Nerdnumbers avatar

The Boxscore Geeks Show: Wins Above Replacement Podcasting



Video Show

Show Notes

Replacement Level Players

Nate Silver started off the WaR discussion using a bit of flawed logic.

Patrick posted a reply to the problems with this.

This brought in some SABR folks to give bad analysis.

And finally Dave Berri, who has a PhD in economics and is a tenured professor, wrote a long piece explaining the philosophies and pitfalls with WaR

Which lead to even more bad (but agreeable) analysis from SABR

We had much more to say on the subject, of course!

Being "average" in the NBA is valuable, as becoming a top NBA athlete is difficult.

If a super "rare" player is available, you take the deal!

In terms on "minimum wage" players, there's tons of salary constraints due to the CBA.

Even overseas players are not as easily attainable as you might think, thanks to FIBA.

The D-League players that have come to the NBA haven't really been that stellar. In short, the D-League isn't a "replacement" level factory.

And of course college players have restrictions on when they can join the NBA and which teams they can sign for.

Basically, there are not just "free markets" to tap for "replacement players".

Something many don't consider is the opportunity cost a player imposes. Every dollar they cost, minute they play, and shot they take is coming from another player.

Excellent quote by Patrick

"If you come at me and tell me something that is factual wrong, and call me dumb for my point of view, then my tolerance level to have an actual, rational conversation with you is going to be pretty low."

Kevin Love

We've been on the Kevin Love bandwagon for a while.


We had quite a long discussion about the ridiculous concept of players "padding their stats" and defense. Tune in to listen.

The Wolves were an above average team in defense last season, for what that's worth.

A "defense" of Klay Thompson is his defense on Golden State, which again, I don't agree with.

I've had various issues with defense. I'm not 100% behind Synergy stats, but I give them more weight because unlike other stats, we know why the rate players how they do.

Shout Outs!

Nick Haugen (@HaugenND) and TangoTiger! Thanks for being a great fan and inspiring good work, respectively.

"Being "average" in the NBA is valuable, as becoming a top NBA athlete is difficult."

Yep. Being average in every sport is valuable.

"In terms on "minimum wage" players, there's tons of salary constraints due to the CBA."

Yep, these would have to be considered when building WAR for the NBA.

"Even overseas players are not as easily attainable as you might think, thanks to FIBA."

Again, yep, must be taken into consideration.

"The D-League players that have come to the NBA haven't really been that stellar. In short, the D-League isn't a "replacement" level factory."

They don't have to be stellar to be our baseline. We just need to know if they'd produce anything at all. Basically, would a team full of minimum contract players win more than 0 games? If they would, then..... "Basically, there are not just "free markets" to tap for "replacement players"." is not true.

"Something many don't consider is the opportunity cost a player imposes. Every dollar they cost, minute they play, and shot they take is coming from another player."

This is basically the whole logic behind WAR.
What you are basically asking is to make a model of several dynamic markets with different pricing structures. That was basically our point on the podcast and the link to Dave Berri's post. This is not a "trivial calculation", nor is it something the current WaR for baseball does.
It should be pointed out that "Wins Produced" is, in fact, a marginal wins produced - that is to say a Wins Above Replacement - metric. The replacement level is implicit in the "Position Adjustment". (There's some fuzziness , but "Wins Produced" has an implicit baseline of player performance around what would be expected in a barely 0-82 team.)

As with any marginal production measurement, there's an issue of what the baseline is - which, in a sense - even comes up in this podcast around 18:40 during the discussion of Monta Ellis' contract when Patrick mentions that the 4 wins that Ellis' gets credit for are not worth an 8 million contract because Ellis plays 3000 minutes to get those wins.

At 2:40 or so, Patrick claims that the 538 article assumes that replacement players would be "NBA average" when that's clearly not the case. Nate Silver mentions that he revised his notion of a 'replacement level team' from 0 wins per sesons to 16 wins per season. (So, I guess Silver agrees with Patrick that Ellis is a bit overpaid.)

It does seem a bit like you guys are disagreeing with Silver just to disagree with him. For example, in February, Forbes published a list of the 10 most overpaid players in Basketball which is subject to all the same criticisms about oversimplified contract evaluation, and you guys had no issue with it.
"If you come at me and tell me something that is factual wrong, and call me dumb for my point of view, then my tolerance level to have an actual, rational conversation with you is going to be pretty low."

