Calculating Championship Odds Based on SRS

ElGee · Post #1 » by **ElGee** » Sun Jul 22, 2012 7:11 am

4/15/13 Update: More data, incorporation of PS data

The goal:

Spoiler:

The Method:

Spoiler:

The Results

Spoiler:

Discussion

Spoiler:

mysticbb · Post #2 » by **mysticbb** » Sun Jul 22, 2012 9:30 am

Couple of notes:

1. The formula for the portabilities is mixed up. You need to switch the one for high and low players.
2. Choosing the point of orgin as 0 SRS on your scale gives you two pretty weird things. First, that means a low, normal and high portability player will have the same effect on an average SRS team. You can use that as a some sort of definition, but I don't think that is a good definition. Players show dimishing returns based on the skillset even with cast based on average players. Second, when you go to a negative SRS team, you will end up saying that a high portability player is actually a worse choice than the low portability player with the same SIO. I doubt that this is something you agree on.
I suggest using a different point of origin here. During the 3pt era the 34 worst teams had an average of -9.7 SRS, and the 34 best teams in average +8.2 SRS. Well, even though that means we don't really have a linear scale, we can still use one. Put the starting point 0 at -9.7 SRS and then raise it to 17.9 for the 8.2 SRS team. Well, you would need a new base (something like 1.1 instead of 1.5) to make it work.
In that way you get a different overall impact for different portability players on average teams while not having the problem with the below average teams.
3. The Bulls in 1986 were an outlier, a 30 win team making the playoffs due to circumstances. Also, without Jordan that team would have missed the playoffs.
4. Longevity vs. Peak. Does your conclusion mean that you agree with my assessment that 12 years of Nowitzki are worth more than 8 years of James? ;)

ElGee · Post #3 » by **ElGee** » Sun Jul 22, 2012 3:24 pm

1. Thanks

2. Totally fair point.

First, let me say that I'm aware of the negatives and thought of having all players use the standard portability on negative teams to avoid this problem (although I could even see the argument that good and bad should flip). In general though, the impact on negative had such little bearing on championships it seemed borderline irrelevant. Which leads to your other point about point of origin -- it's a good idea, although I'm not sure your concern is as common as you think.

Using 0 SRS is easy to conceptualize, and it's also right near the median for team distributions. If you play around with, it seems pretty fair -- especially concerning teams that are above average.

3. I suppose using the 86 Bulls and "average" in the same sentence was a misstep.

4. Yes. If we exclude 2012, I'd take Dirk now.

Post #4 » by **Doctor MJ** » Sun Jul 22, 2012 7:54 pm

Great stuff. As we do the Peaks project, I'll be assembling my subjective ratings, and look forward to see how they come out when sent through this process.

Post #5 » by **mopper8** » Sun Jul 22, 2012 8:21 pm

mysticbb wrote:4. Longevity vs. Peak. Does your conclusion mean that you agree with my assessment that 12 years of Nowitzki are worth more than 8 years of James?

I think his conclusion has to be, roughly speaking, that any top 10 player all-time for say 5 years is a worse bet than players, say, 15-25 for 10 years. No?

mysticbb · Post #6 » by **mysticbb** » Mon Jul 23, 2012 7:03 am

ElGee wrote:(although I could even see the argument that good and bad should flip).

That would depend on the definition of "portability". For me a player which has his main strength in skills which are rather often avaiable (like a small player who can handle the ball and can shoot, there are tons of those players), is not very portable. But then again, it also depends on how SIO is defined. In such a case level over replacement sounds a more reasonable choice. That's also how I calculate PW48, when I make the fit, I come up with -2.9 as replacement level, which would equate to a -14.5 per 100 possession team. Looking at the team results, -2.2 might be a better "choice", well ...
That might be a good starting point for your SIO here. If you have that defined as level over average, adding something between 2.2 and 2.9 might be reasonable when you start putting the 0 at the -10 team level.

ElGee wrote:In general though, the impact on negative had such little bearing on championships it seemed borderline irrelevant.

