APBRmetrics

mark kieffer wrote:There are some interesting patterns I see:

1. You are operating under the assumption that a replacement level team would win 0 NBA games in a season... In baseball, a replacement level team has an estimated winning percentage of something like 29%.... There have been teams in MLB history that have win at that rate or worse. Obviously 29% is too high for basketball, but if you took the 10 worst records in the history, one could choose maybe something like 12-15% winning percentage (10-12 wins or so)

2. Your average WAR is about 2.6 wins (assuming I have no typos), and median is 1.13 wins.... In baseball they are close to zero... Now, I know in baseball there are more players on a team, so maybe that's the reason.


Either way, I think if the assumption was a replacement level team would win some games, I think the WAR numbers for LeBron for example might come down (even though it would still be the highest).

Looking at basketball-reference, LeBron's win shares are about 15-20 depending on the season. That just feels more right, than saying 28 or 30, IMO.

You are correct in all your points - my replacement player bar is quite low. I am well aware that a team full of my replacement players would win a very small handleful of games (by pure luck) - I actually just go by the assumption that ANY smidge above replacement level starts winning games - while a team of NOTHING but replacement players wins zero. We all know that's not a real world assumption - that's just the assumption I make so the results tie in perfectly to team wins. This is obviously why ALL player results might seem a little high - in REAL life no player's minutes would be completely replaced by replacement players - and a solely replacement players team still would have a win % (albeit, a tiny one).

Like what I posted before - I was much more concerned about the rankings making sense - and the results making sense compared to each other (so compiling and rankings careers makes sense). I mean - do any of us take baseball WAR as perfectly literal representation in terms of added (or subtracted) wins. I don't - especially since there isn't ONE accepted baseball WAR - I just like to see the rankings and differences. Trout being a +10 WAR - while another is +5 - to me that's Trout making twice the seasonal impact. Well, in my NBA WAR - LeBron made almost twice the impact as Westbrook (it would take almost two Westbrook seasons to equal the LeBron season in significance when compiling careers). Durant made about twice the impact of David West. Curry made about twice the impact as Kemba Walker.

But, yes, I knew the numbers across the board would be larger than Win Shares - by my setting replacement player at 80% lg average and setting replacement player wins at zero - and not, say, 7 per 82 games.

All your points are well taken.

I guess I don't really understand what the obsession over WAR really is... In theory, a replacement level player isn't playing...

In general things I want to know:

1) What makes a team win vs. what keeps a team from winning?

2) Should a team sign x player for y dollars, or z player for aa dollars, etc

3) Does the personnel on the team match the philosophy of the team?

4) Who should a particular team draft and why?

5) etc., etc., etc.

Debating how many more wins the Heat would have playing LeBron vs. playing a D-League guy is moot because it's never going to happen.

And for the record, I feel the same way about baseball too. With baseball it's interesting to know how much a player might help a team win, but looking at Albert Pujols' WAR in his prime is silly because the Cardinals weren't going to replace him with a replacement player.

Kevin Pelton wrote:

Statman wrote:It makes no sense to me to have poor Austin at some fairly solid negative WAR - if he truly was much worse than replacement, his coaches (and the guys that decided to draft him) would have to be some serious morons giving him all those minutes.

Speaking as someone who has Rivers rated so poorly, I would disagree. First, if the experience he's getting potentially translates into future above-replacement play, that's a reason to continue playing him despite the short-term ramification. Also, to say that no player can play below replacement is to ignore a lag time in evaluation and sample-size issues. Just because a player rates below replacement level during a given timeframe doesn't mean he's truly sub-replacement talent.

I'm feel ya Kevin. When we are talking the bottom of the league, it just feels like rearranging chairs on the Titanic. The point really is - I don't want a bad production guy who played a good # of minutes to be killed BECAUSE he played a good number of minutes (with a negative WAR). Rivers and his 1418 minutes shouldn't get a worse (negative) result than me and my 0 minutes in the NBA. We ALL know NBA ratings initially based on box score stats can be clouded - and playing time (ie, do enough things not found in the box score to warrant PT over higher production guys) is a good way to reign the results in closer to maybe actual real world impact.

I don't want NBA stars to have their career WAR lower when they hang on a year two longer. I think zero is bad enough - the rest (positive WARS) can be adjusted accordingly.

