Ben Gulker is a Program Director for a statewide AmeriCorps program in his home state of Michigan, where he was born and raised. He’s a lifelong Pistons fan that grew up during the Bad Boy era, suffered through the “Teal” era, and spends his current fan time reading up on top 10 draft picks, thanks to the recent demise of his beloved team. If you ever want to talk Pistons, you can find him among other Piston die-hards over at Detroit Bad Boys.
For a large portion of the last decade, the Detroit Pistons were consistently among the Eastern Conference elite. The “Going to Work” Pistons, as they were known among Pistons faithful, were defined by lock-down defense, tenacious rebounding, and slow-paced but efficient offense. As fans, we took pride in this approach to the game; the blue-collar ethos of the Pistons resonated with the Michigan populous.
However, the current Pistons squad consists of a markedly different pool of talent that has produced markedly different results. Regular readers of this forum will know that a great deal of prior Piston success can be linked directly to three players: Ben Wallace, Chauncey Billups, and Antonio McDyess. Although Ben Wallace returned to Detroit and led the team in Wins Produced in 2009-2010 (something some of us Pistons fans like to call the “Benaissance”), the holes left by McDyess, Billups and others have not been filled by equally productive players.
As a result, the once mighty Pistons won only 27 games this past season and currently hold the #7 pick as a result of the Draft Lottery.
To make matters worse, there does not appear to be an immediate solution to the team’s problems. Joe Dumars signed Ben Gordon and Charlie Villanueva to large, lengthy contracts last summer, investing approximately $95 million in these two players combined through 2013-2014. Both players have been average at best to this point in their careers, and both made unimpressive debuts in Detroit. Additionally, Richard Hamilton – whose best basketball appears to be behind him — is owed approximately $25 million over the next two seasons ($12.5 million more is partially guaranteed in 2012-2013).
In the summer of 2011, the Pistons will need to make a decision on another of their “important” players – Rodney Stuckey. It has become clear that Joe Dumars and the Pistons think highly of Stuckey. In November of 2008, Chauncey Billups was traded, and Rodney Stuckey was handed the reigns of the franchise. This trade was made for a variety of reasons, but Dumars’ high opinion of Stuckey and his talents is indisputably one of them. In fact, Keith Langlois, editor of http://www.Pistons.com, has intimated that Stuckey is actually “at the center of the next phase in Pistons history.”
More recently, Langlois compared Stuckey to the remarkably productive Rajon Rondo, arguing that Stuckey’s 2009-2010 performance “isn’t much behind where Rondo was a year ago (2008-2009).” Presumably on the cusp of a breakout season, Dumars has a “tough call” to make concerning Stuckey: give him a Rondo-like extension now (5 years, $55 million)? Or, wait and let Stuckey become a free agent in 2011 and be forced to match a potentially larger offer sheet then?
In my opinion, if Stuckey is on the brink of becoming as productive as Rondo, the ink on that extension can’t dry fast enough. Point guards as productive as Rondo are rare and can alter the course of a struggling franchise for years to come. However, the story told by Wins Produced — told the two tables below — suggests that Stuckey simply isn’t as productive as Rondo and is very unlikely to ever become so. Therefore, signing Stuckey to a similar contract would be a significant mistake.
In 2008-2009, Rondo was a phenomenally productive player on one of the top-performing teams in the NBA, producing 17.8 wins out of team’s 60.9 Wins Produced. Again in 2009-2010, Rondo was exceptional, producing 17.9 out of 50.5 total Wins Produced. Although many perceive this season as a “breakout” season for Rondo – especially in light of his recent heroics in the playoffs – his performance is remarkably similar to what we saw in 2008-09. In short, Rondo has been consistently exceptional.
So, how does Stuckey measure up?
In 2009-10, Stuckey logged 2,499 minutes – and according to the Wins Produced numbers provided by Andres Alvarez – produced 2.6 wins. The position algorithm Alvarez employs, though, allocates Stuckey (one of the bigger guards on Detroit’s roster) to shooting guard. Stuckey, though, is primarily a point guard. And when he is compared to point guards he only produced about 1.6 wins. Either way you look at it, what Stuckey did this past season is a far cry from what Rondo did each of the past two seasons. Furthermore, this level of production should not have been surprising given Stuckey’s previous performance. Last season, for example, Stuckey produced merely 4.0 wins for the Pistons while posting .077 WP48 [Wins Produced per 48 minutes]. In short, Stuckey has been consistently below average (an average player posts a WP48 of 0.100).
