Why Stats are Good, Revisited (257 views)

[eric]

In a previous thread I looked up how many top five finishes players had in the regular and post seasons in the Win Shares stat. I will now repeat that analysis with Wins Produced, VORP, and Estimated Wins Added, all stats courtesy of b-r. Note that VORP can only be calculated until 1974, and WP can only be calculated until 1978. Let us therefore focus on people from that era first:

ws wp vorp ewa name
6 3 3 9 Shaquille O'Neal
7 3 4 5 Tim Duncan
8 2 6 11 Kareem Abdul-Jabbar
5 5 7 6 Larry Bird
7 6 10 9 LeBron James
8 10 10 6 Magic Johnson
8 7 9 9 Michael Jordan
These are the only players with at least three in every category. (Kareem only has two in Wins Produced, but given that he had a Finals MVP in '71, a sensational Finals run in '74, and five MVP regular seasons in the time before Wins Produced, I consider it safe to assume he would have gotten at least one more top five finish in each category.) Only one player has even two in every category, which I think is a pretty convenient dividing line. Also worth mentioning is that for Julius Erving's NBA career alone he rated 3 1 4 4.

For older people we can only look at WS and EWA, so let's look at people with at least four in each:

ws wp vorp ewa name
5 ? ? 6 Bob Pettit
9 ? ? 4 Bill Russell
10 ? ? 10 Wilt Chamberlain
6 ? ? 8 Jerry West
Another pretty solid list. Because Wins Produced obsessively favors rebounds, I think we can safely assume Russell and Wilt would have dominated that stat. Thus, for a list of all time greats this method indicates in chronological order:

Bill Russell
Wilt Chamberlain
Kareem Abdul-Jabbar
Larry Bird (5)
Magic Johnson (6)
Michael Jordan (7)
LeBron James (6)

As magnificent a seven as you could hope to see, with a second tier of Pettit/West/Erving/Shaq/Duncan. Also note that the method declares Michael Jordan the greatest player of the measurable era, and that Kareem could surpass him given that he had six playoffs before WP but he was quite poor in 73 and 72 so it's doubtful he'd even tie.

Then there is Moses with exactly 2 in every category and thirteen guys with at least 1 from the measurable era:

ws wp vorp ewa name
1 1 4 1 Clyde Drexler
1 3 2 4 Charles Barkley
2 1 2 6 Hakeem Olajuwon
4 1 3 6 Karl Malone
1 1 5 1 Scottie Pippen
1 1 2 2 Kevin Garnett
2 1 2 1 Pau Gasol
1 1 1 1 Dwight Howard
2 1 3 4 Dwyane Wade
1 1 1 2 Chris Paul
2 1 3 4 Kevin Durant
1 1 1 1 Steph Curry
2 1 2 1 James Harden
Notable by his absence: Kobe Jellybean Bryant. Lest you think the top five requirement was chosen unfairly, note that Kobe is held back by never even finishing in the top ten for regular season Wins Produced.

[eric]

So we know there are 21 players to have at least one top five finish in the regular and post seasons in all four stats, but are there any players to lead outright across the board in a single year? As it turns out, yes!

We start with Win Shares because it goes the furthest back and we get 22 players to complete the double: six Jordan, four LeBron, three Kareem, two Wilt, then one each by seven other players. Estimated Wins Added goes back to 1956 and cuts the field down to 11 player-seasons, including only one from before 1978, which is nice because VORP and Wins Produced don't go back that far. For the period they do cover we are left with only three seasons: 2012 LeBron, 2000 Shaq, 1989 Jordan, and the prehistoric year is 1970 West. Let's look at what happened besides stats in those years:

2012 LeBron won MVP, Finals MVP, and led the Heat to the #2 seed in the East and the title. Pretty solid year, but nothing historic.

2000 Shaq won MVP and was one vote shy of unanimous, Finals MVP, All Star MVP (tied), and led the Lakers to the #1 overall seed and the title. Their record of 67 wins remains the highest since the Bulls (at least for a couple more months). Finished second in DPOY voting for good measure.

1989 Jordan didn't win jack. After a historic MVP as the #3 seed the year before, his Bulls slumped to the #6 seed and he finished a close but not that close second to Magic Johnson in the MVP voting. Also failed to defend his DPOY title with a fifth place finish. He wildly overachieved in the playoffs but still only reached the Conference Finals.