No matter how bad his production - I'll never have a Bruce Bowen season below a 0 WAR. Also, for that matter, the rankings based off my ratings (HnR & HnI) will never have Bruce Bowen ranked worse than he will be in any other "production" metric (say, PER & the such)- because my ratings account for PT (in relation to team quality and teammate quality) as a factor in the results.

So, because of this, Austin Rivers in 2012-13 will just be a higher minute purely replacement player by my WAR - even if maybe he was even worse than that.

mark kieffer wrote:I guess I don't really understand what the obsession over WAR really is... In theory, a replacement level player isn't playing...

In general things I want to know:

1) What makes a team win vs. what keeps a team from winning?

2) Should a team sign x player for y dollars, or z player for aa dollars, etc

3) Does the personnel on the team match the philosophy of the team?

4) Who should a particular team draft and why?

5) etc., etc., etc.

Debating how many more wins the Heat would have playing LeBron vs. playing a D-League guy is moot because it's never going to happen.

And for the record, I feel the same way about baseball too. With baseball it's interesting to know how much a player might help a team win, but looking at Albert Pujols' WAR in his prime is silly because the Cardinals weren't going to replace him with a replacement player.

Agree with ALL of this honestly. I am doing this for the more general fan - offering something for the NBA they are familiar with somewhat in baseball.

But EVERYTHING you say is true. NO NBA team would hire me to crunch out NBA WAR for them - it won't really tell them what they really want to know.

I'm doing it for fun, to attempt to quantify the history of the NBA player results for rankings. I think the results will often make sense - and may be fun when looking at it from day to day. The results do take into account about everything that is measured in traditional stats (box score stats, team results, team pace, playr minutes, etc). I think the results will make as much (or more) sense in general than either baseball WAR.

I think the career totals will probably measure up quite well with any NBA top player rankings (which has been an "in" topic here recently).

But, I WILL NOT be using WAR for player or team projections. I won't even be using WAR for college - since the diversity of SoS's & team quality would not make it work. WAR won't be involved in my college to pro projections.

This isn't supposed to be some higher level analytic - I'm posting the results here because I know you guys WOULD be more critical because almost all of you are looking for higher level (groundbreaking) analytic. I know that. I'm OK with it.

I just want to have fun with it and have results that most can live with - completely agree with - or somewhere in between.

Statman wrote:

mark kieffer wrote:There are some interesting patterns I see:

1. You are operating under the assumption that a replacement level team would win 0 NBA games in a season... In baseball, a replacement level team has an estimated winning percentage of something like 29%.... There have been teams in MLB history that have win at that rate or worse. Obviously 29% is too high for basketball, but if you took the 10 worst records in the history, one could choose maybe something like 12-15% winning percentage (10-12 wins or so)

2. Your average WAR is about 2.6 wins (assuming I have no typos), and median is 1.13 wins.... In baseball they are close to zero... Now, I know in baseball there are more players on a team, so maybe that's the reason.


Either way, I think if the assumption was a replacement level team would win some games, I think the WAR numbers for LeBron for example might come down (even though it would still be the highest).

Looking at basketball-reference, LeBron's win shares are about 15-20 depending on the season. That just feels more right, than saying 28 or 30, IMO.

You are correct in all your points - my replacement player bar is quite low. I am well aware that a team full of my replacement players would win a very small handleful of games (by pure luck) - I actually just go by the assumption that ANY smidge above replacement level starts winning games - while a team of NOTHING but replacement players wins zero. We all know that's not a real world assumption - that's just the assumption I make so the results tie in perfectly to team wins. This is obviously why ALL player results might seem a little high - in REAL life no player's minutes would be completely replaced by replacement players - and a solely replacement players team still would have a win % (albeit, a tiny one).

Like what I posted before - I was much more concerned about the rankings making sense - and the results making sense compared to each other (so compiling and rankings careers makes sense). I mean - do any of us take baseball WAR as perfectly literal representation in terms of added (or subtracted) wins. I don't - especially since there isn't ONE accepted baseball WAR - I just like to see the rankings and differences. Trout being a +10 WAR - while another is +5 - to me that's Trout making twice the seasonal impact. Well, in my NBA WAR - LeBron made almost twice the impact as Westbrook (it would take almost two Westbrook seasons to equal the LeBron season in significance when compiling careers). Durant made about twice the impact of David West. Curry made about twice the impact as Kemba Walker.