That was my first thought as well, but it is relevant, because you get the same values for each of the different portability players, if you put them on an average team. I just don't see that this is true.

ElGee wrote:Which leads to your other point about point of origin -- it's a good idea, although I'm not sure your concern is as common as you think.

From 2001 to 2012 the average championship team was -2.3 with the "best" player off, while being +10.7 with the "best" player on. I think it is really common. Because when we transport your method into the reality we see that a low portability player is in average more valuable for winning a championship than a high portability player. I don't think that this is true.

ElGee wrote:Using 0 SRS is easy to conceptualize, and it's also right near the median for team distributions.

Completely agree. It is easy to apply, that's why it is pretty common to set things to the league average. It is just in conflict with the used power law in your method in order to determine impact on different strong teams.

ElGee wrote:4. Yes. If we exclude 2012, I'd take Dirk now.

Why would you exclude 2012? 2005 to 2012 for James vs. 2001 to 2012 for Nowitzki makes 8 seasons for James vs. 12 seasons for Nowitzki, which equates overall to a higher amount of "championship odds" for Nowitzki. The difference in performance level between those two is just too small in average in order to come up with a different conclusion.

mysticbb · Post #7 » by **mysticbb** » Mon Jul 23, 2012 7:14 am

mopper8 wrote:I think his conclusion has to be, roughly speaking, that any top 10 player all-time for say 5 years is a worse bet than players, say, 15-25 for 10 years. No?

Yes, that's how it looks like. But well, I would put those "Top 15-25" player ahead of the "Top 10" player in such a case.
It might also be not as easy to see the longevity for different players, when longevity is not just the amount of years those players were in the league, but rather some sort of "All-NBA-like" seasons, a season in which the player can be the cornerstone (clearly best player) of a contending team. And when I look at a player like Moses Malone, for example, I have a hard time following the notion that he had such great longevity. Well, which has also something to do with the problem of converting raw per game numbers into real impact ...

ElGee · Post #8 » by **ElGee** » Mon Jul 23, 2012 4:53 pm

mysticbb wrote:
ElGee wrote:(although I could even see the argument that good and bad should flip).

That would depend on the definition of "portability". For me a player which has his main strength in skills which are rather often avaiable (like a small player who can handle the ball and can shoot, there are tons of those players), is not very portable. But then again, it also depends on how SIO is defined. In such a case level over replacement sounds a more reasonable choice. That's also how I calculate PW48, when I make the fit, I come up with -2.9 as replacement level, which would equate to a -14.5 per 100 possession team. Looking at the team results, -2.2 might be a better "choice", well ...
That might be a good starting point for your SIO here. If you have that defined as level over average, adding something between 2.2 and 2.9 might be reasonable when you start putting the 0 at the -10 team level.

Very good points. As you know, value over replacement is a grand question in basketball statistics and not easily handled. Something closer to -2 might be a better "base" point for a future iteration.

ElGee wrote:In general though, the impact on negative had such little bearing on championships it seemed borderline irrelevant.

That was my first thought as well, but it is relevant, because you get the same values for each of the different portability players, if you put them on an average team. I just don't see that this is true.

I see I did use "normal" portability below 0 for bad port players -- wasn't sure last time. For some perspective, the good portability players have their title odds changed by a few tenths of a percent at the high end, and once we get to a +5 SIO player there is no difference.

ElGee wrote:Which leads to your other point about point of origin -- it's a good idea, although I'm not sure your concern is as common as you think.

From 2001 to 2012 the average championship team was -2.3 with the "best" player off, while being +10.7 with the "best" player on. I think it is really common. Because when we transport your method into the reality we see that a low portability player is in average more valuable for winning a championship than a high portability player. I don't think that this is true.

I don't using raw on/off for all the reasons that confound the stat. Good teams will have good players playing together. They will then rest together, especially during blowouts. This alone will stretch the net number.