Pistons fans — such as myself — are left asking the question, “What is it that Dumars sees in Rodney Stuckey?” Unfortunately, I suspect the answer is all too familiar. If we compare the box score statistics that each of these players has generated throughout their careers, we can see that Rondo grabs more rebounds, dishes out more assists, gets more steals, and scores more efficiently than Stuckey. But, Stuckey scores more points; and this season, Stuckey averaged over 16 points per game. As a result, it would appear that Stuckey can anticipate a hefty payday.
If that payday comes from the Pistons, and if it is similar to what Boston offered Rondo, Pistons fans are likely to be disappointed. The reason? Detroit would have just over $40 million per season committed to Ben Gordon, Charlie Villanueva, Richard Hamilton, and Rodney Stuckey. This quartet combined to produce fewer than five wins this past season.
Stuckey is still young, and it’s still possible that he could improve. But he isn’t that young, and he has quite a way to go before he achieves average production. It also seems likely that at least one of those other players will be traded in the near feature, freeing resources for more productive players. So maybe we won’t be lottery-bound for too much longer.
But regardless of what other moves the Pistons do make, the “tough call,” as Langlois describes it, isn’t that tough at all. Stuckey simply isn’t worth a Rondo-like contract because Stuckey simply can’t do what Rondo does.
Ironically, in June of 2009 Adrian Wojnarowski reported that Boston offered Rajon Rondo and Ray Allen to Detroit for Tayshaun Prince, Richard Hamilton, and Rodney Stuckey – a deal that Detroit “immediately rejected.” Apparently, Danny Ainge came to his senses and made the right decision. As a Pistons fan, I can only hope Dumars can do the same.
– Ben Gulker
The WoW Journal Comments Policy
brgulker
May 30, 2010
Hanks for the opportunity, Dr. Berri. It’s a blast to participate.
brgulker
May 30, 2010
*Thanks :)
Arturo
May 30, 2010
Ben,
Good post. As for Stuckey, here are his ranks among all guards for last season:
WINS 56
WP48 90
NBAEFF 26
NBA48 45
And he’s 23. There’s no way he’s worth more than 5 million a year (2.6 wins * 1.7 million per win). Dumars should try for a sign and trade.
brgulker
May 30, 2010
Arturo, I agree 100%.
Tommy_Grand
May 30, 2010
Good article
Jimbo
May 30, 2010
Awesome article. What has Dumars done? For what its worth I don think Ainge would have ever traded Rondo and Ray for those three guys. He must read this site, right?
Can’t wait to see a finals prediction based on playoff WP48 and PAWS
Joe
May 31, 2010
You don’t need stats to understand that Rondo is better than Stuckey and that Joe Dumar has not been making good decisions as of late.
Your whole post was a waste of space, expressing in an essay what could be expressed in a sentence.
Dre
May 31, 2010
Joe,
More than a little harsh man and definitely uncalled for. Last summer it was a big possibility of Rondo being traded. Stuckey is a big reason Dumars traded Chauncey (Thanks Joe!). So clearly the idea of Rondo > Stuckey is not obvious to all people.
Bgulker,
Awesome post! There may be some hope for your Pistons. The Nuggets have had 30-40 million invested in overpaid contracts the last couple of years and at least been perpetual playoff members. Also, Detroit’s come back from losing Rodman and Lambier, you may just have to wait a while :)
Charles
May 31, 2010
The trade was rejected in large part because Rondo’s contract would expire at the end of this year and there is little chance he would have re-signed with the Pistons. The Pistons would have about 27 million in cap space, but a terrible team unlikely to attract the top free agents. They’d end up overpaying for guys like Joe Johnson and David Lee.
ilikeflowers
May 31, 2010
Dre,
That is Joe the Troll. If it’s a number, he hates it.
Dre
May 31, 2010
Ilikeflowers,
Joe the number hating troll eh? Then he will definitely hate these numbers. I have done some advanced analysis for teams and found a set of “candidates” based on a list of numbers. Joe should hate these, they may be cursed.
