Rockets fans and commentators are rightly excited about the upcoming season, but maybe – and you don’t hear this often – not as excited as they should be. Most of the talk has been around whether the Rockets can sneak into the third or fourth spot in the West, thus earning home court advantage in the first round of the playoffs. But if the predictions from one of my favorite stat geeks, Arturo Galletti, are to be trusted, then brace yourself for some much loftier questions…

- Will anyone in the West come close to challenging the Rockets for the top seed?
- Will the Rockets be the best team in the NBA? Better even than the LeBrons?
- Will the Rockets be the best team in recent NBA memory?

And, most uncomfortably, …

*Will the Rockets challenge the NBA record of 72 wins set by Michael Jordan’s 1995-96 Bulls?*

Gasp. I went there. Or more accurately, Arturo’s statistics went there. And his statistics are as good of an objective measure as you are going to find. He uses David Berri’s WP48 measure, which does some complicated manipulations of simple box score statistics to produce a single measure of player efficiency. WP48 is both predictive and stable across time, thus satisfying two crucial performance-metric criteria which many of its competitors do not. Arturo adjusts the WP48 metrics for the players’ age and expected minutes per game, and then sums these adjusted estimates across all players to get win predictions for each team. (I’m simplifying a bit, but that’s the gist.) The table below shows the resulting predictions. (*Note that these are updated predictions that Arturo shared with me over email, so they are slightly different than the ones at the link above.*)

How to interpret this: Even if the Rockets have a “bad” season, they will still likely finish in the top four or five in the NBA. If, on the other hand, they have a very good season, they may win as many games as any team in NBA history.

I will pause to give you time to gulp.

When such brazen predictions are laid before us, the natural reaction is to hunt for holes to poke. That’s why David Berri has a 6,400-word FAQ page to address all your hole-poking queries. It turns out WP48 can be accused of many things, but being poorly thought out is not one of them. Nevertheless, the first maxim of statistics is that statistics never tell the whole story, and I can think of several factors that might be skewing the Rockets’ statistics. But brace for more gulping because the factors suggest that the predictions might be *underestimating* the Rockets.

- Factor 1. The WP48 measure, like nearly any available measure, does not adequately account for a player’s defense. +3 for the Rockets because they have the most important defensive position manned by a three-time Defensive POY and by Omer Asik, whose defense I do not have enough superlatives to describe. (In my nerdier fantasies I envision discovering a way to measure Omer’s off-ball defense.)
- Factor 2. The predictions are sensitive to the allocation of minutes, and it’s very hard to predict how that allocation will shake out due to injuries and player dynamics and baffling coaching decisions. +2 for the Rockets because Arutro’s model assumes that Omer will be limited to 18 minutes a game when in fact if the towers can function as twins, he may be seeing closer to 30 min/g.
- Factor 3. The predictions treat players as independent parts, but the first lesson of team sports is that the whole can sometimes be much greater or much less than the sum of its parts. +1 for the Rockets because smart money says that the pairing of Harden and Howard – historically two of the league’s most efficient players, particularly in the pick and roll – will yield even greater individual efficiency, at least if Howard is willing to engage in heaping helpings of pick n’ roll.

Let me reiterate: Arturo’s WP48-based predictions suggest that this year’s Rockets may be historically great and my logic suggests that his model may be underestimating them. I feel enormously uneasy about this whole thing. They can’t really be *that* good, can they? No team without Michael Jordan can be *that* good. So then where/how is the model overvaluing the Rockets? Arturo graciously provides the data and assumptions for all to see, so if you’re feeling uneasy like I am, go snoop around for yourself. He also shared with me these detailed projections:

———

**Postscript** — Arturo had these interesting comments after reading a draft of this post:

I have Asik in a sixth man role similar to what he did in Chicago and the role Gortat played with Dwight in Orlando. Given the amount of perimeter shooting on this Rockets team, it should be dead easy to play D12 in the post with Asik just outside the post ready for putbacks and rebounds, two shooters in the corners and a shooter/slasher at the top of the key. If I can figure this out, Daryl’s guys most likely have too.

The Rockets have done an extremely good job of building a deep roster of good young players. Even with injury, the second and third tier guys are all serviceable NBA players. This is a team built for the grind that is the regular season.

The best case scenario for the Rockets is the return of pre back surgery Dwight. If that happens, they will destroy people. Keep in mind that the Magic won 52, 59, 59, and 52 games in their last four seasons with a healthy Dwight.

View this discussion from the forum.

{ 0 comments… add one }

Login to leave a comment.You will be prompted to login with your forum username. Please login and return to this page to leave a comment.