The quote is a bit concerning, if understandable. Mostly because as Dave Berri is point out on Twitter today:
"1) Of all sports fans, I suspect the percentage that care about how to measure productivity of athletes is small. 2) Of those that care to measure productivity, the percentage that understand how to do this is also quite small. 3) So when you are discussing productivity measures in sports, the size of the audience that gets what you are saying is very, very small."

Taken together, there is an infinitesimally small amount of individuals 100% capable of having a conversation to the level being requested. That doesn't mean they are all below the intellectual threshold, some simply do not have the background or experience. There is a slightly larger group of people wishing to read about the conversation, though not capable of the analysis themselves (and some venturing comments as well).

Now I'm not telling you how to run your website or how to react, but as a dedicated reader I find it much more interesting when facts that are not accurate are corrected in a way that makes sense and is teachable. Then as an uninformed person I can understand where things went wrong and learn. A great example is Dave's latest post on WAR in response to this discussion.

I would gain much more out of it as a reader if the logical follow through was done instead of cut off. Just food for thought when you think about your reader base and future articles with similar discussion.
I agree its not a trivial calculation, and 'm not asking you to do it personally, but choosing to do it would make your analysis better. And even a "dumbed down" version would work better than the current method of ignoring it altogether. But the main issue is that a lot of the comments from the "anti-WAR" side aren't saying it would be complex to create, they are questioning the basic concepts. Even Berri, until his most recent post (which while a nice leap forward, still had some mistakes in analyzing WAR as an idea), wasn't even acknowledging how valuing players over replacement value is important. I get that you guys pride yourselves on not being "saberists", but if you are going to do sports analysis, there is a lot to be gained from researching the work done by "saberists" in both basketball and other sports. These things have been carefully considered by many smart people already and some good ideas have come out of it that might not be intuitive to everyone.
Nate, good point. WP already uses a baseline of comparison to come up with marginal wins. And given the "opportunity cost" concept they bring up, they clearly get the main idea. I feel that in practice (whenever referencing WINS PRODUCED), they use the WAR idea, but just disagree with where the "replacement player" baseline should be. The difference is that inn baseball's WAR, you shouldn't pay/hire a negative WAR player (since a replacement could be had who could generate 0 WAR, an improvement), but since the baseline for WINS PRODUCED is average, a player could have negative wins produced and still be worth paying (because they are still better than a replacement level player). That shouldn't be the case.
"Replacement Level" is difficult to determine in any sport, industry, etc.. It's difficult in baseball -- where the CBA also distorts the market -- and it's difficult in basketball for some of the reasons you cite. I imagine it's difficult at the car lot too.

If you asked the stat guys from the 30 MLB teams how they calculate replacement level, you'd might get 29 different answers. (The Phillies would just look at you weird.) But each would correspond roughly to what a bench player or minor leaguer could be expected to produce if thrown into a starting role.

The same is true for basketball. Replacement level is an imperfect estimate with a messy dataset and there's no single way to do it right. But you can eyeball it. You can say replacement level is what the worst starters in the league produce. Or what the average of the last six guys on the roster produce. And you'll come up with a reasonable metric.

And measuring players value above a baseline of replacement level will give you a better approximation of their actual value than measuring them against average production or above a baseline of zero.

Nate Silver -- who explicitly did NOT confuse replacement level with average level in his article on Carmelo -- implies a team of replacement level players would win 16 games in the NBA, which equates to a WP48 of .039 for a replacement player, as opposed to .100 for an average player.

What I would like to see from this blog and from Dave Berri and everyone else is a discussion about the best way to estimate replacement level in the NBA. It may be messy, or you may be able to find a simple method that does the job.

But debating whether replacement level and is a useful concept because of the peculiar circumstances of the NBA? Come on.
"(The Phillies would just look at you weird.)"


..... And also, great post. I intuitively think Silver's replacement level season wins estimate is a little high at 16, but I think it's definitely greater than 0. (8-10 feels about right to me, but I withhold the right to be proven totally wrong on that).
Also as briks says, there's a lot to be learned from SABRists.