Chris Bosh 4
Carlos Arroyo 8
Carmelo Anthony 15
Peja Stojakovic 16
Lebron James 23
David Lee 42
Joe
May 31, 2010
I don’t hate all numbers; I just laugh at the ones that say Andrew Bynum, adjusted per-minute, has played better than Kobe Bryant in these playoffs, the ones that say Jamario Moon, adjusted per-minute, played better than Kobe Bryant in the regular season, and the ones that seem to command utterly no respect outside this blog.
ilikeflowers
May 31, 2010
Joe the Troll,
What’s that? A response!?
Given your past claims that wp48 gets no respect from the blogs that you frequent, why are you here?
Regardless, if one can believe anything from a troll, apparently Joe only hates ‘disrespected’ numbers and/or those that disagree with his teenie-tiny-sample set non-predictive eyeball model.
You don’t pass the laugh test Joe the Troll. I believe there’s a letter around here for you somewhere.
robbieomalley
May 31, 2010
Gil Meriken
May 31, 2010
ilikeflowers – what are the supposed predictive powers of Wins produced?
Have you taken the time to compare how the season wins predictions using WP have fared versus a Las Vegas sportbooks’ season win lines?
I know there are various methods to allocate minutes and such, but have any of those methods favored well against any established sportbooks’ odds?
I’m only using a sportsbook as a point of relative comparison. Instead of saying “look how close the predictions come to what actually happened”, I want to see if the predictions are better than the oddsmakers’.
ilikeflowers
May 31, 2010
Gil,
what are the supposed predictive powers of Wins produced?
Stability (i.e. prediction) of per minute production from year to year regardless of team. Accuracy of calculating finished season win loss records from last season’s wp48.
Have you taken the time to compare how the season wins predictions using WP have fared versus a Las Vegas sportbooks’ season win lines?
No. I have no information on any player minute prediction models which would be necessary for any such comparison. An accurate minutes prediction model would be very nice to have.
I’m only using a sportsbook as a point of relative comparison. Instead of saying “look how close the predictions come to what actually happened”, I want to see if the predictions are better than the oddsmakers’.
Aren’t the oddsmakers’ lines trying to accomplish something different from just predicting who’s going to win? Aren’t they trying to maximize their take based upon who people think the winner is going to be vs who they think the winner is going to be? I’m not a gambler so I can’t really address this, but I imagine that the oddsmakers incorporate efficiency differential and a host of other predictive factors as well.
Without a good minute prediction model, wp48 is really more of a strategic tool for determining who a team should sign and who should get the most minutes.
Gil Meriken
May 31, 2010
@ilikeflowers
Regardless of the oddsmakers’ motives, it’s something convenient to compare against, since the oddsmakers establish their lines before the season begins.
I believe that in 2008 there was a post or site that gave WP season win predictions using three different minute allocation methods. As a fortunate coincidence, this was posted just around the same time that another gambling site posted their line for season win totals. So presumably both WP and the oddmakers had access to the same information regarding the teams.
Which one do you think out of all four (3WP, one oddsmaker) came closest to the actual win totals in aggregate? Another way to look at it is, could you make money using WPs predictions against the oddsmaker’s line?
Dre
May 31, 2010
Gil,
Enjoying the new direction of the thread. Ok, so here’s a common thought the WoW journal puts out about Models. The first is simplicity and the second is what it says.
In terms of estimating team performance using differential and home court advantage is a pretty good model (Dr. Berri mentions using this in his Playoff challenge that he won last year). The problem is that it doesn’t tell us why a team is doing well.
Wp48 does a good job of estimating performance of teams, while also telling us why (looking at player performance). That said, I would probably use a simpler model to bet on team performance than looking at all players across all teams. However, as a GM looking to improve my team, I would use WP48 at least over any other models I know of.
ilikeflowers
May 31, 2010
Gil,
I would hope that ‘the oddsmakers’ would be best at predicting team wins in aggregate since they are humans using models that are designed for that purpose. But I don’t know which of the hybrid models that you mention were best for any specific year.
If you can find an accurate enough player minutes prediction model you might be able to selectively make money, especially when players change teams or anywhere else that a league aggregate method is likely to have specific weaknesses. If ‘the oddsmakers’ have a superior per minute model coupled with a superior likely minutes model then you’re out of luck, or at least you’ll need to find a better/different model.