Consider that a title team plays in about 4 more blowouts per year than the average team in the league, and they often rest players for another game or two at the end of the season. That alone can create maybe ~100 of the 800 minutes the "best" player will be off the court. In those 100 "garbage time" minutes the title team would need it's scrubs to be outscored by 38 points to actually be 0 for the rest of the year...and that still doesn't include lineup combinations. So I don't think that's a very accurate indicator.

Post #9 » by **Doctor MJ** » Wed Jul 25, 2012 12:35 am

fwiw, I like the idea of using essentially the worst imaginable actual team as a baseline for replacement players. So that's about -15 SRS and thus -3 per player.

mysticbb · Post #10 » by **mysticbb** » Wed Jul 25, 2012 6:15 am

ElGee wrote:So I don't think that's a very accurate indicator.

But your extreme example is? And also, if your idea would be true, the NET for the starters would be essentially the same for those teams, but it is not the case. It is obious that the difference is not just created in garbage time. It also could very well end up the other way around, and thus lowering the NET, both things can be rationalized.

Matter of fact is that the championship teams showed a big difference between the time with and without their respective best player. And given the structure of your method you will run into a problem, if you use league average as the starting point.

ElGee · Post #11 » by **ElGee** » Sat Jul 28, 2012 1:26 am

Doctor MJ wrote:fwiw, I like the idea of using essentially the worst imaginable actual team as a baseline for replacement players. So that's about -15 SRS and thus -3 per player.

For my calculations, I'm using -2. I don't know the "right" answer here, but given that the majority of the work was calculating teams between -8 and +8 SRS (most teams fall within this range), and that there have been nine -11 teams in NBA history, and even this year's Bobcats didn't reach the -15 mark, I'm not too worried about looking at VORP at -2 instead of -3. Using Normal Portability (which assumes a slightly smaller dent the better the team), which seems reasonable to me, the different between -2 and -3 is 0.65% (1.85% down to 1.20%) championship odds. At that point, it's barely shifting players more than a spot or two on my own list.

SideshowBob · Post #12 » by **SideshowBob** » Thu May 16, 2013 6:26 am

What happened to the odds for Low/High players?

EDIT: I was able to find them through the Internet Archive, but I assume that you've updated your research on those as well.

SideshowBob · Post #13 » by **SideshowBob** » Tue Jun 4, 2013 6:26 pm

Also, what would be the odds for a normal +10.5 and +11.0 player? I'd guess about +4-4.5% for each 0.5 increment, but I figure the jump may be even steeper at that level

ElGee · Post #14 » by **ElGee** » Wed Apr 23, 2014 2:34 am

Updated Tables for Portability and Health:

ElGee · Post #15 » by **ElGee** » Wed Apr 23, 2014 2:38 am

Hmm -- I think it cut the 100% healthy line. :/

Normal Port:
1.1%
1.2%
1.6%
2.2%
2.2%
3.7%
3.7%
3.8%
6.1%
6.1%
8.4%
9.7%
11.1%
14.0%
15.3%
18.4%
21.4%
24.8%
27.2%
32.0%
35.7%
41.1%
44.8%
49.3%
54.1%
59.4%
63.2%

High Port:
0.9%
1.2%
1.2%
2.2%
2.2%
3.7%
3.7%
3.8%
6.1%
6.1%
9.4%
9.7%
12.8%
14.2%
17.2%
19.6%
22.6%
25.8%
28.8%
33.0%
36.9%
41.3%
44.8%
50.0%
51.8%
57.1%
60.0%

Low Port
1.4%
1.6%
2.1%
2.2%
2.5%
3.7%
3.7%
3.8%
5.7%
6.1%
7.3%
8.7%
10.4%
12.5%
14.2%
17.2%
19.5%
22.6%
26.0%
30.4%
34.9%
39.2%
43.5%
48.3%
52.8%
59.5%
64.5%

All of those are -3 to +10 with 1/2 point increments.

ElGee · Post #16 » by **ElGee** » Wed Oct 8, 2014 5:18 am

So I've finally incorporated what I always considered to be the elephant in the room with this model, which is essentially salary. (Although ego plays a part.)