It's not as if MLB doesn't have dumb GMs and a lot of unnecessary weirdness in the CBA. But that doesn't keep SABRists from estimating replacement level, because replacement is a useful concept.
> "...since the baseline for WINS PRODUCED is average..."

Wins Produced is scaled so that the baseline is a team with -30 point differential - well below NBA average. The worst NBA team to date is probably the 2011-2012 Bobcats with a differential of -14.

(I made the same mistake earlier, but thought I'd corrected it before posting. If a retraction somewhere is appropriate, please let me know.)
Briks, thanks for the hat tip! Right back at you.

I agree with you that intuitively 16 wins seems like a high bar for replacement level in the NBA. Seems like every year there's a team that only wins 16 or so games. But who knows -- I'd love to see Silver's math. And maybe the low win totals are a result of teams intentionally playing below replacement level.

And I think that part of the problem convincing basketball statheads that replacement level is a worthy concept is that replacement level is a lot closer to zero wins in basketball than it is in baseball, where the very worst teams can still be expected to win 1/3 of their games.

But it's not zero wins. And so it is a useful concept.
Nate: the baseline is more like -15. Each point is worth .033 wins, so a -15 team would project to a win% of approximately .000.
This is should be a topic more discussed in the news. Its an option to get away with paying a guy below minimum salary. Player's union should look into this because the league is going to catch on.
Also, there are 78 million basketball players in the world but only 450 jobs. So it is likely that there is a population of players that have either been in the league and have played at a decent production to be pushed out for whatever reason(Fazekas and Balkman come to mind) or undrafted players(this groups up all the leagues, NCAA included) that were misevaluated for players with more prestige. So I agree with what is said. Muscala was drafted by the hawks and stash in the ACB then brought back for under the league minimum for last year. *my comment above points to this being a trend that the league should take advantage of
There is not any question about what the marginal production is when you can add or subtract workers. If you hire someone to sell cars and car sales go up by 2, his marginal production is 2. Period. We know this because 'nobody' sells 0 cars, and 2 - 0 is 2.

We have no idea what someone's marginal production is when you substitute workers. If you fire a worker and replace him with someone else, and car sales go up by 1, then we know that the new worker is 1 car more productive than the previous employee - but we *do not* know what either person's marginal production is. We only know the difference.

In the first case, you are solving one equation for one unknown, which frequently has a well defined answer. In the second case, you are solving one equation for two unknowns, which you cannot do. There is a residual degree of freedom that you cannot account for.

In the context of a basketball team, we know that 2 teams of 5 players playing 1 game produces 1 win. We do not know their individual productivity, however. We know they sum to 1 win, leaving 9 degrees of freedom. We know, roughly, how much better one team was than the other (via point margin), leaving 8 degrees of freedom. We can also compare opposing players at the same position against each other (to some degree), and controlling for that, we have 4 residual degrees of freedom.

Those 4 degrees of freedom correspond to the 5 position adjustments (if you set 4, you know the 5th). Without pulling in some additional information, you can set them to *anything you want* and get the same answer.

You can set them all to 0, and use AdjP48. Or you can set them all such that average production is equal to 0.10, and use WP48. Or you could examine minimum contract players and set minimum contract production at 0.039 (~16 wins over the whole season). These will all give you the same answer for predicting team productivity, and they will all do just as well at predicting wins.

They all have very different implications about which players are being exploited, however, and what contracts are 'fair'.
Couple things, in response to DooDoo_Jump in the last thread.
First of all, in the last thread there was a good deal of discussion about the Picasso analogy that Patrick employed. I pointed out that there are no “replacement” Picassos- no paintings of remotely comparable value on the market at a similar price. DooDoo_Jump responded, “Nobody's arguing about $200 Picassos, they're saying you should judge the value of the Picasso against the value of putting an empty frame that just displays a patch of your wall.”

I don’t want to get bogged down in arguments over analogies, and after this I want to get back to the numbers, but I think it’s important to straighten out our assumptions here. DooDoo_Jump’s “empty frame” is obviously not a tenable analogy to anything like the sort of “replacement” player Silver and TangoTiger are discussing. By Silver’s definition, a team of replacement players should win 16 games, for a WP48 of around .039 each. If Kevin Love is a quality Picasso, then a .039 WP48 NBA player is an inferior painting, but still a painting; I mean, the difference in value between a Picasso and an empty frame is so great as to be not worth considering.