But, yes, I knew the numbers across the board would be larger than Win Shares - by my setting replacement player at 80% lg average and setting replacement player wins at zero - and not, say, 7 per 82 games.

All your points are well taken.

A suggestion might be to change the name of the stat you are calculating. Why not called them your player ratings or something like that, fit them to an exponential distribution or something, and change the values to fit some universal scale (like 1-10 or 0-100, etc)?

That way you can compare the impact of players similar to how you are doing now, and you won't have to defend WAR like you are now...

Statman

My thoughts on a LeBron-less Heat weren't related to your WAR metric. I just saw an interesting tangent being formed, and commented.

You may be correct that if the Heat lost LeBron, and FORCED a replacement level player to play his minutes, then they would seriously suffer. I was operating under the assumption that the Heat wouldn't have to play said replacement player much, if at all.

Your WAR ratings look good. I'm not too big on "Replacement Level" metrics, because the whole concept of "replacement level" seem so subjective to me.

mark kieffer wrote:
A suggestion might be to change the name of the stat you are calculating. Why not called them your player ratings or something like that, fit them to an exponential distribution or something, and change the values to fit some universal scale (like 1-10 or 0-100, etc)?

That way you can compare the impact of players similar to how you are doing now, and you won't have to defend WAR like you are now...

Well, I pretty much already do that - I have my Hoops Nerd Rating. 100 is average. It's my go to rating really - for both college and pro. Look at the WAR rankings - it's the rating in green on the right (HnR) with league ranking. LeBron 165, Durant 154, Paul 141, then 10 guys tightly bunched between 133 & 127. It'd be similar to PER, except 100 average insteade of 15.0, playing time in relation to team quality is a definite factor - and the summed player totals will equal the team rating. The concepts behind HnR are extended to WAR - 80 is replacement level, so the higher the player is above an 80, the more he played, and the more his team won all factor in to WAR.

I actually think HnR is a more "true" rating for a player - WAR adds a bigger emphasis to actually playing a ton also to get those "wins".

BUT, WAR is really a counting rating - I can simply ADD individual season WARs for career WAR. I can't do that for HnR. I LOVE the history of the NBA - and I wanted a better way to quantify careers (both season and playoffs). WAR seemed a perfect fit for that (playing great matters, playing a lot matters, contributing to winners matters). Plus, the term WAR is becoming familiar to the casual stat friendly fan. The casual stat fan can understand the concept of simply adding each season of a rating for a career total. So, I created it.

I'm gonna let it marinate - do the past seasons, do the upcoming season updated every day. I want to see if it strikes any chord with some fans, or it just remains nebulous or unknown to most. I hope all the past results will help people trust my work behind it, and thus trust the future results/rankings.

Just had ear surgery, so I'll not be going to work for almost a week & a half. So, ear throbbing permitting, I'll be posting historical player rankings on my site back to the 1979-1980 season on this page:

http://hoopsnerd.com/?page_id=460

I will do seasons earlier than 1979-80 as well as all playoff seasons, but not until I'm fully done with my draft model & career projection model. I need all the regular season data from '80 on, since it includes all our modern box score stats (threes started in '79-80 in the NBA).

All my historical player ratings will adjust for pace & style, as well as playing time and team quality. My HnI rating (far right on every PDF) will be the go to rating for career projections. I can break that rating down into the very minutiae of statistical subsets AFTER pace/etc, and track how each subset trends from year to year based on age & quality (how good or poor the player is in each skillset).

For example, a young post player who rebounds & block shots a ton but doesn't do too much else won't see his career curve peak nearly as highly (due to rebounding & shot blocking remaining more steady year to year) as a similarly rated higher usage PG who is already showing a nice assist rate & scoring diversity. But, maybe that rebounder/shot blocker will project above average to a later age despite not peaking nearly as highly in his prime than the fancy PG.

I'll be including the last 18 seasons of college basketball player ratings as well on my site, but that probably won't be finished until early June. That'll obvioulsy be the basis for the draft model, in which I'll be able to project rookie NBA ratings, and then project their career based on that statistical breakdown projection & their age. So, I'll be ranking draftees by their projected rookie season, by their projected peak, and by their projected longevity. Can't wait.

Anyway, I'll update here when I add more historical seasons, hopefully I'll feel well enough to crank them out in a timely manner.

Best wishes for your recovery.

Crow wrote:Best wishes for your recovery.