Total comments:24NorEastern is referring to TrueHoop’s Stat Geek Smackdown, which asks some of the most well-known/respected gurus to predict the outcome of each playoff series. Their performance is compared to “The Crowd” (ESPN’s panel of 100+ experts), to Henry Abbott’s mom, and to each other. I love the Stat Geek Smackdown, but there are two problems with tying it to this post in an attempt to evaluate Arturo’s model: (1) Unlike what NorEastern says, the Smackdown shows no real gold standard of which models are most accurate. Performance varies greatly from year to year. (It’s that pesky random variation again.) For example, the WoW model did no better than Henry’s mom last year, but it holds the all-time Smackdown points record with its performance in 2009. (2) Preseason predictions and postseason predictions are different animals, for many reasons which I won’t get into here.

Most importantly, the other models do not publicly offer preseason predictions (as far as I’m aware). Or if they do, they certainly don’t offer the data, math, and assumptions for all to see. That’s why I like Arturo’s model: it’s as transparent as can be, and it has been polished through the flames of much public criticism.

The math that WoW uses (Wages of Wins) is not complicated. Actually it is trivial. For instance there is no calculation of something like linear regression involved. But the basic algebra is long. Seriously long. There are more adjustments, added data calculations and rerigs than I can count. How successful has it been over the years. The best prediction for team standings in the playoffs and beyond comes from crowd sourcing. Then Vegas and Hollinger. Halfway down the list is WoW.

I use WoW as an input. But being a Rocket fan I cannot fault this prediction. I think it is perfect. Homerism.

That's interesting. At first it seems to underestimate the better teams while overestimating the bottom feeders, but the longer I looked at it the less severe that seems to be. It should be noted that the only teams not to make the playoffs that outperformed the Arturo prediction were Dallas, Charlotte, and Detroit.

Love the cartoon by the way. :)

P.S. - Here is a plot showing how last year's predictions panned out.

Well said, Sir Thursday! You capture, in really precise and cogent terms, my sentiments.

Justin: I'd love to hear your thoughts in reply to Sir Thursday. Not in the interest of stirring controversy, but out of genuine curiosity about the usefulness of these particular predictors.

Alright, I feel like I need to have a go at providing a grain of salt.

I've never been sold on the Wages of Wins results. While their approach is aesthetically pleasing (I like the elegance of putting exact values on all of the various elements of the box score via regression), I just don't see how it's possible to make the assumption that all events of the same type that occur across an entire season are of equal value and expect to end up with an especially accurate result. The regressions they did show decent accuracy across a league-wide sample and will therefore turnout to be pretty good predictors on average. However, as soon as you start considering subsets of the data that are not representative samples of the set as a whole (eg. all players from a specific team rather than a mix from all of them), it doesn't seem quite so likely to follow that the model will hold.

There are a variety of factors that warp the values of events. As Johnny Rocket points out, in a close game a point is more valuable than in a blowout. In a game with a high number of possessions, a turnover is less of a negative than in a game with a small number of possessions. A block that keeps the ball in play is more valuable than one that spikes the ball out of bounds. A steal that leads to a fast break is more valuable than one that requires you to set up the offense. And so on. WoW assumes that the nature of these events averages out to give an overall value, but in fact they will probably differ from player to player and team to team.

The Rockets are probably overestimated by a number of the averages in the WoW model. There's the blowouts vs. close games point. There's the reduced marginal value of statistics in a fast-paced game that isn't totally accounted for (I believe the Team Defensive adjustment is an attempt at this, but the further away you get from the league average pace the less accurate it's going to be). Dwight Howard has gaudy block totals but has a tendency to hit them out of bounds (although he seems to be improving in that department). Galetti's age model (which I quite like, despite my nitpicking) also falls under the category of something that will vary from team to team. It assumes that players' growth curves will be broadly similar over time. But of course, a huge part of how a player improves comes down to team environment. Will a team full of young players (as Houston was last year) expect to see the same growth curves as a team with a wide distribution of ages? It seems unlikely to me. My guess is that the Rockets are probably overrated by that to a degree as well.

Furthermore, treating a team as the sum of its parts is all well and good (and I understand the reasons for doing it), but it means that if there is a 'team chemistry' effect then it's not going to be present in the model. The Rockets will be playing a very different style this year to how they did last season due to the presence of Howard. It will take time for that cohesion and systemic familiarity to sink in, and that will likely be borne out by the team getting lower win totals than the sum of its parts would suggest. I disagree with Justin's optimism on that side of things, anyway.

As such I'd strongly expect that the Rockets are on the lower side of that projection rather than the good one. Even with that, it's still looking like it's going to be a really successful season...just not a record-breaking one.