A “replacement-level” .039 WP48 NBA player is someone who is definitely able to make money playing high-level professional basketball, whether in the NBA or abroad. It’s silly to use call such players the NBA equivalent of an empty frame compared to the Picasso of Kevin Love. The idea that huge numbers of players not already in the NBA would be able to post .039 WP48 over a season is completely off-base. Once we get into the realm of the truly unproductive, loss-producing players below .00WP48, we might be talking about the NBA equivalent to an empty frame. But a team comprised wholly of such players would be expected to win only a handful of games at best, and is a far cry from the level of production displayed by even players around .05 WP48. Players at even that mediocre level of productivity are very hard to find, like a Degas as compared to a Picasso. Only woefully unproductive players make sense in DooDoo_Jump’s response to my argument, and such players are a far cry from those on Silver’s hypothetical 16-win team.

In short: Kevin Love is a Picasso available for $500. There are no Picassos available for $200; all that will buy you is a Degas etching like Channing Frye (this actually works out quite nicely, 500/200=2.5, 13.67/5.6=2.44). But hey, Degas etchings are still pretty rare, to the point that their prices can vary widely over time. Tony Wroten is an empty frame. A team of Tony Wroten-level players don’t win 16 games.

Glad we cleared that up. There are no replacement players.

More later, don’t wanna get censored.
One issue with estimating WAR in the NBA is also the allocation of minutes.

If some teams played players they are not currently playing, they might get better without making any roster changes. If you go with the "worst starters per position" method, you probably have players worse than the top 150 players (even accounting for position) as your marginal player, whereas taking an average of bench player production might produce results better than some starters.

The inefficient allocation of NBA minutes means that the estimation of WAR, even if we assume a more simple market than exists, is a non-trivial question without a minutes allocation model.

The largest "cheap" gains in results available in the NBA still probably come from identifying absolute levels of productivity accurately and then playing those players.

If you think the freely available players would win a "handful of games", then yes, there are replacement players. If they would go 0-82, then you are correct, there are no replacement players. Nate Silver's estimate of 16 wins by replacement level players seems high to me, as I've stated. I'm also pretty a team composed of players all making the minimum would win more than 0 games, so there are indeed replacement players.
If you believe WAR, then it's obvious there are teams that could be composed of minimum contract players that would win far more than zero games... look at the FA rating article on this website vs. what Aminu just signed for, for instance.

The problem is that salary is not necessarily correlated with productivity, and as a result, if you have your pick of minimum players, it would be trivial to create a >0 win team. Ignoring how we would address rookie contracts in regard to what a minimum price is (there is obviously value in that space as well).
> the baseline is more like -15. Each point is worth .033 wins, so a > -15 team would project to a win% of approximately .000.

Right. So the 2011-12 Bobcats are close, but if we naively extend the strike-shortened season we'd expect around 8.5 wins from them.

To be clear, when I wrote that a team of unproductive, roughly .00 WP48 players would win "a handful of games at best," I meant that they might win 5 or 6 games in a good season with the benefit of streaky, inefficient shooters having a few good games by sheer luck (anybody can hit a bunch of long twos and threes in close succession if they're given enough opportunities), but by "at best" I meant that such a team would likely win fewer than even a handful. A team of players who each have a WP48 of .005 has an expected mean wins per 82 games of 2.05, but there's only a 12% chance that such a team would actually fail to win a single contest over 82 games. I'm diverging from .00 so as to produce manageable figures, but such players could easily post WP48s of -.005 or some similar number without playing fundamentally differently at all; variations of this sort in individual players happen all the time, and there's no reason to think that a .005 player is much different from a -.005 player.

I don't disagree with you at all that there are large numbers of players readily available to teams who could post around .00 WP48, but by definition these are completely unproductive players to whom no NBA team with aspirations of winning should be giving minutes. Such players cannot be described as "NBA replacement players" because they cannot be genuinely described as "NBA-level" players at all.