Thanks Crow. I'm doing ok for now, a bit of swelling, some pain, and a TON of rushing blood noise going on in that left ear. I can't sit at the computer for very long yet.

Statman wrote:Just had ear surgery, so I'll not be going to work for almost a week & a half. So, ear throbbing permitting, I'll be posting historical player rankings on my site back to the 1979-1980 season on this page:

http://hoopsnerd.com/?page_id=460

I will do seasons earlier than 1979-80 as well as all playoff seasons, but not until I'm fully done with my draft model & career projection model. I need all the regular season data from '80 on, since it includes all our modern box score stats (threes started in '79-80 in the NBA).

All my historical player ratings will adjust for pace & style, as well as playing time and team quality. My HnI rating (far right on every PDF) will be the go to rating for career projections. I can break that rating down into the very minutiae of statistical subsets AFTER pace/etc, and track how each subset trends from year to year based on age & quality (how good or poor the player is in each skillset).

For example, a young post player who rebounds & block shots a ton but doesn't do too much else won't see his career curve peak nearly as highly (due to rebounding & shot blocking remaining more steady year to year) as a similarly rated higher usage PG who is already showing a nice assist rate & scoring diversity. But, maybe that rebounder/shot blocker will project above average to a later age despite not peaking nearly as highly in his prime than the fancy PG.

I'll be including the last 18 seasons of college basketball player ratings as well on my site, but that probably won't be finished until early June. That'll obvioulsy be the basis for the draft model, in which I'll be able to project rookie NBA ratings, and then project their career based on that statistical breakdown projection & their age. So, I'll be ranking draftees by their projected rookie season, by their projected peak, and by their projected longevity. Can't wait.

Anyway, I'll update here when I add more historical seasons, hopefully I'll feel well enough to crank them out in a timely manner.

I did finish ranking ALL the NBA players season by season since 1979-80, as well as compiling the career totals from that 35 time span.

Here's a little write up involving the 35 seasons of results: http://hoopsnerd.com/?p=542

And here are all the results: http://hoopsnerd.com/?page_id=460

Well, this is a nice pile of work. Good luck with adding playoffs and 'adjusting' for strength -- seriously.

At first glance, and upon closer study, I like your rankings by WAR better than b-r.com's Win Shares. But your career totals are about 30% higher, at least among the top careers.

I took the top WS careers since 1980 and entered your WAR for 73 players with >100 WAR.
Here are players more 'liked' by WAR, and more 'liked' by WS:I'm puzzled by a few things:
- I didn't come to any players with negative WAR. And the players above are mostly given higher WAR than WS. Do teams (and leagues) total more WAR than there are games played?
- The WAR/48 shown here are just WAR*48/Min. Your charts show other numbers under WAR/48, but I don't know what they are. 82 game * 48 mpg WAR equivalents?
- It looks like Kevin Durant this year has 27.45 WAR *48/3122Min = .422 WAR/48 . I thought his .295 WS/48 was pretty high -- as good as 3 average players?

Mike G wrote:Well, this is a nice pile of work. Good luck with adding playoffs and 'adjusting' for strength -- seriously.

At first glance, and upon closer study, I like your rankings by WAR better than b-r.com's Win Shares. But your career totals are about 30% higher, at least among the top careers.

Thanks Mike, I was really hoping someone like yourself would look deeper into the results, because I'm very proud with how they've been turning out (for both the NBA & college).

My WAR will add up to exactly match team wins (IF 82 games are played). The only caveat are strike seasons, I use 82 game projected totals to make the season carry the same weight as all other past seasons.

Yes, that means my replacement level is assumed 0 wins - and ANY smidge above replacement level begins the winning process.

Also, my WAR adjusts for changes in league statistical frequencies & pace. This will be more obvious w/ 60s rankings - my rankings, while center heavy (they dominated the game then), won't be nearly as center heavy as any other metric out there, since it'll adjust better to the plethora of rebounds and relative lack of assists (in relation to points scored). Bob Cousy, for example, should rate MUCH higher by my method than WS or PER (or, if you could imagine how bad it'd be if Berri did those seasons - Wins Produced, which would probably have Cousey WAY below league average).

Bill Russell may rate better too, since WAR is directly tied to team wins - but I still can't imagine his regular season WAR will be nearly that of Wilt or Kareem. BUT, his playoff adjusted WAR might only see Jordan as a comp.