1970 West kept the Lakers afloat despite regular season injuries to Baylor and Chamberlain and reached the Finals where he lost for the seventh time in a row. Finished a close second to Willis Reed in MVP, though the Knicks vote was slightly split to Frazier. Probably would not have led the playoffs in Wins Produced seeing as how Wilt averaged a 22/22 on 55% shooting.

All players were All-NBA 1st team and All-Stars. All players except Shaq were All-Defense 1st team (he was 2nd).

.

For me it has to be 2000 Shaq as the best year ever. West and Jordan didn't get the ring, LeBron didn't have the domination (and/or lack of competition), nobody else has the stats.

[eric]

What about ESPN's RPM? There are two big issues with it:

1. Its methodology is not publicly available. This is a Problem.
2. ESPN only gives values since the 2014 regular season. This is a Bigger Problem.

However, we can take those values and stack them up against Win Shares (c/o b-r) to see how well they match the wins the teams generated in those two seasons. There are three additional but small issues:
-ESPN does not split player values up if they were on multiple teams, it just gives the total RPM value. I have addressed these players by splitting their RPM Wins up by that percentage of their minutes they played with each team.
-ESPN sometimes just doesn't have players on their website. Lou Amundson played 185 minutes in 2014, but he is just not on the ESPN RPM page for 2014. It's not a MP requirement thing either because JaVale McGee is on there with only 90 minutes. The only players I found this happen for him and Chris Johnson 2015 at 526 minutes, and both would be pretty close to 0 wins so it wouldn't make much difference if any. I would calculate their RPM manually but I can't: see point (1).
-ESPN gives a value for "wins" on their site that is actually "wins over replacement", and they at no point specify what replacement level is. I am going to use a replacement level value that minimizes the root mean squared error for this sample and thereby make the best possible case for RPM, possibly even better than the actual model.

Here are the results:
actual team rpm ws
60 ATL 43.3 54.9
38 BKN 19.5 32.0
40 BOS 26.1 34.0
33 CHA 17.5 24.6
50 CHI 39.4 50.5
53 CLE 44.3 53.0
50 DAL 37.5 45.5
30 DEN 18.1 32.3
32 DET 25.4 39.0
67 GS 63.4 66.8
56 HOU 41.3 50.9
38 IND 26.1 42.4
56 LAC 51.7 58.7
21 LAL 8.3 23.4
55 MEM 38.6 49.7
37 MIA 20.1 33.9
41 MIL 30.2 43.8
16 MIN 1.1 17.0
45 NO 34.0 43.4
17 NY -1.8 17.5
45 OKC 37.5 47.2
25 ORL 11.0 27.2
18 PHI 0.7 18.8
39 PHX 30.2 36.9
51 POR 40.5 50.3
55 SA 49.3 57.6
29 SAC 19.7 32.8
49 TOR 35.6 44.0
38 UTAH 28.3 44.2
46 WSH 33.0 45.4

38 ATL 29.0 39.9
44 BKN 34.9 38.6
25 BOS 18.5 30.2
43 CHA 22.4 41.6
48 CHI 33.1 48.2
33 CLE 22.2 34.1
49 DAL 42.5 48.3
36 DEN 21.4 36.5
29 DET 25.1 32.0
51 GS 48.9 53.9
54 HOU 45.7 54.7
56 IND 36.4 53.6
57 LAC 50.4 59.1
27 LAL 11.0 24.2
50 MEM 33.1 46.1
54 MIA 46.3 54.6
15 MIL 4.0 20.8
40 MIN 44.6 47.0
34 NO 16.3 36.0
37 NY 28.0 39.3
59 OKC 53.1 59.4
23 ORL 5.6 28.2
19 PHI -6.0 15.7
48 PHX 43.2 48.7
54 POR 36.3 51.4
62 SA 55.7 60.0
28 SAC 12.2 34.9
48 TOR 27.1 51.9
25 UTAH 7.4 21.9
44 WSH 32.6 45.3
With which we get RMSE of 5.43 for RPM (replacement level 11.8 wins) and 3.61 for WS. Clearly WS is a far superior metric with just under two thirds of the error. It is possible this happens to be a particularly good sample for WS and particularly bad for RPM, but I actually found a lower (better) RMSE for WS in the 2014-2004 sample so if anything it looks like this was a tough stretch for WS. I would check more for RPM but again, that is simply impossible for me to do.