One more analogy: a .005 WP48 player in the NBA is like an unskilled programmer with a job at Google, who is able to code to some extent in the context of computer science courses or simple web development, but does not have the ability to contribute meaningfully towards Google engineering projects because of the disparity between his skill level and those of his co-workers. It's not that he can't program; it's just that, in the context of Google engineering, he's not a good enough programmer to be productive. There are plenty of people with limited programming skills, but Google should not consider people who can't contribute to projects to be the "replacement" level against which to measure their engineers. Even a below-average Google engineer is a very good programmer in the grand scheme of things, like a .05 WP48 NBA player.
It's pretty straightforward to identify who a replacement level player is in baseball. Teams have a deep minor league system with quite a few 'veterans' who have played a lot of games at a not-good-enough-for-the-majors level. Our belief is that those players are not good enough, and we have a lot of confidence in that belief. This makes it easy to calibrate a scale to their production.

Basketball doesn't have an analogue. Players on minimum contracts tend to be young and not play a lot of minutes. Our belief is that they aren't very good, but we have *very little* confidence in that belief. Some of them turn out to be very good players, some of them are awful. We don't know, and since they don't get many minutes we continue to not know. So it is very difficult to calibrate a scale to their production - even if we take their average performance, there's an 'uncertainty premium' built in that is not easy to account for (and cannot simply be ignored).

The concept still makes sense, it's just way, way harder to actually implement. Anyone know of data on how baseball values uncertainty?

"It should be pointed out that "Wins Produced" is, in fact, a marginal wins produced - that is to say a Wins Above Replacement - metric. The replacement level is implicit in the "Position Adjustment". (There's some fuzziness , but "Wins Produced" has an implicit baseline of player performance around what would be expected in a barely 0-82 team.)"


You read it wrong. This is not what the position adjustment does at all.
> Patrick wrote:
> No.
> You read it wrong. This is not what the position adjustment does at all.

Let's, for the sake of discussion suppose that there are "replacement players" available with an expected WP48 of 0.000. It should be pretty obvious that if we calculate the "Wins Above Replacement" for players based on that assumption and the usual Wins Produced performance metrics, we're going to get the players' Wins Produced.

That means that there is a "Wins Above Replacement" model which produces the exact same numbers as Wins Produced. So there "Wins Above Replacement" model that is mathematically identical to Wins Produced. So the distinction between Wins Produced and one particular "Wins Above Replacement" model is one without a difference.

I don't think it's necessarily harder to identify replacement level in basketball than it is in baseball -- it's messy in both sports! In baseball, you have minor leagues and 200+ players under contract to a team, which means there's not a fluid market of replacement-level talent once a season begins, and the skill levels of the players in the #2, 3, and 4 spots on the depth chart differ dramatically from team to team. And you have a reserve clause which teams can extend to seven years of control, giving teams a major incentive to value younger players under team control over older players with more rights and equivalent skillets. These factors and many others effect who a team has available when a starter goes down and distort the market.

Basketball has a messy CBA too. But here's the thing -- there's absolutely no need to get hung up on the CBA or contract type when determining replacement level. There are quite a few guys raking it in who play at replacement level, and quite a few guys on minimum deals playing well-above replacement level. I would ignore contracts entirely when estimating replacement level.

Instead, build a simpler model looking at players are past a certain spot on the bench, as Tom Tango did in one of his posts. Or playing time earned. Or players who are out of the league the next season. Or whose minutes decrease substantially over the course of a season.

Plot the career stats of every player who's played in the NBA in the last 10-20 years. Plot their total number of minutes played on the X-axis, and their career WP48 on the Y-axis. You'll probably notice that guys that don't earn a lot of minutes have a lower WP48 than guys who have long careers, and you might even discover an inflection point that gives you some idea where "replacement level" resides.

Points are: 1) while there's no one correct way to estimate replacement level, there a plenty of easy ways to get you in the ballpark/arena and move the debate forward, as Tom Tango does as well as anyone analyzing any sport; and 2) the methods of estimating replacement level aren't likely to be any messier for basketball than they are for baseball or any other sport. They're messy everywhere! It's a messy concept.
Another point about replacement level in baseball vs. basketball.

In MLB, a team composed entirely of replacement level players is expected to win 40-45 games/season, or about 1/4 of its games.

In the NBA, I would estimate purely from intuition that a team composed entirely of replacement level players is around 8-12 wins/season (Nate Silver has it at 16, Briks agrees w/ me), or about 1/8 of its games.