In the previous model, every player had the same odds of being on the same team. In other words, it assumed that the likelihood that a superstar landed on a +8 team was the same likelihood that a +1 player would land on such a team. (This was done for simplicity.) In reality, this is not the case...I think. I did not have the time to use any real salary data, and perhaps won't be too confident until I do so, but I changed the model to examine what would happen if the odds became increasingly less likely that the better a player become, the lower his odds of a playing on a good team would become.

To put in perspective the adjustment I basically "made up" without having the official salary data, the odds of being on a positive SRS team are about 50%. I shifted those all the way up to 65% for weak players and all the way down to 35% for elite players; ultimately, at the edges, I made it 2x less likely than the average and over 3x less likely than a mediocre player to land on a great team.

The actual use-cases for this kind of adjustment are as follows: Imagine a player who thinks he THE MAN and also clearly commands a max salary on the open market. The odds of this player landing on an elite team in free agency are very small, and in fact, only Wilt (69), Moses (83) and LeBron (11) exist as counterexamples. Players can be traded, but rarely are trades so lopsided (and involving a max-contract player!) that the team landing the superstar would be a title contender without him. Finally, teams can land players through the draft, and while occasionally a super-elite young player might develop on a great team (e.g. 98 Duncan), there are still forces that prevent this happening (namely the lottery).

All told, whether I have the distribution odds correct, here are the results for a "normal portability" player with full health:

9 44.0%
8.5 38.7%
8 33.8%
7.5 29.5%
7 25.6%
6.5 22.1%
6 19.1%
5.5 16.6%
5 14.6%
4.5 13.0%
4 11.9%
3.5 11.2%
3 11.0%
2.5 10.1%
2 8.7%
1.5 7.5%
1 6.3%
0.5 5.3%
0 4.4%
-0.5 3.6%
-1 2.9%
-1.5 2.3%
-2 1.8%
-2.5 1.5%
-3 1.2%

Portability
Adjusting for portability can be thought of as a +/- 1% adjustment up to about +1 players, +/- 2% up to about +4 players, and up to +/- 3% as players approach +9.

Health
Adjusting for health can be though of as ~50% value lost if a player plays ~10% of the season and still plays in the playoffs. This drops to ~30% value loss at 40% of a full season played, and over 50% loses no more than 20% value typically. 80% of the season played carries ~90% of a players value.

Comparing Lists
So now I have a career odds calculator that takes into things that might cause on uneven distribution of odds to land on certain teams, and another that assumes all players have the same odds of just randomly plopping down on a team, no matter how good. What does that do to career value?

The "salary" list now becomes incredibly longevity heavy. The following players moved up the most in my top 50:
Allen (+12 spots)
Parish (+11)
Havlicek (+11)
Hayes (+10)
Kidd (+8)
Stockton (+8)
Miller (+7)

These are longevity-driven players, all without elite peaks (I'm not sure any of those peaks are in my top-50.) On the other hand, all the high-peak, low-longevity players dropped:
Paul (-16)
Frazier (-15)
Mourning (-11)
McGrady (-11)

colts18 · Post #17 » by **colts18** » Fri Oct 10, 2014 4:56 am

Elgee, do you have a career top 50 list that you can post?

ElGee · Post #18 » by **ElGee** » Fri Oct 10, 2014 12:52 pm

colts18 wrote:Elgee, do you have a career top 50 list that you can post?

Using which model?

colts18 · Post #19 » by **colts18** » Fri Oct 10, 2014 3:26 pm

ElGee wrote:
colts18 wrote:Elgee, do you have a career top 50 list that you can post?

Using which model?

The new one. Or the old one if you still have it

RSCD3_ · Post #20 » by **RSCD3_** » Fri Oct 10, 2014 8:26 pm

colts18 wrote:
ElGee wrote:
colts18 wrote:Elgee, do you have a career top 50 list that you can post?

Using which model?

The new one. Or the old one if you still have it

Also would like to see it

Sent from my iPhone using RealGM Forums