Replacement level in the NBA

[colts18]

What would you gauge as the replacement level to judge NBA players? I'm trying to calculate RAPM over replacement level but I still haven't figured out a good replacement level value. I'm think -2 would be good since that is the equivalent of a -10 team. Very few teams are worse than that.

[Crow]

-3 seems to get used a lot.

[colts18]

-3 seems to get used a lot.

Shouldn't it be higher because stars in the NBA don't have minutes replaced by the 12th-15th man on the bench? In baseball, when the starter goes out, a bench player who is close to replacement level replaces him. In the NBA, the #2-#9 guys get those minutes spread out to them.

[Statman]

-3 seems to get used a lot.

Shouldn't it be higher because stars in the NBA don't have minutes replaced by the 12th-15th man on the bench? In baseball, when the starter goes out, a bench player who is close to replacement level replaces him. In the NBA, the #2-#9 guys get those minutes spread out to them.

There's a difference between bench player and replacement player imo.

Replacement player means the level of guy that could be easily found for a 10 day contract - replaces a roster spot.

Honestly, there are hundreds of guys that probably aren't too far from each other in overall talent from about the 9th spot and well beyond on every roster that are all replacement playerish. Some are just lucky enough (or good enough earlier in their careers) to have guaranteed contracts to lock up a spot - but aren't really any better than replacement player.

The situation you are referring to would be value over average player - or value over quite close to average player. Many don't do this because it ends up making big minute lowish production guys who are non box score types (Bruce Bowen, Shane Battier, etc.) look like the very worst players in the NBA. I think RP levels are set fairly low partly as a nod to understanding there is a limitation to linear weights systems based on box score stats - and playing time obviously does matter to some extent in maybe identifying players who add value beyond the box score.

[DSMok1]

Replacement level is not related to minutes played. It is related to replace-ability of the player on the open market for minimum salary. If you can't get a similar player for minimum, he is by definition above replacement level, and thus worth more than replacement level.

Here is a visualization I have created (but hitherto not published) for dynamically generating replacement level estimates based on parameters you define:

http://public.tableausoftware.com/views ... l?:embed=y

Based on some thought on the matter, I systematically exclude rookies & second year players from my estimates, because they are getting played minutes in the hopes that they will develop, not based on whether they are above replacement level or not.

[colts18]

Replacement level is not related to minutes played. It is related to replace-ability of the player on the open market for minimum salary. If you can't get a similar player for minimum, he is by definition above replacement level, and thus worth more than replacement level.

Here is a visualization I have created (but hitherto not published) for dynamically generating replacement level estimates based on parameters you define:

http://public.tableausoftware.com/views ... l?:embed=y

Based on some thought on the matter, I systematically exclude rookies & second year players from my estimates, because they are getting played minutes in the hopes that they will develop, not based on whether they are above replacement level or not.

What kind of player do you see as a -1 player? What about -2 player? I'm trying to figure out the right baseline. Maybe a -1 player represents the average 7th man, while -2 is average 10th. Maybe I should compare a player to his bench replacement, not the replacement level since the 13th man on the bench doesn't get to play.

[Jacob Frankel]

Is there any consensus on whether there are different replacement levels for different positions? I know Hollinger uses different replacement levels. I haven't looked at the data, but I'd imagine there is. Anybody else looked at this/use different replacement levels?

(Side note that doesn't really belong in this thread: wanted to link to this awesome piece on Usage/Efficiency that uses the methodology Eli Witus used with a few changes and 5 years of data: http://ascreamingcomesacrossthecourt.bl ... iency.html)

[Dr Positivity]

The number you are looking for is 0.5 Ws. Coming to this in a few ways:

- The average team salary is 67 million, which divided by 41 Ws for an average team, leads to a number of about 1.6 million per win on average. I then took the average individual salary of about 5.1 million and divided it by 1.6, the resulting number states the average player is worth about 3.2 wins. Now we can divide the average player salary of 5.1 million by an estimate of 800k, around what I'd expect a "replacement player" to deserve salary wise. That replacement salary of 800k is just under 16% of the average salary of 5.1 million. Multiplying the average individual wins of 3.2 by .16 = 0.51, or just over 0.5 wins for a replacement player.

- It's a fair guesstimate that a team full of replacement players wouldn't fall below 5-7 Ws over an 82 game season at least by what their point differential indicates they deserve. Clearly a regular sized roster of players at a 0.5 W rate, will end up around there. Using an estimate of 13 players per team (67 million / 5.1 million = just over 13), 13 players at an average of 0.5 wins is 6.5 Ws. That sounds about correct. Furthermore using the above salary numbers of 800k/5.1 million to state a replacement player is worth 15-16% of an average player, we can multiply .16 by the 41 wins number for an average team. That punches out at 6.6 wins rounded up, almost exactly where the 0.5 W and 13 player estimate pegged it.

[AcrossTheCourt]

Is there any consensus on whether there are different replacement levels for different positions? I know Hollinger uses different replacement levels. I haven't looked at the data, but I'd imagine there is. Anybody else looked at this/use different replacement levels?

(Side note that doesn't really belong in this thread: wanted to link to this awesome piece on Usage/Efficiency that uses the methodology Eli Witus used with a few changes and 5 years of data: http://ascreamingcomesacrossthecourt.bl ... iency.html)

Thanks. That's mine.... Some time I'll do an updated version with some tweaks like focusing on specific players (Carmelo, anyone?) Also need to do the lineup's opposing defense.

On a related note, there were some players who had no offensive ratings because they had no baskets. So I needed to do a replacement level score and used the 10th percentile for Ortg, Usage, etc. This only happened with a small handful of players and didn't have too much bearing on the results, but I thought it was an appropriate estimate.

[v-zero]

The number you are looking for is 0.5 Ws.

That's equivalent to a -3 player in +/- (per 100) terms, which is basically what everybody and their dog has more or less settled on with an error of about 0.5 or so.

[DSMok1]

The number you are looking for is 0.5 Ws.

That's equivalent to a -3 player in +/- (per 100) terms, which is basically what everybody and their dog has more or less settled on with an error of about 0.5 or so.

Based on my research, I think that might be a little too low--people at that level play, but usually only if they are rookies or 2nd year players that teams are trying to develop.

Of course, it all depends on the spread of your statistic. RAPM tends to be regressed such that the spread isn't as wide as, say, ASPM. For ASPM, I'm using something around -2.35 (I used to use much lower before I did the research posted above).

[Mike G]

What would you gauge as the replacement level to judge NBA players? I'm trying to calculate RAPM over replacement level but I still haven't figured out a good replacement level value. I'm think -2 would be good since that is the equivalent of a -10 team. Very few teams are worse than that.

The Bobcats last year were one of the worst teams ever, about -14 per game, and they had some players who were considerably better than replacements.

Actually their ORtg-DRtg = -15.2
So I'd say -3 is a bit too high to describe players who are available to replace those on any roster.

I'd also discourage the idea that replacement level varies from team to team, or by situation. If your 9th guy goes down, who takes his minutes? If it's an important game, most likely your 1 thru 8 guys get more minutes. Are they therefore 'replacement level'?