So when you conflate replacement value with zero value in estimating the value of baseball players, your errors are going to be far more substantial than when you do the same for basketball players.

So when you place a value on a basketball player, you can get away with ignoring replacement level (ie -- conflating it with zero) because it's relatively close to zero. But you're still missing an easy opportunity to optimize your analysis.

As I'm sure you know, the position adjustment corrects against average production; taking an ADJP48 of .00 as "replacement-level" ignores the huge difficulties in assembling a team of players above that quality, especially at the larger positions. As I've said, once we've converted ADJP48 to WP48, we find that the only group of players readily available to teams on a consistent basis is those players whose WP48 is around 0 (ADJP around -.1); "replacement players" capable of constituting a 16-win team do not exist. As we know, winning teams cannot afford large expenditures of minutes on literally unproductive players (WP48 of .00).

As I have said, WP48 of .00 is an unacceptable standard for a "replacement-level" NBA player, because while such players are readily available, they are completely unproductive and cannot play many minutes without seriously damaging their teams' performances, to the tune of a generous 2-3 wins per year, on average.

The sort of NBA players conjured by the term "replacement-level" and who can supposedly win 15-20 games per season are exceedingly rare in terms of absolute population. There is a short supply of tall people, and an almost-as-short supply of superbly athletic people with the basketball skill to play guard in the NBA. These statements are true even of garden-variety below-average NBA players with career WP48s in the neighborhood of .05. Totally unproductive players cannot be described as "replacement-level" regardless of their availability.
Man, I wish we had an edit button. I meant to write "to the tune of a generous 2-3 wins per year for a team of such players" to indicate their overall lack of contributions.
"A “replacement-level” .039 WP48 NBA player is someone who is definitely able to make money playing high-level professional basketball, whether in the NBA or abroad...The idea that huge numbers of players not already in the NBA would be able to post .039 WP48 over a season is completely off-base."

This is an argument that the productivity of a replacement player has been set too high, not that it doesn't exist at a conceptual level.

Marginal productivity implies a counterfactual - "what would your production process have looked like had you not employed Mr X?" If your answer to that question in this case is "those minutes would have been played by no one" then any player you consider as the marginal player is responsible for all that team's wins, since you are imagining them completing the starting lineup and avoiding a forfeit.

Obviously, that's not credible - we're clearly all modelling marginal additions to a team from an equilibrium where that team already has a functioning roster, meaning that marginal productivity is a measure of difference in productivity. Or "wins above replacement" if you will. Who's getting replaced? Tough to measure, but it's not "no-one".
> As I have said, WP48 of .00 is an unacceptable standard for a
> "replacement-level" NBA player ...

It doesn't really matter whether it makes sense to field a 'replacement level player'. Rather, replacement level is a minimum which every team should be able to reach, and thus a reference point for determining performance and cost.

For example consider players like Andrea Bargnani who have significantly negative Wins Produced. Anyone who believes that players at 0.000 WP48 are readily available at minimum salaries would suggest putting Bargnani on the bench and starting the replacement sight unseen.

Or consider Monta Ellis. He's produced 4.4 wins above 0 in 3000 minutes. That means his performance is below average for the NBA (which is closer to 6 wins) but he's still worth more than the minimum since his contribution is better than replacement.
0.039 WP48 is *not* what corresponds to replacement level. A team of replacement level players that would be expected to win 16 games, which puts the average of your 5 guys at 0.039 WP48 - but there is no reason why all positions must have the same replacement level WP48. Centers that can play at a 0.039 level are not replacement level, for example - they are rare and get paid well.

