RealGM

Back in 2021, Odinn21 started a project on this board in which the top 12 seasons between pairs of players were ranked. Odinn21 chose the following players, reasoning that they held steadily as the board's top 10 of all-time in the Top 100 project:

- LeBron James
- Michael Jordan
- Kareem Abdul-Jabbar
- Bill Russell
- Tim Duncan
- Wilt Chamberlain
- Magic Johnson
- Shaquille O'Neal
- Hakeem Olajuwon
- Larry Bird (note: Bird was replaced by Kevin Garnett in the top 10 of the 2023 iteration of the Top 100 project)

A single thread in this project would be focused on ranking the top 12 seasons between a select two of these players. 10 players makes 45 combinations. It was an ambitious project, and it "only" got through 22* comparisons before Odinn stopped posting. I was always sad that the project didn't make it to the end because it was a great idea and executed well.

But a significant portion of the project *was* completed, and importantly, a pretty balanced variety of comparisons were covered. Each player was scheduled nine comparisons with other players; all ten players had four or five of those comparisons completed. So in the data we do have, no one is particularly overrepresented. Thus, we can actually fill in the blanks pretty nicely.

I compiled all of the responses to the project and used the Bradley-Terry model to generate a master ranking of all player seasons included. The gist is that each respondent's ranking of player seasons represents a bunch of pairwise comparisons. If I rank twelve seasons 1-12, I am saying that Season 1 is better than Seasons 2-12, Season 2 is better than Seasons 3-12, Season 3 is better than Seasons 4-12, and so on. And using all of those implied pairwise comparisons from all of the project's responses, we can estimate the strength of each season.

In a survey of wine preferences, for instance, it might be difficult for respondents to give a complete ranking of a large set of wines, but relatively easy for them to compare sample pairs of wines and say which they feel is better. Based on a set of such pairwise comparisons, the Bradley–Terry model can then be used to derive a full ranking of the wines.

Through 23* comparison threads, 163 responses were submitted; an average of about seven responses per thread. The number of responses per thread mostly varied from five to 11, with the exception of the last thread which received four. All 163 individual rankings were compiled to generate 4810 implied pairwise comparisons. Fed into the Bradley-Terry model, we get:

Results



A reminder: this is an aggregation of the opinions of some people on this board approximately three years ago. Nothing more, nothing less.

As some may notice, the top of the ranking does not align with this board's GOAT Peaks project. LeBron might stand out the most. A few points on that:

- 2013 LeBron was ranked #2 in both the 2019 and 2022 iterations of the GOAT Peaks project, and that was with more people's responses considered. Here, 2013 LeBron is not in the top 10. Clearly, we got a sample of people who were higher on 2009 and 2012 LeBron than those in the GOAT Peaks project.
- In the LeBron vs Chamberlain thread, only five responses were received and 3/5 picked 1967 Chamberlain over any LeBron season; not a representative opinion on the board, but just how the sample worked out.
- In the LeBron vs Duncan thread, there were eight responses, and four different seasons received two first place votes apiece: 2003 Duncan, 2009 LeBron, 2012 LeBron, 2013 LeBron. The lack of consensus on his peak doesn't help.

All in all, there's not much reason to look at the top of this list when the GOAT Peaks project exists. This was really just a fun statistical case study and a look back on a cool project, take this with a grain of salt.

Project Participants

For the sake of knowing whose opinions are being considered in this ranking along with crediting those who gave us data to work with, here are the project participants including the number of threads they submitted a ranking to (out of 23). Obviously the opinions of the kind folks who submitted 20+ rankings have a great deal of influence on these results.



Data Source & Links

Original project master thread, including individual comparison threads

* Note that after the project was seemingly dead, Djoker compiled the results of Odinn's last comparison thread and then started (and compiled the results for) another thread in an attempt to continue the project. So thanks to Djoker, we actually have 23 player comparisons covered here rather than 21 (as I would've been too lazy to compile the results of Odinn's last thread, #22).

Djoker's spreadsheet

Visualization showing the model's score estimates for each player's seasons. Calculated the median score estimate of each player's top 8 (a rough attempt at an "average quality of prime" measure), marked by those horizontal lines. Didn't add error bars for those, but there is some uncertainty there.



Same disclaimers in the OP apply w.r.t to this simply being an opinion aggregation and not based on anything outside of that.

I really wish the project was finished, but without Odinn it just wasn't the same anymore.

Thank you very much for this work though, it's very interesting to look at the aggregated results presented in a pleasant form.

Yea I even tried to continue the project but it was some time later and the ship had sailed.

70sFan wrote:I really wish the project was finished, but without Odinn it just wasn't the same anymore.

Thank you very much for this work though, it's very interesting to look at the aggregated results presented in a pleasant form.

Anyone heard from him? He just disappeared completely.

Would be a neat project to see done again at some point, maybe with a couple more players (KG? Steph?)