'm curious how it stacks up correlation wise. expected winning pct = (off eff ^14) / (off eff ^14) + (def eff ^14) Some use 16.5 instead of 14 as the exponential power. Also ppg scored and allowed can be substituted for the off & def eff. Whatever the case the expected winning % found on knickerblogger.net and on basketball-reference.com is based on this estimation (or some form of this since you can do even more sophisticated stuff with it to get better prediction).
I choose pathetic defense Keep up the interesting debate yall. I am learning a lot about how statistics can be calculated and evaluated.
Jayz: I'm going to agree to disagree . I can't find a better stat measure that actually takes into account pace,pts allowed, fg% allowed (embedded in the calc), ft's, to's and rb's and accounts for all of collectively any better than a per-possession stat. I agree that points allowed per game is a measure of defense. It's just not a good measure for making relative comparisons regarding effectiveness of defenses between teams though in my opinion.
You're right, it doesn't tell you the whole picture. It doesn't tell you how much they depend on three pointers, it doesn't tell you how often they get to the free throw line, it doesn't tell you the pace they usually play at, and innumerable other things. But suppose you knew, definitevely, that team X was a better defensive team than team Y. That doesn't tell you anything about those tendencies either. In your example, you SHOULD have a good idea of which team is better. Team A scores 99 per possession but allows 100. Team B scores 99 per possession but allows only 98. And remember, as we're defining "possession" both the team and its opponent will have roughly the same number of possessions. Based on that information, team B is likely better. Why? Well, point differential highly correlates with winning%, and team B is outscoring its opponents while team A isn't. Simple.
true, true also true, but then you would already know which team is better defensively. all likely true. my point is ONLY that ppg allowed per possession does NOT take into account PACE - instead, it normalizes it, and in doing so, negates its impact. From a purely defensive standpoint, using the stat to compare one team to another is probably the most useful. but it doesn't paint a complete picture....and that should be kept in mind.
The correlation between Expected Winning% (with 14 as the exponent), for the last 5 seasons, and winning% was 0.963. Same using 16.5 (well, 0.0003 more ). In fact, just looking at: Per possession Point Differential = off eff - def eff the correlation to winning% was 0.962.
Lets say team A allows 1 pt/poss. Then if their opponent has 80 possessions to score against team A, team A will give up 80 points. Do you agree with this? Lets say team B allows 2 points/poss. Then if their opponent has 80 possessions to score against team B team B will give up 160 points. Do you agree with this? Do you realize that team B will ONLY give up less pts than team A if they slow down their pace big time (ie allow less than half the possessions of team A)? So now lets say team B decides to slow down the pace. Would you then call them a better defensive team than team A simply because they made the game much slower? NO!!! However, team B is allowing less PPG than team A?! Here is a situation for you: Lets say its crunch time, game is TIED, 24 seconds to go, Game 7 NBA finals. Your opponent has the ball for the last shot. Would you rather be team A defending or team B. You know that team A gives up 1 pt/poss while team B gives up 2 pts/poss. Make the right decision JayZ!!!!! Your statistics diploma depends on it!
It is what it is -- a measure for how effective a team's defense is. It doesn't tell you the manner in which the team plays defense, obviously, but it does tell you how good that defense is. Everything that contributes to a working defense (limiting scoring from the field, limiting scores from the line, defensive rebounding, forcing turnovers) is encapsulated by it. All else being equal, if a team does something that makes their defense more effective overall, then that difference will show up somehow in the per possession metric. So, in that sense, it does paint a complete picture. But what it wouldn't tell you, of course, is the defensive strategy employed by the team.