Sam Cohen
May 31, 2010
Dre- I enjoyed your list of “candidates.” Not sure that everyone got the joke, but I definitely had a nice chuckle…
Gil Meriken
May 31, 2010
@ilikeflowers – maybe the bookies are using WP as their basis!
But if not, shouldn’t the predictive power of WP (and a good allocation of minutes) be superior to those posted by another other method over the long run?
If WP is so solid, shouldn’t it be a better method at prognosticating the total wins of a team than any other given the same access to information?
@Dre – I thought that was one of the strengths of WP, is that you could determine theoretically how many wins a team should have, especially accounting for roster additions and deletions (given a good allocation of minutes method).
ilikeflowers
May 31, 2010
Gil,
Not necessarily. An example might be a model based on efficiency differential and team lineup stability. This might work better as a simple win/loss predictor model. Another example would be a model that assigns individuals values that are more variable individually yet taken as a whole are more or similarly accurate at predicting team records. The first model doesn’t try to get at individual values and the second isn’t as predictive at the individual level. If I’m a GM and I’m trying to make my team better, I need a tool that tells me which player will make my team better, not which tool predicts league wide win distribution the best – although the first model would be of interest.
If you’re trying to predict league wide wins or a specific teams’ total wins from one season to the next there are likely better models that are optimized to do exactly that. Said model will deal with the details of minute estimation, injury likelihood, regression to the mean, efficiency differential etc… in a statistical fashion all at once without accumulating errors due to using distinct models together.
If you’re trying to compare wp48 against some other player productivity per minute measure you need to isolate the per-minute estimation model. The best way to do this is to just remove it entirely and use the actual minutes played for a concluded season with the previous seasons’ player per minute measures.
If there’s a per minute (or per game) model out there that produces per minute measures that better predict team records and are less variable individually than wp48 then it could be a better tool for a GM (per game measures having additional constraints for usefulness).
As for gambling, I’m betting that there are better models (probably tightly held) out there. I wouldn’t use a per minute model except in specific cases where I thought that it might have an advantage over a different model’s approach.
Gil Meriken
May 31, 2010
@ilikeflowers, So then how best to measure the power of WP48 as a tool to tell a GM which player will make his (or her, someday) team better, besides anecdotally?
Is it even possible, given that we would have to consider player moves and transaction that did not happen, as well those that did?
It seems too convenient to celebrate the model when it seems to be validated by the improved performance of a team, only to ignore or explain away instances in which a team’s performance deteriorates when WP48 would indicate that the team should have gotten better.
Would you agree then that measuring the effectiveness of WP48 as a “GM tool” as you say, is a subjective enterprise? How then, can one make claims about its quality? This is not like one of those basic science experiments that can test and prove things empirically (I remember measuring the speed of a falling object in physics lab to estimate gravitational acceleration. That was fun.)
Dre
May 31, 2010
Gil,
I feel we’ve gotten off track. You asked a question: “Is WP48 the best tool for gamblers”. WP48 takes a decent amount of work, whereas there are simpler methods that take far less work. So we are not saying WP48 is a bad predictor. Quite the contrary, we think it is a good model, but a person should use the simplest model possible.
Also look over the last few years of data. Both this site and my site(listed on the right) have these. WP48 has definitely shown Allen Iverson’s trade wouldn’t really help the Nuggets, Turkolu wouldn’t help the Raptors and that the Rockets would do better than expected.
This model has done a very good job using players as the focal point of explaining team wins. Now in terms of predictive power, you should be careful with any model. One of the radio shows Dr. Berri was on pointed out small things turn the instance of a franchise, Jordan doesn’t play baseball, Artest doesn’t go into the stands, Garnett is healthy last year etc.
Also I appreciate your arguments, but I will stress as you start questioning the model, please supply examples. For instance, where has a team performance deteriorated where WP48 thought it would get better?
ilikeflowers
May 31, 2010
Gil,
That’s not what I’m saying at all. I’m saying that if you want to beat Vegas, a per minute measure is probably not the best model to use. What a GM needs and beating Vegas win/loss lines are two different things as I have detailed.