I would not be surprised if replacement level at center corresponded to a *negative* WP48, while a replacement level shooting guard is likely to be positive under WP48.
@BPS - Agree with you 100% on positional valuation.
It still doesn't make sense to me how a productive swing man like Josh Childress has struggled (the last couple of years) to find a team that will give him consistent playing time, but a scrub like Derek Fisher managed to play upwards of 20 mins his entire career (safe for in his inaugural season and exaugural season). It infuriates me to no end when productive bigs like Ed Davis, Jordan Hill, DeAndre Jordan (when Del Negro was in LAC), et al, are benched in favor of less productive players. It's down right criminal imo. So much so, that I really feel there should be some type of consequence levied against the coaching staff, management, the organization. Especially when said productive player winds up out of the league (I mean, at that point, you're pretty much fucking w/ a man's earning potential).
I agree. These jobs are so political too so if you can be easy and personable to work with, it gives you staying power but if you are a distraction (legal or illegal) then that gives teams an out if you a bad player or an overreaction if you are a good player (birdman)
I think that Nate brings up a valid point in that if we assume they produce zero wins, and subtract zero from their wins produced we simply get wins produced.

the question is first and foremost is it useful to know the difference between an entry level player and an average player and a superstar for each position.

In this sense, I think there is far more to be gained by simply directly comparing wins produced and directly comparing salaries. and I dont see subtracting a set amount of wins from every players production as revealing any more data that could be seen as useful.

in the end it is simply going to mimic the usefulness of whatever win metric we use.
I think that Berri's point that pay is not well correlated with wins is prescient. The effective replacement level in the NBA is closer to 0.150 than 0.040. Focusing on replacement level is premature at this point in the NBA. If the market ever becomes more efficient (that is at minimum, when it starts becoming difficult to acquire above average players for the vet minimum salary), then the idea of replacement level might start to be relevant to the NBA. For now, there are too many negative producers seeing heavy minutes, and too many jobless above average producers for the idea to be very relevant. Until more teams give up their bias toward per game scoring totals, a given team's replacement level will be dominated by its ability to recognize productive players.
I agree with Shilelea, the question should be if a WARP model is more useful than a Wp one or WS or any other, not if it´s logic match the reality instead of being logically consistent(thing that i don´t know for either case).

I liked Dr. Berri´s analysis but just like you guys he made the suposition that replacement player= avg. player, and for what i understand that´s not true. In baseball a RP is close to a "AAAA" kind of player, a guy that can produce wins like an FA would produce paying him the minimum, discounting, like Silver did, rookies because they make a lot less money than his value and they can´t change that(until arbitrage) . The last time i checked a win in baseball cost around $6 millions/year, so that would be the price we´re talking about.

BTW, if really WP is a marginal win model, and WARP is one too(sustracting the RP value) then you can transform one in the other, like Nate said, WP is a special case of WARP
About patrick´s comment: i hate trolls too, or the people that attack you because you have a different opinion, but it´d be easier for us to learn and have a good discussion if you make your arguments with more clarity, i´m not asking you to be teachers or giving us a course just understand our position too, because if everyone gets a different idea of a post, we can´t have a useful discussion. I hope you take this as a constructive critic i really like BSG
Whether or not a "replacement-level" player should exist "at a conceptual level," (in Lacemaker's phrasing), if we define the term- as this thread seems to have done by consensus- to denote players with about .00 WP48 in the NBA, then the term "replacement-level" does us no more good than our current recognition of unproductive players as identified by WP48 results around .00. If we define "replacement-level" players in this way, the concept just doesn't do any work in helping illuminate the NBA labor market. This is aside from the more basic confusion about the distinction between the evaluation of marginal productivity and the pricing of labor in the market which has characterized these threads.

On the subject of, yknow, basketball, did anyone else notice that the Rockets picked up Jeff Adrien? Completely unbelievable, esp. with the Capela and Ariza signings. They are clearly using something strongly correlated with Wins Produced to make decisions down there (even if it's a "holistic" process, they're clearly systematic in evaluating players). I have found a correlation about .72 between Wins Produced and Win Shares, which in general is pretty strong, but Morey really appears to be using something very, very similar to Wins Produced.

Except Bosh and Melo. I don't really think there's any way to reconcile the Rockets valuation metrics with Wins Produced with such huge swings at pitches so obviously out of the strike zone (that metaphor checks out, I think :D )

A part of me thinks that Morey publicly goes after these guys because he's supposed to go after them, but then sabotages the deal by making contractual demands that he knows the player won't agree to, or something since he mostly strikes out on the huge mistakes and seems to land the legit guys. But then my more rational side tells me that this is a crazy theory that's probably not true. It's not impossible, but it's probably more likely that their process surfaces targets that are effective and targets that aren't. I think we can definitely say that they aren't quite as biased toward inefficient scoring as the rest of the league. They seem to appreciate rebounding rate, shooting efficiency, and a lot of other things that are good. But I think they also seem to discount a player's shortcomings (like Bosh's lack of rebounding) that make his overall effect on a team not very impressive.
When talking about the Rockets keep in mind that they were a 50-55 win team with salary cap space and the ability to match on Chandler Parsons when making those offers. Even if you assume Bosh or Melo are merely ~0.150 WP48 players those contracts aren't totally insane; they'd push the team over the cap but also up into the 60 win range that threatens for a championship for a couple years.