Mike G wrote:I'm puzzled by a few things:
- I didn't come to any players with negative WAR. And the players above are mostly given higher WAR than WS. Do teams (and leagues) total more WAR than there are games played?
- The WAR/48 shown here are just WAR*48/Min. Your charts show other numbers under WAR/48, but I don't know what they are. 82 game * 48 mpg WAR equivalents?
- It looks like Kevin Durant this year has 27.45 WAR *48/3122Min = .422 WAR/48 . I thought his .295 WS/48 was pretty high -- as good as 3 average players?

Real quick, no negative WAR. Players rated at or below replacement level (80 HnI) will be 0. I think it's discussed some earlier in this thread. I don't think someone who has played in the NBA should ever be considered WORSE than guys that don't. Plus, I don't want players to FALL down a list of WAR rankings because they played poorly at the end of their careers, even though they were good enough to still be playing. So, lowest single season WAR is 0.

My WAR/48 is simply how much WAR the player would have if he played EVERY available minute for his team that season. I call it WAR/48 because I don't really know a better way to title it.

I think I remember you being shocked by LeBron's WAR from last season. This is wins above replacement, so it's how many more wins the player created for his team assuming his minutes were replaced solely by a replacement level player. If a player puts up superstar stats relative to his teammates - and his team wins a ton - his WAR will be very large. OKC obviously wouldn't be 27.5 games worse w/o Durant, since at the very least they'd find a way to replace his 3122 minutes with at least something closer to an average player. Maybe the team is 20 or 21 games worse w/o Durant, real world? Maybe that's the surplus of wins you are seeing among star players relative to WS?

Mike G wrote: I took the top WS careers since 1980 and entered your WAR for 73 players with >100 WAR.
Here are players more 'liked' by WAR, and more 'liked' by WS:

Code: Select all

ws/war   WS likes      WS/48  war48     ws/war   WAR likes    WS/48  war48
1.154  Horace Grant    .147   .127      .567   Allen Iverson   .126   .223
1.095  Buck Williams   .136   .124      .580   Hakeem Olajuwon .177   .305
1.079  Detlef Schrempf .156   .145      .582   Chris Webber    .132   .227
1.067  Reggie Miller   .176   .165      .588   Patrick Ewing   .150   .254
1.042  Jeff Hornacek   .154   .148      .607  Shaquille O'Neal .208   .343
.975   Eddie Jones     .147   .151      .620   Isiah Thomas    .109   .176
.972   Terry Porter    .150   .154      .629   LeBron James    .243   .386
.958  Chauncey Billups .176   .183      .647   Dwyane Wade     .192   .297
.954   Dikembe Mutombo .153   .160      .654   Larry Bird      .203   .311
.925   Adrian Dantley  .192   .208      .656   Tim Duncan      .211   .322
.916   Kevin McHale    .180   .197      .662   Tracy McGrady   .152   .230
.912   Manu Ginobili   .209   .229      .676   Carmelo Anthony .138   .205
.894   Jason Terry     .122   .137      .678   Kevin Garnett   .185   .273
.892   Chris Bosh      .161   .181      .681   Jason Kidd      .133   .195
.891   Ray Allen       .150   .169      .682   Kobe Bryant     .182   .267
.887   Larry Nance     .171   .193      .693   Michael Jordan  .250   .362
.882   Rasheed Wallace .139   .158      .694   Tim Hardaway    .133   .192
.879   Jack Sikma      .149   .169      .694   Alonzo Mourning .166   .239
.875   Robert Parish   .157   .180      .697   Karl Malone     .205   .295
.862   Andre Miller    .122   .141      .699   Dwight Howard   .179   .257

Also, I grew up in Michigan a Piston fan. It has driven me crazy how the known metrics have undervalued Isiah Thomas in general. He was the star player, the ring leader, on a strong 5 year run and 2 championships. Now, I don't believe at all he's anywhere near Bird/Magic/Jordan et al, but WS has him barely above an average player. Notice WS rates EVERY guy you list on the left (even Jason Terry, Andre Miller, & Buck Williams) much higher than Isiah per minute.

WS also has all those guys on the left except Terry & Miller above Kidd also. Jason Kidd did EVERYTHING so well, he just didn't score much.

WAR only likes Ginobli, Dantley, McHale, Nance, Billups, Bosh, & Parish more than Isiah. Only Ginobli, Dantley, McHale above Kidd. That just seems much more realistic to me.