APBRmetrics

I just finished listening to the Dallas@Clippers game, which the Clippers had led somewhat comfortably (12 points I think around the end of the 3rd quarter), only to go through a horrendous spell -- something like a 15-2 scoring run by the Mavs, followed by further pounding until the Mavs led by 16 points in the 4th -- and then the Clips staged a counter-comeback and ended up winning in a game which went down to the wire.

We users still need to be given some basic stats -- mean and standard deviation at a minimum -- before we can even begin to make sense of the "Excitement" and "Comeback" ratings on the Inpredictable WPA webpage. But based on the figures that I've seen, and using this game as a benchmark, we're starting to see what the distribution looks like:

This game had a Comeback rating of 999.9. Even without knowing the underlying definition or calculations nor the empirical distribution, we can say "yup". This game was about as high on anybody's comeback scale as a game could get, with perhaps the exception of the super comebacks for 20+ points down.

This game's Excitement rating was 7.4. The website frequently shows games with ratings above 7.0 -- and they always seem to be close ones, so 7.4 may be getting at least somewhat close to the max that we can expect to see.

I should add that now I see that the triple OT game had a much higher Excitement rating: 14.2, which is way higher than any other rating I've seen. But, well triple OT, so okay. That game's comeback rating was 14.2, also high but not close to the comeback that the Mavs staged and the counter-comeback that the Clips staged.

I don't quite understand your rationale that its not right to add WPA from FTs, rebounds, etc. to the official WPA. Do you mean that the "win probability" added by an offensive rebound or made free throw is of a different kind than that added by a made field goal?

mtamada wrote:We users still need to be given some basic stats -- mean and standard deviation at a minimum -- before we can even begin to make sense of the "Excitement" and "Comeback" ratings on the Inpredictable WPA webpage. But based on the figures that I've seen, and using this game as a benchmark, we're starting to see what the distribution looks like:

This game had a Comeback rating of 999.9. Even without knowing the underlying definition or calculations nor the empirical distribution, we can say "yup". This game was about as high on anybody's comeback scale as a game could get, with perhaps the exception of the super comebacks for 20+ points down.

This game's Excitement rating was 7.4. The website frequently shows games with ratings above 7.0 -- and they always seem to be close ones, so 7.4 may be getting at least somewhat close to the max that we can expect to see.

I was thinking of adding a ranking next to each game's Excitement and Comeback number, showing how that game compared against all games in the season. Here are some stats based on the 2013-2014 season so far:

Average Excitement Index: 5.9
Average Comeback Factor: 7.8

deepak wrote:I don't quite understand your rationale that its not right to add WPA from FTs, rebounds, etc. to the official WPA. Do you mean that the "win probability" added by an offensive rebound or made free throw is of a different kind than that added by a made field goal?

To clarify, free throws do count towards a player's win probability added (getting to the foul line, making free throws, and missing free throws all get counted).

My rationale for not including rebounds (and assists, blocks, and steals) is due to the fact that those are "one way" stats. The player gets credit for pulling in the rebound, but there is no easy way to assign blame for not getting a rebound, short of distributing the WP loss evenly amongst the players on the floor.

Put another way, my WPA stat attempts to measure what a player does with a possession, when entrusted with it. They could: turn the ball over (the worst), miss a field goal (bad, but better than a turnover), get to the foul line and make a decent percentage of their free throws (better), or make a field goal (the best).

This is still a work in progress, so I'm open to suggestions. I'm also trying to better articulate the rationale underlying my current win probability added definition.

boooeee wrote:

deepak wrote:I don't quite understand your rationale that its not right to add WPA from FTs, rebounds, etc. to the official WPA. Do you mean that the "win probability" added by an offensive rebound or made free throw is of a different kind than that added by a made field goal?

To clarify, free throws do count towards a player's win probability added (getting to the foul line, making free throws, and missing free throws all get counted).

My rationale for not including rebounds (and assists, blocks, and steals) is due to the fact that those are "one way" stats. The player gets credit for pulling in the rebound, but there is no easy way to assign blame for not getting a rebound, short of distributing the WP loss evenly amongst the players on the floor.

Put another way, my WPA stat attempts to measure what a player does with a possession, when entrusted with it. They could: turn the ball over (the worst), miss a field goal (bad, but better than a turnover), get to the foul line and make a decent percentage of their free throws (better), or make a field goal (the best).

This is still a work in progress, so I'm open to suggestions. I'm also trying to better articulate the rationale underlying my current win probability added definition.

Thanks for the explanation.

On the WPA player page, the WPA shown is just the sum of WPA due to SHOT and TO. That's why I assumed it didn't include FTs (though the mouse-over description says differently).

http://stats.inpredictable.com/nba/wpPlayer.php

New Stat added to the player summary page: "Kitchen Sink" Win Probability

Basically, this sums up every win probability contribution that shows up in the player's box score. It's an "unofficial" stat in my mind because I feel it's adding apples and oranges, but it was easy enough to add. Stephen Curry is #1 according to this measure (Lebron #2, Kevin Durant #8).

I only recently became aware of the website Gambletron 2000 which tracks win probabilities in a manner similar to what Inpredictable does (and Fangraph's WPA graphs for baseball). But Gambletron 2000 does it for all of the major sports including at the college level. They also have measures of "excitement" and "comeback". Interestingly, according to them the sports league with the highest average excitement per game is ... the NBA.

mtamada wrote:I only recently became aware of the website Gambletron 2000 which tracks win probabilities in a manner similar to what Inpredictable does (and Fangraph's WPA graphs for baseball). But Gambletron 2000 does it for all of the major sports including at the college level. They also have measures of "excitement" and "comeback". Interestingly, according to them the sports league with the highest average excitement per game is ... the NBA.

Nice find. That's a nice shortcut to producing in-game win probabilities. Similar to my team rankings: don't worry about building everything from the ground up, let the market do the hard work for you.

My guess is that the frequent scoring, alternating possession format of basketball adds a lot to the excitement index. The WP graphs almost become fractal-like compared to the smoother NFL graphs, or really smooth graphs for hockey or soccer.

A new feature added to the box scores: Win Probabilty Added and the Four Factors. Basically this tells you how the game was won (from the floor, at the line, on the boards, or carelessness with the ball).

New model and new features launched today (mostly "entertainment" features):

-Option to display win probability as a function of real time.
-Option to display probabilities callibrated to the pre-game odds (as implied by the point spread).
-"Leverage" calculated for all shots, free throws, and rebounds.

New location too: http://stats.inpredictable.com/nba/wpBox.php