A GM wants the most accurate player-based model possible. It needs to be accurate at the team level as well – but it doesn’t need to beat Vegas. Vegas presumably, either doesn’t have, isn’t using, or won’t sell you their per minute models so you have to look elsewhere. According to the first book WP48 predicted team records for the 03/04 season with an average error of 1.67 wins with a max error of 4.41 wins. That’s pretty good. The methodology and data is freely available to run numbers for additional seasons. IIRC, player wp48 varies by about 10% year to year meaning that is it likely capturing the individual’s performance separate from teammate effects within 10%. Some of this variability is due to aging, so that 10% is likely overstating the variability.
Is it even possible, given that we would have to consider player moves and transaction that did not happen, as well those that did?
Yes, it is possible via historic wp48 and how many minutes the GM thinks that a player will play. If you have a choice between your old player A and a new player B, if player B’s wp48 is 0.100 more than player A’s then over 2952 minutes (36min*82) player B will likely get you 3 additional wins that season (actually more like 2 – 4). That’s critical info. Now a GM can say with better certainty whether or not player B is worth the money and what other players need to be signed to reach the desired wins.
It seems too convenient to celebrate the model when it seems to be validated by the improved performance of a team, only to ignore or explain away instances in which a team’s performance deteriorates when WP48 would indicate that the team should have gotten better.
Where did this come from? What specific instances are you talking about?
Would you agree then that measuring the effectiveness of WP48 as a “GM tool” as you say, is a subjective enterprise?
I disagree completely for the reasons given above. Its empirical accuracy is readily determined by comparison to actual outcomes. One can measure it for themselves or accept the analysis of those whom they trust to be experts.
ilikeflowers
June 1, 2010
Just a note. Those same season win errors when coupled with 10% player variability in wp48 are enough to establish the accuracy of the GM’s individual player decision making calculations given above. For league wide win predictions we actually want to know the season win errors using previous season wp48’s. This is what one would compare against ‘Vegas’.
Assuming that the same-season win errors represent the accuracy limit of wp48 at the team wins prediction level one can tell whether or not wp48 with even a perfect minutes model is likely to beat Vegas. If it can’t in this best case scenario then no minutes prediction model that you use will fare any better and you need to take a different approach.
brgulker
June 1, 2010
This is sort of stating the obvious, but I have a really hard time understanding why people can’t grasp that there isn’t and never will be a statistical model that will be able to predict future team wins and/or individual player performance with 100% accuracy.
Anyone who’s watched the game of basketball at any level for any amount of time (much less investigated the various advanced models out there) should be able to observe that players aren’t 100% consistent from game to game or season to season.
I’ve avoided it up until this point, but to respond to joe’s initial observation — it’s not obvious to everyone that Stuckey isn’t as good as Rondo. Simply read the articles by Langlois to which I linked. It’s clear that the Pistons think Stuckey is/will be a very good player.
Daniel
June 1, 2010
Here’s some WOW gambling successes. A .95 correlation of WOW to team wins means that by and large, predicting team successes and failures is extraordinarily easy. When adding many fallacies in the conventional wisdom surrounding the NBA, it’s not difficult to pick some great values.
Don’t gamble. It has great potential to lead to a terrible lifestyle, and it is unbelievably addicting. If you must, you should use a consistent predictive model, and you should only bet when the stats are overwhelmingly on your side.
The big winners in past seasons have been the OVER with the post-Iverson trade 76ers (AI= WOW loser), the OVER with the post-KG trade Celtics (KG= WOW champ), the UNDER with this year’s Nets (signed all bad WOW players), the OVER with this year’s Bucks (shed unproductive RJ), the UNDER with this year’s Pistons(signed unproductive players), the UNDER with this year’s 76ers (lost Andre Miller). These all happened with gigantic margins, like an average of well over 10 games.
(The Wizards were WAAAAY off, though, and Arenas’ gun trouble couldn’t have been predicted. At 41.5, it was a definite stay-away at the time, as they would have barely reached the .500 mark if EVERYTHING went right).
Gil Meriken
June 1, 2010
@brgulker – the question is not whether there is a model that can predict the future, but how to measure the quality of the a model relative to another model.
If Model A has a better record of predicting outcomes than Model B over a large sample ( and continues to have a better record), wouldn’t that imply that it is a Model A is better than Model B?
ilikeflowers
June 2, 2010
Of course it does Gil, everyone knows that.
Here is a summary of your posts so far:
Models A, B, and C are less predictive (by some unspecified amount) than Model D at predicting Apples, therefore Model E isn’t of much use in predicting Oranges.