For a team winning 52, 53 games a year getting an additional couple wins is huge, it means deep playoff runs and the money that comes with it; paying 3 million per win in that position is totally sane.

They don't offer those contracts without the ability to match Parsons - and they don't even match Parsons, who is comparable to Bosh and Melo in production as well as a bit cheaper. Once they couldn't spike their payroll to make a championship run, Morey went back to playing tight looking for underpriced talent while keeping room to make a run at the next underpriced max guy.

If you're critiquing my critique by saying that my version of the analogy isn't perfectly "to scale", you might want to check your numbers. After all, Picasso's work is some of the most valuable of all time. Kevin Love is fantastic, but he's not a top-5 player all-time.

Substitute blank canvas, reprint, one of those crappy Mondrian paintings, whatever. The thing I was trying to get across is that nobody is arguing (as you implied the first time around) that there are plenty of Picassos to be had, but many people would argue that sub-Picasso paintings also have some value, and, furthermore, the pro-WAR people are arguing that it's reasonable to try to find out how much more value you get from the Picasso than the available alternative, and to compare that to the difference in their prices.

Another plausible explanation for the behavior of the Rockets is that they can be blinded by bias and popular opinion.

This would be much more apparent regarding perceived top-tier FAs (for whom there is a ton of press) than lesser named players (nobody really cares if you signed Jeff Adrien vs. Evan Turner, except people on blogs like this one), so the impact of bias should be lower in lower profile moves, and thus you'd expect higher quality moves as they stick to the process.

There are a few teams, recently, that seem to be employing some variation of stats that correlate well with the moves they are making (Rockets, 76ers, Grizzlies, Spurs, Mavs); they don't all have the same philosophy but I think some organizations are clearly skewing that direction while others (Lakers, TWolves) are not...
First off, I really like this site, but I think the authors need to pump the brakes a little on the current disagreement.

Regarding tolerance for rational conversation, obviously dialogue tends to be more productive when everyone is civil. But in order to keep everything civil, its useful to accurately characterize the other side's argument. I think some of the incredulity from the SABR-response to this series of discussions comes from the fact that you misattribute Silver's definition of "replacement level" as "NBA average." That is an unforced error that throws much of the following discussion off track.

Also, I've noticed a tendency of those around the Bradbury-Berri school to make appeals to authority -- "if an economics professor says it, it must be right." Here, Dre notes that Berri "has a PhD in economics and is a tenured professor." That's all well and good, but I'm not sure that this appeal to authority is going to work on its own terms when you go up against Nate Silver, best-selling author and perhaps the most famous pop statistician around today.

This is particularly so in instances such as when Patrick relied on the JC Bradbury (also an econ professor I believe) chart purporting to reveal that there are more baseball players who perform at the MLB average level than there are who perform at a well-below average level. But there is no accounting for all the players in AAA, who obviously would fill that chart in on the "below average" side. The chart is such a self-evidently incorrect model of the real world that would be rejected out-of-hand on a Stats 101 exam.

To carry it over to the basketball realm, when considering Silver's concept of replacement level on its own terms, it is obvious that there are far more basketball players out there capable of performing at the "12th man" level ( I think pretty synonymous with Silver's or SABR's concept of replacement level) than there are who are capable of performing at the NBA-average level. And in fact, the D-league IS a "replacement-level factory" -- you can see this by the number of guys swapping in-and-out between the NBA and D-league in any given season.
James1: excellent post.

Patrick Minton:

What you said at the 7 minute mark IS the replacement level player we are talking about!

What you said at the 5 minute mark is NOT the replacement level and no one is suggesting otherwise. Not a single person.

This is a strawman.

Sign in to write a comment.