In conclusion: RPM does not do a good job of describing what happened in a season in terms of how much a player did to help his team win. Some people have argued that it isn't designed to, and that ESPN is Shaqtin a fool by having Wins on their RPM leaderboards. I can't speak to that either way.

[eric]

It's interesting to think about where Curry and Anthony Davis will end up on this list, and to do that we need a rough estimate of when players stop making top five finishes. Let's look at Win Shares and see when every player's last top five was, given as year of career regular season / postseason.

LeBron, Curry, Harden, Durant, Paul are conceivably TBD
Wade 7/9
Howard 7/5
Gasol 10/9
Duncan 10/17
Garnett 11/13
Shaq 9/11
Pippen 10/11
Karl 16/13
Jordan 13/13
Olajuwon 10/13
Barkley 9/9
Drexler 5/11
Bird 9/8
Magic 12/12
Moses 8/6
Kareem 12/15

It's interesting to see how the ability to stay relevant in the playoffs leans towards staying relevant longer there but not that dramatically: 8 were more, 3 the same, 5 less. This makes sense, older players sit out more and simply play less hard in the regular season but can turn it on for the playoffs.

It's also interesting to see how soon in some players' careers they stopped being elite players. Dwight Howard's last relevant year by this analysis was 2011, and I don't think anyone would have guessed at that point the trajectory his career would take from there. It's easy to say "well Wade had knee problems, Bird had back problems, etc." but Curry's got ankle problems, Davis has never played 70 games in a season. Everybody's got problems. It's only obvious in retrospect that a guy was going to fall apart.

Anyway, Curry is through six years with exactly one in every category. He needs to get to five in every category to tie with Bird, and eleven productive years is the midpoint for the above list (9 guys made it, 7 guys didn't). That means four out of five in every category, regular and postseason, which definitely has precedent but obviously is a circular question: the guys who did it are the guys he's chasing. TBD

[eric]

The Pelicans will probably not make the playoffs this year, and will certainly not get far if they do.

This is Anthony Davis' fourth year. He has one top five regular season finish in Win Shares and two in Estimated Wins Added, none in any stat for the playoffs. Looking at the twenty players to finish with at least one top five in all (and play no part of their career in the ABA), here are the years in their careers they got at least one for a single stat in the regular and post season:

ws wp vorp ewa earliest name
9 2 9 3 2 Charles Barkley
3 3 3 3 3 Chris Paul
7 7 7 7 7 Clyde Drexler
5 5 6 5 5 Dwight Howard
6 6 3 3 3 Dwyane Wade
9 9 9 2 2 Hakeem Olajuwon
4 6 4 6 4 James Harden
1 x x 1 1 Kareem Abdul-Jabbar
7 9 9 7 7 Karl Malone
4 6 4 4 4 Kevin Durant
13 13 9 9 9 Kevin Garnett
4 3 2 4 2 Larry Bird
4 4 3 4 3 LeBron James
3 1 1 3 1 Magic Johnson
5 5 4 4 4 Michael Jordan
8 10 8 10 8 Pau Gasol
10 8 5 9 5 Scottie Pippen
3 8 3 3 3 Shaquille O'Neal
6 6 6 6 6 Steph Curry
2 2 2 2 2 Tim Duncan

ws wp vorp ewa earliest name
1 x x 1 1 Kareem Abdul-Jabbar
3 1 1 3 1 Magic Johnson
9 2 9 3 2 Charles Barkley
9 9 9 2 2 Hakeem Olajuwon
4 3 2 4 2 Larry Bird
2 2 2 2 2 Tim Duncan
3 3 3 3 3 Chris Paul
6 6 3 3 3 Dwyane Wade
4 4 3 4 3 LeBron James
3 8 3 3 3 Shaquille O'Neal
4 6 4 6 4 James Harden
4 6 4 4 4 Kevin Durant
5 5 4 4 4 Michael Jordan

5 5 6 5 5 Dwight Howard
10 8 5 9 5 Scottie Pippen
6 6 6 6 6 Steph Curry
7 7 7 7 7 Clyde Drexler
7 9 9 7 7 Karl Malone
8 10 8 10 8 Pau Gasol
13 13 9 9 9 Kevin Garnett
While there are some 4- guys who don't belong, I don't see anyone in the 5+ who can make a case of being snubbed (necessary but not sufficient). Everybody in the 5+ level is or will surely be a Hall of Famer, it's not like I'm saying they're bums. I'm just saying that careers are pretty short, and you've gotta get the ball rolling early if you want to end up an all time great.