ST

Justin,

Care to do a comparison of Harden and Howard to 2006 Dwade and Shaq? I think at best, Harden and Howard can match those two in 2006. Its a worthwhile analysis because the pairs play the same positions, 2 and 5, and 2006 is relatively modern/recent

Ha, so true. I have reached the same conclusion when analyzing point margins in relation to winning %.

I'd love to see what last year's lakers were predicted to do. Things like this are fun and are great for all kinds of stuff (Like video game simulations), but it is ultimately fiction. Just like nobody thinks a roulette wheel will hit the same number three times in a row, hit black 12 spins in a row, etc.--there are just too many anomalies to calculate in an NBA season that statistics ignore. I know this is not a popular opinion (stats are like gospel in these parts), but these predictions will be interesting to re-visit at season's end. (somebody should remember to do that B) )

How did Arturo's predictions, with the same metrics, pan out on teams from last season and the season before? What is the precedent set for the accuracy of such predictions?

Myself, also being an Arturo from Italian descent, would have a tendecy to agree 100% with his predictions. Unless proven otherwise (of course).

55 to 59 wins for the young Rox.

follow @HoopsReportCard

Johnny: I don't have the statistical chops to control for pace, but here's a couple of related posts: while this guyobserves the strong correlation in general, he notices some interesting recent outliers from OKC and MIA. And this piece from 2011observed that age, pace, and defensive effectiveness all had some effect (at some point, though, you're really just saying winning basketball games is an effective predictor of winning basketball games).

I agree

Very interesting take, and good explanation. I hope its a "best" case scenario we're looking at at the end of the season.

Great post! Let me try to articulate why the Rockets probably won't win 70 games. My first point is notice the discrepancy between the number of game the Rockets were PREDICTED to win last year (50) and the games they actually won (45). That's pretty sizeable, and my guess is that it has to do with the distribution of Rockets points. Win they played well, they really crushed teams (I'm thinking of the huge win at Utah), thus inflating the team's advanced stats, but then they lost a number of close games. It might be called the blitzkrieg effect. When the Rockets are running and gunning to their heart's content, the other team becomes demoralized, players foul out, and things snowball out of control. But against better teams in which the Rockets have to engaged in the basketball equivalent of trench warfare, things don't work out so well, especially for a young team.

Now I realize that the advance-stats community (per Hollinger) believes that margin of victory is a big deal, and I agree. But what I'm saying is that for teams that play a certain really fast style that could produce huge blowouts, the margin of victory is less effective predictor. The closest analogy I have is Oregon in college football. Thanks to their offensive system, they have produced some stunning blowouts this year, but would you really that much better than Alabama or LSU? So I think that the Rockets will be very, very good this year, but not stunningly great, though I'm willing to be surprised!

That would be awesome for us to just hit our nominal mark of 64 wins, but I find it hard pressed for us to get to the 70 win plateau in year 1 of the H

^{2 }era.I'm surprised by alot of team's nominal win totals (DET, ATL, BOS, GSW, and IND just to name a few) but I find it most interesting to see Miami's good season would only net them 53 wins after winning 66 last year. Arturo has both Atlanta and Detroit beating them and Cleveland is nipping at their heels. I guess that is where age comes into it, but in that case, how are the Spurs going to win 65 games in a good season?

Perhaps the most depressing thing from that graphic would be if the Lakers truly ended up the second worst team in the league. You can call me a conspiracy theorist if you want but if the Lakers end up anywhere near the bottom five teams, I'll go to Vegas and put money on them landing the #1 pick in the draft, and that would be a travesty.

even I'mnot going out on a limb and say we will win 70 games :).............but shooting for 58-62 is more like it. I believe some have let all the negative press following Howardaffect how they view D-12 as a basketball player. I am convinced he will return to his Orlando form and him and Harden will wreck havoc in the league. Yes we have some issues to resolve, but we have problems other teams only wish they had. we have so many options we need to figure out which one works best. that's aluxury. we are loaded with tradable assets to repair any holes and we are young and healthy with durable playersto top it off. Yes we are a powerhouse. to me it just a matter of how soon we can figure our rotation out, which I don't expect to take that long :P

Amazing article....it's not really about poking the holes, it just seems....unreal. Last year we were surprisingly good, but I'm not entirely sold yet that we can win more than 60 games this season (actually, my range goes from 55 to 60, likely falling in 57-25 record just looking at the schedule). All in all, I just don't think that having DH12, and by demoting Asik to backup role, having 48 minutes of elite rim-protection, will net us 15 more wins. Maybe McHale comes up with and effective gameplan other than "run the whole game and sucumb to Harden going Iso at the end of close games", maybe that will net us a couple more, but I think that we are one move away from getting over 60...

But I'd love to be completely wrong (I do trust in numbers, they just seem too good to be true).....