Sports Reference Blog

New Hall of Fame Prediction Model

Posted by sean on October 23, 2014

We maintained a Hall of Fame prediction model that, overall, works pretty well. This looks like a good list, yes?

hof

 

However, keep scrolling and you'll see that a few players are slipping through the cracks. Miller is already in and good money is on both Mutombo's and Nash's eventual inductions.

nashmutom

 

 

So we've tweaked our model a little bit. It's no less simple than what we've been doing, but it casts a wider net around those factors that voters and the media prioritize for player inductions. Its criteria include All-Star games, leaderboard appearances, peak dominance, championships, and the player's height (yes, height!). Many of these variables are considered as well by our old model.

A necessary disclaimer is that Hall voters, to varying extents over the years, have used criteria that are difficult to account for in a simple predictive model such as college stardom, FIBA success, foreign careers, and so on. In a nutshell, we're considering plaudits earned while playing in the NBA so adjust as you wish those careers of players like Manu Ginobili and Pau Gasol who have racked up FIBA medals in addition to their pro career.

Anyway, here's the list of the top 25 retired and active players...

 

RETIRED:                                                             ACTIVE:

hof2_car hof2_act

Except for Mutombo (6.4% to 86.8%) nothing changes too dramatically. Ben Wallace gets back in the race, and Nash becomes a virtual lock. Chauncey Billups and Shawn Marion move up significantly while Carmelo falls 10+ points.

To gauge your thinking about how many players "should" get in, here are the number of players in every NBA season between 1979-80 and 1999-00, inclusive, that were eventually enshrined:

active_num

We used the logistic regression method (same as before) and found that this works fairly well:

my $logit = -0.20303
              + -0.14203 * height_inches
              +  0.80573  * championships
              +  0.01594  * leaderboard_points
              +  0.41568  * peak_ws
              +  1.02443  * all_star_games

I'll get the documentation page up shortly which will explain a few of the nuances, for example how we determine "leaderboard points" and a few other particulars.

5 Responses to “New Hall of Fame Prediction Model”

  1. The 10-man rotation, starring what Steve Nash gave us and what we’re going to miss | So Stadium Status Says:

    […] Sports-Reference.com. Basketball-Reference just tweaked its Hall of Fame Prediction Model, and a certain two-time […]

  2. The 10-man rotation, starring what Steve Nash gave us and what we’re going to miss | Sports Says:

    […] Sports-Reference.com. Basketball-Reference just tweaked its Hall of Fame Prediction Model, and a certain two-time […]

  3. Ben Says:

    Can you expand on the leaderboard points? Is it 10 pts for 1st, 1 point for 2nd-10th? Is it for season totals or per game average subject to a game minimum?

  4. David Corby Says:

    @Ben - It's 10 points for 1st, 9 points for 2nd, 8 points for 3rd, and so on. So 10th in a leaderboard category is 1 point.

    This is for season totals, no per/g averages or minimums applied.

  5. Ben Says:

    Thanks, David.

    I had misread Tony Parker's page when I was looking at his example which led to my confusion.